When T-cell numbers persist, can immune recognition still narrow?

Thymic input, peripheral clonal expansion and cell composition jointly shape observable TCR diversity; neither a cell count nor a diversity score is a complete measure of immune function. We reanalysed public molecular counts from 25 older cardiac surgery patients. A positive association between a proxy for thymic output and diversity remained after standardizing sampling depth. Composition, measurement and functional endpoints constrain what that association means.

01Cell numbers, richness and evenness

Most conventional T cells recognize antigens through a receptor composed of alpha and beta chains. This article concerns those T-cell receptors, or TCRs. Cell division can produce more descendants of an existing receptor clone without creating new receptor sequences. More cells, more clonotypes and a more even distribution are therefore different changes. Human study and measurement Longitudinal evidence

Here, a clonotype is a sequence-defined type; identical sequences do not guarantee a single common cellular ancestor. This also differs from the driver-mutation clones discussed in clonal hematopoiesis. Clonal expansion is a normal part of an immune response.

02What is actually in the data?

Sandhar and colleagues' 2025 study obtained tissue during cardiac surgery and measured blood T-cell subsets. Thymus-like structures persisted in mediastinal fat in some older patients. The study also examined a blood proxy enriched for recent thymic emigrants, abbreviated RTE. Primary study

We obtained and checked the 25 public whole-blood TCR workbooks and clinical workbook. All sequenced donors link uniquely to the main clinical table: ages 54–84, with 16 men and 9 women. Each has separate alpha and beta data. Two chains from each person do not create 50 independent participants. Public data

These are UMI-corrected RNA molecule counts. Unique molecular identifiers help distinguish original molecules and correct amplification and sequencing errors. A cell can contribute multiple RNA molecules, so molecular abundance is not cell abundance. Rearrangement abundance and average reads per UMI cluster are also different fields. Our analysis begins with processed counts and does not rerun error correction on raw sequencing reads. Author software documentation

With potentially productive V/J/nucleotide-CDR3 combinations defining clonotypes, samples contain approximately 20,000–91,000 alpha molecules and 33,000–162,000 beta molecules. Such differences in sampling depth matter. Productive annotation is only a sequence-level classification; it does not verify correct pairing, cell-surface expression or recognition of an antigen.

The source audit exposed a consequential detail. The final row of one alpha workbook contains only the number 33,321, no sequence identifier, and exactly equals the sum of preceding abundances. Counting that total as a clonotype seriously distorts entropy. Removing it allows 49 of the 50 source-figure entropy values to be reproduced from valid sequence rows. One beta file still differs from the figure data; we retain the file result and assess whether omitting that donor changes the conclusion. Workbooks Supplement and figure values

The reproducible source entropies include nonproductive rearrangements. Our primary analysis separately retains sequences annotated as potentially productive, so it addresses a somewhat different measurement.

03Does the association remain at the same sampling depth?

We rarefied every donor-chain repertoire to 15,000 molecules, drawing without replacement 200 times and calculating effective Shannon diversity. Inferential resampling used donors, not thousands of sequences treated as independent participants.

After Holm correction for the two primary associations, permutation p values were approximately 0.0023 for both chains. This supports a positive association in this patient sample. The intervals remain broad: sample correlations are not precise population parameters or estimates of a treatment effect from increasing thymic output.

The direction persisted at 5,000 and 10,000 molecules. Across leave-one-donor-out analyses, alpha correlations ranged approximately 0.57–0.69 and beta correlations 0.57–0.70. Excluding the donor with discrepant source entropy gave approximately 0.61 and 0.62, respectively. The finding therefore does not depend on a single donor.

Age did not show a clear association with these measures in this cohort. All participants were older surgical patients, however; a small sample with a restricted age range cannot establish that age has no effect on TCR diversity across life.

04Equal molecule counts do not mean equal sample completeness

When a few clones occupy much of a repertoire, shallow sequencing repeatedly captures them. When many clones are rare, the same number of molecules provides a more scattered sample.

Approximate completeness based on singleton counts ranged from 0.29 to 0.96 for alpha and 0.38 to 0.94 for beta. This concerns molecular probability mass, not the percentage of all a person's clonotypes that have been identified. Coverage method

As a secondary comparison, we standardized estimated coverage to approximately 25%. RTE correlations were approximately 0.57 and 0.54 for alpha and beta, preserving the direction. But the required depths differed enormously: 38–23,437 alpha molecules and 44–36,371 beta molecules. Estimates based on very small molecular samples have greater sampling variability. This comparison cannot remove uncertainty about unobserved sequences.

The two approaches serve different purposes. Equal depth controls the amount sampled; equal estimated coverage aims to compare similar sample completeness. Neither reconstructs all T cells in the body or supplies the true paired alpha-beta repertoire.

05The RTE proxy itself contains composition information

RTE percentage in this study approximately equals the naive fraction of CD4 cells multiplied by the CD31-positive fraction within those naive cells. It is not a direct measurement of cells leaving the thymus per hour.

Part of the RTE-diversity association may therefore reflect the mixture of naive and memory cells. We fitted small models containing RTE alone, RTE with age and sex, or RTE with naive-CD4 proportion, using HC3 robust intervals.

The positive association retained support after adding age and sex. After adding naive-CD4 proportion, the effective-diversity ratio associated with a 10-percentage-point higher RTE value was approximately 1.23 (95% interval 0.92–1.66) for alpha and 1.23 (0.86–1.77) for beta. Both intervals crossed one.

This does not establish that the thymus has no role. Composition can be an alternative explanation or lie on a pathway from thymic activity to peripheral cells. Including it changes the question. The 25-person whole-blood dataset also lacks complete absolute CD4/CD8 composition and CMV status, preventing isolation of a definite causal thymic effect.

The main clinical sheet and a second selection sheet disagree about smoking labels for eight former smokers. We did not force those labels into a common coding and claim to have controlled smoking confounding.

06Input, expansion and loss can change different aspects of diversity

To separate these processes, we constructed a hypothetical finite-population model. Total cell number stays at 10,000, initially divided equally among 1,000 clonotypes. In each replacement round, some cells enter as new clonotypes and the remainder inherit peripheral parents' clonotypes. Selected scenarios give a few clones a reproductive advantage.

Under the specified conditions, neutral maintenance, reduced new input and selective expansion produce different outcomes. With persistent selection, restoring input can increase the number of clonotypes still present without restoring evenness to the level observed without a selective advantage. A parameter grid also shows that outcomes depend on the relative strength of input and selection.

This is not a simulation of human age: a replacement round is not a year, and input fractions and fitness advantages were not fitted to the 25 donors. Fixed size, synchronous replacement and distinct singleton entrants are simplifying assumptions.

A separate illustration holds within-naive and within-memory clonal distributions fixed while changing their mixture. Overall effective diversity changes substantially. It additionally assumes no shared clonotypes between compartments; real naive and memory cells can share receptors, so this curve cannot be directly applied to them.

07How human and mouse studies constrain the mechanisms

Qi and colleagues' 2014 study sorted cells from four healthy young and five older donors and used multiple cell aliquots from each person to estimate richness. Older participants retained a large naive sequence repertoire, alongside substantial expansion of selected clones. Some expanded clones also proliferated more readily in response to homeostatic cytokines in vitro. This supports a role for peripheral selection alongside thymic involution. Qi study and actual supplement

Egorov and colleagues also found sequence changes in sorted naive cells, including CDR3 length and amino-acid composition. A computational interaction-strength metric derived from those features is not measured antigen affinity. Persistence of fetal-origin clones, peripheral selection and antigen-associated movement out of the naive pool remain competing explanations. Egorov study

Sun and colleagues' 2022 longitudinal analysis examined 30 donors sampled approximately nine years apart. Changes differed by cell subset and receptor chain, with naive CD8 cells particularly affected. Some richness and paired-receptor quantities were model projections, not a direct census of every receptor in the body. Repeated measurements provide stronger time information while remaining observational. Longitudinal study

Large clones also need not displace every other clone. In the context of CMV, Lindau and colleagues found that parts of the CD8 pool could expand; conditional distributions after removing large clones resembled those in seronegative participants. The deeply characterized older subset included only eight people, with absolute counts available for six. This does not establish absence of other CMV harms. It does show why fixed carrying capacity is a model assumption, not a universal biological rule. CMV study

A 2026 mouse transplantation study directly tested thymic environment. Older Rag1-deficient thymus grafts generated a less diverse peripheral repertoire than young grafts in athymic recipients. That supports a contribution from the thymic environment. However, older Rag1-deficient tissue had long lacked normal lymphocyte feedback, and lymphocyte reconstitution after transplantation is an unusual setting. It does not validate thymus-regeneration treatment in humans. 2026 mechanistic study

08Functional measurements exist, but answer different questions

The 2025 study contains more than sequence data: it includes cell stimulation, influenza antibodies and an observational analysis of postoperative respiratory infection. Yet mature naive CD31-negative cells showed greater proliferation and some cytokine responses to synthetic-superantigen stimulation. Being closer to recent thymic output does not mean a stronger response in every assay. Primary study

Higher-RTE groups had higher antibody signals in some time groups, but this is not a paired receptor-recognition experiment in the same people. The methods also select samples by an antibody-signal threshold. Public tables lack donor identifiers needed to recover repeated follow-up and a reconstructable prevaccination baseline. We calculate descriptive summaries only: these points do not establish 78 independent people or individual antibody-decay curves. The duplicated Low group labels in the source workbook were resolved only after checking the published legend and numerical distributions. Figure values and supplementary methods

Postoperative infection findings are observational. Outcome denominators and modelling details in the supplement are insufficient for us to reconstruct a complete infection-risk comparison. They cannot be directly converted into infections prevented by increasing RTE percentage.

Receptor pairing introduces another boundary. The same two alpha and two beta chains can form two paired types or four: identical marginal chain counts can hide different complete repertoires. Matching individual chains to an antigen database likewise cannot replace actual pairing, HLA context and functional experiments.

Finally, the paper's mortality-risk associations use methylation measures such as GrimAge, not observed survival follow-up. Associations between proxies do not establish life extension from an immune process.

09What this reanalysis supports

In these older surgical patients, the positive association of the RTE proxy with effective TCR diversity survives equal molecular depth, alternative clonotype definitions and omission of individual donors. But RTE contains composition information, molecular counts do not recover total cells, and single-chain sequences do not recover complete antigen-recognition capacity.

A stronger next study would track absolute cell counts, sorted subsets, paired receptors and actual antigen responses together, then examine their relationship with infections in the same people. Our calculations and conditional models explain how these layers can separate. They do not establish personal immune age or the clinical effect of a thymic intervention.

Download code, numerical inputs, figures and verification results