Could rats and dogs detect disease better than the finest lab equipment?

Seven days a week, a team of motorcycle couriers fan out to 80 clinics across several regions of Tanzania. They pick up sputum samples taken from people with symptoms of tuberculosis (TB) and deliver them to a central laboratory in Dar Es Salaam run by the Belgian non-profit organization APOPO. There, the samples are heat inactivated to kill the bacterium that causes TB before a team of analysts examines the samples for evidence of the disease.

Part of Nature Outlook: Medical diagnostics

This work, and similar programmes in Mozambique and Ethiopia, typically increase TB detection rates by about 40% compared with those of local clinics. But the highly trained sample analysts in APOPO’s labs that are responsible for this success probably don’t care. They are not scientists or physicians working out of humanitarian concern or for scientific glory — they are African giant pouched rats (Cricetomys ansorgei), and they identify TB-positive samples by smell. Occasional sips of a banana and avocado smoothie are all the motivation they need.

The rats — which Tefera Agizew, head of APOPO’s TB programme, thinks could help to eradicate the disease — are part of a trend that has been unfolding over the past two decades in which animals are used to detect pathogens and diseases as varied as lung cancer, malaria and COVID-19. These efforts have progressed from anecdotal reports and proof-of-concept studies to more advanced trials and even some clinical applications.

The rats offer several benefits over searching for the TB bacterium in sputum under a microscope. First, the animals’ noses are more sensitive at identifying TB. Moreover, they work fast, easily evaluating 100 samples in just 20 minutes. “This might take four days for a lab technician,” Agizew says. Rat diagnosis is also very cheap, he adds, ringing in at just over US$1 per sample.

But developing animal-based diagnostics is far from straightforward. Although the latest studies are fortifying initial suggestions about animals’ astonishing olfactory capabilities, they are also underlining how tricky it is to scale up animal-based diagnostics and roll them out in real-world environments.

Signatures scents

The idea of animals sniffing out disease might sound unlikely, but the underlying mechanism is well established. The body produces an array of volatile organic compounds (VOCs) that escape through the skin, in body fluids and in breath, wafting through the air until they encounter an odour receptor in a nose that is sensitive enough to perceive them.

All kinds of disease can alter which VOCs are present and at what concentration. Virtually everything that goes wrong in the body “leaves a mineable volatile trace”, says Andreas Mershin, chief scientific officer at RealNose, a start-up firm in Boston, Massachusetts, that is developing an electronic nose to diagnose prostate cancer and eventually other diseases. “The world of molecules is so rich in information.”

In fact, smell has long been a clue to disease diagnosis. The scent of acetone on a person’s breath heralds diabetic ketoacidosis, and the aroma of rotting flesh emanating from a wounded limb is an unmistakable sign of gangrene. The African giant pouched rats aren’t even the first to identify a distinctive smell associated with TB. In Dutch, the disease is called tering, which refers to the tar-like smell that physicians noticed on the bodies of their patients.

But many animals have a much keener sense of smell than do humans. Canine noses, for example, can be as much as 100,000 times more sensitive than those of humans. Dogs have 200 million or more olfactory receptors in their noses compared with human’s mere 5 million, and they have proportionally larger brain areas devoted to olfactory processing.

APOPO began as a programme to use rats to help detect land mines; the animals have helped to make more than 240 square kilometres of land in former conflict zones safe for local inhabitants. But sniffing out a disease is much more challenging than finding explosives or contraband, even for the most sensitive animal noses, says Cynthia Otto, director of the University of Pennsylvania’s Penn Vet Working Dog Center in Philadelphia. “It’s a minute difference in a really noisy background,” she says, likening the dogs’ task to a particularly challenging game of puzzle book Where’s Wally.

How exactly animals respond to a disease’s chemical calling card is not well understood. Researchers have searched for VOC signatures of various diseases in body products such as sweat and urine, but these investigations do not always yield clear results. The APOPO team, for example, identified 13 VOCs that are unique to TB¹. When any one of them is presented to rats in isolation, the animals do not recognize it as the smell of TB. Nor do they respond to combinations of three such molecules. But mix at least six of the molecules together — seemingly in any combination — and the rats identify the smell of TB that they have been trained to expect will lead to a banana-and-avocado-smoothie reward.

Why six is the magic number isn’t yet known, but some people argue that it doesn’t matter. “A lot of people are like, ‘You can’t train a dog if you don’t know what the chemical is.’ And the answer is, ‘oh yes we can’”, Otto says. “The dogs don’t care what the chemical is — they can figure out the pattern.”

Canine cancer finders

Anecdotal reports of animals detecting disease emerged in 1989 in the form of dogs who repeatedly sniffed at or licked spots on their owners’ bodies that turned out to be skin cancers. The first formal scientific investigation appeared in 2004, when researchers demonstrated that dogs could consistently detect bladder cancer by smell². Two of the researchers involved in that study founded Medical Detection Dogs in Milton Keynes, UK, in 2008, one of the earliest organizations devoted to the effort. Several years later, one of the detection dogs — a Labrador retriever named Daisy — alerted co-founder Claire Guest to an aggressive tumour in her breast. It was caught early and treated successfully.

A diverse bouquet of such studies has now shown that dogs, rats and even invertebrates such as locusts, ants and roundworms can identify the smell of infectious diseases — TB, methicillin-resistant Staphylococcus aureus (MRSA), malaria, urinary tract infections, Clostridium difficile, COVID-19 and Helicobacter pylori have all been sniffed out. Many other cancers have also proved to be detectable by animals in a person’s urine, breath, saliva, skin secretions, cell cultures, faeces, blood and tissue samples.

In a 2019 study, researchers in France asked 36 women with breast cancer and 51 healthy volunteers to sleep with absorbent pads against their breasts³. They then delivered the pads to a lab, in which two dogs were trained to pick out one sample from a woman with cancer from a group of four — similar to a police line-up. The dogs could smell the signature of the disease in the women’s sweat 90% of the time.

Such a procedure could be used as a screening tool that is less uncomfortable than mammography, or even instead of breast imaging in low-resource areas, says Isabelle Fromantin, a nurse at the Curie Institute in Paris and head of the KDOG cancer detection project, which aims to develop dog-based breast cancer screening.

But moving to the clinic has proved difficult. In a screening situation, many ‘line-ups’ will consist entirely of cancer-free samples, says Michelle Leemans, who worked on the KDOG project and is now a postdoctoral researcher at the French National Institute of Health and Medical Research (INSERM) in Créteil. “This increases the difficulty of the exercise for the dogs,” she says. Keeping them motivated to give an accurate assessment when only the positive samples, which are few and far between, get them a reward is a challenge. Unpublished data from a larger KDOG trial suggest that the dogs struggle in these real-world conditions.

Researchers and handlers must also make sure they don’t give the animals inadvertent clues. The KDOG researchers have used colour-coded cups to keep track of positive and negative samples, which they keep secret from the dogs in an adjoining room. But one of the dogs learnt that he could spot the cups through a window between the rooms if he stood on his hind legs and propped his front paws on a person. “We think, ‘Oh, this is a nice dog. It wants to see me and to say hello’,” says Fromantin. “No — it wants to see the positive sample.” Researchers including Fromantin and Leeman conducted a review of more than 60 studies of animal-disease detection and found that only a handful used rigorous blinding procedures to avoid similar pitfalls⁴.

Efforts to put rats to work are, in some ways, more developed. A 2021 meta-analysis⁵ of seven rat-based TB detection studies pegged the true positive rate at 81%, and the avoidance of false positives at 73%. “Rats are much more content to do repetitive work than dogs are, so they tend to stay much more reliable throughout the session and across days,” says Cindy Fast, APOPO’s head of training and behavioural research. In addition, because the rats are smaller and cheaper to care for than dogs are, the organization can have teams of 5–10 rodents evaluate each sample, boosting the reliability of the results.

Between 2013 and 2023, the programme screened nearly one million sputum samples and identified more than 30,000 cases of TB that had been missed by local clinics. The organization estimates that the rats have helped to prevent almost 364,000 follow-on infections in family members or work colleagues of the people who were found to have TB.

In a bid to further improve the reliability of the test, APOPO has developed automated cages so that the rats evaluate samples without any interaction with humans. The organization hopes that this strategy will help in its effort to gain accreditation from the World Health Organization; until then, the method will remain a second-line screening technology to be followed up by conventional laboratory methods.

Doing without the dog

One major obstacle to the scale-up of animal-assisted diagnostics is the limited supply of detection animals — especially dogs. It takes time and resources to train them, leading to concern that there won’t be enough for widespread cancer detection and screening.

The KDOG researchers are trying to pinpoint the VOC signature of breast cancer, with the aim of producing synthetic odour samples. These could provide an easily accessible, uniform source of samples to train the dogs with, and therefore boost the feasibility of the dog programme.

Other researchers, however, are hoping to do without the animals altogether. “Let’s get rid of the dog packaging, bring it into our technology,” says Mershin of RealNose.

Various teams have sought to translate the chemical signatures of a range of diseases into VOC-based diagnostic tests, including several ongoing efforts to develop a breath test for lung cancer. These aim to analyse the ingredients of a sample. But, Mershin instead wants to “create machines that understand what something smells of, not what something is made of” — the same way that the aroma of coffee or a new car is instantly identifiable to many people without them knowing the molecular sources of the scents. “We are working in perceptual space as opposed to analytical chemistry space,” Mershin says.

RealNose’s concept uses copies of mammalian olfactory receptor proteins attached to a circuit board to detect odours, and machine-learning algorithms to interpret them. In a study⁶ published in 2021, Mershin — then a physicist at the Massachusetts Institute of Technology in Cambridge — collaborated with Medical Detection Dogs and scientists from several other organizations and universities to use dogs to help develop a prostate-cancer test. Their investigation examined 50 urine samples. Some of these were from people with prostate cancer, and some were from people without the disease. The researchers used the dogs’ diagnoses to train an artificial neural network to find patterns in the data from positive samples recorded by gas chromatography–mass spectrometry. These patterns, rather than the presence or absence of biomarkers, represent the signature ‘smell’ of prostate cancer that the dogs perceive intuitively.

RealNose secured just over $1 million in venture capital funding in early 2024 — enough to begin to build a prototype of the electronic nose and validate it over the course of a year on about 100 prostate cancer samples.

Life in the old dog yet

Not everyone is willing to give up on animals just yet, however. For some researchers, the benefits of the ‘dog packaging’ outweigh the problems it presents.

That’s the logic behind the Canines for Care programme at Vancouver Coastal Health, an organization in Canada, at which detection dogs have been identifying sources of C. difficile contamination in hospitals since 2016. C. difficile is the most common cause of hospital-acquired gastroenteritis. In immunocompromised and older people, the infection can be serious and even fatal.

“There is no logistically feasible way” to search an area as large as a hospital room for C. difficile contamination without the dogs, says Teresa Zurberg, the programme’s co-founder and canine-detection specialist. It would require laboriously swabbing the walls, then waiting a week for samples to grow out on culture plates. “I can take a dog to that same patient room, and within five minutes they can screen that whole patient room,” she says.

In a 2017 study⁷, the Canines for Care team reported that the dogs had correctly identified all of the C. difficile positive samples hidden in a hospital ward. They also flagged 83 potential locations of C. difficile contamination in a clinical unit over 49 working days. “We approach this just like we would any laboratory evaluation, with the same scientific rigour,” says microbiologist and Canines for Care co-founder Elizabeth Bryce. “Did the dog, just like a lab test, have the required sensitivity and specificity? Was it reliable?”

Like humans, dogs are typically only vulnerable to infection with the bacterium if they are immunocompromised, so there’s little risk to healthy working dogs. And when they do find contamination, the dogs don’t usually trigger the defensive response from the staff that an admonition from a supervisor might. “Everybody loves the dogs,” Bryce says.

The love goes both ways. This year, Otto’s team launched an effort to give back to the dogs, by training them to detect haemangiosarcoma: a stealthy cancer that is common in working-dog breeds and has taken the lives of several animals in the Penn Vet programme. “It was just like, ‘Hey, we should do this’,” she says, her affection for the animals clear in her voice. These dogs that are training to keep people healthy might soon be able to look out for each other too.