Understanding repertoire sequencing data through a multiscale computational model of the germinal center

The germinal center (GC) is crucial for the adaptive immune response, responsible for affinity maturation of B-cell receptors (BCRs). The GC reaction begins with antigen (Ag)-specific B cell activation, clonal expansion forming the dark zone (DZ). Centroblasts (CBs) undergo somatic hypermutation (SHM), differentiating into centrocytes (CCs) migrating to the light zone (LZ). In the LZ, CCs interact with follicular dendritic cells (FDCs) and T follicular helper (Tfh) cells for selection, returning to the DZ for further rounds of proliferation and SHM. Memory B cells (MBCs) and plasma cells (PCs) are the output cells (OCs). MBCs generally have lower affinity than PCs, produced early in the GC reaction, while higher-affinity PCs are produced later. Mammals possess a vast immune repertoire, estimated at 10¹⁵ naive B cells, with BCR diversity stemming from V(D)J recombination and SHM. Next-generation sequencing profiles immune receptor repertoires, yielding clone abundances. High-abundance clones, or dominant clones, are candidates for further characterization. However, measuring affinities for all clones is infeasible. Also, RNA sequencing may inflate clone abundance due to high immunoglobulin RNA content in PCs. Previous ODE models showed limited correlation between clonal abundance and affinity, but lacked individual clone analysis and adequate representation of low-frequency subclones. This study aimed to extend a multiscale model to investigate the relationship between abundance and affinity, intra-clonal affinity variation, and the effect of PCs on dominant clone identification by RNA-seq.

The literature extensively documents the germinal center reaction and B-cell repertoire sequencing. Studies highlight the importance of germinal centers in affinity maturation and the generation of diverse antibody repertoires. Methods like next-generation sequencing have advanced our understanding of BCR and TCR repertoires, enabling analysis in various contexts. However, limitations remain. Challenges in linking sequencing data to functional properties like affinity and specificity are well-established. The impact of high immunoglobulin mRNA levels in PCs on the interpretation of RNA-seq data has also been noted. Existing models have made contributions, but lacked the detail required for comprehensive analysis. This review sets the stage for a more detailed multiscale model.

This research employed an extended multiscale (eMS) model of the germinal center, building upon a previous multiscale model. The eMS model integrates an agent-based model (ABM) for cellular dynamics and a system of ordinary differential equations (ODEs) representing a core gene-regulatory network (GRN) involved in PC differentiation. Key improvements included a BCR sequence representation for each B cell and OC, using a SHM fate tree to determine mutation region and type. The BCR representation focused on the immunoglobulin heavy chain (IgH). Affinity was determined based on distance in a continuous shape space. Nine simulations generated DNA and RNA-based repertoires. Dominant clones were defined as either the upper 25% of most frequent clones or those representing ≥0.5% of total sequences. Results were compared with experimental repertoires from blood, tissue, and single GCs. The model tracks frequencies and affinities of all clones and subclones. The number of clones and subclones was tracked across the GC reaction (21 days), analyzing their affinities. The relationship between (sub)clonal abundance and affinity was assessed. The effect of a 100-fold higher immunoglobulin mRNA abundance in PCs on dominant clone identification was also investigated by comparing DNA- and RNA-based repertoires.

Simulations revealed a limited correlation between clonal abundance and affinity, with substantial affinity variability within clones. High-affinity clones contained low-affinity B cells. A weak trend showed increasing median affinity with abundance up to approximately 100 cells, but plateaued for higher abundances. The 100-fold increase in BCR mRNA in PCs did not substantially affect the number of dominant clones identified from RNA-seq data of single GCs. Simulated repertoires from single GCs showed better agreement with experimental single-GC datasets compared to blood and tissue repertoires. The number of clones in simulations (4-18) was significantly lower than in most experimental datasets, likely because the model represents a single GC. The number of dominant clones was more comparable to experimental data but still on the lower side. The D50 index (fraction of clones accounting for 50% of sequences) was larger in simulations than in experimental data. The model captured the progression of clonal size, with dominant clones outgrowing the population average at approximately 120 hours. The number of subclones remained steady after the clonal expansion phase. The model revealed heterogeneity in clonal affinities at day 21, with high-affinity clones exhibiting low-affinity B cells. Low-abundance clones might have high affinity but are difficult to select experimentally.

The findings address the research question by demonstrating the limitations of relying solely on clonal abundance to infer affinity and function from repertoire sequencing data. The large affinity variation within clones highlights the incompleteness of characterizing only a single subclone from a dominant clone. This suggests that low-abundance, high-affinity clones could be overlooked when focusing solely on abundant clones, but their identification is challenging experimentally. The finding that PC abundance doesn't strongly affect dominant clone identification in single GCs is unexpected but likely due to the proportional increase in clone abundance from all cell types. This differs from blood or tissue samples, where PC proportions and clone dominance could be skewed by high BCR RNA. The model's limitations (single GC representation, simplified SHM and PC differentiation mechanisms, and the use of abstract shape space) need consideration for interpretation. Despite these, the model's improvement over previous versions enhances our ability to understand the complexities of B-cell repertoire data.

This study presents an advanced multiscale model that simulates germinal center dynamics and generates realistic B-cell repertoires. The findings highlight the limited correlation between clonal abundance and affinity, the considerable affinity variability within clones, and the minimal impact of PC abundance on dominant clone identification in single-GC RNA-seq data. These insights offer valuable guidance for interpreting repertoire sequencing data and designing targeted experiments. Future directions include refining the model to simulate various antigen-specific responses and incorporating more detailed mechanisms for B-cell differentiation and SHM, improving its ability to reflect the diversity and complexity of experimental repertoires.

The model has limitations such as representing only a single GC, simplifying SHM and PC differentiation, and using an abstract shape space to determine affinity. The model's inability to precisely control clonality at the end of the GC reaction might also influence the interpretation of results. These factors may influence the generalizability of the model's predictions. Furthermore, the lack of detailed consideration of MBC differentiation might affect the balance of OC types generated, impacting the influence of PCs on dominant clone identification. The limited number of simulations (nine) might not fully capture the range of possible outcomes.