Stabilizing Selection Appears Widespread Across the Human Plasma Proteome
Stabilizing selection is often discussed as a force that keeps traits near an evolutionary optimum, but its reach across human molecular phenotypes has been difficult to measure directly. In this preprint, investigators turned that question toward plasma proteins and asked whether protein abundance carries signatures of constraint across both evolutionary time and contemporary fitness-related outcomes. The answer was strikingly consistent across methods: signals pointing to stabilizing selection were found for a large share of the proteins examined.
The analysis covered 2,923 plasma proteins in up to 54,000 UK Biobank participants. Two complementary frameworks were used. One captured historical selection through the frequency-dependent architecture of protein levels using genome-wide common variants with LDpred2 and a newly developed rare loss-of-function burden framework. The other examined contemporary phenotypic consequences by testing quadratic relationships between protein abundance and three proxies of fitness: reproductive success, somatic health, and parental longevity.
A single biological theme ran through both arms of the analysis. Protein levels that drift too far from their typical range appear to carry costs.
Convergent Signals From Common and Rare Variant Architecture
The common-variant analysis and the rare-variant analysis did not merely point in the same direction; they did so with remarkable uniformity among the significant estimates. Every significant estimate in the common-variant analysis supported stabilizing selection. In the rare-variant-based analysis, 92.4% of significant estimates supported the same pattern. This is an important distinction because the two approaches reflect different genetic timescales and different ways of perturbing protein expression. Common variants are expected to reveal more subtle, long-running shifts in allele frequencies, while rare loss-of-function burden captures stronger disruptions that can expose dosage sensitivity.
The convergence across those lenses suggests that plasma protein abundance is not a neutral trait space. Many proteins seem to occupy narrow operating windows, and departures from those windows may be pruned by selection over generations. In practical terms, protein homeostasis looks less like a broad plateau and more like a landscape with steep edges.
Contemporary Fitness Proxies Reinforce the Same Pattern
The phenotypic analysis added another layer of support by linking protein abundance to three measures tied to fitness. Among significant estimates, 99.8% supported stabilizing selection for somatic health, 97.7% for parental longevity, and 89.3% for reproductive success. Those values indicate that deviation in protein concentration is not simply a molecular curiosity; it lines up with outcomes that matter for survival and transmission.
Quadratic associations are especially informative here. A linear model would miss the possibility that both low and high levels can be unfavorable, but a quadratic pattern can capture precisely that U-shaped or inverted-U-shaped biology. For plasma proteins, that form is what stabilizing selection predicts: too little function may compromise physiology, and too much may be equally costly because it can disrupt signaling, coagulation, immune balance, or metabolic control. The bedside analogy is simple even if the study is not clinical: the body often needs the right dose of a protein, not just any dose.
Rare-Variant Estimates Track Independent Measures of Constraint
The rare-variant-based estimates aligned with several independent markers of evolutionary constraint. Their correlation with health-score-derived coefficients was ρ = 0.36 with p = 5.2 × 10⁻⁶. They also correlated with protein heritability at ρ = 0.20, p = 5.2 × 10⁻⁵, with loss-of-function intolerance at ρ = −0.19, p = 6.1 × 10⁻⁵, and with cross-species transcript evolutionary rate at ρ = 0.15, p = 2.4 × 10⁻².
Those relationships are biologically coherent. Proteins that are more constrained by biology and evolution should be less tolerant of disruption, and the observed direction of association fits that expectation. The negative relationship with loss-of-function intolerance is especially intuitive: proteins that resist damaging variation are the ones most likely to sit under tighter selection. At the same time, the modest size of the correlations suggests that no single metric fully captures the underlying biology. Stabilizing selection emerges here as a layered phenomenon, not a one-variable explanation.
Why Protein Dosage Matters for Human Biology
The study places protein dosage at the center of human fitness. Many circulating proteins act in narrow physiological ranges because they are embedded in tightly regulated pathways. Coagulation factors, cytokines, transport proteins, hormones, and enzymes can all become problematic when their abundance shifts away from equilibrium. A small perturbation may alter binding kinetics, receptor occupancy, feedback loops, or proteolytic turnover. Over generations, selection would be expected to preserve protein levels that support function while discarding alleles that push abundance too far in either direction.
This framework also helps explain why protein-level constraint may be easier to detect than one might expect from sequence data alone. Selection does not need to eliminate every variant; it only needs to keep the phenotype close enough to its optimum that large deviations remain rare. The present analysis suggests that human plasma proteins are a prime example of that principle.
What the Triangulation Adds Beyond a Single Test
The value of the study lies in triangulation. Historical selection and contemporary fitness consequences are not interchangeable, yet they converged on the same broad conclusion. That convergence strengthens the inference that stabilizing selection is a pervasive force shaping the human plasma proteome. It also demonstrates a methodologic path forward: combining genetic architecture with phenotype-based fitness proxies can connect long-term evolutionary constraint with measurable human variation.
This is still a preprint, so the findings remain outside formal peer review. Even so, the dataset is large, the approach is biologically grounded, and the internal consistency across common variants, rare variants, and external constraint measures is difficult to ignore. The main unanswered issue is not whether stabilizing selection appears present, but how differently it acts across specific protein classes, pathways, and disease-relevant contexts. The next step will be to determine which proteins are most tightly constrained, where compensatory biology can absorb variation, and how these evolutionary signatures map onto human disease risk when protein abundance shifts out of range.