Replaying germinal center evolution on a quantified affinity landscape
William S. DeWittiD, Ashni A. VoraiD, Tatsuya ArakiiD, Jared GallowayiD, Tanwee AlkutkariD, Juliana BortolattoiD, Tiago B. R. CastroiD, Will DummiD, Chris Jennings-ShafferiD, Tongqiu JiaiD, Luka MesiniD, Gabriel OzorowskiiD, Juhee PaeiD, Duncan RalphiD, Jesse D. BloomiD, Armita NourmohammadiD, Yun S. SongiD, Andrew B. WardiD, Tyler N. StarriD, Frederick A. Matsen IViD, Gabriel D. VictoraiD
Abstract
Darwinian evolution of immunoglobulin genes within germinal centers (GCs) underlies the progressive increase in antibody affinity following antigen exposure. Whereas the cellular mechanics of how competition between B cells produces increases in affinity are well established, the evolutionary dynamics of this process are less clear. We developed an experimental evolution model where we "replay" over one hundred monoclonal GC reactions, assigning affinities to each cell using deep mutational scanning. Our data reveal how GCs achieve predictable outcomes by means of noisy but persistent selection on an affinity landscape whose exploration is heavily constrained by somatic hypermutation biases. We infer a fitness landscape that quantitatively recapitulates the affinity maturation trajectory of our clone and find that apparent features of GC selection such as permissiveness to low-affinity lineages and rapid plateauing of affinity are likely artifacts of survivorship biases that distort our view of how B cell affinity progresses over time.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 30
- FWCI
- 75.17
- Field percentile
- top 1%
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Sonnet 4.6 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (12 May 2026) · CC BY 4.0
Across 119 monoclonal germinal centers, B cell affinity rises reliably despite divergent phylogenies. Selection is noisy but persistent; mutability limits exploration. A DMS-based landscape shows saturating fitness and survivorship biases.
Predicting germinal center outcomes matters for vaccine design, but the paper says the dynamics behind clonal bursts and the coexistence of high- and very low-affinity B cells are poorly understood. Replaying identical-founder GCs with deep-mutational-scanning affinities yields a quantitative fitness landscape and points to survivorship bias as the source of apparent low-affinity permissiveness and early plateaus.
Key results
ShownGerminal centers (GCs) seeded by one B cell clone diverge widely in tree shape but converge on affinity gain, beating neutral-drift simulations.
Median affinity exceeded naive in 117 of 119 GCs (0.87 ± 0.38 at 15 dpi; 1.00 ± 0.34 at 20 dpi). Neutral simulations gave -0.47 ± 0.24 and -0.70 ± 0.26. Data: 8,744 B cells from 119 GCs. · Parallel replay of evolutionary trajectories in…; GCs select consistently for increases in…
ShownSumming single-replacement effects from deep mutational scanning (DMS) predicts the affinity of multi-mutant clone 2.1 antibodies well enough for population-level analysis. Replacements were far more likely to be deleterious than affinity-enhancing.
DMS covered 4,158 of 4,161 replacements; 35.6% were deleterious versus 3.6% gaining at least 0.3 log10 affinity. Biolayer interferometry (BLI) of a 17-Fab ladder gave R^2 = 0.91 (slope 1.33). Mean absolute error was 0.17 for single and 0.64 for multiple mutants. · Determining the effects of somatic mutation on…
ShownSomatic hypermutation (SHM) targeting and codon accessibility, more than affinity benefit, determine which replacements GCs find. Many affinity-enhancing replacements stay unexplored even after prolonged selection.
Mutability predicted replacement frequency better than Δaffinity (Spearman 0.67 vs 0.22). Only 156 (1.8%) of 8,744 cells carried any of 64 beneficial multi-nucleotide replacements. No low-mutability ones appeared in top cells at 70 dpi. · Selection of individual aa replacements across…
ShownAffinity maturation comes from persistent, imperfect selection rather than clonal bursts. Affinity-losing lineages are rapidly terminated.
Largest bursts arose from cells at mean affinity percentile rank 66 (range 30-94) at 15 dpi and 64 (21-97) at 20 dpi. No marked high-affinity enrichment in the burst sector. The max-REI node beat its sisters in affinity on aggregate, imperfectly. Affinity-losing nodes were enriched at leaves. · Low-affinity B cell lineages are efficiently…; Clonal bursts are not the primary drivers of…
SuggestedA fitted affinity-fitness landscape for clone 2.1 rises linearly then saturates. Survivorship-bias modeling reproduces the early affinity jump and plateau in tree reconstructions, suggesting these are artifacts.
Initial slope 0.74 (a 10-fold affinity advantage gives 2.09-fold growth per day), saturating near x = 2 (400 pM) and plateauing at x = 3 (40 pM). Reweighting the model by lineage survival probability gave early-jump-then-slowing ancestral trajectories consistent with those from Bayesian time-resolved trees (100 posterior trees per GC). · A fitness landscape for the temporal evolution…; Phylodynamic survivorship bias distorts…
Methods, data and reproducibility
Monoclonal GCs were seeded by transferring chIgY B cells into GC-deficient hosts, photoactivated individually, and sequenced; affinities were assigned to each cell by an additive model built on yeast-display DMS (Tite-Seq). A PDE-based fitness landscape was fit by maximum likelihood to bulk time-course data, and extinction probabilities were derived from a birth-death-mutation master equation.
- Paired Igh/Igk sequences from individually photoactivated monoclonal GCs at 15 and 20 dpi (replay experiment) · mouse · n = 8,744 B cells from 119 GCs (52 GCs, 18 mice at 15 dpi; 67 GCs, 6 mice at 20 dpi)
- Yeast-display DMS (Tite-Seq) of clone 2.1 scFv, binding affinity and surface expression · mouse (antibody clone 2.1 from chIgY mice; displayed on yeast) · n = 4,158 of 4,161 single-aa replacements, duplicate libraries
- Passenger-allele (non-functional Ig) bulk sequencing of GC B cells after Plasmodium chabaudi infection, to measure intrinsic SHM mutability · mouse · n = 7.5 x 10^5 sorted GC B cells; pooled from 3 mice (Igh) and 2 mice (Igk)
- Bulk-sorted GC B cell time course, 10x single-cell Ig sequencing at 5, 8, 11, 14, 17, 20, 30 and 70 dpi · mouse · n = 4 mice per time point; 9 mice at 70 dpi
- Cryo-EM structure of a high-affinity 2.1 Fab bound to chicken IgY · mouse Fab; chicken antigen · n = 4.0 Å reconstruction · EMD-70353; PDB 9ODB
- BLI-measured recombinant Fabs for validating DMS-based affinity predictions · mouse (clone 2.1 variants) · n = 17-step ladder; 13 Fabs from clonal bursts
- Methods & tools
- gctree v4.3.0, partis, BEAST v1.10.4, Nextflow v24.04.3.5916, Tite-Seq, Cell Ranger, pRESTO, cryoSPARC, Phenix, Rosetta, Octet biolayer interferometry, custom Crank-Nicholson PDE solver
- Code
- https://github.com/matsengrp/gcreplay/https://matsen.group/gcreplay/key-files/#manuscript-figureshttps://github.com/jbloomlab/Ab-CGGnaive_DMS/blob/main/Titeseq-modeling.ipynb
- Data & accessions
- https://matsen.group/gcreplay-viz/
- Reagents & environment
- gctree v4.3.0; BEAST v1.10.4; Nextflow v24.04.3.5916; Cell Ranger v6.0.1, v7.0.1 or v8.0.1; cryoSPARC v3.2; Relion 3.0; FlowJo v10.
- Reproducibility gaps
- No accession for raw sequencing reads is printed; cryo-EM maps and coordinates are deposited. Reagents are available from the lead contact on reasonable request.
Limitations (4)
- Only one B cell clone (2.1, affinity about 40 nM) was studied. Features such as the burst-affinity link and the strictness of counterselection could differ where mean affinity is lower. · stated by the authors
- GCs were engineered to lack polyclonal competitors, so the system tests intraclonal but not interclonal competition, ignoring factors such as antibody-mediated feedback. · stated by the authors
- The additive DMS model ignores epistasis; error was larger for multi-mutant antibodies (0.64 vs 0.17 for single mutants), and BLI could not resolve affinities beyond about Δaffinity 2.5. · noted in this summary
- The fitness landscape and survivorship-bias correction come from a deterministic model fit to pooled bulk time-course data from one clone. The bias is not tested by a direct experimental manipulation. · noted in this summary
Who should read:Read this if you study affinity maturation, want a quantitative fitness landscape from DMS plus phylogenies, or must account for survivorship bias in lineage-tree analyses.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- Inference of germinal center evolutionary dynamics via simulation-based deep learning
Duncan Ralph, Athanasios G Bakis, … Frederick A. Matsen IV·eLifePeer-reviewed·Apr 28, 2026
- Antibody maturation in germinal centers selects mutants based on BCR-antigen bond mechanical resistance
Lama Awada, Jad Sleiman, … Philippe Robert·bioRxiv·Mar 13, 2026
- Proliferation and mutation as humoral immune response
Isaac Meilijson, Laura Lea Sacerdote·bioRxiv·Nov 8, 2026
- Germline-encoded specificities and the predictability of the B cell response
M. C. Vieira, A.-K. E. Palm, … S. Cobey·PLOS PathogensPeer-reviewed·Aug 25, 2023
- Quantitative modeling of the effect of antigen dosage on B-cell affinity distributions in maturating germinal centers
M. Molari, K. Eyer, … R. Monasson·bioRxiv·May 22, 2020
- Modeling the Synergetic Dynamics of B cells and TFH cells in Germinal Center Reactions
Andrew G. T. Pyo, Julia Merkenschlager, … Ned S. Wingreen·bioRxiv·Nov 27, 2025