Animal Instincts

As far as working conditions go, Midas is probably one of the happiest laboratory employees there is. She works a five-day week, with no requirement to go in on weekends. She works for about 20 minutes at a time, with a maximum of an hour a day spent on the job. She’s home by 3.30 pm every day. And, best of all, she gets praise, treats and rewards every time she does her job – namely, detecting cancer in urine samples. So how did Midas negotiate such a great deal for herself? It probably helps that Midas, a Hungarian wirehaired viszla, is cute. It probably also helps that no laboratory instrument yet developed is capable of the sophisticated detection Midas is trained to do.

The idea that dogs can smell cancer has been around for quite a long time, although until recently, the claims were all unsubstantiated. Claire Guest, co-founder of the charity Medical Detection Dogs, shared an anecdote about a Dalmatian whose repeated licking alerted its owner to the presence of a mole on her calf that turned out to be malignant melanoma – a cancer that was unlikely to have been caught so soon by any other means. This was the story that caught Guest’s attention and started her thinking about the potential implications. A psychologist working at Hearing Dogs for Deaf People, she was fascinated by canine behavior. Did cancer really have a smell? And if so, could dogs really detect – and possibly even diagnose – the disease?

Getting started 

It was years later that Guest first heard John Church, the “maggot man,” interviewed on the radio. Church’s claim to fame is the introduction of maggot treatment to the UK National Health Service, using medical maggots to remove infected or necrotic tissue from wounds. In his interview, though, he also mentioned that he’d heard a plethora of stories like the one Guest encountered, and that he was intrigued by the possibilities. Did anyone out there have the interest and expertise to train a “cancer dog,” he wondered? The answer was yes – and by the end of the day, Church and Guest were sitting together, making plans for the first-ever proof-of-principle study to see whether or not dogs could be trained to detect human cancers by odor.

But how could they begin training an ordinary dog to sniff out cancer? They needed a sample – something that would contain the odor of cancer, but not the cancer itself. After all, to a dog, a tissue sample is essentially a piece of meat. After weeks of deliberation, Guest and Church settled on testing urine samples from patients with bladder cancer; they knew that tumor cells are often shed into urine and can be detected via microscope, so they suspected it was possible that a volatile odor might also be present. The big question: could they teach a dog to detect it?

As it turned out, the six dogs they trained were able to identify the urine of bladder cancer patients in 22 out of 54 cases – a mean success rate 14 percent higher than would have been expected by chance alone (1). One of the dogs was Guest’s own pet and cancer-sniffing pioneer, Tangle, who was able to detect cancer with 56 percent accuracy. In comparison with modern hospital tests, that success rate doesn’t necessarily inspire confidence – but the dogs had been exposed to only about 50 samples in total, half of which were positive. With a larger training set, the researchers were convinced the rate of detection would improve.

And they weren’t the only ones. Suddenly, attention was drawn to the fact that cancer apparently had unique volatiles that could be identified through odor. For Guest, it was the beginning of an era – one in which she envisioned that well-funded training programs would produce teams of biodetection dogs to save countless lives. Although in practice, engagement has been somewhat slower, significant steps have been made since that day: electronics scientists have begun to analyze urine components in search of detectable volatiles, there have been breakthroughs in “electronic nose” diagnostics, and researchers have continued to build a strong evidence base supporting dogs’ noses as biosensors. Perhaps most significantly in Guest’s case, she and her first trustees set up Medical Detection Dogs and began a serious exploration of the possibilities of this off-the-wall, yet promising idea.

The nose knows

How good is a dog’s nose really? Humans have approximately five million sense receptors in their noses. Dogs, in comparison, have about 300 million – 60 times the sensitivity. “Some humans think they can smell a teaspoon of sugar in a cup of tea,” says Guest. “A dog with Midas’ nose would be able to smell a teaspoon of sugar in the equivalent of two Olympic swimming pools of tea.” And it’s not just the nose that does the job; dogs also have a vomeronasal organ, the organ of Jacobson, near the backs of their palates. When sniffing, they draw air into their throats so that it passes over this organ – which actually sends information to a different part of the brain. Sensory information from both the nose and the organ of Jacobson is consolidated in the brain before, ultimately, the dog makes a decision based on the scents it can detect.

Guest knew she needed to determine exactly how well the dogs could smell – but the available literature varied widely. So rather than use “head space” (blowing air with volatiles into the dogs’ faces), she and her colleagues mimicked the technique they wanted to use by providing samples and asking the dogs to investigate them. “We wanted to find out how low the dogs could go,” she says, “and then I wanted to do some manipulation, by varying the temperature or how long the sample had been out, to see if I could improve their abilities.”

The dogs began by detecting amyl acetate diluted in mineral oil. The initial goal was to get them to “threshold,” the point at which they were experiencing a 50 percent success rate, before beginning manipulations – but the longer they worked, the better they got. No literature reflected that discovery, but experienced handlers had long suspected that, much like marathon training, the dogs needed to build up their abilities over time. Sure enough, the threshold point continued to decrease week after week, until after six months of training, the dogs were detecting at a threshold dilution of 1:5,000,000,000 – one part amyl acetate per half billion of mineral oil.

But the real uniqueness of Medical Detection Dogs’ work came with their training system. Most systems are based on positive finds – dogs earn rewards for detecting something, but not if there’s nothing to detect. With cancer, though, the trainers didn’t know what the dogs were looking for, and they didn’t want to encourage a positive bias by rewarding the dogs only when they alerted. So they rewarded blank runs equally – resulting in a false positive rate of under 5 percent. That’s better than many medical tests, and at a fraction of the cost! At the moment, the charity is training a team of dogs to find prostate cancer; their accuracy using a 1 mL urine sample is over 90 percent, with a false positive rate below 5 percent. That means the biodetection dogs are actually performing better than prostate specific antigen (PSA) testing in the clinic (2)!

The demand for the dogs reflects their performance. “We’re already in a position where medics are asking us – can we send samples to your center?” says Guest. Medical Detection Dogs has a partnership with nearby Milton Keynes Hospital, which sends hundreds of samples to the charity for screening. Each sample is screened by a minimum of two dogs – usually three – to ensure accuracy and then the outcome is sent back to the hospital. Clinicians use those results in concert with symptoms, scans and PSA tests to decide whether or not to send patients for biopsy.

Bringing up biodetectors

What’s it like to train as a biodetection dog? “We have to teach the dogs when they’re young that this urine is basically a soup – it’s got volatiles, it’s got protein, it’s got everything – and the dog has to ignore about 98 percent of it completely.” Guest likens the process to training humans to pick up visual cues. “Imagine if I show you Monet paintings, and you get £100 for each Monet painting you spot with a small red flower in the left-hand corner. But I’m not going to tell you that; you’ve got to work out what it is in that painting that earns you the reward. Of course, the more Monet paintings I show you, the better you get at it. To start with you’re just guessing!”

But what makes the dogs better than electronic noses or other detection devices? “A dog can recognize a ‘red flower’ even if it changes slightly – even if the wind is blowing it slightly to the left or there’s a yellow flower in front of it,” Guest explains. “The electronic machines aren’t quite so good at that. Mass spectrometers are very good at finding specific things, but if you don’t know exactly what you’re looking for, it becomes much more complicated. Dogs are intuitive biosensors; they can tell you that they see a red flower even though it may not look exactly like the red flowers they’ve seen before. And that’s what makes them particularly good at 
this work.”

Branching out

In 2011, the group published a second study (3). By that time, the highest-performing dogs showed sensitivities of over 70 percent, and the average dog could spot cancer in a sample two-thirds of the time. What’s more, the results weren’t specific to a type of cancer; dogs trained on bladder cancer samples and then presented with prostate cancer reacted in very similar ways. “When they went up to the prostate cancer samples, they went, ‘Oh, yes, I think that’s it, but not quite.’ You have to reward them a couple of times before they’re certain. So I think there's a big part of the odor that’s the same, but not all of it.” At the moment, Medical Detection Dogs has three separate sets of biodetection animals in training for bladder, prostate and kidney cancers – so that if a dog alerts, consultants will know to look for that dog’s particular specialty.

As engagement increases, Medical Detection Dogs is expanding the animals’ repertoire even further. Their sights are now set on breast, ovarian, colorectal and even lung cancer. The latter two are particularly exciting. Lung cancer, responsible for almost one in five cancer deaths, is the most common cause of cancer death worldwide (4). Over three-quarters of lung cancers are diagnosed at an advanced stage (5), and the charity is optimistic that, with a new breath test in development, the dogs will be able to improve early diagnosis of lung cancer in the same way that they’ve shown an ability to detect early bladder and prostate cancers. Colorectal cancer detection is equally useful – the disease is the second-most common cause of cancer death in Europe and the fourth-most common worldwide (6), and over half of patients are diagnosed at an advanced stage (7). “I asked the colorectal surgeon, ‘Do we really need better testing for this cancer? You have stool screening – don’t they know if you’ve got blood in your stool?’” says Guest. “He said that most people die of colorectal cancer because they won’t do the stool sample. There’s much higher resistance to doing a stool sample than a urine sample. And people are dying because of it. So if we could detect colorectal cancer in a urine sample, a lot more people would be screened.”

Biodetection dogs clearly have potential. Although there’s always a need for more funding, more studies and more evidence, these hardworking animals have already carved out a niche in the world of early cancer diagnosis – and the results keep on coming. As Medical Detection Dogs expands its remit to include more diseases both within and outside the cancer sphere, it looks like the dogs are here to stay. Could pathologists one day share their laboratory space with furry, four-legged disease detectors? It seems unlikely, but you never know...

An Interview With Claire Guest

What role do volatiles play in disease diagnosis now?

Interest in volatile detection in cancer has grown hugely since our 2004 paper (1). It’s vital to develop a good evidence base, but that requires clinical trials. It’s a chicken-and-egg problem, because you have to start small and do a proof-of-principle study. Then you can do a larger trial with more patients, and if that’s successful, you can do multicenter studies to examine things like sensitivity and specificity. The challenge for us has been moving from the small proof-of-concept studies – which have had a lot of interest, but also invite skepticism because of the small sample size – to trials with a much larger sample size that yield a stronger evidence base. That’s what Medical Detection Dogs has spent the last decade doing, so we were pleased recently to get ethical approval for a breast cancer study, and for a three-year study on prostate, kidney and bladder cancer.

That one is a very significant study, because we’re not only looking at the dogs’ sensitivities and specificities over much larger sample sizes, but also observing them longitudinally. Patients in active surveillance come back every six months for check-ups, and every time a clinician sees the patient, a dog examines a sample. So we’ll be monitoring the progress of the dogs and the patients together, and we’ll be able to see whether the dogs can indicate the presence of cancer earlier than standard methods. That’s important because prostate biopsy currently has such a high false-negative rate; we need a more reliable test.

I think the detection of human disease by volatiles has largely been forgotten, although it was something that people talked about historically. In other parts of the world where testing isn’t as advanced as ours, people still talk about it. We’re not saying that our dogs are the ultimate answer for every disease and condition – but we are saying that there’s a huge area of diagnostics that is being overlooked, and our dogs are leading the way.

How might volatiles affect the way a diagnosis is achieved?

For every disease where early diagnosis is a struggle, detection through associated volatiles could become part of the diagnostic process. It may be for some conditions, that associated volatile pattern may be relatively straightforward to uncover once you know it’s there. For instance, the fact that a dog might detect a particular early-stage disease with 95 percent reliability on a skin pad tells scientists that the volatile pattern is easy to find, occurs before symptoms are seen, and can perhaps be detected with an electronic nose. Ultimately, that knowledge translates into a new diagnostic instrument for the disease.

In other conditions, the pattern might be much more difficult to detect. It might take a decade or two for a machine to do it reliably – and in the meantime, dogs could continue to screen those samples for physicians to use as part of the diagnostic picture. At this point, we’re not talking about screening populations; we’re testing symptomatic individuals, so it’s a small group of patients rather than thousands of samples a day. Later on, there’s a possibility that we could expand to screening high-risk groups (for instance, people who are predisposed to lung cancer and who are coughing), but there simply aren’t enough trained dogs to go out and screen healthy populations right now.

Why wouldn’t everyone want a canine colleague in the lab?

That’s a very good question. Some healthy skepticism at the beginning of our investigation was appropriate. But I think that, if we continue to produce this strong evidence base, then we have to ask: If this work is saving lives and helping patients avoid the painful and invasive treatments that come with late-stage disease, why wouldn’t you want to work with biodetection dogs?

The way we envisage it – and we have plans to develop this in the future if our next three years are successful – is that we’d have a hospital that included a patient-free area where a handler and a dog would come in on a daily basis and screen samples. That way, we could also use dogs to spot diseases where speed matters – for example, highly infectious conditions where you don’t want the patient to leave without a diagnosis in case they spread the disease. That’s been very successful with African pouched rats in Tanzania. The rats test sample for tuberculosis, and they spot patients that would otherwise rejoin the population and pass on the infection. We think it’s possible that similar programs could be instituted for sexually transmitted diseases and other infections.

But in order to do any of this work, it has to be funded – and medical economics can be quite complicated. Recently, numbers have begun to emerge showing that early cancer diagnosis saves money because of the high cost of therapy for advanced disease. And I think that’s the important point. Reliable early diagnosis allows us to treat diseases before they become hugely complicated and expensive. We could improve current diagnostic methods, too; screening tests and biopsies cost clinicians and patients time and money – and if the tests have a high false-positive rate, it’s not time and money that’s being used wisely. I think that’s where our dogs come in… to help identify the right patients at the right time.