Issue #55 // Molecular Moonlighting
On Biological Context, Networks, and the Illusion of Molecular Meaning
Liked this piece? If so, you’ll want to check out other articles in Sequence & Destroy’s Wearable Technology & Biometrics collection. Additionally, you can show your support by tapping the 🖤 in the header above. It’s a small gesture that goes a long way in helping me understand what you value and in growing this newsletter. Thanks!
Issue № 55 // Mølecular Møønlighting
I spend my research time shuttling between two seemingly unrelated worlds— biodefense and precision cancer therapy. One day I’m building models to predict the physiological effects of novel pathogen exposures. The next, I’m trying to identify network biomarkers that predict how well a tumor will respond to treatment. You might expect these two domains to speak different molecular languages, but they don’t. They speak the same language— and that's exactly the challenge.
Last month, while reviewing protein expression profiles from COVID-19 patients, I noticed something that made me pause. Interleukin-1/6, TNF-α and TNF-β, and C-reactive protein — the classic "fever proteins”— were dramatically elevated compared to our healthy control subjects. Standard pyrogenic response, exactly what you’d expect. But, then I pulled up data from micro-dissected tumor samples in our cancer cohort, and there they were again. The same fever proteins, all up-regulated to similar degrees, in patients who had never run a temperature above 98.6 degrees Fahrenheit. These molecules, which are all well known players in the febrile response, were moonlighting in oncology.
The plot thickened when I looked at the inverse relationship. Several well-characterized oncoproteins— P13K, STAT3, EGFR, MYC, AKT— showed significant up-regulation in COVID-19 patients, yet none of these patients had cancer. Is this surprising? Not necessarily. We've been cataloging genes and proteins like a Victorian naturalist collecting insects, pinning each one down with a single functional label. "This is a fever protein." "That is an oncogene." But, these labels reflect the limitations of our observational framework, and the context in which a given gene or protein was first cataloged rather than fundamental biological truth.1
The uncomfortable reality is that proteins don't inherently "do" anything. They exist in dynamic networks where function emerges from context, relationships, and timing2. A kinase that phosphorylates substrate A during cell cycle progression might phosphorylate substrate B during stress response, creating entirely different downstream cascades. The protein hasn't changed—its relational context has.
This realization has profound implications for how we think about biomarkers, both in clinical medicine and the increasingly popular world of consumer health tracking. The quantified self movement has embraced the seductive simplicity of single-molecule monitoring— track your cortisol for stress, measure your CRP for inflammation, monitor your glucose for metabolic health3. But these approaches commit what I'm starting to think of as the fundamental attribution error of molecular biology—assuming that proteins have stable, context-independent meanings.
Take cortisol, the poster child of stress hormone tracking. Wearable devices now promise to monitor your stress levels by tracking cortisol sweat. But cortisol's biological meaning depends entirely on circadian timing, co-occurring hormones, receptor sensitivity, and environmental context. For example, morning cortisol elevation might signal normal circadian function while evening elevating could indicate circadian misalignment. Alternatively, a late day bump in cortisol could be from an evening workout, reflecting a normal adaptive response. The same molecule, the same concentration, completely different biological implications.
The aging biomarker industry has fallen into similar traps. Companies tout panels measuring inflammatory cytokines, oxidative stress markers, NAD+ levels, or telomere length as comprehensive aging assessments. But aging isn't a disease with a uniform pathological signature that can be defined by a single set of biomarkers. This isn't to say that individual biomarkers are useless. Some biomarkers are strong signals of physiologic states across contexts, just as some proteins maintain relatively stable functions across contexts. For example, insulin generally promotes glucose uptake and hemoglobin reliably carries oxygen. But these exceptions prove the rule. Most proteins are contextual actors in biological networks, not agents with fixed properties.
Consider the implications of this for drug development, where therapies have historically been designed to act on single targets. Block this receptor, inhibit that enzyme, etcetera. If a protein’s function emerges from its context within a network, then might effective interventions require modulating the network state rather than individual molecules? And if so, could this explain why so many promising single-target cancer drugs show initial efficacy but eventually fail as tumors "rewire" their networks?
In my biodefense work, I’ve seen how pathogens exploit this contextual flexibility, hijacking entire network modules rather than individual host proteins. The same is true in cancer. The difference between tumor cells with high HER2 expression and a healthy cells from the same patients isn’t just the delta in HER2 levels — it’s the re-wiring of protein co-expression and interaction networks. Viruses and tumors succeed by understanding something we’re still learning, which is that biology is fundamentally relational.
We need to embrace this complexity rather than simplifying it away with easily sellable stories. Instead of asking whether an elevated protein, metabolite, or biomolecule in general is "good" or "bad", we need tools that can decode what functional role it’s playing in real-time4. For consumers, this shift demands a more nuanced relationship with biomarker data, especially as direct-to-consumer blood tests and health panels become increasingly popular. Seeing a change in one biomarker, like C-reactive protein, shouldn’t be inherently alarming5— it should be a conversation starter. What other signals are present? What's the temporal pattern? How has the network changed over time? The meaning emerges from the pattern, not the individual measurement.
The quantified self movement promises that measurement leads to insight, and insight leads to optimization. But measurement without context is just sophisticated numerology. True biological insight requires understanding not just what we're measuring, but how those measurements relate to the dynamic networks that actually govern health and disease. The oncogenic proteins elevated during fever aren’t confused— we are. But, that confusion, when properly embraced, could be the beginning of true understanding.
The gene Notch1 provides a perfect example. The "Notch" gene, including its homolog "Notch1," was first named in studies of the fruit fly Drosophila melanogaster with wing mutations, specifically the "notched" wing phenotype. Yet, in humans, Notch1 plays an important role in cell cycle regulation.
Going back to the IL-6 example, we see that IL-6 can prevent cancer (in the context of acute IL-6 production following exercise), promote cancer (chronically elevated IL-6 creates a pro-tumorigenic microenvironment), or can be involved in the febrile response among many other things. The same context dependency can be found at the signaling pathway level as well. Take the PI3k/Akt/mTor pathway, which is frequently enriched in cancer (and is associated with tumor "invasion"), yet is also activated by resistance training and stimulates muscle hypertrophy (and is associated with an anti-tumorigenic response in this context).
I’ve written about the pitfalls of this approach and alternatives (distributed sensing + network biomarkers) in previous articles Distributed by Design and Network Biomarkers Will Define 21st Century Medicine.
In practice this could mean integrating expression data with information about post-translational modifications, sub cellular localization, and temporal dynamics to understand not just what proteins are present, but how they're relating to each other.
One of the main limitations of direct-to-consumer lab tests (as they are currently sold) is their lack of nuance. Often, continuous biomarker measurements get turned into categorical measurements, like red/yellow/green light representations, that fail to capture changes in physiological states in a meaningful way. While this type of simplification, or compression of information, makes the product easier for the consumer to use (in turn making it more sellable), it removes the necessary context needed for true understanding. An alternative approach could be to not only display individual biomarkers, but also how they change over time (and how these rates of vary over time, which I discussed in The Wolf Of Wetware), how they are related to other biomarkers (including cause-and-effect relationships) and how these relationships change over time. This information could then be fed into a reasoning model, which could act as an interpretation layer between the raw measurements and the end user.




I'm curious. It is already a theory or concept to this approach? I would like to learn more about it. Biology is fascinating, thank you for sharing
Hey, great read as allways. So, P13K connection... how?