Sniffing out cancer


We've know for some time that dogs can detect the scent of diseases including cancer, but new machines, some taking their lead from the canines, may hold the best hopes for dramatically improved early detection rates. Kate Ravilious and Elizabeth Finkel report.


Dogs trained over five years show they correctly sniff out bladder cancer in urine samples 93% of the time. – Corbis

Sometimes life dishes out cruel ironies. In 2007, Claire Guest co-founded a program to train dogs to detect cancer in humans. One of the more enjoyable tasks was taking her ragtag bunch of Labradors, spaniels and retrievers to a Buckinghamshire park for a run. On this particular day, however, she opened her car to release her eager charges and found that her star pupil Daisy wasn’t interested in exercise. Instead, the Golden Labrador leapt at her chest. It caused a deep pain, like a bruise, she recalls. When she pressed her fingers into the area she felt a lump. A trip to the doctor and several scans later revealed a deep-seated malignant tumour, which was quickly removed.

Guest says Daisy had been acting anxiously around her for a couple of weeks. “You can say Daisy’s behaviour was a mere coincidence but the fact remains that it’s because of her I found my cancer,” she says.

Guest, a behavioural psychologist specialising in human-dog relationships, is accustomed to raised eyebrows. “Scepticism is not something I shy away from.” She acknowledges the miasma of shaggy dog stories and that anecdotes are not science. Which is why she founded “Medical Detection Dogs”. It’s a curious institution with two missions. One is to train dogs to detect the odour of illness. It could be a diabetic whose blood sugar is falling dangerously low or a person with Addison’s disease whose cortisone levels are plummeting, or someone with a lurking cancer. The other is to publish scientific papers on the results of that training.

Claire Guest, who founded the charity Medical Detection Dogs, with Daisy. – Medical Detection Dogs

“I didn’t want dog-whispering; if this was going to be integral to medicine it had to be evidence-based,” says Guest. “But we now have too much data to ignore. I haven’t had one person who’s seen the data, go away and not think there’s some value in this work.”

So far the results are impressive. Starting from modestly encouraging results published in the British Medical Journal in 2004, where dogs got cancer samples right almost half the time, the latest results on dogs trained over five years show they correctly sniff out bladder cancer in urine samples 93% of the time. Guest says she is seeing similar results with prostate cancer. By contrast, the current means of screening for this cancer by looking for the Prostate Specific Antigen is considered little better than a coin toss.

There is no doubt dogs are extraordinarily gifted when it comes to detecting smells. Their noses are up to 10,000 times more sensitive than our own. But they are also capable of being trained to home in on the signature of a particular scent. At Medical Detection Dogs, the puppies are presented with a set of eight urine samples. If after sniffing each one purposefully they sit down next to the sample from a patient with bladder cancer, they are rewarded. After years of training, their brains have become exquisitely tuned. “The dog is much more than a machine; it’s doing some intelligent processing,” says Guest.

‘I didn’t want dog-whispering. If this was going to be integral to
medicine it had to be evidence-based’

She has some high profile supporters. “The data from many groups now is beyond statistical doubt,” says Hossam Haick, a Professor of Chemical Engineering and Nanotechnology from Israel’s Technion Institute of Technology, and inventor of an electronic nose known as “NA NOSE”. But it is hard to see dogs being rolled out into hospitals for mass screening. For that you need devices. And in the past decade, devices that can detect cancer by analysing a patient’s breath, much like a breathalyser detects blood alcohol, are raising hopes of a new era in screening. These devices have proven their merit in small studies but if they pan out in larger populations, we may well be on the threshold of a revolution in cancer diagnosis. Haick, for instance, is leading a consortium of European Union researchers that imagines a $20 “sniffing chip” for your mobile phone that will send your breath test to your doctor once a year. “Our aim is to make science fiction a reality,” says Haick.

Hossam Haick is confident of using machines to sniff out disease. “Our aim is to make science fiction a reality,” he says. – Hossam Haick

An accurate breath test for cancer is a prize worth aiming for. It is rapid and no blood or tissue need be drawn. It has the potential to save a fortune in medical costs, while providing a much earlier warning and a better chance of survival. Many cancers, notoriously lung and ovarian, are discovered too late, accounting for the poor prognosis – for lung cancer, 14% survival after five years; for ovarian it is 47%.

It is not an easy mission though. A person’s breath is a cocktail of thousands of different molecules, each present at the vanishingly low concentration of parts per billion. There is also tremendous variety in the breath components of normal people, and even people with the same type of cancer will no doubt have quite different signatures. But “difficult doesn’t mean impossible”, says Michael Phillips, a professor of medicine at New York Medical College and founder of the US breath analysis company, Menssana Research.

Jobi gets down to work. – Medical Detection Dogs

Ever since the father of medicine, Hippocrates, exhorted his students to “smell your patient’s breath”, doctors have been aware of the smell of disease. When organs fail, the body’s chemistry changes and those changes are transmitted to the blood. As the blood courses through the capillaries that enmesh the air-filled sacs of the lung, volatile chemicals percolate out and vent through the breath, like an engine’s exhaust. Doctors have long known that a fishy reek signals liver failure, the result of the build-up of sulphur compounds. Diabetics give off the fruity whiff of acetone because they burn fat instead of sugar. Patients with failing kidneys smell of ammonia, and a person with typhoid may exude the aroma of freshly baked bread.

These odours are distinctive because they result from high levels of a small number of molecules. “They are the low-hanging fruit,” says Phillips. But cancer is something else. Unless it has actually damaged an organ, the problem is merely one of overgrowth – a small outpost of cells amongst trillions, doing what they normally do, but more so.

Terence Risby is a sceptic who wonders how we can isolate the smell of cancer from background odours. – Terence Risby

Terence Risby, an emeritus professor at the Bloomberg School of Public Health in Baltimore, US, with a reputation as a sceptic in the field, has a problem with that. How, he asks, does one detect the signal above the noise? It seems much like trying to determine the sound of a small group of fans roaring extra loudly in a football stadium. Having been involved in breath testing for 30 years, he knows the challenges. His groundbreaking research used breath tests as a convenient way to monitor the release of damaging bursts of unstable molecules known as free radicals, which can damage cells and are possible markers of early cancer. The breath signature of these free radicals is a molecule called ethane. But as he points out, he could do this because he was comparing ethane levels in an individual’s breath before and after the treatment. But if he compared the ethane levels between different people, even healthy ones, the levels varied wildly. Mostly he had no idea why, but sometimes there were hints. Cigarette smokers and people with infections had raised levels. But exercise or even just breathing quickly could do it. This makes it exceedingly difficult to establish a normal level against which to base any mass screen test.

Risby is baffled by the idea that one could detect a unique breath signature for lung cancer patients especially since 90% of them are smokers. “Any compound you can think of is in cigarette smoke and it is loaded with free radicals.”

In any case, it is all too easy for passive vapours to contaminate breath for months. Risby recalls a case where one week after nurses had been in the operating theatre their breath still tested positive for anaesthetic, inhaled second-hand from their patients’ breath. Even though breath tests seem to work in the small studies to date, Risby suspects these are phantom signals - no more real than seeing a face in the clouds. Once tested in larger populations, they might well disappear.

Michael Phillips
It took Michael Phillips 40 years to develop a breath test for cancer. – AP Photo/Mike Derer

Michael Phillips, like Risby, has been in the sniffing business a long time. The two have developed a reputation as sparring partners but “we actually agree more than we disagree”, says Phillips. Phillips’ research has indeed identified a signature for cancer. And it has taken him only 40 years.

Trained as a medical doctor in Western Australia, Phillips spent most of his career as a clinical pharmacologist in US hospitals. He vividly recalls his introduction to breath testing in the 1970s. A colleague at the University of California, San Francisco, needed to check whether alcoholic patients were taking a medication known as “Antabuse”. Many patients simply didn’t take the pill as it caused violent illness if they drank. To test compliance they were asked to blow into a plastic bottle. If they’d been good, two minutes later the fluid in the bottle glowed a bright golden yellow. Phillips was amazed by the power of the simple breath test to show the body’s chemistry and wondered what else it could reveal.

In the library, he came across Linus Pauling. The brilliant chemist’s far-reaching interests had already earned him two Nobel prizes, but in his latest exploits, he had begun probing the contents of breath, using a powerful new technique called gas chromatography that separated out its individual components. He reported finding hundreds of molecules in breath. “I was hooked,” says Phillips. But it was to be a long and initially lonely journey. The field was “not even a backwater back then. There were no conferences and you ended up talking to yourself”.

'The breath of the healthy is more like a garbage dump for everything a person has experienced without and within.'

In 1992, Phillips penned an enthralling account of his progress for Scientific American. Like Pauling, Phillips used a gas chromatography column to separate the components of breath and a sophisticated mass spectrometer to identify the molecules as they marched off the column one-by-one. In stark contrast to these slick machines, he had a homemade device for collecting the breath – an ungainly plywood box with tubing splaying out in every direction. Amateurish though it was, it was critical to his success. It had to collect chemicals present in parts per billion and concentrate them a million-fold. His procedure had to be incredibly clean as any tiny contaminant would also be concentrated.

Until three years ago, he was able to identify some 200 breath components this way. Now, using a second separating column, he sees 2,000 compounds in breath – a collection he has dubbed the “volatome”. But the volatome is not like a blood sample where the chemistry is precisely regulated and remarkably constant in healthy people. The volatomes of the healthy are more like a garbage dump for everything a person has experienced without and within. So how is it possible to pull out a signature of disease?

Breathlink machine prototype
Phillips' prototype Breathlink machine is designed to be used in the field. – Menssana Research

Phillips acknowledges Risby’s concerns. And he admits that so far the numbers of patients tested is small. But the way to make sure he is not seeing “faces in clouds”, he says, is to employ sophisticated statistical algorithms. In the studies so far, he is able to identify a breath signature for lung and breast cancer with an accuracy of 75-85%. Phillips says the tests are more accurate at ruling cancer out than ruling it in as some of the positive results end up being false. But even if a simple breath test could rule out cancer, that could have an extraordinary impact, slashing the need for expensive CT scans and mammograms. “We’ll be able to say with a high degree of accuracy who does not need further tests. If you can exclude 70% of the population this way, you save lots of money and spare people from radiation.”

The lab equipment Phillips uses represents the Rolls Royce of breath detection. It is expensive and time-consuming to operate. But that is not what he envisages for mass screening.

To see what the future might hold, Menssana’s breath test for tuberculosis may be a guide. TB can be very hard to diagnose, especially in a child with common symptoms such as a cough, fever and malaise. In many parts of the world where TB is prevalent, CT scans to detect the telltale nodules TB leaves in the lungs are not an option. Sputum or blood tests to culture the bacteria take weeks that a person many not have. So a rapid breath test could revolutionise treatment. In the Menssana lab in New Jersey, researchers identified a breath signature for a person infected with TB, part of it representing the chemistry of the bug, part the chemical response of the host. Out of the 2,000 or so compounds in the breath, the signature was represented by a dozen or so compounds.

Having established that signature in the lab, the researchers designed a much simpler gas chromatogram for the field. The prototype machine called Breathlinks is battery-powered, can be wheeled around on a cart and comes with a US$30,000 price tag, a tenth that of the lab-version, and costs could drop rapidly if it proved its worth. So far it has been tested in four sites: the United Kingdom, Mumbai and two sites in the Philippines. The test requires a patient to don a nose clip and breathe naturally into a clean disposable tube for two minutes. The machine then analyses the contents and sends the data via an internet connection to New Jersey.

Overall accuracy is around 80%, but again the test does best at ruling TB out, with a greater than 99% certainty of a negative diagnosis. Being able to quickly rule out TB could be invaluable for doctors who need to make quick treatment decisions on very ill children.

The model that Phillips envisages for ruling out cancer is a fast, affordable Breathlinks-style system. “It’s going to be mainstream in medical diagnosis, I’m quite sure of that.”

Hossam Haick is modelling his system more closely on the dog. “Their nose is the most sensitive in the world, but it’s not just their nose, it’s their brain.”

A dog owes its remarkable sense to both. Their noses, like ours, are lined with nerves for scent-detection, like ours each is tipped with a specific receptor. But not only do dogs have 100 times the density of these nerves, they also have double the variety of specific receptors. Added to that, the percentage of the dog’s brain that is devoted to analysing smells is 40 times larger than that of a human. Dogs can identify smells somewhere between 1,000 to 10,000 times better than can nasally challenged humans.

Dog or human, when receptors detect odours they excite the nerve, which sends an electrical signal to the brain. The combination of scent molecules in the odour ends up relaying a pattern of electrical impulses that the brain learns as the signature of that odour. What might look like a cacophony of signals ends up being the exquisite aroma for coffee.

Haick is modelling his NA NOSE on the dog’s nose. – Hossam Haick

Haick has fitted his electronic nose with scent receptors made of gold nanoparticles coated with “thiols”, molecules that respond to scent. That response triggers a change in electrical resistance. Each sensor in the array responds to the mix of molecules in an odour and a pattern of electrical signals is generated that represents its signature. Coupled to that array is a microprocessor that can be taught to recognise different electrical signatures. Haick can get away with a relatively small number of artificial receptors because he has already ascertained the scent signature for particular cancers using the Rolls Royce level equipment in the lab. The electronic nose itself cannot identify the compounds in the breath. It has been trained to respond to a certain overall signature. “Just like the dog,” says Haick.

But just what are cancers producing that is so distinctive? The answer appears to be higher concentrations – up to 10 times - of commonly found molecules. Instead of being present at one to 20 parts per billion, they are present at 10 to 100 parts per billion. And why are particular cancers overproducing these molecules? “I’m afraid that no one knows the exact answer,” says Haick who has just finished composing a comprehensive review on the subject. But broadly speaking, fast-growing cancer cells outgrow their supply of oxygen, and that can result in the production of free radicals, changes to liver enzymes, glucose and fat metabolism.

Since beginning in 2007, Haick’s progress has been rapid, thanks in no small part to the buy-in from the European Union that is funding the “LCAOS” project to the tune of 5.4 million euros. Haick leads a team of 36 researchers across three companies and five universities and hospitals around the world.

Dogs' noses are up to 10,000 times more sensitive than our own thanks, in part, to the many more scent-receptor cells they have.

The electronic nose has had some failures in specific diseases, but the trials have shown 88% accuracy for lung cancer and 90-93% for gastric cancer. Trials in hospitals in the US, Europe and China, suggest that the latest version of the nose can even identify the difference between benign, early and late stage lung tumours, or the difference between a gastric ulcer and early stage cancer.

Now on its third iteration, Haick’s NA-NOSE has shrunk from a 20cm device to just 2cm. The next one will measure a mere 5mm: quite small enough to insert into your smart phone. In a few years time, Haick imagines we’ll be blowing into our smart phones and our GPs will receive the breath print. It certainly beats a mammogram.

Or we might just end up patting a dog. Mirroring Claire Guest’s work in the UK, veterinarian and researcher Cindy Otto at the University of Pennsylvania, is training canine bio-detectors to home in on the scent of ovarian cancer in blood samples. Whatever it is they are detecting is in turn being used to train an artificial nose created by George Preti’s group at the Monell Chemical Senses Center in Philadelphia. “Maybe, one day the artificial nose will be as good as the dogs,” says Otto.

Subscriber Exclusive The remainder of this article is exclusive to Cosmos subscribers

To continue reading this article, please subscribe for unlimited access or log in
Latest Stories
MoreMore Articles