The Continuum
Decades of failed drug trials. For decades the field has debated whether knee osteoarthritis is one disease or several — a debate sharpened by repeated drug-trial failures. Molecular endotypes were the leading candidate answer: find them, match treatments to them, and the trials would finally work. But as the STEpUP OA authors themselves note, that hypothesis “has never been formally assessed” at scale, and the clinical phenotypes proposed in the literature “have not been validated as clinically useful stratification tools.” Then the largest proteomics study ever run on synovial fluid — led by Tonia Vincent at Oxford, published June 2026 — found something the data refused to cluster into: no distinct subgroups. Instead, a single continuous biological landscape, tracking obesity status, biological sex and inflammation.
That finding doesn't just shift the treatment strategy. It asks a harder question: How many times has your field been wrong about the load-bearing pillar? And how would you know?
Sample: 1,300+ patients with established knee arthritis + 36 healthy controls. Data: 7,596 SomaScan v4.1 aptamers (targeting 7,289 unique human proteins) measured in synovial fluid per person; between 5,278 and 6,558 aptamers survived quality filtering depending on the analysis. Hypothesis tested: If OA is really multiple distinct diseases, patients should cluster into separate protein profiles. What happened: The dots refused to cluster. They smeared into one continuous cloud. What the null can and cannot say: the authors state the analysis was “powered to identify several endotypes across the entire OA population and to detect two distinct endotypes when considering only non-advanced radiographic disease” — so it rules out large, common endotypes, not small or rare ones. They also flag that SomaScan use in synovial fluid is “still relatively new.”
Expected vs. Found
Multiple distinct diseases
The leading hypothesis: knee OA is several separate endotypes, each with its own biology — structural-damage, inflammatory and low-tissue-turnover subtypes have all been proposed in the phenotype literature. Find the boundaries, match therapies to boxes. STEpUP OA was built to test that hypothesis; the authors note it had never before been formally assessed at this scale. The trials keep failing because we're mixing incompatible diseases in one protocol.
One biological continuum
All 1,300 patients share a common molecular fingerprint — a shared core pathway of tissue injury and repair. Biological variation exists, but it's continuous, not categorical. The variation is continuous and tracks obesity status, biological sex and systemic inflammation (C-reactive protein) — not hidden disease subtypes. Age is a covariate the models adjust for, not a driver. The continuum itself is driven primarily by epithelial–mesenchymal transition.
The visual shape of the difference
Researchers expected to see this when they plotted each patient's protein profile:
What they actually found:
One continuous cloud. 1,300 patients spread along a gradient, not separated into boxes. The dots refused to cluster. That visual difference — from compartments to continuum — rewrites how you treat everyone carrying knee pain.
The Gatekeeping Moment
Tuesday, 2 June 2026 (the paper drops)
Perry et al. publish the largest proteomics study ever run on knee fluid. No evidence of distinct endotypes. One disease, continuous variation. The findings point to a different treatment strategy entirely.
Gatekeeping response (predictable): "It's one cohort. Other studies have found endotypes using different methods. This doesn't erase years of previous work. Maybe the protein signal is just nosier than we thought."
The opening question: Can independent labs replicate the finding using different patient cohorts, different proteomics platforms, different geographies? If yes, the foundation cracks. If no, Perry et al. becomes a methodological curiosity.
The reckoning: Decades of endotype-hunting would need re-costing, and trial designs premised on finding boxes would need rethinking. What would not need rewriting is clinical practice — the endotype model was a research programme, not a treatment protocol. And the authors keep stratification rather than abolishing it: they argue their pathways give “a rationale for stratification in clinical trial design,” on different axes. The gatekeepers who said "no" were doing their job 90% of the time. Just not this time.
The impossible position
Right now — this week, today — you cannot know whether this is a crank's overreach or a Copernicus moment. The data supports it, and Perry, Vincent and the STEpUP team stand behind it. The replication will decide. But for the next 12-24 months, the field exists in a state of genuine uncertainty, and the people who say "wait for replication" are being cautious in the right way and also possibly gatekeeper-ing a paradigm shift that's already real.
This is why paradigm shifts always look like they could be noise, right up until they aren't.
Update · 14 July 2026 — replication still pending
As of today, no rebuttal or Nature Communications “Matters Arising” has appeared, and no independent replication has yet landed. A July 2026 Bone Research review still advocates a multi-endotype model but does not engage this paper — the standing paradigm continuing, not a response to it. The finding remains a live challenge awaiting external replication. Note that STEpUP OA is itself internally replicated: cohorts were assigned a priori to a Discovery set (N=708) and a Replication set (N=653), and the continuum finding held in both. What is still outstanding is a different consortium, a different platform, a different geography. Source: Perry et al., Nature Communications, 2 June 2026, doi 10.1038/s41467-026-71632-4 (STEpUP OA consortium, Oxford).
Where do you stand?
What's your read on the STEpUP OA finding?
How confident are you?
What would move you?
The Swing and The Continuum
Cosmology (The Swing)
Expected: The universe is isotropic at large scales — no preferred direction. Matter is randomly clumped everywhere.
Found: A possible preferred direction in the large-scale structure — a "cosmic axis." When you slice the universe thin, it looks filamentary in a way a random mock doesn't.
The gatekeeper question: Is this a real anisotropy or just noise in a single thin slice? How many mock universes would you have to run to prove the universe isn't supposed to look like this?
Orthopedic Medicine (The Continuum)
Expected: Knee OA is multiple distinct diseases — separate endotypes, each with its own biology.
Found: One continuous biological condition with shared core pathways, variation tracking obesity status, sex and inflammation; the continuum driven primarily by epithelial–mesenchymal transition.
The gatekeeper question: Is this a real continuous distribution or just noise in one large proteomics study? How many independent cohorts would you have to run to prove that endotypes weren't real subtypes?
What this means
Paradigm shifts aren't rare. They're a feature. And you can't know from inside the moment whether you're watching a crank or a Copernicus. The only honest move is to ask: what would it take to move me? And then wait for the answer without claiming certainty you don't actually have.
See also: The Swing — cosmology's parallel moment · The Agency Lab — the dissociation in the clinic