Following intravenous infection with Salmonella, despite disruption of splenic lymphoid architecture, the number of CD4⁺ T cells inhabiting spleen B cell follicles increased (Figure S1A in Supplementary Material). As assessed by flow cytometry, these cells expressed the typical Tfh cell markers CXCR5, PD1 (Figure S1A in Supplementary Material), and ICOS (not shown). The kinetics of Tfh appearance was similar to that noted for SRBC antigens, peaking in the first week and declining slowly thereafter (Figure S1B in Supplementary Material). Interestingly, during Salmonella infection, almost all Tfh cells expressed intermediate to low levels of PD1 (PD1^lo Tfh-like cells), while the PD1^hi population seen in the response to SRBC was largely missing (Figure S1 in Supplementary Material). This is in contrast to SRBC immunization, where both populations of PD-1^lo Tfh-like cells and PD-1^hi Tfh are formed within first 7 days p.i. (Figure S1C in Supplementary Material). To investigate whether the PD1^lo Tfh-like cells generated in response to Salmonella were dependent on B cells [as previously shown for PD1^hi Tfh responses (7, 20, 21)], we depleted mice of B cells using anti-CD20 monoclonal antibody injections at different times postinfection (Figure 1). B cell depletion at day 2 and day 6 (Figure 1B) postinfection caused the complete loss of PD1^lo Tfh-like populations by day 11 (Figure 1A). Mice depleted of B cells at day 10 postinfection showed a partial loss of both Tfh populations by day 11 (Figure 1A). However, by day 16 postinfection, Tfh cells further decreased to background levels (1–2%) as detected in mice genetically deficient of B cells (μMT) (Figure 1B). These data demonstrate clearly that the PD1^lo Tfh-like cells generated after Salmonella infection are also exquisitely dependent on the presence of B cells for their continued survival (Figure 1B).

It is important to recognize that the Tfh response to a classical antigen, such as SRBC and antigen-pulsed DC, consists of both PD1^lo and PD1^hi populations at day 7 after immunization (Figure 2A, Figure S1A in Supplementary Material). The PD1^lo Tfh-like population often dominated over PD1^hi Tfh cells at relatively early time points, accounting for up to >60% of Tfh cells, especially seen in the case of antigen-pulsed DC (Figure 2A). Having established that the PD1^lo Tfh-like are, similar to PD1^hi Tfh, B cell dependent, we wished to know whether these two populations had similar differentiation requirements and whether they had a precursor–product relationship.

We first investigated whether PD1^hi and PD1^lo Tfh-like populations had similar requirements for antigen presentation by B cells. The generation of PD1^hi Tfh is initiated by antigen presentation from DC (21, 31), but a subsequent cognate interaction with B cells is required to complete Tfh differentiation (7, 15, 17, 20). This conclusion is based on the demonstration that recognition of immunizing antigen via the BCR is required for Tfh formation (20); however, a direct role for B cell antigen presentation has not been shown. To address this question, we made mixed BM chimeras (20% MHC II^−/− + 80% μMT BM into irradiated μMT recipient mice) in which the B cell compartment completely lacked expression of MHC class II (B-MHC II^−/−). To account for the fact that 20% of other lineages in these chimeras also lack MHC class II, we made control MHC II^20%−/− chimeras in which 20% of all lineages (including B cells) lacked MHC class II expression (as described in Section “Material and Methods”), as well as wild-type (B-WT) control chimeras. After reconstitution (8–10 weeks), we immunized chimeric mice in order to compare the Tfh response elicited by two antigens, SRBC and ovalbumin peptide-pulsed DC, at day 7 postimmunization. We found that the PD1^hi Tfh population failed to develop in the B-MHC II^−/− chimeras with both immunization protocols (Figures 2A,B). In contrast, the development of PD1^lo Tfh-like population was only slightly impaired in the B-MHC II^−/− chimeras compared to WT control chimeras (Figures 2A,B). This tallies with the observation that the migration of CD4 T cells in the follicles of B-MHC II^−/− chimeras was unaffected (Figure 2E), as documented previously (32). The lack of B cell antigen presentation and the subsequent loss of the PD1^hi Tfh population led directly to an inability of these mice to mount a GC reaction. No GC B cells (as detected by co-expression of GL-7 and the GC-specific transcription factor, Bcl-6) were found in the spleens of B-MHC II^−/− chimeras (Figures 2C,D) and histologically no GCs were observed in the spleen (data not shown). Thus, while the generation of PD1^hi Tfh requires cognate interactions with B cells, a large proportion of the PD1^lo Tfh-like population is generated independently of interactions involving B cell MHC class II.

To ask whether PD1^lo Tfh-like cells could differentiate into PD1^hi cells, we again used MHC II mixed-BM chimeras, in which the B cell compartment lacked MHC II. In the first set of experiments, we adoptively transferred reconstituted MHC II-sufficient (MHC II⁺) or -deficient B cells in reconstituted B-MHC II^−/− chimeras just prior to immunization with SRBC. Figures 3A,B show that transfer of a relatively small cohort of MHC II⁺ B cells could drive reconstitution of the PD1^hi Tfh compartment after SRBC immunization. The transfer of MHC II⁺ B cells into B-MHC II^−/− chimeras had no augmenting effect on the PD1^lo Tfh-like population (Figure 3B, right panel) and, as expected, transferring MHC II-deficient B cells had no effect on PD1^hi Tfh numbers. Interestingly, although all of the immunized chimeras exhibited Tfh cells expressing intermediate levels of Bcl-6, only those receiving MHC II⁺ B cells showed a Bcl-6^hi Tfh population (Figure 3C).

This result did not, however, demonstrate that the PD1^hi Tfh were derived from PD1^lo cells; it was equally possible that the transferred MHC II⁺ B cells elicited de novo generation of PD1^hi Tfh from naïve T cells. To address this, we next sorted PD1^lo Tfh-like cells from either B-WT chimeras or from B-MHC II^−/− chimeras (both CD45.1⁺) into CD45.2⁺ WT recipients 6 days postimmunization with SRBC. WT recipients were boosted 2 days after transfer and analyzed 3 days later (Figure 4A). Figure 4B shows that adoptively transferred PD1^lo Tfh-like cells isolated from either a WT or B MHC II^−/− chimera give rise to significant proportion of PD1^hi Tfh (CD45.1⁺) (Figures 4B,C). In this respect, it should be noted that the majority of cells derived from the donor PD1^lo Tfh-like cells were characterized as CXCR5-negative activated/effector T cells (Figures 4B,C). As in the previous experiment (Figure 3), the PD1^lo-derived PD1^hi Tfh cells had upregulated the expression of Bcl-6 to levels higher than found in the PD1^lo cells (Figure 4D).

Few studies have shown that the CXCR5 is not essential for the entry of T cells into follicles (7, 33); instead, interaction with non-cognate B cells via ICOS–ICOSL receptor was found required for the T cells to locate to the B cell follicles (33). To understand at what level CXCR5 was important for the differentiation of PD1^hi Tfh, we constructed mixed-BM chimeras in which the T cell compartment was completely deficient in CXCR5 (20% CXCR5^−/− + 80% CD3ε^−/− BM into irradiated CD3ε^−/− mice, T-CXCR5^−/−). In these mice, >80% of the B cells expressed CXCR5 and so B cell follicles formed normally. Control chimeras had a 20% deficit in CXCR5 expression in all cell lineages, including T cells (CXCR5^20%−/−). In a second control group, T-WT chimera, all the cells of hematopoietic compartment had WT expression of CXCR5. In the T-CXCR5^−/− chimeras, Tfh could not be identified on the basis of CXCR5 expression (Figure 5A) and so we stained CD4⁺ T cells additionally for ICOS and PD1. Figure 5A shows that, in the control chimeras, the frequency of cells expressing ICOS and PD1 corresponded well with that expressing CXCR5 and PD1, in both the PD1^hi and PD1^lo populations. In the T-CXCR5^−/− chimeras, the frequency of PD1^hi Tfh was reduced (although not to 0), while PD1^lo Tfh-like cells remained unchanged in frequency (Figure 5B). In line with the reduced number of PD1^hi Tfh, in T-CXCR5^−/− chimeras, the number of follicles containing GC was significantly reduced to less than 10% down from over 30% seen in controls (Figures 5C,D). In parallel, GC B cells (as detected by GL7 and Bcl-6 co-expression) were correspondingly reduced in T-CXCR5^−/− chimeras compared to control mice (Figure 5E). This is in agreement with previous studies which show that CXCR5 expression is not sufficient for the T cell entry to the follicle, while it is required for optimal B cell responses (7, 34, 35). Thus, PD1^hi Tfh responses were observably decreased in the absence of CXCR5, while the generation of PD1^lo Tfh-like cells was seemingly unaffected.

We wondered if the transition through the PD1^hi Tfh population was transient, after which the cells reverted to a PD1^lo phenotype. To test this, we transferred sorted PD1^hi Tfh raised in CD45.1⁺ WT chimeras after SRBC immunization into CD45.2⁺ congenic recipient mice, boosted the recipients with SRBC 48 post transfer, and assessed donor T cell subsets 3 days later (as presented in Figure 4A). After transplantation in CD45.2⁺ WT hosts, the CD45.1⁺ PD1^hi Tfh donor cells (or their progeny) gave rise to two populations: PD1^hi Tfh and CXCR5-negative activated/effector T cells (Figure 6). Very few donor cells were seen in the PD1^lo Tfh-like subset. We, therefore, concluded that the differentiation of PD1^lo Tfh-like toward PD1^hi GC–Tfh cells was mostly unidirectional and not reversible.

To gain insight into the function of the PD1^lo Tfh-like population, we returned to the Salmonella infection where the Tfh response stalls in the PD1^lo stage. The predominant response to Salmonella is IFN-γ and IFN-γ-producing effector cells start playing crucial role in elimination of pathogen 2 weeks postimmunization with Salmonella (9, 36). Therefore, we assessed the production of IFN-γ in T cell subsets following infection. As can be seen in Figure 7A, the PD1^lo Tfh-like cells made up the largest population of IFN-γ-producing T cells at day 7 postinfection (69% of IFN-γ-producers, compared to 27% of the non-Tfh cells, in the example shown). Conversely, most of the IFN-γ-producing T cells were PD1^lo CXCR5⁺ Tfh cells at this time point (Figure 7A, lowest two panels). However, this changed over time postinfection (Figure 7B). Beyond day 7, the proportion of PD1^lo Tfh-like cells making IFN-γ declined, while non-Tfh effector cells increased IFN-γ secretion totally dominating the response by day 20. Thus, a rapid, early but transient cytokine response is made by the PD1^lo Tfh-like cells. This is reminiscent of the “memory precursor” population of CD8 effector T cells defined by Kaech and colleagues (37).

To investigate the possibility that the PD1^lo Tfh-like population represented a “memory precursor effector,” we performed mRNA expression profiling using microarray on sorted PD1^hi and PD1^lo Tfh-like populations following SRBC immunization (>98% and >96% pure, respectively) and the PD1^lo Tfh-like cells arising during Salmonella infection (>95% pure). These subsets were compared to non-Tfh effectors (PD1⁺ CXCR5⁻) in the relevant response (both populations 98% pure). As an additional control, we have sorted naïve CD4⁺ T cells from unimmunized mice (CD4⁺ CD62L^hi CD44^lo, 99% pure). The full data set is available at: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE62961. We independently confirmed the microarray data by performing Q-RT-PCR to measure “effector/memory”-associated gene expression in these subsets, as illustrated in Figure 8.

The PD1^hi Tfh generated upon SRBC immunization showed a quite distinct profile from all the other “Tfh” subsets. They expressed by far the highest levels of Bcl6 mRNA and were the only population exhibiting expression of the recently described Tfh-specific transcription factor, Ascl2 (Figure 8). In the SRBC response, the PD1^hi Tfh cells were expressing significantly more IL-21 mRNA than seen in SRBC PD1^lo Tfh-like and effector cells; however, this was not the case in the Salmonella response where both effectors and PD1^lo Tfh-like transcribed this cytokine gene to comparable levels, as detected in SRBC PD1^hi Tfh cells (Figure 8). So, although the SRBC PD1^lo Tfh-like and Salmonella Tfh cells shared some characteristics with PD1^hi Tfh, our microarray analysis (Figure S2 in Supplementary Material) strongly suggests that PD1^lo and PD1^hi Tfh have potential to be functionally distinct. With regard to markers that identify memory versus effector potential, in both responses Prdm1 (BLIMP1) and Klrg1 were expressed most highly in the effector cells. Even if PD1^lo Tfh-like showed some degree of Prdm1 upregulation with respect to naïve T cells, levels remained approximately 50% lower than seen in effectors, while Klrg1 expression remained stably low in PD1^lo Tfh-like cells. Tox2, previously associated with memory potential, was most highly expressed in PD1^hi Tfh (SRBC). However, PD1^lo Tfh-like cells also expressed elevated Tox2 levels compared to effectors. Importantly, while the PD1^hi Tfh expressed little message for Il7r—indicating little potential to enter the memory pool (37, 38)—PD1^lo Tfh-like cells maintained comparatively high levels of Il7r mRNA. Thus, Salmonella and SRBC PD1^lo Tfh-like expressed genes characteristic of both memory (Tox2, Il7r) and short-lived effector cells (Klrg1, Prdm1), although the latter were expressed at much lower levels than in bona fide effector cells. A wider range of genes expressed in the sorted Tfh populations is shown as a microarray heat map in Figure S2 in Supplementary Material. Collectively, our data confirm the distinct nature of the PD1^hi Tfh population (the classical GC–Tfh) and emphasize that the PD1^lo Tfh-like cells (Salmonella and SRBC) express an mRNA signature reminiscent of memory precursors, including differential expression of Tox2, Slamf6, Id2, Bcl6 (in comparison to naïve T cells) (39), and Il7r.

C57Bl/6, CD45.1, μMT, CD3ε^−/−, CXCR5^−/−, and MHC II^−/− mice were bred and maintained at the School of Biological Sciences Animal Facility, University of Edinburgh (Edinburgh, UK). Mice were aged 6–10 weeks at the start of procedures. Experiments were covered by a Project License granted by the UK Home Office under the Animals (Scientific Procedures) Act of 1986 and approved locally by the University of Edinburgh Ethical Review Committee. Mice with a B cell-specific MHC II or T cell-specific CXCR5 deficiency were generated using the mixed-BM chimera system as described previously (18, 50). In brief, irradiated μMT or CD3ε^−/− recipients were reconstituted with a mixed inoculum of BM (20% MHC II^−/− or 20% CXCR5^−/− BM and 80% μMT or CD3ε^−/− BM, respectively). In a second control group, B-WT and T-WT chimera, all the cells of hematopoietic compartment had WT expression of MHC II or CXCR5 (20% WT + 80% μMT or 80% CD3ε^−/− BM in a μMT or CD3ε^−/− host, respectively). To control for a partial deficiency in the non-B/T cell compartment, a third control group with a 20% deficiency in all hematopoietic cells was made (designated MHC-II^20%−/− and CXCR5^20%−/−; 20% MHCII^−/−; or CXCR5^−/− BM + 80% WT BM in a μMT or CD3ε^−/− host, respectively). Chimerism was confirmed by flow cytometric analysis of B cell MHC-II expression or T cell CXCR5 expression.

The aroA attenuated strain of Salmonella enterica serovar Typhimurium (SL3261) was used for all infections (18, 51). Bacteria were grown as stationary-phase 16-h cultures in Luria-Bertani (LB) broth (Difco Laboratories, Surrey, UK). Animals were injected intravenously (i.v.) with ~1 × 10⁶ colony-forming units, diluted in PBS. Infectious dose was determined by plating bacteria onto LB agar plates and culturing overnight at 37°C. Experiments involving Salmonella were carried out according to Biosafety Level 2 requirements. Ag-pulsed DC immunizations were carried out using BM-derived dendritic cells, generated by a modified version of the procedure developed by Inaba et al. (52). Briefly, BM was harvested from femurs and tibia of C57Bl/6 mice. Approximately, 2 × 10⁶ cells were cultured in 10 ml RPMI, further supplemented with 10% FCS and 20 ng/ml granulocyte macrophage colony-stimulating factor (GM-CSF, Peprotech EC Ltd., London, UK). Plates were incubated at 37°C in humidified 5% CO₂ atmosphere. Medium was replaced with fresh GM-CSF containing medium on days 3, 6, and 8. On day 10, DC were harvested and cultured at 2 × 10⁶ cells/ml with OVA peptide (323–339) at 10 µg/ml and LPS at 1 µg/ml for 16 h. The following day, cells were harvested and extensively washed prior to transfer. Approximately, 2 × 10⁶ cells were injected i.v. in a total volume of 200 µl. SRBC immunizations were carried out as previously described (53). Briefly, SRBC were washed twice with PBS prior to immunization with 1 × 10⁹ SRBC per mouse via the intraperitoneal route.

B cells were depleted by single i.v. injection of 250 µg anti-mouse CD20 (clone 18B12) or isotype control (clone 2B8, both from Dr. Marylin Kehry, Biogen IDEC Inc.) on day 2, day 6, or day 10 of infection. Depletion was confirmed by quantifying peripheral blood B cells by flow cytometry based on CD19 and B220 expression.

Before staining, cells were washed in ice-cold PBS and stained with Live/Dead fixable aqua fluorescent stain (Invitrogen, Carlsbad, CA, USA) for 20 min at 4°C. After washing in FACS buffer (PBS with 0.05% sodium azide and 3% FCS), cells were stained with anti-CXCR5 biotin, anti-GL7 FITC, anti-ICOS PE, anti-PD1 PE-Cy7, anti-CD4 APC-Cy7, and CD19-pacific blue (BD Biosciences) in the presence of Fc receptor blocking antibody (clone 2.4G2, in-house) for a further hour at 4°C, then washed. Streptavidin-APC was added, and samples incubated for 30 mins on ice. Staining for intracellular transcription factors and/or cytokines (Bcl-6 and IFN-γ) were performed following surface staining using fix/perm buffer set (eBioscience, San Diego, CA, USA). Samples were acquired on a LSR II flow cytometer (BD Bioscience, San Diego, CA, USA) using BD FACS-Diva software, and data analyzed using FlowJo software (Tree Star Inc., San Carlos, CA, USA).

Single cell suspensions were prepared from the spleen by manual disruption in IMDM plus 2% FCS and penicillin/streptomycin. For isolation of individual Tfh populations, flow cytometry associated sorting was used. Tfh populations were isolated at day 6 post SRBC immunization from WT C57Bl/6 mice, MHC II-deficient mice (MHC II^−/−), or mixed-BM chimeras. The CD4 T cell fraction was enriched by Dynabeads (Invitrogen, Carlsbad, CA, USA) depletion of other cell populations prior to sorting; splenocytes (1 × 10⁸/ml) were labeled with anti-MHCII (clone M5114) and anti-CD8 antibodies (clone 53.6.72) prior to incubation and separation with anti-rat IgG Dynabeads (bead:cell ratio 1:2). This procedure enriched T cells to around 65%. Thereafter enriched T cells were stained using monoclonal antibodies directed against CD4, CXCR5, and PD1 necessary to identify Tfh subsets which were then sorted on a FACS ARIA II (BD Biosciences) as indicated in Figure 4. Specific gates were used to further distinguish and sort the PD1^hi and PD1^lo populations in the SRBC response as indicated. In adoptive transfer experiments (Figures 4 and 6), 0.4 × 10⁶ sorted T cells (purity >95%) per mouse were injected i.v. in 200 µl of sterile PBS (Sigma-Aldrich, St. Louis, MO, USA). For Q-RT-PCR and microarray analysis, the enriched CD4 T cell populations were FACS-sorted as above. Tfh cells sorted for RNA isolation were stored as cell pellets or in TRIzol reagent (Invitrogen) at −80°C until extraction process. B cells were isolated by magnetic separation (positive selection) with anti-CD19 MicroBeads (Miltenyi Biotec) as per manufacturer’s instruction. Isolated B cells were over 97% pure.

Spleens were placed in cryo-molds (VWR, Lutterworth, UK) in OCT-embedding medium (VWR) and stored at −80°C until required. Tissue sections of 5 µm in thickness were cut, dried, and fixed in acetone, before being stored at −20°C until use. Sections were blocked in 1% BSA for 15 min and then stained with antibodies diluted in 1% BSA. Between staining with primary and secondary antibodies, and after secondary antibody staining, slides were washed 3 times for 5 min each by immersion in PBS. After staining, cover slips were mounted using Mowiol (Hoechst, Frankfurt, Germany) containing 2.5% DABCO (Sigma-Aldrich, St. Louis, MO, USA). Slides were viewed on an Olympus BX50 microscope under reflected light fluorescence, and images captured using OpenLab software (Improvision, Walthman, MA, USA). Tfh cells were analyzed using sections stained with anti-IgM Texas-Red (Southern Biotechnology Associates, Birmingham, AL, USA) and anti-CD4 biotin (Biolegend) for 1 h followed by Streptavidin FITC (Southern Biotechnology Associates, Birmingham, AL, USA) for 30 min. Follicular size was measured using OpenLab software, and CD4⁺ cells residing within the follicles were counted. Quantification of GC was performed by staining with anti-IgM Texas-Red, anti-CD4 biotin, and PNA FITC (Vector Laboratories Inc., Burlingame, CA, USA) before Streptavidin AlexaFluor 350 (Invitrogen, Carlsbad, CA, USA) was added. Sections were measured and the number of GC per square millimeter was calculated.

Total RNA was extracted from sorted T cells stored as a cell pellet or in TriZol (Invitrogen) using RNeasy mini kit (Qiagen). For the microarray analysis, for each subset of interest, cells were sorted in three independent biological repeats. The quality of any recovered RNA was assessed using the Bioanalyser 2100 on a RNA-Nano chip (Agilent technologies), and only preparations with RNA integrity number >8 were considered acceptable for further processing. Extracted RNA was converted to cDNA and hybridized to Affymetrix Mouse Gene ST 1.1 array at ARK Genomics Technologies (The Roslin Institute, University of Edinburgh) according to manufacturer’s protocol. Arrays were scanned with a GeneChip Scanner 3000 (Affymetrix) according to manufacturer’s protocol. Data were processed using rma and limma packages in R/Bioconductor. Pairwise group comparisons were undertaken using linear modeling. Subsequently, empirical Bayesian analysis was applied, including vertical (within a given comparison) p value adjustment for multiple testing, which controls for false discovery rate, using the limma Bioconductor package. The normalized data are log2 values for the probes reliably detected on all arrays, processed using rma followed by quantile normalization.

In Q-RT-PCR experiments, from extracted RNA, cDNA was synthesized using Superscript First Strand Synthesis System (Invitrogen, Thermo Fisher Scientific). Gene expression was examined with TaqMan assays (Applied Biosystems, Thermo Fisher Scientific), and Q-PCRs were run on a LightCycler machine (Roche). Ribosomal RNA, 18S and ubiqutin-C mRNA from GeNorm kit (Primer Design) were used as reference genes to normalize mRNA levels. Differences between the experimental samples were analyzed using delta delta Ct method (54) and final mRNA expression values were calculated relative to the transcripts found in naïve T cells (CD62L^hi CD44^lo) isolated from unimmunized mice. PrimeTime Q-PCR assays (IDT) were used for the gene detection (full list of assays is available in Table S1 in Supplementary Material).

The Student’s unpaired t-test and two-way ANOVA tests were used to calculate significance values where appropriate. Throughout, p-values are illustrated as follows: *p = 0.01 to 0.05, **p = 0.001 to 0.01, ***p < 0.001, NS, not significant (p > 0.05).