Do immune correlates of protection depend on exposure frequency?

The protection threshold moved rightward with transmission

The paper's central measurement is a shift in where protection appears on the antibody scale. Fitting scaled logistic models separately to low- and high-transmission periods, the authors estimated the midpoint of the antibody–infection risk curve — the titer at which observed risk sits halfway along its fitted decline — at roughly 3.5 log10 BAU/mL during low transmission and roughly 5.0 log10 BAU/mL during high transmission. In other words, a titer that looked strongly protective in a quiet period corresponded to weaker observed protection when transmission was intense.

That midpoint is a fitted summary of the observed antibody–risk relationship within a specific exposure window and surveillance population, not a biological cutoff. The authors state this explicitly, and the same caution applies to the risk curves themselves: they describe infection probability over the enrollment window conditional on antibody level and transmission group, not susceptibility to a single infectious contact.

Infected people had higher titers when transmission was high

The clearest empirical signature is a narrowing gap between infected and uninfected participants. During low transmission (biweekly PCR positivity below 15%), PCR-positive participants had a geometric mean titer of 29 BAU/mL (95% CI 9–95) versus 427 BAU/mL among PCR-negative participants, a 14.7-fold difference. During high transmission (positivity at or above 15%), the positive group's GMT rose to 105 BAU/mL (72–152) while the negative group's fell to 260 BAU/mL, shrinking the gap to 2.5-fold.

The direction of the negative group's titers matters for interpretation: background antibody levels were actually higher in the low-transmission stratum, so the narrowing is not simply everyone's titers rising together. A linear model on log10 titers with a PCR-status by transmission interaction confirmed the attenuation (β = 1.77, p = 0.0047). The authors' explanation is selection through exposure: when exposures are few, mostly the least protected get infected; when exposures are many, even moderately protected people accumulate enough exposure opportunities to be infected, compressing the observed difference across antibody levels.

Febrile illness surveillance supplied paired antibody and PCR data

Participants aged 2 years and older presenting with undifferentiated fever, or new-onset anosmia or ageusia, were enrolled prospectively at two acute febrile illness surveillance sites in the Dominican Republic between March 2021 and August 2022. Of 2,271 enrolled, 1,828 met inclusion criteria; 62.1% were female, median age was 31 years (IQR 21–45), and 59.3% had received at least two COVID-19 vaccine doses. Anti-spike antibodies were measured with the Roche Elecsys assay and reported in BAU/mL aligned to the WHO International Standard; acute infection was assessed by RT-PCR.

Because antibody and infection status were captured at a single enrollment visit, the design links measured titers to infection detected at that moment rather than following participants forward for new infections. Analyses were restricted to April 23, 2021–August 17, 2022, when each epidemic wave was dominated by a single variant, and transmission intensity was proxied by biweekly percent PCR positivity. Sensitivity analyses used a lagged positivity proxy and a leave-one-out estimate excluding the index participant, addressing dependence between individual outcomes and the surveillance stream that generated the transmission measure.

Higher antibody quartiles still meant lower infection odds

Before examining transmission effects, the study reproduced the expected antibody–protection relationship. After adjusting for age, vaccine history, time since last vaccination, predominant circulating variant, and community transmission intensity, participants in the highest anti-spike antibody quartile had 84% lower odds of PCR-confirmed infection than those in the lowest quartile (OR 0.16; 95% CI 0.10–0.26), with a consistent gradient across quartiles.

The vaccine analysis adds a useful distinction. In models without antibody adjustment, more vaccine doses were associated with stepwise lower infection risk, but that association was fully attenuated after adjusting for antibody titers. This is consistent with titers carrying the measurable protective signal in this dataset, though it does not by itself establish that titers are the causal mechanism rather than a marker of it.

Repeated exposures alone can bend the protection curve

To test whether exposure frequency alone could generate the observed pattern, the authors ran discrete-time simulations under leaky, non-sterilizing immunity, where each exposure carries some residual infection probability. Cumulative infection probability over a window was expressed as 1 − (1 − p(t))^E, separating per-exposure susceptibility p(t) at antibody level t from exposure burden E. Simulations reproduced the rightward shift of the protection curve as transmission rose, without altering per-exposure susceptibility, and the pattern held across different population antibody distributions and when heterogeneity in exposure risk was included.

A separate 5,000-person leaky-vaccine simulation varied per-exposure infection probability, vaccine efficacy, and exposure frequency. Cumulative risk stayed relatively stable across transmission intensities when per-exposure risk was low and vaccine efficacy high, but the compounding effect of repeated exposures markedly increased infection risk as per-exposure risk rose or vaccine efficacy declined. These are model scenarios illustrating a mechanism, not forecasts of what happened in any particular setting.

One per-exposure curve could reconcile setting-specific thresholds

The authors then asked whether a single per-exposure risk function, independent of exposure frequency, could be recovered from the data. They fitted a mechanistic model to the test-negative dataset in which cumulative infection probability equals 1 − (1 − p(t))^E, with p(t) a scaled logistic function of antibody titer and E approximated as a scaling constant times biweekly PCR positivity. Parameters were estimated jointly by maximum likelihood, with uncertainty from 1,000 draws of the estimated covariance matrix.

The fitted model reproduced the observed rise in infection risk with transmission. The authors present this as proof of concept: if a per-exposure curve is established, cumulative infection probability could be calibrated for another setting by inserting that setting's exposure frequency, so a titer conferring 50% protection in a low-exposure setting could be translated rather than assumed to hold. The approach rests on explicit assumptions — exposure burden scaling approximately linearly with PCR positivity, independent contributions of repeated exposures, and comparable effective exposure burden among people sharing a biweekly proxy — and individual exposure counts were not observed.

The thresholds are context-bound, not universal cutoffs

The empirical estimates come from symptomatic, care-seeking individuals with PCR-confirmed SARS-CoV-2 infection in this surveillance context. The authors note that exposure-dependent correlates may differ in magnitude for asymptomatic infection, severe disease, a different exposure window, or pathogens with different transmission dynamics and immune mechanisms. The 15% positivity cutoff separating low and high transmission was chosen for model stability, and sensitivity analyses with alternate cutoffs gave consistent findings, but the binary split remains an analytic choice.

The mechanistic claim is broader than the empirical one: for any pathogen with leaky, non-sterilizing immunity, repeated exposures within a defined window will raise cumulative infection risk even if per-exposure susceptibility is unchanged. That generality belongs to the simulation framework, not to the measured antibody thresholds.

Compare correlates only alongside their transmission context

The practical priority this paper establishes is a comparison rule rather than a single number: an antibody correlate of protection, a vaccine effectiveness estimate, or an immunobridging result should be read together with the transmission intensity under which it was measured. In this dataset the same antibody scale mapped to different observed protection depending on whether biweekly PCR positivity sat below or above 15%, with the fitted midpoint moving from about 3.5 to about 5.0 log10 BAU/mL.

The most consequential distinction is between protection against a single exposure and protection inferred from infections accumulated over an observation window. The paper's simulations show that the second can shift substantially while the first is unchanged, which is why the authors propose a per-exposure curve as the harmonizing quantity. For readers comparing correlates across studies, the first question is therefore not only which titer was reported, but how much exposure that titer was tested against.

Sources used in this answer