Exploratory positional differences in observed CDRH2 lineage-state transitions following HIV-1 germline-targeting vaccination

Abstract

Background: Germline-targeting HIV-1 vaccination recruits VRC01-class B-cell precursors, but the positional patterns of observed longitudinal B-cell receptor lineage-state changes remain incompletely described. We compared CDRH2 profiles between homologous and heterologous boosting regimens using published IAVI G002 data. Methods: Participant-specific Week-8 lineages were prospectively matched to Week-24 observations. Germline-to-nongermline changes in sampled lineage-modal residues were evaluated at seven IGHV1-2*02 CDRH2 positions using participant-balanced, mean-centered profiles and exact participant-label permutation inference. Sampling-depth and sorting-probe sensitivity analyses assessed observability. A separate participant-held-out benchmark evaluated antibody-language-model representations for predicting observed Week-24 lineage detection. Results: The CDRH2 analysis included 332 lineages, 15 participants, 2,179 eligible lineage-position observations, and 241 observed state changes. The regional profile statistic was 0.027066 (exact p = 0.001709; six-region adjusted q = 0.010256), with descriptive differences at V-relative positions 51, 54, and 55. Most eligible longitudinal observations had one assessable cell per visit. In seven Group-1 participants, Week-24 dual-probe sampling recovered 181 baseline lineages with eOD-GT8 and 143 with Cg28v2, including 60 detected by both probes. In a separate cohort of 2,739 lineages from 17 participants, IgBERT and AbLang-2 representations did not improve participant-balanced average precision over a four-feature biological baseline. Conclusions: Observed CDRH2 lineage-modal-state profiles differed between regimens in this exploratory secondary analysis. Sparse sampling, potentially probe-related ascertainment, and absent linked functional validation preclude inference about mutation timing, functional improvement, or causal booster effects.