How ‘Super Smellers’ and Artificial Noses Are Sniffing Out Disease

Illustration by Janice Chang

Last month, a 65-year-old retired Scottish nurse named Joy Milne created a media frenzy when she was seemingly able to diagnose Parkinson’s disease using her sense of smell even earlier than scientists could diagnose it with regular medical technology. Milne’s gift apparently manifested several years ago, when she started smelling a strange, musky odor coming from her husband, which she initially dismissed as the normal stench of an overworked man. But after doctors diagnosed Milne’s husband with Parkinson’s disease years later and she started to attend meetings with Parkinson’s patients, she noticed the same odor wafting off of others afflicted with the disease. When she mentioned her observation in passing to a cadre of scientists, they decided to put her ability to the test, giving Milne six shirts worn by Parkinson’s patients and six by healthy individuals. Amazingly, she managed to smell a correct diagnosis (they thought) in 11 of 12 cases—though even the one false positive she’d made came from a man who manifested Parkinson’s eight months after the small-scale test.

Milne’s 100 percent accuracy and eerie predictive powers have inspired a new line of Parkinson’s research, geared toward identifying other “super smellers” and using their gifts to isolate the compounds coming out of human skin that seem to presage the disease. Parkinson’s is actually tough to diagnose, so developing an early diagnosis method that could be completed using only a skin swab and a human nose or mechanical substitute could go a long way in advancing research and treating earlier stages of the disease.

That’s a promise worth celebrating. But it’s not as unique as the reporting on Milne’s talented nose might have implied. We’ve known for ages that many (if not all) diseases have a smell, manifested in our breath or sweat and caused by illness-triggered changes in our body chemistry. Most of us can’t detect them until the ailments are already long diagnosable (save for super smellers like Milne; others claim they have been able to smell people’s cancer before it manifested). Still, this longstanding knowledge has inspired a venerable literature on how we might use scent to make earlier or stronger diagnoses for a host of disorders. Some call the tiny field metabolomics, the study of human chemical scent patterns as they relate to illness. This research has been stunted for years by the lack of diagnostic tools more reliable than fickle and fallible noses, human or animal. But in recent years we’ve made huge technological advances that may soon yield reliable, supersensitive artificial noses, unlocking the potential of smell. This olfactory development could lead to far less invasive and earlier diagnoses than we currently have, all ideally cheap and accessible as well.

Doctors have also historically used smell to make diagnoses. Diabetes, when untreated, famously causes a patient’s breath to smell of rotting apples. Some diseases like fish odor syndrome or maple syrup disease are clearly named for and primarily known by the smells that they create (in the first instance by an inability to digest choline in foods and in the second case by a failure of the metabolic system to process certain amino acids). And clinic workers who spend their entire lives around individual diseases have catalogued a host of unique scents: Cystic fibrosis leads to acidic breath. Diphtheria smells sweet, some tuberculosis smells like beer, and typhoid fever smells like baking bread. Strep-throat breath smells metallic, yellow fever smells like a butcher’s shop, and liver failure smells like raw fish. Cholera, pneumonia, schizophrenia, smallpox—everything has a scent. Yet while smell is commonly used to diagnose a few diseases (and some veterinarians still use smell to noninvasively diagnose cats and dogs who can’t talk about their symptoms), the human nose is usually too weak to discern this data without long habituation. And even then, smell-based information on diseases usually comes through to an average human nose too late to be of use.

In the past, scientists eager to explore smell’s potential have turned to dogs. With olfactory senses that are 10,000 to 100,000 times more powerful than our own (regardless of breed), the hope is that they might be able to smell the earliest twinges of disease in humans. Based on anecdotes of dogs nipping at melanomas before they manifested to the human eye, some scientists long suspected that, as with Milne and Parkinson’s, these chemical developments would be detectable via smell to dogs long before they’re detectable to modern medical diagnostics. And to date, research on dog-smell-based disease diagnostics holds up those hopes and assumptions: A series of studies over the past couple of decades have shown that dogs trained to sniff out specific cancers can do so correctly quite often—90 percent of the time for ovarian cancer, 95 percent for colorectal via patient breath and 98 percent via stool, and 99 percent for lung cancer. At worst, dogs were able to identify bladder cancer 60 percent of the time by the scent of urine samples, and breast cancer 88 percent of the time. Although small studies, these dog tests support the idea of smell’s diagnostic promise.

Image by Pixabay user Snufkin

Yet even if we can train dogs to make early smell-based diagnoses with some reliability (at time equal to or greater than that of more invasive diagnostics) we can’t depend on them as a tool. Training dogs to smell a single disease at a time is a huge investment of effort and energy and could have uneven accuracy from dog to dog, depending on their noses or temperaments. Dogs may still have potential as cheap screening tools for large groups of people, especially in regions with minimal medical facilities, sniffing out individuals for more invasive and traditional examinations. And they can already reliably act as service dogs for people with diabetes or seizures, alerting their owners to impending attacks. But as a surefire way to increase the accuracy and early detection capacities of medical diagnoses at the single-patient level, they fall short of our needs for consistency and reliability.

Some researchers have tried to study other animals that might be cheaper and more reliable to train and employ. A German team, for example, thinks you can detect patterns in the antennae movements of fruit flies exposed to cancerous tissues. But rather than rely on transient and fallible organic entities, the ideal for the medical field would be to develop electronic noses, capable of smelling the same changes as a dog or super smeller via a cheap, dedicated machine that every doctor can use with confidence. (The data yielded from studies on the smells Milne and other super smellers identify would ideally be used to program such a machine, which would detect Parkinson’s disease after doctors feed it a skin swab to digitally “smell.”)

You might think it’d be pretty easy to develop an electronic nose. But you would be wrong. We do have machines that can detect the presence of particular chemicals in the air or in a substance at particular levels. But most of the chemical changes we’re talking about occur in minuscule parts per trillion and are part of a whole constellation of organic compounds that make up human odor. Detecting an entire mixture of compounds, especially amid the unique, ever-changing smell print of an individual human whose biochemistry can be influenced by a ton of confounding factors, is a difficult task. Most existing machines capable of doing that sort of detection are massive and incredibly expensive multipurpose rigs. In the face of that reality, the prospect of creating marketable, manageable devices geared toward the same ends, or even toward individual diseases, has long seemed a bit out of reach to many medical professionals.

These hurdles haven’t stopped scientists from trying, though. For the past couple of years it seems like there’s been a boom in research on experimental robotic noses, many of which have started to yield promising results. Basic mechanical noses have been tested for their ability to sniff out numerous types of cancer, organ failures, gut bacteria disorders, and sundry other diseases. Some seem to have about the level of sensitivity of a dog’s nose (without their organic shortcomings), and a few new devices seem to show dog-like effectiveness in detecting bowel, breast, and lung cancer. There’s also research evolving on the potential of artificial noses to sniff out tuberculosis and heart disease in remote settings.

Artificial noses are years off from perfection and widespread distribution. But they seem on track to unlock a whole new realm of diagnostics, the potential of which people like Milne and various dog studies have pointed to for years. The early alert power of smells could be extremely important for research, allowing us to try new and (until recently) unimaginable types of interventions. The noninvasive, cheap, and reliable nature of smell diagnoses could allow more people to get tested, improving our ability to care for people, and ideally lowering the mortality of all sorts of diseases. The portability of the technology means that it could be of use all around the world, especially in places where more invasive tests are hard to perform. The promises of smell for medicine are vast and so tantalizingly close that you can—well, you can almost smell it.