This interview is part of The Analytical Scientist’s feature exploring how analytical technologies are changing the science of sport – from metabolomics, microsampling, and wearables to anti-doping, recovery, muscle growth, and precision medicine.
Travis Nemkov and Angelo D’Alessandro, researchers at the University of Colorado School of Medicine, use omics, microsampling, and machine learning to study elite athletes under extreme physiological stress. By profiling cyclists and ultra-endurance runners before, during, and after major efforts, they are using sport as a stress test for human physiology – a way to examine fatigue, inflammation, red blood cell turnover, recovery, and the boundary between beneficial adaptation and overload.
Here, they discuss how studies in elite sport could inform precision exercise prescriptions for cancer survivors, healthy aging, and metabolic health.
How did your work in sports metabolomics begin, and what have been some of the key findings so far?
Travis Nemkov: This program really started around 2015, when I met Inigo San Millán, who at the time was working as a cycling performance coach and later became performance director at Athletic Bilbao. He was also faculty at the University of Colorado, where we’re based, and we started talking about how our mass spectrometry and omics expertise could potentially be applied to sports science.
At the time, most of our work focused on defining molecular signatures of disease. We were using metabolomics, lipidomics, and proteomics to study patient populations and clinical cohorts. But we realised we hadn’t really spent much time asking the opposite question: what does optimal human health actually look like at the molecular level?
That became the starting point for the sports work. Inigo was then working with UAE Cycling, and around 2016 we started collecting samples from elite cyclists during training camps. We analyzed blood samples taken before and after different training tests and found that we could distinguish not just good performers from poor performers, but even good performers from the very best performers within an already elite population.
Using metabolomic signatures linked to things like lactate threshold, we identified differences in amino acid metabolism, fat metabolism, and coenzyme A metabolism. It was striking how diagnostic these molecular signatures turned out to be.
The technology itself is also important here. We were using mass spectrometry together with touch-activated blood collection devices, so we no longer needed traditional venepuncture. Instead, we could collect around 10 microlitres of blood from a fingertip or shoulder-based microsampling device and still measure 3,000–4,000 molecules spanning metabolites, lipids, proteins, and their modifications.
We then extended that work into real competition settings. One of the first studies of its kind involved collecting samples from cyclists during a World Tour stage race. We sampled riders before and after stages and compared those profiles with their training camp data. What we found was that the best-performing cyclists were able to maintain their fat-burning capacity throughout the race, even as fatigue accumulated. While some molecular features progressively deviated over time, the top performers consistently recovered much closer to baseline between stages.
That kind of information could eventually be used to personalize training or nutrition strategies. If certain depleted metabolites relate to vitamin pathways or recovery mechanisms, for example, those could potentially be targeted directly.
More recently, we expanded the work into ultra-endurance running. In a study involving runners competing in the UTMB – a 171-kilometre ultra-marathon around Mont Blanc – we combined metabolomics, lipidomics, proteomics, and metallomics with haemodynamic and haematological measurements. These athletes were running continuously for roughly 24 to 45 hours, and we observed an enormous inflammatory response together with clear evidence of red blood cell damage and turnover.
The body was responding by releasing reticulocytes – immature red blood cells – into circulation. Interestingly, the red blood cells themselves began to resemble cells aged in stored blood bags used for transfusion medicine, which gave us insight into how extreme endurance stress alters blood physiology.
We also analysed samples from runners in the Trans Europe Foot Race, a 4,700-kilometre ultramarathon stretching from southern Italy to northern Norway over 64 consecutive days. There, we saw even more pronounced inflammatory and vascular effects, including arterial stiffening. Plasma from those runners impaired vascular endothelial cell function in ex vivo experiments and increased oxidative stress markers.
What this work is really helping us understand is where the boundary lies between adaptive training and physiological overload – when exercise shifts from being beneficial to potentially harmful.
But importantly, our ultimate goal is not really about helping athletes break world records. We’re using elite athletes as a model system for understanding optimal human physiology. The long-term objective is to apply those insights to patient populations, particularly in areas like exercise oncology.
For example, we’re now studying whether these molecular signatures can help personalize exercise prescriptions for cancer survivors. Many patients benefit enormously from structured exercise programmes after treatment, but not everyone responds equally well. We want to understand whether metabolomic profiling can help identify who will benefit most, and how to tailor exercise interventions more precisely.
Why has your work focused so heavily on endurance sports, and how does that connect to broader questions around health and disease?
Nemkov: Initially, the simple answer is that those were the samples we had access to first. From a logistical standpoint, endurance sports were the easiest place for us to begin. But scientifically, there’s also a very strong rationale for focusing there.
When we think about healthy ageing and reducing morbidity across the general population, cardiorespiratory fitness consistently emerges as one of the strongest predictors of long-term health. So endurance sports give us a model system for studying the cardiovascular and respiratory systems operating at their absolute limits.
A large part of our laboratory’s expertise also centres on red blood cell physiology and biology. During prolonged endurance exercise, red blood cells are under constant stress – repeatedly loading oxygen in the lungs, delivering it to tissues, and carrying carbon dioxide back for exhalation. Over time, that process accumulates oxidative damage. Endurance athletes therefore provide an ideal system for understanding how the body responds to sustained physiological stress.
That said, we’re certainly not limited to endurance studies. There’s also important work comparing endurance and resistance training – including through the large Molecular Transducers of Physical Activity Consortium – but for us, when somebody offers the opportunity to work with a professional cycling team that includes future Tour de France winners, you take that opportunity.
Angelo D’Alessandro: A lot of my own entry into sports science actually came through blood storage and transfusion medicine rather than exercise physiology directly.
Before moving to the United States, I worked with the Italian National Blood Center studying how blood quality changes during storage. Blood transfusion is one of the most common medical procedures worldwide, and we became interested in how blood ages over time in blood bags and why that ageing process varies between individuals.
At the same time, those same biological systems are highly relevant in elite sport because blood manipulation and erythropoietin use are forms of doping. During my PhD, researchers connected with the World Anti-Doping Agency became interested in our work, particularly around detecting autologous blood transfusions and microdosing strategies that are otherwise very difficult to identify.
What emerged from that work was the idea that blood provides an incredibly powerful window into overall metabolic health. Clinicians have always relied on blood tests to understand disease, but by applying metabolomics and multiomics approaches to tiny blood samples, we can now study human physiology at a far deeper level.
One of the major goals for us has been scalability. We wanted to move mass spectrometry beyond being a niche research technology and make it accessible at population scale.
To do that, we developed extremely high-throughput metabolomics, lipidomics, and proteomics workflows. Some of our methods now run in roughly one minute per sample while still capturing central carbon and nitrogen metabolism alongside exposure markers such as caffeine metabolites, smoking-derived compounds, and nutritional biomarkers.
Ultimately, the vision is that elite athletes become an early adopter population for this technology. We can collect blood from a simple finger prick or touch-activated microsampling device – even during competition – and use those data to monitor fatigue, injury risk, training adaptation, inflammation, and recovery in real time.
But the long-term ambition is broader than sport. We want these tools to become accessible to the general population as part of preventive and personalized healthcare.
Artificial intelligence is also becoming central to this work. We’re now applying machine learning approaches – elastic net models, random forests, large language models – to integrate these enormous datasets with real-world performance measurements. That allows us to identify molecular signatures that predict performance, fatigue, and physiological stress, and eventually to create digital twins of athletes that can model how somebody might respond to training, competition, or recovery protocols over an entire season.
In the future, you could imagine monitoring an elite sports team longitudinally throughout a season, using repeated microsampling to identify who is approaching injury risk, who may need reduced workload, or who is recovering optimally. But equally, the same principles could apply to ageing, chronic disease, rehabilitation, or exercise prescriptions for the general public.
What’s exciting is that seemingly unrelated areas of biology are beginning to connect. For example, we’re also studying mammalian hibernation and how certain species preserve muscle mass through winter. Those adaptations may ultimately inform exercise physiology and muscle preservation in humans.
Do you think we already have the analytical tools and biological understanding needed to start applying these approaches in real sporting environments?
Nemkov: Yes, I think we do. One of the interesting things we’ve seen is that when we compare the blood of elite cyclists after exhaustive exercise testing with blood from clinical patient populations, we actually start seeing overlapping molecular signatures. For example, some of the metabolites associated with extreme fatigue in athletes mirror markers we observe in patients with long COVID or in cancer patients experiencing severe cancer-associated fatigue.
That’s important because it reinforces the idea that these molecular signatures are genuinely informative biomarkers of fatigue and physiological stress.
So in terms of the analytical technology and the biology, I think we already have enough to begin applying these approaches in practice. We can already measure biomarkers linked to fatigue, oxidative stress, incomplete fat oxidation, amino acid metabolism, and a range of other exercise-relevant pathways.
Importantly, though, these technologies operate in two modes simultaneously. There’s an application mode, where we already know enough to report actionable biomarkers, but there’s also a continuous discovery mode happening in parallel. Every time we collect these large-scale datasets, we’re also learning more about biochemical pathways and uncovering new markers that may become useful in the future.
So I think we’re already at the point where these tools can extend well beyond traditional metrics like lactate alone. Lactate is useful, of course, but we now have access to a much richer molecular picture that includes cofactors, amino acids, oxidative stress markers, and many other pathways that shift during fatigue and recovery.
D’Alessandro: What Travis is describing is essentially a precision metabolomics approach – where you begin with a working hypothesis, collect data, and then continuously refine that hypothesis as more information becomes available.
We already have molecular signatures associated with specific outcomes of interest, but as datasets become larger and more diverse, we can start understanding how well those signatures translate from elite athletes into broader populations.
Our expectation is that elite athletes will likely become the first major adopter group, and we’re already seeing interest from Olympic committees and professional sports organisations. After that, the next wave will probably be semi-professional and recreational athletes before these technologies eventually move into broader public health applications.
The long-term vision is really a “precision metabolic health” platform – where individuals can monitor their metabolic trajectories over time and understand how interventions like exercise, diet, or recovery strategies influence long-term health outcomes. Rather than a static snapshot, you begin building a dynamic model that tracks how physiology changes over time.
I realise some of this still sounds futuristic, but the theoretical and analytical foundations are already there. Now it’s really about testing these systems in the real world and determining how effectively they translate into practical decision-making.
What gives us confidence is that we’re already applying similar approaches successfully in clinical medicine. For example, we’ve used these same metabolomics and machine learning strategies in trauma patients arriving at emergency departments and shown that we can predict outcomes – including bleeding complications, inflammatory complications, respiratory distress, and even mortality trajectories – with very high specificity and sensitivity.
We’ve also applied these methods in acute myeloid leukaemia to predict treatment response with very high accuracy, and those approaches are already being deployed clinically at our institution.
So from our perspective, the underlying analytical chemistry infrastructure already exists. The question now is not whether these technologies can work – it’s how broadly and how quickly they can be translated into sports medicine and eventually into precision health for the wider population.
How often would athletes need to be tested for molecularly informed training to become useful – and is there a trade-off between what’s scientifically ideal and what’s practical?
D’Alessandro: From a practical standpoint, the key advantage is that we don’t need a standard 5–10 mL blood draw. We can do this from 10–50 microliters of blood, which means the sampling can be done with a minimally invasive finger-prick or touch-activated device.
That makes it much easier to collect samples in the field, and it avoids issues around drawing larger blood volumes from athletes. Technically, that means you could test almost daily if you wanted to. In practice, something like twice a week – before and after a game, or before and after a key performance – might be more realistic.
Nemkov: The testing schedule would probably depend on the context. During training camp, for example, you might test at the beginning, middle, and end – or more frequently if you’re trying to understand how someone is adapting. During a season, maybe it becomes monthly or quarterly monitoring.
If an athlete is injured and you’re trying to track recovery, you might test more often to see how quickly they’re returning to baseline. So it’s adaptable. We don’t yet have one prescribed interval.
But technically, there’s no limitation stopping us from testing daily. The main constraints would be cost, logistics, and deciding what level of information is actually useful for the athlete or coach.
What are the main barriers to wider adoption? And do you think this kind of testing could become much more commonplace over the next five to 10 years?
Nemkov: Right now, the biggest barrier for me is too much data. People receive reports with measurements on 20, 40, or even 100 different parameters, and they don’t always know how to act on that information. As more data are collected and as our understanding improves, I think we’ll be able to refine these signatures down into individual panel scores, rather than reporting on five to seven metabolites that are all involved in the same pathway.
This is really a machine learning problem to solve. It’s akin to polygenic risk scores, where you’re trying to understand someone’s genetic propensity to develop a certain disease. You might measure specific variation across 200 to 400 different genes, but collectively that gets consolidated into a single polygenic risk score. I think that’s where this is going in this context as well.
But we really can distinguish athletes and performance based on these technologies, so I do think it’s going to be adopted. It’s such a good prognostic measurement for how training and competition are going that I would be surprised if, in 10 years, this wasn’t being used.
In a way, it’s already being used, if you consider lactate measurement as a way of guiding training and performance. This is just measuring multiple pathways that are tangential to lactate accumulation, which could potentially provide more precision when prescribing a different training regimen or nutritional protocol.
D’Alessandro: We have an informal idea of what could be useful, and there are going to be surprises along the way. Some things we expect to be useful in the real world may turn out to be only marginally useful, or not useful at all. And there may be unexpected signatures that prove predictive of other outcomes.
Ultimately, we may find that, within this deluge of data, the relevant information is limited to a subset of structured data. From there, a more targeted lab-on-a-chip approach – perhaps a wearable device that can sense a handful of relevant features, such as lactate, succinate, and others, in a similar fashion to the continuous glucose monitors we already have – may replace mass spectrometry for the masses with a more targeted and precise approach.
I think the two approaches are going to go hand in hand, but I’m optimistic that sooner or later we’re going to start understanding what we should at least try in the real world.
Newsletters
Receive the latest analytical science news, personalities, education, and career development – weekly to your inbox.
