Skip to main content

Certain diseases give off a distinctive smell, mostly undetectable to humans – but could dogs guide us to new ways of detecting cancer? Emma Young investigates.


The voice comes from behind a tall screen in a room at the University of Pennsylvania’s Working Dog Center. Pam, the trainer, is talking to Tsunami, a German Shepherd with a personality described as “hyper-vigilant”. Obediently, Tsunami sets off walking around a large metal disc mounted horizontally on a stand in the centre of the room.

Around the disc are 12 numbered compartments with gridded lids, each designed to take a glass jar. Right now, each of the jars holds a drop of human blood. Ten of the samples are from healthy people. One was collected from a woman with a benign ovarian disease. The final sample came from a woman with ovarian cancer.

Tsunami is one of three dogs being trained here to detect ovarian cancer. Her task today is to sniff the compartments and sit down by the one containing the cancerous sample. While she’s taken behind the screen for a brief play with Pam as a reward, the disc will be spun. She’ll then come back out, and sit by the repositioned target sample.

I’m in another room with George Preti, a researcher at the Monell Chemical Senses Center. We’re watching via Skype so we don’t interfere with Tsunami’s session. She is a very sensitive dog, I’ve been told. She’d have been terrible at police work, explains Cindy Otto, the director of the Working Dog Center, but at sniffing ovarian cancer from a single drop of blood, she’s brilliant.

Tsunami’s moving steadily around the wheel, nose close to it, her tail in the air. She sits down. After a few moments, Pam calls, “Seek!”

It seems she sat by the wrong compartment. This mistake will get recorded as a false positive. On average, Otto says, Tsunami gets it right about 90 per cent of the time. The two other dogs aren’t quite as good, but they’re around the 80 per cent level.

Tsunami gets up and moves around the wheel, then sits down.

Another pause. “Seek!”

Preti shakes his head. “This is probably why you won’t have a dog in the clinic! Even a good dog has a bad day. On Friday, she was getting everything right.”

The goal of this research programme is to take advantage of the fact that different diseases lead to tiny but highly specific changes in the chemicals our bodies produce, and these chemicals often have characteristics that give them a particular scent. So if we can sniff out these changes, we might be able to pick up fatal diseases in the early stages, when a cure may still be possible. But, as Preti says, it’s unlikely to be dogs doing the detecting – instead, researchers are working on ‘electronic noses’ to do the job. 

Artificial noses

Charlie Johnson proudly shows off the gleaming nanofabrication unit at the University of Pennsylvania: a calm, gloriously marigold-lit clean room, in which researchers in white overalls and masks move noiselessly about their cutting-edge business. Then he takes me down to the basement. After passing through a network of dim tunnel-like corridors, strewn with all kinds of research detritus, cardboard boxes and half-coiled cables, we reach his lab.

Johnson is a physicist, and director of the university’s Nano/Bio Interface Center. Down here, on a bench against the wall, is the technology that he and George Preti hope will evolve into a commercial detector capable of picking up ovarian cancer at the earliest stages. 

A graduate student, Nick Kybert, shows me how it works. A sample of blood is heated in a water bath. Volatile compounds evaporate and are channelled along a tube to a wafer consisting of little sensor arrays.

This is the ‘electronic nose’. Each array contains thousands of carbon nanotubes. Each nanotube is wrapped in single-stranded DNA. In total, ten different DNA sequences are used, making for ten types of sensor. Electricity flows through the network. If the DNA on any of the nanotubes binds to a molecule in the gas from the sample, there’s a change to the electrical conduction, which is detected and recorded. “The DNA basically absorbs molecules out of the air, the way your own nose would do it,” Johnson says.

While there are different approaches to e-noses, this is the one that works most like the human olfactory system, he says. The sensors respond differently to different chemicals. “They don’t just pick up one thing. They tend to respond to multiple targets, but some strongly, and some weakly.” Just as we use the pattern of activation of different types of odour receptor in our nose to perceive complex smells, the pattern of activation of the ten sensor types produces the final reading.

Johnson would like to develop an electronic version with 100, rather than ten, different sensors. “Then we’d have a number that’s getting comparable to the human nose,” he says. But even with ten, the preliminary data on ovarian cancer looks promising. “You can see quite readily that the readings from healthy individuals land in one spot, as it were,” he says. “The benign tumours land in another spot, which is distinct. The cancer is well differentiated from the healthy controls, and is differentiated – but not as reliably – from the benign tumour.”

Johnson stresses these are just the very first results. And the DNA strands used around the nanotubes were selected at random. It’s possible to shuffle around the bases in a strand of DNA, but no one really knows at the moment which sequences would work best for odour detection. Still, it would certainly be possible now to screen sequences against cancer samples, to find ones that react most strongly, Johnson says. This will probably be the next step in the research. And then, the barriers to turning the lab bench set-up into a handheld detector would be small, he says: it would just take money.

Ovarian cancer is most commonly detected at stage three of the disease, when it has started to spread to other tissues. But some cases are detected early. As well as developing the sensors, the team want to do more testing with early-stage samples, hopefully to develop a detector that can pick up the disease before it has spread, and so is much more easily treatable. “That’s the whole point, right,” Johnson says. “One wants to flip that ratio so they’re mostly detected in the early stage.”

Return back to top of the page