Detecting disease in breath

Paul Thomas, an analytical chemist at the breath diagnostics company Bioxhale, concluded his presentation at the 2020 Breath Biopsy Conference with a cartoon by Gary Larson. The cartoon features two beavers gnawing at a lumberjack’s wooden leg. One beaver says to the other, “Hey! They’re edible! This changes everything!” 

“The point is that we jump to conclusions so readily based on such limited observations and data,” Thomas said. “It’s just a warning to anybody who's listening to me that I could be that beaver, wildly mistaken and misinformed about the nature of my observations, and that some humility and skepticism is entirely appropriate in all of these things.”

This sage sentiment, while applicable to scientific research in general, is especially relevant to the pursuit of methods to detect disease in exhaled breath. The notion that breath might provide an unexpected source of health information can be traced as far back as the Greek physician Hippocrates, who sniffed out liver disease in patients with fishy-smelling breath (1). 

Translating this ancient anecdotal practice to a rigorous modern diagnostic tool, however, has proven challenging. To pinpoint unique features in complex breath samples that can be reliably traced to disease, researchers are exploring new directions in biomarker discovery, data processing, and breath sampling workflow. In doing so, they aim to finally unleash breath’s diagnostic potential, making detection of conditions ranging from COVID-19 to cancer as easy as breathing.  

Just breathe

In a long line of biofluids that researchers have explored as diagnostic media — including blood, sweat, tears, and urine — the appeal of targeting the involuntary, nonstop process of breathing is clear. Collecting breath is completely noninvasive, and there’s no limit to how much a person can provide.  

It turns out that Hippocrates was onto something when it comes to breath’s diagnostic value. As diseases disrupt key biochemical pathways and metabolic processes, they leave behind a trail of molecular markers in their wake, and those that evaporate end up in exhaled breath vapor. These volatile organic compounds (VOCs) are produced directly within the respiratory system or gastrointestinal tract or carried to the lungs from circulating blood, enabling the composition of exhaled breath to indicate distress throughout the body. “Anything that can escape into the breath and that we can measure can tell us a lot about the state of health or disease of an individual,” said Raed Dweik, a pulmonologist and breath diagnostics researcher at the Cleveland Clinic.

Anything that can escape into the breath and that we can measure can tell us a lot about the state of health or disease of an individual.
- Raed Dweik, Cleveland Clinic

Breath-based signals of disease aren’t blaring sirens that sound out of dead silence, though. “Almost all the biomarkers we find in breath exist in [healthy] people. It’s just much higher or much lower in a disease state,” Dweik said. Just as blood tests measure the concentrations of certain chemicals relative to a normal range, breath testing looks for changes in the levels of VOCs.  

These changes can be challenging to detect in the rich mixture of breath, where thousands of VOCs comingle at concentrations comparable to a single spoonful of sugar in an Olympic-sized swimming pool. In addition to the many endogenous sources, external substances yield additional compounds in exhaled breath that can interfere with the detection of biomarkers of disease. As anyone who’s ever eaten garlic knows, ingested foods leave their own mark on the breath. Similarly, “As we walk around or drive around, we are sadly the vessels. Everything in the environment that we breathe in, we eventually breathe out,” Dweik said. “All these confounders I think need to be solved for to look for the morsel of information we want about the health of the metabolic state.”

Mining for morsels

Thomas’s career in breath research originally focused on detecting the environmental toxins that people breath in and out. But when his project measuring radiation exposure was derailed by the COVID-19 pandemic, he wondered if the SARS-CoV-2 virus might create its own VOC signature in the breath. To see if they could distinguish COVID-19 from other diseases, his team collected breath samples from patients diagnosed with COVID-19 as well as asthma, chronic obstructive pulmonary disease, bacterial pneumonia, and cardiac conditions. The researchers analyzed the breath samples using a technique that separates the various VOC components and detects each one via its movement through a gaseous electric field. Using statistical methods, they developed an equation that provides a COVID-19 breath score based on the levels of six VOCs and found that it could predict infection status with about 80 percent accuracy (2).

The ability to differentiate COVID-19 from other respiratory diseases comes from looking at the virus’s systemic effects beyond the lungs, according to Thomas. “The reason it works is that COVID is such a powerful disease; it affects just about everything in your body,” he said. For example, the team detected high levels of aldehyde compounds, which are linked to oxidative damage and inflammation, high amounts of ketone metabolites, indicating irregular energy metabolism, and a low concentration of methanol, suggesting disrupted activity in the gut microbiome. “When you put all of those effects together, you get a pattern of chemicals in breath that changes quite significantly because of all of this insult and dysregulation in someone who's ill with this disease,” Thomas said. 

Since Thomas’s 2020 study, the delta and omicron variants emerged, bringing with them not only spikes in cases, but potential changes in the COVID-19 VOC profile. Cristina Davis, an interdisciplinary engineer at the University of California, Davis, wondered whether variants may lead to unique changes in the constellation of VOCs in breath. Her team collected breath samples from healthy people and those with COVID-19 during the delta wave (through November 24, 2021) and the omicron wave (after January 11, 2022). 

They analyzed the samples using a technique that concentrates the VOCs onto a solid adsorbent surface, applies heat to dislodge the compounds, and separates and identifies each one by its mass. To control for environmental confounders, the researchers compare their human samples to environmental samples and delete any ambient compounds from the dataset. They can also ensure that participants haven’t ingested anything for at least an hour prior to giving a breath sample to minimize detection of ingested VOCs.  

Davis’s team measured the abundance of hundreds of VOCs in the breath samples and developed statistical models to predict COVID-19 infection using data from all of the COVID-19 patients versus just those with the delta or omicron variants (3). They found that the models developed for each specific variant showed higher accuracy than their general COVID-19 model.

“In the volatile organic compound profiles, there were some pieces in common, but part of it shifted during the omicron strain that proliferated,” Davis said. “That’s something we need to be aware of because we may need to rebuild those models if you get a big enough variant shift.” The researchers are currently evaluating their ability to distinguish COVID-19 in all its forms from the flu now that flu cases have reemerged from a two-year lull due to social distancing, masking, and hand washing.

Davis’s team also examines VOCs emitted by cultured cells to complement their measurements in exhaled human breath, helping them tie their biomarkers back to biochemical mechanisms (4). “When you're using cell culture studies, you can actually understand those signal transduction pathways and where the biology is and where it’s coming from so that it's not just phenomenology that you're measuring,” Davis said. 

When you're using cell culture studies, you can actually understand those signal transduction pathways and where the biology is and where it’s coming from so that it's not just phenomenology that you're measuring. 
- Cristina Davis, University of California, Davis

Breath-based diagnostic tests could have a major impact on reducing transmission of a variety of viruses. Davis’s COVID-19 study relied on symptomatic participants, but she is interested in investigating if VOC analysis can also pinpoint asymptomatic infections — something her cell culture studies predict is possible. “We're looking at what we believe is the host response to infection rather than looking for the virus itself, which may or may not have enough copies for the PCR tests to actually detect it yet,” she said. “The host response to infection starts as soon as infection starts, and so the question is, based on that cell culture data, is there a point at which we can measure those biomarkers before the traditional measures would be available?” 

Thomas sees COVID-19 as a case study for how breath tests can protect underserved populations that don’t have access to traditional testing technology. His VOC detection instrument is relatively portable at about the size of a printer, but “the ultimate goal for breath testing is to be like the breathalyzer for alcohol,” Dweik said. “If you can do it on the side of the street, you can do it anywhere else.” 

Once researchers have identified the critical VOC biomarkers of a disease, small sensors that measure specific compounds could translate their diagnostic discoveries to point of care applications. “Being able to know what compounds are in the breath in a handheld device would be very powerful,” Dweik said. “That will require collaboration between scientists, physicians, technicians, and analysts.” 

A pattern of disease

While breath testing technology might not need to be quite so miniaturized for applications in diseases such as cancer, it could provide a more accessible screening tool for primary care settings. “We envision you could just walk down the street to either your general practitioner’s office or to a pharmacy and give a breath sample that way,” said Stephen Graham, a pharmacist and CEO of the breath diagnostics company Breathe BioMedical. Breath testing may even be able to pick up on early metabolic changes that signal cancer before traditional imaging techniques would detect a tumor.

Breathe BioMedical’s sampling device lets breath from the upper respiratory system pass through before collecting alveolar breath from deep in the lungs, minimizing the presence of ingested compounds in the sample. VOCs in the alveolar breath are trapped onto a solid adsorbent surface, liberated using heat, and injected into an optical cavity with super reflective mirrors on either side. As laser light bounces off the mirrors about 1,000 times, the VOCs absorb more and more of the light with each pass. This highly sensitive technique generates a spectrum of an individual’s breath sample that captures minute shifts in VOC abundance. Acquiring these breath spectra from a group of individuals with the same disease provides a distinct breath print of the disease.

Researchers at Breathe BioMedical use machine learning algorithms to analyze the complex spectra and identify distinguishing features for a disease, accelerating the development of diagnostic models from a collection of breath samples. “We don't look at exact concentrations of specific VOCs,” Graham said. Rather, “The model learns, and it says, ‘Okay, these are the features that look like cancer, and these are the features that look like noncancer.’” Graham compares this approach to disease sniffing dogs. “The dog doesn't smell cancer and say, ‘Oh, yes, I smell so many parts per billion of isoprene,’” he said. “It just recognizes that the features of that smell are what cancer is.” 

It’s easy to lose sight of metabolic underpinnings in a mountain of mysterious peaks in a spectrum, however. Dweik recalled an incident in which his team uncovered a data feature that appeared to give perfect diagnostic accuracy — “almost too good to be true,” he said. Indeed, when they investigated the compound behind the feature, they discovered that it arose from a VOC in the handwash at the hospital where the disease cohort was treated. “You could fall into the trap of thinking you found the test, but really, it's a contaminant,” Dweik said.

The Breathe BioMedical team compares every breath spectrum to a sample of room air to control for potential ambient confounders. They also process their spectra to identify and quantify specific VOCs, enabling them to confirm that they are measuring biologically relevant compounds. 

Our vision is that 10 years from now, you'll be able to go in, give a breath sample, and then receive a panel of screening tests for a number of different diseases from the same breath sample. 
- Stephen Graham, Breathe BioMedical

In a recent study, the researchers collected breath spectra from healthy people and those with lung cancer and analyzed them using both breath print feature recognition with machine learning and standard VOC composition analysis (5). They observed that the breath print method gave better diagnostic accuracy. While researchers have struggled to define a common VOC profile in different individuals with lung cancer, machine learning can reveal subtler features of the disease across the entire spectrum. “We think that we can solve some of that problem with heterogeneity by taking a more comprehensive approach to looking at the data through the absorption,” Graham said. 

Analyzing a breath sample comprehensively may also enable it to provide a general snapshot of health rather than a test for one specific disease. “Our vision is that 10 years from now, you'll be able to go in, give a breath sample, and then receive a panel of screening tests for a number of different diseases from the same breath sample,” Graham said.

As the team continues to validate their method against more traditional biomarker discovery approaches, Graham hopes that spectral feature recognition using machine learning will become the standard in the breath diagnostic space. “It's a shift in thinking, and I think we're the first ones who are really spearheading that shift,” he said.  

Probing deeper

The inability to identify a VOC signature for lung cancer may also stem from the fact that many of the compounds result from nonspecific responses to disease. “If you look at the altered metabolism of lung cancer and at some of the processes that relate to the production of VOCs, things like oxidative stress and inflammation, you're probably not going to get the specificity that you would need for a test for something like early cancer detection,” said Billy Boyle, an engineer and CEO of the breath diagnostics company Owlstone Medical. 

Boyle wondered if simply measuring the endogenous VOCs in the body that end up in exhaled breath was the best strategy for complex diseases such as cancer. “Part of it was from the frustration that there’s got to be a better way if we take a more bottom-up type approach to understand the biology and try to target specific pathways,” he said. 

A breakthrough came from a compound called limonene, which is found in citrus. “We would see it pop up all the time in our analysis, and we would discount it and say, ‘Oh, limonene just comes from diet,’” Boyle said. However, when the team explored the journey limonene takes in the body, they discovered that it’s metabolized by liver enzymes (6). “Compounds that we previously thought, ‘they’re exogenous, so therefore they’re probably chemical noise’ —  what if we introduced them into the body deliberately?’” Boyle said. “What if we start to think about the exogenous compounds as being probes for particular pathways?”

In searching for unique characteristics of cancer they could target with an external compound, the researchers came across extracellular β-glucuronidase enzymes in the environment of solid tumors, such as those found in lung cancer. They designed a deuterium isotope-labeled molecular probe to administer into the body, where any β-glucuronidase enzymes at the tumor site would cleave a bond, yielding deuterated D5-ethanol that comes out in exhaled breath (7). This product can be detected by its heavier mass, which distinguishes it from ethanol in exhaled breath due to alcohol consumption, providing an unambiguous VOC marker of cancer. “The hope is you’d see almost this very clear binary signal of the presence of D5-ethanol in breath, signifying the glucuronidase and the tumor microenvironment,” Boyle said. 

Administering the exogenous probe by inhalation enables it to specifically detect lung tumors over other types of solid tumors and gastrointestinal bacteria that also produce β-glucuronidase, said Anil Modak, a medicinal and pharmaceutical chemist and scientific advisor to Owlstone Medical. Delivering exogenous probes through inhalation or ingestion rather than intravenous injection is also critical for upholding the basic tenet of breath diagnostics as a noninvasive platform, he added. 

The Owlstone Medical researchers found that β-glucuronidase is present in the tumor environment of 90 percent of early-stage lung cancer tissue samples. However, in its first stages, “the tumor is extremely small and the amount of β-glucuronidase enzyme increase is very minimal, so the amount of D5-ethanol that is produced may be almost at undetectable levels. …Can that D5-ethanol be detected at very small amounts in exhaled breath?” Modak said. “The clinical unmet need is identifying the lung cancer prior to it being detected by a CT scan or PET scan.”

Boyle aims to boost sensitivity by introducing an excess of the exogenous probe, allowing one enzyme to cleave multiple substrates into product, and by extending the breath collection period to enrich the sample. He is deeply aware of the need for early cancer detection, having lost his wife to colon cancer that was diagnosed at a late stage. “I really understood why early detection matters in terms of survival outcome,” Boyle said. “That was a key moment for me to say I know what I want to do with the rest of my life, which is to try to hopefully find ways to develop tests that pick up disease sooner.”

The clinical unmet need is identifying the lung cancer prior to it being detected by a CT scan or PET scan. 
- Anil Modak, Owlstone Medical 

While Boyle is blazing a new frontier in early cancer detection with exogenous probes, he is also honing the fundamentals of breath diagnostics for other diseases using endogenous VOCs. “Often, it comes down to getting the basics right to ensure you are detecting chemicals that are truly coming from breath, understand what they are, understand what the ranges of those compounds are in the healthy population, and then think about applying it to the disease population,” he said. 

Toward this end, the Owlstone Medical team compiled the Breath Biopsy^® VOC Atlas documenting the concentration ranges of approximately 150 compounds in the exhaled breath of a heterogeneous group of healthy people (8). The goal of this catalog is to help researchers establish a normal VOC baseline, separate biomarkers from confounding compounds, and ultimately draw well informed conclusions that don’t leave them vulnerable to the beaver trap. “This will be a growing resource over time,” Boyle said. “That's something that we're investing in doing, but with a view to make it available to the broader community.”

References

1. Francesco, F.D., Fuoco, R., Trivella, M.G., & Ceccarini, A. Breath analysis: trends in techniques and clinical applications. Microchem J   79, 405-410 (2005). 
2. Ruszkiewicz, D.M. et al. Diagnosis of COVID-19 by analysis of breath with gas chromatography-ion mobility spectrometry - a feasibility study. eClinicalMedicine  29, 100609 (2020). 
3. McCartney, M.M. et al. Predominant SARS-CoV-2 variant impacts accuracy when screening for infection using exhaled breath vapor. Nat Med Comm  2, 158 (2022). 
4. Yamaguchi, M.S. et al. Headspace sorptive extraction-gas chromatography-mass spectrometry method to measure volatile emissions from human airway cell cultures. J Chromatogr B Analyt Technol Biomed Life Sci  1090, 36-42 (2018). 
5. Larracy, R., Phinyomark, A., & Scheme, E. Infrared cavity ring-down spectroscopy for detecting non-small cell lung cancer in exhaled breath. J Breath Res  16, 026008 (2022). 
6. Ferrandino, G. et al. Breath biopsy assessment of liver disease using an exogenous volatile organic compound — toward improved detection of liver impairment. Clin Transl Gastroenterol  11, e00239 (2020). 
7. Labuschagne, C.F. et al. Breath-based detection of lung cancer using exogenous volatile organic compound targeting β-glucuronidase in the tumour microenvironment. At < https://www.owlstonemedical.com/media/uploads/files/2022-10_Evolution_poster.pdf>. 
8. Owlstone Medical. Breath Biopsy^® VOC Atlas. At < https://www.owlstonemedical.com/science-technology/breath-biopsy-voc-atlas/>.