Bcl-6 is the nexus transcription factor of T follicular helper cells (T_FH) via repressor-of-repressor circuits

Abstract

T follicular helper (T_FH) cells are a distinct type of CD4⁺ T cells that are essential for most antibody and B lymphocyte responses. T_FH cell regulation and dysregulation is involved in a range of diseases. Bcl-6 is the lineage defining transcription factor of T_FH cells and its activity is essential for T_FH cell differentiation and function. However, how Bcl-6 controls T_FH biology has largely remained unclear, at least in part due to intrinsic challenges of connecting repressors to gene upregulation in complex cell types with multiple possible differentiation fates. Multiple competing models were tested here by a series of experimental approaches to determine that Bcl-6 exhibited negative autoregulation and controlled pleiotropic attributes of T_FH differentiation and function, including migration, costimulation, inhibitory receptors, and cytokines, via multiple repressor-of-repressor gene circuits.

INTRODUCTION

The formation of germinal centers (GCs) is essential for the development of high-affinity memory B cells and antibody secreting long-lived plasma cells in response to pathogen infections or vaccinations¹. Follicular helper T cells (T_FH) provide key signals to antigen-specific B cells for the development of germinal center B (B_GC) cells^1,2. CD4⁺ T cells receiving T_FH inductive signals upregulate Bcl-6, the lineage defining transcription factor (TF) of T_FH cells^3–5. Upregulation of Bcl-6 is associated with expression of the chemokine receptor CXCR5 and reduction of CCR7 and PSGL1, among other molecules, allowing for migration to the T-B border and GCs¹, the sites where T_FH and then GC-T_FH cells interact with antigen-specific B cells. T_FH and GC-T_FH cells express many surface and secreted molecules that serve as positive markers and contribute to the differentiation (ICOS, IL-6Rα, PD-1), migration (CXCR5, CD69), and function (IL-21, IL-4, CXCL13, SAP, ICOS, PD-1, CD200, CD40L) of T_FH and GC-T_FH cells. GC-T_FH cells provide IL-21, IL-4, and CD40L that are required for B_GC cell survival, proliferation, and somatic hypermutation^1,2,6.

Bcl-6 function is critical in T_FH differentiation^3–5. Multiple TFs in addition to Bcl-6 have been identified that regulate T_FH differentiation^2,7–18. Inhibition of Blimp-1 (encoded by Prdm1) by Bcl-6 is required for T_FH differentiation³. Tcf-1 and Lef-1 are involved in early induction of Bcl-6 and repression of Blimp-1^19–21. Downregulation of Id2 is important for CXCR5 expression by releasing E protein TFs such as E2A and Ascl2^22,23. While the importance of Bcl-6 in T_FH cell development is clear, it is still unclear how Bcl-6 controls T_FH cell biology. Two studies using Bcl-6 ChIP-seq in human GC-T_FH cells and murine T_FH cells provided insights into Bcl-6–bound genes^24,25, but functional roles have remained largely untested, and there is no consensus on a mechanistic model of how Bcl-6 regulates T_FH cell biology. Bcl-6 and Blimp-1 are reciprocal antagonistic regulators of each other’s genetic loci³. That interaction provides a powerful mechanism for a genetic switch in cell differentiation, as co-expression of Bcl-6 and Blimp-1 is a metastable state²⁶. However, from an experimentalist perspective, their mutual antagonism confounds experimental designs to probe Bcl-6 (and Blimp-1) functions in CD4⁺ T cells. Additionally, the putative nature of Bcl-6 as a repressor in CD4⁺ T cells adds an extra layer of complexity for understanding gene regulation, as many signature T_FH genes are upregulated in the presence of Bcl-6. In the present study, we took a first principles based approach to defining and testing hypothetical models of how Bcl-6 may control T_FH biology.

RESULTS

T_FH differentiation is not a default pathway

One proposed model of T_FH differentiation is that T_FH differentiation is the default pathway for naive CD4⁺ T cells activated by antigen presenting cells. In this model, the primary role of Bcl-6 would be to inhibit Blimp-1 to allow activated CD4⁺ T cells to undergo a default T_FH differentiation pathway²⁷ (Extended Data Fig.1a–b). We tested this model by utilizing Bcl6^f/fPrdm1^f/fCre^CD4 mice. If T_FH differentiation is a default setting in activated CD4⁺ T cells, then when Blimp-1 is absent Bcl-6 would not be required. Bcl6^f/fPrdm1^f/fCre^CD4 CD45.1⁺ SMARTA cells were transferred into C57BL/6 mice, as were wild-type (WT), Bcl6^f/fCre^CD4, or Prdm1^f/fCre^CD4 SMARTA cells. Host mice were immunized with KLH-gp₆₁ in alum + cGAMP adjuvant (Fig.1a and Extended Data Fig.1c–d). Wild-type SMARTA cells differentiated into non-T_FH (CXCR5^loSLAM^hi), T_FH (CXCR5⁺SLAM^lo or CXCR5⁺PSGL1^int/lo), and GC-T_FH cells (CXCR5^hiPSGL1^lo or CXCR5^hiPD-1^hi) after KLH-gp₆₁ immunization (Fig.1b). Prdm1^f/fCre^CD4 SMARTA cells predominantly differentiated into T_FH and GC-T_FH cells. Bcl6^f/fCre^CD4 CD4⁺ T cells did not differentiate into T_FH cells^3,28. Notably, Bcl6^f/fPrdm1^f/fCre^CD4 CD4⁺ T cells failed to differentiate into T_FH and GC-T_FH cells. Similar results were observed in the context of an acute viral infection (see Supplementary Note; Extended Data Fig.1e–g). Adoptive transfer of Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells demonstrated that the T_FH differentiation defect was antigen-specific and CD4⁺ T cell-intrinsic²⁹. Signature T_FH surface markers were examined to determine whether Bcl6^f/fPrdm1^f/fCre^CD4 CD4⁺ T cells became bona fide T_FH cells in response to acute LCMV infection. Expression of T_FH signature surface proteins was dysregulated (Fig.1c), indicating that Bcl-6 has important functions in gene regulation beyond repression of Blimp-1 that are necessary for T_FH differentiation in both immunization and viral infection contexts.

Figure 1. T_FH differentiation is not the default pathway.

a, Schematic of the SMARTA cell transfer system used for KLH-gp₆₁ immunization. wild-type, Bcl6^f/f, Prdm1^f/f, and Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4⁺ T cells were transferred to C57BL/6 host mice, followed by immunization of the host mice with KLH-gp₆₁ in alum + cGAMP adjuvant, and analyzed 8 days later. See Fig. 1b and Extended Data Fig.1c–d.

b, Representative flow cytometry of GC-T_FH, T_FH and non-T_FH SMARTA cell subsets from draining LNs (dLNs) of KLH-gp₆₁ immunized mice. Three independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

c, wild-type, Bcl6^f/f, Prdm1^f/f, and Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4⁺ T cells were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days later. Relative protein expression of GC-T_FH core signature markers, gated on CXCR5⁺ T_FH cells. gMFI value of each gene was normalized to wild-type. Three independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test. See Extended Data Fig.1e.

d, Schematic of the SMARTA cell transfer system used for LCMV_Arm infection or KLH-gp₆₁ immunization. Wild-type, Bcl6^f/f, Prdm1^f/f, and Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4⁺ T cells were transferred to Bcl6^f/fPrdm1^f/f Cre^CD4 host mice, followed by infection or immunization of the host mice with LCMV_Arm or KLH-gp₆₁ in alum + cGAMP adjuvant, and analyzed 8 days later. See Fig.1e–i and Extended Data Fig.1h–i.

e, Representative flow cytometry gate of B_GC cells from spleens of LCMV_Arm infected mice. Three independent experiments performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

f, Representative flow cytometry of B_GC cells and B_PC from dLNs of KLH-gp₆₁ immunized mice. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

g, Histology of dLNs at d8 after KLH-gp₆₁ immunization in Fig.1f. Magnified images from Extended Data Fig.1g are shown. Blue, TCRβ; red, GL7; green, IgD; white, CD45.1 SMARTA. SMARTA cells were presented with large dots for clarity. Scale bar, 200 μm.

h, Quantification of results in Fig. 1i. Total counted number of SMARTA cells were indicated.

i, Serum antigen-specific IgG endpoint titers at d8 after KLH-gp₆₁ immunization in Fig.1f. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

To assess the migration and function of Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells, we transferred SMARTA cells into Bcl6^f/fCre^CD4 mice, followed by infection of the host mice with LCMV_Arm (Fig.1d). Bcl6^f/fCre^CD4 mice were used as recipient mice to eliminate endogenous T_FH help to B cells. Bcl6^f/fCre^CD4 mice receiving Bcl6^f/fCre^CD4 SMARTA cells did not generate B_GC cells (FAS⁺PNA⁺) in response to LCMV_Arm infection. Notably, mice receiving Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells failed to generate B_GC cells, in contrast to mice receiving wild-type or Prdm1^f/fCre^CD4 SMARTA cells (Fig.1e). Similarly, B_GC and plasma cell (B_PC; IgD^loCD138^hi) responses were negligible in mice receiving either Bcl6^f/fPrdm1^f/fCre^CD4 or Bcl6^f/fCre^CD4 SMARTA cells in response to KLH-gp₆₁ immunization (Fig.1d,f). ~50% of wild-type SMARTA cells migrated into B cell follicles and GCs (Fig.1g–h and Extended Data Fig.1h–i). Bcl6^f/fCre^CD4 SMARTA cells were mostly excluded from the B cell follicle³. Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells did not generate any histologically observable GCs and exhibited a migration pattern indistinguishable from Bcl6^f/fCre^CD4 SMARTA cells. IgG titers were significantly decreased in mice receiving Bcl6^f/fPrdm1^f/fCre^CD4 or Bcl6^f/fCre^CD4 SMARTA cells (Fig.1i). Altogether, we conclude that differentiation into T_FH cells is not the default pathway of activated CD4⁺ T cells, and Bcl-6 has important activities beyond inhibition of Prdm1 for instructing functional GC-T_FH and GC development.

Bcl-6 is an autoregulatory repressor in CD4⁺ T cells

In B cells, Bcl-6 is generally considered an obligate repressor of transcription, but Bcl-6 mechanisms of action have been controversial in CD4⁺ T cells. While Bcl-6 expression positively correlates with expression of many genes in T_FH cells, including genes with Bcl-6 binding sites^24,25, a mechanistic connection between Bcl-6 binding and gene regulation has been lacking. One example target gene of interest is Bcl6 itself. Bcl-6 binds to its own promoter in human and mouse GC-T_FH cells^24,25. This Bcl-6 binding site (Bcl6 Promoter Site 1; BPS1) sequence is conserved among mammals (Extended Data Fig.2a). Given that Bcl-6 expression positively correlates with T_FH differentiation, Bcl-6 has been considered a plausible candidate for positive regulation by Bcl-6. In contrast, there is evidence in B cell tumor lines that BCL-6 exhibits negative autoregulation³⁰. To test whether Bcl-6 acts as a repressor or an activator of its own expression in CD4⁺ T cells, we first utilized a self-inactivating (SIN) retroviral vector (RV) to measure Bcl6 promoter activity (Fig.2a and Extended Data Fig.2b). SMARTA cells were transduced with the wild-type Thy1.1-RV (an RV construct containing the proximal Bcl6 promoter upstream of a Thy1.1 reporter) or ΔBPS1 Thy1.1-RV (a mutated Bcl6 promoter construct with an 8-nt deletion mutation), transferred to recipient mice, and Bcl6 promoter activity was analyzed in T_FH and T_H1 cells after acute LCMV infection (Extended Data Fig.2c–d). Wild-type Bcl6 promoter activity (Thy1.1 expression) was reduced in T_FH cells compared to T_H1 cells. ΔBPS1 Bcl6 promoter activity was increased in T_FH cells in comparison to the wild-type Bcl6 promoter (Fig.2b). Thus, Bcl-6 appears to repress Bcl6 promoter activity in T_FH cells by binding of Bcl-6 to the BPS1 locus.

Figure 2. Bcl-6 exhibits direct negative autoregulatory feedback.

a, Schematic diagram of Bcl6 promoter RV plasmid. Wild-type or ΔBPS1 Bcl6 promoter Thy1.1-RV were generated based on pQdT SIN vector.

b, Representative flow cytometry and quantification of flow cytometry gate of Thy1.1 reporter positive cells, gated on CXCR5⁺ T_FH or CXCR5^lo T_H1 cells from spleens of C57BL/6 host mice given SMARTA CD4⁺ T cells transduced with wild-type-RV or ΔBPS1-RV, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days after infection. Three independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test. See Extended Data Fig.2c–d.

c,d, Phenotyping of wild-type and ΔBPS1 SMARTA cells from C57BL/6 host mice given wild-type or ΔBPS1 SMARTA CD4⁺ T cells, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days after infection. Representative flow cytometry of T_FH and GC-T_FH SMARTA cell subsets from spleens of LCMV_Arm infected mice. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test. See Extended Data Fig.2 for experimental scheme (g) and quantification of gene expression level (h).

e, SMARTA CD4⁺ T cells transduced with shCd8-RV or shNcor1-RV were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days after infection. Representative flow cytometry of GC-T_FH RV⁺ SMARTA cell subsets from spleens of LCMV_Arm infected mice. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

To test whether Bcl-6 binding to the endogenous Bcl6 promoter affects Bcl-6 expression in vivo, we generated a new CRISPR mouse line possessing the 8-nt ΔBPS1 mutation. ΔBPS1 or littermate control (WT) SMARTA cells were transferred into C57BL/6 mice and then host mice were infected with LCMV_Arm (Extended Data Fig.2e–f). ΔBPS1 SMARTA cells had highly increased Bcl-6 expression only in T_FH cells, not in T_H1 cells, indicating that BPS1 acts as a cis-regulatory element of Bcl6 expression only when cells express elevated Bcl-6. Deletion of BPS1 increased the frequency of T_FH cells (CXCR5^hiSLAM^lo) and GC-T_FH cells (CXCR5^hiPSGL1^lo or CXCR5^hiBcl-6^hi; Fig.2c–d). Changes in T_FH-associated proteins were observed specifically in T_FH cells (see Supplementary Note; Extended Data Fig.2g).

Ncor1 is a Bcl-6 corepressor³¹. BCL-6 binding at the BCL6 promoter locus overlapped with NCOR binding in a human B cell line (Extended Data Fig.2h). To determine whether the Bcl-6 autoregulation in CD4⁺ T cells involved Ncor1, we transferred SMARTA cells expressing an shRNAmir targeting Ncor1 (shNcor1-RV) or a negative control shCd8 into B6 mice followed by an acute LCMV_Arm infection. shNcor1⁺ SMARTA cells exhibited enhanced GC-T_FH cell development and Bcl-6 expression (Fig.2e). In sum, these data indicate that Bcl-6 represses its own expression in CD4⁺ T cells, mediated in conjunction with co-repressor Ncor1, in a negative autoregulatory feedback loop at the Bcl6 promoter, dampening T_FH and GC-T_FH cell accumulation.

Simple circuitry repressor-of-repressors model of Bcl-6

The findings above excluded the simplest model of Bcl-6 regulation of T_FH differentiation. A logical model for how Bcl-6 functions as the lineage defining TF of T_FH biology is that Bcl-6 instructs positive T_FH gene expression by a repressor-of-repressors mechanism. A simple circuitry model can be proposed (Fig.3a) wherein Bcl-6 inhibits a set of repressor TFs (“Bcl6-r” TFs, directly inhibited by Bcl-6) that in turn repress genes positively associated with T_FH biology (“Bcl6-rr” genes, genes inhibited by repressor TFs targeted by Bcl-6). Alternative cell fates (i.e., non-T_FH or T_H1/T_H2/T_H17/iT_REG) and genes downregulated as part of T_FH cell migration or function (e.g., Selplg, encoding PSGL1) may be downregulated by Bcl-6 directly in this simple gene circuitry model. Testing this model is difficult because of the mutually antagonistic relationship of Bcl-6 and Blimp-1. We reasoned that CD4⁺ T cells deficient in both Bcl-6 and Blimp-1 would be needed to gain insights into TFs directly regulated by Bcl-6.

Figure 3. A testable simple circuitry model of T_FH differentiation.

a, A hypothetical model of Bcl-6 on the regulation of non-T_FH and T_FH genes. Bcl6-r, genes repressed by Bcl-6. Bcl6-rr, genes repressed by repressors that are repressed by Bcl-6.

b, Schematic of the SMARTA cell transfer system used for RNA-seq analysis. T_H1 (CXCR5^loSLAM^hi) populations from wild-type and Bcl6^f/f Cre^CD4 SMARTA cells, T_FH (CXCR5^hiSLAM^lo) populations from wild-type and Prdm1^f/f Cre^CD4 SMARTA cells, T_H1-like (CXCR5^loSLAM^int) and T_FH-like (CXCR5⁺SLAM^int) populations from Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA cells were sorted from spleens of C57BL/6 host mice given wild-type, Bcl6^f/f, Prdm1^f/f, and Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4⁺ T cells, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days later. Naive SMARTA cells were isolated as CD44^loCD62L^hiCD45.1⁺ from uninfected mice. Representative flow cytometry of T_FH, T_H1, T_FH-like, and T_H1-like subsets from three independent experiments.

c, Upper, scatter plot of genes upregulated (red) or downregulated (blue) in T_FH cells relative to their expression in T_H1 cells (1.4-fold cut off, Adj. P <0.05). Lower, volcano plots of gene expression changes between wild-type T_FH cells and Bcl6^f/fPrdm1^f/f T_FH-like cells (horizontal axis) against Adj. P-value (vertical axis). Numbers indicate total and percent of those genes upregulated in wild-type T_FH cells (top left) or Bcl6^f/fPrdm1^f/f T_FH-like cells (top right). T_H1 cell-associated genes, upregulated in wild-type T_H1 cells versus wild-type T_FH cells (WT T_H1 > WT T_FH); T_FH cell-associated genes, upregulated in wild-type T_FH cells versus wild-type T_H1 cells (WT T_FH > WT T_H1). Adj. P-values for multiple test correction were determined using Benjamini-Hochberg algorithm.

d, Gene expression changes were clustered by MAP-DP analysis. Scale, row z-score. DKO, Bcl6^f/fPrdm1^f/f Cre^CD4.

e, Left, a hypothetical model of Bcl-6 regulation of Bcl6-rr T_FH⁺ genes in Cluster 4 by inhibition of Bcl6-r TFs in Cluster 1. Right, a hypothetical model of Blimp-1 regulation of Blimp-1-rr T_H1⁺ genes in Cluster 2 by inhibition of Blimp-1-r TFs in Cluster 3.

f, Multiple gene set enrichment analyses (GSEA) for identifying and comparing gene signatures in CD4⁺ T cells between subpopulations. Blue indicates a negative association, and red indicates a positive association. Circle size is proportional to NES (scale: 1.5–3.0). Tint indicates adjusted P-value that were corrected using Benjamini-Hochberg algorithm.

g, GSEA of BCL-6 bound genes from human tonsillar GC-T_FH²⁴ compared to Cluster 1 genes (left) or Cluster 4 genes (right) differentially expressed between Bcl6^f/fPrdm1^f/f T_FH-like cells and Prdm1^f/f T_FH cells. The ticks below the line correspond to the rank of each gene that is defined by p-value of gene expression change between Bcl6^f/fPrdm1^f/f T_FH-like and Prdm1^f/f T_FH. NES, normalized enrichment score; FDR, false discovery rate.

h, GSEA of Blimp-1 bound genes from activated CD8 T cells⁵⁴ in comparison of Cluster 3 genes differentially expressed between Bcl6^f/fPrdm1^f/f T_H1-like cells and Bcl6^f/f T_H1 cells.

To identify the putative set of Bcl6-r TFs, we conducted RNA-seq gene expression profiling of Bcl6^f/fPrdm1^f/fCre^CD4, Bcl6^f/fCre^CD4, Prdm1^f/fCre^CD4, and wild-type SMARTA T_FH and T_H1 cells generated in response to acute LCMV_Arm infection or KLH-gp₆₁ immunization (Fig.3b and Extended Data Fig.3a). As a first analysis the effect of Bcl6/Prdm1 double-deficiency on the T_FH biology, we assessed expression of a broad curated set^19,22,25 of T_FH-associated genes across all samples from RNA-seq gene expression profiling. Bcl6^f/fPrdm1^f/fCre^CD4 T_FH-like cells lost expression of positively T_FH-associated genes in comparison to wild-type T_FH cells or Prdm1^f/fCre^CD4 T_FH cells. Conversely, Bcl6^f/fPrdm1^f/fCre^CD4 T_H1-like cells had a gene expression profile different from wild-type T_H1 or Bcl6^f/fCre^CD4 T_H1 cells (Extended Data Fig.3b). Principal component analysis provided similar findings (Extended Data Fig.3c), supporting the overall hypothesis that T_FH is not a default differentiation pathway of CD4⁺ T cells and Bcl-6 has important activities beyond inhibition of Prdm1. We next characterized the effect of Bcl6/Prdm1 double-deficiency on the expression of all T_FH- and T_H1-associated genes (Extended Data Fig.3b–c). Bcl6^f/fPrdm1^f/fCre^CD4 T_FH-like cells had reduced expression of ~88% of genes upregulated in wild-type T_FH cells (Fig.3c). These data indicate that the vast majority of genes upregulated in T_FH cells required Bcl-6 for proper induction. Furthermore, Bcl6^f/fPrdm1^f/fCre^CD4 T_FH-like cells had increased expression of ~90% of genes upregulated in wild-type T_H1 cells, suggesting that Bcl-6 was required to properly repress most T_H1-associated genes. Taken together, Bcl-6 is broadly important for both induction of T_FH genes and repression of non-T_FH genes, consistent with, and expanding upon, previous observations^3–5,24,29.

To identify genes likely to be directly repressed by Bcl-6, we analyzed patterns of gene expression changes between the six different T_FH and T_H1 populations and naive CD4⁺ T cells. K-means analysis (K=10) and hierarchical clustering bioinformatic approaches both readily separated four distinct major gene expression patterns (Extended Data Fig. 3d–e). To obtain gene lists associated with those four major cluster patterns, MAP-DP clustering was performed (Fig.3d). We then attempted to apply our simple circuitry model of Bcl-6 function to the MAP-DP clustering outcomes. Given that the data sets also include modulation of Blimp-1 expression, and that Blimp-1 has major effects on T_H1 versus T_FH differentiation, we posited that a similar circuitry model of Blimp-1–mediated gene regulation may need to be included (Fig.3e). We therefore assessed whether the four major gene expression patterns regulated in T_FH and T_H1 cells could be largely accounted for by this simple circuitry model of Bcl-6 and Blimp-1 functioning as repressors. The model predicts that T_FH upregulated genes (Bcl6-rr) are upregulated via a Bcl-6 repressor-of-repressors mechanism (Fig.3e). If the model was accurate, these Bcl6-rr genes would correspond to Cluster 4, as expression of such genes would be reduced in Bcl6^f/fPrdm1^f/fCre^CD4 T_FH-like cells compared to Prdm1^f/fCre^CD4 T_FH cells (Fig.3d–e). Indeed, many genes upregulated in wild-type T_FH cells and important for the differentiation and function of T_FH cells were observed in Cluster 4, including Cxcr5, Icos, Cd200, Pdcd1, Sh2d1a, Tcf7, Lef1, Tox, Tox2, Il6ra, IL4, and Il21.

The model predicts that genes directly repressed by Bcl-6 (Bcl6-r) would fall into Cluster 1 (Fig.3e). Genes associated with alternative cell fates directly repressed by Bcl-6 would also group in Cluster 1. Important genes associated with non-T_FH fates do indeed group in Cluster 1²³ (Fig.3d). Bcl6^f/fPrdm1^f/fCre^CD4 T_FH-like cells had increased T_H1, T_H2, and T_REG signature genes expression (Fig.3f). Thus, multiple analytical approaches identified major gene networks and expression changes consistent with the proposed Bcl-6 repressor-of-repressors model.

If the Bcl-6 repressor-of-repressor model was accurate, a substantial proportion of Cluster 1 genes should represent genes directly repressed by Bcl-6 binding (Bcl6-r), while Cluster 4 genes should largely represent genes not directly bound by Bcl-6 (Bcl6-rr). To test this, we used the gene set of BCL-6-bound genes in human GC-T_FH cells identified by BCL-6 ChIP-Seq²⁴ and performed GSEA against Clusters 1 & 4 (see Supplementary Note). BCL-6–bound genes were highly enriched in Cluster 1 (Fig. 3g). In contrast, BCL-6-bound genes were not enriched in Cluster 4 (Fig.3g). Similarly, BCL-6-bound genes also were not enriched in Cluster 2 & 3 (Extended Data Fig.3f). Our model predicted that Cluster 2 & 3 would contain genes regulated by Blimp-1 (Blimp1-rr and Blimp1-r), with Cluster 3 genes directly targeted by Blimp-1 (Fig.3e). GSEA using Blimp-1-bound gene sets showed that Cluster 3 genes were highly enriched for Blimp-1-bound genes (Fig.3h and Extended Data Fig.3g), consistent with the model proposed. While the analyses do not exclude the possibility of some activity of Bcl-6 as an activator (see Supplementary Note), taken together, the data support the proposed model that Bcl-6 primarily acts as a repressor in regulating T_FH biology.

Bcl-6, Blimp-1, and Id2 relationships regulate Cxcr5

Among genes directly repressed by Bcl-6, TFs were of particular interest. Id2 was identified in the clustering analysis as a Bcl6-r TF (Fig.4a). We previously demonstrated that Id2 is an important regulator of T_FH differentiation²². Bcl-6 directly represses Id2, and Id2 inhibits CXCR5 expression via complexing with E proteins^22,23. GSEA showed that E2A-bound genes were enriched in Cluster 4 (Fig.4b). Therefore, a minimalist T_FH differentiation model would be that Bcl-6 initiates and controls T_FH biology primarily via repression of two inhibitory TFs: Blimp-1 and Id2 (Fig.4c). To test the model, we generated Bcl6^f/fPrdm1^f/fId2^f/fCre^CD4 SMARTA mice. Bcl6^f/fPrdm1^f/fId2^f/fCre^CD4 CD4⁺ T cells exhibited substantial increases in CXCR5 expression in comparison to Bcl6^f/fPrdm1^f/fCre^CD4 CD4⁺ T cells (Fig.4d–e). However, GC-T_FH differentiation by Bcl6^f/fPrdm1^f/fId2^f/fCre^CD4 CD4⁺ T cells remained extremely defective (Fig.4f–g and Extended Data Fig. 4a–g). These data indicated that Bcl-6 likely represses multiple TFs in addition to Blimp-1 and Id2 to control T_FH biology.

Figure 4. Bcl-6 drives CXCR5 expression via repression of Id2-E2A pathway.

a, Gene expression of Id2 from RNA-seq data of LCMV_Arm infected mice or KLH-gp₆₁ immunized mice. Each data point was collected from three (LCMV_Arm) or four (KLH-gp₆₁) independent experiments.

b, GSEA of E2A bound genes from thymocytes^22,55 in comparison to Cluster 4 genes between Bcl6^f/fPrdm1^f/f T_FH-like cells and Prdm1^f/f T_FH cells.

c, A hypothetical model of Bcl-6 regulation of T_FH genes primarily via inhibition of Id2 and Prdm1/Blimp-1.

d, Schematic of the SMARTA cell transfer system used for KLH-gp₆₁ immunization. Wild-type, Bcl6^f/fPrdm1^f/f, or Bcl6^f/fPrdm1^f/fId2^f/f Cre^CD4 SMARTA CD4⁺ T were transferred to C57BL/6 host mice, followed by immunization of the host mice with KLH-gp₆₁ in alum + cGAMP, and analyzed 8 days later. See Fig. 4e,f and Extended Data Fig.4a.

e,f, Representative flow cytometry of CXCR5^hi T_FH, CXCR5^hiPD-1^hi and CXCR5^hiPSGL1^lo GC-T_FH SMARTA cell subsets from dLNs of KLH-gp₆₁ immunized mice in Fig.4d. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

g, Quantification of CXCR5^hiSLAM^lo T_FH and CXCR5^hiPSGL1^lo GC-T_FH cells, gated on SMARTA cells from spleens of LCMV_Arm infected mice. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test. See Extended Data Fig.4b–g.

Identification of Bcl-6 target TF candidates

Which TFs in addition to Blimp-1 and Id2 are key repressors downstream of Bcl-6 that control genes upregulated in T_FH cells? The simple circuitry repressor-of-repressors model predicts that such TFs should be present in Cluster 1 (Fig.3e). 307 TFs were identified in Cluster 1 (Fig.5a). We developed an analytical approach to identify candidate TFs by integrated analysis of the composite RNA-seq data with both BCL-6 ChIP-seq data from human tonsillar GC-T_FH cells²⁴ and ATAC-seq of T_FH and non-T_FH cells from multiple genetically modified mice. Among the Cluster 1 TFs, 119 TFs represented BCL-6–bound gene loci in human GC-T_FH cells, confirming that these 119 TFs are direct targets of BCL-6 (Fig.5b).

Figure 5. Integrated analysis of multiple genetic backgrounds and data types.

a, Volcano plot of gene expression changes of Cluster 1 TFs between Bcl6^f/fPrdm1^f/f T_FH-like cells and Prdm1^f/f T_FH cells (LCMV_Arm infection) or between Bcl6^f/fPrdm1^f/f non-T_FH cells and Prdm1^f/f T_FH cells (KLH-gp₆₁ immunization) (1.4-fold cut off, Adj. P<0.05); 76 TFs from LCMV_Arm infection; 287 TFs from KLH-gp₆₁ immunization; combined 307 TF). Adj. P-values for multiple test correction were determined using Benjamini-Hochberg algorithm. Each data point was collected from independent experiments. Select genes of interest are labeled.

b, Schematic of the integrated analytical approach. The composite RNA-seq data with both BCL-6 ChIP-seq data from human tonsillar GC-T_FH cells and ATAC-seq of T_FH and T_H1 cells from wild-type, Bcl6^f/f, Prdm1^f/f and Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA cells. Among the Cluster 1 TFs in Fig.5a, 119 TFs represented BCL-6-bound gene loci in human GC-T_FH cells. ATAC-seq and TF motifs scanning filters top Bcl6-r TF candidates.

c, Schematic of the experimental plan to generate ATAC-seq data. T_H1 (CXCR5^loSLAM^hi) populations from wild-type and Bcl6^f/f Cre^CD4 SMARTA cells, T_FH (CXCR5^hiSLAM^lo) populations from wild-type and Prdm1^f/f Cre^CD4 SMARTA cells, T_H1-like (CXCR5^loSLAM^int) and T_FH-like (CXCR5⁺SLAM^int) populations from Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA cells were sorted from spleens of C57BL/6 host mice given wild-type, Bcl6^f/f, Prdm1^f/f, and Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA cells, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days later. Naive SMARTA cells were isolated as CD44^loCD62L^hiCD45.1⁺ from uninfected mice. Representative flow cytometry of T_FH, T_H1, T_FH-like, and T_H1-like subsets from three independent experiments.

d, Genome-browser tracks depict ATAC-seq chromatin accessibility and TF occupancy. Peak calls indicated below each track. A Bcl-6 liftover peak from human to mouse reference genome is indicated. * and ** indicate DEseq2 raw P-val ≤ 0.05 and ≤ 0.01, respectively, in comparison between wild-type T_FH and T_H1. Gene expression from RNA-seq data of LCMV_Arm infected mice is graphed. Each data point was collected from three independent experiments. DKO, Bcl6^f/fPrdm1^f/f Cre^CD4.

e, ChIP-qPCR analysis of Bcl-6 at Selplg E1 or negative control region among chromatin prepared from CXCR5^hi T_FH cells from spleens of C57BL/6 host mice given SMARTA CD4⁺ T cells transduced with myctagN-Bcl6-RV, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days after infection. ChIP was performed using an anti-myc IgG or control IgG. Three independent experiments were performed. Each data point is from an independent experiment (n=3) and presented as a percent of input. Data are mean ± s.e.m., unpaired two-tailed Student’s t-test.

The most functionally important Bcl6-r TFs would repress Bcl6-rr genes, based on the repressor-of-repressors Bcl-6 model (Fig.3a). We reasoned that candidate Bcl6-r TFs could be functionally connected to transcriptional regulation of genes upregulated in wild-type T_FH cells in a Bcl-6-dependent manner by testing for enrichment of candidate TF DNA binding motifs in differentially accessible chromatin regulatory regions of T_FH-associated genes, particularly between Prdm1^f/f T_FH and Bcl6^f/fPrdm1^f/f T_FH-like cells, as those chromatin changes would be dependent on Bcl-6 expression. We therefore conducted ATAC-seq of T_FH or T_H1 populations of Bcl6^f/fPrdm1^f/fCre^CD4, Bcl6^f/fCre^CD4, Prdm1^f/fCre^CD4, and wild-type SMARTA cells in the context of acute LCMV_Arm infection (Fig.5c). Several genes of interest were examined as a first test. An E2A binding motif was identified in a differential T_FH ATAC-seq peak in a downstream enhancer of Cxcr5, consistent with Bcl-6 control of CXCR5 expression via inhibition of Id2 (Extended Data Fig.5a–b). Selplg, Tbx21, and Gata3 are known Bcl-6-bound genes²⁴. Substantial changes in chromatin accessibility were observed for each of these genes. BCL-6 binding sites of these genes in human GC-T_FH cells were conserved in mouse by syntenic analysis (Fig.5d and Extended Data Fig.5c,d). A differential T_FH ATAC-seq peak in an Selplg intron overlapped with a large human GC-T_FH BCL-6 ChIP-Seq peak²⁴ centered on a BCL-6 DNA binding motif (SELPLG E1) (Fig.5d and Extended Data Fig.5d). To evaluate whether SELPLG E1 BCL-6 binding is conserved in mouse T_FH, we performed Bcl-6 ChIP with Myc-tagged Bcl-6 expressing (myctagN-Bcl6-RV⁺) Bcl6^f/fCre^CD4 T_FH cells. Human BCL-6 bound SELPLG E1 in GC-T_FH was indeed a site bound by Bcl-6 in mouse T_FH cells (Fig.5e). Together, these results indicated the ATAC-seq data were high quality and could be used for broader TF motif scanning.

We then applied the differential chromatin accessibility plus TF motif analysis to all genes, across all ATAC-seq data sets, to identify TF motifs that were enriched within regions that underwent differential chromatin remodeling. Chromatin accessibility in the seven cell populations was distinct (Fig.6a). We scanned for 566 known TF-binding motifs within each differentially accessible ATAC peak in the genome. Motifs recognized by TFs Runx, Ets, T-box and Klf families were most highly enriched in chromatin regions with increased accessibility in Bcl6^f/fPrdm1^f/f T_FH-like cells compared against Prdm1^f/f T_FH cells (each p<1.0 × 10⁻¹⁶) and wild-type T_FH cells (Fig.6b). In contrast, in the same cell type comparisons, these TF motifs were depleted from chromatin regions with reduced accessibility (Extended Data Fig.6a). Changes in abundance of Runx DNA binding motifs were particularly significant (p<9 × 10⁻⁷⁵). Runx and Tbet footprints were present (Fig.6c and Extended Data Fig.6b), indicating Bcl-6 expression prevents Runx and T-box family TFs from binding these sites, most likely by direct Bcl-6 transcriptional repression of Runx and T-box family genes. Runx DNA binding sites were observed in enhancer regions of T_FH-associated genes including Pdcd1 and Icos (Fig.6d and Extended Data Fig.6c). All three Runx TFs are expressed in CD4⁺ T cells (Fig.6e), and each Runx TF is known to be competent for binding consensus Runx motifs³². Runx2 and Runx3 were grouped in Cluster 1 in the gene expression MAP-DP clustering analysis (Fig. 3d), making them the more likely Runx candidates for repression by Bcl-6. BCL-6 bound robustly to RUNX2 and RUNX3²⁴ enhancers in human GC-T_FH cells (Fig.6f). Bcl-6 bound Runx2 E1, Runx2 E2, Runx2 E3, and Runx3 E1 in mouse T_FH cells, confirming conservation of Bcl-6 binding to Runx2 and Runx3 loci (Fig.6g).

Figure 6. Identification of candidate TFs.

a, tSNE analysis of ATAC-seq chromatin accessibility.

b, Heatmap plots representing the frequencies of the most enriched TF motifs in regions of increased accessibility (relatively more open in first group than second group, DEseq2 raw P-val < 0.05). Scale, motif frequencies (%).

c, TF footprints derived from ATAC-seq reads over representative TF motifs within accessible ATAC-seq regions.

d, Genome-browser tracks depict ATAC-seq chromatin accessibility and TF occupancy. Peak calls indicated below each track. ** indicates DEseq2 raw p-val ≤ 0.01 in comparison between Bcl6^f/fPrdm1^f/f Cre^CD4 T_FH-like and Prdm1^f/f Cre^CD4 T_FH.

e, Gene expression of Runx1, Runx2, and Runx3 from RNA-seq data of LCMV_Arm infected mice or KLH-gp₆₁ immunized mice. Each data point was collected from three (LCMV_Arm) or four (KLH-gp₆₁) independent experiments.

f, Genome-browser tracks depict BCL-6 ChIP-Seq peaks at RUNX2, RUNX3, and KLF2 loci. Peak calls indicated below each track.

g,i. ChIP-qPCR analysis of Bcl-6 at Runx2 E1, E2 and E3, Runx3 E1, or Klf2 P1 and DE among chromatin prepared from CXCR5^hi T_FH cells as shown in Fig.5e. Three independent experiments were performed. Each data point is from an independent experiment (n=3) and presented as a percent of input. Data are mean ± s.e.m., unpaired two-tailed Student’s t-test.

h, Gene expression of Klf2 from RNA-seq data of LCMV_Arm infected mice or KLH-gp₆₁ immunized mice. Each data point was collected from three (LCMV_Arm) or four (KLH-gp₆₁) independent experiments.

Enrichment of Klf DNA binding motifs in Bcl6^f/fPrdm1^f/f T_FH-like cells was examined further. Several Klf-family TFs are expressed in CD4⁺ T cells, with Klf2 representing the dominant member (Extended Data Fig.6d). Klf2 exhibited a Cluster 1 type gene expression pattern in T_FH cells in the context of KLH-gp₆₁ immunization, while it exhibited a more complex gene expression pattern in acute LCMV infection (Fig.6h). The TF footprints for several Klf motifs were enriched in wild-type T_H1 cells over wild-type T_FH cells (Extended Data Fig.6b,e). Bcl-6 bound the Klf2 promoter (P1) and a putative distal enhancer (DE), confirming that Bcl-6 binding sites at Klf2 loci are conserved between humans and mice (Fig.6f,i). Klf binding sites were observed in open chromatin regions of multiple signature GC-T_FH genes (Fig.6d and Extended Data Fig.6c). These results identified Klf2 as a Bcl6-r TF candidate. GATA-3 is constitutively expressed in CD4⁺ T cells. Gata3 was defined as a Cluster 1 gene, with reduction of GATA-3 expression in T_FH cells compared to T_H1 cells (Fig.3d and Extended Data Fig.5c). Taken together, integrated bioinformatic analyses revealed Runx2, Runx3, GATA-3, and Klf2 as strong potential Bcl6-r TF candidates that repress T_FH genes.

Bcl-6 repressor-of-repressors circuits

To test the in vivo roles of candidate Bcl6-r TFs, we optimized a system for direct transfection of SMARTA cells with target gene CRISPR RNA (crRNA) and Cas9 ribonucleoprotein (RNP) complexes to disrupt target genes (Extended Data Fig.7a–b). RNP⁺ SMARTA cells were adoptively transferred into host mice subsequently infected with LCMV_Arm. Bcl6 and Prdm1 were first tested as positive control genes (Fig.7a–b and Extended Data Fig.7c). crRNA-mediated deletion of TFs was efficient and was a suitable experimental system for exploring the genetics of T_FH biology in vivo. We then disrupted Gata3, Runx2, Runx3, and Klf2 as candidate Bcl6-r T_FH repressors, with crRNA RNPs in Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells, and examined their differentiation in the context of acute LCMV infection. crCd8⁺ wild-type SMARTA cells were used as controls. crGata3⁺Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells exhibited increased CXCR5 and reduced PSGL1 expression in comparison to crCd8⁺Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells (Fig.7c and Extended Data Fig.7). These results suggest that GATA-3 is a Bcl6-r TF that is a repressor of CXCR5 and a positive regulator of PSGL1.

Figure 7. Identification of Runx2, Runx3, and GATA-3 as repressors of T_FH genes, acting downstream of Bcl-6.

a, Schematic of the CRISPR/Cas9-mediated gene knockdown of SMARTA cell system used for testing Bcl6 and Prdm1 in LCMV_Arm infection. SMARTA CD4⁺ T cells transfected with crCd8, crBcl6, or crPrdm1 were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 6 days later.

b, Representative flow cytometry of GC-T_FH SMARTA cell subsets from spleens of LCMV_Arm infected mice in a. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

c, Quantification of results from crCd8⁺ and crGata3⁺ SMARTA cells from spleens of LCMV_Arm infected mice. Three independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test. See Extended Data Fig.7h,i,l,m.

d, Schematic of the CRISPR/Cas9-mediated gene knockdown of SMARTA cell system used for testing Runx2 and Runx3 in LCMV_Arm infection. wild-type or Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4⁺ T cells transfected with crCd8, crRunx2, or crRunx3 were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 6–7 days later. See e–h and Extended Data Fig.7j–k.

e,g, Representative flow cytometry of T_FH SMARTA cell subsets from spleens of LCMV_Arm infected mice in d. Three independent experiments were performed; each dot represents one mouse (e, n = 5;g, n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

f,h, Quantification of GC-T_FH core signature markers, gated on SMARTA cells in d. CD44^lo naive CD4⁺ T cells were used as a negative control.

i,j, Representative flow cytometry of T_H1, T_FH and GC-T_FH RV⁺ SMARTA cell subsets from spleens of C57BL/6 host mice given SMARTA CD4⁺ T cells transduced with pMIG (GFP-RV⁺), pMIG-Runx3myc (Runx3-RV⁺ [Low]), or pMIG-Runx2myc (Runx2-RV⁺ [Med]), followed by infection of the host mice with LCMV_Arm, and analyzed 7 days after infection. Two independent experiments were performed; each dot represents one mouse (n = 5). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

Roles of Runx family TFs in T_FH biology are largely unknown. We first examined the impact of Runx2 disruption on T_FH gene regulation (Fig.7d–f and Extended Data Fig.7). We observed significantly greater frequencies of CXCR5⁺ T_FH-like cells among crRunx2⁺Bcl6^f/fPrdm1^f/fCre^CD4 CD4⁺ T cells compared to crCd8⁺ control cells. Expression of ICOS and CD200 was upregulated in the absence of Runx2 (Fig.7e–f). Runx3 disruption also increased the development of CXCR5⁺ T_FH-like cells on the Bcl6^f/fPrdm1^f/fCre^CD4 background, as well as expression of ICOS and CD200 (Fig.7g–h), paralleling the T_FH gene regulation observed by Runx2.

We next investigated whether Runx2 and Runx3 act predominantly upstream or downstream of Bcl-6 in T_FH differentiation. The proposed Bcl-6 repressor-of-repressors model predicted that Runx2 and Runx3 would act downstream of Bcl-6. To test this, we transduced wild-type SMARTA cells with RVs expressing GFP alone (GFP-RV⁺), Runx3 (Runx3-RV⁺), or Runx2 (Runx2-RV⁺), transferred the cells into B6 mice, and analyzed T_FH differentiation 7 days after acute LCMV_Arm infection (Extended Data Fig.8a). Enforced Runx2 expression resulted in reduced T_FH and GC-T_FH differentiation relative to control cells. Expression of ICOS and CD200 was also reduced in Runx2-RV⁺ T_FH cells, consistent with the Runx2 gene disruption data. Constitutive Runx3 expression caused more severe disruption of T_FH differentiation than did Runx2 (Fig.7i, Extended Data Fig.8). Most notably, Bcl-6 expression was not affected by enforced Runx2 or Runx3 expression, indicating that Bcl-6 is indeed upstream of Runx2 and Runx3 in T_FH cells (Fig.7j). Disruption of Runx2 expression results in a gain of T_FH gene expression similar to that of Runx3 gene disruption (Fig.7e–h), indicating that both Runx2 and Runx3 are relevant targets of Bcl-6 in vivo for T_FH development.

Klf2 has been connected to T_FH differentiation in both mice and humans, downstream of ICOS signaling^13,14. Klf2 represses CXCR5 expression and Klf2 binds the Prdm1 locus, but different models were proposed for how Klf2 influences T_FH differentiation^13,14. Thus, we investigated the effect of Klf2 on T_FH gene expression using the crRNA RNP SMARTA system to test the model that Klf2 may act downstream of Bcl-6 as a Bcl6-r TF in a repressor-of-repressors circuit (Fig.8 and Extended Data Fig.9a–d). In the Bcl6^f/fPrdm1^f/fCre^CD4 background, expression of PD-1, ICOS, CD200, and IL-6Rα were all significantly upregulated in crKlf2⁺ CD4⁺ T cells versus crCd8⁺ (Fig.8a–c). More surprisingly, expression of the T_FH cytokine IL-21 was significantly increased (Fig.8d and Extended Data Fig.9b). Given that result, we examined expression of IL-4, the other major cytokine expressed by T_FH cells. IL-4 expression was substantially increased in antigen-stimulated crKlf2⁺ versus crCd8⁺ CD4⁺ T cells in KLH-gp₆₁ immunization (Fig.8e and Extended Data Fig.9d). This occurred even though GATA-3 expression was not changed in the absence of Klf2^13,14. Disruption of Klf2 did not affect the expression of Maf, a TF known to have a role in Il21 and Il4 expression^33,34 (Fig.8f and Extended Data Fig.9c).

Figure 8. Identification of Klf2 as a repressor acting downstream of Bcl-6 regulating major T_FH genes.

a, Schematic of the CRISPR/Cas9-mediated gene knockdown of SMARTA cell system used for testing Klf2 in LCMV_Arm infection. wild-type or Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4⁺ T cells transfected with crCd8 or crKlf2 were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 6–7 days later. See Fig.8b–c and Extended Data Fig.9a,b.

b, Representative flow cytometry of T_FH and GC-T_FH cells, gated on SMARTA cells from spleens of LCMV_Arm infected mice in a. Three independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

c, Quantification of GC-T_FH core signature markers, gated on SMARTA cells from spleens of LCMV_Arm infected mice in a. CD44^lo naive CD4⁺ T cells were used as a negative control.

d,e, wild-type or Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4⁺ T cells transfected with crCd8 or crKlf2 were transferred to C57BL/6 host mice, followed by immunization of the host mice with KLH-gp₆₁ and alum + cGAMP, and analyzed 8 days later. Representative flow cytometry and quantification of gp₆₆-restimulated IL-21⁺ and IL-4⁺ SMARTA cells from dLNs of KLH-gp₆₁ immunized mice. Two independent experiments were performed; each dot represents one mouse (n = 5). Data are mean ± s.d., unpaired two-tailed Student’s t-test. See Extended Data Fig.9d for experimental design.

f, Quantification of expression of Tcf-1, GATA-3, and Maf, gated on SMARTA cells from dLNs of KLH-gp₆₁ immunized mice in Extended Data Fig.9d.

g, A circuitry model of the regulation of T_FH genes upregulated by Bcl-6 through repression of repressor TFs.

These observations indicated that Klf2 is a negative regulator of the expression of PD-1, ICOS, CD200, IL-6Rα, IL-21, and IL-4. The regulation of IL-6Rα and ICOS by Klf2 was reminiscent of the role of Tcf-1 (the product of Tcf7), but opposite^19–21. Multiple Klf motifs were observed in open chromatin regions of the Tcf7 gene locus in T_FH cells, suggesting that Tcf7 may be a Klf2-targeted TF (Extended Data Fig.9e). Therefore, we examined Tcf-1 expression in crKlf2⁺ Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells. Expression of Tcf-1 protein was substantially increased in the absence of Klf2 (Fig.8f and Extended Data Fig.9c). In mouse T_FH cells, Tcf-1 binding was observed at the promoter and enhancers of multiple GC-T_FH signature genes including Pdcd1, Il6ra (Extended Data Fig.9f), Icos and Il21, all of which are upregulated in crKlf2⁺ Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA cells. These data support a Bcl-6 −| Klf2 −| Tcf7 pathway for upregulation of T_FH genes (Fig.8g). Klf and Tcf-1 binding motifs were observed concomitantly in open chromatin of multiple Bcl6-rr genes including Pdcd1, Cd200, and Il21 (Fig.6d and Extended Data Fig.6c,9g), suggesting that Klf2 represses Bcl6-rr genes through a combinatorial mechanism via direct binding and by repression of Tcf7. Taken together, we conclude that Bcl-6 is a nexus for control of positive T_FH gene expression by repression of multiple repressors (Fig.8g).

DISCUSSION

T_FH differentiation is a multistage, multifactorial process¹. How Bcl-6, the lineage defining TF of T_FH cells, accomplishes control of T_FH differentiation and function has remained unclear, at least in part because of the complexity of the biology, the antagonistic relationship between Bcl-6 and Blimp-1, and intrinsic challenges of studying repressors. The present study advances our mechanistic understanding of how Bcl-6 controls T_FH differentiation and function. We resolved several obstacles through a series of logical approaches, integrating analyses of multiple data sources from multiple genotypes and genetically modified cells. Our observations excluded the two simplest models of Bcl-6 regulation of T_FH differentiation via repression of Blimp-1 alone or Blimp-1 and Id2. Bioinformatic analyses identified numerous candidate Bcl6-r TFs, and experiments demonstrated that Runx2, Runx3, GATA-3, and Klf2 are Bcl-6 target TFs that regulate T_FH differentiation and function. We conclude that Bcl-6 is a nexus for control of positive T_FH gene expression by repression of multiple repressors.

Bcl-6 is an obligate repressor in B cells^30,35,36. We find that Bcl-6 regulates its own expression by a negative autoregulatory loop in T_FH cells. Based on these and other data, we proposed that Bcl-6 drives upregulation of canonical genes of T_FH differentiation and function via repressor-of-repressors mechanisms. In addition, Bcl-6 represses alternative, non-T_FH, cell fates. Bcl-6 clearly inhibits non-T_FH differentiation fates via inhibition of Prdm1^2,3,9. BCL-6 can also block T_H1/T_H2/T_H17 differentiation by repression of lineage-defining TFs TBX21, GATA3, and RORA^4–6,24,37, as well as by repressing genes central to those cells types (e.g. Il17a, Il17f, Ifng, Il2ra, and Ifngr1^4,5,24). In contrast, testing the repressor-of-repressors model of Bcl-6 function required identification of Bcl6-r TFs that repress positive features of T_FH biology, downstream of Bcl-6. Our analytical approach to identify candidate Bcl6-r repressor TFs integrated RNA-seq, ChIP-Seq, and ATAC-seq. Id2 is one Bcl6-r TF clearly important for regulation of Cxcr5. We further identified Runx2, Runx3, GATA-3, and Klf2 as Bcl6-r TFs that repress important T_FH genes including PD-1, ICOS, CD200, IL-6Rα, IL-21, and IL-4. These observations demonstrate there are multiple repressors downstream of Bcl-6 that control T_FH genes. While the results do not exclude additional TFs, or additional mechanisms of action including potential direct activator activity of Bcl-6 (see Supplementary Note), the overall structure of the repressor-of-repressors Bcl-6 gene regulatory network can explain why T_FH differentiation is fully dependent on Bcl-6. In the absence of Bcl-6 no TF appears to substitute.

One might consider whether Bcl-6 acting as an obligate repressor in CD4⁺ T cells is a rare case for a lineage defining TF. Bcl-6 was, of course, originally identified as the master regulator of B_GC differentiation, and it has been well characterized as an obligate repressor in B cells^30,35,36, though efforts have predominantly focused on genes downregulated in B_GC cells. Foxp3 is a second lineage-defining TF that mainly acts as a repressor³⁸. RORγT appears to act substantially through repressor activity in T_H17 cells, in concert with Maf³⁹. The primary actions of T-bet in CD4⁺ T cells may also be predominantly repressive⁴⁰. As such, it has been proposed that a general primary function of lineage defining TFs may be to limit, by direct repression or other mechanisms, the number of genes that are induced broadly by TCR and cytokine signaling⁴¹. Bcl-6 is a clear example of this direct repression model, but the work here adds to that by demonstrating how genes positively associated with a cell type can be upregulated downstream of a lineage-defining TF via repressor-of-repressor mechanisms. Our results thus establish an overall structure of T_FH differentiation and gene regulation in a parsimonious model of Bcl-6 serving as the apex of a repressor-of-repressors network. This may also provide future insights into the biology of dysregulated T_FH or T_FH-like cells present in a range of biomedically relevant diseases such as atherosclerosis and autoantibody-mediated autoimmune diseases¹.

METHODS

Mice.

C57BL/6J (B6) and Cre^CD4 and CD45.1⁺ mice were obtained from the Jackson Laboratory. Mouse strains described below were bred and housed in specific pathogen–free conditions in accordance with the Institutional Animal Care and Use Guidelines of the La Jolla Institute. SMARTA mice (T cell antigen receptor (TCR) transgenic for I-A^b-restricted lymphocytic choriomeningitis virus (LCMV) glycoprotein (gp) 66–77 peptide)⁴², Bcl6^f/f43, Prdm1^f/f44, and CD45.1⁺ congenic mice were on a full B6 background. Bcl6^f/f or Prdm1^f/f mice were crossed to the SMARTA, Cre^CD4, and CD45.1⁺ strains to generate Bcl6^f/fCre^CD4 CD45.1⁺ SMARTA and Prdm1^f/fCre^CD4 CD45.1⁺ SMARTA mice. Bcl6^f/fPrdm1^f/fCre^CD4 CD45.1⁺ SMARTA mice were generated by crossing Bcl6^f/fCre^CD4 CD45.1⁺ SMARTA and Prdm1^f/fCre^CD4 CD45.1⁺ SMARTA strains. Blimp-1-YFP⁴⁵ or ΔBPS1 (described below) mice were crossed to the CD45.1⁺ SMARTA strain to generate Blimp-1-YFP CD45.1⁺ SMARTA or ΔBPS1 CD45.1⁺ SMARTA mice. Both male and female mice (6–15 weeks of age) were used throughout the study, with sex- and age-matched T cell donors and recipients. All animal experiments were performed under protocols approved by the Institutional Animal Use and Care Committees of the La Jolla Institute for Immunology.

Adoptive cell transfer, infection, and immunization.

Adoptive transfer of congenically marked cells (CD45.1⁺) into recipient mice (CD45.2⁺) was performed by intravenous injection via the retroorbital sinus. For LCMV Armstrong (LCMV_Arm) infection, 10 × 10³ naïve, transduced RV⁺, or crRNA⁺ CD4⁺ T cells were transferred into each mouse. Recipient mice were injected intraperitoneally with 2.0 × 10⁵ plaque-forming units (pfu) of LCMV_Arm in plain DMEM. For protein immunization, 50 × 10³ naïve or crRNA⁺ CD4⁺ T cells were transferred into each mouse. A total of 10 μg of keyhole limpet hemocyanin (KLH) conjugated with LCMV gp61–80 peptide (KLH-gp₆₁) was prepared in alum (Alhydrogel) only, alum + LPS (1 μg), alum + Poly (I:C) (10 μg), or alum + cyclic [G(3′,5′)pA(3′,5′)p] (3′,3′-cGAMP; 10 μg, Invivogen) adjuvants in a total volume of 20 μL and injected into each footpad of recipient mice. Alum + 3′,3′-cGAMP was chosen as the adjuvant combination because alum alone is a poor inducer of Blimp-1 (Extended Data Fig.1a). Transferred cells were allowed to rest in host mice for 1d (naive cells) or 3–4 d (RV⁺ or crRNA⁺ cells) before infection or immunization.

Plasmids and retroviral transduction.

pMIG (contains an IRES-GFP), pMIG-Bcl6, pMIG-Runx2myc, pMIG-Runx3myc retroviral (RV) plasmids, and pQCXIP (contains Thy1.1 and pGK-GFP) self-inactivating (SIN) RV plasmid were described previously^28,46,47. pMIG-myctagN-Bcl6 plasmid was generated by insertion of sequences encoding the Myc-tag to those encoding the N-terminus of Bcl-6 with short-linker sequences (5′-GATCTGAATTCGGAATCTACC-3′). pQCXIP plasmid was further modified with deletion of the IRES-Puro^R cassette and the CCAAT box and the TATA box of 3′-UTR to reduce background reporter expression (pQdT, Extended Data Fig. 2b). Constructs containing either wild-type or an 8-nt deletion that removes the Bcl-6 binding motif (+18; TCTAGGAA) in the proximal Bcl6 promoter region (−709 to +272 from transcription start site) were cloned into the pQdT plasmid at the upstream of the Thy1.1 reporter to generate wild-type Thy1.1-RV or ΔBPS1 Thy1.1-RV, respectively (Fig.2a). Virions were produced by transfection of the Plat-E cell line. Culture supernatants were collected 24 and 48 h after transfection, filtered through a 0.45 μm syringe filter and stored at 4 °C until transduction. CD4⁺ T cells were isolated from whole splenocytes by negative selection (Stemcell Technologies) and resuspended in R10 (RPMI1640 + 10% FBS, supplemented with 2 mM Glutamax, 100 U/ml Penicillin/Streptomycin, and non-essential amino acids (Gibco)) with 2 ng/ml recombinant human IL-7 (Peprotech) and 50 μM β-mercaptoethanol (2-ME). 0.5 × 10⁶ cells were stimulated in 24-well plates pre-coated with 8 μg/ml anti-CD3 (17A2; BioXcell) and anti-CD28 (37.51; BioXcell). At 40 and 48 h after stimulation, cells were transduced by adding RV supernatants supplemented with 50 μM 2-ME and 8 μg/ml polybrene (Millipore), followed by centrifugation for 90 min at 524 x g at 37 °C. Following each transduction, the RV-containing medium was replaced with R10 + 50 μM 2-ME + 10 ng/ml human IL-2. After 72 h of in vitro stimulation, CD4⁺ T cells were transferred into six-well plates in R10 + 50 μM 2-ME + 10 ng/ml human IL-2, followed by incubation for 2 days. One day before transfer, the culture medium was replaced with R10 + 50 μM 2-ME + 2 ng/ml human IL-7. Transduced cells were sorted based on GFP expression (FACSAria; BD Biosciences).

Enforced Runx3 (Runx3-RV⁺ [High]) or Runx2 (Runx2-RV⁺ [High]) expression disrupted T_FH differentiation relative to control cells (GFP-RV⁺) (Extended Data Fig.8b,d). Since overexpression of Runx3 at much higher than the physiological level resulted in a negative effect on CD4⁺ T cell accumulation (Extended Data Fig.8b,c), we made methodological improvements by performing similar experiments with lower constitutive Runx3 expression by sorting the bottom 10% GFP⁺ cells (Runx3-RV⁺ [Low]) instead of total GFP⁺ SMARTA (Runx3-RV⁺ [high]) for adoptive transfer (Extended Data Fig.8e–f). With lower enforced expression of Runx3, SMARTA cell proliferation was enhanced to levels similar to that of Runx2-RV⁺ [Med] cells (Extended Data Fig.8g). Nevertheless, Runx3-RV⁺ [Low] SMARTA still showed a stronger impaired T_FH and GC-T_FH development compared to Runx2-RV⁺ [Med] SMARTA (Fig. 7i and Extended Data Fig.8h).

Generation of ΔBPS1 mice using CRISPR-Cas9 gene editing.

The in vitro molecular and cellular biology was performed by Ingenious Targeting Laboratory, Inc. Guide RNAs were selected using optimized CRISPR design by CHOP-CHOP (https://chopchop.cbu.uib.no/)⁴⁸. Guide RNAs (gRNA1, 5′-caccgTCTAGGAAAGGCCGGACACC-3′ and 3′-cAGATCCTTTCCGGCCTGTGGcaaa-5′; gRNA2, 5′-caccgTGGTGATGCAAGAAGTTTCT-3′ and 3′-cACCACTACGTTCTTCAAAGAcaaa-5′) were cloned into px459-Cas9-puromycin plasmid. Lipofectamine transfection of each gRNA with a control gRNA was performed on Neuro2A cells (duplicate set per gRNA). The control gRNA was designed upstream of the gRNA. After 24 h, cells were selected for puromycin resistance for 3–5 days and then lysed. PCR analysis was performed to verify the cleavage efficiency of each gRNA. The cleavage efficiency of gRNA was 10–15%. An injection mix of 30 ng/μl Cas9 protein, 0.6 μM gRNA, and 20 ng/μl oligonucleotide (template DNA for the repair to delete the 8-nt Bcl-6 recognition motif) was injected into 150–250 fertilized eggs from C57BL/6J mice by the UCSD Stem Cell Core. These eggs were implanted into C57BL/6J surrogate mothers, and pups were genotyped by DNA sequencing. DNA sequences were analyzed and diagrammed using MacVector (Extended Data Fig.2e). ΔBPS1 mice were healthy, and immune cell development appeared grossly normal.

CRISPR/Cas9-mediated gene deletion of murine CD4⁺ T cells.

High-ranked guide sequenes with the highest on-target and off-target scores were selected by CHOP-CHOP. crRNA and ATTO-550 conjugated tracrRNA were purchased from Integrated DNA Technologies, Inc. Purified S. pyogenes Cas9-NLS protein was purchased from QB3 Macrolab of University of California, Berkeley. crRNA and tracrRNA were duplexed by heating 95 °C for 5 min. Ribonucleoprotein (RNP) complexes were generated by mixing crRNA-tracrRNA duplexes (240 pmol) and Cas9-NLS protein (80 pmol) for 10 min at 24–26 °C. Isolated CD4⁺ T cells were stimulated in 24-well plates pre-coated with 8 μg/ml anti-CD3 (17A2) and anti-CD28 (37.51) for 2 days. The cells were then transfected with a RNP mixture by electroporation using MaxCyte ATX with Expanded T cell-4 protocol (MaxCyte, Inc, MD). The transfected cells were cultured in R10 + 50 μM 2-ME + 10 ng/ml human IL-2 without TCR stimulation for 1 day, followed by culture for an additional day in R10 + 50 μM 2-ME + 2 ng/ml human IL-7. Transfection efficiency and cell viability were measured using LSRII or LSR Fortessa. RNP transfection efficiencies were consistently greater than 90%, with high viability (Extended Data Fig.7a,b). crRNA sequences used in the study: crCd8, 5′-GCAGGTTCAGCGACAGAAAG-3′; crBcl6, 5′-TCAAGATGTCCCGACTCCGG-3′; crPrdm1, 5′-TTGGAACTAATGCCGTACGG-3′, crRunx2, 5′-ACCATGGTGCGGTTGTCGTG-3′; crRunx3, 5′-GCTAAGCGCGCAGGCAACCG-3′; crGata3, 5′- TGTACGAATGGCCGAGGCCC-3′; crKlf2, 5′-CTGGCCGCGAAATGAACCCG-3′.

Wild-type or Bcl6^f/fPrdm1^f/fCre^CD4 SMARTA CD4⁺ T cells transfected with crCd8, crGata3, crRunx2, crRunx3, or crKlf2 were transferred to C57BL/6 host mice, followed by infection with LCMV_Arm or immunization with KLH-gp₆₁ of the host mice, and analyzed 6–7 days later. SMARTA cells were sorted by flow cytometry from spleens or lymph nodes of host mice. Gene knockdown efficiencies were measured by mRNA qPCR (for Runx2 using a primer set: fwd, 5′- CACGACAACCGCACCAT-3′ and rev, 5′- CACGGAGCACAGGAAGTT-3′), flow cytometry (for Bcl-6 and GATA-3), or immunoblot analysis (Runx3 and Klf2). shRunx3-RV⁺ (gene knockdown) SMATRA cells and pMIG-Klf2-RV⁺ (enforced expression) SMARTA cells were used as positive controls for immunoblot analysis.

Flow cytometry and cell sorting.

Single-cell suspensions of spleens or draining popliteal lymph nodes were prepared by standard gentle mechanical disruption. Surface staining for flow cytometry was done with monoclonal antibodies to CD4 (RM4–5), CD8 (53–6.7), SLAM (TC15–12F12.2), ICOS (C398.4A), CD200 (OX-90), CD138 (281–2) (from BioLegend), B220 (RA3–6B2), CD45.1 (A20), PD-1 (J43), CD44 (IM7) (from eBioscience), PSGL1 (2PH1), Fas (Jo2), IgD (11–26) (from BD Biosciences), PNA (FL-1071) (Vector Laboratories), and Fixable Viability Dye eFluor780 (eBioscience). Staining was performed for 30 min at 4 °C in PBS supplemented with 0.5% bovine serum albumin (BSA), unless specified otherwise. CXCR5 staining was done using biotinylated anti-CXCR5 (SPRCL5; eBioscience) for 30 min, followed by BV421 or PE.Cy7-labeled streptavidin (BioLegend) at 4 °C in PBS supplemented with 0.5% BSA. Intracellular staining for TFs was performed with monoclonal antibodies to Bcl-6 (K112–91, BD Biosciences), Tcf-1 (C63D9, Cell Signaling), T-bet (4B10), GATA-3 (TWAJ), Maf (Sym0F1) (from eBioscience) using the Foxp3/Transcription Factor Staining Buffer Set (eBioscience). For measurement of cytokines, the cells from spleen or lymph node were cultured in vitro for 5 h in 10 μg/ml gp_66–77 peptide (gp₆₆) and Brefeldin A. Intracellular staining for cytokines was performed with monoclonal antibody to IL-4 (11B11, eBioscience) and recombinant mouse IL-21 receptor Fc (R&D), followed by anti-human IgG (Invitrogen), using the Fixation/Permeabilization buffer kit (BD Biosciences). Stained cells were analyzed using LSRII, LSRFortessa, or FACS Celesta (BD) and FlowJo software v.10.6 (Flowjo, LLC). For cell sorting, CD45.1⁺ SMARTA cells were pre-enriched with PE conjugated anti-mouse CD45.1 Ab (eBioscience) and anti-PE microbead (Miltenyi). All sorting was done on a FACSAria or FACSAria Fusion (BD Biosciences). Cellular data were presented as frequencies of cell populations; absolute numbers of cells were also determined. Frequency and absolute number conclusion were equivalent unless stated otherwise. Spleen and LN size were equivalent between samples unless stated otherwise.

ELISA.

Nunc MaxiSorp plates (Thermo Fisher Scientific) were coated overnight at 4 °C with 1 μg/ml KLH-gp₆₁ (Genscript) in PBS. Plates were blocked with PBS + 0.05% Tween-20 + 0.5% BSA for 90 min at 25 °C. After washing, mouse serum was added in a serial dilution in PBS + 0.05% Tween-20 + 0.5% BSA (PBST-B) and incubated for 90 min. After washing, horseradish peroxidase (HRP)-conjugated goat anti-mouse IgG (Thermofisher Scientific) was added at 1:5,000 in PBST-B for 90 min at 25 °C. Colorimetric detection was performed using a TMB substrate kit (Thermofisher Scientific). Color development was stopped after approximately 5–10 min with 2 N H₂SO₄, and absorbance was measured at 450 nm.

Immunofluorescence staining of lymph node.

Popliteal lymph nodes from mice immunized with KLH-gp₆₁ were snap frozen in OCT medium (Sakura Finetek, USA), and 5–8 μm sections were prepared using a cryostat. Lymph node sections were fixed with acetone-methanol, and stained with antibodies to TCRβ (BV421), IgD (AlexaFlour488), GL7 (PE), and CD45.1 (AlexaFlour647) to reveal the T cell zone, B cell zone, GCs, and SMARTA cells, respectively. Sections were fixed and mounted with ProLong gold antifade reagent (Invitrogen), and imaged by Zeiss AxioScan Z1 Slide Scanner. SMARTA cell localization was analyzed by ImageJ software (v2.0.0-rc-69/1.52p, NIH). The T-B border was defined as ±15 μm region at a boundary line between T cell zone and B cell zone (Extended Data Fig.1h).

RNA-seq.

RNA-seq was performed with a method previously described⁴⁹. Spleens or LNs were isolated and pooled from 4–8 mice per group. 25,000–100,000 CXCR5⁺SLAM^lo/int T_FH, CXCR5^loSLAM^hi T_H1 SMARTA cells (CD45.1⁺CD4⁺CD8^–B220^– singlets), or naïve SMARTA cells (CD4⁺CD8^–B220^–CD44^loCD62L^hi) were sorted using a FACSAria into Trizol LS (Invitrogen). RNA extraction was performed using miRNeasy micro kits (Qiagen) for downstream RNA-Smart-Seq2 input requirements⁴⁹. cDNA was purified using AMPureXP beads (1:1 ratio; Beckman Coulter). One nanogram cDNA was used to prepare a standard Nextera XT sequencing library (Nextera XT DNA sample preparation and index kits; Illumina). Quality control steps were included to determine total RNA quality and quantity, the optimal number of PCR preamplification cycles, and fragment size selection. All samples passed the quality control. Libraries were sequenced using a HiSeq2500 to generate 50-bp single-end reads (TruSeq Rapid Kit; Illumina), generating median of >13 million (LCMV_Arm infection) or >7 million (KLH-gp₆₁ immunization) mapped reads per sample. Three (LCMV_Arm infection) or 4 (KLH-gp₆₁ immunization) biological replicates were generated.

RNA-seq analysis.

The single-end reads that passed Illumina filters were subsequently filtered for reads aligning to tRNA, rRNA, adapter sequences, and spike-in controls. The reads were aligned with the mm10 reference genome using TopHat (v1.4.1., library-type fr-secondstrand-C) and the RefSeq gene annotation downloaded from the University of California – Santa Cruz Genome Bioinformatics site. DUST scores were calculated with PRINSEQ Lite (v0.20.3) and low-complexity reads (DUST > 4) were removed from the BAM files. The alignment results were parsed via the SAMtools to generate SAM files. Read counts to each genomic feature were obtained with the HTSeq-count program (v0.7.1; -m union -s yes -t exon -i gene_id). After removing absent features (zero counts in all samples), the raw counts were imported to R/Bioconductor package DESeq2 (v3.1). P-values for differential expression were calculated using the Wald test and then adjusted for multiple test correction using Benjamini-Hochberg algorithm⁵⁰. We considered genes differentially expressed between two groups of samples when the DESeq2 analysis resulted in an adjusted P-value of <0.05 and the difference in gene expression was 1.4-fold. Genes with raw counts less than 6 as minimum cutoff (an average raw count of Cd8a in all samples; Cd8a is considered to be not expressed, or very weakly expressed in peripheral, splenic CD4⁺ T cells.) were excluded from differential expression analysis. TPMs were calculated from raw count data by dividing by the number of counts by exon length in kilobases (RPK), dividing the total number of counts by 1 million (CM) and then dividing the RPK by CM. Principal Component Analysis (PCA) was performed using the ‘prcomp’ function in R.

We firstly assessed expression of a broad curated set^19,22,25 of T_FH-associated genes across all samples from RNA-seq gene expression profiling. Sequential clustering analyses were performed to analyze the patterns of gene expression changes between the six different T_FH and T_H1 populations. K-means analyses were firstly performed using K=10 for overall gene expression change patterns by ExpressCluster v1.3 software from CBDM Laboratory in Harvard School of Medicine (https://cbdm.hms.harvard.edu/). Total DEG genes (Fc > 1.4; adj. P-val < 0.05; pre-filtered with minimum cutoff) were subjected to the K-means clustering. Four major clusters (n= ~300 or more genes) of gene expression change were apparent. To obtain gene lists associated with those 4 major cluster patterns of gene expression, Maximum a-posteriori Dirichlet process mixtures (MAP-DP) clustering was then performed with predefined cluster centers according to the result from K-means analysis by R-ClustMAPDP⁵¹. We chose MAP-DP instead of K-means clustering because MAP-DP analysis efficiently separates outliers from the data and is statistically rigorous⁵¹. We used the following criteria for the 4 major clusters: (1) wild-type T_FH and Prdm1^f/f T_FH have lower expression than the other populations; (2) wild-type T_H1 and Bcl6^f/f T_H1 have higher expression than the other populations; (3) wild-type T_H1 and Bcl6^f/f T_H1 have lower expression than the other populations; (4) wild-type T_FH and Prdm1^f/f T_FH have higher expression than the other populations (Fig. 3d). Total DEG genes (Fc > 1.4; adj. P-val < 0.05; pre-filtered with minimum cutoff) were subjected to the MAP-DP analysis, as was done for the K-means clustering.

Hierarchical clustering analysis was performed with the genes upregulated or downregulated in T_FH cells relative to their expression in T_H1 cells (Fig. 3c; 1.4-fold cut off, Adj. P-val <0.05) using hclust function from stats package in R and the heatmap was generated using heatmap.2 function from gplots package in R.

Gene Set Enrichment Analysis (GSEA) was run on gene lists pre-ranked by DESeq log P-value multiplied by the sign of the log-fold change⁵² with GSEA v3.0 (Broad Institute, Inc.).

Bcl-6–bound, Blimp-1–bound, and T_FH, T_H1, T_H2, T_H17, and T_REG signature gene lists were collected from previous studies^{19,22,24,25,53,54} and from Ingenuity Pathway Analysis database (Qiagen). E2A target-gene list was generated by Shaw et al.²² using ChIP-seq results from thymocytes by E2A Bio-ChIP method in previously described⁵⁵. In multiple GSEA (Fig.3f), the nominal P-values were corrected with Benjamini-Hochberg algorithm.

ATAC-seq.

ATAC-seq was performed with a modified method from previously described⁵⁶. Spleens were isolated and pooled from 3–5 mice per group. 5 × 10⁴ CXCR5⁺SLAM^lo/int T_FH, CXCR5^loSLAM^hi T_H1 SMARTA cells (CD45.1⁺CD4⁺CD8^–B220^– singlets), or naive SMARTA cells (CD4⁺CD8^–B220^–CD44^loCD62L^hi) were sorted using a FACSAria. Cells were pelleted and resuspended in 25 μl lysis buffer, and spun down. The nuclear pellet was resuspended into 25 μl transposition reaction mixture containing Tn5 transposase from Nextera DNA Sample Prep Kit (Illumina) and incubated at 37 °C for 30 min. Then the transposase-associated DNA was purified using MinElute Purification kit (Qiagen). To amplify the library, the DNA was amplified for twelve cycles using KAPA Real-Time Library amplification kit (KAPAbiosystems) with Nextera indexing primers. The total amplified DNA was purified using AmPureXP beads. The quantity and size of amplified DNA was examined by TapeStation to confirm that independent samples exhibited similar fragment distributions. The libraries were sequenced using Hiseq4000 with paired-end sequencing (Illumina). Replicates were generated from 3 independent experiments.

ATAC-seq analysis.

Fastq reads were aligned with the mm10 reference genome with Bowtie2 (-p 15 -m 1 -best -strata -X 2000 -S –fr –chunkmbs 1024). PCR duplicates were removed by SAMtools. Peaks were called with MACS2 (macs2 callpeak –t inputfile –f BED –g mm –n outputfile –nomodel −1 0.01 –keep-dup all –call-sumits -B). Bigwig files for ATAC-seq signal visualization on UCSC genome browser was generated by converting MACS2 output read pileup file into bigwig files with bedgraphtobigwig from UCSC tools. Peaks from all samples were merged to create a common reference peak set using Homer (mergePeaks -d 200). Peaks localized within 100 kb upstream of the transcription start site (TSS) and 100 kb downstream of the transcription end site (TES) were annotated to a gene⁴⁶. hTseq-count was used to calculate Tn5 insertion site number in peaks⁵⁷. Scikit-learn was used for t-SNE analysis of Tn5 count matrix to visualize sample similarity⁵⁸. For differential analysis, DEseq2 was used to compare Tn5 insertion site numbers in peaks for each condition after filtering out low read peaks (count number <5 in all samples in comparison). The relative open / close regions were defined by using DEseq2 to compare the Tn5 insertion site number within ATAC-seq peaks among samples (DEseq2 raw P-val < 0.05). Motif enrichment was performed by Homer (findMotifsGenome.pl -size given -mis 3 -mask), with combination of Homer curated known motif set, motifs from MEME motif database, and JASPAR database. Frequencies of the enriched TF motifs were determined in regions of increased or decreased accessibility (DEseq2 raw P-val < 0.05). Significance for the differences of TF motif enrichment across comparisons was determined by Chi-squared tests. Family of transcription factors were characterized with information from tFclass database⁵⁹. Transcription factor footprint analyses were performed with RGT-hint⁶⁰ using JASPAR database. JASPAR motifs are described by a position weight matrix (PWM) providing information on the frequency of usage of each nucleotide in the motifs derived from experimental data, such as ChIP-seq⁶¹. Although two related TFs (e.g., Runx2 and Runx3) can bind the identical consensus sequences, use of the PWMs in JASPAR allows for non-identical assignments if the data support differential PWMs. PageRank analysis was performed with Taiji as previously described method⁵⁶, as a separate bioinformatic algorithm based on differential gene expression and TF binding motifs, with compatible results with differential ATAC-seq analysis (Extended Data Fig.6f), although the nature of the PageRank algorithm is more effective for identifying activator TFs instead of repressors.

ChIP-seq analysis.

Raw sequencing reads for Blimp-1 (GSE75724, GSE79339)^53,54, Tcf-1 (GSE103387)⁶², and BCL-6 and NCOR (GSE29282)³⁵ were downloaded from the SRA database. Reads were aligned to UCSC mm9 (Blimp-1), mm10 (Tcf-1), or hg19 (BCL-6 and NCOR) with Bowtie (v1.1.2) using options (-S–-fr -p 3 -m 1 -k 1–-best –strata) and peaks were called using MACS with default settings. The UCSC tracks and peak calls of BCL-6 ChIP-seq data of human tonsillar GC-T_FH (GSE59933) were from the data of the original paper²⁴. Peaks were annotated based on the RefSeq database. Peaks localized ±2 kb of the TSS were defined as promoter peaks, peaks localized ±2 kb of the TES were defined as 3′ end peaks, and peaks >2 kb away from genes were defined as intergenic²⁴. Syntenic analysis of human Bcl-6-binding sites in the mouse genome was performed using the Liftover tool of UCSC Browser with default settings (https://genome.ucsc.edu/cgi-bin/hgLiftOver). The human BCL-6 binding peaks firstly converted from hg18 to mm9, and then converted from mm9 to mm10 (7,673 peaks were successfully converted to mouse genome from the original human 8,523 peaks). Bcl-6-binding in mouse genome was evaluated by ChIP-qPCR for some genes of interest.

ChIP-qPCR.

To validate whether Bcl-6 binding in human GC-T_FH cell is conserved in mouse T_FH cell, Bcl-6 ChIP-qPCR of mouse T_FH cells was performed using tagged-Bcl6 RV transduction. We confirmed that Bcl-6 protein from pMIG-Bcl6 (Bcl6-RV) was expressed at levels similar to that of endogenous Bcl-6 in T_FH cells (Extended Data Fig.5e). We then validated that an N-terminal Myc-tagged Bcl-6 fusion protein (myctagN-Bcl6-RV) was functionally comparable with non-tagged Bcl6-RV (Bcl6-RV), based on tag-Bcl6-RV rescue of T_FH differentiation and function of Bcl6^f/fCre^CD4 CD4⁺ T cells in LCMV infected mice (Extended Data Fig.5e,f). To perform myctagN-Bcl6 ChIP-qPCR, Bcl6^f/fCre^CD4 SMARTA cells transduced with myctagN-Bcl6-RV were transferred to C57BL/6 mice, followed by infection of the host mice with LCMV_Arm. Seven days later, spleens were isolated and pooled from 30 mice, and pre-enriched CD45.1⁺GFP⁺ SMARTA cells were further sorted to obtain CXCR5⁺SLAM^lo T_FH cells. 9 × 10⁶ T_FH cells were fixed in 1% formaldehyde for 2.5 min and then quenched with 125 mM glycine for 5 min. Cells were lysed using truChIP Chromatin Shearing Kit and sonicated to generate < 500-bp fragments using E220 Focused-ultrasonicator (Covaris). Fragmented DNA was used as input control. Magnetic Dynabeads (45 μL) were washed with IP buffer (50 mM NaCl, 5 mM EDTA, 50 mM Tris pH8.0, and 0.1% NP-40) and then mixed with 7.5 μg anti-Myc tag or goat IgG (Abcam) antibodies in 300 μl IP buffer and rotated for 6 h at 4 °C. The sonicated lysates were diluted in IP buffer at 1:4 ratio and precleared with Dynabeads for 2 h at 4 °C. The precleared lysates were added to antibody-conjugated Dynabeads and incubated overnight at 4 °C. The beads were washed with IP buffer once, Wash Buffer I (150 mM NaCl, 0.5% sodium deoxycholate, 1% NP-40, 0.1% SDS, 1 mM EDTA, and 50 mM Tris pH8.0) twice, Wash Buffer II (500 mM NaCl, 0.5% sodium deoxycholate, 1% NP-40, 0.1% SDS, 1 mM EDTA, and 50 mM Tris pH8.0) twice, Wash Buffer III (250 mM LiCl, 0.5% sodium deoxycholate, 1% NP-40, 0.1% SDS, 1 mM EDTA, and 50 mM Tris pH8.0) twice, and TE buffer twice for 5 min each. The beads were resuspended in 200 μl Elution Buffer (100 mM NaHCO₃ and 1% SDS) and reverse-crosslinked at 65 °C for 30 min and then treated with rNase A for 30 min at 37 °C and Proteinase K at 65 °C for overnight. DNA was purified using AMPureXP beads, and eluted in nuclease-free water. The eluted DNA were further diluted in water and subjected to perform qPCR using primers: Selplg E1 F, 5′-CGCACAAACACACACAACTC-3′; Selplg E1 R, 5′-TCAGACCCTCCAAACTACCT-3′; Runx2 E1 F, 5′-AGATCGCTCACTCGACTCAT-3′; Runx2 E1 R, 5′-CTTCTTCTACTTCCGCCACAC-3′; Runx2 E2 F, 5′-TCCTTGTCTCTTGCTCTCTTTC-3’; Runx2 E2 R, 5′- ACAGGTAGTGGCATAGAGGA-3’; Runx2 E3 F, 5’- GCTGTGTGT TCTTGCTCTTCT-3′; Runx2 E3 R, 5′-CTAATGAGATGCTGTCGCTGAA-3’; Runx3 E1 F, 5′- GAGAGCCTTTGAGGTCTCTTTG-3′; Runx3 E1 R, 5′- CTCAACAGTGCACACCTTCT-3′; Klf2 P1 F, 5′-AGCAAGGTACCAGGCTACA-3′; Klf2 P1 R, 5′- TCCCACAGCCTGAAGTCTAA-3’; Klf2 DE F, 5′-CTATCTCAGGCAACCCAATCA-3′; Klf2 DE R, 5′-ACCGCTGAAGTTTCTAGGTAAA-3′; Neg F 5′-GCCGCTCTATCATCCGAAAT-3′; Neg R 5′-CCAGCTGCAAGATTAACACAAC-3′. Negative control region was arbitrary selected approximately 40 kb upstream of Sleplg E1 site (Fig.5e).

Immunoblot Analysis.

Equivalent cell numbers were lysed in 2X Laemmli sample buffer and boiled for 10 min. The proteins were resolved by NuPAGE 4–12% Bis-Tris gels and transferred onto PVDF membrane in NuPAGE transfer buffer (Invitrogen). Membranes were incubated with primary antibodies: anti-Myc tag (Cell Signaling), anti-Klf2 (EMD Millipore), anti-Runx3 (HRP-conjugated, Santa Cruz Biotechnology), and anti-GAPDH (Santa Cruz Biotechnology). After incubation with HRP-conjugated secondary antibody, target proteins were detected by ECL prime detection kit (GE healthcare) and OdysseyFc Imaging System (LI-COR). The band densities were quantified by ImageStudio Lite software (v.5.2.5; LI-COR).

Statistical analysis.

All RNA-seq and ATAC-seq were performed independently in 3–4 replicates. All graphs represent mean and standard deviations (SD) unless otherwise noted (Prism 8.0, GraphPad). Comparison between two groups was determined by unpaired or paired Student’s t-test with a 95% confidence interval. Statistical details of each experiment can be found in the figure legends and specific methods.

Extended Data

Extended Data Fig. 1. T_FH differentiation is not the default pathway.

Related to Fig. 1.

a, Schematic diagram of a mutually antagonistic relationship of Bcl-6 and Blimp-1. UCSC browser tracks from BCL-6 ChIP-seq and Blimp-1 ChIP-seq were indicated below.

b, A null hypothesis of a default T_FH differentiation pathway model.

c, SMARTA CD4 T cells were transferred to C57BL/6 host mice, followed by immunization of the host mice with KLH-gp₆₁ in alum only, alum + LPS, alum + Poly(I:C), or alum + cGAMP adjuvants, and analyzed 7 days later. Representative flow cytometry of T_FH and non-T_FH SMARTA cell subsets from draining LNs (dLNs) of KLH-gp₆₁ immunized mice. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

d, Quantification of results from Fig.1b.

e, Schematic of the SMARTA cell transfer system used for LCMV_Arm infection. WT, Bcl6^f/f, Prdm1^f/f, and Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4⁺ T cells were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days later.

f, Representative flow cytometry of GC-T_FH, T_FH and non-T_FH SMARTA cell subsets from spleens of LCMV_Arm infected mice in Extended Data Fig.1c. Three independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

g, Quantification of results from Extended Data Fig.1f.

h, Representative histology section to define GC, B cell zone, T cell zone, and T-B border. T-B border was defined as ±15 μm region at a boundary line of T cell zone and B cell zone. Scale bar, 200 μm. Related to Fig.1g.

i, Histology of draining LNs at d8 after KLH-gp₆₁ immunization in Fig. 1d. Blue, TCRβ; red, GL7; green, IgD; white, CD45.1 SMARTA. SMARTA cells were presented with large dots for clarity. Scale bar, 200 μm. Related to Fig.1g.

Extended Data Fig. 2. Bcl-6 exhibits direct negative autoregulatory feedback.

Related to Fig. 2.

a, Sequence homology at BPS1 locus between various species. The numbers indicate region of mouse chromosome 19 (mm10).

b, Schematic diagram of self-inactivating (SIN) vectors. pQdT SIN vector were generated from original pQCXIP SIN vector by removing IRES-Puro^R cassette and additional deletion of CCAAT box and TATA box.

c, Schematic of the RV⁺ SMARTA cell transfer system used for LCMV_Arm infection. SMARTA CD4 T cells transduced with WT Thy1.1-RV or ΔBPS1 Thy1.1-RV were adoptively transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days after infection. Related to Fig.2a–b and Extended Data Fig.2d–e.

d, T_FH and T_H1 gates used for Fig.2b and Extended Data Fig.2e.

e, Chromatogram of DNA sequencing from WT or ΔBPS1 mouse.

f, Schematic of the SMARTA cell transfer system used for LCMV_Arm infection. WT or ΔBPS1 SMARTA CD4 T cells were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days after infection. Related to Fig. 2c–d.

g, Quantification of expression of CXCR5, SLAM, and PSGL1 in Fig.2d, gated on CXCR5^hiSLAM^lo T_FH or CXCR5^loSLAM^hi T_H1 cells. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

h, Genome-browser tracks depict BCL-6 or NCOR ChIP-Seq peaks at BCL6 locus from OCI-LY1 cell line. Peak calls indicated below each track.

Extended Data Fig. 3. RNA-seq of T_FH and non-T_FH populations from multiple genetic backgrounds.

Related to Fig. 3

a, Schematic of the SMARTA cell transfer system used for RNA-seq analysis in KLH-gp₆₁ immunization. Non-T_FH (CXCR5^loPSGL1^hi) populations from WT, Bcl6^f/f Cre^Cd4, Prdm1^f/f Cre^CD4, or Bcl6^f/fPrdm1^f/f Cre^Cd4 SMARTA cells and T_FH (CXCR5^hiPSGL1^lo) populations from WT or Prdm1^f/f Cre^CD4 SMARTA cells were sorted from C57BL/6 host mice given WT, Bcl6^f/f, Prdm1^f/f, or Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4 T cells, followed by infection of the host mice with KLH-gp₆₁ in alum + cGAMP, and analyzed 8 days later. Naive SMARTA cells were isolated as CD44^loCD62L^hiCD45.1⁺ from uninfected mice. Representative flow cytometry of T_FH and T_H1 subsets from four independent experiments.

b, Heatmap of gene expression of curated T_FH-associated genes. Scale, row z-score. DKO, Bcl6^f/fPrdm1^f/f Cre^CD4. As a first analysis the effect of Bcl6/Prdm1 double-deficiency on the T_FH biology, we assessed expression of a broad curated set of T_FH-associated genes across all samples from RNA-seq gene expression profiling. Bcl6^f/fPrdm1^f/fCre^CD4 T_FH-like cells lost expression of positively T_FH-associated genes in comparison to WT T_FH cells or Prdm1^f/fCre^CD4 T_FH cells. Conversely, Bcl6^f/fPrdm1^f/fCre^CD4 T_H1-like cells had a gene expression profile different from WT T_H1 or Bcl6^f/fCre^CD4 T_H1 cells.

c, Principal component analysis of differential gene expression from RNA-seq of LCMV_Arm infection. Principal component analysis provided similar findings as (b), supporting the overall hypothesis that T_FH is not a default differentiation pathway of CD4⁺ T cells and Bcl-6 has important activities beyond inhibition of Prdm1.

d, Gene expression changes were clustered by K-means clustering analysis (K=10) of LCMV_Arm infection.

e, Hierarchical clustering analysis of genes upregulated or downregulated in T_FH cells relative to their expression in T_H1 cells (1.4-fold cut off, Adj. P <0.05) of LCMV_Arm infection shown in Fig.3c. Scale, row z-score.

f, GSEA of BCL-6 bound genes from human tonsillar GC-T_FH compared to Cluster 2 genes (left) or Cluster 3 genes (right) differentially expressed between Bcl6^f/fPrdm1^f/f T_FH-like cells and Prdm1^f/f T_FH cells. NES, normalized enrichment score; FDR, false discovery rate.

g, GSEA of Blimp-1 bound genes from CD8 and T_H17 cells in comparison of Cluster 2 and Cluster 3 genes differentially expressed between Bcl6^f/fPrdm1^f/f T_H1-like cells and Bcl6^f/f T_H1 cells.

Extended Data Fig. 4. Bcl-6 drives CXCR5 expression via repression of Id2-E2A pathway.

Related to Fig. 4

a, Quantification of GC-T_FH core signature markers in SMARTA cells from dLNs of KLH-gp₆₁ immunized mice in Fig. 4d.

b, Schematic of the SMARTA cell transfer system used for LCMV_Arm infection. WT, Bcl6^f/fPrdm1^f/f, or Bcl6^f/fPrdm1^f/fId2^f/f Cre^CD4 SMARTA CD4 T were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days later. Related to Fig 4g and Extended Data Fig. 4c–g.

c,e, Representative flow cytometry of T_FH and GC-T_FH SMARTA cell subsets from spleen of LCMV_Arm infected mice. Two independent experiments performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

d, Quantification of expression of CXCR5 in Extended Data Fig. 4c, gated on SMARTA cells.

f, Quantification of frequency of CXCR5^hiPD1^hi GC-T_FH cells in Extended Data Fig.4e, gated on SMARTA cells.

g, Quantification of GC-T_FH core signature markers in Extended Data Fig.4e, gated on SMARTA cells.

Extended Data Fig. 5. ATAC-seq of T_FH and non-T_FH populations from multiple genetic backgrounds.

Related to Fig. 5

a, Representative flow cytometry of naive SMARTA cells (CD44^loCD62L^hiCD45.1⁺) used for ATAC-seq analysis.

b, Genome-browser tracks depict ATAC-seq chromatin accessibility at Cxcr5 locus. Peak calls indicated below each track. * indicates DEseq2 raw p ≤ 0.05 in comparison between WT T_FH and T_H1.

c, Genome-browser tracks depict ATAC-seq chromatin accessibility at Tbx21 and Gata3 loci.

Peak calls indicated below each track. Bcl-6 liftover peaks from human to mouse reference genome are indicated. * and ** indicate DEseq2 raw p-val 0.05 and 0.01, respectively, in comparison between WT T_FH and T_H1. Gene expressions from RNA-seq data of LCMV_Arm infected mice are graphed. DKO, Bcl6^f/fPrdm1^f/f Cre^CD4.

d, Genome-browser tracks depict BCL-6 ChIP-Seq peaks at SELPLG locus. Peak calls indicated below the track.

e, Representative flow cytometry and quantification of the level of Bcl-6 in RV⁺ T_FH or non-T_FH SMARTA cell subsets from spleen of LCMV_Arm infected mice. SMARTA CD4 T cells transduced with GFP-RV or Bcl6-RV were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 3 days after infection. Non-T_FH: GFP-RV⁺ and T_FH: GFP-RV⁺ indicate the expression level of endogenous Bcl-6 in each population. Each dot represents one mouse (n = 4). Data are mean s.d., unpaired two-tailed Student’s t-test.

f, Representative flow cytometry and quantification of GC-T_FH and B_GC subpopulations from spleen of LCMV_Arm infected mice. Bcl6^f/f Cre^CD4 SMARTA CD4 T cells transduced with Bcl6-RV or myctagN-Bcl6-RV were transferred to Bcl6^f/f Cre^CD4 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 8 days after infection. We validated that an myctagN-Bcl6-RV was functionally comparable with non-tagged Bcl6-RV (Bcl6-RV), based on myctagN-Bcl6-RV rescue of T_FH differentiation and B cell help function of Bcl6^f/f Cre^CD4 CD4 T cells in LCMV infected mice. Each dot represents one mouse (n = 4). Data are mean s.d.

Extended Data Fig. 6. Identification of candidate TFs.

Related to Fig. 6

a, Heatmap plots representing the frequencies of the most enriched TF motifs in regions in decreased accessibility (relatively less open in first group than second group, DEseq2 raw p-val < 0.05). Scale, motif frequencies (%). Related to Fig.6b.

b, Motif analysis of the center of TF footprint (green bar) in Fig.6c and Extended Data Fig.6e from ATAC-seq reads.

c, Genome-browser tracks depict ATAC-seq chromatin accessibility and TF occupancy at Icos and Cd200 loci. Peak calls indicated below each track. ** indicates DEseq2 raw p-val ≤ 0.01 in comparison between Bcl6^f/fPrdm1^f/f Cre^CD4 T_FH-like and Prdm1^f/f Cre^CD4 T_FH.

d, Gene expression of Klf genes from RNA-seq data of LCMV_Arm infected mice or KLH-gp61 immunized mice.

e, TF footprints derived from ATAC-seq reads over representative TF motifs within accessible ATAC-seq regions.

f, Heatmap plots of relative PageRank scores. Scale, row z-score.

Extended Data Fig. 7. CRISPR/Cas9-mediated gene knockdown of SMARTA cells.

Related to Fig. 7

a, Schematic of the CRISPR/Cas9-mediated gene knockdown of SMARTA cell system. Details are in the Method section.

b, Representative flow cytometry of RNP transfection from more than four independent experiment; each dot represents one RNP+ group (n = 9). Data are mean ± s.d. Quantification of cell viability and transfection efficiency of RNP⁺ and RNP⁻ (electroporation without RNP) SMARTA cells.

c, Frequency of crRNA⁺ CD45.1⁺ SMARTA cells among total CD4 T cells from spleen of LCMV_Arm infected mice in Fig.7a. Two independent experiments were performed; each dot represents one crRNA⁺ group (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

d,e,f,g, WT or Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4 T cells transfected with crCd8, crGata3, crRunx2, crRunx3, and crKlf2 were transferred to C57BL/6 host mice, followed by infection with LCMV_Arm (e,f,g) or immunization with KLH-gp₆₁ (d) of the host mice, and analyzed 6–7 days later. crRNA⁺ SMARTA cells were FACS sorted from spleens (LCMV_Arm infection) or dLN (KLH-gp₆₁ immunization) of host mice. Gene knockdown efficiencies were measured by flow cytometry (d), mRNA qPCR (e), or Western blot analysis (f,g). shRunx3-RV⁺ SMATRA cells and pMIG-Klf2-RV⁺ SMARTA cells were used as positive controls. Two independent experiments were performed; Each dot represents one mouse (n = 5). Data are mean ± s.d., unpaired two-tailed Student’s t-test. Details are in the Method section.

h, Schematic of the CRISPR/Cas9-mediated gene knockdown of SMARTA cell system used for testing Gata3 in LCMV_Arm infection. WT or Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4 T cells transfected with crCd8 or crGata3 were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 6–7 days later. Related to Fig.7c and Extended Data Fig.7i,l,m.

i,j,k, Frequency of crRNA⁺ CD45.1⁺ SMARTA cells among total CD4 T cells from spleen of LCMV_Arm infected mice in Fig.7d and Extended Data Fig.7h. Two (i) or three (j,k) independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

l, Representative flow cytometry of T_FH SMARTA cell subsets from spleen of LCMV_Arm infected mice in Extended Data Fig.7h. Two independent experiments were performed; Each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

m, Quantification of expression of PSGL1 in Extended Data Fig.7h, gated on SMARTA cells. CD44^lo naive CD4 T cells were used as a negative control.

Extended Data Fig. 8. Identification of Runx2, Runx3, and Gata3 as repressors of T_FH genes, acting downstream of Bcl-6.

Related to Fig. 7

a, Schematic of the RV⁺ SMARTA cell transfer system used for LCMV_Arm infection. WT SMARTA CD4 T cells transduced with pMIG (GFP-RV⁺), pMIG-Runx3myc (Runx3-RV⁺), or pMIG-Runx2myc (Runx2-RV⁺) were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days after infection. Related to Fig.7i–j and Extended Data Fig.7o–u.

b, Total GFP⁺ RV⁺ SMARTA cells [RV⁺ high] were FACS sorted from in vitro culture. Whole cell lysates were analyzed by immunoblotting using anti-myc tag and anti-GAPDH antibodies. Relative Runx expressions are indicated.

c, SMARTA CD4 T cells transduced with pMIG (GFP-RV⁺), pMIG-Runx3myc (Runx3-RV⁺ [High]), or pMIG-Runx2myc (Runx2-RV⁺ [High]) were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days later. Frequency of CD45.1⁺ SMARTA cells among total CD4 T cells from spleen of LCMV_Arm infected mice. Two independent experiments were performed; each dot represents one mouse (n = 4). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

d, Representative flow cytometry of T_FH RV⁺ [High] SMARTA cell subsets from spleen of LCMV_Arm infected mice. Quantification of results from left panels.

e, Bottom 10% [Low], bottom 10–30% [Med] and remaining 30–100% [High] GFP⁺ Runx3-RV⁺ SMARTA cells, and total GFP⁺ RV⁺ SMARTA cells were FACS sorted from in vitro culture. Whole cell lysates were analyzed by immunoblotting using anti-Runx3 and anti-GAPDH antibodies. Relative Runx expressions are indicated.

f, Bottom 10% GFP⁺ Runx3-RV⁺ (Runx3-RV⁺ [Low]), bottom 30% GFP⁺ Runx2-RV⁺ (Runx2-RV⁺ [Med]), and total GFP⁺ RV⁺ (GFP-RV⁺) SMARTA cells were FACS sorted from in vitro culture. Whole cell lysates were analyzed by immunoblotting using anti-myc tag and anti-GAPDH antibodies. Relative Runx expressions are indicated.

g, SMARTA CD4 T cells transduced with pMIG (GFP-RV⁺), pMIG-Runx3myc (Runx3-RV⁺ [Low]), or pMIG-Runx2myc (Runx2-RV⁺ [Med]) were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 7 days later. Frequency of CD45.1⁺ SMARTA cells among total CD4 T cells from spleen of LCMV_Arm infected mice. Two independent experiments were performed; each dot represents one mouse (n = 5). Data are mean ± s.d., unpaired two-tailed Student’s t-test.

h, Representative flow cytometry of T_FH Runx3-RV⁺ [Low] and Runx2-RV⁺ [Med] SMARTA cell subsets from spleen of LCMV_Arm infected mice. Quantification of results from left panels.

h, Representative flow cytometry of T_FH Runx3-RV⁺ [Low] and Runx2-RV⁺ [Med] SMARTA cell subsets from spleen of LCMV_Arm infected mice. Quantification of results from left panels.

Extended Data Figure 9. Identification of Klf2 as a repressor, acting downstream of Bcl-6 regulating major of T_FH genes.

Related to Fig.8

a-c, WT or Bcl6^f/fPrdm1^f/f Cre^Cd4 SMARTA CD4 T cells transfected with crCd8 or crKlf2 were transferred to C57BL/6 host mice, followed by infection of the host mice with LCMV_Arm, and analyzed 6–7 days later (n=4–5 mice per group). (a) Frequency of CD45.1⁺ SMARTA cells among total CD4 T cells from spleen of LCMV_Arm infected mice. (b) Representative flow cytometry of gp₆₆-restimulated IL-21⁺ SMARTA cells. Quantification of results from left panels. (c) Quantification of expression of Tcf-1, T-bet, GATA-3, and Maf, gated on SMARTA cells. Three independent experiments were performed; each dot represents one mouse (a, n = 4; b-c, n = 5). Data are mean ± s.d., unpaired two-tailed Student’s t-test. Related to Fig.8a–c.

d, Schematic of the CRISPR/Cas9-mediated gene knockdown of SMARTA cell system used for testing Klf2 in KLH-gp₆₁ immunization. WT or Bcl6^f/fPrdm1^f/f Cre^CD4 SMARTA CD4 T cells transfected with crCd8 or crKlf2 were transferred to C57BL/6 host mice, followed by immunization of the host mice with KLH-gp₆₁ and alum + cGAMP, and analyzed 8 days later. Related to Fig. 8d–f.

e,g, Genome-browser tracks depict ATAC-seq chromatin accessibility and TF occupancy. Peak calls indicated below each track. * and ** indicate DEseq2 raw p-val ≤ 0.05 and ≤ 0.01, respectively, in comparison between Bcl6^f/fPrdm1^f/f Cre^CD4 T_FH-like and Prdm1^f/f Cre^CD4 T_FH. Gene expression from RNA-seq data of LCMV_Arm infected mice and KLH-gp₆₁ immunized mice are graphed.

f, Genome-browser tracks depict Tcf1 ChIP-Seq peaks at Pdcd1 and IL6ra loci from murine T_FH cells.

h, Gene expression of Tox and Tox2 from RNA-seq data of LCMV_Arm infected mice or KLH-gp₆₁ immunized mice.

i, Representative flow cytometry of T_FH cells, gated on SMARTA cells from spleens of LCMV_Arm infected mice (n=4 mice per group) in Fig. 8a. Quantification of results from left panels.