human ear mouse experiment

The History of the Lab Rat Is Full of Scientific Triumphs and Ethical Quandaries

Lab rodents have been used in animal testing for more than 150 years, and the number of rodent-based studies continues to grow

Sam Schipani

More than 20 years ago, two Harvard University medical researchers, Joseph and Charles Vacanti, led a team that successfully grew a human-ear-shaped piece of cartilage on the back of a lab mouse. The experiment used an ear-shaped mold filled with cartilage cells from a cow. The “ear” was first placed into an incubator, and once it began to grow, it was transplanted into the body of a nude mouse (a species of laboratory mouse with a genetic mutation that causes a degraded or absent thymus organ, inhibiting the animals’ immune system and ability to reject foreign tissues).

“Earmouse” or the Vacanti mouse, as the animal has become known, continued to grow the piece of tissue out of its back until it resembled the size and shape of a human ear. The team published their research in Plastic and Reconstructive Surgery in 1997. The experiment was designed to test the viability of growing tissues for later transplant to human patients. And just last year, human children in China suffering from a genetic defect called microtia, which prevents the external ear from growing properly, received new ears grown with their own cells —a similar process to growing the “ear” on earmouse.

The mouse with a human ear on its back may have been one of the more bizarre and visually unsettling experiments carried out on a rodent, but mice have been used for scientific experiments since around 1902, when a quirky and enterprising breeder named Abbie E. C. Lathrop recognized the animals' potential for genetic research. The first use of rats in experiments started even earlier, with records dating back to the 1850s. Scientists purchased their subjects from professional breeders known as “rat fanciers” who prized the creatures as pets for their unique coats and personalities. For decades, lab rats and mice have been used to make great scientific and medical advances , from cancer drugs and HIV antiretrovirals to the yearly flu vaccine.

Lab mice—most often of the species Mus musculus, or house mouse—are biomedical swiss army knives, with genomes that are easily manipulated for genetic studies. The physiology of the human body, however, is more closely mimicked in Rattus norvegicus , or the Norway rat , and its various strains. Rats are also easily trainable and perfectly suited for psychological experiments, especially considering their neural networks so closely resemble our own . (In the 1950s and '60s, for example, researchers studying the biological underpinnings of curiosity noted that lab rats, devoid of any other stimulus or task, prefer to explore the unknown parts of a maze .)

Rats are also much larger than mice and have thicker tails and blunter snouts. But it's the characteristics shared by mice and rats that make them both scourges of the city and the perfect scientific guinea pigs, so to speak.

“They reproduce quickly, they are social, they are adaptable, and they are omnivores, so they’ll eat pretty much anything,” says Manuel Berdoy, a zoologist from Oxford University. Additionally, the rodents’ diminutive size allows relatively easy storage in labs, and their shared evolutionary roots with humans mean the species’ genomes overlap overwhelmingly.

As a result, rodents have all but taken over our labs, making up nearly 95 percent of all laboratory animals . Over the past four decades, the number of studies using mice and rats more than quadrupled, while the number of published papers about dogs, cats and rabbits has remained fairly constant. By 2009, mice alone were responsible for three times as many research papers as zebra fish, fruit flies and roundworms combined.

Studies with rodents address everything from neurology and psychology to drugs and disease. Researchers have implanted electronics into mice brains to control their movements , repeatedly tested the addictive properties of cocaine on mice , administered electric shocks to rodents as a negative stimulus , implanted human brains in mice skulls , and sent mice and rats scurrying through endless labyrinths of tests . NASA even keeps lab mice aboard the International Space Station for experiments in microgravity.

For all that lab mice and rats have helped humans accomplish, the day-to-day experience of the animals takes place largely out of the public eye. But the life of lab rodents may be key to understanding and improving their role in the course of scientific discovery.

Scientists must complete animal handling and ethical training before they are permitted to work with laboratory animals, though the rules vary depending on where the experiment takes place. While Canadian and European scientists are overseen by a national governing body, the rules in the United States vary by institution with some overall guidance from the National Institute of Health . (The U.S. Animal Welfare Act , which protects most animals used for research, excludes mice and rats.)

Most universities offer a training course on how to handle the animals in a way to best reduce stress and suffering. The best practices have been updated over the years to reflect a changing understanding of the rodents and their needs. After a 2010 study published in Nature showed that handling lab rats by the tail causes more anxiety than guiding the animals through a tunnel or lifting them with cupped hands, labs around the world abandoned the previously common technique.

Scientists who want to experiment with rodents are required to fill out a detailed application explaining why the work requires animal subjects. The applications are judged based on a framework known as the three R’s : reducing the numbers of animals used, replacing the use of animals when possible, and refining the experiments in order to improve animal welfare.

“A rat or a mouse is not a test tube on legs,” Berdoy says. Housing conditions for the rodents, for example, has become a raison d’etre for lab animal welfare proponents. Most lab mice are kept in shoebox-sized cages (for rats, the space is about doubled) with a few squeaky companions. And although having fellow rodents satisfies the social needs of the animals, most laboratory housing lacks any sort of environmental enrichment objects to occupy the subjects. The size of their confinements also means they are restricted from natural behaviors like burrowing, climbing or even standing up straight.

Even though lab mice and rats are, at this point, genetically distinct from their wild counterparts, they retain many of the same instincts. Repressing these needs could cause undue stress on the animals and compromise scientific findings. Berdoy’s film, The Laboratory Rat: A Natural History , details how lab rats released in the wild behaved and interacted in a similar way to their wild ancestors. Scientists, he believes, should consider the nature of rats when designing experiments to get the best results. “If you are going to do experiments,” Berdoy says, “you need to go with the grain of biology rather than against it.”

In some cases, the impacts of going against the biological grain have already been observed. While the genetic homogeneity of lab rodents helps to remove distracting variables from focused experiments, it may also, more subtly, be skewing scientific results. In a 2010 study on the impacts of intermittent fasting diets, Mark Mattson, chief of the laboratory of neuroscience at the National Institute of Aging, observed that the positive neurological impacts that “metabolically morbid” lab rats derived from the diet regime did not translate to healthy, active humans. The results were only applicable to “couch potato” critters in a “bubble boy type scenario where … their immune systems are not being challenged with different viruses or bacteria.” As Mattson succinctly notes, “What you discover may not be reflective of a healthy animal.”

In other words, the use of static, homogenous, sheltered animals may not always be the best way to accomplish the ultimate goal of using lab rodents: to better understand, and in some cases cure, the human body and mind.

In general, the process of transitioning an experiment from rodents to humans is not haphazard. Besides the reams of paperwork, new drugs are required to be tested on two different animals—a small one, like a mouse or rat, and then a large one, usually a pig, dog or primate—before they move to human trials. According to the Pharmaceutical Research and Manufacturers of America, only one out of every 250 compounds tested on animals moves to human trials. For those that make it to approval, the entire process usually takes 10 to 15 years.

Even after the long road to human trials, many drugs and procedures that work on mice and rats do not work on people. The rodents' "couch potato" lifestyles could influence the results, or perhaps the slight differences between rat, mouse and human genomes produce different responses to drugs. In Alzheimer’s studies, for example, mice and rats are artificially given a condition that resembles the disease because they do not develop it naturally.

When a drug doesn’t work, the results are often disappointing and costly, but sometimes mistakes can be tragic. Thalidomide, a drug used to treat morning sickness in the 1950s and 60s, caused deformities in human babies despite being successfully and harmlessly tested in rats. The drug breaks down much faster in rats, and their embryos have more antioxidant defenses against its nastier side effects. In many cases, however, the reasons for a failed drug remain mysterious.

“This is one of the questions at the heart of medical research. No one has a good answer to it, and there may not be a good answer to it,” says Richard Miller, a professor of pathology at the University of Michigan. “There are enough success stories that people are optimistic, but not everything that will work in the animals will work in people.”

Whether an experiment will end successfully may be uncertain, but one thing is always guaranteed: death of the lab rodents. The body count is unavoidable; an estimated 100 million lab mice and rats or more are killed every year in U.S. labs for the sake of science. While some of the bodies are creatively repurposed as snacks for birds in sanctuaries , most are frozen and incinerated with the rest of the biological waste.

Rats and mice used in aging studies often live out their natural lives, but most lab rodents are terminated at the end of a study. Some are killed via lethal injection or decapitated with strict guidelines to reduce pain and suffering, but most often, they are suffocated in cages with carbon dioxide.

For some time CO 2 has been considered the most ethical end of life practice for these lab animals, but Joanna Makowska, adjunct professor at the University of British Columbia and Lab Animal Advisor for the Animal Welfare Institute, believes there is a better way. The carbon dioxide poisoning, she says, mimics the feeling of running out of air when you are holding your breath underwater, which causes undue fear and anxiety. “It’s not a good death. Anesthesia is more humane, but people are not really doing that because carbon dioxide is more practical and cheaper.”

In general, Makowska believes researchers should be making more of an effort to meet the “reduction” principle of the three R’s . “That really should be the first R ,” she says. At Harvard, scientists made an organ on a chip to help study drugs and model disease without using animal subjects. Researchers have even developed computer algorithms based on thousands of animal trials that can accurately predict the way tissues will react to certain compounds.

But these lab rodent reduction-based advances have yet to take off, and the number of studies using the animals continues to grow. And while animal rights groups will raise hell over the treatment of our other furry friends, the lab rat rights fight has yet to make a splash.

“I think it comes down to how much we like them,” Makowska says. “People invest themselves much more in non-human primates. When it comes to dogs and cats, we have relationships with these animals. We are much more likely to acknowledge that they suffer.”

After all, if a mouse or rat escapes the lab to the streets of the city, it is considered a pest; anyone can kill it with impunity.

Get the latest Science stories in your inbox.

Environment
More…

This site is being redeveloped. For all the latest ABC Science content click here .

Site navigation.

Ancient Worlds
Being Human
Energy & Transport
Environment & Nature
Health & Medical
Innovation & Technology
Space & Astronomy
News in Science
Great Moments in Science
Dr Karl on triplej
Sleek Geeks
School's video prize
Ask an Expert
Bernie's Basics
Quizzes & Games
Lesson Plans
Demonstrations
Ace Day Jobs
Talking Science
Catapult on triplej

Dr Karl › Dr Karl's Great Moments In Science

Mouse with human ear

It's said the camera never lies. But sometimes the caption on a photo can be wickedly misleading. Dr Karl remembers one famous instance of someone taking the Mickey.

By Karl S. Kruszelnicki

Subscribe to the Great Moments in Science Podcast

Latest dr karl's great moments in science.

Dr Karl: Did life begin on an invisible mountain range?
Sorry, not all animals take 12 seconds to poo
The real costs of air pollution
Why isn't cannibalism more common?
Can you get pregnant during a pregnancy?

Follow ABC Science

About | Contact | ABC Science

Teaching Science
All in the Mind
Body Sphere
FutureTense
The Health Report
Ockham's Razor
The Philosopher's Zone
The Science Show
ExperiMentals
Sleek Geeks video prize
Time Traveller's Guide to Australia
Voyage to the Planets

More science sites

Catchment Detox
Surfing Scientist
Walking with Dinosaurs

Related ABC sites

ABC Environment
ABC Health and Wellbeing
ABC Technology and Games

Search ABC Science

A–Z subject and location library
Science archives
Search ABC Science by keyword

Stay updated

Abc science newsletter.

Get ABC Science’s weekly newsletter Science Updates

Follow ABC Science on Twitter

ABC Science
Doctor Karl
ABC Health Online
More ABC programs and people on Twitter»

About the ABC
ABC Careers
ABC Contacts
Terms of Use
Privacy Policy
Accessibility
© 2019 ABC

Human Atlas

The Mouse with an Ear Growing out of its Back

Note: This article was first published in University College London’s MedTech Society Portal. The link can be found here .

In 1997, the scientific world got its first glimpse of the ‘ Vacanti mouse ’ - an eerie, hairless mouse with a human-sized ear growing out of its back. Despite the public controversy that the experiment initially faced over animal ethics, the techniques used by the researchers to achieve this feat brought medical research one step closer to an exciting goal: generating organs in the lab.

Tissue engineering is a recently new field in biology, and researchers have been using its potential to synthesise ‘ bioartificial ’ tissue, which consists of both living and manufactured material. Bioartificial tissue can be implanted into an organism, and the assimilation of this material within the organism’s existing tissues can be a powerful tool to modify cellular functioning, division, and growth. With healthcare systems consistently facing a severe shortage of suitable organ donors, building organs in the lab could be a challenging but potent solution.

Extracting Chondrocyte Cells from Cows

In order to build the human-shaped ear and implant it into a mouse, a team of researchers led by Joseph and Charles Vacanti first isolated cartilage fragments from cows under sterile conditions and used the cells found within the cartilage to model the human-shaped ear. Cartilage is a type of connective tissue that consists of cells known as chondrocytes , which are surrounded by an extracellular matrix of water and proteins like collagen and proteoglycans. These bovine cartilage fragments were treated with collagenase , an enzyme that digests collagen, and later filtered and centrifuged. Centrifugation allowed the researchers to obtain suspensions containing the desired chondrocytes within the pellet , which refers to the heavy components of the mixture that sink to the bottom of the centrifuge tube (see figure 1).

Figure 1: A sketch illustrating how chondrocytes are obtained post-centrifugation

Building a Scaffold

However, without a guide, the chondrocytes cannot grow into a specific shape such as an ear. This is where a scaffold comes in handy. A scaffold can be shaped into the form of a human ear, acting as a 3D template for the cultured cells to grow in. The polymer biomaterial forming the scaffold can either be derived from a natural source or produced synthetically, but researchers need to consider two important factors before making their choice: the polymer’s biocompatibility and biodegradability .

Biocompatibility refers to how successfully the implanted tissue is able to assimilate into the new biological system and interact with neighbouring tissues. This can vary based on its structural properties, such as its surface chemistry or porosity. Researchers must consider the properties of the target tissue type and use a compatible scaffold material to minimise toxic effects on the target tissue type.

Additionally, once the cultured cells have grown into their desired shape, the scaffold is no longer required. Therefore, the material that makes up the scaffold needs to be biodegradable. This allows the scaffold to degrade into waste products like carbon dioxide and water once it has fulfilled its purpose within the body.

In the ‘Vacanti mouse’ experiment, the researchers created a plaster mould using a 3-year-old child’s ear for reference. To create the polymer construct (see figure 2), they used a biodegradable polyester known as polyglycolic acid and submerged it in an organic solution for a few seconds. Following this procedure, the polymer construct was sculpted into an ear-shaped scaffold using the plaster mould. The bovine chondrocytes were then planted into the polymer construct and placed into an incubator.

Figure 2: A sketch displaying the human ear-shaped scaffold

Implantation of the Tissue-Polymer Construct

Once cellular growth begins in the incubator, the tissue-polymer construct is ready for implantation into the subdermal region of the mouse’s back. However, if researchers implant cells from a cow into a healthy mouse, their experiment will unfortunately fail. This is because healthy immune systems are trained to recognise unfamiliar cells and destroy them using antibodies, which means that the body of a healthy mouse will reject an implant made of foreign bovine tissue.

To overcome this obstacle, the researchers decided to experiment on athymic nude mice, which lack functioning immune systems and hair. An athymic mouse lacks a thymus gland and is therefore unable to produce white blood cells called T lymphocytes , which act as soldiers in the fight against infection. As a result, implanting bovine tissue into an immunodeficient mouse will not cause an immune response, allowing the implanted tissue to continue growing in its new environment (see figure 3).

12 weeks after implantation, the researchers sacrificed the mice using an overdose of anaesthesia in order to study the composition of the ears. Through careful examination using different stains like eosin and haematoxylin, researchers found evidence of more cartilage growth in the implants.

Figure 3: An image of the lab mouse, post-implantation.

Applications of Tissue Engineering Techniques

Current tissue regeneration techniques can be used both therapeutically to replace damaged tissues or cosmetically to change the appearance of certain body features. For instance, alloplastic implants , which are similar to prosthetic devices and made of materials like silicone , can be used to reconstruct the cartilage of the ear or the nose. However, these implants are usually temporary because they are vulnerable to infection and may interact negatively with the patient’s immune system.

A more common material used to alter the shape of the external ear and the nose is autologous cartilage , which is extracted from the patient itself, specifically from the costal cartilage in the ribs. While this can be incredibly effective, the consistency of the surgical outcomes may vary depending on the surgeon’s skillset, and extracting costal cartilage from the patient can sometimes lead to donor site morbidity or complications like scars.

The potential benefits of using autologous cartilage within a scaffold could not only include a consistent, predesigned shape for tissue implants but also reduce the operative time taken during current surgeries. Furthermore, using the patient’s own tissue is usually more practical than finding a suitable organ donor, since autologous tissue will not be rejected by the patient’s immune system and the possibility of infection is also minimised.

A current limitation in tissue engineering is that the tissue being prepared for implantation often lacks vascularisation or a blood supply, and relies mostly on diffusive processes to gain nutrients from the culture medium. This limits the resulting size of the tissue, which is a major challenge for generating larger organs in the lab.

However, there have been incredible recent advancements in tissue engineering such as the use of stem cells (undifferentiated cells that can divide and form specialised cells) to build miniature organs in the lab called organoids . It may even be possible to integrate blood vessels into organoids, which could increase their lifespan. Alternatively, seeding stem cells into a scaffold could be a potential method for building organs with numerous types of tissues. Significant scientific progress has also been made using 3D bioprinters which have been able to create liver tissue containing an embedded network of blood vessels. The addition of vascularised systems to these tissues is a major accomplishment that could allow scientists to synthesise larger organs in the future, revolutionising the field of bioengineering.

Ready for more?

The auriculosaurus. The ‘‘human-ear-bearing’’ mouse developed by the Vacanti laboratory that went onto epitomize Tissue Engineering. License to reproduce the image was obtained from the a (2002) BBC Photo Library. Color images available online at www.liebertpub.com/teb

Contexts in source publication

ANTONELLA CHESCA

Ho Fai Chan

Dietmar Werner Hutmacher
G. Abdulina
Nicole M. Vecin
Robert S. Kirsner
BIOMED ENG ONLINE

M.I. Gumeniuk

Recruit researchers
Join for free
Login Email Tip: Most researchers use their institutional email address as their ResearchGate login Password Forgot password? Keep me logged in Log in or Continue with Google Welcome back! Please log in. Email · Hint Tip: Most researchers use their institutional email address as their ResearchGate login Password Forgot password? Keep me logged in Log in or Continue with Google No account? Sign up

Human-like ears 3D-printed inside mice as surgery-free spare parts

By Alice Klein

5 June 2020

Replacement ears could be 3D-printed beneath the skin

SEBASTIAN KAULITZKI/SCIENCE PHOTO LIBRARY/GETTY IMAGES

Human-like ears have been grown on the backs of mice using 3D printing. The technique could potentially be used to construct new ears or other body parts in people without the need for surgery.

3D printing is increasingly being used to custom-build new body parts, like jaws , ribs and spinal vertebrae. But these parts must be printed outside the body and then surgically implanted, which carries an infection risk.

Now, Maling Gou at Sichuan University, China, and his colleagues have shown that body parts can be 3D-printed inside the body, at least in mice, so that surgery is not required.

First, the researchers injected a “bio-ink” made of hydrogel particles and cartilage cells into the backs of mice. Next, they shone ear-shaped patterns of near-infrared light onto the ink. The light caused the hydrogel particles to stick together and develop layer-by-layer into ear-shaped structures.

Over the next month, the cartilage cells grew around the hydrogel structures, eventually resembling the cartilage structures of real human ears. The mice had no significant inflammation or other side-effects.

The famous Vacanti mouse of the 1990s also had a human-like ear grown on its back , but it was made by implanting a pre-made plastic scaffold seeded with cartilage cells underneath the skin, rather than 3D printing the scaffold directly at the site.

Sign up to our weekly newsletter

Receive a weekly dose of discovery in your inbox! We'll also keep you up to date with New Scientist events and special offers.

More from New Scientist

Explore the latest news, articles and features

3D-printed egg shells can help bones regrow

Subscriber-only

Tiny brain sensor implanted without surgery dissolves after weeks

Bioelectronic patch uses living bacteria to treat psoriasis in mice

Woman touching her stomach in pain Shutterstock ID 2022089378; purchase_order: -; job: -; client: -; other: -

Swallowable sensor unfurls in stomach to monitor gut health

MOUSE GROWING HUMAN EAR IS TISSUE TRIUMPH

By associated press.

It sounds like something from a carnival side show: "The Mouse With A Human Ear On Its Back." But it's real. It's alive.

That mouse, and others of its kind, are at the leading edge of a science known as tissue engineering, which allows laboratories to grow skin and cartilage for transplant in humans.The mouse in question, in the laboratory of University of Massachusetts anesthesiologist Dr. Charles Vacanti, is helping researchers refine the technology that someday will allow them to regrow ears and noses for people.

Linda Griffith-Cima, an assistant professor of chemical engineering at Massachusetts Institute of Technology who helped Vacanti grow the first ears on mice, said she did it at the request of a plastic surgeon from Children's Hospital, Dr. Joe Upton.

"He said, `I see these kids who are born without ears. And I have boys who come in whose ears have been chewed off in playground fights, and I can't sew them back on because they're so chewed up,' " Griffith-Cima said.

So she set about creating an ear-like scaffolding of porous, biodegradable polyester fabric. Then she and Vacanti distributed human cartilage cells throughout the form and implanted the prototype ear on the back of a hairless mouse.

The mouse, specially bred to lack an immune system that might reject the human tissue, nourished the ear as the cartilage cells grew to replace the fiber. The mouse remains healthy and alive after the ear is removed, the researchers said.

"You end up with a piece of cartilage in the shape of an ear," Griffith-Cima said.

Griffith-Cima's and Vacanti's research follows in the footsteps of Vacanti's older brother, Dr. Joseph Vacanti, a surgeon who does liver transplants at Children's Hospital, and his close friend Dr. Robert Langer, professor of chemical engineering at MIT.

Twelve years ago, when Joseph Vacanti became head of the hospital's transplant program, he started searching for ways to grow new liver tissue in sick children instead of waiting for donor organs. Too many of his patients died before they could get transplants.

Now Joseph Vacanti can implant a polymer scaffolding in a diseased rat's liver and transplant new liver cells. The new liver will grow and function for up to six weeks, he said.

Langer, the Vicantis and other scientists now have managed to grow liver, skin, cartilage, bone, ureters, heart valves, tendons, intestines, blood vessels, and breast tissue on such polymers, Langer said.

Although no such tissue products have yet become available to the public, skin products are in the advanced stages of clinical testing on humans, and heart valves are in the early phase of clinical trials.

Someday, ears and noses will be grown in a test tube using the patient's own cells on a custom-designed scaffold. Other tissues will be grown from donated cells on polymer devices placed in the patient's body.

"Some tissues, like cartilage, we can grow all the way to perfect tissue before putting it in," Joseph Vacanti said. "In other tissues, we only grow it for a short time, then we implant it and the body takes over."

Dr. Michael Miller, an associate professor of plastic surgery at the University of Texas' Anderson Cancer Center, said the technology is promising.

"In fact, I think the next major advances to come in the field of reconstructive surgery are going to be due to tissue engineering," said Miller, who is working with Rice University scientists trying to grow bone tissues in shapes that are useful for plastic surgery.

The chemical engineers say their job now is to create better polymers. It's one thing to grow cartilage that holds a shape for cosmetic surgery, another to grow cartilage that could mend a shattered knee, Griffith-Cima said.

Come for the science, stay for the dirty jokes.

Daily mos: the vacanti mouse.

July 9, 2021 SciBabe Daily Moment Of Science 1

Just look at this bleb, this floof, this mouse that’ll grow an extra ear to listen to all your problems. Back in the 1990s the phrase ‘face transplant’ was, at best, technology from a Nicholas Cage movie. Organ and tissue transplants had been around for a spell, but the field had a long way to go.

Rather than trying to match recipients to donor ears- which brings up quite the mental image- scientists conjured what could be described as the most dystopian version of those Disney souvenirs.

Today’s Moment of Science… Mouse ears.

Dr. Charles Vacanti was an anesthesiologist before he joined his brother Jay, likewise a doctor, in researching tissue regeneration. He researched growing cartilage on a biodegradable scaffold in the late 1980s. One of the biggest challenges in growing tissue- after getting it to grow at all- is getting it to grow into something useful. The scaffold material was made of a woven mesh of polyglycolic acid and polylactic acid that would break down harmlessly into water and carbon dioxide in the body. The mesh gives the cartilage cells room to grow.

When Vacanti submitted his work for publication to a top tier journal, he was pretty shocked when it was turned down with the rationale that there were no practical applications for it.

In the absence of the technology to grow a literal middle finger, he’d find the most practical body part to grow.

Vacanti consulted with surgical colleagues and they largely agreed that the ear was the most complicated cartilage structure in the body to fix and recreate. He successfully seeded the biodegradable scaffold with cartilage tissue from a cow, and shaped it into the mold of an ear. Now he just needed to give this jumble of cow knee cartilage somewhere to fulfill its destiny of turning into a real ear.

For this, no ordinary mouse would do.

The nude mouse gets its name from its lack of hair, and because it abides no pants Wednesday. A random mutation back in the 1960s produced this critter. Along with no hair, it has no functional thymus, a vital part of the immune system. This means the nude mouse won’t mount a rejection response to organ and tissue transplants, making it incredibly useful in research.

For purposes of this experiment, it would show proof of concept, that the bioscaffolds could support growth of a cartilage structure in a non-human animal. Did they have to do it this way? Probably not. Did it get the desired effect? Absolutely. The mouse had a goddamn ear growing on its back made of cow. People lost their fucking minds.

‘It’s a violation of animal rights!’ ‘Humans are playing god!’ ‘Karen wants to speak to science’s manager!’

Charles tried explaining that he only wanted to show that it could be done, not that he wanted a whole ear-backed-mouse army.

According to an interview with Dr. Jay Vacanti, the mouse was perfectly fine after the ear was removed (which I really want to believe). It went on to live a normal lifespan for one of these mice, which is about a year. It wasn’t genetically modified, but the pictures of that mouse have been and will be used in anti-GMO propaganda until the heat death of the universe.

In 2018, a team in China successfully transplanted ears for several children with microtia, a disorder that causes underdeveloped or absent outer ears. The ears were grown using cells of the children’s own cartilage, building on the techniques that Vacanti was told had no practical applications nearly three decades prior.

Four of the Vacanti brothers are doctors and pioneers in the field of tissue regeneration and transplantation. In the decades since then, their research has moved on from the backs of mice and continued to push the field forward. They’ve developed their techniques in growing lung, pancreas, and even nerve tissue. They’ve made bespoke chest plates, tracheas, and even a thumb using a scaffold of coral.

But the only thing named after them is the Vacanti Mouse.

This has been your daily Moment of Science, pretty sure that if this tech becomes open source someone’s gonna grow a pair of Vulcan ears.

To get the daily MOS delivered directly to your inbox and to check out a giveaway happening right now, head to patreon.com/scibabe.

Join the discussion! Cancel reply

This site uses Akismet to reduce spam. Learn how your comment data is processed .

Discover more from Scibabe

Subscribe now to keep reading and get access to the full archive.

Type your email…

How the mouse with human ears changed the world?

Twenty years ago, a mouse with a human ear on its body caused waves of anger and criticism, but the reality later proved the value of this experiment.

During the 20th century, mankind witnessed great advances in the field of science. It is known that, as technology advances, medicine will advance to the point where every part of the human body will be grown externally and implanted like the replacement of parts of a broken machine. . Although these scientific theories have been debated for decades, no one is rational enough to wait for it to happen.

The peculiarity of the ear

Plastic surgery has developed very rapidly at the end of the 20th century, but in the human body, the ear is still the most difficult part to regenerate, because it is made of cartilage. Although cartilage can be made, it is difficult to make from human tissue. As a result, many people who have had an ear-related accident will have to live with an abnormally shaped ear or be permanently absent.

In the late 1990s, doctors Charles Vacanti, Joseph Vacanti and Bob Langer wanted to create human organs in the laboratory. They experimented with creating "biodegradable scaffolds" or structures that could dissolve inside a body. One day, Joseph Vacanti heard his colleague complain that it was difficult to create new ears for patients who lacked ears, because the ears had odd and complex shapes.

Picture 1 of How the mouse with human ears changed the world?

That irony fact prompted the birth of the 'mouse-ear' project in the 1990s, led by Charles Vacanti, an expert in stem cell and tissue engineering. In the same year, Charles Vacanti, with the help of his brother Joseph Vacanti (tissue regeneration expert), attempted to grow a small piece of human cartilage on a biodegradable scaffold.

The scientists decided to make a scaffold shaped like a human ear and put cartilage cells from a cow on it (cartilage is a semi-rigid tissue found in the human ear, nose and chest). The scientists then selected a line of mice that was immunocompromised, meaning its immune system did not reject foreign bovine cells. They anesthetized the mouse, made an incision, and placed an ear shape under its skin. As expected, the body of the mouse feeds on bovine cartilage cells and when the scaffolding dissolves, the mouse is left with the shape of an artificial ear, without an eardrum.

The world was amazed when all the major news agencies shared a picture of a mouse - known as 'Vacanti mouse' - Vacanti mouse, 'earmouse', or 'ear-mouse' - wearing a human ear . Some expressed great excitement, most expressed fear, others expressed outrage, attributing the ethical aspects of such experiments.

A movement against genetic engineering exploded in the Western world due to the misunderstanding that the experiment involved genetic engineering, that the DNA of the rat had been genetically engineered to create human ears on its back. The misinformation is also due to some news agencies using such keywords to advertise the photo, not knowing that the actual testing began nearly 10 years before the photo was available, and without genetic engineering. which were used in this experiment.

In fact, the truss frame is made from a synthetic material polyglycolic acid, which is commonly used in plastic surgery. The fibers of this material are molded into a loose mesh membrane in the shape of an ear with 97% air, which leaves plenty of space for the cells to fill. This material will dissolve into carbon dioxide and water as tissue begins to grow in the affected area.

If doctors perfect the technique in mice and then in large animals, maybe one day they will help humans grow missing body parts. It sounds simple, but the process takes about 8 years until it is ready to be introduced into an organism to develop. Attachment to human tissue would not be effective because it would not regenerate quickly enough before the original cartilage had dissipated. Another problem is that all immune systems in all organisms will identify this cartilage as a foreign body and will attempt to eliminate it.

Special mouse

The rat used for this experiment is called a 'naked mouse' because of its lack of hair. A random mutation the species has acquired leaves them hairless and immune system. The coat doesn't make much of a difference, but the lack of an immune system is what makes this rat so special and perfect for the experiment. Without an immune system to fight off the foreign body, the cast cartilage can be filled with cells until it has fully grown into the ear.

There are no special requirements for human cells, as long as the cells are healthy and growing fast enough. Synthetic ear cushions are created to recreate the ear of a 3-year-old child. This ear after the transplant will grow again as the child grows up. Synthetic ear cartilage was surgically placed on the back of the mouse and kept there for 12 weeks until the scaffold was filled with living cells.

This artificial ear is up to 90% similar to the natural human ear, which is very surprising since the experiment did not involve any genetic engineering with human DNA. The ear was then successfully implanted in a child.

Despite the misunderstandings, success has demonstrated the capabilities of medicine and science in general. People understand that the future will surpass sci-fi movies. The Vacanti mouse is not simply an exercise to help scientists understand how to develop human organs, using skin cells and cartilage.

In January 2018, doctors in China and Japan published a study: In 2015, they recruited children with a deformed ear. The scientists scanned their normal ears, inverted their shape using a computer, and 3D printed a biodegradable scaffold. They then added cartilage cells from the patient and placed the scaffolds under the skin. As a result, cases of malformations in one ear now have two almost normal ears.

Without the experimental rat-ear, medical advances of this kind might not have occurred. Today, the 'implantation' of the human ear has been successful, what awaits us in 20 years and beyond?

Successfully nourish human ears on mice
Create human ears with apples - The marvelous project of humanity
Amazing things from ears
Hover your mouse as lightly as the air
New finding: Mice don't laugh with their mouth but with their ears
1001 wondering: How much does the world's largest mouse weigh?
See artificial ear from human ribs
Reconstructing ears from patient cells
How has Google changed our brains?
Science has created a mouse carrying human cells
How to handle when insects get into their ears
Upgrade the mouse intelligence

Influencers
Become an Affiliate
Accessibility
Terms of Use
Privacy Policy
Modern Slavery Report
All Stories

Why Scientists Put An Ear On A Mouse

Haven't you heard it's called the earmouse..

The “Vacanti Mouse.” The “earmouse.” That freaky thing you saw in a biology textbook or email chain. Whatever you call it, one thing’s for sure: the mouse with the ear on its back is an icon of science, and it has been for more than 20 years .

Maybe it even represents mad science. However, many misunderstand how and why the mouse was created in the first place.

Why Scientists Put an Ear on a Mouse

In the late ’90s, doctors Charles Vacanti, Joseph Vacanti, and Bob Langer wanted to create human body parts in a lab. They had experimented with creating “biodegradable scaffoldings,” or structures that would dissolve inside a body, in various shapes. One day, Joseph Vacanti heard his colleague complain that it was so hard to create new ears for patients who are missing them, as ears have such peculiar and complicated shapes. That’s when he decided to make a scaffolding in the shape of a human ear.

Shaping the Scaffolding

The researchers created an ear-shaped scaffolding and put cells of cartilage from a cow on it. Cartilage is a type of semi-rigid tissue found in your ears, nose, and chest.

Then, the scientists took a strain of mouse that was immunocompromised, meaning that it didn’t have an immune system that would attack the foreign cow cells. They put the mouse under anesthetic, made a surgical incision, and placed the ear shape under its skin.

Screen-Shot-2019-10-07-at-10.34.00-AM-1024x730.png

COURTESY OF THE LABORATORY FOR TISSUE ENGINEERING AND ORGAN FABRICATION, MASSACHUSETTS GENERAL HOSPITAL, BOSTON, MA, USA, DR. JOSEPH P. VACANTI, DIRECTOR

As predicted, the mouse’s system fed the cow cartilage cells, and as the scaffolding dissolved, the mouse was left with an artificial shape of a human ear. Although, It was only the outside part of the ear with no eardrum, making the function of the ear completely obsolete.

Then, the researchers repeated the process again and again, as they often do with experiments. “There were lots and lots of animals, because it was science,” said Vacanti. That means that the iconic earmouse was just one of many earmice !

When the scientists had important results, they published a study in the journal Plastic and Reconstructive Surgery .

Rumors and Hoaxes

In 1998, the BBC aired a program with an earmouse in the trailer. And, to say the least, the world was stunned. Many were impressed by the feats of science, but some were concerned, leaving rumors and explanations to run amuck.

In 1999, a group took out a full-page ad in the New York Times with a picture of the mouse and the question: “Who Plays God in the 21st Century?” The ad suggested that the mouse was a product of genetic engineering, but that’s incorrect. The ad also says that Biotech companies are “blithely removing components of human beings, and other creatures, and treating us all like auto parts at a swap meet.”

However, the ears on the mice never came from, nor went to, any humans. Instead, the project on the mice was intended to be practice. If the doctors could perfect this technique in mice, and then in large animals, maybe one day they could help humans grow their own missing body parts.

Helping Humans Today

The Vacanti mouse was not simply an exercise in creating Kronenberg-style horrors. It was meant to help scientists understand how to grow body parts in humans, using their own skin and cartilage cells.

In January of 2018, doctors in China and Japan published a study showing that they had achieved just that. Two-and-a-half years prior, they had recruited children with one malformed ear each. The scientists scanned their normal ears, reversed the shape using a computer, and 3D-printed a new biodegradable scaffolding. They added cartilage cells from their patients and put the scaffoldings under the skin. As a result, the children now have two ears that are mostly normal .

Without the strange, pink mouse in the biology textbook, these types of medical advancements may have never happened.

Share This Story

About The Author

Ripley's Believe It or Not!

Step into the world of Ripley’s Believe It or Not!, where truth is always stranger than fiction! Bui…

By this author

Meet Bandit, the Turtle Who's Stealing Our Hearts!

10 Most Underrated Animals at Ripley’s Aquarium of Canada

The Mysterious World of Jellies

Say "Shell-o" to Ripley's Aquarium of Myrtle Beach's newest superstar.

These under-the-radar critters are more JAW-some than you realize!

A deep dive into one of the Aquarium’s most mysterious creatures.

More on Animals

Or explore our categories, have an amazing story.

At Ripley’s, we’re always in search of the unbelievable – maybe it’s you! Show us your talents. Tell us a strange story or a weird fact. Share your unbelievable art with us. Maybe even sell us something that could become a part of Ripley’s collection!

Cartoon of the Day

In 1958, Bruce Lee became the Crown Colony Cha-Cha Dancing Champion of Hong Kong.

Robert Ripley began the Believe It or Not! cartoon in 1918. Today, Kieran Castaño is the eighth artist to continue the legacy of illustrating the world's longest-running syndicated cartoon!

Share full article

Supported by

It’s Possible to Grow a 3-D Printed Ear on a Mouse’s Back

By Nicholas St. Fleur

Feb. 22, 2016

This unattached ear bathed in pink goo may look like a freaky find from Frankenstein’s laboratory, but it’s actually the product of a decade worth of medical research with 3-D printing.

Bioengineers from the Wake Forest Institute for Regenerative Medicine in North Carolina crafted the ear along with a jawbone, skull bone and skeletal muscle using what they call an “integrated tissue and organ printing system.” They then implanted them into mice and rats and found that the 3-D printed biological structures not only stayed alive for several months, but grew.

The team members published the blueprints behind their constructions this week in the journal Nature Biotechnology . In the past, medical researchers have created similar chunks of tissue and organ prototypes using 3-D printers loaded with live cells, but many of those prints were either structurally unfit for transplantation or unable to survive within a host.

Scientists have placed human-size ear structures into rodents before, but those ears were not 3-D printed, or did not keep their structure for long or did not grow cartilage and blood vessels as this one did.

Though we are still a ways from implanting 3-D printed hearts and kidneys into mice or rats, these are steps toward creating replacement organs that can be transplanted into people. More than 120,000 Americans are currently waiting for lifesaving transplants according to the U.S. Department of Health & Human Services , and the researchers said their printing technique could one day produce clinically useful tissues and organs.

The new machine works similarly to traditional 3-D printers in that it squirts out layer upon layer of material to form its product. But rather than use plastic, this printer uses a mix of human, mice, rat or rabbit cells and gelatin as its ink. The machine combines its primary building material, called hydrogel, with a biodegradable plastic to produce a stable structure strong enough to keep its shape upon transplantation. Both materials allow oxygen and blood to flow through the printed tissue, keeping it alive within its host.

In this experiment the researchers inserted the ear beneath the skin of a mouse’s back and found that several months later the rodent created blood vessels that attached to the printed ear and enabled it to thrive. The researchers also inserted muscle tissue printed from their machine into a rat, which later developed both blood vessels and nerves to the implant.

The team hopes that similar outcomes will occur if they implant their 3-D printed biostructures into people, and they plan to conduct human trials next.

Human ear grown on the back of a rat

Posted: by UAR News on 29/01/16

25/01: Human ear grown on the back of a rat Japanese scientists have grown a human ear on the back of a rat in order to help children born with facial abnormalities and adults who have suffered accidents. The ear was grown by turning stem cells into cartilage cells which were placed in inside plastic tubes shaped like a human ear on the rat’s back. The framework dissolved after two month leaving behind a two-inch ear lying flat against the animal’s back. Currently, replacement ears are sculpted from cartilage taken from the patient’s ribs, however this requires multiple operations including the painful removal of the cartilage from the chest which never fully heals. The new technique is one of several being perfected around the world, in the aim of making bespoke replacements for body parts damaged by accidents, ravaged by disease or malformed at birth. http://www.dailymail.co.uk/news/article-3414756/Scientists-grown-human-EAR-rat-say-able-use-humans-five-years.html

26/01: New treatment effectively halt progress of Diabetes in mice American researchers have managed to halt Type I Diabetes in mice for six months by using insulin-producing cells. The researchers used human stem cells to create insulin-producing islet cells, to treat the condition. Such a method might be a route to effectively cure Type I diabetes, which affects around 400,00 people in the UK and currently requires them to take daily injections. According to The Times: “After implantation in the mice, the cells began to produce insulin in response to blood glucose levels, which remained within a healthy range for the length of the study. The findings are published in the journals Nature Medicine and Nature Biotechnology .” http://www.thetimes.co.uk/tto/health/news/article4674277.ece Also in Telegraph: http://www.telegraph.co.uk/news/science/science-news/12120141/Harvard-and-MIT-close-to-cure-for-Type-1-diabetes-which-will-end-daily-injections.html

27/01: Robotic screening tool is a step closer towards animal free toxicity testing An in vitro robotic screening tool able to screen thousands of chemicals in human cell lines has been developed by researchers at the National Institutes of Health's National Center for Advancing Translational Sciences. The tool has the ability to test environmental chemicals found in drugs, food and food packaging, consumer products, and chemicals produced during manufacturing and industrial processes using cell-based assays and is working towards reducing animal testing whilst predicting a chemical’s effects on human health. 10,000 chemicals were screened through 30 different automated, cell-based assays and the team were able to partly predict animal and human toxicity however the information is not perfect and additional chemical structure data is needed for more accurate predictions. http://www.the-scientist.com/?articles.view/articleNo/45173/title/Animal-Free-Toxicity-Testing/

Fertility effects observed when pregnant rats are given paracetamol In a study where rats were given paracetamol or the aspirin-like drug indomethacin, female animals gave birth to smaller litters of offspring that had smaller ovaries and fewer eggs than those not exposed to the medicines. Males were affected too, having fewer cells that make sperm later in life. Even though foetal development is slower in humans than compared to rats, scientists said the findings were significant given the similarity of the two species' reproductive systems. Paracetamol is widely used to treat headaches, while prescription-only indomethacin reduces inflammation therefore this study could have implications for pregnant women. http://www.bbc.co.uk/news/uk-scotland-edinburgh-east-fife-35418779

28/01: Faster test for schistosomiasis in mice

A piece of kit that quickly multiplies the DNA of parasitic worms could detect infections by schistosome species more than six times faster than the most accurate existing method. Test results published in PLOS Neglected Tropical Diseases last week show that the device can detect infections by Schistosoma mansoni in mice with high accuracy from a drop of blood — and it can do so after only a week of infection. This parasite is the main cause of the neglected tropical disease schistosomiasis (bilharzia). - See more at: http://www.scidev.net/global/disease/news/faster-schistosomiasis-diagnosis-disease.html#sthash.5TmBCBTZ.dpuf

29/01: A new e-learning resource focusing on the assessment of laboratory animal welfare has been launched to help researchers and animal care staff to identify signs of good and poor welfare in research animals.

Created by Professor Paul Flecknell and his team at Newcastle University, with funding from the NC3Rs, it is the second scenario-based training module to be added to the FLAIR e-learning site. Recipients of an NC3Rs Infrastructure for Impact award in 2013, the group from Newcastle are developing a range of web-based tutorials on best practice in the refinement of animal experiments.

http://www.nc3rs.org.uk/news/launch-nc3rs-funded-welfare-assessment-resource

Last edited: 9 March 2022 10:41

Back to News

Could AI replace animal research?

Openness in Animal Research Conference Returns in 2026

New fact-checking video series from UAR

Subscribe to our newsletter.

Get the latest articles and news from Understanding Animal Research in your email inbox every month. For more information, please see our privacy policy .

Distillations Podcast

Distillations podcast

The mouse that changed science.

A tiny animal with a big story.

In April 1988 Harvard University was awarded a patent that was the first of its kind. U.S. Patent Number 4,736,866 was small, white, and furry, with red beady eyes. His name was OncoMouse.

The mouse, genetically engineered to have a predisposition for cancer, allowed researchers to study the disease in an intact living organism. It promised to transform cancer research, but not everyone was happy. Most critics were wary of patenting life forms at all. But academic scientists were also worried about the collision of commercial and academic science. It forced them to face difficult questions: Who should pay for science? Who does scientific knowledge belong to? And should science be for the good of the public or for profit?

Hosts: Alexis Pedrick and Elisabeth Berry Drago Senior Producer: Mariel Carr Producer: Rigoberto Hernandez Reporter: Jessie Wright-Mendoza Photo illustration by Jay Muhlin Additional audio production by Dan Drago Additional music courtesy of the Audio Network

Research Notes

Elizabeth Popp Berman, Associate Professor of Sociology, SUNY Albany, and author of Creating the Market University: How Academic Science Became an Economic Engine .
David Einhorn , House Counsel, Jackson Laboratory .
Harold Varmus , Professor of Medicine, Weill Cornell Medicine.
Ken Paigen, Executive Research Fellow and Professor, Jackson Laboratory .
Adler, Jerry. “ The First Patented Animal Is Still Leading the Way on Cancer Research .” Smithsonian Magazine , December 2016.
Chakrabarty, Ananda. Microorganisms having multiple compatible degradative energy-generating plasmids and preparation thereof. U.S. Patent 4259444A , filed June 7, 1981, and issued March 31, 1981.
Diamond v. Chakrabarty, 447 U.S. 303 (1980).
“ Fortune Names Its ’88 Products of the Year .” Associated Press , November 17, 1988.
Hanahan, Douglas, Erwin Wagner, and Richard Palmiter. “ The Origins of Oncomice: A History of the First Transgenic Mice Genetically Engineered to Develop Cancer .” Genes and Development 21 (2007), 2258–2270.
Leder, Philip, and Timothy Stewart. Transgenic non-human mammals. U.S. Patent 4736866A , filed June 22, 1984, and issued April 12, 1988.
Leonelli, Sabina, and Rachel Ankeny. “Re-Thinking Organisms: The Impact of Databases on Model Organism Biology.” Working paper, University of Exeter, April 5, 2011. Published in Studies in History and Philosophy of Science Part C 43:1 (2012), 29–36.
Morse, Herbert C. III, ed. Origins of Inbred Mice . New York: Academic Press, 1978. Google Books.
Murray, Fiona. “ The Oncomouse That Roared: Resistance and Accommodation to Patenting in Academic Science .” Working paper, Massachusetts Institute of Technology, 2006. Published in American Journal of Sociology 116:2 (2010), 341–388.
National Association for Biomedical Research. “Mice and Rats.” Mice and Rats . Washington, DC, 2018. nabr.org.
National Museum of American History. “ OncoMouse .” Washington, DC, 2018. americanhistory.si.edu.
Palmer, Brian. “Jonas Salk: Good at Virology, Bad at Economics. ” Slate , April 13, 2014.
Rader, Karen. “ The Mouse People: Murine Genetics Work at the Bussey Institution, 1909–1936 .” Journal of the History of Biology 31:3 (Autumn 1998), 327–354.
Russell, Elizabeth. “Origins and History of Mouse Inbred Strains: Contributions of Clarence Cook Little.” Jackson Laboratory, Bar Harbor, Maine. informatics.jax.org.
Schneider, Keith. “New Animal Forms Will Be Patented.” New York Times , April 17, 1987.
Specter, Michael. “ Can We Patent Life? ” New Yorker , April 1, 2013.

Archival Sources

Achbar, Mark, and Jennifer Abbott, dir. The Corporation . Canada: Big Picture Media Corporation, 2003.
Albert and Mary Lasker Foundation. “Lasker Archives: Passion and Optimism in Scientific Research.” April 9, 2017, laskerfoundation.org. On the 1987 Albert Lasker Basic Medical Research Award.
Murrow, Edward. See It Now (Jonas Salk). CBS, April 12, 1955. paleycenter.org
Potter, Deborah, and Dan Rather. “Animal Patents.” CBS Evening News , April 12, 1988.
Ronald Reagan Presidential Library. “ Candidacy for Presidency: Ronald Reagan’s Announcement for President of U.S. ” November 13, 1979. youtube.com.

The Mouse that Changed Science

Today the job of building this nation geographically is completed. There are no new frontiers within our borders. So to what new horizons can we look now? Where are tomorrow’s opportunities? What’s ahead for you? For your children? The frontiers of the future are not on any map. They are in the test tubes and laboratories of the great industries.

Alexis : Hello, and welcome to Distillations , a podcast powered by the Science History Institute. I’m Alexis Pedrick.

Lisa : And I’m Lisa Berry Drago.

Alexis : Each episode of Distillations takes a deep dive into a moment of science-related history in order to shed some light on the present. Today we’re talking about a small animal that became a big story.

And a big controversy in the 1980s.

<1988 CBS news clip>

Something really new today from the U.S. Patent Office for the first time a patent for an animal.

Alexis: Chapter One. Patenting Life.

Lisa : In April 1988, Harvard University was awarded a patent that was the first of its kind. Patent Number 4,736,866 (Four million, seven hundred thirty-six thousand, eight hundred and sixty-six) was small, white, and furry. With little red beady eyes. His name was OncoMouse [ONKO-mouse].

CBS Host: The inventors didn’t build a better mouse. They changed one genetically in the lab. Deborah Potter reports on the latest Future Shock Over money medicine and ethics and whose life is it anyway.

CBS Reporter: Scientists at Harvard University genetically altered the mouse by adding cancer genes to make it more sensitive to cancer causing substances. It will be used along with its offspring to test new drugs.

Alexis: Mice and rats are ubiquitous in biomedical research labs throughout the world. They account for 95% of the animals used in scientific research. They’re really important tools for science, but they’re not necessarily that exciting. But this mouse, this mouse was different. This mouse was special. In 1988 it was on Fortune Magazine’s “product of the year” list — alongside E.T. on videotape, Rogaine, and the personal fax machine.

Lisa : OncoMouse was transgenic. This genetic engineering made it predisposed to getting cancer. And it also ensured that it would pass those cancer genes on to its offspring. Cancer researchers really needed this mouse. They needed to study cancer in an intact living organism—rather than just cell lines in petri dishes.

Alexis : The mouse was developed by molecular geneticist Phil Leder with help from Harvard researcher Timothy Stewart. And also funding from DuPont, the company that brought us Teflon and Kevlar. Phil Leder had already made significant contributions to understanding the genetic basis of cancer and he developed OncoMouse specifically to model the behavior of breast cancer in humans. He was optimistic about how the mouse would advance cancer research.

Phil Leder: For the first time there is hope for mankind that some of the age- old scourges can be treated.

CBS Reporter: But critics say that patenting lab animals is just a first step toward allowing scientists to play God.

Critics: The whole range of the animal kingdom is now open to anybody who can afford a patent application for genetic altering of animals for whatever purpose.

CBS Reporter: The patent office says it’s up to Congress to resolve the dispute over where to draw the line. As the law stands now, the only animal that can’t be patented is a human being.

Lisa : OncoMouse was the first mammal to be patented. But it wasn’t the first life-form to be patented. That happened seven years earlier, in 1981.

Justice Burger: We will hear arguments next in Diamond, Commissioner of Patents v. Chakrabarty.

Wallace: Mr. Chief Justice, and may it please the Court. The question before the court in this case is whether a living organism is patentable subject matter under Section 101 of the patent law.

Alexis : Diamond versus Chakrabarty was a Supreme Court case that opened the door to patenting lifeforms. It all started with bacteria.

Lisa : Oil-eating bacteria, to be exact. Microbiologist Ananda Chakrabarty [CHA-krah-BAR-tee] was working for General Electric when he developed a bacteria that could break down crude oil—making oil spill clean-ups a lot easier. He applied for a patent for it, but he was turned down by the U.S. patent office, on the grounds that this bacteria was a product of nature. But the case went all the way to the Supreme Court. And the Supreme Court saw things very differently. Here’s a clip from the 2003 documentary, The Corporation.

The Corporation: And they said this microbe looks more like a detergent or reagent than a horse or a honeybee. I laugh because they didn’t understand basic biology. It looked like a chemical to them. Had it had an antenna or eyes or wings or legs, it would have never crossed their table and been patented. Chief Justice Warren Burger said, sure some of these are big issues but we think this is a small decision. Seven years later the

U.S. patent office issued a one sentence decree. You can patent anything in the world that’s alive, except a full-birth human being.

Alexis : Scientists in the field had quickly recognized that their discoveries could also be commercial products—but only if they had patents.

Lisa: By 1987 the U.S. Patent office decided that they’d start patenting higher level organisms. And they wanted their first pick to be symbolic. So they chose OncoMouse. Because of the ways this little creature could help humanity.

Alexis: Most people who were concerned about OncoMouse were worried about the ethics of patenting life at all. But the academic science community found it alarming for entirely different reasons. Here’s Phil Leder in an interview from 1987.

Leder: What really received a lot of publicity wasn’t the fundamental science that was generated by creating these animal models, but the fact that they were patented and that was grist for a lot of cartoonists mills…It’s both amusing of course and also ominous when you when you think about it…

Interviewer: Well let’s talk about that issue because Leder: Sure

Interviewer: because it’s a controversial issue wherever you come down on it. The fact is the private sector has an enormous amount of control now.

Leder: That’s our system. You know you may like it you may not like it. But that is our system.

Interviewer: But does private sector bring sources bring resources to the table that otherwise would not be?…

Leder: That’s absolutely. Private resources in this case supported the research. They

made the original investment…somebody has to feel that they can get a return on this.

Alexis : The story we’re going to tell you today is about this tension. Between commercial and academic science.

Lisa : Patenting OncoMouse set in action a chain of events that fundamentally changed academic science. It changed how academic scientists did science. And it forced them to face difficult questions, like who should pay for science? And who does scientific knowledge belong to? And should it be for the good of the public, or for profit?

Alexis: Jessie Wright-Mendoza has been reporting on this story for Distillations , and she’s going to take it from here.

Lisa: Chapter Two. Who Patents?

Jessie : In the world of commercial science patenting was—and still is—the norm. So patenting OncoMouse was a no-brainer for DuPont. But in academic science patenting was not the norm. Which isn’t to say that academic science didn’t have its own kind of reward system. It did. It was publishing. Which put your work in the spotlight and gave it—and you—validation. Again, here’s Phil Leder:

Leder: What you work for in this business is perhaps as the most valuable reward is the grudging appreciation of a few colleagues who really understand what you’re doing.

Jessie : Beth Berman is an associate professor of sociology at SUNY Albany. She wrote a book called Creating the Market University: How Academic Science Became an Economic Engine .

Beth Berman: Historically things had not necessarily been patented. So like the polio vaccine for example was notoriously not patented.

< 1952 archive audio>

Edward R. Murrow: Who owns the patent on this vaccine?

Jonas Salk: Well, the people I would say. There is no patent. This is…could you patent the sun? [Laughs]

Beth Berman : And Jonas Salk said how could you patent the sun? That it would just be completely inappropriate to patent something that was a scientific discovery that was meant to share and so that idea of openness was really in conflict with the idea that you would patent something and then use it to make money off of.

Jessie : The polio vaccine wasn’t alone. Penicillin was also intentionally not patented for ethical reasons. So that it could be accessed by as many people as possible.

Jessie : When OncoMouse was patented it really threw one particular group of academic scientists who were especially invested in lab mice.

Alexis: Chapter Three: The Mouse Club.

Pinky and the Brain: “They’re Pinky and the Brain! They’re Pinky and the Brain! One is a genius, the other’s insane. They’re laboratory mice! Their genes have been spliced!”

Alexis : This story is about how OncoMouse disrupted science. But within it there’s a small story and a big story. Now the small story is about how it affected a little group of scientists called mouse geneticists. Scientists who develop mouse models and use them to study illness. These scientists were a subset of the larger genetics community and they stuck together. They even had a club—The Mouse Club of America. Now that’s right. The Mouse Club. And they had a newsletter: “The Mouse Club Newsletter” that shared information on mouse strains and mouse happenings. You get the point.

Lisa: Mouse club was an established group with an established culture. And it all started at the beginning of the 20th century when the field of genetics was brand-new. Most research was done on plants, flies, and mice. If you wanted to study with mice, there was really one place you could do it – the Bussey Institute at Harvard, where zoologist W.E. Castle trained most of the first generation of mouse geneticists. Castle’s program was always short on time, money, and mice, so the early mouse men learned to share resources, including mice. When Castle’s students went off to start their own programs at other institutions, they carried this ethos of openness and sharing with them. It became part of the culture. Something else that was part of the Mouse Club culture was that they didn’t patent their mice. But OncoMouse changed everything.

Ken Paigen: The culture of the research community in the mouse genetics world was that if you had a mouse that had been genetically constructed once you had published the first description you were obligated by custom to distribute that mouse to anybody else who asked for it and people did that very freely. So there was a completely open exchange of mice among laboratories. And that worked.

Jessie : Ken Paigen is a research scientist at Jackson Laboratory in Bar Harbor Maine. Jackson Labs, or JAX, as it’s known, is an institution in the world of mouse genetics. For nearly a century, the lab has developed and distributed specialized mice for biomedical research. Today, mice from JAX go out to over 60 countries, to virtually every major research institution in the world.

David Einhorn: I’m David Einhorn. I was the House counsel for the Jackson Laboratory for about 23 years, I’m now retired. If I may add, if I may add on Ken, he’s still a very active researcher at the lab despite the fact that he’s passed 90 years old.

Jessie : Ken Paigen became the director of Jackson labs in 1989. About a year after Harvard was granted the OncoMouse patent. He didn’t know it then, but the lab was about to get swept right into the middle of the controversy over OncoMouse.

Jessie : Before I go any further I want to talk a bit more about the history of model mice and why they’re so crucial to science.

Ken : Well the primary mammalian model that we have as a surrogate for human research has been the laboratory mouse. All mammals have very similar structures to their genomes. Their basic compliment of DNA. And since it’s so difficult to do so many experiments if not morally impossible to do them in human beings we need some animal surrogate that can serve as a substitute.

Jessie : A mouse model is a representation of a human disease or illness. Just like an architectural model is a representation of a building. It’s really the only way for researchers to observe how diseases act in a living being. You can do similar studies on the cellular level, but cells don’t really have a lifecycle like a human do. But a mouse is born and goes through stages of development. It reproduces and has babies. It ages just like we do. So a researcher can see how cancer cells, for example, how they behave, what activates them. And where, how, and why they spread to other parts of the body.

Jessie : The first mouse models for research were created by selectively inbreeding mice to create a colony that all shared the same trait. Jackson Lab was founded in 1929 by CC Little. He had studied under W.E. Castle at the Bussey Institute. At that time mouse models were made through selective inbreeding, or they were spontaneous mutants. These were mice born with some kind of abnormality. Tumors for example. They would be separated from the colony and bred to create a new colony of mice to share that mutation. The whole process was pretty hit or miss. Researchers usually had to bread multiple generations before that mouse family would reliably display that desired characteristics. Most scientists aren’t particularly interesting in moonlighting as mouse breeders, but sometimes it’s necessary. Phill Leder created OncoMouse because he wanted to study a particular type of cancer. The mouse he needed didn’t exist, so he made it. But maintaining mouse colonies is extremely time consuming and expensive. That’s where JAX comes in. CC Little realized he could make a business out of making, caring for and distributing mice to other scientists. And that business could underwrite the research of Jackson Lab scientist.

Ken : It’s really a problem to maintain a colony…if you have a mouse which is popular and is needed in large numbers around the world… it’s difficult to do it from an individual research lab. And so over time the custom had become for people to deposit their mutants, their mice here at Jackson Lab and then we would serve as the central distribution facility for the rest of the world. That relieved the individual researcher of the problem of maintaining shipping conditions, health status and all the rest of it.

Jessie : Scientists doing biomedical research need laboratory mice, and the mouse genetics community was built on the sharing of mouse models. They can get them from the repository at JAXS or get a breeding pair from another researcher. In 1947 JAXS was devastated when a fire swept through the town of Bar Harbor. Tens of thousands of mice were lost in the blaze. Within days scientists from around the country started sending breeding pairs that had come from Jackson’s own stock. That’s just how the community worked. So the patenting—and subsequent restricting of OncoMouse—a very hard pill to swallow. Harold Varmus is a Nobel Prize-winning cancer researcher who also worked with model mice.

Harold Varmus: It would never have occurred to me that that I should provide any kind of protection of that property. I didn’t think it was property. I thought of it as the creation of mouse strains that others might use to try to understand how these genes work and how breast cancer arises. It was a research tool.

Jessie: The OncoMouse patent made it difficult to obtain what was a crucial research tool for mouse geneticists and other scientists. For one thing, the mice were expensive—$50 a piece as opposed to $5 from JAXS or free from a colleague. Many researchers felt like the carpet had been swept from under them.

Alexis : Maybe it seems like this shouldn’t have been such a big deal. I mean it’s just the Mouse Club, right? Why are we even talking about this?

Lisa : It seems like that. But for them it was a huge paradigm shift. It completely changed how they had to work. Also, they have kind of a silly name, but the Mouse Club equals Cancer Research. And I think we can all agree that that’s important. And patenting OncoMouse slowed that down.

Ken : And the truth was it was all said to be enormously inhibitory of research in the field of cancer.

Alexis : And from there scientists were left wondering, “Whoa? What’s next? What are you going to patent next?”

Lisa : Which brings us to the big part of the story. The effects of patenting OncoMouse radiated out of the mouse club and into academic science as a whole. And forever changed that culture too.

Lisa: Chapter Four: Commercial Science. Academic Science.

Jessie : Outside of the mouse club many other academic scientists were concerned. They saw patenting as incompatible with the idea of open science. One unencumbered by money. One that was for the good of humanity. Not just for making a profit. Here’s sociologist Beth Berman again.

Beth : So universities tended not to be very oriented towards this kind of commercial activity. So they, it was it was kind of unfamiliar and there was a lot of resistance to it. But at the same time there are also a lot of financial opportunities that were very appealing for universities.

Jessie : Part of what made OncoMouse possible was an unusual relationship between academic and commercial science. At least it was unusual at that time.

Everything is preventative through chemistry. That’s the promise of DuPont.

David : Phil Leder had been funded by DuPont in his research and Harvard in return for that funding had given DuPont exclusive rights to commercialize any invention made from that funding.

Jessie : That was David Einhorn. He was Jackson Lab’s legal counsel at the time.

David : So they had the entire rights to commercialize once the patent was granted. I think a decision that Harvard regretted later on when it became very controversial and when other academic researchers weren’t able to obtain the mice. Because Harvard had given up all the rights.

Jessie : If OncoMouse had been created a decade earlier Phil Leder and Harvard would have gained the typical prestige from making such an innovative research tool. But they wouldn’t have gotten a patent for it. Two things had to happen in 1980 for that patent to be possible in 1988. The first was Diamond v. Chakrabarty, the Supreme Court case that opened the door to patenting life. The second was a new law that redefined ownership of scientific innovations.

Beth : The Bayh-Dole Act was legislation that was passed in 1980 in order to make it easier for universities to patent the results of their research. There were certain kinds of inventions that were hard to actually get into use without a patent. So if you had some kind of discovery, for example, something like an early stage drug but you couldn’t patent it, it would take a lot of additional investment in order to get it to the point where it could reach the market. And so, there were at least some things that were not reaching the market because there was no incentive for anybody to invest in actually producing them. So the Bayh-Dole act gave them the right by default to patent anything that was funded by the federal government. And then to license it and to keep the revenue from that.

Jessie : Computers, semiconductors, GPS, and Google were all products developed by research that was funded by federal dollars…taxpayer dollars.

Jessie : Harvard received substantial amounts of federal funding from places like the National Institutes of Science, the National Cancer Institute, and the National Science Foundation. And before the Bayh-Dole Act any product or innovation that came from government-supported research de facto belong to the federal government. But many people, including Phil Leder, found that problematic.

Leder : It used to be the policy at the National Institutes of Health—to pursue patents… to obtain patents on their discoveries and then to dedicate the patent to the public domain. Now that sounds terrific. What could be better than that? The government has achieved a patent and it has then dedicated to the public domain that anybody can use. The problem with that is as I’ve encountered it was that nobody was interested in unprotected patents. Nobody was ready to make the investment in the utilization of that technology. You know I wish the world were different, but that’s what it seems to be like.

Jessie : A little bit of background here: the government really started to pour money into funding university science during World War II. After the war, agencies like the National Science Foundation and the National Institute of Health were created to oversee the distribution of funds to academic researchers. The 1950s and 1960s were the golden age of American science. But the economic malaise of the 1970s did not spare the science community. There was a feeling in the air that the pot of government funding seemed to be getting smaller.

Beth : In the late 70s things were kind of flat. It wasn’t really decreasing but the flatness had come after this really long period of growth so it was perceived as a decrease even though it wasn’t. So there was this sense that things were getting tighter but it was mostly about the big post war boom ending. And another big piece was by the late 70s you have the biotech revolution starting to take off and that really creates all these new opportunities for money.

WGBH documentary: Scientists and the public are trying to come to terms with a dramatic new technique. A technique that gives scientists unprecedented power to manipulate nature.

Jessie : It was called recombinant DNA. It was developed in the 1970s when scientists figured out how to identify and target specific genes or sections of DNA and introduce that genetic material into the DNA of an organism from another species. This 1977 documentary from Boston broadcaster WGBH describes the outcome of an experiment using recombinant DNA technology on bacteria.

WGBH documentary: The few bacteria that have swallowed recombinant plasmids now contain the transplanted genes from a different microbe. They can be picked out and allowed to multiply. With this experiment man this barrier that prevents different species from exchanging genes.

Jessie : For mouse geneticists this meant that instead of relying on the imprecise nature of selected inbreeding or spontaneous mutation they can now target the human gene they wanted to study. Introduce it into the genetic line of a mouse, and create a transgenic mouse that was a better more reliable model of human decease. It was a huge leap forward in understanding and treating any disease with a genetic basis, like cancer, Alzheimer’s, mental illness, or cystic fibrosis. Diamond v. Chakrabarty was the watershed moment for biotech. Entrepreneurs and venture capitalist flocked the biotech industry and that’s where Bayh-Dole fits in.

Beth: So there’s definitely a big financial incentive in it for universities as well by the end… And then the other thing that’s going on the policy side is that policymakers are really interested by the late 70s in trying to make American business more competitive and so they’re really interested in promoting technological innovation.

Jessie : And then it’s 1980:

Reagan montage: no problem that we face today can compare with the need to restore the health of the American economy and the strength of the American dollar;

Jessie : The Reagan years were the era of bootstraps capitalism. The free market was king. Government involvement was rarely the answer.

Reagan montage continued: The people have not created this disaster in our economy; the federal government has. It has overspent, overestimated and overregulated. It has failed …

Jessie : Bayh-Dole opened the door to money making opportunities. Scientists didn’t take much notice of it, but university administrators definitely did. The agreement between Harvard and DuPont required Phil Leder to disclose any products discovered in his research.

Leder : As an obedient employee of Harvard Medical School, Tim and I, reported this invention to our office of technology transfer licensing as an invention and a discovery which we disclosed to them and which potentially would be patentable. And they consulted with a patent attorney and the matter was patented.

Jessie : Dr. Leder is referring to Harvard’s Office of Technology Transfer. A tech transfer office is responsible for commercializing the products that come from a university’s research labs. It’s their job to pursue patents for products or technologies and license them to private companies. They started appearing on campuses of elite schools like Stanford, Harvard, and MIT in the early 1980’s after Bayh- Dole made it possible for them to patent, and now any research university worth its salt has one.

David : if you were a tech transfer office, you’re a tech transfer person – many of whom come from industry. Their success is to increase as much as possible the return from the inventions and the institution and their success is not a value within the basis of the sharing ethos, but in terms of how much money they’re making for the institution. That’s the problem.

Harold : And that’s where things really got hot. When Harvard sought a patent and then license it to DuPont and DuPont tried to extract licensing agreements and payments for licenses that were unprecedented and thought by most of us to be inappropriate.

Alexis: Chapter Five. DuPont vs. The Mouse Club.

Alexis : Scientific breakthroughs don’t usually happen all at once. And they’re rarely the result of just one person’s ideas. They overwhelmingly come from building off of other people’s research, using other people’s techniques and tools. And OncoMouse was no different.

Lisa: Phil Leder wasn’t the only person making a transgenic mouse for cancer research in the 1980s. In fact, two other geneticists, Ralph Brinster and Richard Palmiter had made their own “OncoMouse” two years before Leder made his. They even published an article about it a few months before Leder published. But then Leder took the next, unconventional step of patenting it.

Harold : Those were mice that got cancer as a result of genetic manipulation. These mice came originally from Brinster and Palmiter and other people, nobody was talking about providing intellectual protection against the the use of those mice by others.

Jessie: OncoMouse was out in the mouse genetics world in 1984, four years before the 1988 patent. And during that time—as was customary in the Mouse Club—scientists began using the animal in their work. They would get the mice from colleagues who were already using them or from central repositories like JAXS. Then in 1988 OncoMouse gets the patent. And DuPont, which was used to operating in the world of commercial science, where every product had a price, came calling to collect their fee. Scientists used to getting transgenic mice for free were suddenly informed that these mice would cost about fifty dollars each, which was ten times the price of getting it from Jackson Labs, and fifty times the price of getting it for free from a colleague. Which brings me to another stipulation that enraged the Mouse Club. The license barred anyone using OncoMouse from sharing it with colleagues.

Ken : In the research community as a whole there was outrage because this totally violated their concept of the freedom of research. And so there was a sense of violation a double violation on the part of the research community: one their own ethical standards of conduct. How one is supposed to proceed and the other one was on research on what was at that time, it still is, a major killer in the United States.

Harold : One of the things that I found particularly offensive about the way the patent was written is that it covered all mice that were genetically manipulated

Ken : I don’t know how familiar you are with patent applications but they have a kind of nested structure in which the application starts out with the broadest claim possible and then it narrows down and successive claims in the hope that the patent office will offer the broadest possible. And the original patent application was written in a way that really covered any possible genetic engineering in a mouse that could result in increased incidence of cancer. It was an incredibly broad application of the patents and they just stretched our scientific imagination.

Jessie : So scientists who had made their own oncomice, lower case “o,” were not allowed to use them. Even those who had made them before Phil Leder. The science world was on alert. DuPont was coming for your oncomice, no matter who made them or how. DuPont had trademarked the name. So it could only be used for the Harvard mouse. In 1992 the Mouse Club decided to take a stand.

Lisa: Chapter Six. The Mouse Club Fights Back.

Jessie : The tensions between DuPont and the mouse club came to a boiling point at the 1992 Mouse Molecular Genetics Conference at Cold Spring Harbor lab in New York. Coincidentally the same lab where CC Little organized the Mouse Club of America decades before.

Ken : Actually, I have a strong visual image of the conference hall. It was one of these bank audiences where people are in successive rows elevating up the auditorium. And I was down in front.

Jessie : After a regularly scheduled session a room full of angry mouse geneticists gathered, lead by Harold Varmus. They were riled up over DuPont’s licensing requirements for OncoMouse.

David : And there was not a very friendly audience of scientists but I do remember very clearly Ken Paigin got up and said don’t worry about it. Send your mice to us and we’ll distribute them as we always have and got a big cheer.

Ken : I just got up and did it. That’s all I can tell you about it except that it was an intense thing at the moment and it turned out to be history.

David : Now we’re in the target of DuPont with respect to distribution. I mean we were the spigot we were the ones who had the mice and distributing the mice. So from DuPont’s perspective getting us to agree would solve their concerns about licensing rather than have to deal with individual academic institutions.

Jessie : It didn’t take long for the cease-and-desist to arrive.

David : DuPont said that we were infringing and they wanted a license from us. “If you’re going to distribute these mice you have to do it under our terms” and the terms that we found onerous and inconsistent…were this claim to review publications.

Jessie : There were disclosure agreements that required researchers to annually update DuPont on what they’d been working on, which scientists were reluctant to do if it hadn’t been published yet, lest their ideas be co-opted by someone else.

David : And secondly they were claiming reach-through rights.

Jessie : Basically, DuPont would be able to claim ownership or royalties on any future products that came from the research.

David : We had to make a decision about whether we were going to subject themselves to a lawsuit.

Ken : I think the thing that was very much in our minds at that point in time was that if we acceded to DuPont claims that we’d be shutting off a considerable amount of Cancer research at basic research institutions. And there was some question about our vulnerability. And David was considerably more concerned about the legal aspects of this. My feeling was that it would be a public relations disaster for DuPont to end up suing the small basic research laboratory in Maine. So we went ahead.

David : Well as a lawyer I didn’t have the same confidence and assurance as Ken did. And I felt that there was a possibility that we could be sued. And so although I appreciate Ken’s thoughts and that he might be correct. I thought it was important that we not take that chance and at least try to get the help of NIH before we went any further. So that’s when we reached out to the NIH.

Jessie : By 1993 Harold Varmus had become the head of the National Institutes of Health.

Harold : I personally had taken the position that… If there were certain mice we weren’t going to get without signing a license I just would use them. And I was encouraging people to basically be lawbreakers because the patent was issued in my view was much too broad.

Jessie : The NIH and DuPont entered into negotiations which would stretch on for several years. In the meantime, many scientists openly rebelled against DuPont’s restraints, with help from Jackson Lab.

David : It was easy to ignore that with impunity as long as we continue to send the mice which we continue to do we never stop distributing the mice.

Ken : We also didn’t stop receiving mice because all during this time this was a very fertile research field and that was a period of researching cancer where genes that cause susceptibility to cancer or play an essential role in particular cancers were being identified at a rapid rate.

Harold : In 1998 after a couple of years of negotiation the NIH had reached an agreement with DuPont that allowed a more sensible approach sensible relationship between DuPont and academic investigators.

Jessie : The NIH and DuPont signed a Memorandum of Understanding that allowed researchers with federal funding free use of the mice, as long as they weren’t commercializing their work.

Harold : The patent had been issued and there’s not much you can do about that. What we’re trying to do in that situation is to find a working relationship that’s acceptable. But there clearly there were inequities that became apparent later on and resulted in further conflict…

Jessie : If the story had ended with the MOU it would be a reasonable and somewhat satisfying conclusion. But it didn’t. Around 2002, disputes arose again between several top tier universities and DuPont. Both MIT and the University of California accused DuPont of overreaching their agreement with the NIH. Harvard tried to extend their patents in 2012, but the courts determined that they had expired in 2005. Not that it mattered. By this point new technologies had made OncoMouse obsolete anyway. But the changes the mouse had brought to the scientific community were permanent. In the years since, universities have become much more focused on the pursuit of profits and tech transfer offices have assumed a mantle of power.

Beth : The joke that people make about Harvard sometimes is that it’s a hedge fund with a research university attached to it. Afraid that that big universities that are talking about big amounts of money are very clearly self-interested economic actors.

David : I think the legacy. I think in the beginning I mentioned how Harvard regrets today having handled DuPont the way it did. Giving them all these rights without protecting the research community. I think the lesson for the commercial companies was that Harvard became the poster boy for not how to handle a patent on a research tool. They got lots of bad publicity. They made very little money on the patent. And I think that was a message for the commercial world not to overreach.

Jessie : And there’s another ironic conclusion to this story. And that is that after OncoMouse academic scientists started to increasingly seek out patents. At a scientific conference several years after the OncoMouse debacle a group of researchers was complaining about patents hindering their work. But then someone asked how many of them held patents. About half of the group stood up.

Jessie : The culture had already changed. To many mouse geneticists welcomed patents as a necessary evil. If they wanted to get their products out into the world and help people they’d have to use them. For Distillations , I’m Jessie Wright-Mendoza.

Lisa : I think on a level we can all understand the choices that mouse club made. There’s something on the surface that seems kind hypocritical. But it’s more complicated than that. Most of us have to negotiate our values with reality.

Alexis: Yeah. I mean it’s clear that OncoMouse changed the dynamic. It brought about a new set of rules and mouse club had to get with those rules or get out.

Lisa: If there is one thing the story tells us is that research paradigms can change. Maybe they will change again in the future. Maybe there will be a move away from patenting – for Jonas Salk’s humanitarian reasons or for some other reasons that we can’t yet imagine.

Alexis: And to be fair: there are also positive things about patents. They designate clear owners and boundaries, they give people credit where in publishing it can get really fuzzy. And looking at biotech specifically, for now, this is how products get out into the world.

Alexis : Distillations is more than a podcast. We’re also a multimedia magazine.

Lisa: You can find our videos, our blog, and our print stories at Distillations DOT org.

Alexis : And you can also follow the Science History Institute on Facebook, Twitter, and Instagram.

Lisa : This episode was reported by Jessie Wright-Mendoza.

Alexis : And it was produced by Mariel Carr and Rigo Hernandez.

Lisa : With additional audio production by Dan Drago.

Alexis : We want to give a special shout-out to MIT sociologist Fiona Murray. To produce this episode we relied heavily on her 2006 article, “The Oncomouse that Roared: Resistance and Accommodation to Patenting in Academic Science”.

Lisa : There’s a lot of research that goes into each episode of Distillations , and we keep a list of everything we read on our website, so check it out for further reading!

Alexis : For Distillations , I’m Alexis Pedrick.

Alexis : And I’m Lisa Berry Drago.

Lisa and Alexis : Thanks for listening.

Listen to more episodes

Collage of images related to ALS and health advocates

ALS Patients Take on the FDA

Research on the deadly disease progressed dramatically after the 2014 Ice Bucket Challenge, but a huge blow came when the most recent treatment was taken off the market.

The Fraud That Transformed Psychiatry

Psychology professor David Rosenhan made waves with his “On Being Sane in Insane Places” study, but decades later its legitimacy was questioned.

Collage of images related to cancer research

Cancer Virus Hunters: An Interview with Gregory J. Morgan

Seen as outcasts, some persistent scientists went against the grain to study viruses they suspected caused cancer.

Never Miss an Episode

Copy the above HTML to republish this content. We have formatted the material to follow our guidelines, which include our credit requirements. Please review our full list of guidelines for more information. By republishing this content, you agree to our republication requirements.

Loading metrics

Open Access

Peer-reviewed

Research Article

Mouse screen reveals multiple new genes underlying mouse and human hearing loss

Roles Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

Affiliations Wellcome Trust Sanger Institute, Hinxton, United Kingdom, Wolfson Centre for Age-Related Diseases, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom

Roles Data curation, Formal analysis, Investigation, Supervision, Validation, Visualization, Writing – review & editing

Affiliation Wellcome Trust Sanger Institute, Hinxton, United Kingdom

Roles Data curation, Investigation, Writing – review & editing

Roles Formal analysis, Investigation, Visualization, Writing – review & editing

Affiliation Wolfson Centre for Age-Related Diseases, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom

Roles Formal analysis, Investigation, Supervision, Validation, Writing – review & editing

Roles Data curation, Formal analysis, Investigation, Supervision, Visualization, Writing – review & editing

Roles Formal analysis, Investigation, Supervision, Visualization, Writing – review & editing

Roles Investigation, Visualization, Writing – review & editing

Roles Investigation, Writing – review & editing

Roles Investigation, Resources, Writing – review & editing

Affiliation Department of Emerging Genetics Medicine, Ambry Genetics, Aliso Viejo, California, United States of America

Affiliation Mid-Atlantic Permanente Medical Group, Rockville, Maryland, United States of America

Roles Data curation, Formal analysis, Investigation, Project administration, Supervision, Visualization, Writing – review & editing

Roles Data curation, Formal analysis, Investigation, Visualization, Writing – review & editing

Affiliation UCL Ear Institute, University College London, London, United Kingdom

Roles Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing – review & editing

[ ... ],

Roles Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

* E-mail: [email protected]

[ view all ]
[ view less ]
Neil J. Ingham,
Selina A. Pearson,
Valerie E. Vancollie,
Victoria Rook,
Morag A. Lewis,
Jing Chen,
Annalisa Buniello,
Elisa Martelletti,
Lorenzo Preite,

Published: April 11, 2019
https://doi.org/10.1371/journal.pbio.3000194
Reader Comments

Adult-onset hearing loss is very common, but we know little about the underlying molecular pathogenesis impeding the development of therapies. We took a genetic approach to identify new molecules involved in hearing loss by screening a large cohort of newly generated mouse mutants using a sensitive electrophysiological test, the auditory brainstem response (ABR). We review here the findings from this screen. Thirty-eight unexpected genes associated with raised thresholds were detected from our unbiased sample of 1,211 genes tested, suggesting extreme genetic heterogeneity. A wide range of auditory pathophysiologies was found, and some mutant lines showed normal development followed by deterioration of responses, revealing new molecular pathways involved in progressive hearing loss. Several of the genes were associated with the range of hearing thresholds in the human population and one, SPNS2 , was involved in childhood deafness. The new pathways required for maintenance of hearing discovered by this screen present new therapeutic opportunities.

Author summary

Progressive hearing loss with age is extremely common in the population, leading to difficulties in understanding speech, increased social isolation, and associated depression. We know it has a significant heritability, but so far we know very little about the molecular pathways leading to hearing loss, hampering the development of treatments. Here, we describe a large-scale screen of 1,211 new targeted mouse mutant lines, resulting in the identification of 38 genes underlying hearing loss that were not previously suspected of involvement in hearing. Some of these genes reveal molecular pathways that may be useful targets for drug development. Our further analysis of the genes identified and the varied pathological mechanisms within the ear resulting from the mutations suggests that hearing loss is an extremely heterogeneous disorder and may have as many as 1,000 genes involved.

Citation: Ingham NJ, Pearson SA, Vancollie VE, Rook V, Lewis MA, Chen J, et al. (2019) Mouse screen reveals multiple new genes underlying mouse and human hearing loss. PLoS Biol 17(4): e3000194. https://doi.org/10.1371/journal.pbio.3000194

Academic Editor: Thomas C. Freeman, University of Edinburgh, UNITED KINGDOM

Received: November 6, 2018; Accepted: March 7, 2019; Published: April 11, 2019

Copyright: © 2019 Ingham et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All mutant mice reported here are available via public mouse repositories for further investigation. All plotted data points are presented in the data files in the Supporting Information. The unprocessed ABR data are available at Dryad ( http://dx.doi.org/10.5061/dryad.cv803rv ).

Funding: This work was supported by The Wellcome Trust (KPS, 098051; 100699; 089622), the Medical Research Council (KPS, MC_qA137918; G0300212), European Commission (KPS, EUMODIC contract LSHG-CT-2006-037188), Action on Hearing Loss (KPS), the Haigh Fellowship in age related deafness (SJD), and Deafness Research UK (SJD). This work made use of data and samples generated by the 1958 Birth Cohort ( http://www2.le.ac.uk/projects/birthcohort , http://www.bristol.ac.uk/alspac/ , http://www.cls.ioe.ac.uk/ncds , http://www.esds.ac.uk/findingData/ncds.asp ) under grant G0000934 from the Medical Research Council and grant 068545/Z/02 from the Wellcome Trust. Genotyping was undertaken as part of the Wellcome Trust Case-Control Consortium (WTCCC) under Wellcome Trust award 076113, and a full list of the investigators who contributed to the generation of the data is available at www.wtccc.org.uk . The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: I have read the journal's policy and the authors of this manuscript have the following competing interests: ZP is employed by Ambry Genetics, and exome sequencing is among the commercially available tests. The other authors have declared that no competing interests exist.

Abbreviations: ABR, auditory brainstem response; DPOAE, distortion product otoacoustic emission; EP, endocochlear potential; ER, endoplasmic reticulum; ES, embryonic stem; f, frequency; FRT, flippase recombinase target; GO, Gene Ontology; IEA, Inferred from Electronic Annotation; IHC, inner hair cell; IOF, input-output function; LacZ, gene encoding β-galactosidase; LoxP, locus of crossover in P1 bacteriophage; MGI, Mouse Genome Informatics; MGP, Mouse Genetics Project; OHC, outer hair cell; OTOTO, osmium tetroxide-thiocarbohydrazide; P, postnatal day; qRT-PCR, quantitative real-time PCR; RT, room temperature; SL, sensation level; SPL, sound pressure level; S1P, sphingosine-1-phosphate; tm, targeted mutation

Introduction

Hearing loss is a very common disorder with a significant social impact, including delayed speech and language development, reduced academic achievement, increased social isolation, and risk of depression, and has recently been reported to be a major risk factor for dementia [ 1 ], adding new impetus to the need to develop therapies. Approximately 1 in 850 children are born with permanent hearing impairment in the United Kingdom [ 2 ], and the number of people affected by adult-onset hearing loss increases with each decade of life, with 60% of people in their 70s having a hearing loss of 25 dB or worse [ 3 ]. Environmental factors including noise or drug exposure play an important role in its etiology, but there is also a strong genetic contribution. Over 360 genes are known to be involved in human or mouse deafness, but ascertainment bias has led to many of these having early developmental effects, and little is known about the genetic contribution to adult-onset hearing loss. We set out to identify further genes underlying deafness, including those with mild effects, using a physiological screen based on the auditory brainstem response (ABR) in a large cohort of newly generated targeted mouse mutants. In this report, we review the findings from this screen and present the new data; both mice and ABR waveform data are available for further analysis. From the unbiased sample of 1,211 genes tested, we found 38 unexpected genes to be involved in hearing impairment. This indicates that around 600 additional genes remain to be found (see later), making deafness an extremely heterogeneous condition, with around 1,000 genes that may contribute. The observed impairments ranged from mild to profound, including several with progressive hearing loss, and with a wide range of underlying pathological mechanisms. The 38 genes represent a range of functions from transcription factors and a microRNA to enzymes involved in lipid metabolism. Eleven were found to be significantly associated with auditory function in the human population, and one gene, SPNS2 , was associated with childhood deafness, emphasising the value of the mouse for identifying genes and mechanisms underlying complex processes like hearing.

Results and discussion

Genes involved in auditory sensitivity.

We used a rapid (15-minute), noninvasive electrophysiological test, the ABR [ 4 ] ( S1 Fig ; S1 Data ) in anaesthetised mice aged 14 weeks old as part of an extensive pipeline of phenotyping tests on a set of new mouse mutants generated from targeted embryonic stem (ES) cells [ 5 , 6 , 7 ]. The allele design was mostly the knockout first, conditional-ready ( tm1a ; targeted mutation, first allele with design type a [ 6 ]) allele, which reduced or eliminated expression of the targeted gene by inclusion of a large cassette designed to interfere with transcription, but a few were the derived tm1b allele with an exon deleted or were edited alleles ( S1 and S2 Tables). A total of 1,211 genes were tested. Of these, 38 genes with no prior association with deafness had raised thresholds for detecting a response to sounds ( Fig 1 ; a small number of these have been published recently, after their discovery in the screen). Using objective criteria (see Materials and methods) we classified these 38 genes into five main groups based on thresholds: 5 showed severe or profound deafness, 10 had raised thresholds at high frequencies only, 2 showed raised thresholds at low frequencies only, 7 had moderately raised thresholds across frequencies, and 14 had a mild hearing impairment ( Fig 1 ). In addition to these 38 unexpected genes, 10 known deaf mutant lines were tested as positive controls ( S2 Fig A-J; S2 Data ), and 9 genes known to be involved in mouse ( Srrm4 ) or human deafness ( MYO7A , MYO15 , USH1C , WHRN , ILDR1 , ESPN , CEP250 , CLPP ) were tested in newly generated alleles; all showed raised thresholds ( S1 Table ; S2K–S2T Fig ; S2 Data ).

PPT PowerPoint slide
PNG larger image
TIFF original image

New mutant mouse lines with hearing impairment categorised as Severe–Profound (A), High frequency (B), Low frequency (C), Moderate (D), and Mild (E) at 14 weeks old. Mean ABR thresholds (±SD) are plotted for broadband clicks (Ck) and 6 to 30 kHz pure tone stimuli. On each panel, the green band denotes the 95% reference range for a large population of control wild-type mice derived from littermates of the mutants generated and tested. Green lines and triangles represent the mean thresholds (±SD) for control mice recorded in the same week as the mutants. Red circles and lines represent the mean thresholds (±SD) for mutant mice. Thresholds for individual mutants are shown by open grey circles and lines. Gene symbols are given on each plot, and when both the tm1a and tm1b alleles were screened, both sets of data are presented, indicated by (a) and (b) suffixes. Mice screened were homozygous mutants except for Brd2 , Srsf7 , and Setd5 , which were screened as heterozygotes due to reduced viability of homozygotes. When no response was detected up to the maximum dB SPL used (95 dB SPL), the maximum sound level used was plotted. Only 12 of these mutant lines (35%) would have been identified as having a hearing defect had startle responses alone been used to screen ( S1 Table ). The number of mutant mice screened of each line is given in column E of S1 Table and on each panel. A few of these lines ( Spns2 , Wbp2 , Lrig1 , Ocm , Mcph1 , Pax9 , or Slc25a21 ) have been characterised and published, and others were listed in a summary report ( Zcchc14 , Adgrb1 , Tram2 , Klc2 , Acsl4 , Gpr152 , Klhl18 , Zfp719 , A730017C20Rik/Minar2 , and Duoxa2 ) [ 7 ] because they were first detected in the current screen, but are included in the group analysis here, as they were not known to be involved in auditory function before the screen. Asterisks indicate lines that were called using the reference range or 20 dB rules described in the Materials and methods but were not significant calls using the Fisher Exact test; ‡ indicates lines that were not significant by the Fisher Exact test but did show raised thresholds when other cohorts were tested later. Plotted data points are given S5 Data . ABR, auditory brainstem response; Ck, click; SPL, sound pressure level.

https://doi.org/10.1371/journal.pbio.3000194.g001

Of the set of 1,211 genes, 3.14% were new associations with raised auditory thresholds. As the genes targeted were an unbiased set showing no significant enrichment for any functional class compared with the total set of mouse genes (see Materials and methods), we can extrapolate to estimate that over 600 further genes required for normal auditory thresholds remain to be found. Added to the 362 human and mouse genes already known and 38 reported here, this indicates that there may be as many as 1,000 genes involved in deafness, a very high level of genetic heterogeneity.

There were several targeted genes screened that we expected to show raised thresholds because they had previously been reported to underlie deafness in either humans ( GSDME/DFNA5 , MYH14 , MYH9 , PNPT1 , PRPS1 , CHD7) or mice ( Barhl1 , Fzd6 , Hmx3 , Nfkb1 , Sgms1 , Sms , Synj2) . However, the new alleles had normal ABR thresholds ( S3 Table ). The lack of raised thresholds could be due to incomplete knockdown of targeted gene expression in the tm1a allele (e.g., raised thresholds of Selk mutants were seen only in the tm1b allele, not in tm1a ); onset of hearing loss after 14 weeks, when we carried out the screening; screening was carried out on heterozygotes due to reduced homozygote viability; the genetic background may have influenced phenotype expression (e.g., Chd7 , where the new allele was not viable on the C57BL/6N background); or the original deafness may have been the result of a specific effect of the mutation on the protein rather than a consequence of reduced expression (e.g., GSDME ) ( S3 Table ). Alternatively, the original allele might have led to deafness via a long-range cis effect on a nearby gene, as in the Slc25a21 tm1a(KOMP)Wtsi targeted mutation, which causes deafness by reducing expression of Pax9 [ 8 ]. Thus, we probably missed additional genes involved in hearing loss, so our calculation of 600 more genes awaiting association with deafness may be an underestimate.

All mutant mice reported here are available via public mouse repositories for further investigation to explore these alternative explanations, and the unprocessed ABR data are available at the Dryad repository: http://dx.doi.org/10.5061/dryad.cv803rv [ 9 ]. Of note, all of the mutant lines that we have studied further following the initial screening results have shown raised ABR thresholds, even those in the mild class, suggesting that the screen has produced robust, reproducible calls (mutant lines studied further: Spns2 , Zfp719 , Ocm , Klhl18 , Wbp2 , Pex3 , Acsl4 , Gpr152 , Mcph1 , Slc25a21/Pax9 , Ywhae , Lrig1 , Klc2 , Usp42 , Srsf7) [ 8 , 10 , 11 , 12 ].

Role in human auditory thresholds

As mouse and human inner ears are very similar in structure and function (e.g., [ 13 ]), the newly identified mouse genes represent good candidates for involvement in human deafness. A child from a United States clinic detected through clinical whole exome sequencing inherited a frameshift mutation of SPNS2 from her father (c.1066_1067delCCinsT: p.Pro356Cysfs*35; CADD phred score 26) and an in-frame deletion of a serine codon in SPNS2 from her mother (c.955_957delTCC: p.Ser319del; CADD phred score 20.9). Neither variant has been reported in the gnomAD database (Genome Aggregation Database; accessed January 2019). Visual reinforcement audiometry at two years old revealed moderate to moderately severe hearing loss between 250 Hz and 4 kHz with no response at 8 kHz in the right ear, and severe hearing loss sloping to profound deafness from 500 Hz to 8 kHz with no response at 4 and 8 kHz in the left ear. Bone conduction testing indicated moderate-severe hearing loss at 2 kHz in the right ear, suggesting a sensorineural (not conductive) impairment, and acoustic reflexes were absent. However, the child had surprisingly good sound localisation performance. The severe level of hearing impairment associated with predicted damaging SPNS2 variants is similar to our findings in the mouse Spns2 mutant ( Fig 1 in [ 11 ]). Furthermore, we have previously reported that deaf children from two families in a Chinese cohort carried recessive WBP2 mutations [ 10 ].

We asked if these new candidates had any role in hearing ability in the general population by a candidate gene association analysis. We tested genomic markers within 0.1 Mb up- and downstream of each gene for association with auditory thresholds measured at age 44–45 in 6,099 individuals born during one week in the UK 1958 British Birth Cohort, using genetic data imputed to the 1,000 Genomes dataset [ 14 ]. Eleven of the thirty-seven candidate genes tested (including SPNS2 ) showed a significant association of markers with threshold at either 1 or 4 kHz or both frequencies ( Table 1 ), indicating that these 11 genes may play a role in normal variation of hearing ability in the human population.

The p -values above the significance threshold of p ≤ 6.76 × 10 −4 are not shown, and 37 of the 38 candidate genes were tested; no human orthologue is known for Mir122 , so this was not included.

https://doi.org/10.1371/journal.pbio.3000194.t001

These findings emphasise the value of the mouse in resolving complex human diseases, because very few significant markers have been reported to be linked to hearing through genome-wide association studies of adult-onset hearing loss directly in humans: GRM7 [ 15 ]; PCDH20 and SLC28A3 [ 16 ]; ISG20 or ACAN and TRIOBP [ 17 ].

Several of the new genes that we found to be involved in deafness in the mouse had human orthologues close to or within unidentified non-syndromic deafness loci ( S4 Table , column J) and so are good candidates for further exploration. These were USP42 within the DFNB90 interval, BRD2 close to the DFNA21 and DFNA31 loci, CAMSAP3 close to the DFNA57 region, and MCPH1 very close to the DFNM2 marker reported.

Broad range of gene functions

The 38 genes newly associated with hearing are involved in a broad range of functions, including transcriptional and translational regulation, chromatin modification, splicing factors, cytoskeletal proteins, membrane trafficking, calcium buffering, peroxisome biosynthesis, thyroid hormone generation, ubiquitination and deubiquitination, kinases, signaling molecules (including Wnt signaling), and proteins with no known or predicted function ( S4 Table ). A microRNA gene, Mir122 , was one of the new genes underlying hearing impairment. Seven of the gene products have a role in lipid metabolism: Fads3 is a fatty acid desaturase; Agap1 and Zcchc14 bind phospholipids; Klc2 transports phosphatidylinositol 3-kinase, which is required for phospholipid processing; Pex3 is involved in biosynthesis of peroxisomes, which are involved in lipid processing; Acsl4 is a long-chain fatty acid coenzyme A ligase converting free long-chain fatty acids into fatty acyl-CoA esters; and Spns2 is a transporter of sphingosine-1-phosphate, a key intermediate in sphingolipid metabolism with a role in signalling.

We carried out a GOSlim (high-level version of Gene Ontology) analysis to ask if the genes newly associated with hearing impairment ( n = 38) showed a similar distribution of gene ontology features to the group of genes previously known to be involved in deafness ( n = 362), and to compare them with all genes tested in this screen ( n = 1,211) and a group of genes associated with ABR waveform defects ( n = 27, described later). The novel genes showed nothing notably different to the full set of genes screened ( Fig 2 ). However, nearly 70% of previously known genes had a Gene Ontology (GO) annotation for developmental processes, which was much higher than in the other groups analysed, suggesting an ascertainment bias for deafness due to early developmental defects in human and mouse. This finding suggests that our unbiased screen for new genes involved in hearing impairment at all levels of severity has revealed a fundamentally different class of genes compared with previously known genes underlying deafness.

GO terms associated with each gene were compared between the five groups: genes previously known to be associated with deafness in human and mouse ( n = 362); novel genes associated with raised ABR thresholds in the current screen ( n = 38); genes with normal thresholds but abnormal waveform features ( n = 27); all genes screened in the MGP screen ( n = 1,211); and all genes listed in MGI ( n = 33,395). The proportion of genes in each group with the high-level GO terms listed are plotted for the three categories: Cellular component, Molecular function, and Biological process. Plotted data points are given in S6 Data . ABR, auditory brainstem response; ER, endoplasmic reticulum; GO, Gene Ontology; MGI, Mouse Genome Informatics; MGP, Mouse Genetics Project.

https://doi.org/10.1371/journal.pbio.3000194.g002

Some of the 38 genes had links to existing pathways involved in deafness. For example, Duoxa2 is required for maturation and transport from the endoplasmic reticulum (ER) to the plasma membrane of Duox2, also known to underlie deafness through its role in hypothyroidism, leading to retarded cochlear development and impaired hearing [ 18 ]. Spns2 is a sphingosine-1-phosphate (S1P) transporter, and our discovery of its involvement in deafness supports the role of the S1P signaling pathway in hearing loss, alongside reports of S1PR2 and Sgms1 mutations causing deafness [ 19 , 20 , 21 , 22 , 23 ]. In contrast, many of the other genes discovered in this screen, such as A730017C20Rik (Minar2) , have no demonstrated role in a biological process and no a priori reason to predict they might be involved in deafness.

Although some of the 38 new genes identified by this screen showed strong expression in cochlear hair cells ( S4 Table , column Q; https://gear.igs.umaryland.edu ), in general expression levels did not show a strong correlation with our ABR findings.

Progressive hearing loss

As our goal in carrying out the screen was to identify new genes involved in adult-onset hearing loss, we carried out recurrent ABR recordings (usually at 4, 8, and 14 weeks old, plus shortly after the normal onset of hearing at 2 and 3 weeks old if thresholds were raised at 4 weeks) on some of the mutant lines. Remarkably, several of the mutants we studied showed relatively normal early development of ABRs followed by progressive increase in thresholds. These included Srsf7 heterozygotes (encoding a splicing factor; Fig 3A–3C ), Gpr152 homozygotes (a G-protein–coupled receptor; Fig 3D–3F ), and Klc2 ( Fig 4A–4E ), plus Spns2 [ 11 ], Wbp2 [ 10 ], Acsl4 , Zfp719 , Ocm , and Klhl18 homozygotes.

Several new mutant lines showed normal or near-normal ABR thresholds at young ages followed by progressive increases in thresholds with age. A-C. Srsf7 heterozygous mutants. A. Four weeks old, wild-type littermates n = 9, Srsf7 heterozygotes n = 12. B. Eight weeks old, wild-type littermates n = 3, Srsf7 heterozygotes n = 7. C. Fourteen weeks old, wild-type littermates n = 10, Srsf7 heterozygotes n = 10. D-F. Gpr152 homozygous mutants. D. Four weeks old, wild-type littermates n = 5, Gpr152 homozygotes n = 5. E. Eight weeks old, wild-type littermates n = 5, Gpr152 homozygotes n = 5. F. Fourteen weeks old, wild-type littermates n = 5, Gpr152 homozygotes n = 5. In all panels, means with standard deviations are plotted in red for mutants and green for wild-type littermate controls. Arrows indicate that there was no response, so the maximum sound level used was plotted. Asterisks indicate significant differences between mutants and wild-type controls (Mann–Whitney U test, p < 0.05). Plotted data points are given in S7 Data . ABR, auditory brainstem response; Ck, click; SPL, sound pressure level.

https://doi.org/10.1371/journal.pbio.3000194.g003

ABR thresholds of Klc2 mutants are close to those of controls at 2 weeks old, and show significant, progressive increases in thresholds from one month onwards, mainly at low frequencies (3–18 kHz), increasing to affect higher frequencies by 6 months old (A-E) (wild types 2 weeks n = 4, 1 month n = 13, 2 months n = 13, 3 months n = 9, 6 months n = 15; Klc2 homozygotes 2 weeks n = 6, 1 month n = 13, 2 months n = 13, 3 months n = 13, 6 months n = 18; Mann–Whitney U test, at 2 weeks old, p = 0.12 overall but p = 0.019 at 3 kHz and p = 0.01 at 6 kHz when testing stimuli separately; p < 0.001 at each later age shown by asterisks). F. DPOAEs at 6 months old show mutant amplitudes (red, n = 3) at the noise floor (grey) across all frequencies tested; wild types (green, n = 3) show normal emission amplitudes. Gi-vi. Scanning electron microscopy revealed extensive loss of OHC hair bundles at P28 in the cochlear regions, corresponding to the worst thresholds in mutants (12 kHz; Gi. heterozygote n = 4, Gii. homozygote n = 11), while there was little sign of hair cell loss at higher-frequency regions (42 kHz; Giii. heterozygote, Giv. homozygote). Remaining hair bundles had a normal appearance (Gv. heterozygote, Gvi. mutant). Scale bars, g-j, 10 μm; k-l, 1 μm. Hi-iv. Confocal imaging at P28 showed that many OHC nuclei were missing in the most affected regions (12 kHz; Hi. heterozygote n = 4, Hii. homozygote n = 5), but most hair cell nuclei were present at less affected regions (30 kHz; Hiii. heterozygote, Hiv. homozygote). Blue, DAPI-labelled nuclei; red, CtBP2-labelled ribbons and IHC nuclei; green, neurofilament-labelled unmyelinated dendrites. Scale bars, 10 μm. Hv. Quantification of OHC nuclei from confocal images demonstrated significant reduction in mutants (red) at best-frequency regions from 6 to 24 kHz and no significant difference with controls (green) at the 30-kHz region. Black line represents ABR threshold elevation in mutants compared with littermate controls. I. EPs in wild types (green) and homozygotes (red) show no significant difference in mutants (homozygotes 113.8 ± 11.5 mV, n = 10; wild-type littermates 116.0 ± 6.5 mV, n = 10; t = 0.518, df = 14, two-tailed p -value = 0.613). Maximum negative potentials during anoxia are significantly reduced in homozygotes (lower part of plot) (homozygotes −10.0 ± 5.39 mV, n = 6; wild-type littermates −33.7 ± 4.8 mV, n = 6; t = −8.050, df = 10, two-tailed p -value = 0.0000111). Ji-vi. Confocal imaging of IHCs at P28 showed no obvious abnormalities of GluR2-labelled postsynaptic densities (green), but less extensive Kcnma1-labelled patches (red) of IHCs in mutants compared with controls at the frequency regions showing the worst thresholds (6 and 12 kHz) ( n = 7 homozygotes, 5 littermate controls). Scale bars, 5 μm. Jvii. Counts of green-labelled GluR2 puncta per IHC show no difference between mutants and wild types (6 kHz, homozygotes n = 7, wild types n = 5, t test, p = 0.7422; 12 kHz, homozygotes n = 7, wild types n = 5, t test, p = 0.0737; 30 kHz, homozygotes n = 5, wild types n = 6, t test, p = 0.7089). Ki. Representation of the allele, with exons in grey, FRT sites in green, loxP sites in red, and lacZ and neo components of the inserted construct labelled. Kii-iii. Expression of Klc2 in the cochlear duct of a heterozygote ( n = 3) using the LacZ reporter system in the allele. Blue-labelled areas show expression in cells surrounding the cochlear duct and spiral ganglion. Kiii shows a higher magnification of the organ of Corti. Scale bar on Kii, 100 μm, on Kiii, 50 μm. Li-ii. Confocal images of the organ of Corti in a wild type (left) and homozygote (right) at P28 labelled with Myo7a antibody (false-coloured green) showing hair cell bodies and DAPI (false-coloured red) showing nuclei. Small images at left and right show the areas marked in white boxes rotated through 90° around the radial cochlear axis to show a mid-modiolar view of the hair cells. Nuclei appear in similar locations in mutants and controls (base of OHCs, middle of IHCs). Scale bars, 5 μm. M. qRT-PCR of Klc2 mRNA from brain at P28 showed complete knockdown of transcript in homozygotes (red, n = 3); heterozygotes (blue, n = 4) showed around half of the wild-type level (green, n = 2). All plots are means ± standard deviation. Plotted data points are given in S8 Data . ABR, auditory brainstem response; DPOAE, distortion product otoacoustic emission; EP, endocochlear potential; FRT, flippase recombinase target; IHC, inner hair cell; lacZ , gene encoding β-galactosidase; loxP, locus of crossover in P1 bacteriophage; OHC, outer hair cell; P28, postnatal day 28; qRT-PCR, quantitative real-time PCR.

https://doi.org/10.1371/journal.pbio.3000194.g004

The finding of multiple new genes underlying progressive hearing loss and/or impairment of responses to high frequencies (Figs 1 and 3 ) is important because progressive hearing loss in humans is very common and often affects high frequencies first, yet we have few clues to the pathological molecular processes involved. The mouse alleles studied here are relatively severe in their effect on protein expression, but variants in the human population may have milder effects on protein function and lead to later onset of hearing loss. Importantly, the finding of genes involved in normal development but later deterioration of hearing identifies molecular pathways likely to underlie adult-onset progressive hearing loss in humans.

Varied pathophysiology underlying deafness

We analysed further a subset of mutant lines and revealed a wide range of pathological conditions underlying hearing impairment. Two examples of contrasting phenotypes are the Klc2 and the Ywhae mutant lines.

Klc2 mutants showed a progressive increase in ABR thresholds with age, mostly affecting low frequencies ( Fig 4A–4F ), with a sensorineural (not conductive) pathology. Klc2 encodes kinesin light chain 2, which, together with kinesin heavy chains (encoded by Kif5 ), forms the kinesin-1 motor complex, a microtubule-associated anterograde transporter. The allele design ( Fig 4Ki ) allowed us to use the gene encoding β-galactosidase ( LacZ ) as a reporter system, showing that Klc2 was expressed in the epithelial cells lining the cochlear duct and strongly in the spiral ganglion ( Fig 4Kii-iii ). The middle ear and gross structure of the inner ear appeared normal. The endocochlear potential (EP) was maintained at a normal level even up to 6 months of age, but the anoxia potential in scala media was significantly less negative in these mutants, consistent with loss of hair cell conductance ( Fig 4I ). At one month old, there was extensive loss of outer hair cell (OHC) hair bundles ( Fig 4Gi-vi ), DAPI-stained OHC nuclei, and CtBP2-labelled presynaptic ribbons of OHCs primarily in the region that normally responds best to 12 kHz (60%–70% of the distance along the cochlear duct from the base), corresponding to the worst ABR thresholds ( Fig 4Hi-v ). There were few signs of inner hair cell (IHC) degeneration, but the increase in threshold was larger than would be expected if only OHCs were affected, suggesting IHC dysfunction. Klc2 is involved in anterograde transport of PI3K, which mediates insertion of AMPA receptors at synaptic membranes [ 24 ] and GluR1/2-containing vesicles to axon terminals [ 25 , 26 ]. We found no abnormality in GluR2-labelled postsynaptic densities below mutant IHCs ( Fig 4Ji-vii ), suggesting other transport systems must move this AMPA receptor to the membrane. Klc2 also interacts with Kcnma1, the calcium-activated potassium channel (BK channel) that underlies the I K,f current required for very rapid responses of IHCs and contributes to protective efferent suppression of OHCs [ 27 , 28 , 29 , 30 ]. We found that labelling of Kcnma1 in IHCs was less extensive in mutants compared with wild-type controls (at 12kHz; Fig 4Ji-vi ), implicating Kcnma1 in the pathological mechanism; however, knockout of Kcnma1 leads to less severe loss of thresholds [ 31 ], so this alone cannot explain the extent of dysfunction in the Klc2 mutants. Finally, kinesin-1 has been implicated in maintenance of the hair cell nucleus in its correct position by interacting with Nesp4 in the outer nuclear membrane [ 32 , 33 ], and Nesp4 mutations lead to location of the OHC nucleus at the top of the cell and subsequent degeneration [ 34 ]. We did not find mislocalisation of OHC or IHC nuclei ( Fig 4Li-ii ), suggesting this was not the mechanism underlying hearing loss in the Klc2 mutants, and that redundancy between kinesin light chains may compensate for loss of Klc2 in nuclear localisation. No disease-associated loss-of-function mutations of human KLC2 have been reported yet to compare with the Klc2 mutant mice, which show a complete lack of Klc2 mRNA ( Fig 4M ). However, a human gain-of-function KLC2 mutation (216-bp deletion upstream of the coding region) leading to increased KLC2 expression causes spastic paraplegia, optic atrophy and neuropathy (SPOAN), a neurodegenerative disorder involving progressive axonal neuropathy [ 35 ].

In contrast, the Ywhae mutants showed increased thresholds across all frequencies associated with variable amounts of accumulated fluid and exudate containing inflammatory cells in the middle ear, suggesting predisposition to otitis media ( Fig 5A–5M ). The middle ear mucosa appeared thickened with granulation tissue in sections ( Fig 5L ), and scanning electron microscopy of the luminal surface showed an open Eustachian tube in mutants, but abundant clusters of goblet cells (presumed to produce mucus) with fewer ciliated epithelial cells, which would normally contribute to the clearing of excess mucus ( Fig 5M and 5N ). The variability in thresholds between individual Ywhae mutants, relatively flat increase across all frequencies, and near-normal ABR waveform support a conductive hearing loss ( Fig 5A–5H ). The surface of the organ of Corti looked normal ( Fig 5O and 5P ) but we cannot exclude a sensorineural component in some of the more severely affected mutants, possibly due to the impact of 129S5-derived alleles in the mixed genetic background, or an effect of persistent inflammation of the middle ear [ 36 ]. Ywhae, also known as 14-3-3ε, is a member of the highly conserved 14-3-3 phosphoserine/threonine binding family, which have many interacting partners and are thought to provide a scaffold allowing coordination of intracellular signaling [ 37 , 38 ]. Ywhae is normally widely and strongly expressed throughout the body [ 39 ] and within the cochlea [ 40 ], and the mutation led to an absence of detectable Ywhae protein in homozygotes ( Fig 5R ). Not surprisingly, Ywhae mutants showed a number of other abnormal phenotypes similar to those reported in another Ywhae mutant [ 41 ] (see S1 Table ), including reduced viability of heterozygotes on a C57BL/6N background (20 heterozygotes out of 192 offspring from heterozygote × wild-type matings at weaning, chi- squared p < 0.0001) and homozygotes on a mixed C57BL/6N and 129S5/SvEvBrd/Wtsi background (53 homozygotes at 2 weeks old from 685 offspring from heterozygous intercrosses, chi-squared p < 0.0001). Ywhae homozygotes also showed reduced growth ( Fig 5S ) and a shortened skull ( Fig 5T–5V ). Craniofacial malformations may affect Eustachian tube structure and function, leading to otitis media, but there are many other possible pathological mechanisms not yet explored. Similar fluid-filled middle ear and conductive hearing loss phenotypes were found in the Mcph1 mutant [ 12 ] and Slc25a21 mutants with reduced Pax9 expression [ 8 ].

ABR thresholds were significantly raised at 8 weeks old in mutants (A,E) (Mann–Whitney U test, p < 0.001) but were highly variable between individuals (grey lines), while click-evoked ABR waveforms were close to normal (B,F show group mean waveforms at 20 dB above threshold), and reduced but nonsignificant amplitude growth with increasing stimulus level in mutants (C,D,G,H) (Mann-Whitney-Wilcoxon test with Holm correction for multiple testing). Ywhae homozygotes (red) and wild-type littermates (green) with individual mutant thresholds (grey) on a 50% C57BL/6N, 50% 129S5 (A-D) ( n = 18 homozygotes, 12 wild types for thresholds; n = 10 homozygotes, 11 wild types for amplitude functions) or a 12.5% C57BL/6N, 87.5% 129S5 genetic background (E-H) ( n = 9 homozygotes, 13 wild types for thresholds; n = 6 homozygotes, 13 wild types for amplitude functions). Heterozygotes have normal ABR thresholds. (I-L) Coronal sections of the middle ear in wild type (I,J) and homozygous Ywhae mutants (K,L). The tympanic membrane is retracted in the mutant, and the middle ear contains inflammatory debris. The mucosa is thicker in the mutant (double-headed red arrow), containing granulation tissue (‘G’), infiltrated immune cells (black arrow in L), and foamy macrophages (red arrowhead). Black rectangles in I and K indicate the areas enlarged in J and L. Scale bar for I and K is 1 mm; in J and L it is 100 μm. n = 2 homozygotes, 2 heterozygotes, 4 wild types. M,N. Scanning electron micrographs of the middle ear epithelium near the opening of the Eustachian tube in wild-type (M) and homozygous Ywhae mutant (N), showing widespread goblet cells and fewer ciliated epithelial cells in the mutant compared with the wild type, which is rich in ciliated cells. Scale bar is 10 μm. n = 5 homozygotes, 5 heterozygotes, 3 wild types. O,P. Scanning electron micrographs of the organ of Corti in a heterozygote (O) and homozygous mutant (P) showing normal appearance. OHCs are shown at the top, each with a V-shaped stereocilia bundle, and IHCs at the bottom. Images taken from 60% of the distance along the cochlear duct from the base. n = 7 homozygotes, 5 heterozygotes, 2 wild types. Scale bar is 10 μm. Q. Design of the Ywhae tm1e(EUCOMM)Wtsi mutation with exons in grey, FRT sites in green, loxP sites in red, and lacZ and neo components of the inserted construct labelled. R. Western blot of Ywhae and Pafah1b1 (Lis1) protein in brain showing no detected Ywhae in homozygous mutants (−/−), while expression of Pafah1b1 (encoded by a nearby gene) was unaffected. Vinculin was used as a loading control. n = 2 homozygotes, 2 heterozygotes, 4 wild types. S. Body weight growth with age in males (left) and females (right), showing significantly reduced weights in homozygotes (red) compared with wild types (green) (males, n = 7 homozygotes, 6 wild types; females, n = 7 homozygotes, 7 wild types; mixed model framework test as described by Karp and colleagues [ 42 ] at 4 weeks, 16 weeks, and area under the curve, p = 6.9 × 10 −3 , 0.013, and 6.9 × 10 −3 respectively). T-V. X-rays of wild type (T) and homozygotes (U) showing shorter skull length (V) in the mutants ( p = 3.9 × 10 −5 ; n = 6 male and 5 female homozygotes and 5 male and 6 female wild types; mixed model framework test, males and females p = 3.9 × 10 −5 ). All plots are means ± standard deviation. Plotted data points are given in S9 Data . ABR, auditory brainstem response; Ck, click; EAC, external auditory canal; FRT, flippase recombinase target; IHC, inner hair cell; lacZ , gene encoding β-galactosidase; loxP, locus of crossover in P1 bacteriophage; M, malleus; MEC, middle ear cavity; OHC, outer hair cell; SL, sensation level; SPL, sound pressure level; TM, tympanic membrane.

https://doi.org/10.1371/journal.pbio.3000194.g005

A third distinct pathology found was a reduction in EP. Normally, a high resting potential in the cochlear endolymph is generated by the stria vascularis. This is necessary for normal sensory hair cell function. Progressive disorganisation of the stria vascularis accompanies the reduced EP in Spns2 mutants [ 11 ]. A fourth example is the Wbp2 mutant, in which abnormal structure of synapses between IHCs and cochlear neurons and swelling of nerve terminals leads to progressive increase in ABR thresholds [ 10 ].

The finding of a wide range of primary pathological processes in these mouse mutants as outlined above suggests that the pathogenesis of hearing loss in the human population may be equally heterogeneous. The limited information gleaned from human temporal bone studies supports the suggestion of heterogeneous pathophysiology underlying progressive hearing loss [ 43 ].

Other features associated with deafness in the new mutants

Many mouse mutants with deafness were originally detected, because they also had a balance defect leading to circling and/or head bobbing; thus, many of the earliest genes to be identified were those involved in early developmental problems such as gross inner ear malformations or sensory hair cell developmental abnormalities affecting both the cochlea and vestibular part of the inner ear. It is notable that none of the 38 new mutant genes we report here showed any sign of leading to a balance defect ( S1 Table ). Nine lines had reduced viability assessed at postnatal day 14, with three of these lines producing so few homozygotes that heterozygotes were passed through the phenotyping pipelines instead ( Brd2 , Srsf7 , and Setd5 ). Six lines had either male or female infertility ( Pex3 , Mkrn2 , Herc1 , Camsap3 , Mcph1 , Usp42 ), which is higher than the expected 5% based on larger panels of mutant alleles. Corneal or lens defects were observed in five lines ( Spns2 , Pex3 , Agap1 , Mcph1 , Usp42 ). Occurrence of anomalous features in other systems tested were generally scattered across mutant lines and phenotypes, with Duoxa2 and Ywhae mutants showing the largest number of other abnormalities ( S1 Table ).

Mutant lines with normal thresholds but abnormal waveforms

By analysis of click-evoked ABR waveforms, we identified 27 additional mutant lines with normal hearing sensitivity, but which had abnormal patterns of neural responses, such as smaller ABR wave amplitudes or prolonged latencies, determined using objective criteria ( S1 Table ; Fig 6 ; S3 Fig ; S3 Data ). Ten further mutant lines from the 38 with ABR threshold elevation also exhibited abnormal ABR waveform shapes ( S1 Table ; S4 Fig ; S4 Data ). The ABR waveform is a complex mixture of voltage changes reflecting the sum of excitatory and inhibitory activity at different times after stimulus onset and different physical locations within the brain relative to the position of the recording electrodes. Wave 1 reflects auditory nerve activity, and later waves reflect activity higher up the central auditory pathways. Some mutant lines had reduced wave 1 amplitudes ( Fig 6 ; S3 and S4 Fig ; S4 Data ; S1 Table ), which may result from desynchronisation of the onset of firing in auditory nerve fibres [ 44 ], or reduced numbers of auditory nerve fibres contributing to the ABR, or a selective loss of high spontaneous rate/low threshold neurons, which have maximal discharge rate at stimulus onset, or inefficient synaptic recovery during the short gap (23.5 ms) between stimuli. Other mutants showed abnormal amplitudes or latencies of later waves, suggesting auditory processing anomalies in the central auditory system ( Fig 6 ; S3 and S4 Figs; S3 and S4 Data). These changes could reflect abnormal inherent excitability of auditory neurons or an alteration in the balance of excitatory and inhibitory inputs onto these neurons, resulting in increased or decreased discharge or synchrony. Mutants showing changes in latency are most easily explained by changes in neural conduction speed, alterations in synaptic delays, or changes in the relative contributions of different components in this complex neuronal pathway. Finally, in the most extreme example ( Bai1 , also known as Adgrb1 ), the mice exhibited clear auditory-evoked responses and measurable thresholds, but the ABRs were so abnormal that it was not possible to determine the equivalent peaks to quantify and compare with control mice ( Fig 6 ). This set of 27 mutants with waveform anomalies will be an interesting group to analyse further, because central auditory function is critical for normal sound perception. Such deficits may translate in humans to altered performance in sound localisation, ability to follow salient acoustic stimuli in background noise, discrimination of specific speech features, and other auditory processing disorders (e.g., [ 45 ]).

Examples of three mutant mouse lines with altered ABR waveforms are shown. ABR thresholds are plotted in the left column, click-evoked ABR waveforms recorded at 50-dB SL in the second column, and IOFs (parameter versus stimulus level above threshold, dB SL) are plotted in the third column. The green band denotes the 95% reference range of control values (defined in Methods). Red lines and circles represent mean responses (±SD) from mutant mice. Responses from individual mutants are shown by grey lines and open circles. Green lines and triangles represent the mean thresholds (±SD) for control mice recorded in the same week as the mutants. A,B. The Bai1 (Adgrb1) mutant line produced ABRs that were grossly abnormal. C-E. The Fam107b mutants produced a mild increase in thresholds but also have ABR waveforms with significantly reduced wave 1 amplitude. F-H. Sesn3 mutants showed normal thresholds but prolonged P3 latency. Plotted data points are given in S10 Data . ABR, auditory brainstem response; IOF, input-output function; SL, sensation level; SPL, sound pressure level.

https://doi.org/10.1371/journal.pbio.3000194.g006

Conclusions

We used a rapid ABR protocol to carry out a high-throughput screen of 1,211 new mouse mutants and revealed a new spectrum of functional deficits in hearing that would not have been detected using simpler screens, such as the startle response. These include mild-moderate degrees of hearing impairment, frequency-specific impairments (low or high frequencies), and a group with abnormal ABR waveforms that likely have deficits in central auditory pathways. In a subset of the new mutant lines, we have examined other ages to establish the time course of hearing loss and investigate the pathophysiological mechanisms underlying the raised ABR thresholds. A broad range of pathologies was found, and many mutants showed normal development followed by progressive hearing loss. We have shown that some of the genes highlighted by this study play a role in human hearing, including 2 genes with mutations that can account for recessive deafness in families and 11 genes that are associated with variation in auditory thresholds in the UK 1958 British Birth Cohort cross-sectional population. Thus, mouse mutants can be an effective means to identify candidate genes for human deafness.

This project has provided insights into the wide range of pathological processes involved in hearing impairment and has revealed a surprising number of unexpected genes involved in deafness, suggesting extreme genetic heterogeneity. For this reason, it is likely that therapies will need to be directed at common molecular pathways involved in deafness rather than individual genes or mutations. Each new gene identified gives insight into the metabolic pathways and regulatory processes involved in hearing and thus provides a rich source of targets for development of therapies for the restoration of hearing.

Materials and methods

Ethics statement.

Mouse studies were carried out in accordance with UK Home Office regulations and the UK Animals (Scientific Procedures) Act of 1986 (ASPA) under UK Home Office licences, and the study was approved by the King’s College London and Wellcome Trust Sanger Institute Ethical Review Committees. Mice were culled using methods approved under these licences to minimise any possibility of suffering.

For human studies, informed consent was obtained from the adult participants and the parents or guardians of children prior to participation, and the experiments conformed to the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report. The US patient provided consent for clinical whole exome analysis and written consent for inclusion as a case report. Testing was conducted during the routine clinical care of a patient in the US; thus, in accordance with US law, this study is exempt from Institutional Research Board approval.

Generation of mice

Mutant mouse lines were generated using targeted mutations in mouse ES cells [ 5 , 6 ]. The viability of new mutants was determined by genotype distribution at weaning. When possible, mice homozygous for the targeted mutation were used for screening. If a mutation proved to be embryonic lethal or had significantly reduced viability at weaning, heterozygous mice were used instead (356 genes out of 1,211; 29.4%). In most cases, the knockout first conditional-ready ( tm1a ) allele for each gene was used, but a subset of the genes were tested using the derived tm1b allele, which had deletion of a critical exon(s), or other mutations ( S2 Table , column B). The tm1a allele is designed to knock down transcription by introducing a large cassette into the gene, but not all genes were completely inactivated (see S2 Table and column X in [ 5 ], for some examples). Most mutant mice were screened on a C57BL/6 genetic background. This included some lines on a mixed C57BL/6Brd- Tyr c-Brd ; C57BL/6N line and others on a pure C57BL/6N line ( S1 and S2 Tables, column D). Mutant data were compared with a large set of wild-type data on the same genetic background. When mice were screened on mixed genetic backgrounds (for example, Ywhae tm1e ), age and strain-matched wild-type mice were used alongside the mutants. Positive control lines known to have a hearing impairment were compared with their littermate controls on the same, varied genetic backgrounds ( S1 Table ). At the age screened, 14 weeks, we found no apparent effect of the known Cdh23 ahl allele carried by C57BL/6 mice on ABR thresholds [ 46 ], and mutant responses were compared with mice of the identical genetic background. The Cdh23 ahl allele may have interacted with any of the new mutations to exacerbate their effect, such that the phenotype was easier to detect at 14 weeks, so this screen could be regarded as a sensitised screen [ 47 ].

Phenotyping pipelines

Details of the full phenotyping pipelines in the Mouse Genetics Project (MGP) have been reported elsewhere [ 5 ]. Two assays required mice to be immobilised and so to minimise stress to the mice, ABR testing was performed under anaesthesia immediately before the DEXA/Faxitron assays. Balance was assessed by observation of gait, head bobbing or circling, the rotarod test, or contact righting test. ABR testing was performed on three phenotyping pipelines used over successive time periods, termed MGP Pipeline 2, Mouse GP, and MGP Select, respectively. ABRs were recorded in mice aged 13 weeks (±3 days) on MGP Pipeline 2, and at 14 weeks (±3 days) on Mouse GP and MGP Select pipelines. Mice were maintained on a normal lab chow diet on MGP Pipeline 2 and MGP Select pipelines, but on a high-fat lab chow for the Mouse GP pipeline.

ABR recordings

ABRs were recorded using the methods described in detail in [ 4 ]. Mice were anaesthetised using intraperitoneal ketamine (100 mg/kg, Ketaset, Fort Dodge Animal Health, KS) and xylazine (10 mg/kg, Rompun, Bayer Animal Health), or a 10% greater dose for the MGP Select pipeline. Mice were placed on a heating blanket inside a sound-attenuating booth. Subcutaneous needle electrodes were inserted in the skin on the vertex (active) and overlying the ventral region of the left (reference) and right (ground) bullae. Stimuli were presented as free-field sounds from a loudspeaker whose leading edge was 20 cm in front of the mouse’s interaural axis. The sound delivery system was calibrated using an ACO Pacific 7017 microphone. For threshold determination, custom software and Tucker Davis Technologies hardware were used to deliver click (0.01-ms duration) and tone pip (6, 18, 24, and 30 kHz of 5-ms duration, 1-ms rise/fall time) stimuli over a range of intensity levels from 0 to 95 dB sound pressure level (SPL, re. 5 μPa) in 5-dB steps. Averaged responses to 256 stimuli, presented at 42.6/s, were analysed and thresholds established as the lowest sound intensity giving a visually detectable ABR response ( S1 Fig ; S1 Data ). Following completion of recording, mice were injected with intraperitoneal atipamezole (1 mg/kg, Antisedan, Pfizer) to promote recovery.

A fixed recording protocol was followed:

1. A series of click-evoked ABRs were recorded, ranging from 0 to 85 dB SPL in 5-dB intervals.

2. Tone-evoked ABRs were recorded for a fixed set of frequencies (6, 12, 18, 24, and 32 kHz) over sound levels ranging from 0 to 85 dB SPL in 5-db intervals. Different SPL ranges were recorded for different test frequencies to improve the time efficiency of the recording process (6 kHz, 20 to 85 dB; 12 kHz, 0 to 70 dB; 18 kHz, 0 to 70 dB; 24 kHz, 10 to 70 dB; 30 kHz, 20 to 85 dB). Responses were recorded in an array, beginning with the lowest stimulus level, in decreasing frequency order before stepping up to the next (5 dB higher) stimulus level. If mice appeared to have hearing impairment, the upper limit of SPLs was extended to 95 dB for each test frequency and for clicks (representing the upper limit of the linear range of our sound system at these frequencies).

For the ABR screen, we aimed to test a minimum of four mutant mice per line (of either sex). For other tests on the pipeline, 14 mutant mice (7 males and 7 females) were required. Phenotyping cohorts were issued as mice became available, such that several partial cohorts were issued, to achieve the required number of 14 mice for each single line. This allowed the ABR assay to pick up further mice from any lines that exhibited any features of interest to extend the number tested beyond the target of four. In addition to mutant mice, at least four wild-type mice from the same matings used to generate the mutants were tested each week from each core genetic background of the mutants tested. These wild-type results formed a local control group for comparison with the mutant lines and also contributed to a large reference range of control data that were used to determine if ABR results from a particular mutant line were significantly abnormal.

We compared ABR thresholds measured in 1,142 wild-type mice (female, n = 583; male, n = 559) from the pure C57BL/6N or mixed C57BL/6J and 6N genetic backgrounds. No sex differences were noted for mice of either genetic background ( S2 Fig U-X; S2 Data ).

Data analysis

We compared mutant data to a fixed population of control mice of the same genetic background (Pipeline 2, C57BL/6J and 6N mixed n = 201; Mouse GP, C57BL/6J and 6N mixed n = 951; MGP Select, pure C57BL/6N n = 742). The reference range for each parameter was defined as a 95% range of control values, from the 2.5 percentile to the 97.5 percentile of the control population, shown as a distribution around the median of the control data.

ABR responses were considered in three phases of analysis as follows:

1. ABR threshold and hearing sensitivity. For each stimulus used (click and five tone frequencies), ABRs recorded over the range of sound levels tested were plotted as a stack, ordered by increasing dB SPL ( S1 Fig ; S1 Data ). Threshold (dB SPL) was estimated by visual inspection of the stacked ABR traces as the lowest sound level at which any component of the ABR waveform was recognisable and consistent with responses recorded at higher sound levels, taking into account the characteristic lengthening of peak latency as threshold is approached. Thresholds for each stimulus were plotted to give a profile of the hearing sensitivity of each mouse.

2. Waveform shape comparisons. Through consistent, reproducible electrode placements, it was possible to compare, qualitatively and quantitatively, the waveform shapes of click-evoked ABRs. Wave 1 is understood to reflect auditory nerve activity, but as the responses represent a complex mixture of responses detected at a single point, there is some uncertainty in ascribing specific brain locations to specific features of the ABR waveform. The free-field binaural stimulation conditions we used complicates interpretation further, because there are binaural interactions even within the cochlear nucleus, e.g., [ 48 ]. The ABR represents the summed electrical vectors detected by the electrodes as synchronised action potential volleys (particularly from onset-responding neurons) traverse the central auditory pathways. As these pathways can be both excitatory and inhibitory, as well as both ascending and descending, and are distributed in a 3D volume, interpretation is complex.

Waveforms recorded to clicks at 20 dB and 50 dB above threshold (sensation level [SL]) were plotted for mutant and control mice, along with an average of the ABR amplitude over time across mice for each genotype. In these responses, we could identify four waves (positive to negative deflections; S1 Fig ; S1 Data ). We first performed a qualitative comparison of click-evoked waveforms recorded at 50-dB SL ( Fig 6 ; S4 Fig ; S4 Data ). The averaged mutant waveforms were compared with both the averaged control waveform and a 95% reference range of waveform amplitudes. We also compared individual mutant responses with the reference range. If both comparisons were in agreement between at least two of three experienced observers, a quantitative analysis was carried out of the peak amplitude, latencies, and intervals of these waveforms, by measurement of input-output functions (IOFs).

3. IOFs. Using click-evoked ABRs, waveforms were analysed in detail to determine the amplitude and latency of positive and negative peaks of the waveform at each stimulus level recorded ( S1 Fig ; S1 Data ). This was performed using software routines developed by Brad Buran and kindly donated for our use by M.C. Liberman (Harvard University). We found wave 1 and wave 3 were highly consistent in control mice. However, whilst wave 2 was clearly present as a single peak at low sound levels, it often split into two components at higher sound levels, making analysis complicated, so we did not include it. From these measures, we calculated the peak-peak amplitude of waves 1, 3, and 4, the amplitude of the N2-P3 component, and the intervals from P1 to P3 and N1 to N3. IOF curves were plotted relative to click threshold for each mouse (i.e., parameter plotted against dB SL). IOFs of individual mice (mutants and local controls) were plotted, together with 95% reference range generated for each parameter for controls.

Significance calls on mutant data

Wild-type control mice of the same genetic background tested in the same week as mutants were used as a local control population. As mutant mice were often tested in separate weeks to obtain the required numbers, the local control population for each mutant line varied in numbers. We compared each mutant population of results with a 95% reference range obtained from a fixed large number of wild-type controls of the same genetic background. Control populations for ABR thresholds of C57BL/6N or mixed C57BL/6N and 6J mice were not normally distributed (Shapiro-Wilk Normality test, p < 0.001). Furthermore, the large disparity in population sizes of the groups invalidates the use of traditional statistical tests giving p -values (e.g., t tests or analysis of variance) [ 42 ]. Thus, we used the following criteria to define parameters considered to be abnormal compared with controls.

1. ABR thresholds. A mutant line was considered to have abnormal ABR thresholds if one of two criteria were met: (1) at least 60% of mutant mice had thresholds for any stimulus outside the reference range, or (2) if the mean mutant threshold for any stimulus was at least 20 dB different from the median of the reference range for that stimulus. Thus, thresholds could be considered abnormal if they were elevated above controls (lower sensitivity) or reduced below controls (enhanced sensitivity). We did not find any mutants with enhanced sensitivity.

2. Waveform shape. Click-evoked ABR waveform shapes were compared at 20 dB and 50 dB SL. These comparisons were used as a subjective triage step in the assessment of whether waveform shapes were normal or perturbed in responses from mutant mice. A dataset was considered potentially interesting if two experienced observers considered the waveforms to be perturbed. In these cases, peak amplitudes and latencies were determined and IOFs plotted.

3. IOFs. IOFs were plotted for peak amplitude, latency, and also for wave 1–3 inter-peak interval as a function of dB SL. Due to the dependency of amplitude and latency on SPL, it is important to plot IOFs relative to stimulus threshold, so that any changes seen are not a result of variation in response threshold. Parameters for a mutant line were considered significant if the mutant mean value was outside of the reference range for at least 40% of the SLs measured (i.e., for at least 5/12 SLs, when comparing over a 60-dB suprathreshold range).

Quality control.

All mice issued for phenotyping were genotyped [ 49 ] at least twice; once was prior to cohort generation from a clip of pinna skin and a second time was at the end of the pipeline from postmortem skin tissue. Only when both results matched was the genotype for an individual mouse ‘locked’ (confirmed), to allow phenotyping results from that mouse to be included for analyses. ABR results were also subject to a quality control process. Visual inspection of the ABR traces recorded was used to look for significant noise or artefact on the recordings, which was accounted for when allocating threshold and other parameters. For example, excessive myogenic/electrical noise on the recordings can effectively mask an evoked potential, artefactually elevating the threshold estimate. Thresholds were allocated by experienced operators using the criteria outlined above. Data from random mice were checked by a second operator.

False positive threshold hits.

Our aim was to report robust effects on hearing that are likely to be reproducible in other laboratories, so we were cautious about calling the positive hits. All threshold calls made according to the two criteria detailed above were assessed by two experienced auditory scientists (one of whom was blinded to the genotype) to identify any false positive hits. A small number of false positive hits were discounted based on a number of principles, including excessive variability between individual mice (in some cases due to segregation of an independent new mutation, e.g., [ 20 ]) or the presence of a clear outlier (possibly due to a new mutation) in the mutant population skewing the mean threshold. On these grounds the number of false positive calls made by criterion 1 (60% rule) was 6 and by criterion 2 (20dB rule) was 12. Of the 1,211 genes tested, this represents a false discovery rate of 0.50% (criterion 1) and 0.99% (criterion 2).

Fisher exact test.

We carried out this additional statistical test for information about which threshold calls might be significant if the data had been normally distributed (see above for reasons why this was not our primary test). A two-way contingency table (using the number of threshold observations that were inside or outside the reference range) was generated for each mutant group compared with the wild-type control group, and this was used to carry out a Fisher exact test, giving a p -value indicating the likelihood of the two sets of data belonging to the same population (see [ 7 ] for details).

Classification of mouse ABR audiograms

Patterns of raised thresholds for ABRs were classified according to the following criteria: Severe-Profound , if no responses were detected (up to 95 dB SPL) for at least two adjacent frequency stimuli, for all mice of that genotype; High-Frequency , if thresholds were elevated at 30 kHz (by >30 dB) and thresholds were not elevated for at least one of the lower-frequency stimuli; Low-Frequency , if thresholds were elevated for 6 and 12 kHz and were normal for at least one of the higher-frequency stimuli (with a minimum mean threshold elevation <15 dB); Moderate , when thresholds were significantly elevated for at least four of the six stimuli tested (with a minimum mean threshold elevation >15 dB); Mild , when mean thresholds were elevated by 30 dB or less for up to three stimuli tested; Normal Hearing , when no stimuli produced altered thresholds.

Secondary phenotyping of selected lines

Methods used for histology, immunolabelling and confocal analysis, EP recording, and associated statistical tests used have been published elsewhere [ 10 , 11 , 12 , 20 , 42 ]. ABR thresholds were compared using the Mann-Whitney test.

Alleles and genetic backgrounds studied further.

The Srsf7 mutants studied further carried one copy of the mutant Srsf7 tm1a(EUCOMM)Wtsi allele, as homozygotes were subviable, on a C57BL/6N genetic background. Gpr152 mutants were homozygous for the Gpr152 tm1b(EUCOMM)Wtsi allele on a C57BL/6N background. The Klc2 mutant allele was Klc2 tm1e(EUCOMM)Wtsi on a C57BL/6N genetic background that had lost the distal LoxP site downstream of its targeted exon ( Fig 4K ). Ywhae mutants studied were homozygous for the Ywhae tm1e(EUCOMM)Wtsi allele, which had also lost its 3′ LoxP site ( Fig 5Q ), and were analysed on a mixed C57BL/6N and 129S5/SvEvBrd/Wtsi genetic background because of reduced viability on the original C57BL/6N background.

Auditory function recording.

ABRs were recorded in new cohorts of Srsf7 , Gpr152 , Klc2 , and Ywhae mutants along with littermate controls at 4 weeks, 8 weeks, 14 weeks, and 6 months old, as indicated in Figs 3 , 4 and 5 , and waveforms were analysed as described previously [ 10 ]. EPs were recorded in Klc2 mutants as described previously [ 10 , 11 ]. The 2f1-f2 distortion product otoacoustic emission (DPOAE) was recorded in Klc2 mutants in response to f2 frequencies ranging from 6,000 to 30,000 Hz in 500-Hz steps, where the f2:f1 frequency ratio was 1.2 and the f1 and f2 tones were presented at 70 dB SPL and 60 dB SPL, respectively. The 2f1-f2 DPOAE amplitude (dB SPL) was plotted against f2 frequency for control and mutant mice, for comparison.

Scanning electron microscopy.

Scanning electron microscopy was used to assess the organ of Corti in Klc2 mutants at 4 weeks old and Ywhae mutants at 8–9 weeks old, along with their littermate controls, and the middle ear mucosa of Ywhae mutants at 8–9 weeks of age. Inner ears were isolated (7 Ywhae homozygotes, 5 Ywhae heterozygotes, 2 wild-type littermates; 11 Klc2 homozygotes, 4 Klc2 heterozygotes, 2 wild-type littermates) and fixed in 2.5% glutaraldehyde and processed using the osmium tetroxide-thiocarbohydrazide (OTOTO) method as described previously [ 11 ]. Middle ear cavities ( n = 5 Ywhae homozygotes, 3 heterozygotes, and 3 wild-type littermates) were prepared by opening the bulla; removing the ossicles, muscles, and ligaments; and fixing in glutaraldehyde and osmium tetroxide, as above. Samples were examined in a Hitachi S-4800 or a JEOL JSM-7610F field emission scanning electron microscope. For the cochlear samples, images of the surface of the organ of Corti were taken at 10% intervals along the length of the cochlear duct or locations corresponding to the best frequency locations of the tones tested in ABR. The frequency areas were determined according to the mouse tonotopic cochlear map described by Müller and colleagues [ 50 ].

Immunohistochemistry and imaging.

Inner ears were fixed in 4% PFA for 2 hours and decalcified in EDTA overnight at room temperature (RT). Following fine dissection, the organ of Corti was permeabilised in 5% Tween PBS for 40 minutes and incubated in blocking solution (4.5 mL of 0.5% Triton X-100 in PBS and 0.5 mL of normal horse serum) for 2 hours. The primary antibodies used overnight at RT were mouse anti-GluR2 (1:200, MAB397, Emd Millipore), rabbit anti-Kcnma1 (1:100, APC-021, Alomone), rabbit anti-Myo7a (1:200, PTS-25-6790-C050, Axxora), mouse anti-Ctbp2 (1:400, BD Transduction Laboratories 612044), and chicken anti-NF-H (1:800, Abcam ab4680). The samples were incubated for 45 minutes at RT with secondary antibodies at 1:500 dilution: goat anti-mouse IgG2a Alexa Fluor 488 (A21131), goat anti-rabbit IgG Alexa Fluor 546 (A11035), donkey anti-mouse Alexa Fluor 594 (Molecular Probes A-21203), and goat anti-chicken Alexa Fluor 488 (Life Technologies A11039) and later were mounted using ProLong Gold mounting media with DAPI and stored at 4°C. Specimens were imaged using a Zeiss Imager 710 confocal microscope interfaced with ZEN 2010 software. The plan-Apochromat 63× Oil DIC objective was used for all the images with 2.0 optical zoom, and confocal z-stacks were obtained with a z-step size of 1 μm for innervation imaging, 0.5 μm for the Myo7a imaging, and 0.25 μm for other images. The frequency areas were determined according to the mouse tonotopic cochlear map described by Müller and colleagues [ 50 ].

The number of postsynaptic densities per IHC was quantified by counting manually the GluR2 puncta in the confocal maximum projection images and dividing it by the number of IHC nuclei (DAPI labelled). The cell-counter plugin in Fiji (ImageJ) software was used for counting. Three-dimensional reconstruction of the hair cell confocal stacks was performed using Fiji software (3D project function). Myo7a was labelled with the secondary antibody Alexa Fluor 546 and the nuclei with DAPI; however, false colours were used in the 3D reconstruction.

X-gal staining.

Heads from mice aged 4 weeks were bisected through the midline, the brain removed, and semicircular canal opened, then fixed in 4% paraformaldehyde at RT for 90 minutes. After decalcification in EDTA, the samples were washed for 30 minutes at RT with rotation with the detergent solution (2 mM MgCl 2 ; 0.02% NP-40; 0.01% sodium deoxycholate; 0.01% sodium deoxycholate in PBS, pH 7.3). X-gal (Promega, cat. no. V394A) was added 1:50 to prewarmed staining solution (5 mM K 3 Fe(CN) 6 and 5 mM K 4 Fe(CN) 6 in detergent solution), and then the samples were stained at 37°C in the dark overnight. The half-skulls were then washed twice with saline at 4°C in rotation for at least 2 hours and stored at 4°C in 70% ethanol until tissue processing and embedding. The samples were gradually dehydrated in ethanol (Leica TP1020 tissue processor) and embedded in paraffin using xylene as clearing agent (Leica EG1150H tissue embedder). The samples were cut at 8 μm thick, counterstained with Nuclear Fast Red (VWR, cat. no. 342094W), and mounted with Eukitt mounting medium (Sigma-Aldrich). Specimens were imaged using a Zeiss Axioskop microscope connected to an AxioCam camera and interfaced with Axiovision 3.0 software.

Skeletal measurements.

High-resolution radiographs were collected from 14-week-old Ywhae homozygotes ( n = 6 males, 6 females) with age-, sex-, and strain-matched controls ( n = 5 males, 6 females) using a Faxitron X-ray cabinet (MX-20, Faxitron X-ray, Wheeling, IL) and assessed using a standard set of parameters, including skull shape, mandible, and teeth [ 51 ].

Middle ear morphology.

External, middle, and inner ear regions were isolated from Ywhae mutants aged 16 weeks (4 homozygotes, 3 heterozygotes, 4 wild types). Left sides were examined by dissection for any signs of malformation or inflammation, including excessive cerumen in the external ear canal; thickening, whitening, sponginess, or vascularisation of the bulla wall; clarity of the tympanic membrane; presence of fluid or inflammatory debris in the middle ear cavity; and ossicle malformation. Right ears were fixed for 24–48 hours in formalin, decalcified in EDTA over 2 weeks, embedded in paraffin wax, and sectioned at 5 μm in a coronal plane. Sections were stained with haemotoxylin and eosin, scanned using a Hamamatsu NanoZoomer (Hamamatsu City, Japan), and examined. Examiners of middle ears and sections were blinded to mouse genotype.

Western blotting.

Brains from 16-week-old mice (2 Ywhae homozygotes, 2 heterozygotes, 4 wild types) were snap-frozen in liquid nitrogen for storage at −80°C. Protein extracts were generated by homogenisation in ice-cold T-PER protein extraction reagent (Pierce, Rockford, IL) containing a protease/phosphatase inhibitor cocktail (Halt, ThermoScientific, Waltham, MA). Protein (50 μg) from each sample was run in 4%–12% Bis-Tris gels (Life Technologies, Paisley, UK). A primary rabbit polyclonal antibody directed against a peptide mapping within a divergent domain of human YWHAE was used to detect the protein (T-16, sc1020, Lot: C1914; 1:200, Santa Cruz Biotechnology, Dallas, TX; Antibody Registry AB_630821) [ 52 ], PAFAH1b1/LIS1 was detected with recombinant rabbit monoclonal raised against a synthetic peptide corresponding to residues in human LISs1 (ab109630, Lot: GR55503-8; 1:750, Abcam, Cambridge, UK; Antibody Registry AB_10861275), and a blot with an additional LIS1/PAFAH1b1 rabbit polyclonal antibody generated against synthetic peptide corresponding to residues surrounding Gly298 of human LIS1 protein was used for validation (12453S Lot:1, Cell Signaling Technology, Danvers, MA). Goat anti-rabbit monoclonal vinculin antibody was used as a loading control (ab129002, 1:10,000, Abcam; Antibody Registry AB_11144129). A goat anti-rabbit IgG HRP-conjugate was used as a secondary antibody (12–348, Lot 2584427; Millipore, Billerica, MA; Antibody Registry AB_390191). The blots were imaged by enhanced chemiluminescence using an ImageLAS 4000 system (GE Healthcare, Chalfont St. Giles, UK).

Bioinformatic analysis of the 1,211 genes screened

GO term enrichment was analysed using FuncAssociate v3.0 ([ 53 ] http://llama.mshri.on.ca/funcassociate/ ), based on 22,644 genes (genespace), with 19,064 GO attributes, downloaded on 22 December 2015. FuncAssociate was configured to exclude computationally predicted GO annotations (Inferred from Electronic Annotation [IEA] evidence code) and run against the 1,211 genes tested by ABR. Revigo was used to reduce the redundancy in GO terms classed as over- or under-represented (by FuncAssociate) by clustering significant terms into more representative subsets ([ 54 ] http://revigo.irb.hr/ ). Only 0.42% GO terms ( n = 80) out of 19,064 were overrepresented in our list of 1,211 genes, and 0.09% ( n = 18) were under-represented. The over- and under-represented GO attributes were not significantly clustered into distinct subsets (visualised as TreeMaps). We also analysed this list of genes with the Reactome pathway database using Reactome V58 ( www.Reactome.org ), released October 2016; 10,168 human reactions were organised in 2,069 pathways involving 10,212 proteins and 10,214 complexes. Of the 1,211 genes tested, 564 were not listed in Reactome (46.6%). Of the 38 new genes underlying increased ABR thresholds, only 12 were included in Reactome. Of the 27 genes associated with abnormal ABR waveforms, only 17 were included in Reactome. Of 362 deafness genes already known, 233 were found in the Reactome databases. The 647 (53.4%) genes represented in the Reactome database out of the initial 1,211 did not produce any overrepresentation of any particular pathways (false discovery rate probability, FDR > 0.839).

Thus, taking the GO term and Reactome analyses together, the 1,211 mouse genes targeted can be considered to be representative of the entire mouse genome.

GO analysis of genes associated with hearing abnormalities

We used the GOSlim analysis tool on the MGI website ( http://www.informatics.jax.org/gotools/MGI_GO_Slim_Chart.html ) to compare the proportions of high-level GO terms associated with genes in each of five lists: new genes associated with raised ABR thresholds ( n = 38); previously known genes associated with raised thresholds, human and mouse ( n = 362); new genes associated with ABR waveform abnormalities ( n = 27); all genes screened by ABR ( n = 1,211); and all genes in MGI ( n = 33,395). We excluded evidence code IEA. The proportions of genes labelled with each high-level GO term in each group are plotted in Fig 2 .

Variant identification in candidate genes in human deafness

The hearing-impaired child was ascertained through the Kaiser Permanente clinic, testing was performed at Ambry Genetics, and the mutations were identified in SPNS2 by sequencing of candidate genes as part of clinical whole exome analysis. Variant pathogenicity prediction was carried out as previously described [ 55 , 56 ]. The link to the mouse study was established through GeneMatcher [ 57 ].

Human population analysis

The 1958 British Birth Cohort and the collection of hearing data and analysis have been described previously [ 58 , 59 , 60 ]. Participants were drawn from 17,638 individuals born in England, Scotland, and Wales in one week of March 1958. Of the original cohort, 9,377 members were revisited by a research nurse for a biomedical follow-up in 2002–2004. Hearing measures consisted of pure tone audiometry at 1 kHz and 4 kHz at age 44–45 years and were adjusted for sex, nuisance variables (noise at test, nurse performing test, audiometer used in test), conductive loss, and hearing loss in childhood. DNA was collected from 6,099 individuals and genotyped on various Illumina and Affymetrix SNP chips (for detail, see http://www2.le.ac.uk/projects/birthcohort/1958bc/available-resources/genetic ). These data were then imputed to the 1,000 Genomes haplotypes (released March 2012) using MACH and Minimac. Measured SNPs with >95% call rate and Hardy–Weinberg p -value >0.0001 were included as the input set. In subsequent analysis, imputed SNPs with low imputation quality (r2-hat < 0.3 or MAF < 1%) were omitted. For the association analysis, hearing thresholds at 1 kHz and 4 kHz were log transformed and adjusted for sex, nuisance variables, and conductive hearing loss in childhood; they were analysed by performing a 1-df ‘per allele’ significance test for association between mean hearing threshold and number of minor alleles (0, 1, or 2) as described previously [ 60 ]. The Bonferroni-corrected p -value for this candidate gene analysis is p < 6.76 × 10 −4 , based on testing 37 genes against two threshold frequencies (1 kHz and 4 kHz) at a significance p -value of 0.05, so all p -values listed are significant after correction for multiple testing. A p -value of p < 6.76 × 10 −4 represents the top 0.06% and 0.08% associations of all approximately 9 million imputed variants for 1 kHz and 4 kHz, respectively.

Supporting information

S1 fig. abrs, threshold estimation, and click-evoked waveform shape and parameters..

ABRs evoked by clicks (A), 6-kHz tones (B), 12-kHz tones (C), 18-kHz tones (D), 24-kHz tones (E), and 30-kHz tones (F) in a normal hearing wild-type mouse (black lines; 0–85 dB SPL) and a Duoxa2 homozygous mutant (red lines; 0–95 dB SPL). Threshold estimates are indicated by thickened black and red lines. In the case of the Duoxa2 mutant, no responses were measurable up to 95 dB SPL, and this sound level was arbitrarily assigned to be threshold. G. Threshold estimates from the responses from the two mice illustrated in A–F are plotted for the normal hearing wild-type mouse (black) and the Duoxa2 mutant mouse (red). The green area plotted indicates the 2.5–97.5 percentile range of thresholds for each stimulus, recorded from mice on the C57BL/6N genetic background, and represents the 95% reference range used to assess whether mutant mice have normal or elevated thresholds. H. ABR waveforms are illustrated for click stimuli at 50 dB above threshold in the normal hearing mouse (black line) and the 95 dB SPL click stimulus in the Duoxa2 mutant mouse (red line). The four waves of the click-evoked ABR are indicated by the grey areas, and the four positive and four negative peaks are labelled P1–P4 and N1–N4. Panels I-M indicate IOFs derived from features of the click-evoked responses recorded over a 60-dB range of stimulus levels above threshold. As above, the green area plots the 2.5–97.5 percentile range of each parameter and represents the 95% reference range used to assess whether mutant mice have normal or abnormal IOFs. I. IOF for wave 1 amplitude (P1-N1 amplitude). J. IOF for wave 3 amplitude (P3-N3 amplitude). K. IOF for latency of P1. L. IOF for latency of P3. M. IOF for the interval between P1 and P3. Plotted data points are given in S1 Data . ABR, auditory brainstem response; IOF, input-output function.

https://doi.org/10.1371/journal.pbio.3000194.s001

S2 Fig. ABR thresholds of positive controls, new alleles of genes known to underlie hearing impairment, and male and female wild-type mice.

ABR thresholds (dB SPL) are plotted against stimulus frequency and clicks. On each panel, green triangles and lines represent the mean thresholds (±SD) for control mice recorded in the same week as the mutants, and red circles and lines represent the mean thresholds (±SD) for mutant mice. Thresholds for individual mutants are shown by open grey circles and lines. Green bands denote the 95% reference range for a large population of control wild-type mice. Top panel: (A-J) Pre-existing mutant mouse lines known to have hearing impairment, tested as positive controls to validate the methodology used in this screen. Mutant lines tested were (A) Chd7 Whi , wild-type versus heterozygous mutants; (B) Atp2b2 Obv , wild-type versus heterozygous mutants; (C) Myo7a Hdb , wild-type versus homozygous mutants; (D) Myo7a sh1-6J , heterozygous controls versus homozygous mutants; (E) Cdh23 v , heterozygous controls versus homozygous mutants; (F) Grxcr1 Tde , heterozygous controls versus homozygous mutants; (G) Mir96 Dmdo , wild-type versus homozygous mutants; (H) Myo6 sv , heterozygous controls versus homozygous mutants; (I) Tmc1 dn , heterozygous controls versus homozygous mutants; and (J) Whrn wi , heterozygous controls versus homozygous mutants. Middle panel: (K-T) New targeted alleles of known deafness genes: (K) Myo15 tm1a(EUCOMM)Wtsi , (L) Myo7a tm1a(EUCOMM)Wtsi , (M) Ush1c tm1a(KOMP)Wtsi , (N) Ildr1 tm1(KOMP)Wtsi , (O) Espn tm1a(EUCOMM)Wtsi , (P) Whrn tm1a(EUCOMM)Wtsi , (Q) Cep250 tm1a(EUCOMM)Wtsi , (R) Srrm4 tm1e(EUCOMM)Wtsi , (S) Clpp tm1a(EUCOMM)Wtsi , and (T) Chd7 tm2a(EUCOMM)Wtsi (included as an example of normal thresholds). Bottom panel: (U-X) Mean (±SD) threshold for male (red) and female (black) mice are plotted for (U) Mouse GP pipeline, mixed C57BL/6Brd- Tyr c-Brd ;C57BL/6N line (B6JTyr/B6N) (female n = 243, male n = 237), (V) Mouse GP pipeline, C57BL/6N line (B6N) (female n = 191, male n = 180), (W) MGP Select pipeline, mixed B6JTyr/B6N (female n = 29, male n = 25), (X) MGP Select pipeline, C57BL/6N (female n = 120, male n = 117). Plotted data points are given in S2 Data . ABR, auditory brainstem response; MGP, Mouse Genetics Project; SPL, sound pressure level.

https://doi.org/10.1371/journal.pbio.3000194.s002

S3 Fig. Genes affecting central auditory system function with normal thresholds.

Mutant mouse lines showing altered ABR waveforms but normal thresholds are shown. ABR thresholds are plotted on the left of each set of three plots. Averaged click-evoked ABR waveforms recorded at 50 dB above threshold are plotted in the middle column of each set. IOFs for selected abnormal features are plotted on the right of each set. The green band denotes the 95% reference range of control values. Red lines and circles represent mean responses (±SD) from mutant mice. Responses from individual mutants are shown by grey lines and open grey circles. Green lines and circles represent the mean thresholds (±SD) for control mice recorded in the same week as the mutants. Plotted data points are given in S3 Data . ABR, auditory brainstem response; IOF, input-output function.

https://doi.org/10.1371/journal.pbio.3000194.s003

S4 Fig. Genes affecting central auditory system function and thresholds.

Mutant mouse lines showing raised thresholds and altered ABR waveforms are shown. ABR thresholds are plotted on the left of each set of three plots. Averaged click-evoked ABR waveforms recorded at 50 dB (unless stated otherwise on the panel) above threshold are plotted in the middle column of each set. IOFs for selected abnormal features are plotted on the right of each set. The green band denotes the 95% reference range of control values. Red lines and circles represent mean responses (±SD) from mutant mice. Responses from individual mutants are shown by grey lines and open grey circles. Green lines and circles represent the mean thresholds (±SD) for control mice recorded in the same week as the mutants. Plotted data points are given in S4 Data . ABR, auditory brainstem response; IOF, input-output function.

https://doi.org/10.1371/journal.pbio.3000194.s004

S1 Table. A summary of features of interest identified from ABR measurements in mutant mice.

On the ‘Gene List’ worksheet, columns (left to right) denote gene symbol, allele tested, genotypes compared, genetic background strain, and number of mutant mice tested. The heat map represents click-evoked, 6-kHz, 12-kHz, 18-kHz, 24-kHz, and 30-kHz ABR thresholds, click-evoked ABR waveform overlay of all mice tested at the same stimulus level above threshold, and IOFs of features of the positive (P) and negative (N) peaks of the ABR waveform, including P1-N1, N2-P3, P3-N3 amplitude; P1, N1, P3, and N3 latency; and P1-P3 and N1-N3 interval. Blue cells indicate a parameter within the normal range. Red cells indicate abnormal parameters (as detailed in Methods). For threshold estimates, a red cell containing ‘20dB’ was called on the basis of the mutant mean deviating by ≥20dB from the reference range median value; all other red cells were called on the basis that 60% of observations or more fell outside the reference range. For click-evoked ABR waveform shapes, a red cell indicates that the observation of the overlaid waveforms suggested an abnormal response and that this was supported by subsequent IOF quantification; an orange cell indicates that overlaid waveforms suggested an abnormal response but this was not supported by input-output analyses. For input-output analyses, a red cell indicates the parameter was outside the reference range for at least 40% of the sound levels tested, and the arrow indicates the direction of change. A cell containing a dash (-) indicates that input-output analysis was not carried out, as waveforms appeared normal. Grey cells (containing ‘n/a’) indicate that data were not available for analysis; for example, thresholds were too high to allow waveform overlays or that waveforms were too abnormal to permit IOF analysis. Genetic background: B6N, C57BL/6N; B6JTyr;B6N, mixed C57BL/6Brd- Tyr c-Brd and C57BL/6N; C3Fe, C3HeB/FeJ; B6JIco;B10, mixed C57BL/6JIco and C57BL/10ScSn; AK;BKS, AKR/J 25.0% and C57BLKS/J 75.0%; STOCK, partly undefined background. Further columns indicate mouse gene symbol, mouse Ensembl ID, human Ensembl ID, and gene symbol of the human orthologue. Other columns summarise the other phenotyping information obtained from the Mouse Genetics Project screen at the Wellcome Trust Sanger Institute. White cells indicate that data are not available; blue cells indicate no significant abnormalities; red cells indicate that one or more parameters varied significantly from control values. This heat map can be examined in more detail by clicking on the links on the top row, which links to other worksheets showing the individual parameters tested in each test area. In these worksheets, only parameters considered to vary significantly from control values are shown, plotted as red cells. Where appropriate, the direction of the change of the parameter is indicated by an up or down arrow. Cells containing ‘M’ or ‘F’ indicate that the change was significant for male or female mice only. White cells indicate that either the data were not collected or the results were not significant. On these worksheets, the link in cell E1 (‘Back to Gene List’) returns the user back to cell A1 of the main Gene List worksheet. ABR, auditory brainstem response; ID, identifier; IOF, input-output function.

https://doi.org/10.1371/journal.pbio.3000194.s005

S2 Table. A summary of ABR results from all mutant lines tested.

See S1 Table legend for details of columns and key to data representation. Additional columns to the right indicate the significant calls for each stimulus threshold, with red where p -value is greater than the appropriate critical value and blue where the p -value is not significant, using a two-way contingency table and the Fisher exact text. For new mutant lines tested, the Bonferroni-corrected critical p -value = 4.1 × 10 −5 , and for positive control lines, the critical p -value = 0.05, as these were compared with littermate controls due to their unique genetic background. Mutant lines tested are grouped in rows in the following categories: (1) Mutant mouse lines screened when gene was previously reported to be associated with hearing impairment. (2) Mutant mouse lines of known deafness genes where the mouse line did not show raised ABR thresholds. (3) Comparison of mutant lines when more than one allele was screened. For both Rnf10 and Selk , the tm1a allele responses were normal, but the tm1b allele responses showed an abnormal feature. For a further two genes, Spns2 and Zfp719 , responses from mice carrying the tm1b allele were marginally more affected than those from mice carrying the tm1a allele. In one unusual case, a targeted mutation in the Slc25a21 gene only produced affected ABR responses for the tm1a allele, where the inserted cassette of DNA was found to interrupt expression of the neighbouring Pax9 gene and result in phenotype differences [ 8 ]. Once this cassette was removed to produce the ‘b, ‘c’, and d’ alleles, expression of Pax9 was restored, and the normal ABR phenotypes were also restored. (4) The main grouping of the table lists a summary of the ABR results for all the mutant lines tested, listed alphabetically. ABR, auditory brainstem response.

https://doi.org/10.1371/journal.pbio.3000194.s006

S3 Table. Known deafness genes with normal ABR thresholds in the MGP Screen.

This table lists the genes that were previously known to be involved in deafness either in mouse or human, but which did not show raised ABR thresholds in the current study. The genotype reported deaf and the genotype screened are indicated. The right-hand column indicates possible explanations for the discrepancy in each case. ABR, auditory brainstem response; DFNA, non-syndromic deafness with dominant inheritance; DFNB, non-syndromic deafness with recessive inheritance; DFNX, non-syndromic deafness with X-linked inheritance; Hemi, hemizygote; Het, heterozygote; Hom, homozygote; MGP, Mouse Genetics Project; tm, targeted mutation.

https://doi.org/10.1371/journal.pbio.3000194.s007

S4 Table. Features of genes identified in this ABR screen.

This table summarises information from a range of sources to indicate features of interest about the genes detected with ABR defects in this screen, illustrating the broad range of types of genes found associated with hearing impairment. Columns A–I and K–P were derived from the MGI database, with links given in column E. For column M, references to phenotype details of alleles reported here are not included, only reports of other alleles of the same gene. For column P, only selected key GO terms are listed. Column J was obtained by comparing the human gene location with unidentified non-syndromic hearing loss loci listed in the Hereditary Hearing Loss Homepage (Van Camp and Smith; http://hereditaryhearingloss.org/ ; accessed January 2019). Column Q was derived from violin plots of single-cell mRNAseq data presented in the gEAR portal (Herzano and colleagues; https://umgear.org ; accessed January 2019). The criteria for inclusion of the genes were either (1) threshold above 95% confidence interval for at least one stimulus in 60% of mice of that genotype tested in the MGP screen, (2) *mean threshold for at least one stimulus more than 20 dB above wild-type mean, or (3) †further data obtained after screen confirmed raised thresholds in a larger n . All hits listed were in homozygotes unless noted as (Het) in column A when heterozygotes were screened. Four or more mice were screened unless a lower number is given in column A. ABR, auditory brainstem response; GO, Gene Ontology; MGI, Mouse Genome Informatics; MGP, Mouse Genetics Project; mRNAseq, messenger RNA sequence.

https://doi.org/10.1371/journal.pbio.3000194.s008

S1 Data. Data for S1 Fig .

https://doi.org/10.1371/journal.pbio.3000194.s009

S2 Data. Data for S2 Fig .

https://doi.org/10.1371/journal.pbio.3000194.s010

S3 Data. Data for S3 Fig .

https://doi.org/10.1371/journal.pbio.3000194.s011

S4 Data. Data for S4 Fig .

https://doi.org/10.1371/journal.pbio.3000194.s012

S5 Data. Data for Fig 1 .

https://doi.org/10.1371/journal.pbio.3000194.s013

S6 Data. Data for Fig 2 .

https://doi.org/10.1371/journal.pbio.3000194.s014

S7 Data. Data for Fig 3 .

https://doi.org/10.1371/journal.pbio.3000194.s015

S8 Data. Data for Fig 4 .

https://doi.org/10.1371/journal.pbio.3000194.s016

S9 Data. Data for Fig 5 .

https://doi.org/10.1371/journal.pbio.3000194.s017

S10 Data. Data for Fig 6 .

https://doi.org/10.1371/journal.pbio.3000194.s018

Acknowledgments

We thank the Sanger Institute Mouse Genetics Project (MGP) team, including Jeanne Estabel, Anna-Karin Maguire, Mark Sanderson, Hannah Wardle-Jones, Simon Maguire, Agnes Swiatkowska, and Ed Ryder for generating mouse mutants and carrying out screening of other phenotypes; James Bussell for mouse facility management; Rosalind Lacey, Amy Gates, and Katherine McGill for contributing to ABR screening; David Adams and Ramiro Ramirez-Solis for support of the MGP screen; David Melvin, Simon Holdroyd, Mark Griffiths, Wei Li, David Richardson, and David Gannon for contributing to software development to support mouse and data management; Natasha Karp for contributing to reference range methodology; M. Charles Liberman and Brad Buran for providing software to analyse ABR waveforms; Tim Folkard for contributing to ABR recording software; Nupur Kain and Johanna Pass for additional characterisation of new mutant lines; Ray Meddis for comments on abnormal waveforms; Adrian Davis, David Strachan, and Russell Ecob for access to the 1958 British Birth Cohort data; and the family for agreeing to participate.

View Article
Google Scholar
PubMed/NCBI
3. Davis AC. Hearing in Adults. London: Whurr; 1995.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

View all journals
Explore content
About the journal
Publish with us
Sign up for alerts
Open access
Published: 08 September 2024

An open-source tool for automated human-level circling behavior detection

O. R. Stanley 1 ,
A. Swaminathan 1 na1 ,
E. Wojahn 1 na1 ,
Z. M. Ahmed 2 &
K. E. Cullen ORCID: orcid.org/0000-0002-9348-0933 1 , 3

Scientific Reports volume 14 , Article number: 20914 ( 2024 ) Cite this article

Metrics details

Behavioural methods

Quantitatively relating behavior to underlying biology is crucial in life science. Although progress in keypoint tracking tools has reduced barriers to recording postural data, identifying specific behaviors from this data remains challenging. Manual behavior coding is labor-intensive and inconsistent, while automatic methods struggle to explicitly define complex behaviors, even when they seem obvious to the human eye. Here, we demonstrate an effective technique for detecting circling in mice, a form of locomotion characterized by stereotyped spinning. Despite circling's extensive history as a behavioral marker, there currently exists no standard automated detection method. We developed a circling detection technique using simple postprocessing of keypoint data obtained from videos of freely-exploring ( Cib2 −/− ;Cib3 −/− ) mutant mice, a strain previously found to exhibit circling behavior. Our technique achieves statistical parity with independent human observers in matching occurrence times based on human consensus, and it accurately distinguishes between videos of wild type mice and mutants. Our pipeline provides a convenient, noninvasive, quantitative tool for analyzing circling mouse models without the need for software engineering experience. Additionally, as the concepts underlying our approach are agnostic to the behavior being analyzed, and indeed to the modality of the recorded data, our results support the feasibility of algorithmically detecting specific research-relevant behaviors using readily-interpretable parameters tuned on the basis of human consensus.

Introduction

Observable actions serve as noninvasive readouts of underlying biological facts—e.g. injury, disease, gene expression, or neural function. This recognition sits at the core of behavioral sciences. Behavioral analysis has historically relied, and largely still relies, on labor-intensive manual behavior coding of real-time or videotaped behavior as its gold standard. Unfortunately, especially when analyzing long sessions or large numbers of sessions, manual coding can suffer due to rater variability, fatigue, or quirks in precise definitions 1 , 2 , 3 . Though recent advances in computer vision 4 and miniaturized sensors 5 , 6 have made quantitative data increasingly available, classifying kinematic data into specific behaviors remains a central challenge. Critically, behaviors which seem clear-cut to a human observer can in reality be noisy and subjective. As a result, while algorithmic behavior detection holds the potential for rapid, objective quantification, automated methods face difficulty in explicitly defining complex behaviors.

As one example, rodent behavioral neuroscience studies frequently report a repetitive spinning behavior known as 'circling'. The utility of circling as a behavioral marker, and the resultant need for objective quantification of the behavior, has been recognized for more than 50 years 7 . Examples include studies of basal ganglia damage 7 , 8 , 9 , 10 , 11 , 12 as well as genetically engineered models of Alzheimer’s Disease 13 , 14 and autism 15 , 16 , 17 . Additionally, a large number of mutant mouse strains displaying circling behavior display dysfunction of the vestibular system, which in healthy animals contributes to maintaining balance, steadying gaze, and keeping track of the body within the environment. These include mutants that exhibit loss of vestibular hair cells 18 , disrupted development of stereocilia 19 , 20 , or disrupted structural development of the inner ear 21 , 22 .

However, despite its long history as a behavioral marker, the existing literature lacks a standard, quantitative definition of circling. Rather, studies often report the simple presence or absence of circling 5 , 23 , 24 , 25 . Studies which quantify frequency of occurrence rely on manual coding and use disparate definitions such as complete rotations 26 , 27 , sequences of complete rotations 13 , or 270-degree turns during which the body travels a minimum distance 28 . Older studies which deployed video analysis relied on tracking the center of mass of a mouse against a high-contrast background 9 , 10 and could thus apply limited analysis, whereas more recent work has incorporated commercial and open source tracking solutions but focused on total amount of rotation 29 , 30 rather than circling per se. Accurate and objective quantification of behavioral parameters such as the frequency of occurrence, duration of bouts of circling behavior, and velocity of movements would facilitate comparison between specific etiologies. This inconsistency reduces the utility of circling as a tool for comparisons across models and setups. Thus, there is a need for a broadly-accessible, quantitative, automated tool for the detection of circling behavior.

Here, we present a technique for detecting circling behavior by tuning algorithmic parameters based on consensus occurrence times among human observers. Specifically, we assessed ( Cib2 −/− ;Cib3 −/− ) dual knockout mice, a mouse strain we and others have previously reported exhibiting circling 31 , 32 , 33 . We track the snout and base of the tail in mice using the open-source software package DeepLabCut 4 (“DLC”), then compare the performance of several behavioral detection algorithms which analyze characteristics of the animals' paths. By identifying the behavioral parameters that result in labels most closely matching human behavioral coding, our technique achieves statistical equivalence to individual observers' independent labels at matching consensus times. Our methodology thus provides a simple, inexpensive process for recording and quantifying mouse circling behavior. Furthermore, the success of our technique suggests its applicability for use in comparing against other etiologies and in detecting and quantifying other research-relevant behaviors.

The goal of the present work was to build and validate a tool to automatically identify mice exhibiting circling behavior during free exploration. Figure 1 illustrates an overview of our data collection and analysis pipeline. We first recorded a set of videos of five wild-type (C57BL/6) mice and five ( Cib2 −/− ;Cib3 −/− ) dual knockout mice in six different recording conditions (enumerated in Methods: “ Data generation ”). To establish a standard against which to compare the quality of automatic circling detection, three human observers independently marked times at which circling behavior occurred in two subsets of these videos: (1) a test set containing all videos of one mutant mouse and one wild type mouse and (2) an equally-sized training set randomly selected from the remaining videos, evenly split between mutant and wild type. These behavior labels were compared to identify consensus instances in which all observers marked circles sufficiently close to one another (0.1 s or less), which subsequently served as our gold standard (see Methods: “ Gold standard development ”).

Data collection conditions and analysis pipeline. We collected videos of five wild-type and five ( Cib2 − / − ;Cib3 − / − ) dual knockout mice exploring a 30 cm-diameter cylindrical arena. Each of 6 combinations of light and distance conditions was repeated 4 times for each mouse, resulting in a total of 236 videos as 4 became corrupted. After behavior videos were recorded, all videos of one mutant mouse and one wild-type mouse were set aside for human behavioral labeling as a test set. For each of these held-out videos, three observers independently marked occurrences of circling behavior. These behavioral labels were compared to produce a set of consensus labels on which all observers agreed. A separate training set of human behavior labels was constructed by randomly selecting 24 mutant and 24 wild-type videos from among the remaining 188 videos. Additionally, positions of the snout and tailbase were manually labeled in 20 random frames from each of these 188 videos. Manually-labeled bodypart locations were used to train a computer vision model using DeepLabCut. This trained model was then used to track animals in the human-scored videos, and the resulting paths were analyzed by three candidate circling detection algorithms. After the parameters of these algorithms were optimized for F1 score on the training set, they were applied to the test set for evaluation.

All videos not used in the test set were used to train and evaluate a computer vision model for tracking two keypoints on the mouse body, the tip of the snout and base of the tail. Once trained, the computer vision model was run on the manually-screened training set videos to generate tracking information. After minimal preprocessing of the resulting keypoint position data, we applied three progressively more sophisticated algorithms intended to identify instances of circling behavior. Parameters for each algorithm were optimized to match consensus times as closely as possible as measured by F1 score on the human-scored training set (see Methods: “ Algorithm development and parameter search ”). After this optimization, the trained computer vision model and finalized detection methods were applied to the manually-screened test set to establish whether these methods generalized to completely novel videos. Finally, to establish whether this large amount of data was required to achieve human-level performance at detecting circling behavior, we trained computer vision models on varied subsets of our overall dataset and applied our most performant algorithm to the test set.

Comparison and consensus among human observers

To obtain a standard against which to measure our behavioral labeling algorithms, we examined the degree of consensus among human labels of circling behavior. To this end, first we selected one mutant mouse and one wild-type mouse at random to form a held-out test set. All videos of these two animals (4 videos in each of 6 conditions; 24 videos each animal) manually screened for instances of circling independently by three observers, who were instructed to mark times at which they noted complete rotations during bouts of circling behavior but not during normal exploratory behavior shared by wild-type mice (e.g. turning around after reaching the edge of the arena). This process was repeated for 24 randomly selected mutant videos and 24 randomly selected wild-type videos from among the 188 not included in the test set. Videos were not labeled with information regarding mouse genotype.

We hypothesized that human observers marking occurrences of circling behavior would show consistent agreement. To assess this, we calculated F1 scores for pairs of observers, treating one as ground truth for another (Fig. 2 A). Contrary to our expectation, we found that although pairwise F1 scores were similar on average (0.53, 0.52, and 0.49), the distribution of those scores varied widely enough that one pair differed significantly from both others (pair CA, p = 3.5E−2 and 1.4E−4 versus pairs AB and BC).

Human F1 scores. ( A ) Treating one independent observer as the gold standard for another reveals that humans show substantial variability in labeling circling behavior. In particular, although average F1 scores for each pair (AB, BC, CA) are similar (0.53, 0.52, 0.49), the distributions of scores across videos differ significantly between one pair and the other two (pair CA, p = 3.5E−2 and 1.4E−4 versus pairs AB, BC respectively) while the other pair did not differ significantly (AB versus BC, p = 0.28). ( B ) Scoring of independent observers' labels against another observer (left columns) or against consensus labels (agreement among 3 observers, right columns) produce similar results (p = 0.2), as does comparing between our two human data subsets (train versus test subset, p = 0.65 and 0.75). Pooled pairwise F1 scores averaged 0.51 (95% CI 0.47–0.55) in the training set and 0.53 (0.41–0.62) in the testing set. Scoring against consensus occurrences, in which all observers mark a complete circle within 0.1 s of one another, produced similar scores of 0.51 (0.44–0.57) in the training set and 0.53 (0.38–0.65). Each point in a column represents a single video. Labeler-video combinations for which F1 score is undefined (i.e., both scorer and ground truth marked no circling instances), are not displayed for either paired or consensus scoring but were included in bootstrapping for purposes of calculating confidence intervals.

We next examined whether human performance differed between our training and test sets as well as how well independent labels matched ultimate consensus circling occurrence times, i.e., cases in which all three observers indicated a circle sufficiently close to one another. Independent observers scored against each other produced an average F1 score of 0.51 (95% CI 0.47–0.55) on the training set and 0.53 (0.41–0.62) on the test set. (not significantly different, p = 0.65; Fig. 2 B, Independent vs Independent) When all observers were scored against a group consensus gold standard, average F1 scores were similar to pairwise scoring: 0.51 (0.44–0.57) and 0.53 (0.38–0.65) on the training and test sets, respectively. (p = 0.75; Fig. 2 B, Independent vs Consensus) We observed no significant difference between these alternative gold standards (p = 0.20). Consensus times were used for subsequent tuning and evaluation of automated behavior detection methods.

Developing and testing algorithmic circling detection methods

Accurately tracking an animal's position, or even the position of many body parts, is insufficient on its own to establish the behavior an animal is exhibiting. Rather, this raw data must be processed once collected. To this end, we applied several candidate algorithms for detecting circling behavior using keypoint tracking data after minimal preprocessing. In our case, this preprocessing consisted of setting a threshold on frame-by-frame keypoint label jumps equal to a velocity of 40× the median length of the labeled tail-snout vector per second, a value selected to be high enough that no plausible physical process could produce it. Datapoints exceeding this limit were replaced with simple linear interpolation.

Each behavior detection algorithm first detects instances in which the path of the mouse's snout intersects itself (see example timelapse, Fig. 3 A) to analyze as candidate circling occurrences. Importantly, not all intersections are the result of circling—many will be produced by normal exploratory head movements made by both mutant and wild-type animals. Each algorithm therefore attempts to exclude false positives (i.e., instances incorrectly marked as circling) using one or more features of the animal's path between the points of collision.

Method parameters and performance levels. ( A ) Timelapse of keypoint-labeled frames of a mouse engaged in circling behavior. ( B ) Parameter distributions and associated exponential and Gaussian fits from two sample videos. To accommodate the substantial variability observed across videos, we relied on a two-step process of Gaussian kernel estimation followed by fitting to a weighted sum of an exponential and normal distribution. This allowed the same technique to account for differences in e.g. average duration (left column, compare blue Gaussian fits) or greater numbers of small collisions likely to be false positives (right column, compare pink exponential fits). ( C ) Illustration of circle detection using each of the described methods. Duration-Only considers only time taken to complete the putative circle, Time-Angle additionally calculates the angle of the tail-to-snout vector for each frame and considers its total net change, and Box-Angle removes duration requirements and instead constraints the geometry of the circle based on the axes of a rectangle bounding the candidate circling instance. ( D ) Examples of false-positive detections using each method. There are clear features which indicate an instance should be filtered out for the Duration-Only (minimal head movement relative to the tail) and Time-Angle (oblong or missized snout path geometry) methods.

We observed that the underlying distribution of putative circling parameters (duration, rotation, and size) could be well described by an initial decay for parameter values near zero and a roughly normal peak. We thus sought to model the distribution of each of these parameters as the sum of an exponential and a Gaussian distribution (see Methods: “ Algorithm development and parameter search ”). Figure 3 B illustrates these distributions and fits for two example videos. Bounds outside of which collisions were excluded as circles were parameterized by considering how far, and in what direction, a candidate circle falls from the center of the fitted Gaussian in terms of standard deviations. For example, a snout path collision produced by an exploratory head movement during which a mouse rotated its body only 10 degrees would fall many standard deviations short of the average and thus be correctly excluded. Values of these features, as measured by F1 score against human consensus labels in our training set, were optimized via an iterative Bayesian search over the range of ± 3 standard deviations from the mean of the Gaussian fit. Table 1 lists the parameters and final values for each method; parameter ranges were searched independently for each method.

Duration-only method

To establish a lower bound on automatic behavioral detection performance, we first assessed a simple detection algorithm which considers only how long a candidate circling instance takes. The ‘Duration-Only’ method locates points at which the path of the mouse's snout crosses over itself and excludes those which are either too short or too long (Fig. 3 C, Row 1). Our parameter search produced an F1 score of approximately 0.21 in the training set. The same parameters applied to the manually-screened test set scored 0.10 (95% CI 0.02–0.17). As expected given its simplicity, this fell well below human performance and was not sufficient to filter out many false positives which appear obvious upon review. Specifically, it incorrectly labeled many cases of head-only exploratory movements as circling, examples of which are illustrated in Fig. 3 D, Row 1. To filter these out more effectively and thereby improve behavioral detection, we next explored incorporating the labeled tailbase position.

Time-angle method

For a mouse that is spinning rather than exploring with its snout alone, the angle of its snout relative to its tail should change noticeably and in a consistent direction. Accordingly, we next considered what we term the 'Time-Angle' method for excluding false-positive instances of circling. As illustrated in Fig. 3 C, Row 2, this method calculates the angle of the mouse's body in each frame using the vector from labeled tailbase position to labeled snout position. It then screens candidate circles using bounds on duration as well as minimum and maximum total rotation.

The optimized Time-Angle method reached an F1 score of approximately 0.38 in the training set and 0.22 (95% CI 0.03–0.47) in the test set. Thus, incorporating the additional information of the tail's position resulted in a substantial increase in behavioral detection performance which nevertheless remained well below human level. However, in examining erroneous circles detected by this method (Fig. 3 D, Row 2), we identified many cases in which false positives were either clearly too small, too large, or distinctly oblong. To counteract this, we sought to incorporate additional geometric information about candidate circles.

Box-angle method

To filter out cases based on the animal's snout path, we implemented a final method which discards constraints on duration and instead contains an additional step in which a candidate circle is fitted to a minimum-area rectangle to provide additional geometric information (see Fig. 3 C, Row 3). The resulting 'Box-Angle' method constrains the side lengths of the resulting fitted rectangle. To avoid relying on specific information about the camera and the geometry of the recording apparatus, the minimum side length is specified relative to the median body length of the mouse as measured by tail-to-snout vector over the course of the video. Notably, during our parameter search we found that the inclusion of temporal information (duration constraints) actually resulted in a slightly worse training set performance (F1 score of 0.39 with vs 0.41 without). The best-performing parameters on the training set produced a test-set F1 score of 0.43 (95% CI 0.21–0.57). We observed that the false positives produced by this more sophisticated method do not suffer the same obvious flaws as those from the less complex techniques (Fig. 3 D, Row 3),

Method performance comparison

As illustrated in Fig. 4 , the straightforward Duration-Only method performs significantly lower than both human performance (p = 1.1E−11, two-tailed Wilcoxson signed rank test of automatic scores versus human scores) and the automatic Box-Angle methods (p = 2.8E−6, two-tailed Wilcoxson signed rank test between automatic scores from each method). The Time-Angle method, in turn, performs on par with the Duration-Only method (p = 0.2) and underperforms both independent human labeling (p = 4.7E−6) and the automatic Box-Angle methods (p = 1.9E−3). Finally, the extensive filtering of the Box-Angle method results in a performance distribution not significantly different from human observers (p = 0.51).

Method performance comparison. After optimizing behavior detection algorithms on the human-labeled training set, each was scored on the human consensus circling labels of the test set. Each column represents one algorithm, with one dot for each test set video with a defined score. Videos for which F1 score is undefined (i.e., the automated method and human consensus both marked no circling instances) were included in confidence interval calculations but not displayed as individual datapoints. The Duration-Only and Time-Angle methods significantly underperformed independent human observers (mean and 95% CI 0.1 (0.02–0.17) and 0.22 (0.03–0.47), p = 1.1E−11 and 4.7E−6, respectively). Only the Box-Angle method reaches statistical parity (mean F1 0.43 (0.21–0.57), p = 0.51).

Differentiation between mutant and wild-type mice

As the present work was aimed to develop a tool for detecting vestibular-mutant mice, the ultimate determinant of success is whether the presence of an automatically-detected circling instance positively identifies a video of a mutant mouse. The Box-Angle method developed above correctly reports zero circles for 20 of 24 test-set videos of wild-type mice, whereas at least one circle was scored in 16 of 24 mutant mouse videos. Thus, in our test dataset, the presence of at least one automatically-detected circle detects videos of ( Cib2 −/− ;Cib3 −/− ) mutant mice with an F1 score of 0.73.

Impacts of dataset size

All the results reported thus far were produced using a DLC-trained convolutional neural network model, termed the "Full Dataset" model, which used 20 manually-labeled frames from each of 188 mouse behavior videos not used for human labeling. This model represents a substantial investment of experimenter effort, raising the question of whether sufficient behavioral detection performance could be achieved more easily. In particular, the publishers of DeepLabCut observed good labeling performance with substantially smaller datasets than what we use for our “Full Dataset” model 4 , but it was unclear a priori what labeling quality would be necessary for successful behavior detection.

In order to investigate the minimum amount of data and thus human labor needed to obtain good automatic behavioral detection performance, we trained several DLC models using different subsets of our manually-labeled frames, detailed in Table 2 . Specifically, ten models were trained for each of several training dataset sizes—half, one-quarter, and one-eighth of the full dataset. Figure 5 A displays the labeling performance (root-mean-squared error, in pixels) for all 10 networks within each dataset size category. For frames within a model's training set, these network families reached errors of 9.29 (8.13–10.73), 9.84 (8.53–11.7), and 11.02 (9.11–12.91) pixels (mean & 95% CI). For unseen frames, these errors increased to 19.37 (16.92–22.28), 12.3 (10.51–14.4), and 14.34 (12.66–15.98). Training data for each model was randomly selected to include all 20 labeled frames from the appropriate number of videos, i.e., data were shuffled by video rather than by frame. Subsequently, the ability of each model to accurately label the locations of the snout and tail-base on previously unseen videos was evaluated on a randomly-selected 50% (94 videos) of the full dataset which did not overlap with that network's training data. Notably, while each family of networks produced larger errors on average than the full dataset model (7.82 pixels, dashed horizontal line), reduced dataset size did not monotonically worsen labeling performance on previously unseen videos.

Dataset size performance comparison. ( A ) Labeling performance (error, in pixels) for each of 10 trained networks on datasets of progressively smaller sizes. All dataset sizes resulted in greater labeling error than the Full Dataset model (dashed horizontal line), particularly for frames not seen during training (test frames). Notably, this trend was not monotonic—the set of quarter-dataset models performed better on test frames, on average, than the set of half-dataset models. Root-mean-squared errors on training set frames were (mean and 95% CI) 9.29 (8.13–10.73), 9.84 (8.53–11.7), and 11.02 (9.11–12.91) pixels respectively. For unseen frames, these errors increased to 19.37 (16.92–22.28), 12.3 (10.51–14.4), and 14.34 (12.66–15.98). Dashed horizontal line represents Full Dataset model training frame error (7.82 pixels). ( B ) To determine whether these changes in labeling quality impacted, we applied the optimized Box-Angle method to the keypoint tracking produced by each network at each dataset size. Within a dataset, the true-positive, false-positive, and false-negative scores for each video were summed to calculate a representative F1 score, plotted here as individual dots in the half-, quarter-, and eighth-sized datasets. The resulting distributions are compared to scores from the Full Dataset network (left column) and to independent human scores (right). As elsewhere, video-net combinations for which F1 score are undefined are included in confidence interval calculations but not displayed as individual datapoints. These smaller datasets underperformed the Full Dataset network (p = 0.03, 0.03, 0.02) as well as human labels (p = 1.7E−4, 1.4E−4, 3.9E−5), indicating that even small reductions in keypoint tracking quality can impact behavioral detection. * p < 0.05, *** p < 0.001.

To determine whether these differences in pixel-wise error impacted the ability of networks to successfully identify circling behavior, we ran F1-score-optimizing parameter searches over the human-scored training videos for each network using the Box-Angle method. As described above, the Full Dataset network achieved an F1 score across the manually-screened videos of approximately 0.43. Model families trained on smaller datasets again showed a non-monotonic relationship between performance and dataset size, with half-, quarter- and eighth-sized datasets producing mean F1 scores across the test video dataset of 0.39, 0.41, and 0.36 respectively (Fig. 5 B). These scores fell below the performance of the Full Dataset model (p = 0.03, 0.03, 0.02) and failed to match human performance (p = 1.7E−4, 1.4E−4, and 3.9E−5), suggesting that even small reductions in tracking quality were functionally significant at the behavioral analysis level.

In the present study, we developed a technique to automatically detect circling behavior in videos of freely exploring mice using readily-available, consumer-grade hardware and open-source software. Importantly, this method is applicable to analyzing keypoint tracking captured by any method, whether open-source or proprietary, so long as it includes snout and tailbase positions. This makes it a convenient, quantitative tool to screen mice for circling behavior according to specific, objective criteria. More generally, our results suggest that similar procedures to develop consensus behavioral labels among human observers could be straightforwardly applied to enable effective, automatic detection of other behaviors of interest. The development of such quantitative tools with low barriers to entry is essential for the comparative analysis of behavior, as it expands the number of research groups able to produce directly inter-comparable data.

Limitations of prior approaches

Studies which report circling have varied widely in methods of detection and analysis, ranging from qualitatively reporting the presence or absence of circling 5 , 23 , 24 , 25 to counting rotations per minute using manual or automated video tracking 9 , 10 , 29 , 30 . Manual methods place substantial demands on human time; in our experience, marking circling behavior in a given video took on the order of four to five times as long as the video itself on average, owing to the need to e.g. play videos slowly to avoid missing instances, pause videos to make notes, and replay sections of videos to ensure the timing of instances were precisely noted. Indeed, both the low overall F1 scores and the inter-video variability in those scores observed in comparisons among human observers (Fig. 2 ) suggest that an automated detection system for circling behavior might be advantageous in that, in addition to not becoming distracted or fatigued 1 , 2 , 3 , it would give a consistent output for a given video.

Software to track rodents in videos is commercially available, but such products face issues of both price and opacity; as closed-source software, they limit the ability of researchers to examine how specific results were generated and to customize or otherwise modify those underlying methods 34 . In the present study, we chose to use the open-source markerless feature-tracking toolbox DeepLabCut to track the positions of keypoints on animals. However, the analysis applied to this positional information is agnostic to the tracking method used (see Discussion: “ Availability and use ”). Existing automated methods using video analysis face limitations in quantifying circling behavior due to false positives arising from grooming or exploratory turns. Our work specifically addresses this issue by incorporating filters for excluding false positives using carefully tuned geometric parameters—by tracking only two key points on the mouse and applying straightforward algorithms, we achieved behavioral labeling performance (F1 score 0.43) statistically similar to that of independent human coders (0.53, p = 0.51, Fig. 5 B).

Availability and use

Our Full-Data model, the Python script to detect circles from keypoint positions, and an associated Anaconda environment file can be found at our GitHub ( https://github.com/CullenLab/CirclingDetection ) along with a step-by-step guide to installing and using the system. Use requires only basic familiarity with the command line or with integrated development environments. To use our code, we recommend installing Anaconda (see anaconda.com ) and using it to create a Python environment from the ‘ circlingmouse.yaml ’ file included in the GitHub Repository (i.e., using ‘ conda env create -n circlingmouse -f circlingmouse.yaml ’). Tracking data in the form of a comma-separated-value (CSV) file, whether from DLC or other sources, should be placed within the downloaded folder. Although we chose to develop our methodology using DLC, the detection algorithm can in principle be employed using any technique to track the snout and tail base. Those utilizing other tracking methods should note that while the ‘ Circling_BoxAngle.py ’ script assumes a particular configuration of columns in the files to be analyzed (described further in the repository documentation), CSVs produced by other tracking methods can be straightforwardly modified to fit this scheme. The code can either be run from the command line (‘ conda activate circlingmouse ’, then ‘ python Circling_BoxAngle.py $Your_Tracking_CSV_Here$.csv ’ from within the downloaded Repository), or run via one of the development environments Anaconda offers.

In our analysis, we found that human-level performance required accurate keypoint tracking and extensive filtering of false positives. As a result, we encourage the use of our pretrained Full Dataset model by future experimenters wherever possible to make best use of human time and effort. Furthermore, regardless of the tracking method used, we recommend applying the included Box-Angle method code for experiments focused on circling, as it incorporates the most information about the animal's movements and was the only technique to reach parity with human observers. As the analysis is applied directly to the labeled coordinates of the snout and tailbase keypoints, the Box-Angle method will be usable alongside any high-quality tracking method. As noted below, however (see Discussion: “ Directions for future work ”), our tools may not be appropriate for e.g. mice of varied colors or in complex environments, nor for behaviors whose detection requires tracking additional keypoints.

Conclusions and implications

Here, we present an open-source tool which identifies circling, a specific, abnormal locomotor behavior which distinguishes between wild-type mice and many mouse models of interest to biological researchers, in this case a strain of vestibular mutant mice. We first revealed the substantial variability exhibited by human observers in manually marking this behavior, then developed a tool which uses simple, directly interpretable parameters to achieve statistical parity with independent human observers when scored against human consensus.

Emerging video-based technologies that facilitate the tracking of key body features are well-suited to the development of accessible methods for objectively quantifying behavior, including the effects of vestibular loss on those behaviors. Such studies are clinically relevant in light of aging populations, as vestibular dysfunction substantially increases fall risk and causes symptoms including dizziness, vertigo, nausea, and blurred vision. In adults over 40, its prevalence has been estimated as ranging from more than one in twenty 35 (using vestibular-specific clinical measures) to more than one in three people over 36 (using broader balance-related symptoms).

Mouse models are advantageous for studies of treatments and causes of vestibular impairment, due in part to the ability of researchers to genetically engineer new strains using increasingly sophisticated tools 37 , 38 . Recent developments in mouse genetic engineering, involving the creation of transgenic and knockout mutant mice, have provided a novel opportunity to study the relationship between genes and behavior 3 . For example, many mutant mice strains that have been characterized with an underlying impairment of peripheral vestibular function display a circling behavioral phenotype. While there exist a number of non-invasive methods to detect the existence of vestibular dysfunction in mice, such as the rota-rod and balance beam tests 6 , the variety of vestibular-loss circling mouse models suggests screening for circling may serve as a convenient screening tool for identifying novel models of vestibular dysfunction. In the present study, we specifically assessed the circling behavior ( Cib2 −/− ;Cib3 −/− ) dual knockout mice 31 , 32 , 33 . CIB2 is found in the stereocilia tips of the receptor cells within the vestibular sensory organs 39 (i.e., vestibular hair cells), suggesting the circling behavior observed in these mice results specifically from deficits in peripheral mechanotransduction. Notably, the strain studied in the present work is just one example of a large number of strains with mutations homologous to subtypes of Usher syndrome 40 , including deaf circler 26 , waltzer 29 , Ames waltzer 41 , and Jackson shaker 42 mice. In the present study, comparing those parameters that are shared across the developed methods (i.e., duration and rotation, Table 1 ) reveals that the parameters which resulted in optimal performance varied depending on what additional information was included; we speculate that this variability arises from the variability of the underlying behavior. Notably, the similar performance observed when applying the optimized Box-Angle method to novel videos (0.41 vs 0.43) suggests we have successfully built a tool which was robust to this variability. Further work automatically and objectively quantifying circling behavior may reveal otherwise undetectable differences in circling parameters (e.g. differences in frequency, rotational velocity, and geometry) between model strains.

An important advantage of video-based approaches to behavioral and especially vestibular neuroscience is that they are non-invasive. This stands in contrast with the emerging use of head-mounted sensors to assess motor impairments in mouse models (see for example 5 , 6 ), which typically require an experimental surgery to securely fix the sensor to the mouse’s skull. We speculate that as the spatio-temporal resolution of readily available, consumer-grade hardware continues to improve, the tradeoff in resolution between a completely noninvasive recording technique (video analysis) and a more invasive technique in which inertial sensors are mounted surgically will become less critical.

More broadly, tools for detecting and quantifying behavior which do not rely on specialized experience with software engineering will allow a wider array of research groups to directly compare analyses of neurodevelopmental differences and intervention effects. The use of shared automatic behavioral analysis tools would increase both the speed and consistency of behavioral labeling, especially when analyzing long or numerous videos. For example, the technique presented in this paper could aid in screening novel mutants for vestibular dysfunction. Common behavior definitions may also provide inter-comparability among studies within and across research groups as well as improving the reproducibility of those studies.

As noted above, the development of this work was motivated in part by substantial variability seen among human observers when labeling behavior independently. The success of the methodology developed here argues for the broader use of tuning behavioral detection algorithms based on consensus among multiple human observers. In particular, by demonstrating the effectiveness of an automated detection system for circling behavior tuned to match group consensus, our results provide a convenient, quantitative screening method for mouse models of vestibular dysfunction, which we hope will serve as an important step toward standardized, automatic measurement of motor dysfunction in mouse models that provide more reliable measurements across studies and laboratories.

Directions for future work

The current study focuses on creating as simple a method as possible for detecting a specific behavior. Importantly, however, this work is limited to first-order features of the tracked keypoint paths, i.e. it does not incorporate information regarding velocity or acceleration which human observers have access to when watching behavior videos. Further refinement of automated methods may take advantage of these or other higher-order features. Additionally, it seems clear that incorporating a larger number of keypoint labels will be critical for richer examination of rodent posture and gait 43 , and has previously been incorporated into other analyses of circling mice specifically 44 . As our Full Dataset model was trained for only the two keypoints of interest to us, future researchers seeking to apply our findings to more complex behaviors will likely be unable to employ the specific computer vision model used here. Our model is further limited in that its training data consists of black-furred mice on white backgrounds (see example frames, Fig. 3 A); we expect this will result in degraded tracking of animals of varied colors or on low-contrast backgrounds.

It is noteworthy that the concepts underlying our approach can be readily applied to other behaviors of interest to researchers. Specifically, by first creating a set of occurrence times coded independently and then constructing a set of consensus occurrence times, we were able to directly quantify human-level performance. In principle, this enables working with a wide range of behaviors which may be difficult to define explicitly ahead of time but which we 'know when we see them'. Indeed, the concept is not limited in application to the analysis of visual information. Rather, in any situation where human observation is currently required to disentangle ambiguity in behavioral data, it is possible to optimize automated analysis to match agreement among multiple independent observers as well as possible to avoid incorporating biases or quirks of any one observer. This is equally true for the identification of freezing behavior 45 , which must be distinguished from simply remaining still just as circling must be distinguished from normal exploration; analysis of social interactions among animals 46 , the dimensionality of which will rise exponentially with the number of individuals being considered; or attempting to detect a particular category of vocalization 47 , 48 , which may occur in noisy environments or be ambiguous as to whether a call represents e.g. a warning of approaching predators versus a warning to deter conspecific competitors. Although differences will arise in the choice of features being used as input to automated detection methods, which will depend on the behavior and modality of interest (e.g. rotational velocity for circling detection versus linear velocity for freezing detection), in all cases the process of generating consensus from independently-marked behavior timing will be similar.

Animal care and housing

Five adult wild-type (C57BL/6) mice (Jackson Laboratories) and five adult mutant ( Cib2 −/− ;Cib3 −/− ) mice were used in this study. Generation of mutant mice is described in a separate paper 33 . Animals were group-housed with their littermates on a 12:12 h light: dark cycle at 20 °C with ad libitum access to food and water. All animal procedures complied with the ARRIVE guidelines and were carried out in accordance with National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publications No. 8023, revised 1978) and were approved by the Institutional Animal Care and Use Committees (IACUCs) at University of Maryland (protocol #0420002).

Data generation

We recorded videos of 5 wild-type mice and 5 ( Cib2 −/− ;Cib3 −/− ) dual knockout circling mice during single-animal exploration of a cylindrical arena 30 cm in diameter, similar to ones used in past studies 8 , 9 , 15 . For each mouse, we recorded four 2-min videos at 60 fps in each of six recording conditions—low (~ 4 lx), moderate (~ 375 lx), or bright lighting (~ 875 lx), each with the camera either near to (60 cm) or far from (100 cm) the animal. These were selected to span a broad range of potential experimental setups and to provide varied data for neural network training; conditions are illustrated in Fig. 1 . Unfortunately, during the course of the study one set of video data (four videos of a mutant mouse in low-light, near-camera conditions) became corrupted; as this mouse was not the one used for human behavioral labeling, we believe the impacts on the study’s results were negligible. Our total set of 236 videos thus consisted of 116 videos of wild-type mice and 120 videos of mutant mice.

Gold standard development

Assessing the effectiveness of our tool required a ground truth against which we could compare the automatic detection of circling behavior. In order to compare potential automatic methods against the current gold standard of human behavior labeling, we set aside all videos of one randomly selected mutant and wild-type mouse for manual screening (4 videos in each of 6 conditions; 24 videos each). Additionally, within the remaining videos, 24 of mutant mice and 24 of wild-type mice were randomly selected for manual screening. Three observers independently marked times at which circling behavior occurred. Observers were instructed to mark complete rotations during bouts of circling behavior but not during normal exploratory behavior shared by wild-type mice (e.g., turning around after reaching the edge of the arena). Videos were not labeled with information regarding mouse genotype. These behavior labels were compared to produce a set of consensus occurrences.

Consensus behavior labels were established by comparing these independently-marked times. To accommodate variations in the precise timing of marked circling occurrences, times within 6 frames (0.1 s) of one another were counted as the same instance, a timeframe chosen to cover 95% of the observed variation between independent observers. If all three observers agreed by this quantitative definition, the average of these matched times was taken as a consensus instance. Independent labels were not subject to modification during this process. Subsequently, circling instances detected by either human observers or automatic methods were counted as true positives if they fell within 0.1 s of a consensus-labeled occurrence.

Algorithm development and parameter search

We compared three algorithms for detecting circling using labeled locations of two keypoints on the body of freely exploring mice. These methods first search the path of the mouse's snout for cases where it crosses over itself as candidates for instances of circling, then apply thresholds to features of the mouse's path to filter out false positives. The 'Duration-Only' method uses only thresholds on minimum and maximum duration. The 'Time-Angle' method additionally excludes candidate circles based on minimum and maximum angular change of the vector from the tailbase to nose. Finally, the 'Box-Angle' method removes the duration constraints and instead considers the tail-nose vector rotation and the lengths of the major and minor axes of a minimum-bounding rectangle fit to the snout path. As described in Table 1 , we thus needed to optimize either 2 or 4 parameters depending on the method being considered. Listed parameter ranges were explored independently for each method rather than the common parameters being frozen as new parameters were added.

The statistics of animal movement can be highly variable. For example, Fig. 3 B illustrates the different distributions of path collision parameters observed in two videos in our training dataset. To ensure robustness against behavioral variability between animals or of the same animal at different times, we selected a method for excluding false positives based on the distribution of duration, total rotation, and size within a given video. Specifically, for a given method applied to a specific video n , Gaussian kernel density estimation k for the parameter(s) $\theta$ of interest is performed on the set of all M collisions detected in that video's tracking data across possible parameter values x .

A combined probability density q consisting of a weighted sum of an exponential with parameter λ and a Gaussian probability density function with mean μ and standard deviation σ, with weights v and w , respectively, is fitted to this density estimate via least-squared-error.

Using the Gaussian component of the fitted distribution, thresholds for rejecting a collision as a circling candidate are then specified in terms of standard deviations above or below the mean.

Lacking an explicit representation of the derivative of F1 score with respect to these threshold parameters, we instead employed constrained Bayesian optimization to identify well-performing parameter combinations 49 . This process first constructs an estimate of the function to be optimized (in our case, F1 score on the training set) based on random sampling of the parameter space, then iteratively explores the location expected the next most exploration-worthy positions in the parameter space and the posterior distributions to find a parameter combination close to the optimal combination. For each parameter combination, we first explore the parameter space via 2000 random samples and then apply Bayesian optimization with 1000 iterations to search for the best parameter combination. Thresholds were constrained to within ± 3 standard deviations. The parameter combination values with the highest train F1 score are reported, evaluated on the test set, and compared with human performance.

Neural network training

We chose to use DLC, an open-source tool for training deep convolutional neural networks to recognize user-labeled image features, to track the locations of mouse-body keypoints due to its accessibility, as it can be straightforwardly used by researchers with little machine learning experience with consumer-grade computing hardware.

As described above, from the 240 free behavior videos originally recorded, 48 were held out as a test set. Four videos unfortunately became corrupted and were not used in the study. From each of the 188 remaining videos not used in manual behavior screening, we labeled 20 random frames with the positions of the mouse's snout and the base of its tail. We utilized data augmentation in the form of the ' imgaug ' dataloader included in DLC, which applies perturbations during network training such as cropping, blurring, and rotating training images. We refer to this as the "Full Dataset'' model, in contrast to models trained on subsets of this data.

To investigate the amount of data necessary to reach a plateau in performance, we used different subsets of our 188-video dataset to train several different DLC models. We compared networks trained with one-half, one-quarter, and one-eighth of the full training dataset (97, 48, and 24 videos, respectively). For each such network, the appropriate number of videos were randomly assigned to a training dataset, with 97 of the remaining videos (50% of the training set) then assigned at random to assess the network's ability to label frames from unseen videos. (Table 2 ) Each model was initialized using a 50-layer pretrained network model (ImageNet-pretrained Resnet50) and trained for 100,000 iterations at a learning rate of 0.001. After training, each DLC network was run on human-scored videos to produce position traces to be analyzed using the Box-Angle method described above.

Statistical analyses

To balance the need to avoid both false positive and false negative errors, we used F1 score to assess three methods of detecting circling behavior in labeled paths, calculated as follows:

Notably, in cases where all three of (True Positives, False Positives, False Negatives) are equal to zero, all these metrics F1 score is undefined. When this occurs, the specific set of scores in question is excluded from p-value but not confidence interval calculations. This is the case on many of the manually-screened behavior videos, in which all three of the human observers marked no instances of circling and well-tuned automatic methods do not score any false positives. P-values for differences between F1 score distributions were calculated using a two-tailed Wilcoxson signed rank tests.

As we could not assume normality of performance distributions a priori, confidence intervals were calculated via bootstrap 50 , i.e., creating a large number of pseudo-datasets of the same size as the original by repeatedly sampling with replacement from the video scores generated by a given method or human labeler. Specifically, this process involves repeatedly drawing N samples with replacement from among the N scores generated by a particular method, individual, or family of DLC models. The F1 score for a particular draw is then calculated from the sum of true positives, false positives, and false negatives for that draw. In all cases where a confidence interval is reported, we sampled one million such pseudo-datasets. Reported intervals are calculated as the 2.5th and 97.5th percentiles of the resulting population of F1 scores.

Data availability

The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.

Segalin, C. et al. The Mouse Action Recognition System (MARS) software pipeline for automated analysis of social behaviors in mice. Elife 10 , e63720. https://doi.org/10.7554/eLife.63720 (2021).

Article CAS PubMed PubMed Central Google Scholar

van den Boom, B. J. G., Pavlidi, P., Wolf, C. J. H., Mooij, A. H. & Willuhn, I. Automated classification of self-grooming in mice using open-source software. J. Neurosci. Methods 289 , 48–56. https://doi.org/10.1016/j.jneumeth.2017.05.026 (2017).

Article PubMed Google Scholar

von Ziegler, L., Sturman, O. & Bohacek, J. Big behavior: Challenges and opportunities in a new era of deep behavior profiling. Neuropsychopharmacology 46 , 33–44. https://doi.org/10.1038/s41386-020-0751-7 (2021).

Article Google Scholar

Mathis, A. et al. DeepLabCut: Markerless pose estimation of user-defined body parts with deep learning. Nat. Neurosci. 21 , 1281–1289 https://doi.org/10.1038/s41593-018-0209-y (2018).

Article CAS PubMed Google Scholar

Ono, K. et al. Retinoic acid degradation shapes zonal development of vestibular organs and sensitivity to transient linear accelerations. Nat. Commun. 11 , 63. https://doi.org/10.1038/s41467-019-13710-4 (2020).

Article ADS CAS PubMed PubMed Central Google Scholar

Chang, H. H. V., Morley, B. J. & Cullen, K. E. Loss of α-9 nicotinic acetylcholine receptor subunit predominantly results in impaired postural stability rather than gaze stability. Front. Cell. Neurosci. 15 , 799752. https://doi.org/10.3389/fncel.2021.799752 (2021).

Ungerstedt, U. & Arbuthnott, G. W. Quantitative recording of rotational behavior in rats after 6-hydroxy-dopamine lesions of the nigrostriatal dopamine system. Brain Res. 24 , 485–493. https://doi.org/10.1016/0006-8993(70)90187-3 (1970).

Hefti, F., Melamed, E., Sahakian, B. J. & Wurtman, R. J. Circling behavior in rats with partial, unilateral nigro-striatal lesions: Effect of amphetamine, apomorphine, and DOPA. Pharmacol. Biochem. Behav. 12 , 185–188. https://doi.org/10.1016/0091-3057(80)90353-6 (1980).

Torello, M. W., Czekajewski, J., Potter, E. A., Kober, K. J. & Fung, Y. K. An automated method for measurement of circling behavior in the mouse. Pharmacol. Biochem. Behav. 19 , 13–17. https://doi.org/10.1016/0091-3057(83)90303-9 (1983).

Bonatz, A. E., Steiner, H. & Huston, J. P. Video image analysis of behavior by microcomputer: Categorization of turning and locomotion after 6-OHDA injection into the substantia nigra. J. Neurosci. Methods 22 , 13–26. https://doi.org/10.1016/0165-0270(87)90084-7 (1987).

Ishiguro, A., Inagaki, M. & Kaga, M. Stereotypic circling behavior in mice with vestibular dysfunction: Asymmetrical effects of intrastriatal microinjection of a dopamine agonist. Int. J. Neurosci. 117 , 1049–1064. https://doi.org/10.1080/00207450600936874 (2007).

Iwamoto, E. T. & Way, E. L. Circling behavior and stereotypy induced by intranigral opiate microinjections. J. Pharmacol. Exp. Ther. 203 , 347–359 (1977).

CAS PubMed Google Scholar

Giuliani, F., Vernay, A., Leuba, G. & Schenk, F. Decreased behavioral impairments in an Alzheimer mice model by interfering with TNF-alpha metabolism. Brain Res. Bull. 80 , 302–308. https://doi.org/10.1016/j.brainresbull.2009.07.009 (2009).

Castillo-Mariqueo, L. & Giménez-Llort, L. Kyphosis and bizarre patterns impair spontaneous gait performance in end-of-life mice with Alzheimer’s disease pathology while gait is preserved in normal aging. Neurosci. Lett. 767 , 136280. https://doi.org/10.1016/j.neulet.2021.136280 (2022).

Portmann, T. et al. Behavioral abnormalities and circuit defects in the basal ganglia of a mouse model of 16p11.2 deletion syndrome. Cell Rep. 7 , 1077–1092. https://doi.org/10.1016/j.celrep.2014.03.036 (2014).

Ryan, B. C., Young, N. B., Crawley, J. N., Bodfish, J. W. & Moy, S. S. Social deficits, stereotypy and early emergence of repetitive behavior in the C58/J inbred mouse strain. Behav. Brain Res. 208 , 178–188. https://doi.org/10.1016/j.bbr.2009.11.031 (2010).

DeLorey, T. M., Sahbaie, P., Hashemi, E., Homanics, G. E. & Clark, J. D. Gabrb3 gene deficient mice exhibit impaired social and exploratory behaviors, deficits in non-selective attention and hypoplasia of cerebellar vermal lobules: A potential model of autism spectrum disorder. Behav. Brain Res. 187 , 207–220. https://doi.org/10.1016/j.bbr.2007.09.009 (2008).

Lee, J. W. et al. Circling mouse: Possible animal model for deafness. Comp. Med. 51 , 550–554 (2001).

Mitchem, K. L. et al. Mutation of the novel gene Tmie results in sensory cell defects in the inner ear of spinner, a mouse model of human hearing loss DFNB6. Hum. Mol. Genet. 11 , 1887–1898. https://doi.org/10.1093/hmg/11.16.1887 (2002).

Article MathSciNet CAS PubMed Google Scholar

Avni, R. et al. Mice with vestibular deficiency display hyperactivity, disorientation, and signs of anxiety. Behav. Brain Res. 202 , 210–217. https://doi.org/10.1016/j.bbr.2009.03.033 (2009).

Wang, W., Chan, E. K., Baron, S., Van De Water, T. & Lufkin, T. Hmx2 homeobox gene control of murine vestibular morphogenesis. Development 128 , 5017–5029. https://doi.org/10.1242/dev.128.24.5017 (2001).

Lv, K. et al. A novel auditory ossicles membrane and the development of conductive hearing loss in Dmp1-null mice. Bone 103 , 39–46. https://doi.org/10.1016/j.bone.2017.06.007 (2017).

Alavizadeh, A. et al. The Wheels mutation in the mouse causes vascular, hindbrain, and inner ear defects. Dev. Biol. 234 , 244–260. https://doi.org/10.1006/dbio.2001.0241 (2001).

Makishima, T. et al. Inner ear dysfunction in caspase-3 deficient mice. BMC Neurosci. 12 , 102. https://doi.org/10.1186/1471-2202-12-102 (2011).

Maskey, D., Pradhan, J., Oh, C. K. & Kim, M. J. Changes in the distribution of calbindin D28-k, parvalbumin, and calretinin in the hippocampus of the circling mouse. Brain Res. 1437 , 58–68. https://doi.org/10.1016/j.brainres.2011.12.009 (2012).

Pan, B. et al. Gene therapy restores auditory and vestibular function in a mouse model of Usher syndrome type 1c. Nat. Biotechnol. 35 , 264–272. https://doi.org/10.1038/nbt.3801 (2017).

Isgrig, K. et al. Gene therapy restores balance and auditory functions in a mouse model of usher syndrome. Mol. Ther. 25 , 780–791. https://doi.org/10.1016/j.ymthe.2017.01.007 (2017).

Nist-Lund, C. A. et al. Improved TMC1 gene therapy restores hearing and balance in mice with genetic inner ear disorders. Nat. Commun. 10 , 236. https://doi.org/10.1038/s41467-018-08264-w (2019).

Zhao, T. et al. Phenotypic differences in the inner ears of CBA/CaJ and C57BL/6J mice carrying missense and single base pair deletion mutations in the Cdh23 gene. J. Neurosci. Res. 99 , 2743–2758. https://doi.org/10.1002/jnr.24905 (2021).

Article ADS CAS PubMed Google Scholar

Rastoldo, G. et al. Quantitative evaluation of a new posturo-locomotor phenotype in a rodent model of acute unilateral vestibulopathy. Front. Neurol. 11 , 505. https://doi.org/10.3389/fneur.2020.00505 (2020).

Article PubMed PubMed Central Google Scholar

Holt, J. R. et al. Putting the pieces together: The hair cell transduction complex. J. Assoc. Res. Otolaryngol. 22 , 601–608. https://doi.org/10.1007/s10162-021-00808-0 (2021).

Wang, X. et al. CIB2 and CIB3 regulate stereocilia maintenance and mechanoelectrical transduction in mouse vestibular hair cells. J. Neurosci. 43 , 3219–3231. https://doi.org/10.1523/jneurosci.1807-22.2023 (2023).

Giese, A. P. J. et al. Complexes of vertebrate TMC1/2 and CIB2/3 proteins form hair-cell mechanotransduction cation channels. bioRxiv https://doi.org/10.1101/2023.05.26.542533 (2023).

Sturman, O. et al. Deep learning-based behavioral analysis reaches human accuracy and is capable of outperforming commercial solutions. Neuropsychopharmacology 45 , 1942–1952. https://doi.org/10.1038/s41386-020-0776-y (2020).

Grill, E. et al. Prevalence, determinants, and consequences of vestibular hypofunction. Results from the KORA-FF4 survey. Front. Neurol. 9 , 1076. https://doi.org/10.3389/fneur.2018.01076 (2018).

Agrawal, Y., Carey, J. P., Della Santina, C. C., Schubert, M. C. & Minor, L. B. Disorders of balance and vestibular function in US adults: Data from the National Health and Nutrition Examination Survey, 2001–2004. Arch. Intern. Med. 169 , 938–944. https://doi.org/10.1001/archinternmed.2009.66 (2009).

Ohlemiller, K. K., Jones, S. M. & Johnson, K. R. Application of mouse models to research in hearing and balance. J. Assoc. Res. Otolaryngol. 17 , 493–523. https://doi.org/10.1007/s10162-016-0589-1 (2016).

Straka, H., Zwergal, A. & Cullen, K. E. Vestibular animal models: Contributions to understanding physiology and disease. J. Neurol. 263 (Suppl 1), S10-23. https://doi.org/10.1007/s00415-015-7909-y (2016).

Giese, A. P. J. et al. CIB2 interacts with TMC1 and TMC2 and is essential for mechanotransduction in auditory hair cells. Nat. Commun. 8 , 43. https://doi.org/10.1038/s41467-017-00061-1 (2017).

Whatley, M. et al. Usher syndrome: Genetics and molecular links of hearing loss and directions for therapy. Front. Genet. 11 , 565216. https://doi.org/10.3389/fgene.2020.565216 (2020).

Alagramam, K. N., Stahl, J. S., Jones, S. M., Pawlowski, K. S. & Wright, C. G. Characterization of vestibular dysfunction in the mouse model for Usher syndrome 1F. J. Assoc. Res. Otolaryngol. 6 , 106–118. https://doi.org/10.1007/s10162-004-5032-3 (2005).

Kikkawa, Y. et al. Mutations in a new scaffold protein Sans cause deafness in Jackson shaker mice. Hum. Mol. Genet. 12 , 453–461. https://doi.org/10.1093/hmg/ddg042 (2003).

Sheppard, K. et al. Stride-level analysis of mouse open field behavior using deep-learning-based pose estimation. Cell Rep. 38 , 110231. https://doi.org/10.1016/j.celrep.2021.110231 (2022).

Vidal, P.-P., Degallaix, L., Josset, P., Gasc, J.-P. & Cullen, K. E. Postural and locomotor control in normal and vestibularly deficient mice. J. Physiol. 559 , 625–638. https://doi.org/10.1113/jphysiol.2004.063883 (2004).

Gabriel, C. J. et al. BehaviorDEPOT is a simple, flexible tool for automated behavioral detection based on markerless pose tracking. Elife 11 , e74314. https://doi.org/10.7554/eLife.74314 (2022).

Lauer, J. et al. Multi-animal pose estimation, identification and tracking with DeepLabCut. Nat. Methods 19 , 496–504. https://doi.org/10.1038/s41592-022-01443-0 (2022).

Blumstein, D. T. et al. Acoustic monitoring in terrestrial environments using microphone arrays: Applications, technological considerations and prospectus. J. Appl. Ecol. 48 , 758–767. https://doi.org/10.1111/j.1365-2664.2011.01993.x (2011).

Kahl, S., Wood, C. M., Eibl, M. & Klinck, H. BirdNET: A deep learning solution for avian diversity monitoring. Ecol. Inform. 61 , 101236. https://doi.org/10.1016/j.ecoinf.2021.101236 (2021).

Gardner J., Kusner M., Zhixiang X., Weinberger K., Cunningham J. Bayesian optimization with inequality constraints. In Proceedings of the 31st International Conference on Machine Learning, Beijing, China (2014).

Carpenter, J. & Bithell, J. Bootstrap confidence intervals: When, which, what? A practical guide for medical statisticians. Stat. Med. 19 , 1141–1164. https://doi.org/10.1002/(SICI)1097-0258(20000515)19:93.0.CO;2-F (2000).

Download references

Acknowledgements

The authors wish to thank Celia Fernandez-Brillet, Olivia Leavitt Brown, Robyn Mildren, Kiara Quinn, and Kantapon Pum Wiboonsaksakul for their insightful feedback on the manuscript, Dale Roberts for assistance with Python, and Ryan Riegel for assistance with recording the initial behavior videos. This work was supported by NIH grants R01DC012564 (ZMA) and 2T32DC000023 (KEC).

Author information

These authors contributed equally: A. Swaminathan and E. Wojahn.

Authors and Affiliations

Department of Biomedical Engineering, Johns Hopkins University, 720 Rutland Ave, Traylor 504, Baltimore, MD, 21205-2109, USA

O. R. Stanley, A. Swaminathan, E. Wojahn, C. Bao & K. E. Cullen

Departments of Otorhinolaryngology-Head and Neck Surgery, Biochemistry and Molecular Biology, Ophthalmology, University of Maryland School of Medicine, Baltimore, MD, USA

Z. M. Ahmed

Departments of Neuroscience, Otolaryngology-Head and Neck Surgery, Johns Hopkins University, Baltimore, MD, USA

K. E. Cullen

You can also search for this author in PubMed Google Scholar

Contributions

O.R.S., A.S., E.W., and K.E.C. designed the study. Z.M.A provided animals and experimental facilities. A.S. and E.W. performed the experiments. O.R.S, A.S, and E.W. served as behavior raters. O.R.S, A.S., and C.B. wrote data analysis code. O.R.S. analyzed the data and prepared the figures and tables. O.R.S and K.E.C. wrote the paper with input from A.S., E.W., C.B., and Z.M.A. All authors reviewed the manuscript.

Corresponding author

Correspondence to K. E. Cullen .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ .

Reprints and permissions

About this article

Cite this article.

Stanley, O.R., Swaminathan, A., Wojahn, E. et al. An open-source tool for automated human-level circling behavior detection. Sci Rep 14 , 20914 (2024). https://doi.org/10.1038/s41598-024-71665-z

Download citation

Received : 15 June 2023

Accepted : 29 August 2024

Published : 08 September 2024

DOI : https://doi.org/10.1038/s41598-024-71665-z

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

Explore articles by subject
Guide to authors
Editorial policies

Along with Stanford news and stories, show me:

Student information
Faculty/Staff information

We want to provide announcements, events, leadership messages and resources that are relevant to you. Your selection is stored in a browser cookie which you can remove at any time using “Clear all personalization” below.

Seeing what’s going on inside a body is never easy. While technologies like CT scans, X-rays, MRIs, and microscopy can provide insights, the images are rarely completely clear and can come with side effects like radiation exposure.

But what if you could apply a substance on the skin, much like a moisturizing cream, and make it transparent, without harming the tissue?

That’s what Stanford scientists have done using an FDA-approved dye that is commonly found in food, among several other light-absorbing molecules that exhibit similar effects. Published in Science on Sept. 5, the research details how rubbing a dye solution on the skin of a mouse in a lab allowed researchers to see, with the naked eye, through the skin to the internal organs, without making an incision. And, just as easily as the transparency happened, it could be reversed.

“As soon as we rinsed and massaged the skin with water, the effect was reversed within minutes,” said Guosong Hong , assistant professor of materials science and engineering and senior author on the paper. “It’s a stunning result.”

Absorption reduces scattering of light

When light waves strike the skin, the tissue scatters them, making it appear opaque and non-transparent to the eye. This scattering effect arises from the difference in the refractive indices of different tissue components, such as water and lipids. Water usually has a much lower refractive index than lipids in the visible spectrum, causing visible light to scatter as it goes through tissue containing both.

To match the refractive indices of different tissue components, the team massaged a solution of red tartrazine – also known as the food dye FD&C Yellow 5 – onto the abdomen, scalp, and hindlimb of a sedated mouse. The skin turned red in color, indicating that much of the blue light had been absorbed due to the presence of this light-absorbing molecule. This increase in absorption altered the refractive index of the water at a different wavelength – in this case, red. As a result of the absorption of the dye, the refractive index of water matches that of lipids in the red spectrum, leading to reduced scattering and making the skin appear more transparent at the red wavelength.

This research is a new application of decades-old equations that can describe the relationship between absorption and refractive index, called the Kramers-Kronig relations. In addition to this food dye, several other light-absorbing molecules have demonstrated similar effects, thereby confirming the generalizability of the underlying physics behind this phenomenon.

Researchers were able to see, without special equipment, the functioning internal organs, including the liver, small intestine, cecum, and bladder. They were also able to visualize blood flow in the brain and the fine structures of muscle fibers in the limb. The mouse’s beating heart and active respiratory system indicated that transparency was successfully achieved in live animals. Furthermore, the dye didn’t permanently alter the subject’s skin, and the transparency disappeared as soon as the dye was rinsed with water.

The researchers believe this is the first non-invasive approach to achieving visibility of a mouse’s living internal organs.

“Stanford is the perfect place for such a multifaceted project that brings together experts in materials science, neuroscience, biology, applied physics, and optics,” said Mark Brongersma , professor of materials science and engineering and co-author on the paper. “Each discipline comes with its own language. Guosong and I enjoyed taking each other’s courses on neuroscience and nanophotonics to better appreciate all the exciting opportunities.”

The potential future of ‘clear’ tissue

Right now, the study has only been conducted on an animal. If the same technique could be translated to humans, it could provide a range of biological, diagnostic, and even cosmetic benefits, Hong said.

For example, instead of through invasive biopsies, melanoma testing could be done by looking directly at a person’s tissue without removing it. This approach could potentially also replace some X-rays and CT scans, and make blood draws less painful by helping phlebotomists easily find veins. It could also improve services like laser tattoo removal by helping to focus laser beams precisely where the pigment is below the skin.

“This could have an impact on health care and prevent people from undergoing invasive kinds of testing,” said Hong. “If we could just look at what’s going on under the skin instead of cutting into it, or using radiation to get a less than clear look, we could change the way we see the human body.”

For more information

Other Stanford co-authors include Betty Cai, member of the Department of Materials Science Engineering ; Zihao Ou, Carl H. C. Keck, Shan Jiang, Kenneth Brinson Jr, Su Zhao, Elizabeth L. Schmidt, Xiang Wu, Fan Yang, Han Cui, and Shifu Wu, who are also with the Department of Materials Science Engineering and Wu Tsai Neurosciences Institute ; Yi-Shiou Duh of the Department of Physics and Geballe Laboratory for Advanced Materials ; Nicholas J. Rommelfanger of the Wu Tsai Neurosciences Institute and Department of Applied Physics ; Wei Qi and Xiaoke Chen of the Department of Biology ; Adarsh Tantry of the Wu Tsai Neurosciences Institute and Neurosciences IDP Graduate program ; Richard Roth of the Department of Neurosurgery ; Jun Ding of the Department of Neurosurgery and Department of Neurology and Neurological Sciences ; and Julia A. Kaltschmidt of the Wu Tsai Neurosciences Institute and Department of Neurosurgery .

This work was supported by the National Institutes of Health, National Science Foundation, Air Force, Beckman Technology, Rita Allen Foundation, Focused Ultrasound Foundation, Spinal Muscular Atrophy Foundation, Pinetops Foundation, Bio-X Initiative of Stanford University, Wu Tsai Neuroscience Institute, Knight-Hennessy, and U.S. Army Long Term Health Education and Training program.

Disclaimer: The technique described above has not been tested on humans. Dyes may be harmful. Always exercise caution when handling dyes – do not consume them, apply them to people or animals, or misuse them in any way.

Media contacts:

Jill Wu, School of Engineering: [email protected]

IMAGES

Bioprinting human ears inside living mice—all without a single surgical
The Vacanti mouse was a laboratory mouse created by Charles A. Vacanti
Mouse grown successfully on human ear
A mouse with a human ear growing out of its back
Stressed Lab Rat Breaking Out In Human Ears
| Mouse surgical preparation and intravital imaging setups. (A) Ear

VIDEO

putting my ear mouse
The Worst Mouse Experiment Ever?
25 mouse Experiment #shorts
Bio lab Mouse experiment.. Mouse shaheed ho gya 🙂#biolab #mouse #iiser #neet #lab #iat #iisc #viral
The Universe 25 Experiment
Universe 25: The Great Mouse Experiment

COMMENTS

Vacanti mouse
The Vacanti mouse. The Vacanti mouse was a laboratory mouse (circa 1996)[1] that had what looked like a human ear grown on its back. The "ear" was actually an ear-shaped cartilage structure grown by seeding cow cartilage cells into biodegradable ear-shaped mold and then implanted under the skin of the mouse, with an external ear-shaped splint ...
Scientists Behind the Lab Mouse With a Human Ear Speaks
They implanted the shape of a human ear in the back of a mouse as part of research to better understand how they could help grow body parts for humans. They published their results in 1997. After ...
The History of the Lab Rat Is Full of Scientific Triumphs and Ethical
The History of the Lab Rat Is Full of Scientific Triumphs and ...
Mouse with human ear › Dr Karl's Great Moments In Science (ABC Science)
The "mouse-ear" project began in 1989, when Charles Vacanti (brother of Joseph) managed to grow a small piece of human cartilage on a biodegradable scaffold. The scaffold was the same synthetic ...
The Mouse That Changed the World of Science
The synthetic ear cartilage was surgically placed on the mouse's back and left there for 12 weeks until it had fully developed with living cells. The ear came close to being 90% similar to a natural human ear which was something very surprising taking into consideration that this experiment didn't involve any genetic engineering with human DNA.
30 years of tissue engineering, what has been achieved?
30 years of tissue engineering, what has been achieved?
The Mouse with an Ear Growing out of its Back
In 1997, the scientific world got its first glimpse of the ' Vacanti mouse ' - an eerie, hairless mouse with a human-sized ear growing out of its back. Despite the public controversy that the experiment initially faced over animal ethics, the techniques used by the researchers to achieve this feat brought medical research one step closer to an exciting goal: generating organs in the lab.
Vacanti mouse
The Vacanti mouse was a laboratory mouse (circa 1996) [1] who had what looked like a human ear grown on their back. The "ear" was actually an ear-shaped cartilage structure grown by seeding cow cartilage cells into biodegradable ear-shaped mold and then implanted under the skin of the mouse, with an external ear-shaped splint to maintain the desired shape.
4 The Vacanti mouse -a human ear grown on mouse? In 1997, Vacanti C. et
In 1997, Vacanti C. et al. reported on the successful growth of human ear-like structure on the back of a mouse by implanting synthetic biodegradable polymer embedded with bovine-derived cartilage ...
The auriculosaurus. The ''human-ear-bearing'' mouse developed by the
The ''human-ear-bearing'' mouse developed by the Vacanti laboratory that went onto epitomize Tissue Engineering. ... Running design experiments using fungi helps to understand the extent ...
Human-like ears 3D-printed inside mice as surgery-free spare parts
The famous Vacanti mouse of the 1990s also had a human-like ear grown on its back, but it was made by implanting a pre-made plastic scaffold seeded with cartilage cells underneath the skin, rather ...
Charles Vacanti
Charles Alfred "Chuck" [1] Vacanti (born 1950) is a researcher in tissue engineering [2] and stem cells and the Vandam/Covino Professor of Anesthesiology, Emeritus, at Harvard Medical School. [3] He is a former head of the Department of Anesthesiology at the University of Massachusetts and Brigham and Women's Hospital, now retired.. He is known for the Vacanti mouse, a mouse created with ...
HUMAN EAR GROWN ON MOUSE
The mouse, specially bred to lack an immune system that might reject the human tissue, nourished the ear as the cartilage cells grew to replace the fiber. The mouse remains healthy and alive after the ear is removed, the researchers said. ``You end up with a piece of cartilage in the shape of an ear,'' Griffith-Cima said.
MOUSE GROWING HUMAN EAR IS TISSUE TRIUMPH
The mouse, specially bred to lack an immune system that might reject the human tissue, nourished the ear as the cartilage cells grew to replace the fiber. The mouse remains healthy and alive after the ear is removed, the researchers said. "You end up with a piece of cartilage in the shape of an ear," Griffith-Cima said.
Daily MOS: The Vacanti Mouse
For purposes of this experiment, it would show proof of concept, that the bioscaffolds could support growth of a cartilage structure in a non-human animal. Did they have to do it this way? Probably not. Did it get the desired effect? Absolutely. The mouse had a goddamn ear growing on its back made of cow. People lost their fucking minds.
How the mouse with human ears changed the world?
Home Medicine - Health Health. Twenty years ago, a mouse with a human ear on its body caused waves of anger and criticism, but the reality later proved the value of this experiment. During the 20th century, mankind witnessed great advances in the field of science. It is known that, as technology advances, medicine will advance to the point ...
Why Scientists Put An Ear On A Mouse
The "Vacanti Mouse." The "earmouse." That freaky thing you saw in a biology textbook or email chain. Whatever you call it, one thing's for sure: the mouse with the ear on its back is an icon of science, and it has been for more than 20 years.. Maybe it even represents mad science. However, many misunderstand how and why the mouse was created in the first place.
Who is Vacanti mouse? with a human ear growing on his back
In order to better understand how they could assist humans to grow body parts, they implanted the shape of a human ear in the back of a mouse. Their findings were published in 1997. Following the ...
It's Possible to Grow a 3-D Printed Ear on a Mouse's Back
In this experiment the researchers inserted the ear beneath the skin of a mouse's back and found that several months later the rodent created blood vessels that attached to the printed ear and ...
Human ear grown on the back of a rat
Japanese scientists have grown a human ear on the back of a rat in order to help children born with facial abnormalities and adults who have suffered accidents. The ear was grown by turning stem cells into cartilage cells which were placed in inside plastic tubes shaped like a human ear on the rat's back. The framework dissolved after two ...
The Mouse That Changed Science
The Mouse That Changed Science
Mouse with a Human Ear
The mouse used in this experiment was referred to as a "Nude Mouse", since it lacked hair due to a random mutation. This same mutation also compromised the mouse's immune system — which is ...
Mouse screen reveals multiple new genes underlying mouse and human
As mouse and human inner ears are very similar in structure and function (e.g., ), ... and the experiments conformed to the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report. The US patient provided consent for clinical whole exome analysis and written consent for inclusion as a ...
An open-source tool for automated human-level circling behavior
Mitchem, K. L. et al. Mutation of the novel gene Tmie results in sensory cell defects in the inner ear of spinner, a mouse model of human hearing loss DFNB6. Hum. Mol.
Researchers make mouse skin transparent using a common food dye
Published in Science on Sept. 5, the research details how rubbing a dye solution on the skin of a mouse in a lab allowed researchers to see, with the naked eye, through the skin to the internal ...

The History of the Lab Rat Is Full of Scientific Triumphs and Ethical Quandaries

This site is being redeveloped. For all the latest ABC Science content click here .

Mouse with human ear

Related Stories

Subscribe to the Great Moments in Science Podcast

You might also be interested in Science and Technology

Follow ABC Science

About | Contact | ABC Science

More science sites

Related ABC sites

Search ABC Science

Stay updated

Follow ABC Science on Twitter

Human Atlas

The Mouse with an Ear Growing out of its Back

Extracting Chondrocyte Cells from Cows

Building a Scaffold

Implantation of the Tissue-Polymer Construct

Applications of Tissue Engineering Techniques

The auriculosaurus. The ‘‘human-ear-bearing’’ mouse developed by the Vacanti laboratory that went onto epitomize Tissue Engineering. License to reproduce the image was obtained from the a (2002) BBC Photo Library. Color images available online at www.liebertpub.com/teb

Contexts in source publication

Human-like ears 3D-printed inside mice as surgery-free spare parts

Read more: Children get new ears grown from their own cells in world first

Sign up to our weekly newsletter

More from New Scientist

3D-printed egg shells can help bones regrow

Tiny brain sensor implanted without surgery dissolves after weeks

Bioelectronic patch uses living bacteria to treat psoriasis in mice

Swallowable sensor unfurls in stomach to monitor gut health

MOUSE GROWING HUMAN EAR IS TISSUE TRIUMPH

Come for the science, stay for the dirty jokes.

Share this:

Join the discussion! Cancel reply

Discover more from Scibabe

How the mouse with human ears changed the world?

Why Scientists Put An Ear On A Mouse

Why Scientists Put an Ear on a Mouse

Shaping the Scaffolding

Rumors and Hoaxes

Helping Humans Today

About The Author

Meet Bandit, the Turtle Who's Stealing Our Hearts!

10 Most Underrated Animals at Ripley’s Aquarium of Canada

The Mysterious World of Jellies

More on Animals

Read More Ripley's

Cartoon of the Day

It’s Possible to Grow a 3-D Printed Ear on a Mouse’s Back

Human ear grown on the back of a rat

Related articles

Could AI replace animal research?

Openness in Animal Research Conference Returns in 2026

New fact-checking video series from UAR

Distillations podcast

Research Notes

Archival Sources

The Mouse that Changed Science

Listen to more episodes

ALS Patients Take on the FDA

The Fraud That Transformed Psychiatry

Cancer Virus Hunters: An Interview with Gregory J. Morgan

Never Miss an Episode

Mouse screen reveals multiple new genes underlying mouse and human hearing loss

Author summary

Introduction

Results and discussion

Role in human auditory thresholds

Broad range of gene functions

Progressive hearing loss

Varied pathophysiology underlying deafness

Other features associated with deafness in the new mutants

Mutant lines with normal thresholds but abnormal waveforms

Conclusions

Materials and methods

Generation of mice

Phenotyping pipelines

ABR recordings

Data analysis

Significance calls on mutant data

Quality control.