Interferon-γ drives T_reg fragility to promote anti-tumor immunity

Summary

Regulatory T cells (T_regs) are a barrier to anti-tumor immunity. Neuropilin-1 (Nrp1) is required to maintain intratumoral T_reg stability and function but is dispensable for peripheral immune tolerance. T_reg-restricted Nrp1 deletion results in profound tumor resistance due to T_reg functional fragility. Thus, identifying the basis for Nrp1 dependency and the key drivers of T_reg fragility could help to improve immunotherapy for human cancer. We show that a high percentage of intratumoral NRP1⁺ T_regs correlates with poor prognosis in melanoma and head and neck squamous cell carcinoma. Using a mouse model of melanoma where Nrp1-deficient (Nrp1^−/−) and wild-type (Nrp1^+/+) T_regs can be assessed in a competitive environment, we find that a high proportion of intratumoral Nrp1^−/− T_regs produce interferon-γ (IFNγ), which drives the fragility of surrounding WT T_regs, boosts anti-tumor immunity, and facilitates tumor clearance. We also show that IFNγ-induced T_reg fragility is required for response to anti-PD1, suggesting that cancer therapies promoting T_reg fragility may be efficacious.

Graphical abstract

Introduction

Regulatory T cells (T_regs), characterized by their expression of the forkhead box transcription factor, Foxp3, are required to maintain immune homeostasis and prevent excessive tissue damage (Fontenot et al., 2005; Kim et al., 2007; Miyara and Sakaguchi, 2007; Vignali et al., 2008). Humans that lack a functional T_reg population develop a lethal autoimmune disorder, termed Immune dysregulation, Polyendocrinopathy, Enteropathy, X-linked (IPEX) syndrome, which can be recapitulated in mice through Foxp3 deletion. While T_regs are required to limit autoimmunity and maintain immune regulation, they can be deleterious in cancer through suppression of anti-tumor immunity (Chaudhry and Rudensky, 2013; Facciabene et al., 2012; Liu et al., 2016). Indeed, high numbers of T_regs and a low CD8⁺ T cell:T_reg ratio are considered poor prognostic factors for many tumor types, including melanoma, head and neck squamous cell carcinoma (HNSCC), ovarian cancer and colorectal carcinoma (Curiel et al., 2004; Drennan et al., 2013; Jacobs et al., 2012; Nishikawa and Sakaguchi, 2010; Saito et al., 2016). Although targeting intratumoral T_regs could be an effective therapeutic approach for multiple tumor types, perturbation of peripheral T_reg number or function could lead to life-threatening autoimmune or inflammatory complications. Therefore, identifying pathways that could be targeted to selectively undermine intratumoral T_regs is essential.

We have previously shown that Neuropilin-1 (Nrp1) is expressed by ∼90% of tumor infiltrating T_regs in mouse models of cancer and is critical for their function in the tumor microenvironment (Delgoffe et al., 2013). Indeed, mice with a T_reg-restricted deletion of Nrp1 are highly resistant to B16 melanoma, which is normally refractory to immune-mediated clearance, yet remarkably do not exhibit any autoimmune or inflammatory disease. Although we have previously described Nrp1-deficient T_regs as ‘unstable’, due to their loss of function (Delgoffe et al., 2013), previous studies and data included here clearly show that they retain Foxp3 expression. Thus, we now refer to this phenotype as T_reg ‘fragility’ consistent with their retention of Foxp3 expression yet loss of function ex vivo (as exhibited by loss of suppressive activity in vitro) and tumor tolerance in vivo (as exhibited by tumor growth reduction/clearance). While our previous data demonstrated the importance of Nrp1 in maintaining intratumoral T_reg function, many questions remain including the fate of these fragile T_regs and their contribution to anti-tumor immunity, the drivers of T_reg fragility, the expression, contribution, and impact of NRP1 on human intratumoral T_regs, and the broader implications for T_reg function and cancer immunotherapy.

Results

Increased NRP1 expression on T_regs in human cancer

While Nrp1 has been shown to prevent T_reg fragility in mice, its presence and role in human T_regs remains unclear. Previous studies have been controversial, with some suggesting peripheral human T_regs do not express NRP1 while other suggest that NRP1⁺ T_regs are potent suppressors (Battaglia et al., 2009; Battaglia et al., 2008; Chaudhary and Elkord, 2015; Chaudhary et al., 2014; Gao et al., 2016; Milpied et al., 2009; Piechnik et al., 2013; Tatura et al., 2015). Indeed, very few human T_regs in peripheral blood lymphocytes (PBL) from healthy donors express NRP1 (Fig. 1A and B). Remarkably, most patients with metastatic melanoma and head and neck squamous cell carcinoma (HNSCC) possessed a reasonably high percentage of intratumoral NRP1⁺ T_regs (Fig. 1A and B). This varied considerably from 3-90% in melanoma and 35-90% in HNSCC. The percentage of NRP1⁺ T_regs in PBL was also substantially enhanced. Interestingly, NRP1 expression in intratumoral T_regs appeared to correlated with poor prognosis in both melanoma and HNSCC (Fig. 1C).

Figure 1. Decreased Nrp1 expression leads to tumor regression and enhanced survival.

(A-C) Lymphocytes were harvested from PBL of healthy donors (n=8) or from PBL and TIL of Head and Neck Squamous Cell Carcinoma (HNSCC) and metastatic melanoma (3-5 experiments, n=16-23) and frozen or stained fresh. Frozen TIL and PBL were thawed and stained directly without stimulation. (D-G) Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP}, Foxp3^{DTR-GFP/DTR-GFP}, and Foxp3^DTR-GFP/+ mice were injected with B16.F10 melanoma tumor cells ID on day 0. Tumor growth was measured with digital calipers every three days. Mice were removed from study when tumor growth reached a diameter of 2cm in any direction or when necrosis was observed, and survival plots were generated (4 experiments, n=9-18). (F-G) Foxp3^{DTR-GFP/DTR-GFP}, and Foxp3^DTR-GFP/+ mice were treated with 100μg Diptheria Toxin IP every three days starting on day 7. (H) T_regs were isolated on day 12 post B16 injection from ndLN and TIL of Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice and cultured with effector T cells and APCs for 72 hours in a classical microsuppression assay. T_regs were pooled from 3 mice with 5-6 mice per group per experiment. Proliferation was measured and percent suppression was calculated as described in methods. Data represent 3-5 (A-C), 4 (D-G), or 3 (H) independent experiments. Error bars represent the mean ± SEM. Student unpaired t test (Fig. 1B, H), 2 way ANOVA (Fig. 1E, G), and Kaplan-Meier tests (Fig. 1E, G) were used (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

Nrp1^−/− T_regs block wild type T_reg function and promote anti-tumor immunity

We have previously shown that anti-Nrp1 substantially limits the growth of a B16 mouse model of human melanoma (Delgoffe et al., 2013). Given the heterogeneous nature of NRP1 expression on human tumor infiltrating T_regs, where only a proportion express NRP1, we questioned what impact Nrp1 loss on only a proportion of mouse T_regs might have on the function of the remaining wild-type (WT) counterparts, and by extension anti-tumor immunity and tumor growth. Also, as Nrp1-deficient T_regs show a reduction in suppressive function but also an increase in effector phenotype (Delgoffe et al., 2013), we questioned whether these cells had an active role in re-shaping the tumor microenvironment, or whether reduced tumor growth was instead due to reduction of a major suppressive cell population.

Foxp3 is on the X chromosome and is thus subject to X inactivation (Briggs and Reijo Pera, 2014; Galupa and Heard, 2015; Lee and Bartolomei, 2013), rendering only one allele active in each T_reg. Consequently in heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} female mice, 50% of T_regs have a Cre-mediated deletion of Nrp1, are marked with YFP [herein referred to as Nrp1^−/− T_regs] and exhibit functional fragility, and the other 50% express DTR-GFP and are WT T_regs, as they carry the Nrp1^L/L allele but not Foxp3^Cre-YFP [herein referred to as Nrp1^+/+ T_regs] (Fig. 1D). We first assessed tumor growth in heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} female mice, with Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} [all T_regs are Nrp1^−/−] and Foxp3^{Cre-YFP/Cre-YFP} [all T_regs are WT/Nrp1^+/+] female mice as controls. Strikingly, Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice exhibited dramatically reduced tumor growth, enhanced survival, and increased intratumoral lymphocyte and CD8⁺ T cell number, phenocopying Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} mice (Fig. 1E and fig. S1A) (Delgoffe et al., 2013). This occurred despite the presence of Nrp1^+/+ T_regs in similar numbers to Nrp1^−/− T_regs in the tumor (fig. S1A). A previous study had suggested that the absence of Nrp1 leads to reduced influx of T_regs into certain tumor types (Hansen et al., 2012). However, there did not seem to be a significant difference in the number of intratumoral Nrp1^−/− versus Nrp1^+/+ T_regs, even in Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice (fig. S1A). One might also argue that the Nrp1^L/LFoxp3^Cre-YFP mutation causes a basal inflammatory state that impacts the establishment of a tumor mass. To rule this out and any potential impact Nrp1 loss may have on T_reg development, migration and function, the impact of Nrp1 temporal deletion in T_regs following the establishment of B16 tumor growth was determined. Nrp1^L/LFoxp3^CreERT2, but not Foxp3^CreERT2, mice exhibited substantially reduced tumor growth following Nrp1 deletion induced by daily tamoxifen treatment on Days 7-11 (fig. S1B-C).

To rule out the possibility that the inability of Nrp1^+/+ T_regs to block anti-tumor immunity and tumor clearance was due to their reduced number in heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice, we also assessed tumor growth in heterozygous Foxp3^DTR-GFP/+ female mice in which diphtheria toxin (DT) treatment reduced peripheral and intratumoral T_reg number by approximately half (Fig. 1F and fig. S1D). In stark contrast to tumor growth in heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice, DT-treated heterozygous Foxp3^DTR-GFP/+ mice exhibited tumor growth that was indistinguishable from the untreated control mice (Fig. 1G). While Foxp3^{DTR-GFP/DTR-GFP} mice treated with DT largely cleared their tumors, they ultimately succumbed to autoimmunity due to an absence of T_regs in contrast to heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} and Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} female mice that never exhibited an autoimmune or inflammatory phenotype (Fig. 1D and 1E, and data not shown) (Delgoffe et al., 2013). Taken together, our data suggest that if half the T_regs are depleted tumors grow unrestrained, as the reduced number of WT T_regs are still capable of blocking anti-tumor immunity. In contrast, if half the T_regs lose Nrp1, tumors are controlled, suggesting that Nrp1^−/− T_regs are playing an active role in re-shaping the tumor microenvironment.

It is possible that Nrp1^−/− T_regs impact the tumor microenvironment by losing Foxp3 and becoming so-called ex-T_regs with an altered functional phenotype (Overacre and Vignali, 2016; Zhou et al., 2009b). Indeed, we have previously shown that Nrp1 contributes to T_reg stability in the tumor microenvironment by undermining the pAKT:FOXO pathway and preventing the expression of T helper lineage-defining transcription factors, raising the possibility that this may also result in the loss of Foxp3 expression. We analyzed Foxp3 fate mapping mice in which T_regs either possessed (Foxp3^Cre-YFPRosa^L-Tom-L-GFP) or lacked Nrp1 expression (Nrp1^L/LFoxp3^Cre-YFPRosa^L-Tom-L-GFP) (fig. S2A). In this mouse model, non-T_regs are Tomato⁺GFP⁻YFP⁻, Foxp3⁺ T_regs are Tomato⁻GFP⁺YFP⁺ and Foxp3⁻ ex-T_regs are Tomato⁻GFP⁺YFP⁻. Interestingly, there were very few (<5%) ex-T_regs present in the periphery or in the tumor, regardless of Nrp1 expression (fig. S2B and C). These data suggest that the absence of Nrp1 does not affect Foxp3 expression and does not result in the generation of ex-T_regs.

Previous reports have shown that T_regs can display alternative functions in vivo while maintaining Foxp3 expression (Hori, 2014; Sharma et al., 2013). We hypothesized that Nrp1^−/− T_regs not only lost their suppressive activity but may also negatively impact the function of surrounding intratumoral Nrp1^+/+ T_regs. In order to assess this possibility, we determined the suppressive capacity of Nrp1^+/+ (GFP⁺) and Nrp1^−/− (YFP⁺) T_regs from Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} heterozygous and control mice using a microsuppression assay (Turnis et al., 2016). While all T_reg populations isolated from non-draining lymph nodes (ndLN) were equally capable of suppressing effector T cells (Fig. 1H), Nrp1^−/− T_regs isolated from homozygous Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} and heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} tumors lacked suppressive activity (Fig. 1H). Interestingly, intratumoral Nrp1^+/+ T_regs isolated from heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice were also unable to suppress. Two hallmarks of T_reg fragility are elevated pAkt and reduced ICOS expression (Delgoffe et al., 2013). Elevated pAkt and reduced ICOS were observed in Nrp1^−/− T_regs in Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} mice and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} heterozygous mice relative to Nrp1^+/+ T_regs in Foxp3^{Cre-YFP/Cre-YFP} mice as expected (fig. S2D and E). However, elevated pAkt and, to a lesser extent, reduced ICOS was also observed in Nrp1^+/+ T_regs in Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} heterozygous mice. Taken together, these data suggest that Nrp1^−/− T_regs have a negative impact on the suppressive function of intratumoral Nrp1^+/+ T_regs.

Fragile and wild type T_regs have a reciprocal impact on their transcriptome

We next used transcriptomic analysis of Nrp1^+/+ and Nrp1^−/− T_regs from Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice to evaluate the cell intrinsic and extrinsic impact of Nrp1 loss (fig. S3). Significant alterations in the T_reg transcriptome were observed between Nrp1^+/+ and Nrp1^−/− T_regs. Principal Component Analysis (PCA) based on differentially expressed genes clearly separated T_eff and T_regs based on both location and genotype. Interestingly, intratumoral Nrp1^+/+ and Nrp1^−/− T_regs from heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice were similar to each other and yet distinct from their genotypically identical counterparts from control Foxp3^{Cre-YFP/Cre-YFP} and Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} mice (Fig. 2A, fig. S4A-B). A similar impact was observed when T_reg signature genes were assessed wherein it became evident that all four populations were district and yet bore a transcriptional relationship (fig. S4C). These data suggest that Nrp1^+/+ and Nrp1^−/− T_regs in heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice impact each other's transcriptome in a reciprocal manner.

Figure 2. Nrp1 alters the T_reg transcriptome.

(A-C) T_regs were purified based on CD4⁺, and GFP or YFP expression from Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice on D12, cDNA and libraries were generated using the Clontech SmartER Ultra-Low and Illumina Nextera XT Library Prep kits. Samples were normalized to 2nM and sequenced on a NextSeq500. (A) Differentially expressed genes are determined by the genes that have q-value of 0.2 between any two of the four T_reg groups in the TIL. PCA was computed using the “prcomp()” R functions using the normalized voom data restricted to the same differentially expressed genes as shown in figure. (B-C) Significant genes were cross-referenced with those that were annotated to “plasma membrane” or “extracellular part” in the Cellular Component Gene Ontology. The Gene Ontology annotations were obtained from mSigDB. A number of genes associated with the Ifng/Il12/Il18 pathways were upregulated in the Nrp1^−/− samples. Data represent 5 independent experiments with 3-5 mice pooled per experiment.

Pathway analysis highlighted a potential role for IFNγ/IL-12-related transcriptional programs. Of particular interest was an increase in Ifng and its targets in Nrp1^−/− T_regs in Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} as well as in Nrp1^+/+ T_regs in Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice (Fig. 2B-C, fig. S4D), implicating a role for the IFNγ pathway in modulating T_reg function and function in the tumor microenvironment.

While previous reports have suggested that a small subset of T_regs produce IFNγ during inflammation (Duhen et al., 2012; Koenecke et al., 2012; Pandiyan and Zhu, 2015), the expression of IFNγ by T_regs in tumors and its impact on their suppressive function remains unclear. Using flow cytometry, we found that there was increased expression of IFNγ by Nrp1^−/− T_regs in both Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice (Fig. 3A-B). Interestingly, an increased percentage of Nrp1^+/+ T_regs from Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice also expressed IFNγ. Interferon-γ receptor 1 (IFNγR) expression showed an elevated trend in Nrp1^−/− T_regs in Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} mice, and both Nrp1^−/− and Nrp1^+/+ T_regs in Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} heterozygous mice (Fig. 3B). In addition, several type 1 helper T cell markers were upregulated in Nrp1^−/− T_regs, such as Tbet, Cxcr3, and Eomes (fig. S5A–F) as well as downstream pSTAT1 in Nrp1^−/− T_regs from Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice and pSTAT4 in all Nrp1^−/− T_regs (fig. S5G, H).

Figure 3. Nrp1^−/− T_regs display increased IFNγ in the tumor microenvironment.

(A-B) Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice were injected with B16.F10 melanoma tumor cells ID on day 0 and sacrificed on day 12. Lymphocytes were isolated from ndLN and TIL of mice noted, stimulated and stained for IFNγ and IFNγR. (n=8-18). (C-E) C57BL/6 mice were injected with B16.F10 melanoma tumor cells ID on day 0. Mice were treated with either Sema4aIg or IgG1 every 3 days until sacrifice starting on day 5 (400ug, 200ug, 200ug, 200ug). (C) Tumors were measured on day 12 for prior to sacrifice (n=10-25). (D) Lymphocytes were isolated from ndLN and TIL, stimulated and stained for IFNγ (n=5-13). (E) Lymphocytes were isolated from ndLN and TIL, and stained for IFNγR (n=5). Data represent 3-4 independent experiments. Student unpaired t test was used. (*p<0.05, **p < 0.01, ***p <0.001, ****p < 0.0001).

In order to determine whether modulation of IFNγ and IFNγR expression also occurred following blockade of the Nrp1:Sema4a axis, we treated B16 tumor-bearing mice with Sema4aIg (Delgoffe et al., 2013). Indeed, Sema4aIg treatment decreased tumor size, and led to an unstable T_reg phenotype as assessed by increased IFNγ production and higher IFNγR expression (Fig. 3C-E). Surprisingly, we found that the majority of the IFNγ⁺ cells in the TIL of Nrp1^L/LFoxp3^Cre-YFP and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} were T_regs (Fig. 4A) and that the total percentage of IFNγ⁺ cells in TIL was small in the absence of T_reg-restricted Nrp1 deletion (Fig. 4B), raising the possibility that IFNγ production may be a dominant feature of T_reg fragility and thus could be affecting surrounding cells in the tumor microenvironment including Nrp1^+/+ T_regs.

Figure 4. Hypoxia sensitizes intratumoral T_regs to IFNγ-mediated fragility.

(A-B) Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice were injected with B16.F10 melanoma tumor cells ID on day 0 and sacrificed on day 12. Lymphocytes were isolated from TIL of mice noted, stimulated and stained for IFNγ (n=5). (C-D) Mice were treated with anti-CD8 or isotype (200ug) every 3 days starting on day 5. Tumor size was measured on day of sacrifice (D12), lymphocytes were isolated from TIL of mice noted, stimulated, and stained for IFNγ (n=7-10). (E) Lymphocytes were isolated from mice noted and used in a microsuppression assay. (F) T_regs were purified, processed, and analyzed as in Fig. 2 (n=5). Heatmap includes genes previously shown to be positive or negative regulators in the Hif1α pathway. Pathway analysis includes all genes in pathway. (G) Foxp3^{Cre-YFP/Cre-YFP} and Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} mice were injected with B16.F10 melanoma tumor cells ID on day 0 and sacrificed on day 12. Lymphocytes were isolated from ndLN and TIL and stained for Hif1α (n=10). (H) T_regs were isolated from LN of Foxp3^{Cre-YFP/Cre-YFP} mice, stimulated for 3 days in hypoxia or normoxia, and stained (n=4-6). Data represent 2-5 independent experiments. Student unpaired t test was used. (*p<0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

While our data show that T_regs were the predominant source of intratumoral IFNγ in Nrp1^L/LFoxp3^Cre-YFP and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice, we could not rule out the possibility that IFNγ production by fragile T_regs was initially triggered, or potentiated by, the altered tumor microenvironment. Indeed, tumor size is greatly reduced and lymphocyte infiltration increased in Nrp1^L/LFoxp3^Cre-YFP and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice (Fig. 1E and fig. S1A). In order to address this possibility, we blocked the anti-tumor immune response and prevented tumor shrinkage by depleting CD8⁺ T cells (using anti-CD8) in B16 tumor bearing mice, and assessed T_reg phenotype and function. We found that CD8⁺ T cell depletion had no effect on the suppressive capacity in Nrp1^L/LFoxp3^Cre-YFP and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice. Indeed, Nrp1^−/− T_regs exhibited increased IFNγ and Tbet protein expression along with reduced in suppressive capacity in an in vitro suppression assay, suggesting that T_reg fragility due to Nrp1 loss is primarily due to a cell intrinsic mechanism rather than an extrinsic environmental effect due to increased CD8⁺ T cell infiltration, the ensuing anti-tumor response, and tumor size (Fig. 4C-E and fig. S5I). However, a role for CD8⁺ T cell-derived IFNγ in promoting T_reg fragility in this system cannot be ruled out.

Although cell-intrinsic processes downstream of Nrp1 loss appeared to drive T_reg fragility, it is still possible that cell-extrinsic, environmental factors facilitated intratumoral T_reg fragility. Further analysis of our RNAseq data highlighted enhanced expression of Hif1a (hypoxia-inducible factor 1 alpha) in Nrp1^−/− compared to Nrp1^+/+ intratumoral T_regs (Fig. 4F). Indeed, the percentage of Hif1α⁺ Nrp1^−/− intratumoral T_regs and Hif1α protein expression within those T_regs was higher than their Nrp1^+/+ counterparts (Fig. 4G). Interestingly, Hif1α has been shown to be upregulated by Akt signaling which in turn led to increased IFN-γ production by T_regs (Dang et al., 2011; Lee et al., 2015). As Hif1α is upregulated in hypoxic conditions, we wondered whether hypoxia was capable of inducing T_reg fragility, analogous to the environment in which intratumoral T_regs reside. Remarkably, LN-derived T_regs from a naïve mouse showed increased IFN-γ and Tbet expression, and an elevated trend in Hif1α expression while retaining Foxp3 expression after being cultured for 3 days in hypoxic versus normoxic conditions (Fig. 4H and fig. S5J and K). These data suggest that the hypoxia:Hif1α axis may prime T_regs to become functionally fragile in the tumor microenvironment.

IFNγ is required and sufficient to drive intratumoral T_reg fragility

In order to test whether Nrp1^−/− T_regs could directly impact the function of Nrp1^+/+ T_regs, and if this was mediated by IFNγ, we co-cultured ndLN- or tumor-derived Nrp1^+/+ T_regs from either Thy1.1 Foxp3^Cre-YFP or Ifngr1^−/−Foxp3^Cre-YFP mice (T_regs which lack the IFNγ receptor) with Nrp1^−/− T_regs from Thy1.2 Nrp1^L/LFoxp3^Cre-YFP mice prior to assessing their regulatory activity in a microsuppression assay (Fig. 5A). All cell populations were also cultured alone under the same conditions as controls, and as expected, T_reg populations cultured alone exhibited the expected suppressive capacity (Fig. 1H compared with Fig. 5B, left side columns). Note that APCs were not included in the 72h pre-culture prior to the T_reg microsuppression assay. Interestingly, TIL-derived Nrp1^+/+ T_regs that were co-cultured with tumor-derived Nrp1^−/− T_regs lost their ability to suppress effector T cells, in contrast to ndLN-derived T_regs (Fig. 5B, right side columns). Loss of suppressive activity did not require cell-cell contact, but was dependent on IFNγR expression (Fig. 5B-D). In order to confirm that IFNγ was the sole cytokine responsible for Nrp1^+/+ T_reg fragility following co-culture with Nrp1^−/− T_regs, we co-cultured Nrp1^+/+ and Nrp1^−/− T_regs in the presence of different concentrations of anti-IFNγ for 72 hours, and then assessed the suppressive capacity of the purified Nrp1^+/+ T_regs in the absence of anti-IFNγ. IFNγ neutralization prevented the loss of tumor-derived Nrp1^+/+ T_reg suppression in a dose-dependent manner (Fig. 5E and fig. S6A). The increased sensitivity of tumor- versus ndLN-derived Nrp1^+/+ T_regs to Nrp1^−/− T_regs appeared to correlate with IFNγR expression (fig. S6B).

Figure 5. IFNγ reduces T_reg suppression.

(A, B) T_regs were isolated from ndLN and TIL of mice noted, stimulated with PMA and Ionomycin, cultured with IL-2 alone or with IL-2 and Nrp1^−/− T_regs for 72 hours and used in a microsuppression assay in absence of cytokine (n=6-7). (C, D) T_regs were isolated and stimulated as in (A), cultured in the bottom of a transwell plate with IL-2 alone or with IL-2 and Nrp1^−/− T_regs in the top well for 72 hours and used in a microsuppression assay in the absence of cytokine (n=6-7). (E) T_regs were isolated from ndLN and TIL of Foxp3^{Cre-YFP/Cre-YFP} mice, co-cultured with Nrp1^−/− T_regs and IL-2 in the presence or absence of anti-IFNγ, re-sorted and used in a microsuppression assay in the absence of cytokine (n=6). (F) T_regs were isolated from ndLN and TIL of Foxp3^{Cre-YFP/Cre-YFP} mice, treated with IL-2 and IFNγ for 72 hours, re-sorted and used in a microsuppression assay in the absence of cytokine (n=6). (G) T_regs were isolated from HNSCC PBL and TIL, cultured with IL-2 +/- IFNγ for 3 days, then used in a microsuppression assay in the absence of cytokine (n=2-14). Data represent 3-5 experiments. Error bars represent the mean ± SEM. Student unpaired t test (A-D) and 2 Way Anova (E-G) were used. (*p<0.05, **p <0.01, ****p <0.0001).

While previous studies have suggested that IL-12 can impact T_reg suppression and induce IFNγ expression (Koch et al., 2012; Zhao et al., 2012), whether IFNγ has a direct effect on T_regs and if so what impact that might have in their function remains obscure. In order to determine whether IFNγ was sufficient to limit suppressive capacity, we treated Nrp1^+/+ T_regs from ndLN and TIL with IFNγ for 72 hours plus IL-2 in stimulating conditions prior to assessing their functional capacity in a microsuppression assay in the absence of cytokine. IFNγ substantially limited the suppressive capacity of TIL-derived, and to a lesser extent ndLN-derived, Nrp1^+/+ T_regs in a dose-dependent manner (Fig. 5F). This effect was lost if Ifngr1^−/− T_regs were used (fig. S6C). Pre-treatment with IFNγ also induced IFNγ expression by WT T_regs but not Ifngr1^−/− T_regs (fig. S6D, and data not shown). Given that IFNγ limits the function of murine T_regs, we asked whether IFNγ could also impact human T_regs and whether this was enhanced by the human tumor microenvironment. Indeed, intratumoral human T_regs showed an increased sensitivity to IFNγ in comparison to PBL T_regs when cultured with the cytokine 72 hours prior to assessment of their suppressive capacity in the absence of cytokine (Fig. 5G). Furthermore, intratumoral NRP1⁺ T_regs appeared to be less sensitive to the effects of IFNγ than NRP1⁻ T_regs. Taken together, these data suggest that IFNγ can undermine the function of murine and human T_regs in vitro.

We next sought to determine if Nrp1^−/− T_reg-derived IFNγ from could drive Nrp1^+/+ T_reg fragility in vivo. To address this, we used Foxp3^−/− mice that lack T_regs and succumb to a scurfy-like phenotype if T_regs are not adoptively transferred within 48 hours of birth (Workman et al., 2011). T_regs of a single genotype or a 50:50 mixture of two different T_reg genotypes were adoptively transferred into two-day-old Foxp3^−/− mice. At 4 weeks of age, mice were injected with B16 melanoma and tumor growth assessed over time (Fig. 6A). Mice were monitored after T_reg injection and removed from study prior to B16 injection if any signs of autoimmunity were observed (Workman et al., 2011). Tumor growth in Foxp3^−/− mice that received either Nrp1^+/+ T_regs, Nrp1^−/− T_regs or a 50:50 mixture of Nrp1^+/+ and Nrp1^−/− T_regs phenocopied tumor growth in Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice, respectively (Fig. 1E and 6B-D and fig. S6E). We then assessed the impact of T_reg-derived IFNγ on tumor growth by transferring either [a] Nrp1^+/+ T_regs that cannot respond to IFNγ with Nrp1^−/− T_regs (50% Nrp1^+/+Ifngr1^−/− + 50% Nrp1^−/−Ifngr1^+/+), or [b] Nrp1^−/− T_regs that cannot produce IFNγ with Nrp1^+/+ T_regs (50% Nrp1^+/+Ifng^+/+ + 50% Nrp1^−/−Ifng^−/−). Strikingly, tumor growth was completely restored with either combination (Fig. 6E-H), revealing a critical role for IFNγ produced by fragile Nrp1^−/− T_regs in mediating Nrp1^+/+ T_reg dysfunction, and thereby facilitating anti-tumor immunity and limiting tumor growth.

Figure 6. IFNγ uptake by T_regs is required for T_reg fragility and tumor clearance.

(A-G) Foxp3^−/− mice were injected with 10⁶ T_regs on day 2 post-birth, and monitored for 28 days for the onset of any autoimmune symptoms [4 of 34 mice were removed from study], no more than 1 per experimental group. B16.F10 was injected ID on day 28 and tumor size was measured every 3 days. (H) Statistics of Foxp3^−/− mice tumor growth. Data represent 5-7 independent experiments with 5-7 mice per experimental group. Error bars represent the mean ± SEM. 2 way ANOVA was used. (ns, not significant, ***p < 0.001, ****p < 0.0001).

IFN-γ-induced T_reg fragility is required for effective PD1-targeted immunotherapy

While our previous data suggested a prominent role for IFN-γ in driving T_reg fragility, the importance for this observation in the broader context of an immunotherapeutic response is unknown. We sought to address this question using mice that lack IFN-γR on T_regs (Ifngr1^L/LFoxp3^Cre-YFP) (fig. S7A and B). Ifngr1^L/LFoxp3^Cre-YFP or Foxp3^Cre-YFP mice were injected with 5×10⁵ MC38 (an anti-PD1 sensitive tumor cell line) subcutaneously (SC) and then treated with either anti-PD1 or Armenian Hamster IgG control (200ug) on Days 6, 9 and 12 post-tumor injection. Strikingly, Ifngr1^L/LFoxp3^Cre-YFP mice were completely resistant to PD1 blockade in comparison to Foxp3^Cre-YFP mice, as exhibited by tumor growth and survival (Fig. 7A and fig. S7C). Consistent with the loss of IFN-γR expression preventing the development of T_reg fragility, no increase in percentage of IFNγ⁺ T_regs was observed in Ifngr1^L/LFoxp3^Cre-YFP mice in contrast to Foxp3^Cre-YFP mice following anti-PD1 treatment (Fig. 7B). Taken together, these data suggest that IFN-γ-induced T_reg fragility is required for an effective response to PD1-targeted immunotherapy.

Figure 7. IFNγ-mediated T_reg fragility is required for antiPD1 response.

(A-B) Foxp3^Cre-YFP and Ifngr1^L/LFoxp3^Cre-YFP mice were injected with MC38 SC on day 0 and treated with either anti-PD1 or isotype on days 6, 9, and 12 (200ug, 200ug, 200ug). (A) Tumor growth was measured with digital calipers every three days. Mice were removed from study when tumor growth reached a diameter of 2cm in any direction or when necrosis was observed, and survival plots were generated. (B) Lymphocytes were isolated from TIL on day 12 from Foxp3^Cre-YFP and Ifngr1^L/LFoxp3^Cre-YFP mice and were stimulated and stained for IFNγ. Data represent 2 independent experiments with 4-11 mice per experimental group. 2 way ANOVA (Fig. 7A), Kaplan-Meier test (Fig. 7A), and Student unpaired t test (Fig. 7B) were used (**p < 0.01, ***p<0.001, ****p < 0.0001).

Discussion

In summary, our data highlight seven key observations. [i] A high proportion of human T_regs expressed NRP1 in two tumor types: melanoma and HNSCC. It is also noteworthy that PBL T_regs from these cancer patients also possessed a clear population of NRP1⁺ T_regs in contrast to healthy donor PBL T_regs, which exhibited little to no NRP1 expression. Interestingly, the percentage of intratumoral NRP1⁺ T_regs appeared to correlate with poor disease prognosis. [ii] B16 tumors were rapidly cleared in mice harboring a 50:50 mixture of Nrp1-deficient and WT T_regs due to increased functional fragility and loss of suppressive activity of both T_reg populations without loss of Foxp3 expression. This was the result of Nrp1-deficient T_reg fragility rather than the generation of Foxp3⁻ ex-T_regs. [iii] T_reg fragility had a reciprocal impact on the transcriptomes of Nrp1-deficient and WT T_regs, highlighting the previously unappreciated fact that T_regs can impact other T_reg populations directly as well as many other cell types. [iv] The induction of IFNγ pathway genes was a dominant feature of T_reg fragility in tumors. Intratumoral T_regs were more susceptible to this functionally fragile phenotype due to the hypoxic tumor microenvironment, which led to increased Hif1α expression and IFNγ production (Lee et al., 2015). While it is possible that IFNγ derived from other sources could lead to T_reg fragility, we have shown that hypoxia promoted IFNγ production and T_reg fragility. [v] IFNγ, exogenously-provided (human or mouse) or intratumoral Nrp1^−/− T_reg-derived (mouse), was capable of driving the fragility of tumor-derived WT T_regs and loss of mouse and human T_reg suppressive activity in vitro. This was a direct effect of IFNγ or Nrp1^−/− T_reg-derived IFNγ as no other cells types were included in the in vitro experiments. Previous studies have suggested that IL12 can impact T_reg suppression and induce IFNγ expression (Dominguez-Villar et al., 2011; Koch et al., 2012; Zhao et al., 2012), while others show that IFNγ can limit T_reg expansion (Deligne et al., 2015; Olalekan et al., 2015; St Rose et al., 2013; Visperas et al., 2014). However, the direct effect of IFNγ on T_reg function in vivo had surprisingly not been appreciated. While some studies have shown that IFNγ⁺ T_regs can maintain suppressive function, this seems to be largely disease specific and has not been carefully assessed in the context of the tumor microenvironment (Oldenhove et al., 2009; Zhao et al., 2011). It is possible that the tumor microenvironment plays a critical role in driving IFNγ-mediated T_reg fragility, as suggested by the role of Hif1α in this process. In addition, while we anticipate that IL-12 may be playing a role in this process given that we see increased pSTAT4 expression, we have shown that IFNγ is capable of driving T_reg fragility both in vitro and in vivo in tumor-derived T_regs exposed to a hypoxic environment. (Figs. 5E-F and 6), suggesting that IL-12 may not be essential. [vi] Intratumoral T_reg fragility was mediated by IFNγ derived from Nrp1^−/− T_regs that acted on WT intratumoral T_regs, thereby leading to their fragility and loss of suppressive activity. This was supported by the fact that inclusion of WT T_regs that could not respond to IFNγ or Nrp1^−/− T_regs that could not produce IFNγ restored T_reg function, block anti-tumor immunity, and promote tumor growth. While these data suggest that Nrp1^−/− T_reg-derived IFNγ is required, we do not yet know if it is sufficient. These observations were consistent with a model in which fragile Nrp1^−/− T_regs produce large amounts of IFNγ in the tumor microenvironment that directly promotes the fragility of intratumoral WT T_regs without loss of Foxp3 expression. Importantly, this occurred without any detectable peripheral T_reg fragility and without impacting their maintenance of peripheral tolerance, suggesting that this was a proximally- and locally-driven event, likely induced by inflammation. While we only observed this in the context of the tumor microenvironment, it is possible that T_reg fragility could indeed occur in other inflammatory settings where exposure to IFNγ is increased. We would argue that the mechanism of IFNγ-induced T_reg fragility is mediated directly between Nrp1^−/− and WT T_regs as either loss of IFNγ or IFNγR expression, respectively, impacts fragility. While it is possible that IFNγ derived from other cell populations, such as CD8⁺ T cells or NK cells, or the ensuing anti-tumor response and altered tumor microenvironment contributed to T_reg fragility, it is noteworthy that the dominant IFNγ-producing cell type was Nrp1^−/− T_regs and CD8⁺ T cell depletion did not impact the enhanced IFNγ production and loss of suppressive activity observed. [vii] IFNγR expression on intratumoral T_regs was required for an effective response to PD1 blockade. Strikingly, Ifngr1^L/LFoxp3^Cre-YFP mice were completely resistant to anti-PD1 immunotherapy. Whereas WT T_regs showed a significant increase in IFNγ expression after PD1 blockade, Ifngr1^−/− T_regs showed no increase in IFNγ, suggesting that IFNγ-driven T_reg fragility may need to be induced for an effective immunotherapeutic response.

A previous study had suggested that the absence of Nrp1 leads to reduced influx of T_regs into certain tumor types (Vegfa^+/+ or Vegfa^−/− fibrosarcomas and Ret melanoma models) (Hansen et al., 2012). However, we did not observe any defect in the migration of Nrp1-deficient T_regs even in the competitive environment of B16 tumors in heterozygous Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} female mice. These discrepancies could be due to the different tumor models analyzed. Alternatively, their study primarily utilized Nrp1^L/LCD4^Cre mice in which Nrp1 would be removed in all T cells, which could have many direct and indirect effects on intratumoral T_regs (Hansen et al., 2012). Indeed, we and others have shown that Nrp1 is expressed on a number of cell types, especially in the tumor (Jackson et al., 2014) (data not shown).

Whether T_reg fragility is a feature of certain diseases and the extent to which this can be prevented or utilized therapeutically remains largely unknown and highly controversial (Rubtsov et al., 2010; Sakaguchi et al., 2013; Zhou et al., 2009a; Zhou et al., 2009b). We speculate that the IFNγ pathway may drive T_reg fragility in certain inflammatory environments. The loss of Nrp1 not only results in T_reg fragility but also results in substantial IFNγ expression, which in turn induces fragility in other T_regs regardless of their Nrp1 expression in a feed-forward manner; a process we refer to as “infectious fragility”. As IFNγ production is a hallmark of a productive T cell-mediated immune response, our observations also raise the possibility that IFNγ-induced T_reg fragility may be a physiologically important regulatory mechanism to locally limit T_reg function and promote a productive immune response. Given the profound consequences of T_reg-derived IFNγ production, our data emphasize the importance of the Nrp1 pathway in limiting T_reg fragility in the tumor microenvironment but also highlights that this pathway can be overcome when sufficient IFNγ is induced. Indeed, our data show that the IFNγ response induced by PD1 blockade appears to be sufficient to drive intratumoral T_reg fragility despite expression of Nrp1. However, loss of IFNγR expression on T_regs, renders mice completely resistant to anti-PD1 immunotherapy. This raises the provocative possibility that an essential component of effective immunotherapy is to induce sufficient IFNγ in the tumor microenvironment to drive T_reg fragility. However, the impact of T_reg fragility and IFNγ expression by intratumoral T_regs on tumor growth and responsiveness to immunotherapy in murine and human tumors needs to be investigated further.

As a high frequency of intratumoral T_regs in cancer patients is largely considered a negative prognostic factor (Facciabene et al., 2012; Ichihara et al., 2003; Knol et al., 2011), identifying approaches to selectively target intratumoral T_regs while maintaining peripheral tolerance is critical. Although expression of NRP1 in peripheral, tissue-resident T_regs remains unclear, our findings highlight the surprisingly extensive and variable expression of NRP1 on human intratumoral T_regs (Battaglia et al., 2009; Battaglia et al., 2008; Chaudhary et al., 2014; Milpied et al., 2009). Importantly, given that our mouse model experiments suggest that NRP1 may not need to be targeted in all human intratumoral T_regs to derive a therapeutic effect, it is possible that targeting NRP1⁺ intratumoral T_regs with an NRP1 mAb may be therapeutic. As we show that the impact of the IFNγ pathway on human T_regs is conserved, it is possible that by blocking NRP1 on T_regs, one could induce functional fragility in surrounding T_regs to further enhance the therapeutic effect and overall outcome. Our identification of IFNγ as the critical mediator of T_reg fragility, whether driven by manipulation of the Nrp1 pathway in T_regs or PD1 blockade highlights the potential importance of this mechanism in promoting anti-tumor immunity and provides a pathway to develop immunotherapeutic approaches that could lead to tumor reduction while maintaining peripheral tolerance.

Star Methods

Contact for Reagent and Resource Sharing

Further information and requests for reagents should be directed to and will be fulfilled by the Lead Contact, Dario A.A. Vignali (dvignali@pitt.edu).

Experimental Model and Subject Details

Mice

Nrp1^L/L mice were obtained from D. Cheresh (UC San Diego). Foxp3^{Cre-YFP/Cre-YFP}, Foxp3^{DTR-GFP/DTR-GFP}, Foxp3^−/− mice were obtained from A.Y. Rudensky (Memorial Sloan Kettering). (Fontenot et al., 2003; Kim et al., 2007; Rubtsov et al., 2008). Ifngr1^−/−, Ifng^−/−, Ifngr1^L/L and Rosa^{L-Tomato-L-GFP} mice were obtained from Jackson Laboratories (Dalton et al., 1993; Huang et al., 1993; Lee et al., 2013). All animal experiments were performed in the American Association for the Accreditation of Laboratory Animal Care-accredited, specific-pathogen-free facilities in Animal Resource Center, St. Jude Children's Research Hospital (SJCRH), and Division of Laboratory Animal Resources, University of Pittsburgh School of Medicine (UPSOM). Female and male mice were used. Animal protocols were approved by the Institutional Animal Care and Use Committees of SJCRH and UP.

Human T-cell populations

All HNSCC and melanoma tissues were acquired under a University of Pittsburgh Cancer Institute Institutional Review Board (IRB)-approved protocol with written informed consent obtained from each patient in conjunction with the University of Pittsburgh Cancer Institute HNSCC and Melanoma SPOREs. There were no restrictions on cancer subtype, smoking status, age, race, or prior adjuvant therapy. Control donor peripheral blood (PBL) was collected through an approved MTA protocol with the Western Pennsylvania Bloodbank. Human HNSCC PBL and TIL samples (unmatched) as well as healthy donor PBL samples were provided by R. Ferris from patients with high-risk, advanced (stage III or IV) resectable HNSCC treated with surgery. Most tumors were from oral cavity or laryngeal sites, and all were HPV-negative. Tumor specimens were obtained at the time of surgical resection, prior to adjuvant therapy. TIL were isolated, frozen, and thawed prior to staining for NRP1. Freshly processed samples were used in functional assays. Human melanoma TIL and PBL samples were provided by J. Kirkwood from an accrual trail (96-099) of patients with metastatic melanoma.

Method Details

Antibodies and flow cytometry

Single cell suspensions were stained with antibodies against CD4 (clone# GK1.5, Biolegend), CD8a (clone# YTS156.7.7, Biolegend; clone# H35-17.2, eBioscience), TCRβ (clone# H57-597, Biolegend), Thy1.1 (clone# OX-7, Biolegend), Thy1.2 (clone# 30-H12, Biolegend), Foxp3 (clone# FJK-16s, eBioscience; clone# 150D, Biolegend), IFNγ (clone# XMG1.2, Biolegend), ICOS (clone# C398.4A, Biolegend), phosphor-Stat1 (Clone# 4a, BD Biosciences) and phosphor-Stat4 (Clone# 38, BD Biosciences). Surface staining was performed on ice for 15min. For cytokine expression analysis, cells were activated with 0.1ng/ml PMA (Sigma) and 0.5ng/ml Ionomycin (Sigma) in RPMI containing 10% FBS and Monensin (eBioscience) for 8hr. For intracellular staining of cytokines and transcription factors, cells were stained with surface markers, fixed in Fix/Perm buffer (eBioscience) for 15 minutes, washed in permeabilization buffer (eBioscience) twice and stained intracellular factors in permeabilization buffer for 30min on ice. For phosphoprotein staining, cells were fixed with 1.5% PFA (Alfa Aesar) at 37°C for 15min, permeabilized with ice cold Methanol for 1hr, and stained on ice for 1hr. Cells were sorted on Aria II (BD Biosciences) or analyzed on Fortessa (BD Biosciences), and data analysis was performed on FlowJo (Tree Star).

Tumor models

Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, or Foxp3^{DTR-GFP/DTR-GFP} mice were injected with B16.F10 melanoma (1.25 × 10⁵ cells intradermally). Tumors were measured every 3 days with digital calipers and tumor size was calculated; this was performed in blinded manner but not randomized. 100ug Diptheria Toxin was injected every 3 days starting on day 7 in Foxp3^{DTR-GFP/DTR-GFP} and Foxp3^DTR-GFP/+ mice. Sema4aIg was injected every 3 days starting on day 5 (400ug, 200ug, 200ug), and anti-CD8 (YTS) was injected every 3 days starting on day 5 (200ug). Foxp3^Cre-YFP and Ifngr1^L/LFoxp3^Cre-YFP mice were injected with MC38 (5×10⁵ cells subcutaneously). Tumors were measured as above and mice were treated with either anti-PD1 (clone G4) or isotype (Armenian Hamster IgG). Tumors and non-tumor draining lymph nodes were collected for analysis on day 12. TILs were prepared with enzymatic (collagenase IV and dispase, 1mg/mL) and mechanical disruption. To achieve reasonable power, at least 5 mice were used in each group, at least 3 mice per group per experiment. Group means were compared with Student's t tests. Tumor growth over time was analyzed using two-way ANOVA with multiple comparisons. Event-free survival (moribund) estimates were calculated with the Kaplan-Meier method. Groups of mice were compared by log-rank test. All p values are two-sided, and statistical significance was assessed at the 0.05 level. Analyses were conducted using GraphPad Prism software.

Foxp3^−/− model

CD45.1⁺ Foxp3^+/− female mice were bred with CD45.1⁺Foxp3^+/+ male mice in timed matings. Male progeny were genotyped at birth for Foxp3^−/− status. T_regs from Thy1.1⁺ Foxp3^{Cre-YFP/Cre-YFP}, Thy1.2⁺ Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, Thy1.2⁺ Ifng^−/−Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, Thy1.1⁺ Ifngr1^−/− Foxp3^{Cre-YFP/Cre-YFP} mice were purified by FACS and 10⁶ cells injected intraperitoneally into Foxp3^−/− male pups within 2 days of birth (Workman et al., 2011). When a 50:50 mixture of T_regs was injected the total was maintained at 10⁶ cells. Mice were monitored for the autoimmune phenotype ‘scurfy’ (scaly skin, eye inflammation, runted phenotype, and lack of mobility) (Workman et al., 2011). Any mice exhibiting any autoimmune or inflammatory symptoms prior to B16 injection, even if mild, were removed from further study. Mice were injected with 1.25×10⁵ B16.F10 cells at 4 weeks of age and tumor growth was monitored every 3 days.

Gene expression profiling by RNAseq and bioinformatics analyses

Tregs (5×10³) were either single (n=3) or double sorted (n=2) and cDNAs were prepared using the SMARTer® Ultra™ Low Input RNA Kit for Sequencing - v3 user manual (Clontech Laboratories). We reasoned that double sorting results in higher purity but has lower yield and may alter the expression profile. Though melanoma genes were found at lower levels in double sorted samples no other substantial differences, such as activation of stress response genes, were observed. T_regs were sorted on the following markers: Foxp3^{Cre-YFP/Cre-YFP} (Nrp1^+/+) on YFP, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} (Nrp1^−/−) on YFP, Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} (Nrp1^−/−) on YFP, and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} (Nrp1^+/+) on GFP. Sequencing libraries were prepared using Nextera XT DNA Library Preparation kit (Illumina), normalized at 2nM using Tris-HCl (10mM, pH 8.5) with 0.1% Tween20, diluted and denatured to a final concentration of 1.8nM using the Illumina Denaturing and Diluting libraries for the NextSeq 500 protocol Revision D (Illumina). Cluster generation and 75bp paired-end dual-indexed sequencing was performed on Illumina NextSeq 500 system.

RNAseq data was aligned to the mm10 genome using the STAR aligner (Dobin et al., 2013) and quantified using featureCounts (Liao et al., 2014). The raw counts data were processed using the “voom” function (Law et al., 2014) in the limma R package (Ritchie et al., 2015; Wu and Smyth, 2012), which normalizes the data and assigns a weight for each measurement for subsequent linear model fitting. Unsupervised analysis of the date revealed a small cluster of melanoma specific genes that reasoned were caused by contaminations. Following a previous approach (Battle et al., 2014), we removed five melanoma specific genes from all downstream analysis (Mlana, Syt4, Tyr, Tyrp1, Dct). To filter for low expression genes we defined a cutoff of 90 reads per gene based on visual inspection of the bimodal count distribution. Only genes that meat this threshold in at least 5 samples (∼11,000 out of ∼23,000) were kept for further analysis. Differential expression was assessed using the limma moderated T statistic. The differences between the intratumoral Treg populations were subtle and in order to increase the power of our study we included technical factors as covariates in our differential expression analysis. Following the approach outlined in a recent human RNAseq study (Battle et al., 2014) we included three Picard RNAseq metrics (“PCT_INTERGENIC_BASES” “MEDIAN_3PRIME_BIAS”, “MEDIAN_CV_COVERAGE”) (Picard) as well GC correlation (computed as the sample specific Pearson correlation between each gene's GC content and its expression value). Normalization for replicate number and technical parameters was also applied directly to the voom result to obtain “normalized counts”, which were used for data visualization. Geneset enrichment was performed using the “RankSumWithCorrelation” function in the limma R package, which automatically corrects enrichment statistic inflation due to correlation among genes (with the immune genesets restricted to those relevant to T cells). p-values were combined from the “Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} (Nrp1^−/−) vs Foxp3^{Cre-YFP/Cre-YFP} (Nrp1^+/+)” and “Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} (Nrp1^+/+) vs Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} (Nrp1^−/−) using the Fisher's “log sum” method [Fisher] to define significant genes. Genesets with a q-value FDR of <0.2 were considered significant. In order to assess the alterations in Treg specific expression profile we relied on the Treg signature genes defined in a previous study (Hill et al., 2007). For pathway analysis bar charts, results of geneset enrichment analysis were depicted with colors representing the effect size and height representing the corresponding p-values. The effect size is defined as AUC (area under receiver operating curve) – 0.5 which provides a normalized ranksum statistic that is comparable across genesets of different sizes. The plot is restricted to the top 10 pathways (based on their Nrp1^+/+ vs Nrp1^−/− significance) from the “canonical” mSigDB geneset. We additionally restricted this analysis to genesets and pathways that were deemed relevant to intercellular signaling (defined at least half of the genes in the geneset having an extracellular or membrane annotation).

In vitro assays

Splenic CD4⁺YFP⁻ (CD4⁺Foxp3⁻ T cells) cells from Foxp3^Cre-YFP mice were sorted as responder cells and labeled with 5μM CellTrace Violet (Life Technology). CD4⁻CD8⁻ splenocytes from Foxp3^Cre-FP mice were treated with 20μg/mL mytomycin C (Sigma) at 37°C for 30min, washed five times with PBS, and then used as antigen presenting cells (APCs). Responder cells (4×10³), APCs (8×10³), and different concentrations of T_regs (1:2-1:16 T_reg:T_eff ratio, 500-2000 T_regs) were activated with 2μg/ml anti-CD3 (Biolegend) in a 96-well round bottom plate with 100ul RPMI for 3 days (Turnis et al., 2016). Suppression was calculated as previously described (McMurchy and Levings, 2012). Briefly, cells were acquired by BD Fortessa, and the division index of responder cells was analyzed using FlowJo based on the division of CellTrace Violet. Suppression was then calculated with the formula % Suppression = (1-DIT_reg/DICtrl) × 100% (DIT_reg stands for the division index of responder cells with Tregs, and DICtrl stands for the division index of responder cells activated without T_regs). Human microsuppression assays were performed similarly to mouse assays with the following changes: 0.5μg/ml anti-CD3 is used for activation, and cells are cultured in assay conditions (200uL) for 4 days.

For co-culture assays, sorted T_reg populations were cultured together in a 96 well round-bottom plate or in a 96 well transwell plate (Millipore) for 72 hours prior to being resorted and used in a suppression assay. Cells were treated with 100ng/mL PMA (Sigma), 500ng/mL lonomycin (Sigma) and 10001U hlL-2 (Prometheus) for co-culture. For some experiments, sorted T_regs were cultured in the presence of 0.3-20ng/mL anti-IFNγ (BioXcell) or 0-200ng/mL IFNγ (Biolegend) to the microsuppression assay.

For the hypoxia assays, T_regs were stimulated with PMA/Ionomycin and IL-2 for 3 days in normoxic (5% CO₂ and 20% O₂) or hypoxic (5.5% CO₂ and 1.5% O₂) conditions at 37°C. Cells were then stained in the same condition in the absence of cytokine.

Quantification and Statistical Analysis

Statistics were performed using Prism v6.07. Student t tests were used in Figures 1B,H; 3B-E; 4B-D, F-G; 5B,D; 7B-C and supplemental figures 1A-B; 2E-F, 5, 6B-D, 7C. ANOVA was used in Figures 1C, E, G; 5E-G; 6; 7A. Kaplan Meier was used in Figures 1C, E, F; 7A. “n” represents the number of mice used in the experiment, with the number of individual experiments listed in the legend. Graphs show individual samples. Samples are shown with the mean with or without error bars showing the SEM. Significance was defined as p=0.05.

Data and Software Availability

The RNASeq datasets have been deposited in the Gene Expression Omnibus (GEO) under code GSE97939.

Supplementary Material

key

Supplemental Figure 1. Temporal Nrp1 deletion leads to reduced tumor growth, Related to Figure 1. (A) Lymphocytes were isolated from TIL of Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice, stained and quantified for cell numbers (n=2-6). (B) Foxp3^CreERT2-GFP Rosa^LSL-YFP or Nrp1^L/LFoxp3^CreERT2-GFP Rosa^LSL-YFP mice were injected with B16.F10 ID on day 0 and treated with Tamoxifen on days 7-11. Tumors were measured every 3 days (n=5-11). (C) Foxp3^CreERT2-GFPRosa^LSL-YFP or Nrp1^L/LFoxp3^CreERT2-GFPRosa^LSL-YFP mice were bled on days indicated and stained. Mice were treated with Tamoxifen on days 1-5 (n=3). (D) Lymphocytes were isolated from ndLN and TIL of Foxp3^{DTR-GFP/DTR-GFP}, Foxp3^{DTR-GFP/DTR-GFP} + DT, and Foxp3^DTR-GFP/+ + DT mice (n=3-6). Data represent 2-4 independent experiments. Student unpaired t test (S1A), 2-way ANOVA and Kaplan-Meier tests (S1B) were used. (*p < 0.05, ****p<0.0001).

Supplemental Figure 2. Nrp1^−/− T_regs maintain Foxp3 expression, Related to Figure 1. (A-C) Lymphocytes were isolated from ndLN, dLN, and TIL of Foxp3^{Cre-YFP/Cre-YFP}Rosa^{L-Tom-L GFP} and Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}Rosa^{L-Tom-L GFP} mice and stained (n=3-5). (D-E) Lymphocytes were isolated from TIL of Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^Cre-YFP/+ mice and stained. T_regs from Nrp1^L/LFoxp3^Cre-YFP/+ mice were stained with Foxp3 and Nrp1⁺ and Nrp1⁻ T_regs were distinguished as YFP⁺Foxp3⁺ (Nrp1⁻) and YFP⁻Foxp3⁺ (Nrp1⁺). This was also verified by Nrp1 staining (n=5). Data represent 2-3 independent experiments. Student unpaired t test was used. (ns, not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

Supplemental Figure 3. RNASeq sorting scheme, Related to Figure 2. T_regs were isolated from ndLN and TIL of Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice and checked for purity. Cells were either single or double sorted. No significant difference was noted between single or double sorted samples. Data represent 5 independent experiments.

Supplemental Figure 4. Nrp1^−/− display an altered T_reg signature, Related to Figure 2. T_regs were isolated from ndLN and TIL of Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, and Nrp1^L/LFoxp3^{Cre-YFP/DTR-GFP} mice and checked for purity. (A) Volcano plots. (B) Global heatmap. All genes that passed a threshold of q-value FDR <0.2 are shown. (C) T_reg signature gene heatmap. Green/red coloring represents genes that are upregulated/downregulated in T_regs. (D) Bar chart of top differentially expressed pathways. Data represent 5 experiments.

Supplemental Figure 5. Nrp1 loss leads to an increase in TH1 markers, Related to Figures 3-4. (A-H) Lymphocytes were isolated from ndLN, dLN, or TIL of Foxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP}, Nrp1^L/LFoxp3^Cre-YFP/+ mice and stained. T_regs from Nrp1^L/LFoxp3^Cre-YFP/+ mice were stained with Foxp3 and Nrp1⁺ and Nrp1⁻ T_regs were distinguished as YFP⁺Foxp3⁺ (Nrp1⁻) and YFP⁻Foxp3⁺ (Nrp1⁺) (n=5-15). (I) Foxp3^{Cre-YFP/Cre-YFP} and Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} mice were injected with B16 on day 0 and treated with isotype or anti-CD8 (n=4-7). Lymphocytes were isolated from TIL and stained. (J-K) T_regs were isolated from Foxp3^{Cre-YFP/Cre-YFP} and Nrp1^L/LFoxp3^{Cre-YFP/Cre-YFP} mice and stimulated in hypoxia or normoxia for 3 days and stained (n=6). Data represent 3-6 independent experiments. Student unpaired t test was used. (ns, not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

Supplemental Figure 6. T_reg suppression is crippled by IFNγ, Related to Figures 5-6. (A) T_regs were isolated from ndLN and TIL of Foxp3^Cre-YFP mice, co-cultured with Nrp1^−/− T_regs and IL-2 in the presence or absence of anti-IFNγ, resorted and used in a microsuppression assay (n=6). (B) T_regs were isolated from the ndLN and TIL of Foxp3^{Cre-YFP/Cre-YFP} mice, treated with IL-12 and IL-2 for 72 hours, sorted based on Ifngr1 expression, co-cultured with Nrp1^−/− T_regs for 72 hours with or without anti-IFNγ, and re-isolated for a microsuppression assay (n=3). (C) T_regs were isolated from ndLN and TIL of Foxp3^Cre-YFP and Ifngr1^−/−Foxp3^Cre-YFP mice, treated with IL-2 and IFNγ for 72 hours, resorted and used in a microsuppression assay (n=3-4). (D) T_regs were isolated from TIL of Foxp3^Cre-YFP mice, treated with IFNγ and IL-2 for 72 hours, washed and stained (n=3). (E) Foxp3^−/− mice were injected with 10⁶ of a 50:50 mix of Nrp1^+/+Thy1.1 and Nrp1^−/−Thy1.2 T_regs on day 2 post-birth. B16.F10 was injected ID on day 28. Lymphocytes were isolated and stained (n=3). Data represent 2-3 independent experiments. Error bars represent the mean ± SEM. Student unpaired t test was used. (ns, not significant, *p < 0.05, **p < 0.01, ***p < 0.001).

Supplemental Figure 7. Ifngr1^L/LFoxp3^Cre-YFP mice are non-responsive to PD1 blockade, Related to Figure 7. (A) Lymphocytes were isolated from ndLN of Foxp3^Cre-YFP or Ifngr1^L/LFoxp3^Cre-YFP mice and stained for IFNγR (n=2). (B) T_conv, T_regs, CD8s, or B cells were isolated from LN and spleen of Foxp3^Cre-YFP or Ifngr1^L/LFoxp3^Cre-YFP mice, DNA was extracted for endpoint PCR testing for the presence or absence of the loxP sites or excision of loxP-targeted sequence (n=2). (C) Foxp3^Cre-YFP or Ifngr1^L/LFoxp3^Cre-YFP mice were injected with MC38 on day 0, sacrificed on day 13 (n=3-6). Data represent 2 experiments. Student unpaired t test was used. (ns, not significant, **p < 0.01).