Phylogenetic analysis of the human antibody repertoire reveals quantitative signatures of immune senescence and aging

Abstract

The elderly have reduced humoral immunity, as manifested by increased susceptibility to infections and impaired vaccine responses. To investigate the effects of aging on B-cell receptor (BCR) repertoire evolution during an immunological challenge, we used a phylogenetic distance metric to analyze Ig heavy-chain transcript sequences in both young and elderly individuals before and after influenza vaccination. We determined that BCR repertoires become increasingly specialized over a span of decades, but less plastic. In 50% of the elderly individuals, a large space in the repertoire was occupied by a small number of recall lineages that did not decline during vaccine response and contained hypermutated IgD+ B cells. Relative to their younger counterparts, older subjects demonstrated a contracted naive repertoire and diminished intralineage diversification, signifying a reduced substrate for mounting novel responses and decreased fine-tuning of BCR specificities by somatic hypermutation. Furthermore, a larger proportion of the repertoire exhibited premature stop codons in some elderly subjects, indicating that aging may negatively affect the ability of B cells to discriminate between functional and nonfunctional receptors. Finally, we observed a decreased incidence of radical mutations compared with conservative mutations in elderly subjects' vaccine responses, which suggests that accumulating original antigenic sin may be limiting the accessible space for paratope evolution. Our findings shed light on the complex interplay of environmental and gerontological factors affecting immune senescence, and provide direct molecular characterization of the effects of senescence on the immune repertoire.

Trial registration: ClinicalTrials.gov NCT01827462 NCT02987374.

Keywords: CMV; UniFrac; aging; antibody repertoire; influenza vaccine.

Conflict of interest statement

Dr. Quake and Dr. Knight were coauthors on Biteen JS, et al. (2015) Tools for the microbiome: Nano and beyond. ACS Nano 10(1):6–37. This was a perspective and did not involve any active research collaboration.

Figures

Fig. 1.

Comparison of repertoire distances between individuals. (A) Study design. PBMCs, peripheral blood mononuclear cells. (B) UniFrac distances between study participants at baseline. Participant labels begin with a two-letter code indicating age group and CMV status. (C) Comparison of between-participant UniFrac values at baseline by age range and CMV status. (D) Within-participant longitudinal UniFrac values.

Fig. S1.

UniFrac analysis at high time resolution. These data are from six healthy volunteers (labeled HTR1–HTR6) in the age range of 17–30 y separate from the main text study, sampled at days −5, −3, 0, 1, 4, 7, 9, and 11 relative to 2012 seasonal trivalent inactivated influenza vaccine administration. BCR libraries were prepared and sequenced using a previously published protocol (11) very similar to the protocol used for the main text cohort. Participants HTR1, HTR2, HTR3, HTR4, HTR5, and HTR6 were aged 18, 28, 25, 24, 19, and 28 y, respectively, and were female, male, male, male, female, and male, respectively. (A) Heat map of UniFrac values between participants and time points, clustered using complete linkage. (B) Within-participant longitudinal UniFrac distances as a function of time interval. Day 7 marks peak vaccine response. Five of 6 participants showed an increase in UniFrac from the smallest time interval to the largest.

Fig. S2.

Nonnormalized unshared branch lengths between study participants at baseline. The displayed quantity is given by UAB=∑G∑i{bi×|I(NiA>0)−I(NiB>0)|×|NiA−NiB|}, where A and B refer to the two participants under consideration (subsampled to 104 sequences each), G refers to the different V-segment/J-segment/CDR3-length groups, i refers to the branches of the phylogenetic tree constructed for group G, bi refers to the length of branch i, NiA refers to the number of A sequences in G descending from branch i, and NiB refers to the number of B sequences in G descending from branch i (details on phylogenetic tree construction are provided in SI Materials and Methods).

Fig. S3.

Longitudinal aspects of the main text study cohort. (A) Estimate of new sequence diversity created from day 0 to day 7 [respectively (resp.) day 28]: Chao1 diversity estimate at day 7 (resp. day 28) minus Chao estimate of shared diversity (22) between day 0 and day 7 (resp. day 28) samples. (B) Day 0 to day 28 UniFrac distance for each participant versus the participant’s day 0 influenza antibody titer. The correlation between the two variables was not statistically significantly different from 0. Here, antibody titers were assessed by hemagglutination inhibition assay for all three strains present in the vaccine, and the geometric mean of the three values was reported. Note that titers were not available for all participants.

Fig. 2.

Dissection of oligoclonality. (A) Lineage composition of repertoires by abundance of RNA molecules. The 20 most abundant lineages across all visits are distinguished by color and vertical order; transcripts belonging to the less abundant lineages are represented in gray. Participants displaying oligoclonality are labeled “oligocl.”. (B) Isotype proportions (weighted by distinct sequences rather than molecular abundance) in the most abundant lineage at baseline. (C) Mutation loads in isotype-unswitched versus switched compartments of the most abundant lineage and of the rest of the repertoire. (D) Average mutation loads in IgD sequences. (E) Percentage of VDJ sequences in the IgD compartment that were not observed in the IgM compartment.

Fig. 3.

Repertoire structure as relevant to vaccine response. (A) Chao1 (21) estimates of repertoire diversity by isotype compartment and isotype proportions of observed sequences. (B) Relative and absolute sizes of naive and antigen-experienced compartments, in terms of numbers of distinct sequences estimated using Chao1. Here, isotype and mutation levels were used as proxies to separate compartments: “Naive” counts were based only on unmutated IgD sequences, and “antigen-experienced” counts were based only on IgA and IgG sequences. (C) Percentage of VDJ sequences in the IgM compartment that were also observed in the IgD compartment. (D) Analysis of within-lineage sequence entropy. The mean entropy per nucleotide was calculated for the sequences in each lineage and then averaged over all lineages with equal numbers of distinct sequences. Curves were smoothed using a moving-average filter of width 5 values. (E) Polyclonal vaccine response at day 7: Of lineages present at both day 0 and day 7, the percentage that increased in abundance from day 0 to day 7 (excluding superlineages) is shown. (F) Distribution of baseline-to-endpoint lineage radius increases. Here, only lineages present at both day 0 and day 28 were considered, and lineage radii from pooled day 0 and day 28 sequences were compared with radii from day 0 sequences.

Fig. S4.

Diversity estimates based on conditional uncovered probability (CUP) (33) computed with look-ahead r = 100 using QIIME (49). (A) CUP-derived estimates of repertoire diversity by isotype compartment. IgE was omitted because raw sequence counts were insufficient to compute the estimate in participants EP1 and EP3. (B) Relative and absolute sizes of naive and antigen-experienced compartments, in terms of numbers of distinct sequences estimated using CUP. Here, isotype and mutation levels were used as proxies to separate compartments: Naive counts were based only on unmutated IgD sequences, and antigen-experienced counts were based only on IgA and IgG sequences. Note that the estimates for “naive diversity” and “% naive sequences” were not feasible for participant EN5.

Fig. S5.

Visualization of lineage radius increase. A lineage from participant YN1 at day 0 and day 28 is shown as a phylogenetic tree based on the edit distance between CDR3 sequences. Each leaf corresponds to one sequence, colored according to the time point at which it was observed. The axis measures distance (branch length) as the number of nucleotide substitutions. In this example, the “lineage radius increase from day 0 to day 28” (largest distance between any two gray or black sequences on the tree MINUS largest distance between any two black sequences) is 2. For each time point, the branches joining two sequences with the maximum edit distance between them are highlighted in bold.

Fig. 4.

Nonsense and missense amino acid mutation analysis. (A) Percentage of IGH sequences that contained at least one premature stop codon, while being in-frame. (B) Average number of V-region somatic mutations among in-frame sequences with a premature stop codon. The P value is from a two-sided Wilcoxon–Mann–Whitney test. (C) Percentage of observed premature stop codons that occurred in the CDR3. (D) Percentage of amino acid mutations in IgA and IgG sequences at day 7 that were considered radical. Similarity between the germline residue and the mutated residue was assessed by IMGT based on three aspects: hydropathy, volume, and chemical characteristics; a mutation was called “radical” if the residue changed class in at least two aspects.

Fig. S6.

Illustration of UniFrac calculations for a pair of young participants and a pair of elderly participants at baseline. Six sequences with V-segment IGHV3-11, J-segment IGHJ4, and a CDR3 length of 45 nucleotides were sampled from each participant. Unshared branches are highlighted in bold. [Tree plots were generated using the R package “ggtree” (50)]. (A) Participants YN2 and YN3. The unshared branch length is 18.66, and the total branch length 20.32. After applying the weighting scheme described in SI Materials and Methods, the weighted unshared branch length is 3.17 and the weighted total branch length is 3.75, giving a UniFrac value of 0.84. (B) Participants EP4 and EN5. The unshared branch length is 21.95, the total branch length is 22.43, the weighted unshared branch length is 4.37, the weighted total branch length is 4.88, and the UniFrac value is 0.90. Notice how the elderly pair’s UniFrac value turns out larger than the young pair’s due to an increased prevalence of highly mutated sequences in unshared clades.

Fig. S7.

Effect of sampling depth on UniFrac’s dynamic range and discrimination potential between young individuals. (A) UniFrac distance matrix computed from 104 sequences per sample. (B) UniFrac distance matrix computed from 103 sequences per sample. Dendrograms were produced using complete-linkage clustering. Notice that cluster samples from the same participant more often fail to cluster together in B than in A. Note also the generalized increase in UniFrac values from A to B.