Mitochondrial Phenotyping Will Improve Drug Discovery

Why measuring mitochondrial function will transform how we develop, evaluate and personalise medicines for complex disease.

Imagine complex diseases like heart failure or Alzheimer's as a car with a failing engine. We try everything to keep the car moving: taking the load off, fixing leaks and blockages, adding more fuel, stopping it from overheating, even manually pushing it along the road. The one thing we don't do is fix the engine.

This engine is the mitochondria – the small parts of every cell that 'power' the body like an engine powers a car.

Right now, we are trying to fix an engine we can't see. There is no dependable way to measure mitochondrial energetics inside the body, leaving us without a clear picture of how mitochondria behave in health or disease. A dynamic, systems‑level measurement would be transformative, especially given how widely mitochondria are disrupted – from metabolic disease to heart failure and Parkinson's.

Moving from static to dynamic biomarkers

Only ~10% of compounds entering phase I ever reach approval, despite more than 5,000 candidates entering development in the past five years. Each successful drug costs ~$3 billion and takes 10-15 years to reach patients, resulting in hundreds of billions lost to failed R&D. Meanwhile, disease burden remains high: cardiovascular disease is still the leading global killer, and about 10% of people over 65 live with neurodegenerative conditions such as Alzheimer's or Parkinson's.

Static readouts are no longer good enough for drug discovery. These just give us a single snapshot of one biomarker at one moment in time. They report abundance, not function – cholesterol concentration, glucose levels, inflammatory proteins. Heart failure therapies are evaluated using cardiac troponin, which marks necrosis, not early functional impairment. They are rarely tissue-specific or dynamic, and ignore metabolic flux. In clinical trials, this translates into reliance on hard outcomes such as hospitalisation, mortality, and MACE, all of which occur late and are expensive to measure.

There are even examples of drugs that successfully corrected a static biomarker of disease but worsened the underlying physiology. This was the case for torcetrapib – a drug that improved cholesterol (lowering LDLs by 25% and raising HDLs by 72%), but actually ended up worsening mortality rates. The effect on the one biomarker did not show the overall systemic impact.

The challenge is that dynamic processes like metabolism are very difficult to measure in a whole body. This is a fundamental limitation of biological models; we need to break down complex systems into simpler ones to understand mechanistic changes, but we lose the overlapping, real-time interactions that shape physiology. For example, cardiac models often exclude autonomic input, even though hormones like adrenaline influence heart rate and contractility. This reductionism leaves us with models that miss out the integrated behaviour of living systems.

To overcome this challenge, we have to change our approach for finding drugs to 'function first' rather than building upwards from the molecular changes. The field is already moving in this direction, with growing emphasis on dynamic, systems‑level biomarkers. One of the most promising areas is real‑time measurement of mitochondrial energetics, a core determinant of cellular and organ‑level function.

Why a mitochondrial readout specifically?

An emerging view of complex diseases is as systemic, multi-organ conditions driven by interconnected biological pathways. These are fundamentally disorders of bioenergetic mismatch and metabolic reprogramming.

We are now at an exciting point where the importance of mitochondria in disease is becoming increasingly clear. These organelles are central to bioenergetics and metabolism; they are responsible for providing the energy we need to carry out all biological processes to keep us alive. And they are so much more than just the 'powerhouse' – they integrate signals from metabolism, inflammation, hypoxia, mechanical stress and ageing.

Their energetic state represents the combined output of many upstream biological pathways rather than a single molecular target. Therefore, a measurement of their function could be the ultimate dynamic biomarker of metabolism. Of course, complex diseases are multifactorial, with many systems shifting simultaneously. But mitochondrial energetics provide a uniquely valuable readout of the functional consequences of diverse biological changes while still providing a measurement that is scalable, high-throughput, and physiologically relevant. The clear, central role of these organelles across the body makes them a simple but biologically meaningful marker of health and disease.

And we need to build this functional readout now. The mitochondrial therapeutic landscape is evolving rapidly. Frontier modalities, such as mitochondrial transplantation and targeted manipulation of mitophagy, are moving from proof-of-concept towards early clinical evaluation, with a landmark accelerated FDA approval of the first mitochondrial-targeted drug elamipretide in 2025. The importance of targeting mitochondria is also being recognised on a national scale, with the £55M ARIA mitochondrial genome-engineering initiative starting this year. With these interventions, the need for quality control of mitochondria is clear.

What a dynamic mitochondria-specific biomarker unlocks

Understanding mitochondrial function in the body could provide an entirely new layer of insight into human physiology and therapeutic response.

We already have seen evidence that measuring energetics can predict clinical outcomes. In 1997, Neubauer et al. used magnetic resonance spectroscopy (MRS) to quantify the cardiac phosphocreatine‑to‑ATP ratio in patients with dilated cardiomyopathy and followed them for two years. Those with depleted energetic reserves had a 40% mortality rate, compared with 11% in patients with preserved energetics. This single measurement was able to predict high‑risk individuals far earlier than conventional markers. However, MRS has not scaled from here: it is technically demanding, expensive, and ultimately provides only an indirect, partially static readout of mitochondrial function.

A more direct, tissue-specific energetic readout could transform this. Tissue specificity unlocks a deeper understanding of how drugs affect different organs – where they deliver benefit and where they cause harm. Consider GLP-1 receptor agonists: a tissue-resolved energetic profile could show the relative impact on skeletal muscle versus fat tissue metabolism, helping disentangle therapeutic effects from potential risks.

Dynamic mitochondrial biomarkers could serve multiple roles across the drug discovery pipeline, including early efficacy readouts, dose-finding tools, go/no-go decision points, as well as predictors of long-term outcomes. This could provide an opportunity to intervene early enough to prevent overt disease rather than manage the symptoms. This is important for any drug that affects metabolism at a systemic level, but particularly as the first wave of mitochondrial-targeting compounds emerge.

Personalised medicine through bioenergetic stratification

Beyond improved biomarkers, functional mitochondrial phenotyping improves personalised medicine, stratifying patients by biology, not symptoms. This would allow us to identify responders and non-responders early, and redefine disease subtypes around energetic dysfunction.

Patients with the same disease diagnosis often have very different underlying biology. Two heart failure patients may both present with preserved ejection fraction (HFpEF) yet have completely different energetic phenotypes: one driven by impaired mitochondrial oxidation, another by microvascular inflammation. Traditional biomarkers would not be able to distinguish these subtypes because they do not measure function.

This leads to a challenging clinical trial scenario: a therapy that is highly effective in a biologically defined subgroup may appear ineffective when tested across a heterogeneous patient population. A drug that is beneficial in 50% of patients but has no effect in the remainder may fail to meet efficacy thresholds and therefore be abandoned altogether. If we could instead identify the patients whose disease is driven by mitochondrial dysfunction, this drug could be a targeted treatment with a much higher probability of success rather than a waste of billions of dollars.

Using these late-stage overt disease state biomarkers, such as mortality or hospitalisation, also means that trials often enrol patients who are too far along in disease progression or have too many comorbidities to respond to treatment. Measuring the efficacy of drugs based on their ability to reverse these symptoms at this point is too late. Most interventions at this stage can only slow further decline rather than restore the underlying system. This is the case with neurodegenerative diseases like Alzheimer's especially, where treatments can modestly slow progression but cannot reverse the disease trajectory.

An earlier toxicity readout

Whilst many drugs are being developed to target mitochondria, an even larger number are trying not to hit them. Late-stage drug failure is primarily due to liver and cardiac toxicity, and a major driver for both of these is mitochondrial toxicity. As much as 50% of all FDA black-box toxicity warnings are thought to be associated with drug-induced mitochondrial dysfunction. Toxicity arises through impaired respiration, mitochondrial DNA damage, impaired fuel utilisation, or excess reactive oxygen species. A classic example is doxorubicin, a commonly used chemotherapeutic agent. Doxorubicin gradually injures cardiac mitochondria, leaving many cancer survivors with heart failure years later because their hearts are no longer able to sustain ATP production. High energy organs like the heart and brain cannot tolerate disruptions to mitochondrial function, they fail as soon as ATP supply declines.

Since we lack a robust measurement of mitochondrial function in the body, the true number of promising drugs that failed in phase II due to unrecognised mitochondrial toxicity is likely far higher than we can prove. The fact that we cannot determine mitochondrial toxicity until so late in the drug discovery pipeline is an extraordinarily expensive problem. A preclinical mitochondrial function readout could flag harmful compounds much earlier, killing toxic drugs before they reach expensive human studies, and saving millions.

How to phenotype mitochondria? Recent advances

Phenotyping mitochondria has never been straightforward. I should acknowledge that we have many existing methods to study mitochondrial behaviour in vitro. These allow high throughput measurements, but are limited in their physiological relevance as they do not translate to a full body system. When it comes to in vivo, we still rely on surrogate markers like lactate, pyruvate, GDF-15, or FGF-15 – signals that reflect cellular stress rather than true mitochondrial function.

A better measurement platform is urgently needed, but it is inherently difficult to build. This is partly because the biology itself is hard – mitochondria are tiny, dynamic organelles within a cell that are challenging to access. In addition, the upfront cost for building hardware required for new measurement tools is high, and often requires significant technical expertise and complex infrastructure. These challenges mean that innovation in this space is lagging behind the growing recognition of mitochondrial dysfunction as a therapeutic target.

Dynamic energetic measurements have advanced, though at high cost, in the areas of high-resolution respirometry, spectroscopy and metabolic flux assays. These tools capture physiological changes as living, adaptive processes rather than a single snapshot, pinpointing exactly when function starts to decline. ³¹P MRS appeared the most promising, but was ultimately rejected by the FDA as a clinical readout due to logistical difficulties and inconsistencies across trial sites.

At the same time, human-relevant disease models are improving, with the development of organoids and engineered microphysiological systems offering more realistic environments. The main limitation of these models, however, is that they often retain an immature metabolic profile, and so their relevance to human disease is debated.

Imaging is advancing even faster. Emerging functional mitochondrial imaging techniques aim to map tissue‑specific energetics across organs, offering a systems‑level view of metabolic resilience and failure. Combined with integrated omics and AI‑driven analyses, these datasets are finally becoming interpretable at scale, and could provide an entirely new level of detail on mitochondrial function.

The success of drug discovery for complex diseases lies in energetic functional phenotyping

Mitochondria sit at the centre of systems biology, and capturing their biological function in a clinically scalable and sensitive way is difficult but important. A dynamic measurement of their function allows three major advantages for drug development: an improved pharmacodynamic efficacy biomarker, better patient stratification to match drugs to the correct disease biology, and earlier detection of mitochondrial toxicity.

We are at a crucial point in drug discovery. For years we have focused on fixing the body like a broken down car, but what will unlock the next generation of therapies is understanding how the engine works – through a dynamic measurement of mitochondrial function.

Solving this challenge could provide a universal physiological readout that bridges molecular mechanisms and systems-level biology, ultimately transforming how we develop, evaluate and personalise therapies for complex disease.