TLMs are significantly less diverse and mutated than other MBC subsets in PLWH

We first sought to understand if TLMs may be blocked in their ability to undergo antigen-driven evolution of their BCRs. We selected a variety of PLWH (with or without concurrent ART), including those in an elite control cohort, and phenotyped their circulating B cells by flow cytometry (Figs 1A and S1). There was inter-donor variation irrespective of whether the donors were in the ART or the elite control group and regardless of treatment status. The majority of MBCs were consistently of a resting phenotype and there was no obvious enrichment for a particular MBC subset within either group as compared to HIV negative controls. As no statistical significance was found between donor groups for each MBC subset in turn (Fig 1A) we randomly selected five donors (two from the ART control group and three from the elite control group) and bulk-sorted TLM (CD21^-CD27^-), resting (CD21⁺CD27⁺) and activated (CD21^-CD27⁺) MBCs based on their isotype (IgG versus IgM). These cells then underwent bulk BCR sequencing of the IgG and IgM heavy chains (methods) to assess affinity maturation, diversity and evolution since these are key indicators of antigen-driven evolution in B cells. The aim of this was to compare BCR repertoire characteristics in different MBC populations during HIV infection. We observed that resting IgG MBCs were significantly (Mann-Whitney U test) more diverse than the other subsets, and that the diversity of IgG TLM and activated MBCs was similar (Fig 1B). A similar trend was observed for IgM MBCs (Fig 1B); however, this was not found to be statistically significant (p > 0.05) and may reflect their wider responsiveness to a range of antigens and distinct temporal germinal centre outputs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. TLMs are significantly less diverse and mutated than other MBCs in PLWH.

A: Flow-cytometric quantification of frequencies of IgM and IgG MBC subsets within IgD- B cells, n = 10 PLWH on ART (triangles), n = 12 for PLWH with elite control (circles) and n = 10 for HIV-negative controls (squares). Gating strategy shown in S1 Fig. MBC subsets are colour-coded as follows: activated (orange), resting (teal) and TLM (purple). Donors that underwent bulk BCR sequencing are highlighted with black symbols. No statistical significance was found in the Benjamini-Hochberg multiple testing adjusted p-values between donor groups for each MBC subset in turn using Mann-Whitney U tests with the compare_means function from the ggpubr package. B: Clonal diversity indices of MBC subsets stratified by IgG versus IgM expression for PLWH (n = 5). 100 BCR sequences were randomly sampled from each donor/subset for comparison. C: VH mutational burden (as % of nucleotides) of BCR sequences analysed in (B). D: Calculation of mutations between B cell subsets for PLWH (n = 5). Lineages containing multiple MBC subsets were selected, and BCR sequences underwent pairwise comparison to identify unique AA mutations (FWR1 - FWR3). The median was taken per comparison for each lineage and then averaged to give a value per donor. The x axis shows the comparison group, and the y axis the average proportion of unique AA mutations per comparison. E-F: BCR phylogenetic network analysis to ascertain MBC subset evolution within PLWH (n = 5). Networks were created for each lineage, with each node representing a BCR (self-linkages excluded). Edges between nodes were drawn if the tip-to-tip distance was below a set threshold, defined as the position of first apex in the rate of change curve for the tip-to-tip distances. The network was directed by only allowing edges from closer nodes to germline to further nodes from germline. Once these networks were created, the hub score (E) and out-degree to in-degree ratio (F) was calculated (hub score representation, panel (E), left hand side), to provide global and local measures of node connectivity, respectively. Throughout, statistical significance was assessed using Mann-Whitney U tests with the compare_means function from the ggpubr package (p > 0.05 not marked, p ≤ 0.05 *, p < 0.01, **, p < 0.001 ***, p < 0.0001 ****).

https://doi.org/10.1371/journal.ppat.1013817.g001

Next, we sought to ascertain the mutational burden of these BCR sequences across MBC subsets. Firstly, the V_H mutation percentage was calculated (Fig 1C). In accordance with previous reports in HIV [25], TLMs were significantly less mutated than the other MBC subsets (p = 4.86x10^-20, p = 4.45x10^-20, p = 8.16x10^-9, p = 2.76x10^-4 for IgG activated vs TLM, IgG resting vs TLM, IgM activated vs TLM, and IgM resting vs TLM respectively) (pairwise Mann-Whitney U test). This may suggest that TLMs are less able to undergo antigen-driven activation and BCR mutation. To further characterize the mutation landscape of these cells, the proportion of unique (de novo) amino acid (AA) mutations between and within each subset was compared. The median unique mutation frequency was calculated per MBC subset for each lineage, and the mean average was taken per donor (Fig 1D). This data illustrates that mutation sharing is significantly greater between TLMs than between other subsets and suggests that they are less able to undergo continued BCR diversification than other MBC subsets since they contain fewer unique mutations overall.

Next, directed phylogenetic networks were created to better understand the evolutionary trajectories of the different MBC subsets. These networks allow evaluation of evolutionary importance of each node (cell) using the hub score (Fig 1E) and the out-degree to in-degree ratio (degree ratio; Fig 1F). The hub score estimates the value of each node’s connections to other nodes within the network, i.e., the score is higher if a cell connects to another cell that is highly connected. We observed significantly lower hub scores for TLMs than other subsets, regardless of isotype (IgG; Activated p = 5.5x10^-7; Resting p = 7.1x10^-5, IgM; Activated p = 2.3x10^-7, Resting p = 4.4x10^-6,.Fig 1E) (Mann-Whitney U test). This illustrates that MBCs are less able to undergo continued BCR evolution upon differentiation into TLMs than other subsets. Similarly, the degree ratios (ratio of outgoing connections divided by incoming connections) were significantly lower for TLMs across both isotypes than the other subsets (Fig 1F, IgG; Activated p = 0.0011; Resting p = 0.0308, IgM; Activated p = 0.0147, Resting p = 0.0005) (Mann-Whitney U test). Taken together, TLMs studied here contained significantly fewer mutations, were less diverse, and were less well connected in each B cell network than the other two MBC subsets that underwent heavy chain BCR sequencing from these 5 PLWH. Such data suggests that TLMs could be subject to additional activation/maturation controls as reported previously [14], a precursor population to other subsets, or produced via extrafollicular responses that undergo weaker affinity maturation [26].

Minor transient spike in HIV viremia triggers a potent antibody response with expanded neutralization breadth

Based on our BCR analysis of MBCs in PLWH, we hypothesised that the TLM subset may have impaired capacity for BCR mutation and diversification. To address how and when TLMs arise, we focussed further study on one donor who controlled virus in the absence of ART but then experienced a loss of control and a transient viral load, where longitudinal samples (pre-, during and post-viral blip) were available. Access to such timepoints allowed us to gain a unique insight into B cell dynamics during a relatively acute and low viral blip as may occur during the early stages following ART cessation. This ART naïve individual (referred to as “index participant”) had a low/undetectable viral load (<50 copies/mL) at the time of recruitment to the elite controller cohort (month 0 – Pre Blip) in October 2013. However, when sampled at month 41, a viral load of 603 copies/mL was detected (Blip). This viral blip was associated with a CD4⁺ T cell count decrease from 950 cells/μL to 650 cells/μL (Fig 2A). Following the viral blip, CD4⁺ T cell counts remained largely stable through to month 65, although they never fully recovered to the pre-blip levels. Similarly, the viral load returned to low levels of ~100 copies/mL at month 52 (Post Blip).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Minor transient spike in HIV viremia triggers antibody response.

A: Dynamics of viral load (V/L, copies/mL) and CD4 + T cell count (per mm3) from an elite controller (index participant). Grey bars indicate timepoints of stored blood (PBMC/plasma) samples. B: Maximum Likelihood Phylogenetic analysis of env nucleotide sequences (amplicons spanning AA35-683, HXB2 numbering) isolated from the index participant. Sequences are coloured according to the sampling timepoint. Red nodes represent bootstrap support > 70%. Clade B sequences (NL4-3 and Ba-L) were included as outgroups for phylogeny rooting. Root branches have been shortened to prevent large branches caused by differences between participant sequences and outgroups. C-D: Maximum Likelihood Phylogenetic analysis of env nucleotide sequences (amplicons spanning AA35-683, HXB2 numbering) isolated from an individual on ART at one timepoint (C) or the index participant at the viral blip timepoint (D). Red nodes represent bootstrap support > 70%. E: Anti-Env IgG titre in the plasma from the index and a HIV-negative participant. Plasma titre was assessed via semi-quantitative ELISA against the 1A3, 1D5 and 2E4 Env proteins (encoded by env sequences isolated from the index participant). F: Pseudovirus neutralization 50% inhibitory dilution (ID₅₀) titres of plasma from the index and HIV-negative participant, respectively. Pseudovirus was expressed using autologous env sequences cloned from the index participant including 6 pseudoviruses from the pre blip timepoint (month 0), 8 pseudoviruses from the blip timepoint (month 41) and 8 pseudoviruses from the post blip timepoint (month 65). The dotted line represents the detection limit of the assay (ID₅₀ = 1:100). If no neutralization was detected, samples were assigned an ID₅₀ of 0. If only a single data point of the plasma dilution series was above 1:100, no accurate ID₅₀ estimation could be determined and thus samples were assigned an ID₅₀ of 1:100. The box plot lines show the median and 10-90th percentile.

https://doi.org/10.1371/journal.ppat.1013817.g002

Analysis of the integrated HIV envelope (env) gene sequences by single genome amplification (SGA) from genomic DNA revealed a high degree of sequence variation within the index participant across pre-blip, blip and post-blip timepoints (Fig 2B). While certain branches of the phylogenetic tree were enriched with sequences from specific time points, the sequences from the post-blip time point were dispersed throughout the tree, suggesting limited selective pressure across the assessed timepoints despite ongoing replication. Interestingly, and in contrast to diverse env sequences from the viral blip (index participant) (Fig 2C), sequences recovered from an individual PLWH who was on ART (Fig 2D) were largely clonal, likely driven instead by the proliferation of infected T cell clones rather than viral outgrowth and diversification [27]. Moreover, the sequence diversity of the index participant viral quasi-species during the blip may have been underestimated as only integrated provirus was evaluated. This was because it was not possible to obtain viral sequences directly from replicating virus in the blood due to insufficient RNA recovery at this low viral load (<1,000 copies/mL).

To profile the B cell response to the viral blip we first determined the anti-Env plasma IgG titre. Env is the only viral antigen exposed on the surface of the HIV virion and is thus the major target for the B cell mediated antibody response, including neutralizing antibodies against HIV that are the goal of many HIV vaccine candidates. Three env sequences representative of the sequence variation and glycosylation features present in the index participant were selected, cloned and expressed as stabilized soluble recombinant proteins (1A3, 1D5 and 2E4) by utilizing SOSIP mutations [28]. The stabilized Env proteins were used to quantify the autologous anti-Env IgG titre and compared with the binding response against two non-autologous Env SOSIP proteins: BG505 N332 (Clade A) and B41 (Clade B) [28,29]. Strikingly, the anti-Env IgG response of the index participant rose to >1 mg/mL at the viral blip (Fig 2E). As expected, the titre against autologous Env proteins (Fig 2E) was higher than that against heterologous proteins and no response was observed in HIV-negative control plasma (S2 Fig). Interestingly, the IgG titre against the heterologous proteins peaked at the viral blip, suggesting that the small viral load increase was capable of boosting both the autologous and heterologous IgG titre.

Subsequently, plasma neutralization was assessed using a pseudovirus assay as previously described [30,31]. The 50% inhibitory dilution (ID₅₀) values highlight that plasma neutralization was broadest and most potent against autologous pseudoviruses at the viral blip (Fig 2F), in line with the ELISA readings (Fig 2E). This is in contrast to weaker and less broad responses at the pre-blip and post-blip timepoints (Fig 2F). Interestingly, the viral blip timepoint plasma was unique in neutralizing a heterologous pseudovirus; B41, albeit at a low level (S2 Fig). Overall, the neutralization profile of the index participant appeared to be the weakest in terms of breadth and potency against autologous pseudoviruses prior to the viral blip (Fig 2F), followed by a ~ 5–10-fold increase in potency and increased breadth associated with the viral blip, before the response appears to partially return to baseline (Fig 2F).

Taken together, these data show that in an elite controller, even a minor spike in viral load (603 copies/mL) can induce a potent but short-lived neutralizing plasma antibody response against HIV. This makes the index participant a suitable case for studying B cell response to HIV viremia, including TLM development with particular relevance to the acute setting of ART cessation.

Transient viral blip induces expansion of Env-reactive IgG⁺ memory B cells

After characterizing the humoral response to this minor viral blip, we next examined potential changes in the MBC compartment. We used fluorochrome-tagged autologous 1A3 and 1D5 recombinant Env proteins to detect antigen-specific MBCs at the blip and post-blip timepoints. Unfortunately, limited PBMC availability precluded such analysis in the pre-blip sample. Env⁺ IgG ⁺ MBCs formed almost 1% of MBCs at the blip timepoint (Fig 3A and 3B). However, this population contracted by 75% at the post-blip timepoint as the index participant regained viral control (Fig 3B), mirroring our ELISA and neutralization data (Fig 2E and 2F). Surprisingly, the flow-cytometric phenotyping of Env-negative and Env-positive IgG⁺ B cells during the blip (and all IgG⁺ MBC post-blip) demonstrated a marked similarity to the MBC phenotypes seen in 15 HIV-negative donors with the resting subset representing >75% and TLM < 10% of MBCs, respectively (Fig 3C). This contrasts with the MBC phenotypes of 11 donors with chronic viremia (Fig 3C) who had significantly expanded TLM and activated subsets and a reduced proportion of resting MBCs compared to HIV-negative controls (Fig 3C). Although there was no enrichment of TLM cells in the Env⁺ subset during the viral blip, as has been previously reported for PLWH with high chronic viremia [25], we did observe a > 2-fold enrichment in the frequency of the activated subset in the Env⁺ MBCs compared to the frequency of the activated subset in Env-negative MBCs [33].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Transient viral blip induces expansion of Env-reactive IgG+ memory B cells.

A: Gating strategy for FACS-sorting of 1A3 Env-PE and 1D5 Env-BV786 specific memory B cells (Aqua live/Dead-, CD4-, CD19 + , IgM- IgG+) from index participant at the blip (month 41) and post-blip (month 65) timepoints, or a HIV negative participant. B: Percentage of Env-reactive B cells within IgG + MBCs stratified by participant. C: Proportion of MBC subsets (resting, activated, TLM, intermediate) within total IgG + MBCs in PLWH with chronic viremia (n = 11) and HIV-negative (n = 15) controls are shown on the left panel. The analysis was based on flow-cytometric phenotyping of the expression of CD21 and CD27. Statistical significance was assessed using Mann-Whitney U tests for each subset in turn between the chronic viremia and HIV-negative donors (p > 0.05 not marked, p ≤ 0.05 *, p < 0.01, **, p < 0.001 ***, p < 0.0001 ****). The right panel shows data on these MBC subsets for the elite controller during the viral blip (divided into Env+ and Env- cells) compared to the post-blip timepoint. D: UMAP visualization of single cell transcriptomes from TLM, activated and resting MBCs (n = 301) sorted from the elite controller post blip and Env+ memory B cells from the blip timepoint. Single cell libraries were prepared via the Smart-seq 2 pipeline [32], a total of 88 resting, 88 activated, 88 TLM and 161 Env + MBC are profiled. Cells are annotated by their sorting identity. E: Expression of selected genes associated with memory B cell subsets profiled in (D). The fraction of cells is shown by the dot size, and the mean expression level is reflected by the colour. F: Similarity of single-cell transcriptomes of Env + IgG+ memory B cells with the resting, activated and TLM IgG + B cell subsets sequenced in (D). Similarity is calculated as probability using the Celltypist annotation tool trained on the sorted post-blip MBC subsets. G-H: IgG subclass usage (G) and nucleotide somatic hypermutation levels (V + J gene), (H)of sequenced memory B cell populations shown in (D). One way-ANOVA with Tukey’s multiple comparison used for statistical testing. *p < 0.032, **p < 0.021, ***p < 0.0002 and ****p < 0.0001.

https://doi.org/10.1371/journal.ppat.1013817.g003

Next, we used FACS to isolate IgG+ cells from all three memory subsets (resting, activated and TLM) as well as Env⁺ IgG⁺ MBCs for single-cell RNA/BCR sequencing to assess their transcriptomic similarity. UMAP projection of the different MBC transcriptomes presented TLMs as a distinct population, whereas there was a significant overlap between the resting and activated MBC subsets (Fig 3D). Interestingly, the Env⁺ cells largely clustered with the resting and activated subsets but not TLMs (Fig 3D), mirroring their flow-cytometry phenotype (Fig 3C). TLM population identity/annotation was confirmed by their expression of bona fide TLM marker genes such as FCRL5 and ITGAX and low expression of CD27 when compared to the resting and activated subsets (Fig 3E). We subsequently used the automated annotation tool Celltypist to quantify the transcriptional similarity of the Env⁺ cells to MBC subsets. This confirmed that the vast majority of Env⁺ cells were transcriptionally most similar to resting MBCs, with only a small proportion of cells resembling TLMs (Fig 3F). Finally, we compared the differences in the use of IgG subclasses and levels of SHM among MBCs. In line with our earlier data (Fig 1C), TLM cells have significantly lower levels of SHM than resting and activated MBCs (Fig 3G). TLM cells also displayed decreased usage of downstream IgG subclasses such as IgG2 and IgG4, suggesting a less mature or more innate-like phenotype (Fig 3H). Again, Env⁺ MBCs had a BCR profile (SHM, IgG subclasses) more similar to resting/activated MBCs (Fig 3G and 3H).

Overall, these data confirm that the viral blip was accompanied by an expansion of activated Env⁺ IgG MBCs. However, this discrete viral blip was not large enough to induce the global MBC dysfunction typically associated with chronic HIV viremia nor dysfunction within Env⁺ MBCs. Instead, the antigen-specific MBCs observed (presumably precursors of anti-Env antibody-secreting cells) resembled resting/activated MBCs both phenotypically and transcriptionally. Although TLM cells were present, their frequency was significantly lower than in chronically infected individuals and more in keeping with HIV-negative donors.

Transient viral blip induces a complex response across both B and non-B cell immune subsets

To examine the transcriptomic changes in the whole B cell and non-B cell immune compartment of the index participant across all three time points in an unbiased approach, we performed single-cell RNA/BCR sequencing of MACS-selected circulating B cells spiked with all PBMCs at 3:1 ratio (favouring B cells). This generated a total of 48,602 single-cell transcriptomes (post-QC) with 13 distinct cellular clusters identified (Fig 4A and 4B). CellTypist automated cell annotation (Figs 4C, S3 and S4) [34] was validated and further refined using a panel of eight CITE-seq markers (S3 and S4 Figs).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Transient viral blip induces a complex response across both B- and non-B cell subsets.

A: UMAP visualization of annotated single-cell transcriptomes (48,602 cells) recovered from circulating MACS-selected B cells spiked with PBMC in 3:1 ratio of the index participant (elite controller) across all timepoints. B: Bar plot showing proportion (% of non-B cells) of non-B cell subsets across all three timepoints. C-D: Dot plot showing expression of selected marker genes (C) and CITE-seq protein markers (D) across annotated cell clusters from (A). The fraction of cells expressing a particular marker is shown by dot size and the marker mean expression level is reflected by the colour. E: UMAP visualization of finely annotated B cell transcriptomes (42,149 cells) extracted from the dataset shown in (A). F: Bar plot showing proportion (% of B cells) of annotated B cell subsets across all three timepoints. G-H: Dot plot showing expression of selected marker genes (G) and CITE-seq protein markers (H) across annotated B cell clusters from (E). The fraction of cells expressing a particular marker is shown by dot size and the marker mean expression level is reflected by the color. I: Box plots showing number of inferred cell-cell interactions (left) and interaction strength (right) across timepoints using CellChat package. Both B- and non-B cells included. J: Isotype usage analysis depicted as % of B cells within each subset and timepoint. K: Violin plot showing VDJ mutation frequency stratified by subset and timepoint. L: GSEA of selected Hallmark and KEGG gene sets based on pre-ranked DEG comparing all B cells from blip (red dots) and post-blip timepoint (blue dots) to the reference (pre-blip B cells). Dot size reflects normalized enrichment score (NES). Vertical black lines indicate the threshold for statistical significance. M: Heatmap showing scanpy expression score of selected HALLMARK, KEGG and GOBP gene sets in each B cell subset and timepoints. Scaled by row.

https://doi.org/10.1371/journal.ppat.1013817.g004

The non-B cell compartment was analysed using differential abundance analysis (DA) with the milopy package across timepoints. Milopy uses the QL F-test with a specified contrast to compute a P value for each neighbourhood and controls for multiple testing as described previously [35]. The DA analysis revealed an expansion of naïve helper T cells during the viral blip (12% in blip to 24% post-blip; Figs 4D and S4). This was unexpected as increased viral load typically correlates with decreased helper T cells due to their direct destruction by HIV [36]. Indeed, the Tem/Effector helper T cells decreased during the viral blip by a greater magnitude (6% in blip to 2% post-blip). Interestingly, in the post-blip phase, we observed an expansion of effector memory (TEM)/ terminal effector memory (TEMRA) cytotoxic T cells (33.3% in blip to 46.8% post-blip, Figs 4B and S4) that likely played a dominant role in regaining viral control.

In the B cell compartment, we annotated eight distinct cell clusters (Fig 4E) based on their marker genes (Fig 4F) and CITE-seq panel expression (Fig 4G): Naive1, Naive2, PLD4 + plasmablasts, TLM1, TLM2, IgM memory, switched memory, and plasma cells. Notably, we identified two distinct TLM populations, both expressing canonical TLM markers although with a different intensity. TLM2 cells had higher expression of ITGAX, TBX21 (T-bet), LILRB1, and CD86 (Fig 4G). A similar expression pattern was also observed at the surface protein level by CITE-seq (Fig 4H). This suggests that TLM1 may be a precursor to the TLM2 population, which resembles previously described TLM populations more closely [9,20]. DA analysis with the milopy package was again used to examine the cell cluster changes in the B cell compartment across timepoints (S4 Fig). This analysis revealed a progressive increase in naïve B cells (20.4%, 32.6%, 51.1%) and that PLD4 + plasmablasts expand during the blip and are maintained post-blip (0.6%, 4.3%, 6.6%) (Fig 4H). The expansion of naïve B cells during and after the viral blip may result from their enhanced generation and/or recruitment to combat the transient viremia. In line with our previous flow cytometry and Smart-Seq2 data (Fig 3), the proportion of TLM cells remained stable during the viral blip.

The active coordination of antiviral immune response during the blip was further evidenced by the increased total number and strength of predicted cell-cell interactions accounting for all annotated immune cells (Figs 4I and S4). Of note, PLD4 + plasmablasts had one of the lowest numbers of predicted interactions at the blip timepoint, which may, together with predominant IGHM expression (Fig 4J) and lower levels of SMH (Fig 4K), indicate their extrafollicular differentiation pathway, allowing for a rapid anti-viral IgM response. Similarly, TLM1 cells predominantly expressed IGHM and had lower SHM burden compared to TLM2 that were mostly switched to IgG subclasses, indicating that TLM1 cells are less mature than TLM2 cells (Fig 4J and 4K). However, TLM2 were in turn less mature than switched memory as evidenced by more restricted SHM. During blip and post-blip, TLMs (unlike the switched memory subset) increased their proportion of unswitched cells with a small drop in median SHM, consistent with an influx of new (less mature) B cells in response to viremia (Fig 4J and 4K).

Analysis of Hallmark gene signatures for the whole B cell compartment across timepoints revealed a significantly enriched TNF-α signature during the blip, which declined post-blip when an enhanced interferon-α and -γ signature emerged instead (Fig 4L). This temporal signature pattern was seen across all B cell subsets, although with variable intensity (Fig 4M). The Naïve1 subset displayed the strongest TNF-α and inflammatory response during blip, while PLD4 + plasmablasts were the top expressor of both interferon signatures (Fig 4M). Unexpectedly, this global blip-induced interferon response was not accompanied by an expected TLM expansion [23], for which persistent viremia might be required. Surprisingly, TLM1 and TLM2 were the most active B cell subset in terms of BCR signalling, activation and differentiation with only slight decrease of these signatures during blip and post-blip phases (Fig 4M). This persistent TLM activation state might indicate evolving early dysfunction, ultimately leading to reduced responsiveness, as documented in TLMs from chronically viremic PLWH [9,14,37].

Taking together, our comprehensive unbiased single-cell data analysis elucidated dynamic changes in B cell and non-B cell compartments in response to a transient viral blip. This viral blip induced TNF-α and subsequent IFN response in all B cell subsets, with naïve B cells and newly emerged (likely extra-follicular) PLD4 + plasmablasts being the most robust responders. Surprisingly, TLM2 cells did not expand in response to the transient viremia but along with TLM1 cells displayed a persistently activated phenotype. Thus, this analysis emphasises the nuanced interplay between viral load and immune modulation in an elite HIV controller.

TLM subsets are heterogenous and display innate-like B cell characteristics

To investigate the clonal relationship and potential differentiation pathway of annotated B cell subsets (in particular, TLMs) we utilised the single-cell BCR-seq data generated in the previous experiment. A principal component analysis (PCA) based on VJ gene usage revealed that TLM1 cells across all three timepoints clustered together and separate from other B cell subsets (Fig 5A). This was driven mainly by a high use of IGHV4–34 segment (Fig 5B), which was also observed, albeit to a lesser extent, in the TLM2 subset (Figs 5B and S5). IGHV4–34 is often found in self-reactive and lymphoma B cells [38–41]. Surprisingly, TLM1 subset also appeared one of the most clonal groups apart from plasma cells, the analysis of which was likely less reliable given their small cell count (Fig 5C). Overall, the high frequency of IGHV4–34 segment use and clonal appearance of the TLM1 subset despite a relatively low SHM burden (Fig 4K) indicates that this subset is likely enriched for innate-like B cells [42]. The clonal overlap analysis showed the strongest link between TLM1 and IgM MBCs, reaching its maximum during the blip (Fig 5D). Interestingly, TLM2 cells overlapped equally with TLM1, IgM memory and switched memory subsets, suggesting a likely heterogeneous origin of these cells (Fig 5D). Indeed, we observed many clones composed of cells from different B cell subsets, although individual clones seemed mostly homogenous in their isotype use (Fig 5E). TLM2 cells within some mixed (cross-subset) clones had the highest VDJ mutation count, suggesting them as one possible differentiation endpoint for TLM1(Figs 5E and S5). To explore the differentiation pathways of the B cell subsets further, we used the Slingshot package to infer pseudotime and developmental trajectories and identified four distinct developmental lineages (Fig 5F). PLD4 + plasmablasts differentiated directly from naïve B cells, which is consistent with their proposed extrafollicular origin. Interestingly, the TLM1 subset appeared to be the branching point for three distinct lineages (IgM memory, switched memory and TLM2). Although TLM1 formed a cluster of transcriptomically similar B cells, they are still heterogenous and poised for different differentiation pathways. These cells likely include innate-like B cells that tend to progress into TLM2, and naïve B cells differentiating into memory cells (either IgM or switched). Finally, we replicated these GEX-based pseudotime analysis findings also using BCR-seq-based pseudobulk VDJ trajectory analysis (Fig 5G). In conclusion, our investigation into the origins and differentiation potential of TLM cells through BCR and 5’GEX data analysis revealed high IGHV4–34 usage, and high clonality with lower SHM burden compared to MBCs. However, both TLM subsets also contained cells from IgM memory and switched memory clones, indicating the potential for TLM1 cells to differentiate into conventional MBCs whilst TLM2 appeared to be the final differentiation endpoint.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. TLM subsets are heterogenous and display innate-like B cell characteristics.

A: PCA based on VJ gene usage by elite controller B cells coloured by B cell subset and timepoint. B: Bar chart showing percentage of B cells using IGHV4-34 segment within each subset and timepoint. C: Bar chart showing clonality (clone size Gini index) of each B cell subset stratified by timepoint. D: Circos-style plots depicting clonal overlap between each B cell subset pair stratified within each timepoint. The thickness of the grey connecting line between circles reflects the number of overlapping BCR clonotypes. E: Single-cell BCR network plots of all pre-blip clones consisting of at least three cells, coloured by B cell subset annotation (left), heavy chain isotype class use (middle) and mutation count (right). Each circle/node represents a single B cell with a corresponding set of BCR(s). Cells belonging to the same clone are connected with a black line. F: UMAP visualization of all B cells described in Fig 4, coloured by GEX-based pseudotime and overlayed by inferred developmental lineages (black arrows), generated by slingshot package. G: UMAP generated by the BCR-seq-based pseudobulk VDJ trajectory analysis (dandelion package) of B cell dataset described in Fig 4, coloured by calculated pseudotime and B cell subset annotation.

https://doi.org/10.1371/journal.ppat.1013817.g005

Ethics statement

The index participant was classified as exhibiting elite control based upon their largely stable low viral load (<100 copies/mL) over 13 years whilst not on ART (patient preference) prior to the viral blip. Samples from PLWH who had detectable viremia or non-ART mediated viral control were collected as part of a protocol approved by local research ethic committee (London – City & East REC 12/LO/1572). Samples from PLWH on ART (suppressed) and HIV-negative participant samples were previously collected and processed with ethical approval by South Central – Hampshire B (REC 19/SC/0423) [2]. Samples from chronic untreated HIV and HIV-negative volunteers were collected with the University of KwaZulu-Natal Biomedical Research Ethics Committee (BREC) approval (BREC/00001275/2020) and the Collection of Sputum, Urine and Blood samples (CUBS) cohort with BREC approval (BE022/13) respectively. Studies complied with all relevant ethical regulations for work with human participants and conformed to the Helsinki declaration principles and Good Clinical Practice (GCP) guidelines. All subjects enrolled into the studies provided written informed consent.

PBMC isolation

Heparinised blood was isolated, diluted with an equal volume of PBS (Sigma Aldrich, Dorset, UK) and layered over 20 mL Histopaque-1077 (Sigma Aldrich). Peripheral blood mononuclear cells (PBMCs) and plasma were then isolated by centrifugation (400x g, 20 minutes, no brake). A 1x 15 mL aliquot and 2x 1 mL aliquots of plasma were recovered and stored at −80°C for future use. The mononuclear layer was then recovered, washed in 40 mL PBS and collected by centrifugation (400x g, 5 minutes). Cells were then washed in complete RPMI (RPMI-1640 media (Sigma Aldrich) supplemented with L-glutamine (Sigma Aldrich), penicillin/streptomycin (Sigma Aldrich) and 10% fetal bovine serum (FBS) (Sigma Aldrich)), pelleted by centrifugation (400x g, 5 minutes), resuspended in complete RPMI and counted using trypan blue exclusion. PBMC were cryopreserved in cryovials for future use by resuspension at 1x10⁷ cells/ml in FBS 10% DMSO (Sigma Aldrich) and stored at -80°C.

Single Genome Amplification (SGA) of env genes

The Qiagen DNAEasy kit (Qiagen, Manchester, UK) was used to extract DNA from 1x107 PBMCs using the spin column protocol with some minor modifications. Cells were lysed for 30 minutes at 56°C rather than the recommended 10 minutes. A longer lysis step was found to lead to increased DNA yield. Additionally, a modification was made to the elution step, whereby 90μl of 5mM Tris-HCL (Sigma Aldrich) was dispensed onto the column, the column was then warmed to 40°C and left for 20 minutes prior to centrifugation at 13,000 RPM for 1 minute. HIV env single genomes were PCR amplified as previously described [17] using isolated genomic DNA as template and primers shown in Table 1. In brief, gDNA was diluted in PBS and nested PCR carried out using High Fidelity Platinum Taq DNA polymerase (Invitrogen, Paisley, UK) with primers Env5out and OMF19, followed by nested primers Env5in and Env3in [59,60]. PCR amplifications were carried out in 1 × High Fidelity Platinum PCR buffer, 2mM MgSO4, 0.2mM dNTPs, 0.2μM of each primer and 0.025 units/μL Platinum Taq, with a final reaction volume of 10 μL. The thermocycler protocol was as follows: 94°C for 2 minutes, followed by 35 cycles of 94°C for 15 seconds, 55°C for 30 seconds and 68°C for 4 minutes, finishing with a 10-minute elongation step at 68°C. 2kb env amplicons were visualised by agarose gel electrophoresis on precast 96 well 1% agarose E-gels (Invitrogen). Wells where a single env amplicon was obtained then underwent one final round of PCR amplification using the Env Cloning Fwd and Env Cloning Rev primers which generate an env amplicon compatible with Gibson cloning into the psvIII expression vector. Conditions for the final PCR were unchanged although reactions were scaled up to 25 μL.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Smart Seq 2 and Env sequencing primer sequences.

https://doi.org/10.1371/journal.ppat.1013817.t001

Envelope cloning

env PCR amplicons were purified using the QIAquick PCR purification kit (Qiagen) following the manufacturer’s recommendations. PsvIII vector containing the HXB2 env sequence was digested with the BamHI-HF and KpnI-HF restriction enzymes (NEB, Hitchin, UK) at 37°C for 3 hours in 1x CutSmart buffer (NEB). The digested plasmid was then visualised by agarose gel electrophoresis and the digested plasmid purified using the QIAquick Gel Extraction Kit (Qiagen) following the manufacturer’s recommendations. Purified inserts were then cloned into the digested psvIII expression vector via Gibson assembly. Briefly, 100 ng of digested psvIII vector was incubated with a 3-fold molar excess of PCR insert in 1x NEBuilder HiFi DNA Assembly Master Mix (NEB) for 2 hours at 50°C. NEB 10-beta Competent E. coli (High Efficiency) were subsequently transomed with the assembly mix and incubated on LB-Agar ampicillin selection plates overnight at 37°C. The following day single colonies were upscaled to 1 mL LB-Broth ampicillin culture for 6 hours at 30°C, 200 RPM. 2 μL of culture was then added to 48 μL H₂O and placed at -80°C for 1 hour before incubation at 95°C for 15 minutes. The resulting supernatant was then used as template for a PCR reaction with the Env Cloning Fwd and Env Cloning Rev primers as described above. env amplicons were visualised by agarose gel electrophoresis on precast 96 well 1% agarose E-gels (Invitrogen). Wells which contained env amplicons were upscaled to 100 mL LB-Broth ampicillin culture 30°C, 200 RPM for 16 hours and then plasmid DNA isolated using the plasmid plus midi sample kit (Qiagen).

Envelope Sequencing and Phylogenetic analysis

Cloned env amplicons underwent Sanger-sequencing using the A-K sequencing primers. Sequence reads were trimmed at the 5′ and 3′ ends until base calls consistently reached a quality score ≥30, resulting in 99.9% accuracy of base calls. Contigs were then generated using SnapGene Software (www.snapgene.com) by alignment to the HXB2 env sequence. env contigs were trimmed resulting in amplicons from amino acid 35 to amino acid 683 (HXB2 numbering). env sequences were then aligned at the protein sequence level using MUSCLE v.3.8.31 [61] and mapped back to nucleotide, with minor manual adjustment. 2 clade B sequences (NL4–3 and B41) were included for reference and 2 clade A sequences (BG505 and 92UG029.EC1) from the Los Alamos HIV Sequence Database (http://www.hiv.lanl.gov/) were included to serve as an outgroup. A maximum likelihood phylogeny was estimated using IQ-tree software [62], with the general time-reversible nucleotide substitution model and gamma-distributed rate heterogeneity. Clade support was estimated from 1,000 nonparametric bootstrap replicate datasets. The maximum likelihood phylogeny was rooted on the outgroup branch and visualized using FigTree v.1.4.4 (http://tree.bio.ed.ac.uk/software/figtree/).

Soluble Envelope protein production

Three diverse env sequences from the maximum likelihood phylogeny: 1A3, 1D5 and 2E4 were selected to be expressed as soluble stabilized proteins. To allow for stabilized expression, SOSIP mutations were introduced into the sequences as described by Sanders et al [28] and sequences ordered as synthetic Gene Fragments with Gibson cloning compatible overhangs (Genewiz, Bishop’s Stortford, UK). p818 plasmid was digested with the BstBI and PstI restriction enzymes and gel extracted, env Gene Fragments were then cloned into the digested vector as described in the “envelope cloning” methods section. Stabilized Env proteins were expressed in HEK293F cells and purified with either PGT145 or 2G12 affinity chromatograph followed by size exclusion chromatography (SEC) using a HiLoad 16/600 Superdex pg200 (GE Healthcare) as described previously [28,63,64]. The Avi-tagged proteins were biotinylated using the BirA enzyme (Avidity, Colorado, USA) according to the manufacturer’s instructions.

Pseudovirus production

HEK-293T cells were co-transfected with Env-expressing psvIII plasmid and a HIV Env deficient backbone plasmid (pSG3ΔEnv) at a mass ratio of 1:2 using polyethyleneimine (PEI) max (Polysciences, Bergstrasse, Germany) at a DNA: PEI mass ration of 1:3 and incubated for 48 hours at 37°C 5% CO₂. The pseudovirus containing supernatant was then harvested and passed through a 0.45 μm syringe filter (Sigma Aldrich), aliquoted and stored at -80°C. Subsequently, frozen pseudovirus was titrated using the HeLa-TZM-bl reporter cell assay with the Bright-Glo Luciferase assay kit (Promega, Hampshire, UK) and the 50% tissue culture infectious dose (TCID_50/mL) calculated.

In vitro neutralization assays

Pseudovirus neutralization assays were carried out using the 96-well plate TZM-bl cell-based assay described by Sarzotti-Kelsoe et al [30], with Bright-Glo luciferase reagent (Promega) and final readout analysed using a BioTek Synergy H1 plate reader (Agilent technologies, Manchester, UK). Briefly, plasma samples were heat-inactivated by incubation at 56°C for 1 h and then titrations of plasma starting at 1:100 incubated with luciferase encoding pseudovirus (at 200 TCID₅₀/mL) in duplicate for 1 hour at 37°C 5% CO₂. 10,000 HeLa TZM-bl cells were then added to each well and incubated for 48 hours at 37°C 5% CO₂ before supernatants were removed and luciferase readout used as an indicator of infectivity. 100% infectivity was defined by wells which contained virus and TZM-bl cells only, with no antibody present.

Semi-quantitative total Ig ELISA

Half-area 96-well MaxiSorp plates (VWR, Leicestershire, UK) were coated with either 25 μL goat anti-human F(ab)’2 (1:1,000) (Jackson immunoResearch, Cambridgeshire, UK) in PBS for total IgG plates, or 25 μL mouse anti-human Ig κ light chain and mouse anti-human Ig λ light chain (1:200 each) (Becton Dickinson Biosciences, Wokingham, UK) in PBS for total IgG3 plates, and incubated overnight at 4°C. Plates were washed with PBS-T and blocked with 100 μL assay buffer (PBS, 3% BSA, 0.05% Tween-20) for 1 hour at room temperature. 25 μL of serially diluted plasma (1:100–1:10⁷) in duplicate or known concentrations of IgG or IgG3 standard in triplicate (Sigma Aldrich) were then applied to the plates. Plates were washed with PBS-T and 25 μL of detection antibody diluted in assay buffer added to each well: goat anti-human IgG-AP (1:1,000) (Sigma Aldrich) or goat anti-human IgG3-AP (1:1,000) (Cambridge Bioscience, Cambridge, UK) and incubated for 1 hour at room temperature. Subsequently, plates were washed 6 times with PBS-T and 25 µL of colorimetric alkaline phosphatase substrate added (Sigma Aldrich). Absorbance was measured at 405 nm after 60 minutes. Total IgG or IgG3 concentrations in plasma were then calculated based on interpolation from the IgG or IgG3 standard curves using a four-parameter logistic (4PL) regression curve fitting model.

Semi-quantitative Env ELISA

Columns 1–9 of half-area 96-well MaxiSorp plates were coated overnight at 4°C with streptavidin in PBS (2 µg/ml per well in 25 µL) (Sigma Aldrich). Columns 10–12 were coated with either 25 μL goat anti-human F(ab)’2 (1:1,000) in PBS for IgG standard curves, or 25 μL mouse anti-human Ig κ light chain and mouse anti-human Ig λ light chain (1:200 each) in PBS for IgG3 standard curves. Plates were then washed with PBS-T and blocked with 100 μL assay buffer (PBS, 3% BSA, 0.05% Tween 20) for 1 hour at room temperature. 25 μL of biotinylated Env protein (2 µg/ml) in assay buffer was then added to streptavidin coated wells and 25 µL of assay buffer added to wells in columns 10–12 for 2 hours at room temperature. Following washing 25 μL of serially diluted plasma (1:100 to 1:10⁷) in duplicate or known concentrations of IgG or IgG3 standard in triplicate were then applied to the plates. Plates were washed with PBS-T and 25 μL of detection antibody diluted in assay buffer added to each well: goat anti-human IgG-AP (1:1,000) or goat anti-human IgG3-AP (1:1,000) and incubated for 1 hour at room temperature. Subsequently, plates were washed 6 times with PBS-T and 25 µL of colorimetric alkaline phosphatase substrate added. Absorbance was measured at 405 nm after 60 minutes. Env reactive IgG or IgG3 concentrations in plasma were then calculated based on interpolation from the IgG or IgG3 standard curves using a four-parameter logistic (4PL) regression curve fitting model.

Flow cytometry and single cell sorting

Fluorescence-activated cell sorting (FACS) of Env reactive cryopreserved PBMCs was performed on a BD FACS Melody (BD Biosciences). Alternatively, memory B cell subsets were sorted using a BD FACS Aria II (BD Biosciences). To generate fluorescently tagged Env probes 2 μg of biotinylated Env protein was incubated with either 0.5 μg streptavidin-conjugated PE (BD Biosciences) or 0.5 μg streptavidin-conjugated BV786 (BD Biosciences) for 30 minutes. Cryopreserved PBMC were thawed on ice and incubated with 100 μL of Zombie Live/Dead dye (diluted 1:400 in PBS) (Biolegend, London, UK) for every 5x10⁶ cells at room temperature for 20 minutes. The dye was then quenched by addition of complete RPMI, cells collected by centrifugation (400x g, 10 minutes) and then washed once in PBS. Cells were then incubated with specific combinations of the phenotyping antibodies + /- biotinylated tetramers listed in Table 2 in a final volume of 100 μL in PBS (concentrations used were for every 5x10⁶ cells). Compensation controls were prepared according to manufacturer’s instructions using Anti-Mouse Ig, κ and negative control compensation particles (BD Biosciences). For Smart-Seq 2 experiments single B cells were gated according to the following successive gates: lymphocyte, single cells, CD19⁺ Aqua live/Dead^- CD4^- and IgM^- IgG⁺. From this population 1x 96 well plate of each of the following subsets was sorted: CD21^- CD27^- (TLM cells), CD21^- CD27⁺ (activated memory cells) and CD21⁺ CD27⁺ (resting memory cells). Alternatively, antigen specific cells were sorted based on positivity for either the PE-tagged or BV786-tagged Env probes. All cells were sorted using “Yield” purity mode into chilled 96-well plates (VWR) containing 4.4 μL of lysis buffer composed of oligo-dT (5′-AAGCAGT GGTATCAACGCAGAGTACT30VN-3′) (1 μM), 0.4% Triton X-100 (Sigma Aldrich), dNTPs (1mM) (Thermo Fisher, Paisley, UK), and RNAse inhibitor (4 units) (Takara Bio, London, UK). After each plate was filled, it underwent centrifugation at 400x g for 1 minute and placed on dry ice immediately before storage at −80°C. For bulk BCR sequencing experiments single B cells were gated according to the following successive gates: lymphocyte, single cells, CD19⁺ Aqua live/Dead^- CD14^- CD3^-, CD20⁺ CD38^lo, IgD^- and either IgM^- IgG⁺ or IgM⁺ IgG^-. From this population the following subsets were sorted: CD21^- CD27^- (TLM cells), CD21^- CD27⁺ (activated memory cells) and CD21⁺ CD27⁺ (resting memory cells). Bulk cell populations were sorted using “Yield” purity mode into chilled 1.5 mL tubes containing lysis buffer (0.3% NP40 and ~6.67mM RNaseOUT). After sorting of each donor was complete each 1.5 mL tube underwent centrifugation at 400x g for 1 minute and was placed on dry ice immediately before storage at −80°C. Data was analyzed on FlowJo v10 (BD).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. B cell phenotyping flow cytometry and CITE-Seq panels. Where suppliers do not provide antibody concentrations the volume of antibody used (μL) is provided.

https://doi.org/10.1371/journal.ppat.1013817.t002

cDNA generation, library preparation, and sequencing (Smart-seq2)

The Smart-seq2 protocol [32] was utilised with some minor adjustments for working with lymphocytes. In brief, 96-well plates containing single cell lysates were thawed on ice. Each well then received 5.4 μL of reverse transcription mix containing Superscript II reverse transcriptase (50 units) (Thermo Fisher), RNAse inhibitor (10 units) (Tarka Bio), 5X first strand buffer (Thermo Fisher), DTT (5mM) (Thermo Fisher), betaine (1M) (Sigma), MgCl2 (12mM) (Sigma Aldrich), Nuclease-free water (Thermo Fisher) and template switching-oligo (5′- AAGCAGTGGTATCAACGCAGAGTACATrGrG + G-3′) (1μM). Reverse transcription (RT) was carried out for 90 minutes at 42°C followed by 10 cycles of: 50°C for 2 minutes and 42°C for 2 minutes to allow for completion/continuation of RT and template-switching, then finally 70°C for 15 minutes to terminate the reaction. Each well then received 15 μL of PCR mix containing Nuclease-free water, and 2x HiFi HotStart ReadyMix (Roche, Welwyn Garden City, UK). Cells were then subjected to the following thermocycler protocol: 98°C for 3 minutes, 24 cycles of: 98°C for 20 seconds, 67°C for 20 seconds, and 72°C for 6 minutes, then finally 72°C for 5 minutes. Following amplification, 0.8 × AMPure XP beads (Beckman Coulter, High Wycombe, UK) were used to purify cDNA and remove primer dimers. The size distribution of cDNA in each well as well as the concentration of cDNA in the 500–5000 bp range was subsequently determined using a High Sensitivity D5000 ScreenTape kit (Agilent Technologies) and readout analysed on an Agilent 4200 TapeStation System. Cells that exhibited cDNA profiles indicative of degraded RNA were excluded from subsequent library preparation. The cDNA from remaining wells were diluted to approximately 0.5 ng/μL before carrying out the Nextera XT DNA library preparation (Illumina, Cambridge, UK) following manufacturer’s instructions. 0.6 × AMPure XP beads (Beckman Coulter) were used to purify cDNA and remove primer dimers. Libraries were then quality checked and quantified once again using the High Sensitivity D5000 ScreenTape system and then libraries were pooled at equimolar concentrations for sequencing. In total, 443 cells were sequenced across two runs using 2 × 75 bp NextSeq 500 High Output v2.5 kits on an Illumina NextSeq 500.

CITE-seq staining for single-cell proteogenomics

Frozen PBMC samples were thawed at 37°C in a water bath. Ice cold compete RPMI was added slowly to the cells before centrifuging at 300x g for 5 min. This was followed by a wash in 5 ml complete RPMI. The PBMC pellet was collected, resuspend in ice cold PBS and the cell number and viability were determined using Trypan Blue. A portion of the PBMC from each timepoint was then placed in 1.5 mL tubes and placed on ice, whilst the remaining PBMC underwent magnetic cell sorting (Miltenyi Biotec, Bergisch Gladbach, Germany) to purify the B cell populations. After isolation B cells were mixed with non-purified PBMC at a 3:1 ratio. The cells were then resuspend in 100 μL of mastermix containing CITE-Seq antibodies (Table 3) for 30 min at 4°C. cells were then washed three times by centrifugation at 500x g for 5 min at 4°C. PBMCs were counted again and processed immediately for 10x 5′ single cell capture (Chromium Next GEM Single Cell V(D)J Reagent Kit v2 with Feature Barcoding technology for cell Surface Protein-Rev F protocol) (10x genomics, Pleasanton, CA, USA). Three lanes of 20,000 cells from each timepoint were loaded onto a 10x chip.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Bulk BCR sequencing primers.

https://doi.org/10.1371/journal.ppat.1013817.t003

Library generation and sequencing

The Chromium Next GEM Single Cell 5′ V(D)J Reagent Kit (V2 chemistry) with Feature Barcoding technology for cell surface proteins was used for single-cell RNA-seq library construction for all samples. GEX and V(D)J libraries were prepared according to the manufacturer’s protocol (10x Genomics) using individual Chromium i7 Sample Indices. The cell surface protein libraries were created according to the manufacturer’s protocol. 5’GEX, V(D)J and cell surface protein indexed libraries were pooled and sequenced on a NovaSeq 6000 S4 Flow cell (paired-end, 150 bp reads) aiming for a minimum of 50,000 paired-end reads per cell for 5’GEX libraries and 10,000 paired-end reads per cell for V(D)J and cell surface protein libraries.

Bulk sorted BCR library construction

Bulk sorted MBC subsets (in lysis buffer) were collected from -80°C and incubated at room temperature to thaw the samples. The lysate was collected at the bottom of the tube by centrifugation. To purify the RNA, 1.7X volumes of room temperature RNAClean XP RNA purification beads were used (Beckman Coulter), binding RNA to beads for 20 minutes at room temperature. Mixtures were magnetically separated and then washed with fresh 80% ethanol. They were then air-dried before and eluted using molecular biology grade water (Ambion) for 10 minutes at room temperature. Clarified RNA containing supernatant was transferred to a clean tube.

To generate bulk BCR libraries from each MBC type, 15µL per RNA sample was combined with dNTPs (0.5mM, Thermo Fisher) and C-region specific primer (100nM) (Table 3) and molecular biology grade water, to a final volume of 17.75µL. Reverse transcription was performed using the SuperScript IV First-Strand Synthesis System, following manufacturer’s recommendations in a final volume of 25µL, using C-region specific primers (100nM) that contained UMIs and the Illumina read 2 site (Table 3). An IgG-specific C-region primer was used for the bulk sorted IgG⁺ MBC populations and an IgM-specific primer was used for the bulk sorted IgM⁺ MBC populations. This enabled us to perform heavy chain only bulk BCR sequencing. The reaction proceeded at 50°C for 45 minutes, prior to enzyme inactivation at 80°C for 10 minutes, followed by degradation of RNA: cDNA hybrids using RNaseH (New England Biolabs). cDNA product was purified using 1.0X AMPURE XP DNA purification beads (Beckman Coulter).

Eluted cDNA was supplemented with a mixture of leader sequence specific forward primers (H1F mix; 500nM), and a tag-specific reverse primer (PT056; 500nM), and 2X Q5 DNA polymerase Mastermix (New England Biolabs), to a final volume of 50µL, before undergoing PCR. This, and all subsequent PCRs, began with an initial denaturation at 98°C (50 seconds), and temperature then cycled between 98°C (10 seconds), 68°C (20 seconds) and 72°C (20 seconds) for a total 10 cycles, before a final extension at 72°C (120 seconds). The product was purified using 0.6X AMPURE XP DNA purification beads, and the eluted PCR product was supplemented with nested FWR1-specific primers containing the read 1 site (H2F mix; 500nM), a tag-specific reverse primer (PT056; 500nM) and Q5 DNA polymerase to a final volume of 50µL, which then underwent PCR. The product was purified using (0.6X) AMPURE XP DNA purification beads, and a final PCR was performed to add Illumina indexes (500nM each), again using Q5 DNA polymerase to the amplified DNA. PCR product was gel extracted using the QIAquick gel extraction kit (QIAGEN) to ensure homogenous size selection, prior to size and concentration quantification using the Agilent TapeStation 4150 (Agilent Technologies) and the Qubit 2.0 Fluorimeter (Thermo Fisher). Libraries were submitted for sequencing using the MiSeq platform, via UCL Genomics, using the following read lengths 325 bp, 8 bp, 8 bp and 275 bp for read 1, the two index reads, and read 2, respectively.

Bulk BCR raw sequence processing

Raw sequencing reads (fastq) were processed using the Immcantation framework via singularity (version 4.4.0) [65,66]. Briefly, reads were filtered (phred score > 20), prior to primer and UMI region identification used to define coarse read groups. These were then more finely clustered using cd-hit-est [67] with a threshold of 90%. Similarly to other published works, we assume that the resulting groups represent reads from unique RNA molecules [68]). A consensus sequence was then generated per read group, and duplicate reads were collapsed to provide a single representative sequence. UMIs comprised of a single read were removed.

V, D and J regions were assigned using IgBLAST [69]. Annotated sequences were then partitioned into coarse clusters based on identical V, J and CDRH3 lengths and then finely clustered based on 90% CDRH3 amino acid identity. Amino acid mutations were identified with shazam [66].

For phylogenetic tree analysis, lineages containing multiple MBC subsets with at least 3 unique sequences were identified and processed using IgPhyML [70,71]. CDRH3 sequence was ignored for this analysis as per the developer’s recommendations.

Bulk BCR bioinformatic analysis

Bulk BCR sequence analysis was carried out using R. Data manipulation and summarization was primarily performed using the tidyverse suite of packages [72], as well as foreach and doParallel for streamlining execution [73,74], and ape for basic analysis of trees [75]. The vegan package was used to calculate Shannon and Simpson diversity indices [76].

De novo mutation calculation

Each lineage that contained multiple MBC subsets was first identified within each donor. Within these lineages, mutations were identified in BCR sequences by comparing each sequence to the lineage germline sequence. Next, BCR sequences underwent pairwise analysis to identify both shared and unique mutations for each sequence, and the proportion of unique mutations identified (unique mutations/ total mutations). The comparisons were then grouped based on the subsets compared (e.g., TLM—Resting, TLM—TLM as multiple sequences from the same subset were also compared), and the median average taken, to provide a single value per comparison, per lineage. Once computed for all lineages, this was then summarised by taking the mean average per comparison, to give a single value per donor, which underwent statistical comparison.

BCR phylogenetic network analysis

To ascertain the evolutionary ‘potential’ of the MBC subsets, phylogenetic networks were created, using igraph [77]. To gain evolutionary information, lineages containing multiple MBC subsets were first identified and phylogenetic trees estimated using IgPhyML. From these trees, distance between tips (sum of branch lengths) was calculated using the distTips function from the adephylo R package [78], and assembled into a square matrix per lineage. To draw edges between related nodes, it was first necessary to define a distance threshold.

For this, the smallest tip-to-tip distance was identified for each tree, and a density curve created. For this curve, the lagged differences in y-axis value were divided by the lagged differences in x-axis, aiming to identify the position where the curve was most rapidly changing. This position was identified by taking the apex of the first identified peak (identified using the pracma package [79]), and this x-axis position was the selected cut-off. By choosing the position where the aggregated tip-to-tip distances were most rapidly changing, we aimed to identify a relevant distance where BCRs were linked to neighbouring BCRs and not linked to distant BCRs across multiple phylogenetic trees. This threshold was used to derive an adjacency matrix, whereby nodes that had a distance smaller than the calculated threshold were connected. The network was then directed by only allowing connections from nodes that had a smaller distance from the inferred germline node to those with a larger distance from the inferred germline node. Statistics were then calculated for each node in each network, namely the hub score (networkR; [80]) and the degree-ratio (igraph). Higher hub scores are achieved when a node has many outgoing connections and is further influenced by the quality of these downstream connections. Therefore, in our phylogenetic networks, nodes with higher hub scores reflect nodes with more close evolutionary relationships and indicates antigen driven SHM. The degree-ratio (number of outgoing connections divided by the number of incoming connections) was calculated using igraph and reflects the local environment of each node in the network. Statistical analysis was performed using Mann-Whitney U tests, via the ggpubr package (p > 0.05 not marked, p < 0.05 *, p < 0.01, **, p < 0.001 ***, p < 0.0001 ****) [81].

Smart-Seq2 single-cell data analysis

Data were then processed using scanpy (v.1.9.8) [82] workflow with standard quality control steps; cells were filtered if number of genes >6000 or <600. Mitochondrial content was determined using scanpy.pp.calculate_qc_metrics function; cells with mitochondrial genes percentage <50% were retained for further analyses. Genes were retained if they were expressed by at least 2 cells. Gene counts for each cell were normalised to contain a total count equal to 10^6 counts per cell. This led to a working dataset of 301 cells. The top 2000 highly variable genes were selected based on Seurat v.3 algorithm (flavor = seurat_v3) with batch key “Sequencing_batch”. Highly variable genes were further refined by removing potentially confounding genes using the following search formula: ‘^HLA|^IG[HKL][VDJC]|^MT|^A[A-Z][0-9]|^B[A-Z][0-0]’. The number of principal components used for neighbourhood graph construction and dimensional reduction was set at 20. Data integration from both donors was performed using the bbknn algorithm [83]. Uniform Manifold Approximation and Projection (UMAP; v3.10.0) [84] was used for dimensional reduction and visualization with all parameters as per default settings in scanpy. For the assessment of transcriptional similarity between Env-reactive B cells and MBC reference subsets, Celltypist [85] probability scores were generated as per default settings.

10X data analysis

Statistics

Neutralization and binding data were analysed using GraphPad Prism. For BCR phylogenetic network analysis statistical significance was assessed using pairwise Mann-Whitney U tests with the compare_means function from the ggpubr package (p > 0.05 not marked, p < 0.05 *, p < 0.01, **, p < 0.001 ***, p < 0.0001 ****). IgG subclass usage and nucleotide somatic hypermutation levels (V + J gene) of sequenced memory B cell populations were analysed using a one way-ANOVA with Tukey’s multiple comparison used for statistical testing. *p > 0.032, **p > 0.021, ***p > 0.0002 and ****p > 0.0001. Comparative analysis of differential gene expression from cell subsets used a Wilcoxon rank-sum test with Benjamini-Hochberg correction used for statistical testing, only genes with adjusted p-value < 0.05 and log-fold change >1 shown.