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

The study addresses how to interpret B-cell receptor (BcR) repertoire sequencing data, which report clone lineages and abundances but not binding affinity or antigen specificity. In germinal centers (GCs), affinity maturation selects higher-affinity B cells, leading to the common assumption that dominant (highly abundant) clones also have high affinity. However, measuring affinities for large numbers of clones is impractical, and PCs may bias RNA-based abundance due to elevated immunoglobulin mRNA. The authors extend a prior multiscale GC model to simulate single-GC repertoires, aiming to quantify the relationship between clonal abundance and affinity, assess affinity variability within clones, and evaluate the impact of high PC mRNA on identifying dominant clones. They also compare simulated single-GC repertoires to experimental datasets from blood, tissues, and single GCs to understand deviations and guide experimental design.

Background covers GC biology, affinity maturation, and BcR structure and diversity. Repertoire sequencing has been widely used in vaccinology, infection, and immune disorders, typically yielding clones and abundances. It is generally assumed dominant clones have higher affinity due to GC selection, but prior ODE-based modeling suggested limited correlation between abundance and affinity and could not resolve individual clones or low-frequency subclones. Elevated immunoglobulin mRNA in PCs could bias RNA-based repertoire dominance calls. Existing repertoire simulators (e.g., SHazaM, AbSim, immuneSIM, immuneML) focus on V(D)J recombination and SHM to generate sequences and clonal lineages but do not simulate GC dynamics, affinity maturation, or output cell types (PCs, MBCs). Single-GC and single-cell studies are emerging, enabling better benchmarking of GC-resolved repertoire features.

The authors developed an extended multiscale (eMS) model of a single germinal center (GC), integrating: (1) a 3D agent-based model (ABM) for cellular dynamics (centroblasts, centrocytes, Tfh), chemokine-guided migration (CXCL12/13), DZ/LZ partitioning, cycles of proliferation, SHM, selection, and differentiation; and (2) an ODE-based core gene regulatory network (GRN) in each B cell for plasma cell (PC) differentiation involving BLIMP1, BCL6, and IRF4 with BCR and CD40 signaling inputs. Simulations run for 504 h (21 days) with 0.002 h time steps. Founder B cells (~200) enter the GC probabilistically over 96 h according to p(influx)=μΔt/(1+e^{-βt}), with μ=2 cells/h and β=6 h, matching early GC seeding. BcR representation and SHM: Each B cell and output cell carries an immunoglobulin heavy chain (IgH) sequence built from partially reconstructed V and J germline segments (CDR3 partly from experimental sequences) with annotated FWR1–4 and CDR1–3. Only IgH is modeled to align with common repertoire data. SHM is activated after 24 h with rate 10^{-3} mutations/bp/division; given ~400 nt IgH, m~Poisson(λ=0.4) mutations per division. A SHM fate tree assigns region (FWR/CDR), mutation type (replacement/silent), and effect (affinity-changing, lethal, neutral), based on mouse κ light-chain non-functional data due to lack of human heavy-chain data. Only replacement mutations in CDRs can change affinity; lethal mutations set affinity to zero causing apoptosis. Approximate probability that a daughter cell accrues an affinity-changing mutation per division is ~0.1. A lookup stores sequence–affinity pairs for consistency across the simulation. Affinity model: A continuous Perelson-style shape space (4D) is used with affinities in [0,1]. SHMs move the BcR in a random dimension by a step s~Normal(μ=1,σ=0.1); the L1 distance to the fixed antigen location is transformed to affinity via a Gaussian kernel. This facilitates affinity maturation but does not represent antigen-specific binding energetics. Differentiation rules: Positively selected CCs recycle to DZ; proliferation and GRN dynamics determine PC fate (PC if [BLIMP1] ≥ 8e−8 M). PCs and memory B cells (MBCs) are output cells (leave GC via DZ). MBCs arise via asymmetric antigen segregation during division (from the ABM) for output cells not meeting the BLIMP1 threshold, reflecting the observed early MBC vs later PC production. Repertoire generation: Nine stochastic simulations (different seeds) were run. DNA-based repertoires at day 21 count each cell (GC B cell, PC, MBC) once. RNA-based repertoires up-weight PCs by a factor of 100 to represent elevated immunoglobulin mRNA content (tested as an extreme upper bound). Clones are defined as lineages sharing a naive ancestor; subclones are groups of cells with identical BcR sequences. Dominant clones are defined by either the top 25% in abundance or those ≥0.5% of total counts. Metrics include D50, Berger-Parker, and Pielou’s evenness. The simulations are compared against multiple experimental repertoires: blood/tissue bulk datasets and single-GC datasets (human single-GC DNA repertoires; mouse steady-state gaGC single-cell RNA-Seq; mouse immunized single-GC single-cell RNA-Seq).

- Clonal counts and diversity: At day 21, simulations yielded a median of 11 clones (range 4–18) surviving from ~200 founders; maximum concurrent clones during the reaction approximated the number of founders (~200). D50 at day 21 was 0.13 (range 0.07–0.25), Berger-Parker index 0.47 (0.15–0.98), and Pielou’s evenness 0.60 (0.07–0.85). The number of dominant clones (≥0.5% threshold) at day 21 had median 9 (range 2–15). - Comparison to experimental data: Blood/tissue repertoires exhibited orders-of-magnitude more clones than single-GC simulations, as expected due to accumulation across many immune responses and many naive singletons. Single-GC datasets (particularly mouse immunized single-GC data) matched simulated ranges at corresponding time points, including metrics such as clone counts, dominant clone fractions, D50, Berger-Parker, and evenness. - Abundance–affinity relationship: There is only a weak trend between clone abundance and median affinity; median affinity increases from very low abundances up to ~10^2 cells and then plateaus, never reaching the maximum affinity (1.0). Many high-abundance clones have modest affinity, and many low-abundance clones have high median affinity. At the subclone level, low-abundance subclones span the full affinity range; higher-abundance subclones tend to have higher affinities with reduced variance. - Intraclone variability: Surviving clones at day 21 are highly heterogeneous in subclone affinities; even dominant clones contain subclones with very low affinity, indicating that selecting a single subclone for functional follow-up can misrepresent the clone’s overall properties. - PC mRNA bias on dominance calls: Up-weighting PCs by 100× in RNA-based repertoires typically did not change the number of dominant clones at day 21 compared to DNA-based repertoires in single-GC simulations. While PC up-weighting increases disparities between clone abundances and can swap some minor cases of dominance status depending on PC content, thresholds shift proportionally across clones because most clones contain mixed cell types. Therefore, within a single GC, high PC mRNA content by itself does not systematically inflate the number of dominant clones.

The simulations directly address whether dominance in repertoire data implies high affinity and whether PC mRNA content biases dominance calls. Results indicate only a limited correlation between abundance and affinity; many low-abundance clones/subclones can be high affinity, while some dominant clones are of moderate affinity. This cautions against assuming dominance equates to functionality and against characterizing clones based on a single subclone. Intraclone affinity diversity can exceed interclone differences, implying that clone-level conclusions from one sequence may be incomplete or misleading. Regarding PC mRNA bias, within a single GC context, higher PC transcript levels do not substantially alter the number of dominant clones because clone composition is mixed and dominance thresholds shift accordingly; however, in blood or tissue, where PCs may accumulate and sampling differs, PC-derived biases may still arise. Comparisons to single-GC datasets show good agreement in clonal metrics across time, validating aspects of the model; in contrast, bulk blood/tissue repertoires deviate as they represent multiple responses and compartments. Overall, the findings refine interpretation of repertoire sequencing by decoupling dominance from affinity and highlighting meaningful low-abundance candidates, guiding more nuanced selection strategies for experimental follow-up.

This work introduces an extended multiscale GC model that simulates BcR sequence evolution, affinity maturation, and output cell differentiation to generate single-GC repertoire features. Key contributions are demonstrating weak linkage between clonal abundance and affinity, substantial intraclone affinity heterogeneity, and a limited impact of high PC mRNA on dominance calls in single-GC RNA repertoires. Simulated single-GC clonality metrics align with single-GC datasets but not with bulk blood/tissue repertoires, as expected. These insights can inform selection of (sub)clones for functional characterization, including low-abundance high-affinity candidates, and improve interpretation of dominance metrics. Future directions include: incorporating antigen-specific structural/biophysical models beyond abstract shape space; expanding the GRN and cytokine inputs for PC differentiation; explicitly modeling MBC differentiation mechanisms; controlling GC selection pressure and founder numbers to tune clonality; and extending to multi-GC, temporal, and compartmental models to better mirror blood/tissue repertoires.

- Affinity modeled via abstract continuous shape space; not antigen-specific and cannot be directly compared to biophysical binding measurements. Only IgH is considered (no light chain pairing). - SHM fate probabilities derived from mouse κ light-chain data due to lack of human heavy-chain non-functional data; mutation hot/cold spots and key/blocking mutations are not modeled. Only CDR replacement mutations affect affinity; potential FWR effects on rigidity/entropy and affinity are ignored. - PC differentiation uses a minimal GRN (BLIMP1, BCL6, IRF4) and thresholding; cytokines and additional transcriptional regulation are not fully represented. MBC differentiation is implicit via asymmetric antigen segregation rather than an explicit molecular mechanism. - Single-GC context: PCs and MBCs exit the GC; accumulation and lifespans across compartments are not modeled, limiting applicability to blood/tissue repertoires. The number of output cells produced by a single GC is uncertain, making quantitative validation challenging. - Founder cell number is at the high end (~200); the model provides limited control over selection pressure to adjust end-of-GC clonality without disrupting GC dynamics.