Assessing evidence of immune imprinting in serological patterns of influenza immune response

Problem Statement
The burden of influenza on public health is exacerbated by the limited efficacy of vaccines, often due to pre-existing immunity from prior infections or vaccinations. A specific phenomenon, termed "immune imprinting," suggests that exposure to specific influenza strains during childhood shapes the immune system's memory, influencing future responses to strains based on their structural similarity to early exposures. While epidemiological data on disease and mortality suggest imprinting drives age-specific patterns of illness, isolating this mechanism from confounding cohort effects is difficult. Specifically, because specific subtypes dominate circulation during specific historical windows, a participant's birth year simultaneously determines their likely "imprinting subtype" and their age at sampling/isolation. This collinearity makes it challenging to distinguish whether observed serological patterns result from a biological imprinting mechanism or simply from age-related variations in immune responses driven by other factors. Furthermore, previous serological investigations have been limited, often relying on cohorts born prior to 1953 (exclusively exposed to H1N1) or defining imprinting as inferred protection rather than differences in immune response.

Methodology
The authors utilized serological data from the Fluscape cohort, a long-term community-based study in southern China involving 2,008 individuals aged 2–95, sampled between 2009 and 2015. Hemagglutination inhibition (HI) titers were measured against 21 A(H3N2) strains (circulating 1968–2014) and three A(H1N1) strains.

To assess imprinting, the authors employed a suite of Generalized Additive Models (GAMs) fitted to log-transformed HI titers. The models varied in how they accounted for age, a critical confounder:

1. No age adjustment.
2. Adjustment for age at sampling only.
3. Adjustment for age at isolation (age when the strain was circulating) only.
4. Adjustment for both age at sampling and age at isolation.

The core variable of interest was a categorical "imprinting variable," encoding the likely subtype of first exposure based on birth year. Two approaches were used to define this variable:

• Non-probabilistic: Deterministically assigns a subtype based on birth year (assuming infection occurs within the first year of life).
• Probabilistic: Draws subtype assignments from empirically estimated probabilities (Gostic et al., 2016) to account for the possibility of first exposure occurring later in childhood. This approach was repeated 1,000 times to propagate uncertainty.

The study also utilized a mechanistic simulation model (Yang et al., 2022) as a null model. This simulation modeled the evolution of immune responses to H3N2 from 1968 to 2014 for 777 participants but explicitly excluded any imprinting mechanism or between-subtype interactions. A "positive control" simulation was also run where a hypothesized imprinting effect was introduced by boosting responses by 50% for participants born on or after 1968 relative to those born earlier. Model performance was evaluated using the Bayesian Information Criterion (BIC).

Key Results

• Empirical Analysis: The best-fitting models (by BIC) for H3N2 titers included age at sampling, age at isolation, and the probabilistic imprinting variable. However, the evidence for imprinting was weak and sensitive to model specification. Only the coefficient for the H2N2-imprinted cohort was consistently negative (indicating lower H3N2 titers), which aligns with the imprinting hypothesis. Coefficients for the H1N1 cohort were not consistently significant or negative. When fitting models to H1N1 titers, no coefficients were consistent with the imprinting hypothesis.
• Simulation Analysis (Null Model): Crucially, when the same statistical models were fitted to the output of the null simulation (which contained no imprinting mechanism), the resulting coefficient patterns closely resembled those found in the empirical data. Specifically, models including only age at isolation suggested systematically lower titers for the H2N2 cohort, and the inclusion of the imprinting variable was strongly favored by BIC even in the absence of a true imprinting mechanism.
• Simulation Analysis (Positive Control): When the positive control simulation (with an explicit imprinting mechanism) was analyzed, the imprinting variable successfully identified the pattern, producing coefficients clearly consistent with the hypothesis. However, the signal was diminished when both age variables were included, particularly for the H1N1 cohort.

Key Contributions
The paper demonstrates that statistical patterns in serological data that appear consistent with immune imprinting (specifically, systematically lower HI titers against H3N2 strains in cohorts likely first exposed to H1N1/H2N2) can emerge even in the absence of an explicit imprinting mechanism. These patterns can be reproduced by the confluence of age-related immune dynamics and temporal phenomena (cohort effects). The study highlights the difficulty of distinguishing true biological imprinting from age-related variations in immune response when using HI titers alone, as the statistical signal can be absorbed by age splines or generated by the complex interplay of birth year and strain circulation history.

Significance and Claims
The authors state that their work "lays the foundations for further research into immune imprinting while accounting for cohort effects." They do not claim to have definitively proven or disproven the existence of immune imprinting. Instead, they argue that the serological patterns observed are "somewhat equivocal." The findings suggest that apparent subtype-subtype interactions may result from the unique experiences of specific cohorts (e.g., timing of infections, time since antigenic cluster introduction) rather than solely from a specific immunological imprinting mechanism. The authors conclude that understanding these complex dynamics is crucial for refining vaccination approaches, but caution that imprinting-like patterns can emerge without the hypothesized immunological mechanisms, complicating the interpretation of serological data in the context of age and temporal dynamics.