Abstract
During chronic infection, HIV-1 engages in a rapid coevolutionary arms race with the host’s adaptive immune system. While it is clear that HIV exerts strong selection on the adaptive immune system, the modes of immune response are still unknown. Traditional population genetics methods fail to distinguish a chronic immune response from natural repertoire evolution in healthy individuals. Here, we infer the evolutionary modes of B-cell repertoire response and identify a complex dynamics where, instead of one winning clone, there is a constant production of new better mutants that compete with each other. A substantial fraction of mutations in pathogen-engaging CDRs of B-cell receptors are beneficial, in contrast to the many deleterious changes in structurally relevant framework regions. The picture is of a dynamic repertoire, where better clones may be outcompeted by new mutants before they fix, challenging current vaccine design and therapy ideas.
The HIV-1 virus evolves and proliferates quickly within the human body [1–3], often recombining its genetic material among different viral genomes and rapidly mutating. These factors make it very hard for the host immune system to control an infection, leading to long-term chronic infection. While it is clear that the virus exerts strong selective pressure on the host immune system, the adaptive immune response during chronic infections remains unknown.
The immune system has a diverse set of B and T-cells with specialized surface receptors that recognize foreign antigens, such as virus epitopes, and protect the organism. We focus on the chronic phase of HIV infection, where the immune response is dominated by antibodymediated mechanisms, following the strong response of the cytotoxic T-lymphocytes (i.e., CD8+ killers T-cells), around 50 days after infection [4]. During the chronic phase, the symptoms are minor and the viral load is relatively stable but its genetic composition undergoes rapid turnover. After an infection, B-cells undergo a rapid somatic hypermutation in lymph node germinal centers, with a rate that is approximately 4 – 5 orders of magnitude larger than an average germline mutation rate in humans [5]. Mutated B-cells compete for survival and proliferation signals from helper T-cells, based on the B-cell receptor’s binding to antigens. This process of affinity maturation is Darwinian evolution within the host and can increase binding affinities of B-cell receptors up to 10-100 fold [6]. It generates memory and plasma B-cells with distinct receptors, forming lineages that trace the evolutionary selection pressures inflicted by the virus [7] (see schematic in Fig. 1A). A B-cell repertoire consists of many such lineages forming a forest of co-existing genealogies.
Immune repertoire high-throughput sequencing has been instrumental in quantifying the diversity of B-cell repertoires [8, 9]. Statistical methods have been developed to characterize the processes involved in the generation of diversity in repertoires and to infer the underlying heterogenous hypermutation preferences in B-cell receptors (BCRs) [9–11]. Deviation of the observed mutations in BCRs from the expected hypermutation patterns are used to infer selection effects of mutations from repertoire snapshots in order to identify functional changes that contribute to the response against pathogens [10, 12].
Recently, longitudinal data, with repertoires sampled over multiple time points from the same individuals, has brought insight into the dynamics of affinity maturation in response to antigens [13–16]. The dynamics of affinity maturation and selection in response to HIV have also been characterized for chosen monoclonal broadly neutralising antibody lineages [3, 17]. Yet, the effect of a chronic infection on the dynamics of the whole BCR repertoire remains unknown.
Here, we compare the structure and dynamics of BCR repertoires sampled over 2.5 years in HIV patients (data from ref. [15] collected through the SPARTAC study [18]) with the repertoire structure in healthy individuals (data from ref. [19]). We reconstruct genealogical trees for B-cell receptor lineages inferred from BCR repertoires in each individual (SI). B-cell lineages of HIV patients, a few examples of which are shown in Fig. 1B, can persist over months to years of infection, which is much longer than the lifetime of a germinal center (weeks), indicating the recruitment of memory cells for further affinity maturation in response to the evolving virus.
Reconstructed lineage trees show a skewed and asymmetric structure, consistent with rapid evolution under positive selection (see Fig. S1A) [20]. To quantify these asymmetries, we estimated two indices of tree imbalance and terminal branch length anomaly. In both HIV patients and healthy individuals, we observe a significant branching imbalance at the root of the BCR lineage trees, indicated by the U-shaped distribution of the sub-lineage weight ratios (see SI), in contrast to the flat prediction of neutral evolution, calculated from Kingman’s coalescent (Fig. 2A). Moreover, we observe elongated terminal branches in BCR trees compared to their internal branches, with the strongest effect seen in trees from HIV patients, again in violation of neutrality (Fig. 2B, Fig. S1). These asymmetric features of BCR trees are clear signs of intra-lineage positive selection. However, they only reflect the history of lineage replication and give limited insight into the mechanisms and dynamics of selection. For instance, tree asymmetry is also observed in unproductive BCR lineages, which lack any immunological function but are carried along with the productive version of the recombined gene expressed on the other chromosome (Fig. 2A,B).
To characterize the selection effect of mutations in more detail, we evaluate the spectrum of mutation frequencies in a lineage, known as the site frequency spectrum (SFS). We evaluate the SFS separately for synonymous and nonsynonymous mutations in different regions of BCRs (Fig. 2C, Fig. S2). We see a signifiant upturn of SFS polarized on non-synonymous mutations in pathogen-engaging CDR3 regions, consistent with rapid adaptive evolution [20], and in contrast to monotonically decaying SFS in neutrality (SI). This signal of positive selection is strongest in HIV patients with an order of magnitude increase in the high end of the spectrum, suggesting that the BCR population rapidly adapts in HIV patients.
To understand the dynamics and fate of these adaptive mutations, we use the longitudinal nature of the data to analyse the temporal structure of the lineages. We estimate the likelihood that a new mutation appearing in a certain region of the BCR reaches frequency x at some later time within the lineage (Fig. 3A), and evaluate a measure of selection g(x) as the ratio of this likelihood between non-synonymous and synonymous mutations [21] (SI). At frequency x = 1 (i.e., substitution), this ratio is equivalent to the McDonald-Kreitman test for selection [22]. Generalizing it to x < 1 makes it a more flexible measure applicable to the majority of mutations that only reach intermediate frequencies. A major reason why many beneficial mutations never fix in a lineage is clonal interference, whereby BCR mutants within and across lineages compete with each other [7]. To quantify the prevalence of clonal interference, we also evaluate the nonsynonymous-to-synonymous ratio h(x) of the likelihood for a mutation to reach frequency x and later to go extinct (SI). In short, g(x) identifies “surges” and h(x) “bumps” in frequency trajectories of clones. These likelihood ratios have intuitive interpretations: g(x) > 1 indicates evolution under positive selection, with a fraction of at least αbenef. = 1 − 1/g strongly beneficial amino acid changes in a given region [23]. On the other hand, the likelihood ratio g(x) smaller than 1 is indicative of negative selection, with a fraction of at least αdel. = 1 − g strongly deleterious changes (see SI for a derivation of these bounds). Likewise, κbenef. = 1 − 1/h or κdel. = 1 − h define a lower bound on the fraction of either beneficial or deleterious mutations that go extinct.
Fig. 3B shows the selection likelihood ratio g(x) in an HIV patient (patient 4) for lineages belonging to a typical V-gene class IGHV2-70D (SI); see Fig. S3 for statistics in all individuals. In this gene family, we detect positive selection (g > 1) in the CDR3 region, with around a two fold larger fraction of non-synonymous compared to synonymous changes that reach frequencies x > 0.6, indicating at least αbenef. = 40% of CDR3 mutations to be strongly beneficial. On the other hand, the likelihood ratio in FWR signals strong negative selection (g <1), where non-synonymous changes reaching frequencies x > 0.6 are two times fewer than the synonymous changes, indicating at least αdel. = 35% of these mutations to be strongly deleterious. Similarly, the interference likelihood ratio h(x) for a V-gene class IGHV510-1 in an HIV patient with interrupted treatment (patient 5) indicates that about κbenef. = 47% of CDR3 mutations in this gene family that go extinct due to clonal competition are strongly beneficial (Fig. 3B). In short, we observe a large fraction of adaptive mutations, and also a substantial amount of clonal interference which prevents some of the mutations from dominating within lineages.
To see how these observations generalize at the repertoire level, we quantify the region-specific fraction of beneficial and deleterious mutations within BCR lineages of distinct VJ-gene classes and also the fraction of selected mutations that are impeded by clonal interference (Fig. 3C and Table I). We infer a larger fraction of VJ-gene classes with positively selected amino acid changes in their CDR regions and negatively selected amino acid changes in FWRs . Moreover, the positively selected beneficial mutations inCDR3 and the pooled CDR1/CDR2 regions are strongly impacted by clonal inference, in contrast to mutations in FWR (Fig. 3C, Table I, Fig. S3). These observations confirm the pervasiveness of clonal interference in the regions of the BCR with the most important functional role.
In patients with interrupted ART, we infer a twice larger fraction of beneficial mutations to rise with strong clonal interference in pathogen-engaging CDR3 regions following the interruption of treatment, compared to the ART-naive patients with a stable chronic infection – such a shift is not present for mutations in CDR1, CDR2 and FWR (Fig. 3 and Table I). This pattern is consistent with the rate of HIV-1 evolution in patients with different states of therapy. Genome-wide analysis of HIV-1 has revealed that evolution of the virus within ART-naive patients slows down during chronic infections with limited clonal interference in viral populations [24]. The antibody response traces the evolution of the virus [1, 7] and forms a quasi-equilibrium balance. On the other hand, rapid expansion and evolution of HIV following the interruption of ART drives a strong immune response and affinity maturation in HIV-responsive B-cell lineages. Evolution of HIV-1 population during viral expansion introduces a time-dependent target for the adaptive immune system and opens room for many beneficial changes in the HIV-engaging CDR3 regions, as indicated in Fig. 3.
Somatic evolution during affinity maturation is complex: there is no one winner of the race for the best antibody. We show that rapid and strong affinity maturation upon sudden pathogenic challenges, and a quasistationary response during chronic infections are a feature of the B-cell response to infections. Somatic evolution of BCRs is similar to rapid evolution in asexual populations where many beneficial mutations rise to intermediate frequencies leading to complex clonal competition and genetic hitchhiking. Such evolutionary dynamics is prominent in microbial populations [25], in viruses including HIV within a patient [24, 26] and global influenza [21,27,28]. In the immune system, clonal competition in BCR repertoires is also observed on short time scales (~ weeks) in response to the influenza vaccine [16].
Clonal interference among beneficial mutations not only makes selection slower and less efficient, but it also makes the outcome of the evolutionary process less predictable [25]. This is of significant consequence for designing targeted immune-based therapies. Currently, the central challenge in HIV vaccine research is to devise a means to stimulate a lineage producing highly potent broadly neutralizing antibodies (BnAbs). A combination of successive immunization and ART has been suggested as an approach to elicit a stable and effective BnAb response; see e.g. ref. [29]. An optimal treatment strategy should account for clonal interference among BCRs during a rapid immune response to antigen stimulation, which could hamper the emergence of a desired BnAb within the repertoire.