Programmed death-1 restrains the germinal center in type 1 diabetes

Abstract

Programmed death-1 (PD-1) inhibits T and B cell function upon ligand binding. PD-1 blockade revolutionized cancer treatment, and while numerous patients respond, some develop autoimmune-like symptoms or overt autoimmunity characterized by autoantibody production. PD-1 inhibition accelerates autoimmunity in mice, but its role in regulating the germinal center (GC) reaction is controversial. To address the role of PD-1 in the GC reaction in type 1 diabetes, we used tetramers to phenotype insulin-specific CD4⁺ T and B cells in non-obese diabetic (NOD) mice. PD-1 or PD-L1 deficiency, and PD-1 but not PD-L2 blockade, unleashed insulin-specific T follicular helper CD4⁺ T cells and enhanced their survival. This was concomitant with an increase in germinal center B cells and augmented insulin autoantibody production. The effect of PD-1 blockade on the germinal center was reduced when mice were treated with a monoclonal antibody targeting the insulin peptide:MHCII complex. This work provides an explanation for autoimmune side-effects following PD-1 pathway inhibition, and suggests that targeting the self-peptide:MHCII complex might limit autoimmunity arising from checkpoint blockade.

Introduction

Programmed death-1 (PD-1) is an inhibitory receptor expressed mainly by activated T and B lymphocytes [1]. Upon binding to ligands PD-L1 and PD-L2, PD-1 recruits SHP2 phosphatase, which then dephosphorylates molecules downstream of the T cell receptor (TCR) and CD28, leading to a block in T cell effector function [2]. Chronically stimulated T cells, such as those infiltrating a tumor, or fighting a persistent viral infection, express high levels of PD-1, and have an exhausted phenotype characterized by diminished capability to produce cytokines, mediate target cell killing, and proliferate [2].

Blocking the PD-1/PD-L1 signaling pathway via monoclonal antibodies can re-invigorate these exhausted T cells, and kick-start anti-tumor immunity [2]. Exciting results from clinical trials testing the efficacy of PD-1/PD-L1 checkpoint blockade led to Food and Drug Administration approvals to treat a wide variety of tumor types [3]. However, a significant proportion of patients do not respond, and many develop immune related adverse events (irAE), including overt autoimmunity such as type 1 diabetes (T1D) [3, 4]. Interestingly, T1D occurs in 1–3% of patients receiving checkpoint therapy. Over 70% of these individuals have HLA alleles associated with T1D risk, suggesting that PD-1 may maintain islet tolerance in that subset of individuals [4]. Increased B cell clonality and increase in plasmablasts are predictive of grade 3 and 4 irAE after combined checkpoint blockade, but there are still no reliable biomarkers that can predict the development of naturally occurring or checkpoint blockade-induced autoimmunity [4]. To develop biomarkers, we must first understand the mechanism by which PD-1 maintains tolerance to self-antigens.

Autoantibody production depends on cognate interactions between CD4⁺ T and B cells in the germinal center (GC) region of the lymph node [5]. T follicular helper (T_FH) cells express PD-1, ICOS, CXCR5 and Bcl-6, and provide IL-4, IL-21 and CD40-ligand stimulation to GC B cells, thus promoting antibody affinity maturation and somatic hypermutation [5]. Increases in circulating T_FH-like cells have been reported in patients with autoimmunity, suggesting that these cells may contribute to disease [5]. T follicular regulatory (T_FR) cells express PD-1, ICOS, CXCR5, CD25, Bcl-6 and Foxp3 and suppress T_FH-B cell interactions to limit autoimmunity [6].

Given that the critical cellular players involved in the GC express PD-1 [5] and PD-L1 [7], it is likely that this pathway plays an important role in regulating CD4⁺ T cell-B cell cross-talk. Indeed, loss of PD-1/PD-L1 in C57BL/6 mice precipitates autoantibody production against ds-DNA and a lupus-like disease [8], while PD-1 deficiency in BALB/c mice leads to autoantibody production against cardiac troponin I and autoimmune cardiomyopathy [9]. Loss or blockade of PD-1 or PD-L1 in non-obese diabetic (NOD) mice accelerates T1D (reflecting irAE after checkpoint therapy), however the effect on autoantibody production is unclear [10]. Several studies showed that PD-1 deficiency or blockade impairs the outcome of the GC, resulting in fewer long-lived plasma cells [7] and lower affinity antibodies [11]. Others demonstrated that PD-1 blockade enhances antibody production [12]. In recent work, these contradicting findings were revisited, as PD-1 blockade was shown to enhance both the T_FH and T_FR CD4⁺ T cells, but their ratio determined the final outcome of the GC during foreign antigen immunization and in experimental autoimmune encephalomyelitis [13].

In this study, we investigated the role of PD-1 in regulating an antigen-specific endogenous, polyclonal GC response in pre-diabetic NOD mice to further reconcile the role for PD-1 in regulating self-antigen-specific T/B interaction and autoantibody responses. Insulin is a critical autoantigen precipitating autoimmune T1D in NOD mice [14], and most insulin-specific CD4⁺ T cells in NOD mice respond to the InsB_10-23 peptide [15, 16]. Thus, we used insulin B_10-23:MHC II tetramers to track insulin-reactive CD4⁺ T cells, and generated insulin tetramers to track cognate insulin-specific B cells. We show that these two cell populations collaborate in pre-diabetic mice to give rise to insulin autoantibodies (IAA), as previously shown with T and B cell receptor transgenic cells [17]. We demonstrate that PD-1 or PD-L1 deficiency, as well as PD-1 but not PD-L2 blockade, impacted both insulin-specific T_FH and T_FR cells, and increased their survival. Using mixed bone marrow chimeric mice, we also demonstrate that PD-1 deficient insulin-specific CD4⁺ T cells had an advantage in the GC over wild type, suggesting that T cell-intrinsic PD-1 restrains the GC. Furthermore, using an antibody that specifically disrupts TCR interactions with insulin peptide:MHCII complex, we show that the effects of PD-1 blockade on the GC are peptide:MHCII-dependent as insulin-specific B cell numbers were significantly decreased in antibody treated mice. We hypothesize that blocking self-peptide:MHCII complexes during checkpoint blockade may circumvent the autoimmune side effects of immunotherapy, while still allowing anti-tumor immunity to reboot and eradicate the tumor.

Methods

Mice

Female NOD (Taconic), BALB/cJ (The Jackson Laboratories, Bar Harbor, ME), NOD.CD45.2, NOD.PD-1KO [16], NOD.PD-L1KO [16], NOD.Tg125 [18], C57BL/6.MD4.Rag1KO [19], and C57BL/6.I-A^g7 (B6.g7) mice [20] were housed in specific pathogen-free conditions. The Institutional Animal Care and Use Committee of the University of Minnesota approved all animal experiments.

In vivo PD-1 pathway blockade

Five-week-old female NOD and B6.g7 mice were treated with 250µg anti-PD-1 (clone J43, BioXCell, West Lebanon, NH), anti-PD-L2 (TY25, BioXCell) or vehicle via intraperitoneal route every other day for a total of five injections, and harvested at indicated time points [10, 21, 22].

Insulin autoantibody ELISA

For IAA ELISA, high-bind 96 well plates were coated with 10 µg/ml human insulin (I2643, Sigma-Aldrich, St. Louis, MO) and incubated overnight at 4C. Plates were washed and blocked for 1 hour with 5% BSA in 1x TBST, washed, and incubated with diluted serum for 1 hour, and subsequently with HRP-conjugated anti-mouse IgG for 1 hour (1:2000 dilution, 405306, BioLegend, San Diego, CA). ABTS substrate (50–66-01, KPL Inc, Gaithersburg, MD) was added to each well for 10 minutes, and the plate was analyzed at 405 nm.

Insulin peptide:I-A^g7 tetramer development and validation

To detect the two major populations of insulin B_10-23-specific CD4⁺ T cells [15, 23-26], we developed I-A^g7 tetramer reagents to HLVERLYLVCGEEG (p8E) or HLVERLYLVCGGEG (p8G) peptide sequences as previously described [15, 27, 28].

B cell tetramer development and validation

To detect antigen-specific B cells, we developed insulin B cell tetramers using described methods with several modifications [19]. Briefly, human insulin (I2643, Sigma-Aldrich) or hen egg lysozyme (HEL; L6876, Sigma-Aldrich) were biotinylated using EZ-Link Sulfo-NHS-LC Biotinylation Kit (21435, Thermo Fisher Scientific, Waltham, MA) at a biotin:protein ratio of 1:1. Excess biotin was removed using 3mL SpinOUT GT-600 size exclusion column (786–721, G-Biosciences, St. Louis, MO). Western blot was performed to determine the molar amount of biotinylated protein. Streptavidin (SA)-PE (PJRS27, Prozyme) or SA-Allophycocyanin (SA-APC) (PJ27S, Prozyme) were added at a 4-fold molar excess to yield tetramers. Free unbiotinylated protein was eliminated from the tetramer solution by overnight dialysis in PBS using 20kDa cut-off Slide-A-Lyzers (66005, Thermo Fisher Scientific). Tetramer specificity was confirmed by staining single cell suspensions from pooled spleen and lymph nodes of NOD.Tg125 or C57BL/6.MD4.Rag1KO mice with insulin or HEL tetramers and pre-incubating samples with insulin prior to tetramer stain (Supplemental Fig. 1E).

Decoy tetramer development

To exclude B cells specific for the tetramer scaffold, we developed decoy reagents that encompassed SA, PE or APC, and biotin [19, 28]. To accomplish this, SA-PE and SA-APC were conjugated to Alexa Fluor (AF) 647 (A20173) or DyLight 755 (1862394) (Thermo Fisher Scientific). SA-PE-AF647 and SA-APC-DyLight 755 concentration was adjusted to 1µM, based on PE and APC absorbance at 565 and 650 nm, respectively. Both reagents were biotinylated at a 6-fold molar excess of D-biotin.

Mixed bone marrow chimera setup

Five-week-old female NOD.CD45.2 mice were lethally irradiated (550 rads twice, 4 hours apart) using RS-2000 X-Ray irradiator (Rad Source Technologies, Buford, GA). Between irradiations, bone marrow was isolated from femurs and tibias of donor NOD.CD45.1/2 and NOD.PD-1KO.CD45.1 mice. Mature T cells were depleted from the bone marrow cell preparations using biotinylated anti-CD90.2 antibody (53–2.1, Thermo Fisher Scientific) and anti-biotin magnetic beads (130–090-485, Miltenyi Biotec, Bergisch Gladbach, Germany) and LS columns (Miltenyi Biotec). Irradiated mice were reconstituted with a 50:50 mixture of WT and PD-1KO cells (10 × 10⁶ cells/mouse) injected intravenously. Antibiotic water with Polymixin B and Neomycin was provided to mice for 3 weeks. Mice were harvested for CD4⁺ T and B cell analyses 5 weeks after reconstitution.

Flow cytometry

Single cell suspensions were stained with PE- and APC-conjugated insB_10-23:I-A^g7 tetramers (p8E, p8G) and anti-CXCR5 for 1 hour at 25C in the presence of Fc block (2.4G2; BioXCell) followed by 30 minute incubation at 4C with cell surface marker antibodies, fixed and permeabilized (Tonbo Biosciences), and stained intracellularly (Supplemental Table 1) [28]. Antibodies, clones, and colors listed in Supplemental Table 1 are from BD Biosciences (San Jose, CA), BioLegend, Tonbo, and Thermo Fisher Scientific. Insulin-specific CD4⁺ T cells were defined as: singlets, live, lineage^neg (B220, CD8, CD11b, CD11c), CD4⁺, PE- and APC-tetramer⁺ cells. To detect B cells, samples were stained with 10nM decoy reagents for 10min at 25C followed by 10nM tetramers for 30min at 4C [19]. Tetramer⁺ cells were magnetically enriched from spleen and other lymph node cell suspensions (EasySep, STEMCELL Technologies), surface stained, fixed, permeabilized and stained with anti-Ig(H+L) (Supplemental Table 1). Insulin-reactive B cells were defined as: singlets, live, lineage^neg (CD11c, F4/80, Gr1, CD90.2), decoy^neg, PE- and APC-tetramer⁺ cells. Flow cytometry was performed on BD Biosciences LSRII or Fortessa cytometers and analyzed using FlowJo software (FlowJo, Ashland, OR; BD Biosciences).

Anti-peptide:MHCII antibody generation

To generate a monoclonal antibody that recognizes both insB_10-23 p8E:I-A^g7 and insB_10-23 p8G:I-A^g7 complexes, we modified a recently published method for generation of anti-peptide:MHCII antibodies [28]. Briefly, BALB/cJ mice were immunized subcutaneously with 25µg insB_10-23 p8E:I-A^g7 monomer in CFA (day 0) and boosted twice (day 28, day 42) with 25µg insB_10-23 p8G:I-A^g7 monomer in PBS intravenously. Animals were harvested on day 45, and insB_10-23 p8E and insB_10-23 p8G:I-A^g7 tetramer⁺ B cells were enriched using anti-PE magnetic beads (18557, STEMCELL Technologies) and fused with SP2/0 mouse myeloma cells using the HY Hybridoma Cloning Kit method A (03800, STEMCELL Technologies). Individual colonies were hand-picked, grown, and tested for specificity as previously described [28]. Supernatants reacting to both p8E- and insB_10-23 p8G:I-A^g7 monomers, but not others, were considered cross-reactive to insulin peptide:I-A^g7 and corresponding hybridomas were further subcloned and isotyped. Clone Ins4G8 was considered specific for both p8E- and insB_10-23 p8G:I-A^g7 and the Ig variable regions of the heavy and light chains were sequenced (U.S. Patent Application No. 15/952,965) using the SMARTer Mouse BCR profiling kit (634422, Clontech).

Antibody staining of peptide-pulsed bone marrow-derived dendritic cells

Bone marrow was isolated from femurs and humerus of donor NOD mice, and cultured in complete RPMI medium supplemented with 10% FCS, β-mercaptoethanol, HEPES, Penicllin/Streptomycin, nonessential amino acids and recombinant murine (rm)GM-CSF (200 U/mL) (415ML010, Biotechne) for 10 days. Non-adherent cells were collected by gentle pipetting, centrifuged at 300g for 5 minutes at 25C, and resuspended in 3mL complete RPMI medium, supplemented with 10ng/mL rmGM-CSF, 1µg/mL lipopolysaccharide (LPS from E. coli J5(Rc); List Biological Laboratories Inc., Campbell, CA), and 40µM peptide. After overnight culture, cells were harvested, counted, and surface stained with Ins4G8-AF647 and antibodies against: I-A^g7 (10–2.16-AF488), CD45.1 (PE Cy7, A20), CD11b (PE, M1/70, Tonbo Biosciences), CD11c (BUV395, HL3, BD Biosciences), and viability Ghost Dye (Violet 510, Tonbo Biosciences).

Immunofluorescence

Pancreata were harvested and frozen in Optimal Cutting Temperature compound (Sakura Finetek, Torrance, CA) as previously described [29, 30]. For insulitis quantification, 7µm-thick pancreas sections were stained with guinea pig anti-insulin (#A0564, Dako, Agilent, Santa Clara, CA) at 1:1000, anti-guinea pig IgG (H + L) (Cy5; #706175148, Jackson ImmunoResearch, West Grove, PA) at 1:1000, anti-CD4 (AF488; GK1.5) and anti-CD8 (PE; 53–6.7; Thermo Fisher Scientific) at 1:100. Slides were mounted with Prolong Gold anti-fade reagent with DAPI (P36935; Thermo Fisher Scientific). Images were acquired on a Leica DM6000B epifluorescent microscope (Leica Microsystems, Buffalo Grove, IL). Islet scoring was performed by a blinded investigator using the following scale: 0 – no insulitis, 1 – peri-insulitis, 2 – less than 25% of islet mass infiltrated, 3 – less than 75% of islet mass infiltrated, 4 – more than 75% of islet mass infiltrated [29].

In vivo Ins4G8 antibody validation

Ten-week-old NOD mice were immunized with 20µg peptide (p8E, p8G or p63) and 50µg LPS on day 0 via intraperitoneal route, and treated with 500µg Ins4G8 or isotype control (IgG1; clone MOPC21) on days 0 and 4. Animals were harvested on day 7 for enumeration and phenotypic analysis of antigen-specific CD4⁺ T cells from the secondary lymphoid organs.

Ins4G8 and anti-PD-1 treatment

Five-week-old female NOD mice were treated with 250µg anti-PD-1 (clone J43, BioXCell) and 250µg Ins4G8 or 250µg IgG1 isotype control (MOPC-21, BioXCell) via intraperitoneal route five times every other day, and harvested one day after the last injection [10, 21].

Statistics

Statistical analyses (paired or unpaired Student’s t-test; One-Way ANOVA with Tukey post-hoc test) were performed in Prism 7 or 8.1 (GraphPad, La Jolla, CA). P-values lower than 0.05 were considered significant. Graphs show mean ± standard error of the mean.

Results

B cell tetramers reliably detect insulin-specific B cells

To detect insulin-binding B cells from peripheral blood of T1D patients, Smith et al. used monomeric biotinylated insulin and magnetic enrichment [31]. We reasoned that a tetrameric reagent would allow enhanced detection of lower affinity autoreactive B cells as well as lower avidity GC B cells, as these typically downregulate their B cell receptor (BCR) [32]. In our approach, we also utilized decoy tetramers to exclude B cells specific for tetramer scaffold components including biotin, SA, or fluorochromes [19]. Therefore, we generated fluorescent tetrameric insulin or HEL reagents and confirmed their specificity by staining single cell suspensions from BCR transgenic C57BL/6.MD4.Rag1KO [33] and NOD.Tg125 mice [18] (Supplemental Fig. 1). Insulin-, but not HEL-tetramers stained >92% of B cells from NOD.Tg125 mice, whereas HEL- but not insulin-tetramers stained >98% of B cells from C57BL/6.MD4.Rag1KO animals (Supplemental Fig. 1A-1D). Next, we verified the specificity of tetramer-binding B cells by pre-incubating splenocytes from NOD.Tg125 with excess monomeric insulin prior to tetramer stain (Supplemental Fig. 1E). Incubation with monomeric insulin blocked insulin tetramer staining by >99% for NOD.Tg125 B cells, suggesting that the tetramer is detecting insulin-specific cells (Supplemental Fig. 1E). Priming NOD mice with insulin (25µg) in Complete Freund’s Adjuvant (CFA) led to a significant ~2-fold expansion of insulin tetramer⁺ B cells (p<0.05), and a corresponding increase in insulin autoantibody (IAA) titer, further showing reagent specificity (Supplemental Fig. 1F and 1G). Having validated the tools to track insulin-specific B cells in a polyclonal setting, we proceeded to examine the effects of PD-1/PD-L1 deprivation on this cell population and on cognate CD4⁺ T cells in the polyclonal repertoire of pre-diabetic NOD mice.

Loss of PD-1 or PD-L1 unleashes insulin-specific CD4⁺ T and B cells

NOD.PD-1KO mice had significantly more insulinB_10-23:I-A^g7 tetramer⁺ CD4⁺ T cells in the pancreatic lymph nodes (pLN) compared to age-matched NOD mice (p=0.0097), consistent with our previous observations [16] (Fig. 1A). Additionally, PD-1 deficiency resulted in an increase in insulin-specific B cells in the pLN, pancreas and other secondary lymphoid organs (spleen+non-draining lymph nodes (ndLNs); axillary, brachial, cervical, inguinal), and an increase in insulin autoantibodies (IAA) (Fig. 1B, 1C). Similarly, NOD.PD-L1KO mice had increased insulin-specific CD4⁺ T cells in the pLN (p=0.0012, Fig. 1D). PD-L1-deficient mice also had a larger population of insulin-specific B cells in the pLN, and spleen+ndLNs, but not in the pancreas (Fig. 1E). The increase in insulin-specific lymphocytes was accompanied with an increase in IAA in NOD.PD-L1KO animals (Fig. 1F). Importantly, deficiency in the PD-1/PD-L1 pathway did not affect CD4⁺ T and B cell populations specific for foreign antigens, as NOD and NOD.PD-L1KO mice had comparable numbers of HEL-specific CD4⁺ T cells [16] and HEL-specific B cells in the secondary lymphoid organs (Supplemental Fig. 1H). Taken together, these results support a model in which insulin-specific CD4⁺ T and B cell populations are held in check by the PD-1/PD-L1 pathway in autoimmune-prone mice.

Figure 1. Loss of PD-1 or PD-L1 results in increased numbers of insulin-specific CD4⁺ T and B cells, and increased insulin autoantibody (IAA) production.

(A, D) Number of insB_10-23:I-A^g7 tetramer⁺ CD4⁺ T cells in the pLN of 5-7 week old NOD, PD-1KO and PD-L1KO mice (n = 9-13 per group). (B, E) Number of insulin-tetramer⁺ B cells in the pLN, pancreas, and other secondary lymphoid organs (spl+ndLNs; spleen, inguinal, axillary, brachial, cervical lymph nodes) in NOD, PD-1KO, and PD-L1KO mice (n = 9-13 per group). (C, F) IAA ELISA OD measurements at 405nm (n=19-22). (A-F) Data are compiled from 3-4 independent experiments. Statistical significance was determined by unpaired Student’s T-test. * p <0.05, ** p < 0.01.

Loss of PD-1 or PD-L1 results in increased insulin-specific follicular T cells and GC B cells

Antibody production was shown to positively correlate with a high T_FH/T_FR CD4⁺ T ratio after foreign antigen immunization [13]. To detect bulk and insulin-specific T_FH and T_FR cells, we utilized a well-established and previously reported gating strategy [13, 34]. Follicular cells were identified as ICOS⁺CXCR5⁺. Within this gate, GITR and CD25 expression were used to identify T_FR cells (GITR⁺CD25⁺). As shown in Supplemental Fig. 1I, >95% of ICOS⁺CXCR5⁺GITR⁺CD25⁺ cells also expressed Foxp3, confirming their regulatory lineage. When examining bulk CD4⁺ T cells in the pLN, we observed >4-fold increase in the T_FH and T_FR cells in NOD.PD-1KO and NOD.PD-L1KO mice compared to age-matched NOD (Supplemental Fig.2A and 2B, respectively). Since CXCR5⁺ CD8⁺ T cells have been described and shown to participate in the GC [35-37], we tested whether these cells were impacted by PD-1 deficiency. In both NOD and PD-1KO mice, <0.3% of CD8⁺ T cells in the pLN expressed CXCR5 (Supplemental Fig. 2F). We next evaluated insulin-specific follicular CD4⁺ T cells, identified as shown in Supplemental Fig. 1I. NOD.PD-1KO mice had a 2-fold increase in insulin-specific T_FH cells and a trend towards increased T_FR cells in the pLN compared to age-matched NOD mice (Fig. 2A). Similarly, NOD.PD-L1KO mice had increased insulin-specific T_FH cells and increased T_FR cells in the pLN (Fig. 2B). The increase in insulin-specific T_FH cells in NOD.PD-1KO and NOD.PD-L1KO mice was accompanied with an increase in GC (GL7⁺) insulin-specific B cells in the pLN (Fig. 2C, 2D; p=0.034), a more modest increase in the spleen+ndLNs (Fig. 2E, 2F; p = 0.057, p = 0.084, respectively), and a trend towards increased isotype-switched insulin-specific B cells. The gating strategy for GC and switched cells is shown in Supplemental Fig. 1J. In summary, enhanced insulin-specific CD4⁺ T and B cell interactions in the absence of PD-1/PD-L1 led to increased output of isotype switched insulin-specific B cells, and likely contributed to accelerated T1D [10, 16, 22, 38].

Figure 2. Loss of PD-1 unleashes a larger anti-insulin GC reaction.

(A, B) Number of T_FH and T_FR insB_10-23:I-A^g7 tetramer⁺ CD4⁺ T cells in the pLN of 5-7-week-old NOD, PD-1KO, and PD-L1KO mice (n=9-13). (C, D) Number of GC (GL7⁺) and isotype-switched (swIg; IgM⁻IgD⁻) insulin-tetramer⁺ B cells in the pLN (n=8-13) and (E, F) spl+ndLNs in NOD, PD-1KO, and PD-L1KO mice (n=9-13). (A-F) Data are compiled from 3-4 independent experiments. Statistical significance was determined by unpaired Student’s T-test. * p <0.05, ** p < 0.01.

Cell-intrinsic loss of PD-1 favors CD4⁺ T cell skewing to the follicular phenotype

We and others have previously shown that 5-week-old NOD.PD-1KO and NOD.PD-L1KO mice have more severe islet infiltration compared to age-matched NOD mice and develop T1D more quickly [10, 16, 22, 38]. We next performed mixed bone marrow chimeras to determine if cell-intrinsic PD-1/PD-L1 inhibition contributed to the enhanced GC reaction. This was also done to exclude antigen availability and systemic inflammation as potential factors contributing to increased insulin-specific CD4⁺ T and B cells. Wild type 5-week-old NOD.CD45.2 mice were lethally irradiated and reconstituted with a 50:50 mixture of CD45.1⁺ PD-1KO and CD45.1/CD45.2⁺ WT cells (10 × 10⁶ cells total). Five weeks after reconstitution, we assessed the number of insulin-specific CD4⁺ T cells in the thymus and the pLN. We did not detect differences in the number of donor-derived insulin-specific CD4⁺ T cells in the thymus; however, in the pLN, PD-1KO cells vastly outnumbered WT cells (Fig. 3A). This was also true for the bulk CD4⁺ T cells; however, the bias towards PD-1KO cells was higher among tetramer⁺ cells (Supplemental Fig. 2D). In the pLN, PD-1KO insulin-specific T_FH and T_FR cells outcompeted WT counterparts, as most detectable follicular cells originated from the PD-1KO donor (Fig. 3B). Similarly, PD-1KO insulin-specific B cells significantly outnumbered WT cells in the pLN and other secondary lymphoid organs, but did not exhibit a strong bias towards GC or isotype-switched phenotype (Fig. 3C, 3D). Importantly, we did not observe a difference in the number of PD-1KO and WT HEL-specific B cells in the pLN (Supplemental Fig. 2E). These results indicate that PD-1 has a cell intrinsic effect on insulin-specific CD4⁺ T_FH and T_FR, and to a lesser extent on insulin-specific B cells, but not on foreign antigen HEL-specific B cells.

Figure 3. PD-1 deficient insulin-specific CD4⁺ T cells outcompete wild type cells in the GC.

Five-week-old female NOD mice were irradiated and reconstituted with a 50:50 mixture (10×10⁶ cells) of congenically disparate wild type (WT) and PD-1KO bone marrow (depleted of mature T cells) (n = 8). Mice were harvested 5 weeks after reconstitution. (A) Number of WT and PD-1KO insB_10-23:I-A^g7 tetramer⁺ CD4⁺ T cells in the thymus and pLN of chimeric mice (n = 8). (B) Number of WT and PD-1KO T_FH and T_FR insB_10-23:I-A^g7 CD4⁺ T cells in the pLN (n = 8). (C) Number of WT and PD-1KO insulin-tetramer⁺ B cells in the pLN and spl+ndLNs (n = 8). (D) Number of WT and PD-1KO GC (GL7⁺) and isotype-switched (swIg; IgM⁻IgD⁻) insulin-tetramer⁺ B cells in the pLN (n = 8). (A-D) Data are compiled from two independent experiments. Statistical significance was determined by paired Student’s T-test. * p <0.05, ** p < 0.01, *** p <0.0001.

PD-1 blockade increases insulin-specific T_FH cells, insulin-specific GC B cell numbers, and enhances IAA production in NOD mice

To better model the clinical scenario of irAE associated with checkpoint blockade, we examined the effects of PD-1 blockade on the GC reaction. Five week-old NOD mice were treated with vehicle or anti-PD-1 (J43 clone) [10, 21], and were harvested on days 4, 10, 15, or 22 after treatment initiation. Anti-PD-1-treated mice had more insulin-specific CD4⁺ T cells in the pLN at day 10 but not at earlier or later time points (Fig. 4A). We analyzed the phenotype of insulin-specific CD4⁺ T cells, and found that anti-PD-1 treated mice had significantly higher total (Supplemental Fig. 2C) and insulin-specific T_FH cells in the pLN at day 10 (Fig. 4B). Previous work has shown that the relative ratio of T_FH/T_FR subsets determines the outcome of the GC reaction [13, 34, 39]. Indeed, anti-PD-1 treatment led to an increase in the T_FH/T_FR ratio among tetramer⁺ cells on day 10 (Fig. 4C). To determine whether PD-1 blockade promoted increased T_FH cell proliferation or survival, we measured Ki67 expression (Fig. 4D), and caspase-3 activation in follicular CD4⁺ T cells on day 10 post treatment initiation. As shown in Fig. 4E, we detected significantly fewer caspase-3⁺ CD4⁺ T_FH cells in mice treated with anti-PD-1 (p=0.0218).

Figure 4. PD-1 blockade increases the number of insulin-specific CD4⁺ T and B cells, and increased insulin autoantibody (IAA) production.

(A) Number of insB_10-23:I-A^g7 tetramer⁺ CD4⁺ T cells in the pLN of control and anti-PD-1-treated mice at day 10 post treatment initiation (n=6 per group). (B) Number of T_FH and T_FR insB_10-23:I-A^g7 tetramer⁺ CD4⁺ T cells in the pLN of control and anti-PD-1-treated mice at day 10 post treatment initiation (n=6 per group). (C) Ratio of T_FH and T_FR insulin-tetramer⁺ CD4⁺ T cells in the pLNs of control and anti-PD-1-treated mice, n=6-12 per group combined from 2-3 independent experiments. (D) Percent of Ki67 positive T_FH and T_FR from pLN of control and anti-PD-1 treated mice at day 10 post treatment start (n=5 per group). (E) Percent of cleaved caspase 3 positive T_FH and T_FR from pLN of control and anti-PD-1 treated mice at day 10 post treatment start (n=5 per group). (F) Number of insulin-tetramer⁺ B cells in the pLN at day 0, 10, 15, and 22 following treatment (n=5-6 per time point). (G) Number of GC (GL7⁺) and isotype-switched (swIg; IgM⁻IgD⁻) insulin-tetramer⁺ B cells in the pLN at day 10 post treatment (n=6 per group). (H) IAA ELISA OD measurements at 405nm at day 0, 10, 15, and 20 following treatment ( n=5 per time point). (I) Insulitis score at d0, 10, 15, and 20 post injection (n=3-6 per group). (J) Number of insulin-tetramer⁺ B cells in the spl+ndLNs in control and treated mice at day 10 post treatment, n=6. (K) Number of GC (GL7⁺) and isotype-switched (swIg; IgM⁻IgD⁻) insulin-tetramer⁺ B cells in spl+ndLNs in control and treated mice at day 10 post treatment (n=6). Data are representative of two independent experiments. Statistical significance was determined by unpaired Student’s T-test. * p <0.05, ** p < 0.01, *** p <0.001.

We next determined if the increased T_FH/T_FR ratio was associated with enhanced insulin specific B cell frequency, IAA and insulitis at d10, d15, and d20/22 post anti-PD-1 injection. We detected significantly higher numbers of insulin-specific B cells in the pLN at all tested time points (Fig. 4F). This was concomitant with an increase in GC B cells (pLN; p=0.0003, Fig. 4G). Importantly, PD-1 blockade led to de novo generation of IAA as treated mice had significantly higher titers after anti-PD-1 at d10, d15, and d20 compared to baseline and compared to vehicle-treated mice (Fig. 4H). Increased IAA production was accompanied with enhanced early islet infiltration at days 10 and 15 post anti-PD-1 administration (Fig. 4I). This increased frequency of insulin-specific B cells was mirrored in the spleen and non-draining LN (Fig. 4J), and isotype-switched B cells (Fig. 4K), mirroring findings with five-week-old PD-1 and PD-L1 deficient mice. Collectively, these findings reflect clinical observations, where some patients treated with checkpoint blockade rapidly developed autoantibodies, and in some cases, T1D [40].

Since PD-1 can also bind to PD-L2 [38], and GC B cells express PD-L2 [7], we examined whether PD-L2 blockade would have the same effect on insulin-specific CD4⁺ T and B cells. Five-week-old NOD mice were treated with vehicle or anti-PD-L2 (TY25 clone) five times, and harvested one day after the last injection [41]. PD-L2 blockade did not increase the number of total or follicular insulin-specific CD4⁺ T cells (Supplemental Fig. 3A). We also did not detect a change in insulin-specific B cells, or in serum IAA (Supplemental Fig. 3B-3D).

We next examined the effects of PD-1 pathway blockade on the GC formation in diabetes-resistant C57BL/6.I-A^g7 (B6.g7) mice [41]. Five-week-old B6.g7 mice were treated with vehicle or anti-PD-1 (J43 clone) five times, and harvested one day after the last injection (day 10 post treatment initiation). As shown in Supplemental Figure 3E-3H, PD-1 blockade did not promote a significant increase in insulin-specific T_FH or T_FR cells, nor lead to an increase in IAA production, and treated mice remained diabetes-free [16]. These results suggest that the pre-existence of effector CD4⁺ T cells is required for PD-1 blockade to cause IAA and insulitis [16].

Effects of PD-1 blockade are reduced by treatment with a monoclonal antibody targeting insB_10-23:I-A^g7 complexes

IAA production in NOD mice depends on CD4⁺ T cell help [42] and specifically CD4⁺ T cell recognition of tyrosine at position 16 in the insulin B_10-23 peptide [14, 17, 43]. Blocking TCR recognition of insulin B_10-23 p8E and p8G peptide variants in the context of I-A^g7 with a monoclonal antibody reduced IAA titers and significantly delayed T1D onset in NOD mice [43]. We therefore reasoned that a similar strategy could be used to dampen the effects of PD-1 blockade on the GC reaction.

To generate a monoclonal antibody against insulin B_{10-23 p8E} and _p8G:I-A^g7, we modified our recently developed approach [28]. We immunized BALB/c mice with insB_10-23p8E:I-A^g7 in CFA (day 0) and boosted them twice (day 28 and day 42) with insB_10-23p8G:I-A^g7 intravenously. Antigen-specific B cells were enriched using insB_10-23p8E and insB_10-23p8G:I-A^g7 tetramers, and fused with SP2/0 mouse myeloma cells to create hybridomas. Out of 31 pre-screened hybridomas, 3 were specific for insB_10-23p8E:I-A^g7 and insB_10-23p8G:I-A^g7, as determined by tetramer binding. We further purified a monoclonal IgG1 antibody from one hybridoma, termed Ins4G8.

To validate Ins4G8 specificity, we pulsed bone marrow-derived dendritic cells with various peptides, and evaluated the ability of fluorescently labeled Ins4G8 to specifically bind to insB_10-23p8E and insB_10-23p8G:I-A^g7 complexes. As shown in Supplemental Fig. 3I, mean fluorescent intensity of Ins4G8 staining was at least 2-fold higher when bone marrow-derived dendritic cells were pulsed with insB_10-23p8E or insB_10-23p8G versus OVA_323–339, BDC2.5 mimetope 1040-p63, or no peptide. To validate this antibody in vivo, we immunized NOD mice with 20μg of 1040-p63, insB_10-23p8E or insB_10-23p8G peptide and 50μg LPS, and treated them twice on days 0 and 4 with 500μg Ins4G8 or isotype (IgG1; MOPC21 clone) antibody. Seven days after immunization, we enumerated cognate CD4⁺ T cells, and found that Ins4G8 treatment did not impact the activation or proliferation of p63:I-A^g7-restricted CD4⁺ T cells, but did reduce the activation , expansion and expression of CD44 and PD-1 of both subsets of insulin-specific CD4⁺ T cells (Supplemental Fig. 3J and 3K).

Given the significant effect anti-PD-1 elicited on insulin-specific B cells and IAA (Fig. 4F and 4H), we next tested the effects of Ins4G8 and anti-PD-1 combination treatment on B cells. In the pLN, anti-PD-1 treatment led to a significant increase in total and insulin-reactive B cells, while Ins4G8 alone had no effect (Fig. 5A, 5B). Importantly, mice treated with both Ins4G8 and anti-PD-1 had fewer total and insulin-reactive B cells in the pLN compared to anti-PD-1 treated mice (Fig. 5A, 5B). Mice that received combination treatment also had a reduced degree of severely infiltrated islets compared to anti-PD-1 treated mice at day 10 post treatment initiation (Fig. 5C; Supplemental Fig. 3L). However, combination treatment did not prevent the development of PD-1 blockade-induced T1D in NOD mice.

Figure 5. Blocking insB_10-23:I-A^g7 complexes reduces the effects of anti-PD-1.

(A) Number of total and (B) insulin tetramer⁺ B cells in the pLN of treated mice, n=5 per group. (C) Frequency of infiltrated islets in mice treated five times with anti-PD-1 alone, or anti-PD-1 with Ins4G8 antibody (n=4-5 per group). Islet scoring: 0 – no insulitis, 1 – peri-insulitis, 2 – less than 25% of islet mass infiltrated, 3 – less than 75% of islet mass infiltrated, 4 – more than 75% of islet mass infiltrated [29]. Data are representative of two independent experiments. One-Way ANOVA with Tukey post-hoc analysis was performed to determine statistical significance. ** p < 0.01, *** p < 0.0005, **** p <0.0001.

Discussion

The role of PD-1 in regulating the GC reaction has been a matter of intense debate. While several studies reported lower antibody production, poor antibody affinity and diversity after PD-1 blockade, others found that PD-1 deficiency or blockade augmented antibody titers in the context of immunization or infection [12, 13, 44]. With reports of rapid autoantibody development after PD-1 inhibition in the clinic, there is an urgent need to determine how PD-1 regulates self antigen-specific T_FH and T_FR CD4⁺ T cells and self-reactive cognate B cells. In this study, we used insulin peptide:MHC II tetramers to track insulin-reactive CD4⁺ T cells and developed a novel insulin tetramer reagent to track endogenous, polyclonal B cells in PD-1-sufficient and -deficient NOD mice.

In the current report we determined that PD-1/PD-L1 deficiency or blockade increased the number of total and insulin-specific T_FH and T_FR cells in the pLN. Since we did not detect changes in foreign antigen-specific CD4⁺ T or B cells, we hypothesize that the overall increase in cellularity in the pLN of PD-1/PD-L1KO mice is due to other islet antigen-specific cell populations. PD-1/PD-L1 deficiency or blockade also promoted an increase in GC B cells and de novo IAA production, reflecting clinical observations [4, 40]. We also observed an increase in insulin-specific T_FH/T_FR cell ratio after PD-1 blockade, and speculate that this accounts for the increased IAA production, similarly to what has been described previously for bulk T_FH/T_FR ratio [13]. Treating mice with anti-PD-1 allowed us to investigate the kinetics of the GC response. The GC reaction peaked at day 10 post-treatment initiation, correlating with increased insulitis and accelerated T1D onset [10]. Importantly, we did not observe changes in insulin-specific CD4⁺ T or B cells, nor an increase in IAA, following treatment with anti-PD-L2, suggesting that PD-1/PD-L1, but not PD-1/PD-L2 interactions, control the GC. Furthermore, using mixed bone marrow chimeras, we showed that PD-1 on CD4⁺ T cells, and to a much lesser extent on B cells, regulates the GC, and appears to limit CD4⁺ T cell survival. CD4⁺ T cell-intrinsic PD-1 might bind to PD-L1 on dendritic cells [34] and/or on B cells [7] and limit T cell dwell time [45], and by extension limit TCR signaling, leading to decreased cytokine production, survival, and reduced T cell help to B cells. In turn, CD4⁺ T cell-intrinsic PD-1 deficiency results in augmented T cell help, and perhaps reduces the selective pressure for affinity maturation, thus leading to greater IAA production but lower overall IAA affinity [11, 44, 46]. It is also possible that increased insulin-specific CD4⁺ T cell help licenses a small population of B cells to mutate towards autoreactivity [47, 48].

Recent work has challenged CD25 expression on T_FR cells. When compared to non-follicular T_reg cells, T_FR cells were shown to express lower CD25 [49, 50]. However, when compared to T_FH cells which are CD25^null, T_FR cells are considered CD25^pos [51]. Further adding to this complexity, there appear to be two populations of T_FR, one that is CD25⁺ and one that is CD25⁻ [51]. Our study is one of few that tracks endogenous, polyclonal self-specific T_FH and T_FR cells using peptide:MHCII tetramers [39]. To identify T_FR cells, we utilized ICOS, CXCR5, GITR, and CD25 expression, as well-established and previously reported [13, 34, 52, 53]. By enumerating ICOS⁺CXCR5⁺GITR⁺CD25⁺ cells, we likely underestimated the number of T_FR. However, in a subset of confirmatory experiments we used Foxp3 to delineate T_FH versus T_FR cells and also observed increased ICOS⁺CXCR5⁺Foxp3⁻ and ICOS⁺CXCR5⁺Foxp3⁺ cells (Supplemental Fig. 2G), thus validating our approach and the use of CD25 to delineate T_FR cells.

Previous work has shown that CXCR5⁺CD8⁺ T cells infiltrate the GC and contribute to its regulation [35-37]. This was largely documented in the context of chronic viral infection, namely HIV/SIV and EBV, where the infected cells are CD4⁺ T cells and B cells, respectively, and as many as 20% of CD8⁺ T cells in peripheral lymph nodes express CXCR5. Here, <0.3% of CD8⁺ T cells in the pLNs expressed CXCR5 (in NOD and PD-1KO mice, Supplemental Figure 2F), suggesting that CD8⁺ T cells likely do not participate in the GC, but rather contribute to beta cell death in the pancreas [54]. Autoreactive B cells, including those responding to insulin, have been shown to be polyreactive [31]. While we observed a significant expansion of insulin tetramer⁺ B cells and a corresponding increase in IAA titer after insulin immunization and after PD-1 inhibition, it is still possible that a fraction of insulin tetramer-binding cells may be polyreactive and not solely bind insulin under physiological conditions. Whether polyreactive insulin tetramer⁺ B cells are more or less pathogenic remains to be determined.

Herein, treatment with an antibody that blocks insulin peptide:MHCII complexes reduced the effects of PD-1 blockade on insulin-reactive B cell expansion, but had subtle effects on islet inflammation in the short term and, ultimately, did not impact T1D incidence. This is likely due to other islet-specific CD4⁺ T cells being unaffected by Ins4G8 treatment, or due to islet-specific CD8⁺ T cells which would be unaffected by Ins4G8 as they are not MHCII-restricted. Recent work has shown that Methyldopa has the ability to block insB_10-23 but not viral peptide docking in the context of the DQ8 molecule in T1D patients [55]. We speculate that antibodies targeting multiple self-peptide:MHCI and II complexes, compounds like Methyldopa, antigen-coupled nanoparticles [56] or antigen-coupled apoptotic splenocytes [57] could limit the autoimmune side effects of checkpoint blockade in at-risk patients, while still allowing anti-tumor immunity to reboot and eradicate tumors.

We have previously shown that PD-1 blockade in mice largely impacts effector, but not anergic or naïve insulin-specific CD4⁺ T cells [16, 30]. The majority of insulin-specific CD4⁺ T cells in B6.g7 mice are naïve; hence, it is not surprising that PD-1 blockade did not augment the GC reaction or promote T1D in this strain. Since most non-diabetic individuals also harbor naïve insulin-specific CD4⁺ T cells [58], we hypothesize that only people with pre-existing effector self-specific CD4⁺ T cells are at risk of developing autoimmunity after PD-1 blockade. Phenotyping self-specific CD4⁺ T cells prior to checkpoint blockade and screening for HLA II or anti-islet autoantibodies might be a useful prognostic tool for the development of autoimmunity, and it might identify a window of opportunity for therapeutic interventions.

Key points.

PD-1 blockade or deficiency unleashes follicular insulin-specific CD4⁺ T and B cells

PD-1 limits T follicular helper cell survival in the germinal center

PD-1 blockade increases insulin autoantibody production in diabetes-prone mice