Which T cell responses correlate with COVID-19 risk after a second booster?

What the COVAIL second-booster analysis measured

The COVAIL trial randomized adults who had already received a primary vaccination series and one ancestral booster to a second booster on Day 1, drawn from three vaccine platforms: mRNA-1273 (Moderna), BNT162b2 (Pfizer-BioNTech), and a pre-S DTM AS03 protein vaccine (Sanofi). Arms differed in whether the insert targeted wild type, Beta, Delta, Omicron BA.1, or Omicron BA.4/5, and in whether they were monovalent or bivalent. Follow-up coincided with the Omicron BA.2 and BA.4/5 waves.

The immune measurements came from a validated 27-color intracellular cytokine staining assay. Peripheral blood mononuclear cells were stimulated for six hours with overlapping peptide pools covering Spike S1 and S2, Spike BA.4/5 S1 and S2, and Nucleocapsid, then stained for cytokines and activation markers. The readouts used here are the percentages of CD4+ or CD8+ T cells producing IFN-γ and/or IL-2 in response to those peptides, plus a COMPASS functionality score summarizing how many distinct cytokine combinations an individual's antigen-specific T cells express.

Participants were classified as non-naive if they self-reported a prior SARS-CoV-2 infection or had detectable anti-Nucleocapsid antibodies at Day 1; otherwise they were naive. This definition matters because it is the variable that splits every correlate result in the paper. The correlates analysis focused on one-dose mRNA recipients and counted COVID-19 cases occurring at least 7 days after the Day 15 visit through 188 days after it: 22 non-naive cases and 122 naive cases, against 117 and 209 non-cases respectively.

Risk fell per 3-fold CD4 increase only in non-naive participants

Among one-dose mRNA recipients, Day 15 CD4+ T cell IFN-γ and/or IL-2 responses to Spike BA.4/5 correlated inversely with COVID-19 risk in non-naive participants, with a hazard ratio of 0.62 (95% CI 0.39, 0.99) per 3-fold increase in response. In naive participants the same measure gave a hazard ratio of 1.03 (0.81, 1.32), an interval that includes no association. The unit here is a 3-fold change in the percentage of responding CD4+ T cells, not a one-unit change, and the endpoint is COVID-19 over 188 days after Day 15.

The contrast is not explained by the booster failing to raise responses in naive participants. Across all one-dose arms, Day 15 Spike BA.4/5-specific IFN-γ and/or IL-2 geometric means ranged from 0.12% to 0.27% in naive participants and 0.12% to 0.37% in non-naive participants, and positive vaccine-response proportions ranged from 31.6% to 80.6% and 8.9% to 85.0% respectively. The paper's discussion states directly that the CD4+ correlate appeared in non-naive but not naive participants despite similar post-booster response magnitudes in both groups.

That combination is what makes the result informative rather than a simple statement that more T cells are better. If the same measured response level carries different risk meaning depending on infection history, the assay value is not acting as a self-contained protective quantity. The authors interpret it as a proxy for unmeasured protective immune mechanisms operating after SARS-CoV-2 infection, which is their explanation rather than a directly measured mechanism in this analysis.

CD4 responses and neutralizing antibodies predicted independently

The paper also asked whether T cell responses add information beyond neutralizing antibody titer. In a Cox bivariate model for non-naive participants, Day 1 Spike CD4+ T cell response and Day 15 neutralizing antibody titer each independently predicted COVID-19, with hazard ratios per standard deviation of 0.81 (0.68, 0.96) for the CD4+ measure and 0.43 (0.27, 0.69) for the titer, and a co-correlate p < 0.001. Note that these two hazard ratios use a different unit from the 0.62 figure above: one standard deviation rather than a 3-fold increase, and the CD4+ measure is taken at Day 1 rather than Day 15.

Prediction models told a related but distinct story. In the non-naive cohort, the top-performing superlearner model combined Day 15 neutralizing antibody titer against BA.1 with Day 1 CD4+ T cell IFN-γ/IL-2 response to Spike BA.4/5, reaching a cross-validated AUC of 0.726 (95% CI 0.580, 0.836). The best model containing a T cell response but no neutralization titer reached only 0.592 (0.446, 0.724) and ranked 16th, while a baseline-factors-only model was not predictive at 0.500. In the naive cohort, the top model used neutralizing antibody titers at both Day 1 and Day 15 (CV-AUC 0.632; 0.552, 0.705), and the top four models all included baseline factors and neutralizing antibody titer without T cell responses.

Read together, these results separate two questions that are easy to conflate. One is whether a marker is statistically associated with risk; the other is how much a prediction model gains from adding it. Neutralizing antibody titer carried most of the discriminative weight in both cohorts, while the CD4+ contribution was clearest in non-naive participants and appeared in combination with titer rather than instead of it.

Polyfunctional CD4 subsets marked lower risk in non-naive participants

The inverse association was not spread evenly across all Spike-specific CD4+ cells. The paper reports that subsets co-expressing three or more cytokines or markers, most commonly CD154, IFN-γ, IL-2, and TNF, appeared to drive the observed inverse correlations of antigen-specific CD4+ T cell responses with COVID-19. Among non-naive participants, high Spike-specific CD4+ T cell responses and high neutralizing antibody titers both marked low risk, as did polyfunctional Spike-specific CD4+ T cells expressing at least three cytokines or markers.

This is a statement about which functional subsets tracked with the outcome, not a demonstration that polyfunctionality causes protection. The COMPASS functionality score summarizes the proportion of distinct antigen-specific T cell subsets an individual produces across eight cytokines and markers (CD154, IFN-γ, IL-2, IL-4, IL-5/13, IL-17a, IL-21, and TNF). The paper reports that Day 15 CD4+ Spike BA.4/5 functionality scores exceeded their Day 1 counterparts in each one-dose arm for both naive and non-naive cohorts, while the CD8+ increase from Day 1 to Day 15 was attenuated or absent, particularly in the protein vaccine stage.

The CD8+ results sit at the edge of significance rather than matching the CD4+ pattern. The discussion describes Spike-specific CD8+ T cell responses as borderline significant inverse correlates of booster-proximal COVID-19, with the associations weakening over the 6 months of follow-up. Nucleocapsid-specific CD8+ responses could not serve as correlates because Day 1 positive response rates in naive participants were 1%, 0%, and 0% across the three stages, failing the prespecified eligibility criterion, and because none of the study vaccines contained Nucleocapsid.

At the same CD4 level, prior infection still separated risk

The most direct test of whether the T cell measure is a proxy rather than the protective mechanism itself comes from comparing risk at matched response levels. The paper reports that at the same high Day 15 CD4+ Spike-specific T cell levels, risk was lower for non-naive than for naive participants. In other words, two participants with the same measured percentage of responding CD4+ T cells did not have the same subsequent COVID-19 risk if their infection histories differed.

This comparison is what supports the authors' interpretation that these T cell responses are a proxy for unmeasured protective immune mechanisms operating after SARS-CoV-2 infection. The measured assay value is a marker that travels with protection in one group, not a quantity that fully carries protection across groups. The paper does not identify which additional mechanism produces the residual difference; that remains an inference from the matched-level comparison rather than a measured pathway.

A second asymmetry points the same way. The discussion notes that where circulating Spike-specific CD4+ responses were similar between naive and non-naive participants after the second booster, CD8+ responses remained higher in non-naive participants, a pattern the authors describe as consistent with a more persistent influence of prior infection on circulating Spike-specific CD8+ cells. Both observations place prior infection as a modifier of what a given T cell readout means for risk.

Where these correlates apply and where they do not

Several boundaries qualify the results. The correlates analyses were restricted to one-dose mRNA second-booster recipients, so the hazard ratios and prediction AUCs do not automatically extend to the protein vaccine arm, even though T cell immunogenicity was characterized across all three manufacturers. The endpoint is COVID-19 from 7 days after Day 15 through 188 days after Day 15, during the Omicron BA.2 and BA.4/5 waves, so the numbers describe that follow-up window and that variant context rather than a longer horizon.

The non-naive group is small in absolute terms: 22 COVID-19 cases against 117 non-cases, compared with 122 cases and 209 non-cases in the naive group. The 95% confidence interval for the non-naive CD4+ hazard ratio of 0.62 runs from 0.39 to 0.99, close to the null at its upper bound. The naive result, by contrast, is estimated with more cases and centers near no association. All correlates models adjusted for a baseline risk score and a standardized force of infection score, which means the reported hazard ratios are conditioned on those covariates rather than being unadjusted descriptions.

The naive and non-naive labels rest on self-reported prior infection or detectable anti-Nucleocapsid antibodies at Day 1, so the grouping captures documented or serologically evident prior infection rather than every possible prior exposure. The paper's own framing of the CD4+ response as a proxy for unmeasured mechanisms is consistent with this: the classification is a stand-in for immune history, and the analysis shows that the stand-in changes the meaning of the T cell readout.

Prior infection status decides how to read the CD4 result

The practical priority from this paper is a conditional one. A Day 15 Spike-specific CD4+ T cell measurement after a second mRNA booster is informative about subsequent COVID-19 risk in previously infected recipients, where a 3-fold higher response corresponded to a hazard ratio of 0.62, and where CD4+ response and neutralizing antibody titer each contributed independently. In SARS-CoV-2 naive recipients, the same measurement at the same post-booster magnitudes carried essentially no risk information, with a hazard ratio of 1.03.

For prediction specifically, the paper's model comparisons put neutralizing antibody titer first in both cohorts, with the best non-naive model pairing Day 15 BA.1 titer with Day 1 CD4+ response and the best naive models relying on titer and baseline factors without T cell responses. The CD4+ measure is therefore best read as an addition to antibody information in previously infected people, not as a standalone substitute for it in either group.

The matched-level comparison is the finding to keep in view: equal Day 15 CD4+ Spike-specific T cell levels came with lower risk in non-naive than in naive participants. That is why the paper's answer to which T cell responses correlate with COVID-19 risk after a second booster is not a single threshold, but a readout whose meaning depends on whether the recipient was infected before vaccination.

Sources used in this answer