The research team, led by scientists at Human Technopole in Milan, purified naive B cells, which have never encountered an antigen, and unswitched memory B cells, which carry the imprint of past infections, from the blood of healthy donors. They then activated both cell types in the laboratory using a cocktail that mimics T-cell-dependent stimulation: a cross-linker that engages the B cell receptor, multimeric CD40L to trigger the costimulatory CD40 receptor, and the cytokines IL-2 and IL-21. Over six days of culture, the researchers captured single-cell RNA sequencing and B cell receptor sequencing at four time points, from the resting state through the first cell division to full differentiation, retaining nearly 119,000 cells for analysis.

The first surprise came from comparing the two cell types at rest. At day zero, naive and memory B cells showed highly similar gene regulatory network activity, with a correlation of 0.89 across transcription factor regulons. Yet as activation progressed, the two populations drifted apart, and by day six the correlation had fallen to just 0.4. The transcriptional response to the very same stimulus was therefore profoundly cell-type-specific, with more than 3,000 genes differentially expressed between the two populations at the final time point.

That divergence translated into radically different fates. Memory B cells, when activated, almost exclusively differentiated into plasmablasts and plasma cells, the antibody factories of the immune system. Their gene expression was dominated by signatures of endoplasmic reticulum stress and the unfolded protein response, pathways such as IRE1α activating chaperones and XBP1(S) activating chaperone genes, which are hallmarks of cells ramping up massive antibody production. Naive B cells, by contrast, bifurcated. One branch produced plasma cells indistinguishable from those generated by memory cells, while the other adopted a germinal center phenotype marked by high expression of BCL6, CD81 and LIMK1, together with markers of memory precursors such as CCR6 and HHEX. An independent computational approach based on optimal transport, called Moscot, confirmed that germinal center cells arose almost exclusively from proliferating naive B cells, whereas plasmablasts could originate from either population.

The engine behind this fork in the road proved to be a pair of antagonistic transcription factors. The researchers applied the SCENIC framework to infer the activity of gene regulatory networks, or regulons, across the time course. They found that the activity of IRF4, the master driver of plasma cell differentiation, rose steadily in memory B cells from the earliest hours of activation and continued climbing throughout, culminating in expression of its downstream effector PRDM1. In naive B cells, IRF4 activity showed only a transient early increase before declining, persisting at high levels solely in the subset of cells that ultimately committed to the plasma cell fate. The activity of SPI1, also known as PU.1, behaved in exactly the opposite manner: strongly suppressed in memory cells, it rebounded in naive cells and peaked by day six, characterizing the germinal center state. Imaging flow cytometry revealed a mutually exclusive pattern of nuclear localization, with cells showing nuclear IRF4 consistently negative for nuclear PU.1 and vice versa, underscoring the antagonism between the two programs.

To test causality, the team deployed CRISPR-Cas9 to knock out IRF4 and PRDM1 in activated B cells, achieving knockout efficiencies of roughly 98 and 90 percent respectively. The consequences were dramatic. Loss of IRF4 blocked plasma cell differentiation in both naive and memory B cells and pushed the cells toward an aberrant germinal-center-like state that expressed canonical GC markers but differed transcriptionally from naturally arising germinal center cells by more than 2,600 genes. Knockout of PRDM1, in contrast, selectively eliminated plasma cells while preserving the capacity to form germinal center cells that were transcriptionally indistinguishable from those arising in control cultures. The conclusion is that PRDM1 is essential only for the plasma cell program, whereas IRF4 plays a dual and indispensable role, coordinating both plasma cell differentiation and proper germinal center development.

Perhaps the most conceptually striking finding concerns clonal behavior. By culturing only about 2,000 naive B cells and sequencing the entire culture at day six, the researchers identified 265 clones with at least ten cells each. A large proportion of these clones contained both plasma cells and germinal center cells, demonstrating that a single ancestral B cell can give rise to both fates. This rules out a simple population-level model in which different cells are pre-committed to different outcomes, and instead points to stochastic fluctuations in IRF4 expression within individual cells as the source of the bifurcation, particularly because the polyclonal stimulation system excludes any selection by antigen affinity.

The clonal analysis also revealed that gene expression itself is heritable. Sister cells within a clone showed significantly correlated expression of key plasma cell genes, including XBP1 and the immunoglobulin heavy chain gene IGHM, whereas random cell pairs from the same donor showed no such correlation. Genes regulated by XBP1 displayed especially strong intraclonal correlation, indicating that transcriptional programs central to the plasma cell lineage are inherited across cell divisions. To quantify this, the researchers trained three machine learning classifiers, logistic regression, a neural network and a random forest, to predict clonal identity from gene expression alone. Using a set of 369 genes with clonal correlation above 0.25, the logistic regression model correctly assigned sister cells with an average accuracy of nearly 88 percent, and the performance held up even after cell cycle and cell state effects were regressed out. Validation on eleven independent in vivo mouse B cell clonal datasets confirmed that the phenomenon is not an artifact of the culture system.

The study also shed light on antibody class switching, the process by which B cells replace the constant region of their antibody heavy chain to produce IgG, IgA or IgE while retaining antigen specificity. Using the sciCSR pipeline to quantify germline transcripts, the short non-coding RNAs that mark switch regions poised for recombination, the researchers found that more than 75 percent of germinal center cells expressed germline IGHG3, compared with only about 25 percent of plasma cells. Yet this priming did not translate into more switching, confirming that germline transcription is necessary but not sufficient for class switch recombination. Critically, by comparing IgM-positive cells from clones that contained switched cells with IgM-positive cells from entirely unswitched clones, the team found no transcriptional differences between the two groups. Belonging to a clonal lineage in which a sister cell has already switched therefore confers no increased predisposition to switch, providing direct experimental support for mathematical models proposing that class switching is a stochastic, cell-intrinsic event.

Together, these findings sketch a model in which each individual B cell is governed by a unique combination of competing gene programs that regulate cell state, clonal gene expression and isotype switching, maximizing the diversity of the immune response. Memory B cells carry an intrinsic bias toward plasma cell differentiation, driven by sustained IRF4 activity, yet they retain the latent capacity to adopt a germinal center fate when PRDM1 is suppressed, a discovery that could open new strategies for directing memory cells toward antibody refinement. Naive B cells, meanwhile, hedge their bets, with stochastic IRF4 dynamics splitting their progeny between immediate antibody secretion and the germinal center where future memory is made. As the authors note, gene regulatory network inference from single-cell transcriptomics remains inherently correlative, and future work with genome-wide transcription factor binding assays and high-throughput CRISPR screens will be needed to map the causal wiring in full. Even so, the study provides the most detailed picture yet of how human B cells decide their destinies, knowledge that is directly relevant to understanding responses to infection, vaccination and autoimmunity.

Subject of Research: Gene regulatory network dynamics controlling human naive and memory B cell differentiation and fate decisions

Article Title: Competing gene regulatory networks drive naive and memory B cell differentiation

Article References: Competing gene regulatory networks drive naive and memory B cell differentiation. (n.d.). https://doi.org/10.1038/s44320-026-00207-8

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00207-8

Keywords: B cells, gene regulatory networks, plasma cells, germinal center, IRF4, PRDM1, single-cell RNA sequencing, CRISPR, class switch recombination, clonal tracking, immune memory, transcription factors