High-affinity, neutralizing antibodies to SARS-CoV-2 can be made without T follicular helper cells

T follicular helper (TFH) cells are the conventional drivers of protective, germinal center (GC)–based antiviral antibody responses. However, loss of TFH cells and GCs has been observed in patients with severe COVID-19. As T cell–B cell interactions and immunoglobulin class switching still occur in these patients, noncanonical pathways of antibody production may be operative during SARS-CoV-2 infection. We found that both TFH-dependent and -independent antibodies were induced against SARS-CoV-2 infection, SARS-CoV-2 vaccination, and influenza A virus infection. Although TFH-independent antibodies to SARS-CoV-2 had evidence of reduced somatic hypermutation, they were still high affinity, durable, and reactive against diverse spike-derived epitopes and were capable of neutralizing both homologous SARS-CoV-2 and the B.1.351 (beta) variant of concern. We found by epitope mapping and B cell receptor sequencing that TFH cells focused the B cell response, and therefore, in the absence of TFH cells, a more diverse clonal repertoire was maintained. These data support an alternative pathway for the induction of B cell responses during viral infection that enables effective, neutralizing antibody production to complement traditional GC-derived antibodies that might compensate for GCs damaged by viral inflammation. Description Complementary TFH cell–dependent and –independent pathways of antibody production mediate neutralizing responses to SARS-CoV-2.


INTRODUCTION
Antibodies mediate protection against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the causative agent of the coronavirus disease 2019 (COVID- 19) pandemic (1). T follicular helper (T FH ) cells are the conventional drivers of protective antibody responses, as they support immunoglobulin (Ig) class switching, germinal center (GC)-based affinity maturation, and long-lived humoral immunity (2). Multiple studies have reported a correlation between circulating T FH (cT FH ) cells-in particular, type 1 CXCR3 + CCR6 − cT FH cells-and neutralizing antibody titers in patients with COVID-19 (3). Although there have been mixed findings on the association of T FH cells with disease severity, several groups have observed an absence of T FH cells and GCs in severely ill patients (4)(5)(6)(7). Despite a loss of T FH cells and GC structures, T cell-B cell interactions and antibody class switching still occur in the secondary lymphoid organs of these patients (6). Findings of enhanced extrafollicular B cell responses associated with severe disease further suggest that noncanonical pathways of antibody production may be operative in these individuals (6,8). However, a causal relationship between antibody provenance and disease severity cannot be established by existing human studies. Thus, it remains unclear whether antibodies produced through noncanonical pathways without T FH cell help are also protective against SARS-CoV-2.
No study has evaluated the requirement for T FH cells in antibody production to SARS-CoV-2. Previous work with a related coronavirus, SARS-CoV-1, has shown that CD4 + T cells are required for neutralizing antibody production (9), but the necessary CD4 + T cell subset has not been identified. For other viruses, as well as bacteria, T FH cells are thought to be required for class-switched, pathogen-specific antibody production, especially at later time points (10)(11)(12)(13)(14)(15)(16)(17)(18). In contrast, different studies have found conflicting results on the requirement of T FH cells during vaccination in mice. Certain vaccine strategies induce robust T FH -independent antibody responses, although of lower affinity (19,20), whereas other strategies fail to induce durable, class-switched, or high-affinity antibodies in mice with T FH cell deficiency or dysfunction (11,(21)(22)(23). In humans, though, cT FH cell populations in the blood correlate with the response to influenza vaccination (24,25). Together, although non-T FH CD4 + T cells can promote effective antibodies in certain contexts, protective antipathogen humoral immunity is largely thought to be T FH cell-dependent.
We therefore directly tested whether non-T FH CD4 + T cells could compensate for T FH cell loss in severe COVID-19, and perhaps during acute viral infection in general, by promoting class-switched antibodies. We also sought to compare the T FH -independent antibody response to SARS-CoV-2 infection with that induced by mRNA vaccination. On the basis of prior work, we hypothesized that antibodies induced through such noncanonical mechanisms would be lower in quantity and quality. To test this, we characterized the titer, isotype, longevity, affinity, epitope reactivity, and function of antibodies to SARS-CoV-2 infection from T FH -deficient mice. We also examined antibody titers and isotypes after SARS-CoV-2 vaccination and influenza A virus infection. Both T FH cells and non-T FH CD4 + T cells promoted class-switched antibodies to SARS-CoV-2 infection, SARS-CoV-2 vaccination, and influenza A virus. Contrary unaffected by the absence of T FH cells ( fig. S3B). Together, these results suggest that both T FH cells and non-T FH CD4 + T cells promote the production of class-switched antibodies to SARS-CoV-2, although T FH cells are specifically able to induce certain subclasses.
To corroborate these findings with a model that does not require AAV pretransduction, we measured serum antibody titers of K18-hACE2 Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice at 14 dpi with SARS-CoV-2 (Fig. 1C). K18-hACE2 Bcl6 fl/fl Cd4 Cre mice produced substantial levels of S-specific IgG antibodies, though reduced compared with K18-hACE2 Bcl6 fl/fl mice (Fig. 1D). K18-hACE2 Bcl6 fl/fl Cd4 Cre mice also demonstrated similar patterns of IgG subclass production to AAV-hACE2 Bcl6 fl/fl Cd4 Cre mice. Whereas S-specific IgG1 and IgG3 were completely abrogated in the absence of T FH cells, S-specific IgG2b, IgG2c, and IgM were only partially reduced ( Fig. 1D and fig. S3C). This alternative model of SARS-CoV-2 infection therefore confirmed that certain IgG subclasses could be generated through T FH -independent mechanisms.
Having observed T FH -independent class-switched antibody production to SARS-CoV-2 infection, we asked whether this could also happen during non-live pathogen-driven immune stimulation such as vaccination. We hypothesized that although the strong and prolonged inflammatory response caused by infection might overcome the requirement for T FH cell help, antibodies to vaccination would still be fully T FH -dependent. We therefore vaccinated mice intramuscularly with a single dose of Moderna mRNA-1273 or Pfizer-BioNTech BNT162b2 mRNA vaccine and evaluated antibody responses at 14 days post-vaccination (Fig. 1E). As with SARS-CoV-2 infection, mRNA vaccination induced both T FH -and non-T FH CD4 + T celldependent S-specific IgG antibodies (Fig. 1F). mRNA vaccination promoted higher levels of S-specific IgG1 compared with infection, and some IgG1 could even be made in the absence of T FH cells, contrasting with the complete T FH cell dependence of infection-induced IgG1. S-specific IgG2b and IgG2c could similarly be made without T FH cell help. S-specific IgG3 and IgM were minimally induced by mRNA vaccination (Fig. 1F and fig. S3D). Therefore, class-switched antibodies to SARS-CoV-2 infection and vaccination are generated through both T FH -dependent and -independent mechanisms.

Influenza virus infection induces T FH -dependent and -independent antibodies
We next determined whether our findings with SARS-CoV-2 infection were generalizable to other models of respiratory viral infection. To this end, we infected mice with mouse-adapted influenza virus A/PR/8/34 H1N1 (PR8) and assessed antibody production at 14 dpi ( Fig. 2A). Similar to SARS-CoV-2 infection, PR8 infection induced both T FH -dependent and non-T FH CD4 + T cell-dependent IgG antibodies, whereas IgM was largely CD4 + T cell-independent (Fig. 2, B and C). Again, PR8-specific IgG1 demonstrated a complete dependence on T FH cell help, whereas PR8-specific IgG2b and IgG2c were promoted by both T FH -dependent and -independent pathways (Fig. 2D). PR8-specific IgG3 was only partially dependent on T FH and CD4 + T cell help (Fig. 2D). Thus, both T FH and non-T FH CD4 + T cells contribute to antibody production in two distinct models of respiratory viral infection.
Similarly, K18-hACE2 Bcl6 fl/fl Cd4 Cre mice produced high-affinity antibodies to S and RBD (Fig. 3B), with a respective 2.6-and 1.7-fold reduction in affinity compared with controls. These data indicate that non-T FH cells can support high-affinity antibody production in two separate models of SARS-CoV-2 infection. However, this was not the case with PR8 infection, as Bcl6 fl/fl Cd4 Cre mice produced IgG antibodies of minimal affinity to both PR8 and PR8 surface glycoprotein hemagglutinin (HA) (Fig. 3C). Therefore, the ability of non-T FH cells to promote high-affinity antibodies may depend on the nature of the viral infection and the antigenic target.
We also evaluated the durability of S-specific IgG antibodies produced by K18-hACE2 Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice after SARS-CoV-2 infection. Previous studies of viral infection have shown that antibody titers in T FH -impaired mice are especially reduced at later time points (10,11,17). In SARS-CoV-2-infected K18-hACE2 Bcl6 fl/fl mice, S-specific IgG levels peaked at 28 dpi and were stable through 84 dpi (Fig. 3, D and E). In K18-hACE2 Bcl6 fl/fl Cd4 Cre mice, S-specific IgG antibodies slowly declined after 28 dpi but, at 84 dpi, still retained 50% of the antibody titer of 28 dpi (Fig. 3E and fig. S4A). AAV-hACE2 Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice displayed a similar pattern of antibody kinetics ( fig. S4, B and C).
Given that both Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice demonstrated persistent S-specific IgG antibodies several months after infection, we investigated whether they had developed a virus-specific long-lived plasma cell compartment. We quantified S-specific IgG antibody-secreting cells (ASCs) in the bone marrow of K18-hACE2 Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice at 85 dpi by enzyme-linked immunosorbent spot (ELISpot). S-specific IgG ASCs were detected in both groups of mice, although they were reduced 10-fold in T FH -deficient mice (Fig. 3F). However, T FH -deficient mice also had fewer total bone marrow plasma cells (BMPCs), so the number of S-specific IgG ASCs per BMPC was similar between Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice (Fig. 3, F and G). Together, these results indicate that T FH -independent antibodies to SARS-CoV-2 can still be high affinity and durabletwo important qualities usually attributed to T FH -dependent responses.

T FH -deficient mice demonstrate similar V gene usage but impaired mutation selection compared with T FH -sufficient mice
To ascertain how T FH -deficient mice generate high-affinity antibodies to S and RBD, we considered two nonmutually exclusive hypotheses: (i) unmutated, germline-encoded BCRs in the murine V(D)J repertoire may already possess high affinity for S and/or RBD and (ii) S-specific B cells may undergo SHM even in the absence of T FH cells and GCs. The former possibility has already been observed for patient-derived SARS-CoV-2-specific antibodies, which demonstrate high potency with minimal SHM (32)(33)(34)(35)(36). The latter possibility is supported by prior reports of SHM occurring at extrafollicular sites during chronic autoimmunity and bacterial infection (37)(38)(39).
To investigate these two possibilities, we isolated S-specific plasmablasts from K18-hACE2 Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice at 14 dpi and performed BCR sequencing ( fig. S5, A to B). After read (D) PR8-specific IgG1, IgG2b, IgG2c, and IgG3 antibody titers at 14 dpi with PR8. Statistical significance was assessed by one-way ANOVA with Tukey's test or Dunnett's test, or Welch's t test with Bonferroni multiple hypothesis correction when sample variances were 0 (B to D). *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Data are expressed as means ± SEM log 10 AU. Each symbol represents an individual mouse. Data are aggregated from two independent experiments with a total of six to eight mice per condition.
preprocessing and V(D)J gene annotation, we clustered the resulting BCR sequences into clonal families, identifying a range of 131 to 694 distinct clones in each sample ( Fig. 4A and fig. S5C). We assessed the relative proportions of the 10 largest clones in each sample, finding that the top-ranked clone in each sample comprised 8 to 56% of total BCR sequences recovered, whereas the top 10 clones and CD138 + BMPC (right) in femur + tibia of K18-hACE2 Bcl6 fl/fl or K18-hACE2 Bcl6 fl/fl Cd4 Cre mice at 85 dpi with SARS-CoV-2. Statistical significance was assessed by one-way ANOVA with Dunnett's test, or Welch's t test with Bonferroni multiple hypothesis correction when sample variances were 0 (A and C); two-tailed unpaired t test or Welch's t test, based on the F test for unequal variance (B, E, and G); two-tailed Mann-Whitney test (F). *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Data are expressed as means ± SEM. Each symbol in (A) to (C) and (E) to (G) represents an individual mouse. Each symbol in (D) represents the mean of nine (Bcl6 fl/fl ) and seven (Bcl6 fl/fl Cd4 Cre ) mice. Data are aggregated from at least two independent experiments with a total of 5 to 15 mice per condition.       accounted for 40 to 76% of total BCR sequences (Fig. 4A). These findings indicate that clonal expansion had occurred in both Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice. To further characterize the clonal architecture of the BCR repertoires, we calculated the diversity of each BCR repertoire using the Shannon entropy score and Simpson's diversity index. The diversity of BCR repertoires in Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice was not statistically different by either metric, though we noted a trend toward increased diversity in the Bcl6 fl/fl Cd4 Cre mice (Fig. 4B). Analysis of Ig heavy-chain variable (V) gene usage frequencies revealed that of all candidate V genes, 22 were used at ≥2% frequency in ≥2 samples. Usage of most V genes was not significantly different between Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice (Fig. 4C). However, we observed that Ighv1-72, Ighv11-2, and Ighv5-6 were used more frequently in BCRs from Bcl6 fl/fl Cd4 Cre mice compared with Bcl6 fl/fl mice. Examining the median usage frequencies, Ighv1-26 was the most frequently used V gene in Bcl6 fl/fl mice, whereas BCRs from Bcl6 fl/fl Cd4 Cre mice most frequently used Ighv1-72. Repeating this analysis on the clone level, rather than the individual BCR sequence level, similarly revealed that usage of most V genes was not significantly different in the absence of T FH cells ( fig. S5D). Thus, although T FH -deficient mice did not exhibit overt changes in V gene usage, there were nevertheless differences that could reflect the distinct nature of the T cell help in these mice.

T DY Y I NWV KQ RPGQG L EW I GW I F PG SG S T Y Y NE K F KG K A T L T V DK S S S T A YML L S S L T S E DS A V Y F CA R Q VQ L QQ SG A E L V RPG T S V K V S CK A SG Y A F T NY L I EWV KQ RPGQG L EW I G V I NPG SGG T NY NE K F KG K A T L T A DK S S S T A YMQ L S S L T S E DS A V Y F CA R Q VQ L QQ PG A E L V K PG A S V KMS C K A SG Y T F T S YW I TWV KQ RPGQG L EW I G D I YPG SG S T NY NE K F K S K A T L T V DT S S S T A YMQ L S S L T S E DS
Studies on patients with COVID-19 have identified potent RBDspecific antibodies with recurrent V gene usage and minimal SHM (32)(33)(34)(35)(36). To explore whether there are murine V genes that can similarly contribute to high-affinity S-or RBD-specific antibodies, we identified the murine V genes expressed in our dataset (≥2% frequency in ≥2 samples) with the greatest homology to frequently used and minimally mutated human V genes ( Fig. 4D and fig. S5, E and F) (32)(33)(34)(35)(36). Human IGHV3-53 and IGHV3-30 demonstrated the highest homology to murine Ighv5-6 and Ighv11-2, which were used more frequently in BCRs from Bcl6 fl/fl Cd4 Cre compared with Bcl6 fl/fl mice (Fig. 4C). Murine V genes most similar to human IGHV1-58, IGHV3-30-3, and IGHV1-69 were used at comparable frequencies between BCRs from Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice. Together, these data suggest that homologous human and murine V genes may contribute to potent SARS-CoV-2-specific antibodies. Furthermore, preferential usage of these V genes in T FH -deficient mice may enable the production of high-affinity antibodies with minimal SHM.
To investigate whether SHM could still occur in T FH -deficient mice, we determined the mutational profiles of each clonal consensus sequence compared with the mouse germline. We found evidence of SHM in S-specific IgG plasmablasts from Bcl6 fl/fl Cd4 Cre mice, albeit at significantly lower frequency compared with plasmablasts from Bcl6 fl/fl mice (Fig. 4E). Overall, 66.28 ± 6.36% (mean ± SEM) of clones from Bcl6 fl/fl mice had somatic mutations, compared with 41.61 ± 3.88% of clones from Bcl6 fl/fl Cd4 Cre mice (Fig. 4F). This contrast was especially pronounced when considering the percentage of clones with two or more somatic mutations: 46.04 ± 6.13% of Bcl6 fl/fl clones and 14.81 ± 1.25% of Bcl6 fl/fl Cd4 Cre clones (Fig. 4F). Similar results were obtained by comparing individual BCR sequences ( fig. S5G).
We further compared mutation selection strength in Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre BCR repertoires using BASELINe, which determines the ratio of nonsynonymous to synonymous mutations compared with a reference model (40,41). Validating the BASELINe analytical approach, selection strength was positive in the complementarity determining regions (CDRs) of BCRs from Bcl6 fl/fl mice and negative in the framework regions (FWRs) (Fig. 4G). This is consistent with the fact that CDRs are the main determinants of antigen specificity and thus enriched in nonsynonymous mutations, whereas FWRs generally serve as structural scaffolds, making them less tolerant of residue-altering mutations. Across both CDRs and FWRs, we found that the ratio of nonsynonymous to synonymous mutations was significantly lower in S-specific IgG plasmablasts from Bcl6 fl/fl Cd4 Cre mice compared with Bcl6 fl/fl mice (Fig. 4G), indicating that T FH cells are required for positive selection of mutated B cell clones against SARS-CoV-2. Collectively, our analysis of the BCR repertoire after SARS-CoV-2 infection suggests that both V gene usage patterns and low levels of SHM, though without mutation selection, may contribute to the production of high-affinity T FH -independent antibodies.

T FH cells focus the antibody repertoire but are dispensable for broad coverage of SARS-CoV-2 epitopes
We next characterized the antibody epitope repertoire of T FHdependent versus T FH -independent responses to SARS-CoV-2. Sera from AAV-hACE2 Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice at 14 dpi were profiled using a bacterial display library of 2410 linear peptides tiling the entire SARS-CoV-2 proteome ( Fig. 5A and data file S1). We first compared the diversity of antibody epitope reactivity, calculating the Shannon entropy, Simpson's diversity index, and the repertoire focusing index within each sample. We observed that antibody diversity assessed by Shannon entropy and Simpson's diversity index was significantly decreased in Bcl6 fl/fl mice compared with Bcl6 fl/fl Cd4 Cre mice, whereas the degree of repertoire focusing was increased ( Fig. 5B and data file S2). These findings were robust to variations in read counts ( fig. S6A). These results suggest that T FH cells help focus the antibody response to particular SARS-CoV-2 epitopes, whereas non-T FH cells promote antibodies to a wider array of targets.
Given these changes in antibody diversity, we next explored whether there were differences in antibody reactivity at the level of individual linear epitopes. After normalizing for read count variations using the median of ratios method ( fig. S6B), we identified epitopes that were comparatively enriched or depleted in Bcl6 fl/fl versus Bcl6 fl/fl Cd4 Cre mice (Fig. 5C). We found that seven epitopes were enriched in Bcl6 fl/fl mice, whereas one epitope was depleted (Fig. 5D). Five of the seven enriched epitopes were derived from S: aa661-672 (proximal to S1/S2 cleavage site), aa801-812 [fusion peptide (FP) 1], aa817-828 (FP2), aa977-988 [heptad repeat (HR) 1], and aa1145-1156 (between HR1/HR2). On the other hand, aa145-156 (N-terminal domain) from S was comparatively depleted in Bcl6 fl/fl mice.
To further investigate alterations in epitope reactivity, we converted the normalized counts to z scores on a sample-by-sample basis, such that the z scores would denote the relative rank of a specific epitope within a particular sample ( fig. S6C). Consistent with our prior analyses, the average z scores in Bcl6 fl/fl versus Bcl6 fl/fl Cd4 Cre mice were similar across most SARS-CoV-2 proteins, with the exception of regions within the S and ORF3a proteins (Fig. 5E). In particular, the regression lines for non-S proteins all closely followed the line of identity, indicating that the relative ranks of epitopes from non-S proteins were largely similar in the presence or absence of T FH cells. These analyses therefore indicate that, although T FH cells are dispensable for antibody production against most SARS-CoV-2 epitopes, they are required to focus the antibody response against certain S-derived epitopes.

RBD-specific antibodies are generated in the absence of T FH cells
Analyzing antibody epitope reactivity along the length of S, we observed that the majority of epitopes enriched in Bcl6 fl/fl mice (17/21 epitopes with differential average z score > 1) were found in the S2 domain (aa686-1273; Fisher's exact test, P = 0.0012) (Fig. 6A), which mediates fusion of viral and target cell membranes (42). These included most of the aforementioned S-derived epitopes that were significantly enriched (Fig. 5C), as well as contiguous epitopes whose enrichment did not reach statistical significance in the epitope-level  S6D). Many of these epitopes are highly conserved across human coronaviruses (hCoVs) as well as the emerging variants of concern (Fig. 6A). The enriched epitopes spanning FP1/ FP2 and preceding HR2 have also been identified in numerous studies profiling the antibody epitope repertoire of patients with COVID-19 (43,44). Given their immunodominance and conservation across hCoVs, these epitopes have been proposed as targets for a pan-coronavirus vaccine.
In contrast, Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice demonstrated similar antibody reactivity to most epitopes within the RBD (Fig. 6B), the target of most neutralizing antibodies (31). However, as most antibodies to RBD likely recognize conformational epitopes (43,44), we also measured RBD-specific antibodies by enzyme-linked immunosorbent assay (ELISA) using full-length RBD. RBD-specific IgG titers normalized by total S-specific IgG were similar between Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice (Fig. 6C). Thus, whereas T FH cells focus the antibody response against immunodominant S2 epitopes, non-T FH cells still promote antibodies against the primary target of neutralization, RBD.
T FH -dependent and -independent antibodies demonstrate similar neutralization potency against homologous SARS-CoV-2 as well as the B.1.351 variant of concern We next evaluated the function of antibodies from Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice after SARS-CoV-2 infection. Although we had observed that T FH -independent antibody responses to the virus lacked IgG1/IgG3 subclasses (Fig. 1, B and D) and S2 epitope focusing (Fig. 6A), these antibodies were still high affinity (Fig. 3, A and B) and could target the RBD (Fig. 6C). We therefore predicted that T FH -independent antibodies would demonstrate similar neutralizing function against homologous SARS-CoV-2 (USA-WA1/2020) as those generated with T FH cell help. Using vesicular stomatitis virus (VSV) pseudotyped with USA-WA1/2020 S protein, we measured the neutralization titer (the reciprocal serum dilution achieving 50% neutralization of pseudovirus infection, NT 50 ) of sera from AAV-hACE2 Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice. Bcl6 fl/fl sera exhibited increased NT 50 (Fig. 7A), which was expected given their higher levels of S-specific IgG antibodies (Fig. 1B). However, by normalizing NT 50 to S-specific IgG levels in each sample, we observed that the neutralization potency indices of Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre sera were similar and actually trended higher for Bcl6 fl/fl Cd4 Cre sera (Fig. 7A). We next tested the same sera against VSV pseudotyped with S protein from the B.1.351 variant of concern. Multiple studies have shown that B.1.351 S mutations, particularly those in the RBD, disrupt binding by neutralizing antibodies and facilitate immune escape (45,46). We therefore hypothesized that increased focusing of T FH -dependent antibodies against conserved S2 epitopes would enable Bcl6 fl/fl sera to better neutralize B.1.351 pseudovirus than Bcl6 fl/fl Cd4 Cre sera. Although Bcl6 fl/fl sera exhibited greater NT 50 than Bcl6 fl/fl Cd4 Cre sera, we found that the neutralization potency index of Bcl6 fl/fl Cd4 Cre sera again trended higher than that of Bcl6 fl/fl sera (Fig. 7B). Nevertheless, both Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre sera demonstrated about a 10-fold reduction in NT 50 against the B.1.351 variant compared with USA-WA1/2020 (Fig. 7C), consistent with previous studies (45,46). Our findings therefore indicate that T FH -independent antibodies exhibit similar, if not increased, neutralization potency to T FH -dependent antibodies in vitro and that S2 epitope focusing does not improve neutralization activity against the B.1.351 variant of concern.
Last, we assessed the function of Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre sera against SARS-CoV-2 in vivo. To this end, we transferred sera from AAV-hACE2 Bcl6 fl/fl or Bcl6 fl/fl Cd4 Cre mice infected with SARS-CoV-2 (USA-WA1/2020) into naïve K18-hACE2 mice (Fig. 7D). Bcl6 fl/fl sera were given undiluted or diluted seven-to ninefold to match the S-specific IgG titer of Bcl6 fl/fl Cd4 Cre sera in a given experiment. One day later, we infected the K18-hACE2 recipients with homologous SARS-CoV-2 and then measured viral burden in the lungs at 4 dpi. Undiluted Bcl6 fl/fl sera led to the greatest reduction in viral burden (Fig. 7, E and F), indicating that, as expected, antibody titer is an important determinant of protection against viral challenge. However, once matched for S-specific IgG titer, Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre sera provided a similar degree of protection (Fig. 7, E and F), corroborating our in vitro findings that T FH -dependent and -independent antibodies exhibit similar neutralization potency. Together, these results demonstrate that T FH -independent antibodies efficiently neutralize both homologous SARS-CoV-2 and the B.1.351 variant of concern but that optimal protection in vivo still relies on T FH cells to generate high antibody titers.

DISCUSSION
Although protective antibodies are generally thought to originate from T FH /GC-dependent pathways, it is unclear what happens to the antibody response when these structures are disrupted, as has been observed in patients with severe SARS-CoV-2 infection (6, 7).
We found that certain class-switched antibodies were reduced but still present in T FH -deficient mice after SARS-CoV-2 or influenza A virus infection, as well as SARS-CoV-2 vaccination. Although BCR analysis demonstrated impairment of SHM and mutation selection, T FH -independent antibodies to SARS-CoV-2 were still high affinity. They were also durable and demonstrated more diverse epitope reactivity compared with T FH -dependent antibodies. T FH -independent antibody responses neutralized both homologous SARS-CoV-2 (USA-WA1/2020) and the B.1.351 variant of concern and were functional in vitro and in vivo ( fig. S7).
Our findings raise the question of which CD4 + T cell subset promotes protective class-switched antibodies in the absence of T FH cells. It has previously been shown in the setting of influenza vaccination that T helper 1 (T H 1) cells make interferon- along with interleukin-21 to induce IgG2c antibodies (19). Although less mutated and of lower avidity, these T H 1-driven antibodies still neutralize influenza virus in vitro and protect from lethal challenge in vivo (19). In addition, a recent study proposed a division of labor between T H 1 cells and T FH cells in promoting IgG2c class switching and supporting GC growth, respectively, during influenza virus infection (47). Therefore, it is possible that lymph node-resident T H 1 cells are also responsible for the IgG2c antibodies we observed to SARS-CoV-2 infection and vaccination (48). Contrary to the consistent T FH -independent induction of IgG2c, IgG1 demonstrated a divergent requirement for T FH cell help between SARS-CoV-2 infection and vaccination. These findings align with prior literature (17,20,23), suggesting that there may be different cellular requirements for IgG1 in the setting of diverse immune stimuli. Similarly, CD4 + T cells are dispensable for IgG3 in response to influenza virus infection (49), but we observed that IgG3 to SARS-CoV-2 was completely dependent on T FH cell help. Thus, the same subclass may be produced by disparate mechanisms even during different viral infections. Last, it is important to consider whether incomplete deletion of T FH cells could contribute to antibody responses in Bcl6 fl/fl Cd4 Cre mice. This same mouse model demonstrates complete abrogation of IgE and IgG1 to allergens (26), and our characterization of cells and structures in the medLN support a lack of T cell-infiltrated, organized GCs in Bcl6 fl/fl Cd4 Cre mice; therefore, we conclude that highly specific and long-lived antibodies can be produced through T FHindependent mechanisms likely dependent on the nature of infection or vaccination.
An unexpected finding from our work was that SARS-CoV-2, but not influenza A virus, induced high-affinity T FH -independent antibodies. As a possible mechanism for this finding, S-specific BCRs from T FH -deficient mice used V genes highly homologous to human V genes that generate potent S-specific antibodies with minimal SHM (32)(33)(34)(35)(36). Furthermore, S-specific BCRs still experienced low levels of SHM in the absence of T FH cell help. Although SHM conventionally occurs in GCs (2), it has also been shown to occur at extrafollicular sites during bacterial infection and chronic autoimmunity (37)(38)(39). However, in the absence of T FH cell help, mutated B cell clones did not experience positive selection, differentiating this process from classical affinity maturation. In this noncanonical pathway of high-affinity antibody production, non-T FH CD4 + T cells may help by selecting naïve B cells that already express potent S-specific BCRs and supporting plasmablast differentiation of lowly mutated B cell clones. Our findings also suggest that long-lived plasma cells can emerge from this pathway, providing a durable source of humoral immunity. Bcl6 fl/fl sera were given undiluted or diluted (dil.) seven-to ninefold to match the S-specific IgG titer of Bcl6 fl/fl Cd4 Cre sera in each experiment. Data are expressed as log 10 PFU per lung lobe by plaque assay (E) or log 10 N1 gene copy number by qPCR, normalized to Actb (F). Statistical significance was assessed by two-tailed Mann-Whitney test (A and B), two-tailed Wilcoxon signed-rank test (C), or one-way ANOVA with Tukey's test (E and F). **P < 0.01; ***P < 0.001; ****P < 0.0001. Data are expressed as means ± SEM. Each symbol represents an individual mouse. Data are aggregated from at least two independent experiments with a total of 8 to 18 mice per condition.
Epitope profiling revealed that T FH cells focus the antibody repertoire against S2-derived epitopes that are highly conserved across hCoVs as well as the emerging variants of concern. These same epitopes have been repeatedly identified in studies profiling the antibody repertoire of patients with COVID-19 (43,44), suggesting that the immunodominance of these epitopes in humans is mediated by T FH cells. It has also been proposed that S2-reactive antibodies in people are primed by prior infections with endemic hCoVs (43). Given our findings in mice, which were not exposed to other coronaviruses before SARS-CoV-2 infection, it is likely that the intrinsic qualities of these S2-derived epitopes also contribute to their immunodominance. In addition, broadly neutralizing betacoronavirus antibodies with S2 specificity have recently been described, suggesting the use of S2-derived epitopes as targets for a pan-coronavirus vaccine (50). However, S2-specific antibodies are generally less potent neutralizers than RBD-specific antibodies and may even demonstrate little neutralization in vitro despite providing protection in vivo (50,51). This provides a possible explanation for why we observed slightly lower neutralization potency indices of T FH -dependent antibodies enriched for S2 epitope reactivity compared with more diverse T FH -independent antibodies. Alternatively, the less focused antibody repertoire of T FH -deficient mice may better neutralize SARS-CoV-2 by targeting multiple sites of vulnerability, similar to previous findings of GC inhibition promoting a broader antibody response and enhanced heterosubtypic immunity to influenza virus infection (52).
Although long-lived, high-affinity, neutralizing antibody responses conventionally depend on T FH cells, we found that antibodies of similar quality, though not quantity, could be generated with the help of non-T FH CD4 + T cells. Therefore, multiple pathways involving different CD4 + T cell subsets likely exist to promote protective antiviral humoral immunity (48). T FH -independent responses may serve as a parallel mechanism for producing protective antibodies in settings of T FH /GC impairment, such as COVID-19-induced inflammation and advanced age (6,53). Understanding this additional axis of antiviral antibody production may therefore inform more effective vaccine design and help broaden our understanding of how T cell-dependent humoral immunity is generated.

Study design
The purpose of this study was to define the T FH -independent antibody response to respiratory viruses, including SARS-CoV-2 and influenza A virus, and SARS-CoV-2 mRNA vaccination. Experiments included intranasal infection or intramuscular vaccination. Analyses included serum antibody profiling, BCR sequencing, T cell and B cell phenotyping from medLNs and bone marrow, and lung viral burden. purchased from the Jackson Laboratory. Bcl6 fl/fl mice were crossed with Cd4 Cre mice to generate Bcl6 fl/fl Cd4 Cre mice. K18-hACE2 mice were crossed with Bcl6 fl/fl Cd4 Cre mice to generate K18-hACE2 Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice. Mice were bred in-house using mating trios to enable utilization of littermates for experiments. Mice of both sexes between 6 and 10 weeks old were used for this study. Animal use and care was approved in agreement with the Yale Animal Resource Center and Institutional Animal Care and Use Committee according to the standards set by the Animal Welfare Act.
AAV-hACE2 transduction AAV 9 expressing hACE2 from a cytomegalovirus promoter (AAV-hACE2) was purchased from Vector Biolabs (SKU AAV-200183). Mice were anesthetized by intraperitoneal injection of ketamine (50 mg/kg) and xylazine (5 mg/kg). The rostral neck was shaved and disinfected with povidone-iodine. After a 5-mm incision was made, the salivary glands were retracted and the trachea was visualized. Using a 31-gauge insulin syringe, 10 11 genomic copies of AAV-hACE2 in 50 l of phosphate-buffered saline (PBS) were injected into the trachea. The incision was closed with 3M Vetbond tissue adhesive, and mice were monitored until full recovery.

Viruses
SARS-CoV-2 P1 stock was generated by inoculating Huh7.5 cells with SARS-CoV-2 isolate USA-WA1/2020 (BEI Resources, NR-52281) at a multiplicity of infection of 0.01 for 3 days. The P1 stock was then used to inoculate Vero-E6 cells, and the supernatant was harvested after 3 days at 50% cytopathic effect. The supernatant was clarified by centrifugation (450g for 5 min), filtered through a 0.45-m filter, and stored in aliquots at −80°C. For infection of AAV-hACE2 mice, virus was concentrated by mixing one volume of cold 4× PEG-it Virus Precipitation Solution (40% w/v PEG-8000 and 1.2 M NaCl) with three volumes of viral supernatant. The mixture was incubated overnight at 4°C and then centrifuged at 1500g for 60 min at 4°C. The pelleted virus was resuspended in PBS and stored in aliquots at −80°C. Virus titer was determined by plaque assay using Vero-E6 cells (55).
Influenza virus A/PR/8/34 H1N1 (PR8) expressing the ovalbumin OT-II peptide was grown for 2.5 days at 37°C in the allantoic cavities of 10-day-old specific pathogen-free fertilized chicken eggs. Harvested virus was centrifuged at 3000g for 20 min at 4°C to remove debris and stored in aliquots at −80°C. Virus titer was determined by plaque assay on Madin-Darby canine kidney cells (56).

mRNA vaccination
Used vials of Moderna mRNA-1273 and Pfizer-BioNTech BNT162b2 mRNA vaccine were obtained from Yale Health within 6 hours of opening. All vials contained less than one full dose per vial, and no vaccines were diverted for the purpose of this study. Mice were anesthetized using 30% isoflurane and administered 1 g of either Moderna or Pfizer-BioNTech vaccine intramuscularly in 50 l of PBS. Vaccine was injected into the right hamstring muscles with a 31-gauge insulin syringe.

Enzyme-linked immunosorbent assay
Sera were incubated with a final concentration of 0.5% Triton X-100 and ribonuclease A (0.5 mg/ml) for 30 min at room temperature to inactivate potential SARS-CoV-2. SARS-CoV-2 stabilized spike glycoprotein (BEI Resources, NR-53257) (57), SARS-CoV-2 spike glycoprotein RBD (BEI Resources, NR-52946), and influenza virus A/PR/8/34 H1N1 HA protein (Sino Biological, 11684-V08H) were coated at a concentration of 2 g/ml in carbonate buffer on 96-well MaxiSorp plates (Thermo Fisher Scientific) overnight at 4°C. PR8 was inactivated with 1% Triton X-100 for 1 hour at 37°C and coated at a concentration of 20 g/ml in carbonate buffer. Plates were blocked with 1% bovine serum albumin (BSA) in PBS for 1 hour at room temperature. Serum samples were serially diluted in 1% BSA in PBS and incubated in plates for 2 hours at room temperature. Antibody isotypes were detected with anti-mouse IgG-horseradish peroxidase (HRP) (1013-05), anti-mouse IgG1-HRP (1073-05), anti-mouse IgG2b-HRP (1093-05), anti-mouse IgG2c-HRP (1077-05), or anti-mouse IgG3-HRP (1103-05) from SouthernBiotech or antimouse IgM-HRP (550588) from BD Biosciences by incubating for 1 hour at room temperature. Plates were developed with TMB Stabilized Chromogen (Thermo Fisher Scientific), stopped with 3 N hydrochloric acid, and read at 450 nm on a microplate reader. Background was determined as twice the average optical density value of blank wells. Pooled sera from mice infected with SARS-CoV-2 or PR8 were used as reference standards to calculate arbitrary units. To measure antibody affinity, serial dilutions of serum samples were plated in duplicate. After incubation of serum samples, a 10-min wash with 5.3 M urea was performed on one set of the samples. Percentage of IgG binding after urea wash was calculated by dividing the area under the curve for each sample with urea wash by that without urea wash.

Enzyme-linked immunosorbent spot
SARS-CoV-2 stabilized spike glycoprotein (BEI Resources, NR-53257) (57) was coated at a concentration of 2 g/ml in carbonate buffer on 96-well MultiScreen HTS IP Filter plates (Millipore) overnight at 4°C. Plates were blocked with complete RPMI (10% heat-inactivated FBS, 1% penicillin/streptomycin, 2 mM l-glutamine, 1 mM sodium pyruvate, 10 mM Hepes, and 55 M 2-mercaptoethanol) for 4 hours at 37°C. Bone marrow cells were isolated from the left femur + tibia of mice infected with SARS-CoV-2. Red blood cells were lysed with RBC Lysis Buffer (BioLegend) for 2 min. Cells were resuspended in complete RPMI and plated in duplicate at three dilutions (1/5, 1/10, and 1/20 of total bone marrow cells) for 20 hours at 37°C. Plates were washed six times with PBS-T (0.01% Tween 20), followed by incubation with anti-mouse IgG-alkaline phosphatase (Southern-Biotech, 1030-04) in PBS with 0.5% BSA for 2 hours at room temperature. Plates were then washed three times with PBS-T and three times with PBS. Spots were developed with the Vector Blue Substrate Kit (Vector Laboratories, SK-5300) and imaged with an ImmunoSpot analyzer (Cellular Technology Limited). Spots were counted manually by a blinded investigator. For cell sorting, medLNs were stained with the aforementioned B cell surface markers and viability dye, along with FLAG-tagged SARS-CoV-2 spike protein (2 g/ml; GenScript, Z03481) for 30 min at 4°C (58). Cells were then stained with both allophycocyanin-and phycoerythrin-conjugated anti-FLAG antibodies (BioLegend, 637309 and 637307) to double-stain spike-specific B cells. Spike-specific plasmablasts (live CD138 hi Spike + ) were sorted into 350 l of Buffer RLT Plus (Qiagen) with 1% -mercaptoethanol. RLT lysate was vortexed for 1 min, frozen on dry ice, and then stored at −80°C. Cell sorting was performed on a FACSAria II (BD Biosciences) in the BSL3 facility.

BCR library preparation
RNA from spike-specific B cells was isolated using the RNeasy Plus Micro Kit (Qiagen) following the manufacturer's instructions. BCR libraries were prepared using the NEBNext Single Cell/Low Input cDNA Synthesis & Amplification Module [New England Biolabs (NEB), E6421] and NEBNext Immune Sequencing Kit (NEB, E6330), with additional reagents provided by NEB to integrate the two kits. High-quality RNA (1 to 20 ng) with RNA integrity number ≥ 8 was used as input. Libraries were analyzed by Bioanalyzer High Sensitivity DNA assay, pooled in equal amounts with PhiX spike-in, and sequenced on an Illumina MiSeq using the V3 kit, with 325 base pairs (bp) for read 1 and 275 bp for read 2.

BCR repertoire analysis
Unique molecular identifier (UMI)-barcoded paired-end sequencing reads were processed and analyzed with the Immcantation suite (Docker image v4.3.0). Read preprocessing was performed with the presto-abseq pipeline interface for pRESTO (v0.7.0) (59) with default settings, using mouse constant region primers, isotype-specific internal constant region sequences, and template switch sequences provided by NEB. The preprocessed reads were aligned to the mouse germline V(D)J genes with the changeo-igblast pipeline interface for Change-O (v1.2.0) (60) and IgBLAST (v1.17.1). Productive sequences aligning to Ig heavy chains were retained for further analysis. Each sequence was annotated with the corresponding germline V(D)J sequence, masking N/P and D-segment regions. Spectral clustering was performed using SCOPer (v1.2.0) (61) to identify clonally related sequences, based on the level of junction region homology and the mutation profiles in the V-J segments. After clonal clustering, consensus germline sequences for each clone were reconstructed as above.
Sequences aligning to constant regions other than IgM, IgA, or IgG were filtered out from further analysis. Relative clonal proportions were calculated by tabulating the number of UMI-barcoded sequences belonging to each clonal family. Clonal proportions were visualized as pie charts, with the top 10 clones each represented individually and all other clones combined into one category. Given the wide variation in the number of clones across samples, to characterize the diversity of each BCR repertoire, 100 clones were randomly sampled from each repertoire to calculate Shannon entropy and Simpson's diversity index. This process was repeated 100 times for each sample, ultimately taking the average value across all repeats.
Downstream analysis was performed using Alakazam (v1.2.0) and SHazaM (v1.1.0) (40,41). Heavy-chain V gene usage frequencies were calculated on either the sequence level or the clone level. For the sequence-level analysis, the frequency of individual UMIbarcoded BCR sequences aligned to each V gene was calculated. For the clone-level analysis, the frequency of clones aligned to each V gene was determined, taking the consensus V gene assignment across all constituent UMI-barcoded sequences belonging to each clone. V genes that were present at ≥0.02 frequency in ≥2 repertoires were retained for visualization purposes. The most frequently used human V genes in SARS-CoV-2 specific antibodies from patients were determined by mining the CoV-AbDab database (62). Of the mouse V genes that were expressed at ≥0.02 clone-level frequency in ≥2 repertoires, those that are most similar to the top-ranked human V genes were identified with BLAST. For visualization, alignment of mouse and human V genes was performed using Clustal Omega and visualized using Jalview with the Clustal X color scheme. Lineage trees were constructed with dowser (63), building the maximum parsimony tree for all constituent BCRs in a clone. Branch lengths were scaled by the number of mutations between nodes. To facilitate qualitative comparisons, the pair of Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre mice with the greatest similarity in the total number of clones and the size of the largest clone were selected for visualization.
Sequences belonging to the same clone were then collapsed, taking the most common nucleotide at each position along the consensus sequence. Clones were annotated for antibody isotypes by majority vote. IgG isotype clones were retained for further analysis of SHM. Nucleotide mutation frequencies of each collapsed clone were calculated across the entire heavy-chain sequence (with N/P and D-segment regions masked), normalizing by the length of the input sequence. To evaluate SHM propensity, the percentage of clones with 1+ or 2+ mutations was determined in each sample and compared across experimental groups. To estimate selection pressure, the collapsed clone sequences from each experimental group (Bcl6 fl/fl or Bcl6 fl/fl Cd4 Cre ) were analyzed using BASELINe (40,41), a method that compares the observed versus expected ratios of replacement to synonymous mutations. Specifically, the observed substitution frequencies in each V segment (excluding CDR3), grouped by CDR or FWR, were compared with the expected frequencies of a reference SHM targeting model using the "focused" test statistic. As a reference model for SHM in murine heavy chains is not currently available, the HH_S5F human heavy-chain 5-mer targeting model was used for analysis, because it has been previously been shown that human heavy versus light chains have considerably different mutational patterns (64). Larger BASELINe values (corresponding to an increased ratio of replacement mutations) are consistent with positive selection.
Pseudovirus production VSV-based pseudotyped viruses were produced as previously described (55,65). Vector pCAGGS containing the SARS-CoV-2 Wuhan-Hu-1 spike glycoprotein gene was produced under HHSN272201400008C and obtained through BEI Resources (NR-52310). The sequence of the Wuhan-Hu-1 isolate spike glycoprotein is identical to that of the USA-WA1/2020 isolate. The spike sequence of the B.1.351 variant of concern was generated by introducing the following mutations: L18F, D80A, D215G, R246I, K417N, E484K, N501Y, and A701V. 293T cells were transfected with either spike plasmid, followed by inoculation with replication-deficient VSV expressing Renilla luciferase for 1 hour at 37°C (65). The virus inoculum was then removed, and cells were washed with PBS before adding medium with anti-VSV glycoprotein (8G5F11) to neutralize residual inoculum. Supernatant containing pseudovirus was collected 24 hours after inoculation, clarified by centrifugation, concentrated with Amicon Ultra Centrifugal Filter Units (100 kDa), and stored in aliquots at −80°C. Pseudoviruses were titrated in Huh7.5 cells to achieve a relative light unit signal of 600 times the cell-only control background.

Pseudovirus neutralization assay
Sera for neutralization assay were heat-inactivated for 30 min at 56°C. Sera were tested at a starting dilution of 1:50 for USA-WA1/2020 pseudovirus and 1:12.5 for B.1.351 pseudovirus, with up to eight twofold serial dilutions. Huh7.5 cells (2 × 10 4 ) were plated in each well of a 96-well plate the day before. Serial dilutions of sera were incubated with pseudovirus for 1 hour at 37°C. Growth medium was then aspirated from the cells and replaced with 50 l of serum/ virus mixture. Luciferase activity was measured at 24 hours after infection using the Renilla Luciferase Assay System (Promega). Each well of cells was lysed with 50 l of Lysis Buffer, and 15 l of cell lysate was then mixed with 15 l of Luciferase Assay reagent. Luminescence was measured on a microplate reader (BioTek Synergy). Half-maximal inhibitory concentration (IC 50 ) was calculated as previously described (66). Neutralizing titer (NT 50 ) was defined as the inverse of IC 50 .

Serum transfer
Sera from SARS-CoV-2-infected AAV-hACE2 Bcl6 fl/fl or Bcl6 fl/fl Cd4 Cre mice at 14 dpi were pooled, and the resulting levels of spike-specific IgG were measured by ELISA. Bcl6 fl/fl sera were left undiluted or diluted with naïve sera to match the spike-specific IgG titer of undiluted Bcl6 fl/fl Cd4 Cre sera. Serum samples were then mixed 1:1 with PBS, and 200 l of serum/PBS mixture was transferred intravascularly by retro-orbital injection into K18-hACE2 mice under anesthesia with 30% isoflurane.

SERA with SARS-CoV-2 linear epitope library
Serum samples for epitope profiling were inactivated with ultraviolet light (250 mJ). The serum epitope repertoire analysis (SERA) platform for next-generation sequencing (NGS)-based analysis of antibody epitope repertoires has been previously described (67). In brief, Escherichia coli were engineered with a surface display vector carrying linear peptides derived from the SARS-CoV-2 proteome (GenBank, MN908947.3), designed using oligonucleotides (Twist Bioscience) encoding peptides 12 amino acids in length and tiled with 8 amino acids overlapping. Serum samples (0.5 l each) were then diluted 1:200 in a suspension of PBS and bacteria carrying the surface display library (10 9 cells per sample with 3 × 10 5 -fold library representation) and incubated so that antibodies contained in the serum would bind to the peptides on the surface of the bacteria. After incubating with protein A/G magnetic beads and magnetically isolating bacteria that were bound to antibodies contained in the serum, plasmid DNA was purified and PCR-amplified for NGS. UMIs were applied during PCR to minimize amplification bias, designed as an 8-bp semi-random sequence (NNNNNNHH). After preprocessing and read trimming the raw sequencing data, the resulting reads were filtered by using the UMIs to remove PCR duplicates. The filtered UMI data (hereafter referred to as reads) were then aligned to the original reference of linear epitopes derived from SARS-CoV-2 and quantified.
From the raw mapped read counts for each of the 2410 linear epitopes represented in the library, we first calculated the Shannon entropy and Simpson's diversity index of each sample using the diversity function in the diverse R package. To calculate the "repertoire focusing index," we used the following formula: 1 − (H′/log 2 (R)), where H′ is Shannon entropy and R is richness, defined here as the number of unique epitopes recognized by a given sample (read count > 0). Statistical differences in these various metrics were assessed by two-tailed unpaired Welch's t test, comparing Bcl6 fl/fl and Bcl6 fl/fl Cd4 Cre conditions.
For further analysis, we normalized the raw count data using the median ratio approach implemented in DESeq2 (68). Differential enrichment analysis was performed using the Wald test in DESeq2, comparing Bcl6 fl/fl versus Bcl6 fl/fl Cd4 Cre samples. Multiple hypothesis correction was performed by the Benjamini-Hochberg method, setting a statistical significance threshold of adjusted P < 0.05. After identifying differentially enriched epitopes, the normalized counts were log 2 -transformed (hereafter referred to as log 2 normalized counts) for downstream visualization and analysis.
For converting the log 2 normalized counts into relative enrichment scores on a sample-by-sample basis, we scaled the log 2 normalized counts within each sample to z scores. In this manner, a z score of 0 would correspond to epitopes that exhibited an average level of enrichment in a given sample; a positive z score would indicate that an epitope is relatively enriched in a sample, whereas a negative z score would denote a low-scoring epitope. Where applicable, statistical significance was assessed on the within-sample z scores by two-tailed unpaired Welch's t test. For visualization purposes, we also calculated the average z scores in Bcl6 fl/fl or Bcl6 fl/fl Cd4 Cre groups.

Statistical analysis
Data analysis was performed using GraphPad Prism 9 unless otherwise indicated. Data were analyzed using one-way analysis of variance (ANOVA) with Tukey's test or Dunnett's test; Welch's t test with Bonferroni multiple hypothesis correction; Student's two-tailed, unpaired t test; Welch's two-tailed, unpaired t test; two-tailed Mann-Whitney test; or two-tailed Wilcoxon signed-rank test, as indicated. P < 0.05 was considered statistically significant.