Although immune checkpoint blockade (ICB) with anti-CTLA4, anti-PD1, anti-PD-L1) has revolutionized cancer therapy, only a small proportion (10–30%) of patients with cancer exhibit a durable clinical response1,2. Consequently, there is intense interest in identifying mechanisms of pre-existing and acquired immune resistance as well as the development of new therapeutic approaches to prevent relapse. It is also imperative to understand the mechanisms underlying the lack of a durable response seen in patients post-ICB treatment, an issue that can be associated with impaired formation of a long-lived memory CD8+ T cell (Tmem) pool capable of conferring durable remission3,4. It has been postulated that the restrained and/or limited Tmem pool that occurs in tumor-bearing hosts is reminiscent of observations in chronic viral infection where severe T cell exhaustion manifests at the expense of memory differentiation5. As a consequence, memory precursors fail to accumulate and thus are unable to form an adequate Tmem repertoire. Exhausted T cells (TEX) represent an independent CD8+ T cell lineage with considerable stability that is distinct from CD8+ T effector (Teff) or Tmem, and are marked by coexpression of multiple inhibitory receptors (IRs), such as PD1, CTLA4, LAG3, TIM3, TIGIT and 2B4 (refs. 6,7). PD1:PD-L1 ICB is insufficient to reinvigorate TEX to become durable Tmem (ref. 6), however, it is unclear what pathways would be necessary for restoration of robust and durable antitumor T cell memory.

NRP1, originally discovered as a neuronal and endothelial cell receptor required for normal embryonic development, axon guidance and vasculature formation8,9,10, is also expressed by several immune cell types where it participates in critical immune functions11. Notably, it has been identified as a marker for thymically derived murine regulatory T cells (Treg cells)12 and is crucial for their suppression of antitumor immunity13,14. NRP1 is elevated in Treg cells from patients with cancer and marks a highly suppressive Treg cell subset14,15,16, suggesting that NRP1 might be a therapeutic target for cancer immunotherapy. However, the functional role of NRP1 on other T cell subsets, in particular CD8+ T cells, remains poorly understood17,18,19.

In this study, we report that NRP1 is induced on tumor-infiltrating CD8+ T cells despite limited expression on naive CD8+ T cells. Although NRP1 expression is highly concordant with multiple IRs, its function is distinct in cell-intrinsically limiting Tmem differentiation without substantially modulating effector activity. Analysis of peripheral blood-derived CD8+ T cells from patients with head and neck squamous cell carcinoma (HNSCC) and advanced skin malignancies showed that increased NRP1 expression, particularly within the effector memory CD8+ T cell compartment, was correlated with the reduced size of the memory CD8+ T cell pool, poor disease outcome and decreased responsiveness to ICB. Our study uncovers a previously unappreciated role for NRP1 in CD8+ T cells and suggests that it is primarily an immune checkpoint of T cell memory. Thus, targeting NRP1 may facilitate the restoration of durable antitumor immunity.


NRP1 limits CD8+ T cell–mediated antitumor immunity to tumor rechallenge and anti-PD1 immunotherapy

NRP1 was robustly induced on CD8+ tumor-infiltrating lymphocytes (TILs) in the B16.F10 (B16) mouse melanoma model, although it is undetectable on naive CD8+ T cells and only expressed on a small fraction of effector or memory CD8+ T cell subsets from unchallenged mice (Fig. 1a). This was observed primarily in the KLRG1CD127 Teff and the KLRG1loCD127hi memory precursor effector cells (MPECs) but remained low in the KLRG1hiCD127lo short-lived effector cells (SLECs) and barely detectable in central memory T cells (TCM) (Fig. 1a and Extended Data Fig. 1a). The upregulation of NRP1 was also found in CD8+ Teff subsets from mice infected with lymphocytic choriomeningitis virus (LCMV), in particular higher in the MPECs induced by clone 13 infection (D30), a chronic infection model known to induce T cell exhaustion20, compared to those induced by Armstrong (D8), a model of acute infection (Extended Data Fig. 1b). Moreover, NRP1 was coexpressed with PD1 on CD8+ T cells both in unchallenged and tumor-bearing mice (Extended Data Fig. 1c), suggesting that NPR1 may act as an IR-like molecule in CD8+ T cells.

Fig. 1: NRP1 limits CD8+ T cell–mediated antitumor immunity to tumor rechallenge and anti-PD1 immunotherapy.
figure 1

a, Expression of NRP1 measured by flow cytometry in subsets of CD8αβ+ T cells isolated from the intestine of naive mice (n = 7, upper) and B16.F10 (B16) tumors recovered on day 12 (D12) postinoculation (n = 9, lower); bar graphs tabulating the percentage of NRP1+ cells within the indicated CD8αβ+ T cell subpopulations (TRM, tissue resident memory; NS, not significant). b, Top, Barnes–Hut Stochastic Neighbor Embedding (bh-SNE) visualization of the expressions of indicated markers by B16 intratumoral CD8+ cells recovered from E8ICre (WT) mice, D18 postinoculation. Bottom, representative flow plots depicting coexpression of NRP1 with other IRs in the same sample shown in the bh-SNE plots. c, SPICE (simplified presentation of incredibly complex evaluations) plots illustrating the coexpression of NRP1 and IRs by B16 tumor-infiltrating CD8+ T cells on D9, D12 and D18 postinoculation. d, Experimental scheme for induction of postsurgical tumor immunity with B16 tumors. e, E8ICre and E8ICreNrp1L/L mice subjected to B16 tumor model were monitored for tumor growth at primary (1°) and rechallenge phases, respectively. Left, survival curve for the 1° B16 tumor growth; middle and right, Kaplan–Meier curves for the percentage of tumor-free mice with the +30-d (middle) and +60-d (right) rechallenge schemes, respectively. f, E8ICreErt2gfp (E8ICreErt2gfp) and E8ICreErt2gfpRosa26LSL.mAmetrine.2A.Nrp1 (E8ICreErt2gfpRosa26LSL.Nrp1) mice were treated with tamoxifen (1.5 mg) by i.p. injection for 5 consecutive days before the B16 tumor model. Left, survival curve for the 1° B16 tumor growth; right, Kaplan–Meier curve for the percentage of tumor-free mice with the +30-d rechallenge scheme. g, Experimental scheme for anti-PD1 immunotherapy with MC38 tumor model. h,i, Tumor growth curve (h) and survival curve (i) of MC38 tumor implanted in the E8ICreNrp1L/L (n = 10) and E8ICre mice (n = 8). All data were pooled from three independent experiments, with n = 5 − 17 mice per group (as specified in the legend for each panel). Error bars, mean ± s.e.m.; statistical significance was determined by one-way ANOVA (a); log-rank test (e,f,i) or two-way ANOVA (h) with correction for multiple comparisons (****P < 0.0001 and other P values as indicated).

Source data

NRP1+CD8+ TILs exhibited higher expression of markers associated with T cell activation (CD44, CD69, CD25) and cell proliferation (BrdU) compared to NRP1 cells, but lower expression of CD62L, CD127 and KLRG1, markers associated with naive/memory or senescent T cell phenotypes (Extended Data Fig. 1d). NRP1 expression, as determined by high dimensional flow cytometry, consistently marked a subset of PD1hi, multi-IR+ CD8+ TILs, which was most prominent on D18 post-B16 inoculation, suggesting that NRP1 may primarily function on TEX cells (Fig. 1b,c).

To interrogate whether NRP1 specifically affects CD8+ T cell function in antitumor immunity, we generated mice with (1) CD8+ T cell–restricted NRP1-deficiency (E8ICreNrp1L/L) (Extended Data Fig. 2a,b), and (2) CD8+ T cell–restricted, temporally controlled, constitutive NRP1 expression (E8ICreErt2gfpRosa26LSL.Ametrine.2A.Nrp1; GFP marks CD8 expression, Ametrine marks Cre activity and Rosa26 driven-Nrp1 expression) (Extended Data Fig. 2c). Deletion of Nrp1 in CD8+ T cells did not substantially alter the composition of the immune infiltrate in primary (1°) B16.F10 (B16) tumors (Extended Data Fig. 3a–d), nor did it result in a notable difference in the growth of 1° B16 tumors (Fig. 1d and Extended Data Fig. 3e). Furthermore, whereas therapeutic vaccination of B16-Ova tumor-bearing mice with attenuated Listeria monocytogenes expressing Ova peptide (LM-Ova) exhibited a substantive reduction in tumor growth, this was comparable between E8ICre and E8ICreNrp1L/L mice (Extended Data Fig. 3f). Collectively, these data indicate that NRP1 loss from CD8+ T cells has no effect on primary tumor growth and the effector T cell response to primary tumors following therapeutic vaccination.

To assess the effect of NRP1 on the recall response to a secondary tumor challenge, the primary tumor was resected at day (D) 12, mice left for 1 or 2 months, and then injected with B16 on the contralateral side (Fig. 1d). E8ICreNrp1L/L mice exhibited notably enhanced protection against a secondary (2°) B16 challenge 30 or 60 d following 1° tumor resection, relative to E8ICre mice (Fig. 1e and Extended Data Fig. 3g). In contrast, 2° B16 tumor growth was slightly exacerbated in E8ICreErt2gfpRosa26LSL.Nrp1 mice in which NRP1 was constitutively expressed on CD8+ T cells following treatment with tamoxifen before 1° tumor challenge (Fig. 1f and Extended Data Fig. 3h). The E8ICreNrp1L/L mice also showed significantly improved sensitivity to anti-PD1 checkpoint blockade, when administered at a suboptimal dose (100 μg per mouse) in the 1° MC38 tumor model, resulting in 70% complete regression compared to 25% complete regression in the E8ICre control mice (Fig. 1g–i). Taken together, these data indicate that loss of NRP1 in CD8+ T cells substantively enhances immunity against secondary exposure to tumors, including relatively ‘cold’ tumors such as B16, and improves sensitivity to anti-PD1 immunotherapy, albeit with little impact on the growth of untreated primary tumors.

NRP1 promotes terminal exhaustion in tumor-infiltrating CD8+ T cells

Tumor-primed T cell memory contributes critically to postsurgical tumor immunity21,22. We thus asked whether NRP1 acts as a CD8+ T cell-intrinsic modulator of this process. At the time of tumor resection (D12), although the number of intratumoral CD8+ TCM (CD44+CD62L+) in the E8ICreNrp1L/L mice was not significantly altered compared with the E8ICre control mice, the number of intratumoral CD8+ Teff (CD44+CD62L) was significantly increased, resulting in a higher Teff to TCM ratio (Fig. 2a,b). We further examined MPECs and SLECs, the bifurcation of which represents the first step of memory T cell commitment, but did not observe a significantly altered ratio between these two subsets, although both showed increased numbers in the E8ICreNrp1L/L mice (Fig. 2a,b). This was also the case when we examined tumor-specific CD8+ T cells that recognized the melanocyte differentiation antigen gp100 (H2-Db gp100+, designated as Db-gp100+ thereafter) (Extended Data Fig. 4a,b). Nevertheless, the frequency of Db-gp100+ CD8+ T cells, as well their TCM phenotype in the spleen of these mice was significantly increased on D12 (Extended Data Fig. 4c,d), consistent with their enhanced protection against tumor rechallenge (Extended Data Fig. 4e). Taken together, these data indicated that loss of NRP1 on CD8+ T cells promoted an intratumoral effector CD8+ T cell pool that was more capable of generating a peripheral memory T cell pool, without any alteration in proportional fate choice between MPECs and SLECs.

Fig. 2: NRP1 promotes terminal exhaustion in tumor-infiltrating CD8+ T cells.
figure 2

a,b, Numeration of CD8+ T cell subsets (TCM, Teff, MPECs, SLECs) infiltrating B16-gp100 tumor on D12, from the E8ICre (n = 7) or E8ICreNrp1L/L (n = 10) mice. Data presented as absolute cell number per gram tumor mass. Data were pooled from two independent experiments with consistent observations. c, Left, representative flow plots for the expression of IRs on CD8+ TILs of 1° B16 tumors on D18. Right, SPICE plots visualization for coexpression of multiple IRs. d, Numeration of 5-IR-coexpressing cells (5-IR+) on D18 (n = 8), depicted as the percentage within CD8+ TILs (left) and absolute number per gram tumor mass (right). Bar represents mean value. e, Representative flow cytometry plot for the expressions of cleaved Caspase 3 (cCasp3), Bcl2, Ki67 and TCF1 on CD8+ TILs recovered from 1° B16 tumors on D18 postinoculation. f, Bar graphs tabulating the mean percentage of cells of indicated phenotype within CD8+ TILs, that is, total cCasp3+ (n = 5), Bcl2hicCasp3 (n = 6), Ki67+ (n = 7) and TCF1+ (n = 8), respectively, on D12 and D18 of 1° B16 tumors. g, Representative flow cytometry plots depicting the subsets within the exhausted CD8+ T cell (TEX) pool of B16-gp100 tumors, harvested on D12 and D18, segregated by progenitor TEX (pTEX, CD44+PD1+TCF1+TIM3) and terminally TEX (tTEX, CD44+PD1+TCF1TIM3+). Bar graphs tabulating the number (mean value) of pTEX and tTEX per gram tumor mass are shown on the right. n = 8 for D12 and n = 5 for D18, pooled from two independent experiments. h, The same pTEX and tTEX analysis as in f, but within the CD8+ TILs specific for gp100-tetramer (Db-gp100+). Only samples with >50 intratumoral Db-gp100+ cells gated for flow cytometry analysis were included. All symbols represent individual mice; error bars, mean ± s.e.m.; statistical significance was determined by two-tailed unpaired Student’s t-test (a,b,d) or two-way ANOVA (f,g) with correction for multiple comparisons; all P values are indicated.

Source data

Apart from generating long-lived memory T cells, MPECs are considered to be the same precursors that give rise to TEX following prolonged antigen stimulation, which commonly occurs in tumor-bearing or chronic virally infected hosts5. The Nrp1−/− CD8+ TILs exhibited a slight increase in IR expression on D12 and an increased proportion of 5-IR expressing cells compared to the Nrp1+/+CD8+ TILs (Extended Data Fig. 4f,g). In contrast, there was a significant reduction of IR expression and reduced proportion of 5-IR expressing cells on D18 (Fig. 2c,d), when most intratumoral CD8+ TILs were functional exhausted, evidenced by the substantially reduced frequency of polyfunctional cytokine-producing cells (Extended Data Fig. 4h,i). Although the Nrp1−/− CD8+ TILs did not show significantly improved polyfunctionality on D18 (Extended Data Fig. 4h,i), they did exhibit alterations in certain TEX related-features, such as reduced cell turnover indicated by lower levels of Ki67 and cleaved Caspase 3 (Fig. 2e,f). Conversely, they showed increased expression of the survival factor Bcl2 and notably TCF1, the transcription factor associated with programming both memory precursor and progenitor TEX (pTEX)23,24. Moreover, TCF1 is also required for the maintenance of intratumoral pTEX pool by driving a transcriptional program that confers stem-like, self-renewal properties25. Indeed, while the number of terminally TEX (tTEX, CD44+PD1+TCF1TIM3+) did not significantly differ between the two genotypes on D18, the number of pTEX (CD44+PD1+TCF1+TIM3) was significantly increased in the E8ICreNrp1L/L mice (Fig. 2g). This was particularly evident for the tumor antigen-specific (Db-gp100+) CD8+ Teff from the B16-g100 tumor-bearing E8ICreNrp1L/L mice, which contained a higher frequency and number of pTEX on D18 but also D12 (Fig. 2h). The early time point (D12) observation is interesting, as it coincided with the increased frequency of peripheral Db-gp100+CD8+ in the E8ICreNrp1L/L mice on D12 (Extended Data Fig. 4c). Altogether, our data indicated that loss of NRP1 in late-stage CD8+ effector T cells restricts terminal exhaustion by reducing cell turnover and importantly, imposing a stem-like self-renewing phenotype that may improve differentiation toward functional, long-lived memory cells.

NRP1 cell-intrinsically limits the in vivo persistence of tumor antigen-specific CD8+ T cells

To accurately track CD8+ T cell differentiation, persistence and fate in vivo, as well as assess if NRP1 functions in a cell-intrinsic or cell-extrinsic manner, congenically marked Nrp1−/− and Nrp1+/+ pMel transgenic CD8+ T cells that are specific for premelanosome protein gp100 (ref. 26) were mixed at a 1:1 ratio and cotransferred into CD45.1 C57BL/6 recipients on D-1 (Fig. 3a and Extended Data Fig. 5a). On D0, mice were injected with B16-gp100 cells, the tumor resected on D12 and mice rechallenged on D42 with B16-gp100 cells. While Nrp1−/− and Nrp1+/+ pMel-T cell numbers were comparable at early stages of both the 1° and 2° tumor, Nrp1−/− cells outnumbered their wild-type (WT) counterparts over time (Fig. 3b,c). The ratio between Nrp1−/− versus Nrp1+/+ donor pMel-T cells positively correlated with the tumor size, consistent with the notion from the polyclonal CD8+ T cell setting that loss of Nrp1 led to a survival advantage in CD8+ Teff under prolonged antigen exposure (Extended Data Fig. 5b). As a result, Nrp1−/− pMel-T cell number was increased on D12 and exhibited delayed contraction compared with their WT counterparts (Extended Data Fig. 5c). Additionally, a roughly two to one ratio of Nrp1−/− to WT pMel-T cells was consistently observed in the peripheral lymphoid tissues, but not in the blood during the 1° tumor phase, which was maintained long-term (Fig. 3b,c and Extended Data Fig. 5d,e). These data indicate that NRP1 restrains in vivo persistence of tumor antigen-specific CD8+ T cells in a cell-intrinsic manner.

Fig. 3: NRP1 cell-intrinsically limits the in vivo persistence of antigen-specific CD8+ T cells.
figure 3

a, Experimental scheme for pMel-T cell cotransfer model. b,c, Representative flow cytometry plots depicting the Nrp1+/+ and Nrp1−/− pMel-T cell donors detected in the B16-gp100 tumors (upper), and the matched NdLNs (lower) during primary phase (n = 5 for D9 and D15, n = 7 for D12 and D21) (b) and recall phase (n = 7 for both time points) (c). Bar graph on the side tabulates the mean percentage of cells of either genotype within the donor cell pool (CD45.2+). Replicates were derived from individual recipients, and three independent time course experiments with consistent observations were pooled. d, Top, representative flow cytometry plots showing the expression of TCF1 and TIM3 by intratumoral Nrp1+/+ or Nrp1−/− pMel-T donors (gated on CD44+PD1+), from B16-gp100 tumor harvested on D12. Bottom, the percentage of pTEX and tTEX within the Nrp1+/+ or Nrp1−/− pMel-T donors were tabulated (n = 8). Each pair of connected dots represented donor cells recovered from the same individual recipient; data were pooled from two independent experiments. e, Left, gating scheme used to identify subsets enriched for MPECs or TCM within the CD44+ donor-derived pMel-T cells found in the periphery between D12 and D42 of the primary phase. Right, frequencies of Nrp1+/+ (n = 5) or Nrp1−/− (n = 5) -derived donor cells of the CD27+CD62L (MPECs), or CD27+CD62L+ (TCM) phenotype within CD45.2+ compartment over time, recovered from NdLN. Each replicate was derived from an individual recipient, and three independent time course experiments with consistent observations were pooled. f, Left, representative flow cytometry plots showing the expression of Bcl2 and IRF4 in the NRP1 and NRP1+ fractions of Nrp1+/+ pMel-T cells of MPECs phenotype, recovered from NdLN on D21 of the primary phase. Right, ratio between the Bcl2hiIRF4lo and Bcl2loIRF4hi subset within the Nrp1+/+ or Nrp1−/− pMel-T cells (n = 3, from individual recipients) recovered from NdLN of CD45.1 recipient, on D12 and D21 of the primary phase. Bar represents mean value. Error bars, mean ± s.e.m.; statistical significance was determined by two-tailed paired Student’s t-test (bd and f) or two-way ANOVA (e) with correction for multiple comparisons (P values as indicated).

Source data

Although NRP1 was highly induced on intratumoral pMel-Teff, it was downregulated in early MPECs cells found in the periphery on D21 and undetectable in established TCM on D56 (Extended Data Fig. 5f), aligned with the observation from polyclonal CD8+ T cells that NRP1 is primarily expressed during the effector phase. Therefore, we hypothesized that the impact of NRP1 on Tmem fate may be selective for effector-to-memory (E→M) transition, rather than Tmem pool maintenance, which occurs later.

Consistent with our observations in E8ICreNrp1L/L mice, intratumoral Nrp1/ pMel cells contained higher frequency of pTEX (CD44+PD1+TCF1+TIM3) compared to their WT counterparts on D12, supporting the notion that intratumoral pTEX phenotype were better preserved in the absence of NRP1 (Fig. 3d). Furthermore, from D12 to D42, during which E→M transition occurs, the percentage of Nrp1/ MPECs within the donor cell pool (CD45.2+) was preferentially sustained compared to Nrp1+/+ MPECs (Fig. 3e and Extended Data Fig. 5g). Consequently, this contributed to an increase in the Nrp1/ TCM pool during this period, in contrast to a smaller, unchanged Nrp1+/+ TCM pool. Consistent with polyclonal CD8+ T cells, the NRP1+ fraction within the WT pMel donor-derived MPECs exhibited a higher percentage of Ki67+ cells compared to their NRP1 counterparts (Extended Data Fig. 5h). Additionally, this was associated with a Bcl2loIRF4hi short-lived effector phenotype, while the NRP1 counterparts had a predominantly Bcl2hiIRF4lo long-lived cell phenotype27. Furthermore, ablation of Nrp1 resulted in a higher ratio of Bcl2hi versus Bcl2lo cells within MPECs during E→M transition, particularly on D21 (Fig. 3f and Extended Data Fig. 5i). Taken together, NRP1 appears to limit proliferative quiescence in antigen-specific cells during the E→M transition, leading to reduced cell survival and inefficient commitment to a memory T cell fate.

To identify downstream target(s) of NRP1 in CD8+ T cells, we performed transcriptomic analysis using bulk-population RNA sequencing (bpRNA-seq) of Nrp1+/+ and Nrp1−/− donor pMel-T cells recovered from B16-gp100 tumors (D12 and D21 posttransfer), along with draining (DLN) and non-draining lymph nodes (NdLNs) (D12, D21, D35 and D63 posttransfer) (sampling scheme, Fig. 4a). Principal component analysis (PCA) of pooled datasets demonstrated that tumor-derived and peripherally derived T cells segregated into two distinct clusters, in line with the effector-dominating and memory-dominating phenotype of cells present in these two locations (Fig. 4b). The peripherally derived datasets further segregated into four clusters (C1–C4), primarily based on the stage of cell activation, with the C1 and C2 consisting of cells recovered from the primary phase (D12 and D21), C3 with cells from the late exhaustion/early memory phase (D35) and C4 with cells from recall/secondary memory phase (D63). Loss of NRP1 had a greater impact on the transcriptome of cells in the primary phase (C1 and C2) and recall/secondary memory phage (C4), which was driven by biosynthesis (C1), cell division (C2) and the effector T cell activity and T or B cell interaction in the secondary effectors (C4), respectively (Supplementary Table 1). When compared to CD8+ T cell-specific gene signatures derived from LCMV infection models, the loss of NRP1 in the tumor-infiltrating effector cells resulted in a significant enrichment of gene signatures associated with a naive or quiescent phenotype and reciprocally a reduction with exhaustion gene signatures, particularly at D21 (Fig. 4c,d). A similar pattern was observed in the corresponding peripherally derived dataset (cluster C2), suggesting that enhanced quiescence retention during E→M transition might contribute to increased memory differentiation in Nrp1/ CD8+ Teff cells (Fig. 4e). Furthermore, effector-, exhaustion- and memory-associated gene signatures were enriched in Nrp1/ pMel-T cells recovered on D63 (C4), indicative of enhanced recall activity and memory cell generation in the absence of NRP1 (Fig. 4f).

Fig. 4: Impact of NRP1 on the transcriptome of antigen-specific CD8+ T cells in vivo.
figure 4

bpRNA-seq was performed on donor-derived Nrp1+/+ (CD45.2+Thy1.1+Thy1.2+) and Nrp1/ (CD45.2+Thy1.2+) pMel-T cells, sorted on D12, D21 and D35 of the primary phase, as well as D21 post-rechallenge (D63) from tumor (D12 and D21 only), DLN and NdLN, respectively. A total of 34 datasets were generated and grouped into five clusters (one tumor-derived and four periphery-derived), by unbiased clustering (described in Methods) for subsequent analysis. Genotype comparison (Nrp1/ versus Nrp1+/+) was performed at pathway level using the GSEA within each cluster against the MSigDB C7. a, Sampling scheme for Nrp1/ and Nrp1+/+ pMel-T cells subjected to bpRNA-seq. b, Left, t-distributed stochastic neighbor embedding (t-SNE) plots illustrating the segregation between tumor- and periphery-derived datasets; right, subclusters within the periphery-derived datasets by PCA. cf, Radar plots depicting the enrichment for CD8-associated phenotypes, that is, naive, effector (EFF), exhausted (ESG) and memory (MEM), that corresponded to the Nrp1–/– pMel-T cell in the tumor-derived cluster (D12 and D21) and periphery-derived cluster nos. 2 and 4, by using the CD8-specific GSEA analysis (described in Methods). gi, The GSEA plots assessing the enrichment of the gene signatures of chronic TCF1 P14 cells (GSE83978) or of TCF1+ P14 cells (GSE114631) (g); the targets of Id3 (GSE23568) between the peripherally derived Nrp1/ and Nrp1+/+ pMel-T cells (h) and the signature of CXCR5+CD8+ T cells (GSE76279) between the tumor-derived Nrp1/ and Nrp1+/+ pMel-T cells (i).

Gene set enrichment analysis (GSEA) revealed that peripheral Nrp1+/+ pMel-T cells (mostly MPECs or Tmem) from D21 B16-gp100 tumor-bearing hosts exhibited substantive overlap with the gene expression profile of Tcf7-GFP P14 cells (‘chronic TCF1 cells’, GSE83978, ref. 24: NB, the gene Tcf7 encodes the protein TCF1) from a LCMV C13 infected host (Fig. 4g), which were reportedly defective in long-term persistence following chronic infection compared to the ‘chronic TCF1+ cells’24. Conversely, the Nrp1/ pMel-T cells possess a phenotype that resembles TCF1+ cells, evidenced by the association with a gene signature corresponding to Tcf7-GFP+ P14 cells from B16-gp33 tumor-bearing host (GSE114631, ref. 25) (Fig. 4g). These peripheral Nrp1/ pMel-T cells also exhibited significant enrichment for a subset of Id3-dependent genes, which were downregulated in Id3/ pMel-T cells recovered from mice that were infected with vaccinia virus encoding the cognate antigen gp100 (gp100-VV) (GSE23568, ref. 28) (Fig. 4h). Coincident with this finding, Id3 expression (measured by an Id3-gfp reporter29) was enhanced in intratumoral Nrp1−/− pMel-T cells compared with WT counterparts on D18 (Extended Data Fig. 6a,b). Within the intratumoral CD8+ T cells, Id3 expression primarily marks a subset of pTEX, rather than TCM, with TCF1 serving as the key lineage-defining transcription factor (Extended Data Fig. 6c). Moreover, this Id3-dependent gene set exhibited considerable overlap with the genes that are differentially modulated between chronic TCF1+ versus TCF1 cells in a subset of cell-cycle regulation genes (for example, Ccnb2, Ccnb1, Nek2 and Prc1) (Supplementary Table 2). This observation not only supported an enhanced naive or quiescence trait in Nrp1−/− pMel-T cells, as determined by the previous global GSEA analysis, but also suggested that TCF1 controlled this transcription program, of which Id3 is known to be one of its key targets24,30,31,32, that may primarily account for the alterations observed in Nrp1−/− pMel-T cells.

In contrast, the intratumoral Nrp1−/− pMel-Teff cells exhibited a significantly enriched signature comprising genes upregulated in CXCR5+CD8+ T cells (GSE76279, ref. 33) (Fig. 4i), a PD1-responsive subset with stem cell-like, self-renewal properties34. This was consistent with the role of TCF1 in programming CXCR5+CD8+ T cells during chronic LCMV infection34. This was also consistent with our observation that E8ICreNrp1L/L mice exhibited enhanced sensitivity to PD1 blockade-induced tumor regression and the enhanced pTEX phenotype in the Nrp1−/− TILs. The latter notion was particularly relevant considering that CXCR5 signature enrichment was more significant on D21 compared to D12 (Fig. 4i), which indicated that the upregulated CXCR5 signature was associated with a reduced exhaustion phenotype in the Nrp1−/− CD8+ TILs from late-stage tumors. This was validated by increased CXCR5 protein expression on Nrp1−/− pMel-T cells recovered from D21 tumors, compared to their WT counterparts. This elevated CXCR5 expression was preferentially enriched in the Ki67 fraction, supporting the notion that a gain of ‘stemness’ was associated with retention of proliferative quiescence (Extended Data Fig. 6d). Taken together, our transcriptomic analysis strongly suggested that the absence of NRP1 drives a TCF1-dependent transcriptional program that promotes memory fate choice in MPECs under settings associated with prolonged antigen stimulation.

NRP1 inhibits c-Jun/AP-1 activation in chronically stimulated CD8+ T cells

To better determine the CD8+ T cell-intrinsic target that is directly modulated by NRP1, we sought to model chronic antigen stimulation in vitro. Naive Nrp1+/+ or Nrp1−/− CD8+ T cells activated with plate-bound anti-CD3 plus anti-CD28 for 48 h were either rested in IL-2-containing medium (acute activation, 1°), or subjected to two- (2°) or three- rounds (3°) of restimulation by anti-CD3/anti-CD28-coated beads (chronic stimulation, Fig. 5a). The NRP1 ligand Semaphorin-4A is highly expressed on CD8+ T cells and thus available for NRP1 ligation in cis or trans in this assay system (Extended Data Fig. 7a)13,35. Compared to acute activation, repetitive stimulation was sufficient to drive some of the hallmarks of in vivo TEX, such as the coexpression of multiple IRs (for example PD1 and LAG3), and altered cytokine production (a switch from IFN-γ+TNF-α+ to partial IFN-γ+GzmB+ producers), which were increased by the end of the 3° compared to 2° restimulation (Extended Data Fig. 7b,c). Surface expression of NRP1 was more robustly maintained in the chronically stimulated cells compared to the acutely activated cells, consistent with the notion that NRP1 expression requires continuous antigen stimulation, similar to many known IRs (Extended Data Fig. 7a).

Fig. 5: NRP1 inhibits c-Jun/AP-1 activation in chronically stimulated CD8+ T cells.
figure 5

a, Experimental scheme for in vitro chronic stimulation of CD8+ T cells. b, Representative flow cytometry plots showing the expression of TCF1, TIM3 and Ly108 in the in vitro chronically stimulated Nrp1+/+ or Nrp1−/− CD8+ T cells. c, The percentage of TCF1+TIM3 or Ly108+TIM3 within the CD44+PD1+ gate, from Nrp1+/+ or Nrp1−/− CD8+ T cells cultured under the indicated conditions. d, Left, representative flow plot depicting the gating strategy for measuring c-Jun expression. Right, representative histogram illustrating the expression of c-Jun by indicated subsets in the CD8+ T cells subjected to chronic stimulation. e,f, Expression of c-Jun, quantified by genomic mean fluorescence intensity (gMFI) by flow cytometry, within the indicated subsets from chronically stimulated Nrp1+/+ or Nrp1−/− CD8+ T cells (n = 5 for each group) (e), or CD8+ TILs from B16-gp100 tumor implanted in the E8ICre (n = 6) or E8ICreNrp1L/L (n = 5) mice (f). Representative histograms for the genotype comparison within the indicated subsets are shown. Error bars, mean ± s.e.m.; data in ae were aggregated from five independent in vitro stimulation assays and data in f were from two independent mice cohorts. Statistical significance was determined by two-tailed unpaired Student’s t-test (c) or two-way ANOVA (e,f) with correction for multiple comparisons. All P values are indicated.

Source data

Although there was no significant difference in the overall multi-IR or cytokine expression between genotypes at any time point with chronically stimulated cells, loss of Nrp1 resulted in an increase in the percentage TCF1+TIM3 pTEX-like cells, as well as an increase in the percentage of Ly108+TIM3 cells (Ly108, encoded by Slamf6, is considered a reliable surface marker for TCF1+ cells24). This was rapidly replaced by the TCF1TIM3+ tTEX-like phenotype within Nrp1+/+ pMel-T cells (Fig. 5b,c). Retention of TCF1 and Ly108 expression in Nrp1−/− cells was more prominent under 2° than 3° restimulation, indicating that NRP1 acts as one of the early regulators that promotes TCF1 loss in chronically activated CD8+ T cells.

It is known that signaling downstream of the T cell receptor (TCR) and costimulation is dampened in T cells subjected to prolonged antigen stimulation, resulting in an inadequate pairing between nuclear factor of activated T cells (NFAT) and the canonical AP-1 family transcription factor (c-Jun/c-Fos heterodimer), thereby driving a gene transcription program that promotes T cell dysfunction36,37. Thus, we next assessed NFAT1 nuclear translocation and induction of c-Jun, a key member of the AP-1 family, in response to TCR restimulation. NFAT1 translocated to the nucleus in ~20% of both Nrp1+/+ and Nrp1−/− pMel-T cells that were chronically stimulated compared to 80% of naive CD8+ T cells (Supplementary Fig 7d,e). The transcriptional activation of c-Jun, an event that requires concerted signaling through the TCR in conjunction with costimulation38, was also reduced in the chronically stimulated cells, the extent of which correlated with the progression from the Ly108+TIM3 to the Ly108TIM3+ stage (Fig. 5d). Therefore, the three subsets of chronically stimulated cells, stratified by Ly108 (surface surrogate for TCF1 expression) and TIM3 expression, were not only phenotypically distinct but also associated with progressive signaling impairment. c-Jun activation was significantly increased in chronically stimulated Nrp1−/− pMel-T cells, in particular for the Ly108+TIM3 and Ly108+TIM3+ cell subsets (Fig. 5e). These data indicated that NRP1 may directly modulate the pairing between NFAT and AP-1 by downmodulating c-Jun.

Prompted by this observation, we asked whether NRP1 also modulates c-Jun induction in tumor-infiltrating CD8+ T cells on restimulation. Indeed, c-Jun activation in the CD8+CD44+PD1+ Nrp1+/+ TILs was only preserved in the Ly108+TIM3 subset, progressively declined as cells differentiated through the Ly108+TIM3+ and Ly108TIM3+ subsets, whereas their peripheral CD8+CD44+PD1+ T cell counterparts (predominantly Ly108+TIM3) uniformly upregulated c-Jun on TCR restimulation (Extended Data Fig. 7f). This indicated that decreased c-Jun activation also correlated with the progression toward exhaustion in vivo. In contrast, the Nrp1−/− CD8+CD44+PD1+ TILs showed improved c-Jun activation on restimulation, particularly at the Ly108+TIM3+ and Ly108TIM3+ stages (Fig. 5f). Taken together, these data indicate that NRP1 suppresses the activation of c-Jun, a mechanism known to counteract the T cell exhaustion transcriptional program39, thereby promoting progression to terminal exhaustion in chronically stimulated CD8+ T cells.

Elevated NRP1+ TEM in patients with cancer is associated with poor survival and decreased response to ICB therapy

Last, we interrogated the clinical relevance of NRP1 expression and its modulation of CD8+ memory T cell differentiation in patients with cancer. We assessed a cohort of peripheral blood lymphocyte (PBL) samples from treatment-naive patients with HNSCC, consisting of 23 early stage disease (T1, T2 and/or N0) and 25 advanced disease cases (T3, T4 and/or N1, N2B, N2C) (Cohort A, Supplementary Table 3). Increased surface NRP1 expression was significantly associated with advanced disease, which was observed in both the CD8+ effector memory (TEM) and terminally differentiated effector (TEMRA) subsets (Fig. 6a–c). Consistent with the NRP1 expression pattern observed on the TILs from the B16 mouse model, NRP1 expression correlated with PD1 expression on CD8+ T cells in patients with HNSCC along with enhanced expression of other IRs, raising the possibility that this also correlated with exhaustion (Fig. 6d). NRP1 expression on CD8+ TEM, but not TEMRA, was enhanced in patients with advanced disease and correlated with disease recurrence and decreased overall survival (Fig. 6e,f), suggesting that the NRP1 expression on CD8+ TEM may be a predictor for poor prognosis. NRP1 may also play a role in limiting the peripheral Tmem pool in the patients with HNSCC, indicated by the inverse correlation between NRP1 expression and the size of the memory CD8+ T cell compartment in the periphery (Fig. 6g). Consistent with this notion, patients with advanced, progressive HNSCC and elevated NRP1 expression on their TEM cells exhibited an even greater reduction in peripheral TEM percentage within CD8+ T cells (Fig. 6h).

Fig. 6: Elevated NRP1+ TEM in patients with cancer is associated with poor survival and decreased response to ICB therapy.
figure 6

ah, CD8+ T cell subsets were analyzed on banked PBL of healthy donors (n = 10), and patients with advanced HNSCC (n = 48, cohort A, Supplementary Table 3). a, Gating scheme of CD8 subsets (naive, TEM, TEMRA) in human peripheral blood (PB) (TEM, effector memory T cells; TEMRA, effector memory RA T cells). b, Representative flow cytometry plots for surface NRP1 and PD1 expression on CD8+ TEM and TEMRA. c, Bar graphs tabulating the percentage of surface NRP1+ cells within PBL-derived CD8+ TEM or TEMRA of healthy donors (n = 10) and the patients with HNSCC stratified by disease stage (early, n = 23 versus advanced, n = 25). Bar represents mean value. d, SPICE plots depicting coexpression of surface NRP1 with other IRs (PD1, LAG3, CD39 and TIGIT) by PBL CD8+ T cells from the advanced patient cohort. e, Bar graphs tabulating the percentage of surface NRP1+ cells within PBL-derived CD8+ TEM or TEMRA, stratified by recurrence (stable disease versus progressive disease). SD, stable disease (n = 16); PD, progressive disease (n = 9). Bar represents mean value. f, Survival curve stratified by high or low (threshold indicated) surface NRP1 expression (by percentage of NRP1+ cells) on CD8+ TEM and TEMRA, respectively. *P = 0.0454. g, Correlation between percentage of TEM within CD8+ T cells and their surface NRP1 expression across all the patients with HNSCC (n = 48). h, The percentage of TEM within CD8+ T cells in subgroups of advanced patients stratified by disease recurrence (SD, n = 16; PD, n = 9). im, PBL were isolated from patients with advanced skin cancers both before (pre) or after (post) the start of ICB therapies (n = 40; cohort B, Supplementary Table 4). Patients were stratified by responsiveness to treatment (responder (RESP) versus progressor (PROG)). i, Surface NRP1 expression on CD8+ TEM (left) and percentage of TEM within PBL CD8+ T cells (right). j, Correlation between percentage of TEM in CD8+ T cells and TEM-NRP1 expression, both from post-ICB measurements, stratified by response. k,l, Within each CD8+ TEMRA subset, the percentage of cells expressing NRP1 (left) and TCF1 (right) (k) and the correlation between post-ICB TCF1 versus surface NRP1 expression (l), stratified by response; error bars, mean ± s.e.m.; statistical significance was determined by one-way ANOVA (c), two-tailed unpaired Student’s t-test (e,h), Wilcoxon test (i,k), and the log-rank test (f) and Pearson correlation analysis (g,j,l). Coefficient r was calculated and a t-test was performed to assess linear association (all P values are indicated).

Source data

The impact of CD8+ T cell expression of NRP1 on responses of patients with cancer to ICB immunotherapy was interrogated in a second cohort (cohort B, Supplementary Table 4) of 40 patients with advanced skin cancers who received single agent anti-PD1 or combinatorial ICB therapies ((1) anti-PD1 alone; (2) anti-PD1 + anti-CTLA4 and (3) anti-PD1 + anti-LAG3). PBL taken before and after 12 weeks of therapy was collected from 20 ICB responders and 20 ICB progressors, stratified based on RECIST 1.1 criteria. The significant increase in surface NRP1 expression on CD8+ TEM post-ICB therapy was only observed in patients with tumors that progressed, although their baseline NRP1 expression before treatment was comparable (Fig. 6i). By contrast, the size of the TEM pool was significantly reduced in the ICB progressors posttreatment (Fig. 6i), which inversely correlated with their surface NRP1 expression (Fig. 6j), suggesting the inability to maintain a sizable TEM pool may be associated with an unfavorable response to ICB therapy. While these observations aligned with the synergy observed between NRP1 deficiency and anti-PD1 treatment in mouse models (Fig. 1g–i), they further suggested that high NRP1 expression by TEM may contribute to acquired resistance to ICB immunotherapy. Of note, a reduction in TCF1 expression by CD8+ TEMRA was also significantly associated with ICB therapy resistance (Fig. 6k), in line with recent reports suggesting that intratumoral TCF1+CD8+ T cells are preferentially targeted for reinvigoration and correlate with better survival40. In addition, a trend toward an inverse correlation between expression of TCF1 and NRP1 by CD8 TEMRA was found in the tumor progressors, which was consistent with the notion that NRP1 promotes terminal exhaustion by negatively modulating TCF1 (Fig. 6l). Taken together, these data indicate that NRP1 was highly expressed on the human TEX in patients with cancer and negatively associated with the size of the memory T cell pool, disease prognosis and responsiveness to ICB therapy.


In summary, our findings highlight NRP1 as a previously uncharacterized immune checkpoint that affects the development and function of intratumoral CD8+ T cells in a cell-intrinsic manner and selectively affects the generation of memory precursors during an antitumor immune response. This may distinguish NRP1 from other IRs, such as PD1, CTLA4 and LAG3, that primarily affect effector T cell development and function. Although loss of NRP1 did not enhance the antitumor response to a primary tumor, it had a substantive effect on the development of T cell memory to subsequent tumor challenge. Nevertheless, CD8+ T cell–restricted NRP1 deletion in combination with ICB therapy did result in enhanced tumor clearance suggesting that targeting NRP1 may serve to enhance priming of the antitumor immune response in conjunction with ICB.

Mechanistically, NRP1 may affect memory T cell development in three ways. First, NRP1 expression correlates with Bcl2 loss, leading to substantive restraint on the frequency of antitumor CD8+ T cells over time, suggesting that NRP1 contributes to antigen-dependent maintenance of CD8+ T cell exhaustion and impaired memory differentiation41. Second, the absence of NRP1 sustains a TCF1+ pTEX subset that originates from phenotypically defined MPECs (CD127hiKLRG1lo) but are transcriptionally and epigenetically distinct from conventional memory precursors42. While intratumoral TCF1+ pTEX exhibit enhanced in vivo persistence, it is unclear whether they are capable of converting to bona fide memory T cells in vivo. Our data support this notion as E8ICreNrp1L/L mice exhibit enhanced protection from secondary tumor challenge, which was associated with an enlarged peripheral TCM cell pool following primary tumor resection on D12. Future lineage-tracing or in situ labeling approaches specifically targeting pTEX would help to definitively address this model. Finally, the increased proliferation observed in the presence of NRP1 may serve as a mechanism to compensate for exhaustion-driven CD8+ T cell dysfunction but at the expense of memory differentiation, which requires proliferative quiescence (slower turnover) while gaining stem cell-like properties (self-renewal and multipotency)43,44. Given the heterogeneity observed within TEX, NRP1 may identify a TEX subset that exhibits inflexibility in terms of potential for memory differentiation and reinvigoration following ICB. The apparent impact of NRP1 on CD8+ T cells in patients with advanced cancer is consistent with our mouse models of cancer wherein increased NRP1 expression correlated with a reduced TEM pool and decreased patient survival. Our findings identify NRP1 as a new immune checkpoint that affects the development of T cell memory to tumor antigens, suggesting that combinatorial blockade of PD1 and NRP1 may lead to more durable, systemic antitumor immunity and long-term remission in patients with cancer.


Cell lines and reagents

The B16.F10 melanoma cells (referred to as B16) were obtained from M.J. Turk (Dartmouth College, New Hampshire), the B16-gp100 cells from A.L. Rakhmilevich (University of Wisconsin, Madison, Wisconsin) and the B16-OVA cells from G. Delgoffe (University of Pittsburgh). The MC38 colon adenocarcinoma cells were obtained from J.P. Allison (M.D. Anderson Cancer Center, Houston, Texas). Tumor cell lines were cultured in RPMI1640 (for B16, B16-gp100 and B16-OVA) or DMEM (for MC38) supplemented with 10% fetal bovine solution (FBS), 100 units per ml penicillin, 100 μg ml−1 streptomycin, 2 mM glutamine, 1 mM pyruvate, 5 mM HEPES, 100 μM nonessential amino acids and 2-ME). The B16-gp100 and B16-OVA cells were cultured in the presence of 0.8 mg ml−1 geneticin (ThermoFisher Scientific). All cell lines and assay cultures were maintained at 37 °C and 5% CO2.

Human patients and specimen processing

Patients were seen in the Department of Otolaryngology at University of Pittsburgh Medical Center (UPMC) (cohort A), and the Department of Oncology at the SKCCC and Bloomberg–Kimmel Institute for Cancer Immunotherapy at Johns Hopkins University School of Medicine (cohort B). Cohort A consisted of a cohort of banked PBL samples from patients with HNSCC who had early stage (n = 23) or advanced (n = 25) disease for correlation with disease progression and survival. Cohort B from SKCCC consisted of banked PBL samples from patients with advanced skin cancers that were ICB therapy responders (complete or partial responsers) (n = 20), or progressors (PD) (n = 20).

Patients diagnosed with HNSCC electing to undergo treatment were offered the option to participate in the University of Pittsburgh Cancer Institute protocol for research. Patients signed an informed consent that was approved by the Institutional Review Board of the University of Pittsburgh.

Blood samples were obtained from patients at time of surgery or at time of clinic visit. Whole blood was centrifuged, and serum/plasma was immediately aliquoted in 5 × 1-ml amounts into sterile cryovials. Peripheral blood mononuclear cells are then separated via Ficoll Paque Plus (GE Healthcare). Viable peripheral blood mononuclear cells were frozen in FBS with 10% DMSO and stored in liquid nitrogen. Once the patient sample cohort was identified, cells were thawed and stained with designated antibodies followed by flow cytometric analysis.


The E8ICre mice were obtained from I. Taniuchi (Rikagaku Kenkyūsho, Japan). The Nrp1L/L mice were obtained from D. Cheresh (University of California, San Diego). The Id3-GFP reporter mice (Id3tm2.1Cmu; MGI, 5305600) were obtained from L. D’Cruz (University of Pittsburgh). pMel-1 mice (B6.Cg-Thy1a/Cy Tg(TcraTcrb)8Rest/J, stock No: 005023) and CD45.1 mice (B6.SJL-Ptprca Pepcb/BoyJ, stock no: 002014) were purchased from the Jackson Laboratory and bred in house. The E8ICreErt2GFP and Rosa26LSL.mAmetrine.2A.Nrp1 mice were generated in house. To generate E8ICreERT2.GFP transgenic mice, the plasmid used to generate the E8ICre.GFP mouse (kindly provided by I. Taniuchi)45 was altered by removing the Cre and inserting iCreERT2. To generate the Rosa26LSL.mAmetrine.2A.Nrp1 mice, the Rosa26.LSL.mAmetrine.2A.Nrp1 targeting construct was made using Rosa26 plasmids, pROSA26-PA (Addgene no. 21271) and pBigT (Addgene no. 21270)46, and mAmetrine-2A-Neuropilin-1 inserted in the ATG start codon within the Rosa26 locus downstream of the lox-stop-lox cassette. The linearized targeting construct was electroporated into JM8A3.N1 embryonic stem cells and neomycin resistant clones were screened by Southern blot analysis using EcoRV and MscI digestions for the 5′ and 3′ ends, respectively. Clones that were correctly targeted and greater than 85% normal diploid by karyotype analysis were injected into C57BL/6J blastocysts. Chimeric mice were mated to C57BL/6J mice and transmission of the targeted allele verified by PCR.

All animal experiments were performed in the American Association for the Accreditation of Laboratory Animal Care-accredited, specific-pathogen-free facilities (temperatures of ~18–23 °C (65–75 °F) with 40–60% humidity, 12-h light/12-h dark cycle) in the Division of Laboratory Animal Resources, University of Pittsburgh School of Medicine (UPSOM). Female and male mice of all the strains mentioned were used at 4–8 weeks of age. Animal protocols were approved by the Institutional Animal Care and Use Committees of University of Pittsburgh.

In vivo mouse tumor models

The E8ICre, Nrp1L/LE8ICre, E8ICreErt2GFP and E8ICreErt2GFPRosa26LSL.mAmetrine.2A.Nrp1 mice were inoculated with B16 cells (1.25 × 105, intradermally) or MC38 cells (5.0 × 105, subcutaneously). Tumors were measured every 3 d with a digital caliper in two dimensions (width and length) and presented as tumor volume (mm3, defined as the w2 × l/2). Antimouse PD1 (Clone 29 F.1A12) was injected intraperitoneally (i.p.) on D6, D9 and D12 at 100 μg per mouse. For attenuated Listeria monocytogenes infection, ActA-deficient L. monocytogenes was grown in tryptic soy broth (Sigma-Aldrich) supplemented with 50 µg ml−1 streptomycin (Sigma-Aldrich) at 37 °C until optical density (OD600) of 0.1 (108 CFU ml−1). Then 107 CFUs in 200 µl phosphate-buffered saline (PBS) were transferred intravenously into mice.

Surgical tumor excision and challenge

Surgical excision of primary tumors was performed as previously described47. Briefly, intradermal primary tumors (implanted in the right flank) were excised on D12 postinoculation, at size of approximately 7–10 mm in diameter. Mice were anesthetized with isoflurane and tumors were removed with a 2-mm perimeter of healthy skin. Incisions were closed with steel wound clips, and mice were given carprofen containing MediGel (ClearH2O) 24 h before and after surgery for pain management. Mice with recurrent primary tumor after surgery (<5%) were removed from study. For tumor challenge, 1.25 × 105 B16 or B16-gp100 cells were inoculated in the left flank on D30 or D60 posttumor resection.

pMel-T cell adoptive transfer

Bulk CD8+ T cells were purified from naive pMel-1×E8ICreThy1.1+Thy1.2+ and pMel-1×E8ICreNrp1L/LThy1.2+ mice by negative selection. Briefly, single cell suspensions from pooled spleen and lymph nodes were incubated with a cocktail containing biotinylated antibodies against CD4 (GK1.5), CD25 (PC61), CD49b (DX5), γδT (GL3), B220 (RA3-6B2), Gr1(RB6-8C5), CD19 (6D5), CD11b (M1/70), CD11c (N418), Ter119 (TER-119), IAb (KH74), CD16/32 (93) and CD105 (MJ7/18). Non-CD8 cells were removed by mixing the labeled cell suspension with the streptavidin-coated magnetic beads (Pierce) at 4 °C for 20 min, followed by separation in a magnetic field. The unbound CD8+ pMel-T cells (of purify >90%) were washed in sterile PBS (1×), with purity determined on a flow cytometer. The Nrp1+/+ and Nrp1−/− pMel-T cells were mixed at 1:1 ratio (postpurity correction, 4 × 105 in total) and injected (i.v.) into the CD45.1 recipient, followed by B16-gp100 tumor inoculation (1.25 × 105, i.d.) within around 18–24 h posttransfer.

LCMV infection

Mice were infected with 2 × 105 PFU of LCMV Armstrong (i.p.) or 4 × 106 PFU of Clone 13 (i.v.) at 8 weeks of age. Lymph nodes and spleens were taken at D8 post-infection(p.i.) for Armstrong, and at D30 p.i. for Clone 13 and processed for flow cytometry.

Flow cytometry and image flow cytometry

Single-cell suspensions were prepared from mouse spleens or tumors as previously described. Briefly, tumor-infiltrating leukocytes (TILs) were isolated by digesting B16 or MC38 tumors with Collagenase IV (1 mg ml−1) at 37 °C for 30 min. After red blood cell lysis, live/dead cell discrimination was performed using Live/Dead Fixable Aqua Dead Cell Stain Kit (Life Technologies). Fc block was performed by staining with anti-CD16/32 to avoid nonspecific binding. Surface staining was performed at 4 °C for 30 min in fluorescence-activated cell-sorting staining buffer (1× PBS/5% FBS/0.5% sodium azide) containing designated antibody cocktails. For transcription factor (Ki67, TCF1) and intracellular proteins (Bcl2, cleaved Caspase 3) staining, cells were fixed and permeabilized using with Foxp3 Transcription Factor Buffer Set (catalog 00–5523, Life Technologies), following the manufacturer’s instructions. To detect the cytokine-producing cells, cells were stimulated with PMA (100 ng ml−1) and Ionomycin (500 ng ml−1) for 5 h in the presence of Monensin (catalog 00-4505-51, Life Technologies) before cell surface staining, followed by standard intracellular staining procedure as above. All flow cytometry data were acquired on a BD LSR Fortessa analyzer (BD Biosciences) and analyzed by FlowJo software (v.10.5.3, Treestar, Inc.). For NFAT1 imaging cytometry, cells were fixed with 1.5% PFA for 10 min at room temperature and permeabilized in fluorescence-activated cell-sorting staining buffer containing 0.1% Triton X-100. Cells were then stained in staining buffer with anti-NFAT1 (NFAT1 (D43B1) XP, rabbit mAb no. 5861, Cell Signaling Technology) for 45 min, washed and then stained with an Alexa Fluor 647-conjugated antirabbit secondary antibody (goat antirabbit IgG (H + L) secondary Ab, Alexa Fluor Plus 647, ThermoFisher Scientific) for 30 min, and the nuclei were stained with DAPI. Data were collected on an Amnis ImageStream X Mark II Imaging Flow Cytometer and analyzed with IDEAS software (EMD Millipore).

BrdU (5-bromo-2’-deoxy-uridine) incorporation

To evaluate cell proliferation in vivo, mice implanted with B16.F10 tumors were injected with 2 mg BrdU 12 h before analysis. Tumors and lymph nodes were harvested, processed to yield single-cell suspension. Cells were stained with surface markers and Ki67 as described above, and then fixed and permeabilized with Cytofix/Cytoperm (BD Biosciences) and permeabilization buffer, respectively. After DNase I treatment at 37 °C for 45 min, cells were stained with allophycocyanin anti-BrdU for 45 min before analyzing on a flow cytometer.

Tamoxifen treatment

Tamoxifen working solution (10 mg ml−1) was prepared by dissolving the tamoxifen in the 5% ethanol-sunflower seed oil (v/v) by shaking overnight at 37 °C and stored at −20 °C, light protected. Mice were given 1.5 mg tamoxifen solution (approximately 75 mg kg−1 body weight) by i.p. injection daily for 5 consecutive days, before tumor implantation.


bpRNA-seq on adoptively transferred pMel-T cells was performed following a protocol developed in laboratory based on the Smart-Seq2 technology48. Briefly, Nrp1+/+ (CD45.2+Thy1.1+Thy1.2+) or Nrp1−/− (CD45.2+Thy1.2+) pMel-T cells were recovered from tumor, as well as the matched draining and NdLNs of the CD45.1 recipient at the designated day(s) post-B16-gp100 tumor challenge. Five hundred cells from either genotype were double sorted (purity >99.5%) directly to individual well of a 96-well plate containing 2 μl lysis buffer (0.2% Triton X-100 with RNase inhibitor at 2 U μl−1). The plate was spun down at 2,000g for 2 min and immediately proceeded with reverse transcription. A mixture (2 μl) of reverse-transcription primers and dNTP (1 mM) was added to each well followed by incubation at 70 °C for 3 min. Denatured templates and reverse-transcription primers were quickly spun down and added with 6 μl reverse-transcription master mix containing MgCl2 (9 mM), first-strand buffer (5×), Superscript II reverse transcriptase (10 U μl−1), DTT (5 mM), Betaine (1 M), RNase inhibitors (1 U μl−1) and template switch oligo (1 μM) followed by the reverse transcription reaction on a PCR cycler (42 °C for 90 min, ten cycles of 50/42 °C and 70 °C for 15 min, hold at 4 °C). An addition of 15-cycle complementary DNA amplification was performed following cDNA synthesis by the KAPA Hot Start II High-Fidelity DNA Polymerase. The amplified cDNA was purified using Ampure XP beads (at 0.6:1 bead to cDNA ratio) and eluted with 17.5 μl elution buffer. cDNA size (peak at roughly 1.5–2 kb) was verified with TapeStation5000 and quantified by the Qbit.

Sequencing libraries were prepared from 1 ng cDNA using the Nextera XT DNA Library Prep kit (Illumina FC-131-1096), following the manufacturer’s instructions. cDNA Libraries were quantified by the KAPA library quantification kit (KAPA KK4854) and size (peak at ~400 bp) verified on TapeStation1000. Ten diluted libraries (2 nM) were pooled and sequenced with the NextSeq 500/550 High Output v.2 kit using 75 bp single read.

In vitro T cell culture

Bulk CD8+ T cells were purified by negative selection from spleen and lymph nodes of naive mice (purity above 95%). They were initially plated at 1 × 106 per well in a 24-well plate, in the presence of plated-coated anti-CD3 (1 μg ml−1) and anti-CD28 (2 μg ml−1) and cultured in RPMI1640 medium supplemented with hIL2 (25 U ml−1) and 10% heat-inactivated FBS. For the ‘chronical stimulation’ scheme, cells were washed off the original plate 48 h after the first activation, counted and replated at 1 × 105 cells per well in U-bottom 96-well plate, with 200 μl fresh RPMI containing latex bead-coated anti-CD3 (3 μg ml−1) and anti-CD28 (5 μg ml−1), at 1:2.5 cell to beads ratio. This was followed by a second and/or third round stimulations every 2 d with fresh batches of anti-CD3/anti-CD28 beads. In each stimulation step, cells in each well were split in half by moving to a new well, with the amount of beads used and the number of viable cells activated being kept proportionally equal. As a control group, cells subjected to initial 48-h activation but without restimulation were rested in IL-2 containing medium for the same duration before analysis.

Transcriptomic (RNA-seq) analysis

RNA-seq reads were aligned to GRCm38/mm10 build of Mus musculus genome using STAR-2.5.2a. The Unique mapped reads were normalized using PORT ( and counts were converted to log2 counts per million, quantile normalized and precision weighted with the ‘voom’ function of the limma package49,50. A linear model was fitted to each gene, and empirical Bayes moderated t-statistics were used to assess differences in expression51. Heatmaps were created using R (v.3.5.1) package pheatmap_1.0.12 and plots for PCA were created using ggplot 2_3.1.0. P values from Bayes moderated t-tests were adjusted to control the global false discovery rate. Genes were called differentially expressed if they achieved an false discovery rate of 0.05 or less.

Gene ontology analysis and gene set tests

Cluster-specific, significantly enriched biological processes (gene ontology terms) were analyzed with QIAGEN’s Ingenuity Pathway Analysis (Qiagen, GSEA52,53 was performed against MSigDB (v.5.1) C7 database from which ID3 ko signature (GSE23568) was identified as one of the enriched signatures.

Gene signatures were generated for CXCR5 Pos (GSE76279) by identifying differentially expressed genes between CXCR5 positive versus CXCR5 negative groups from a published dataset33. For CD8-specific gene signature comparison, GSEA were run against collected datasets corresponding to traits of naive, effector, memory and exhaustion, as previously described in the LCMV infection model54. The normalized enrichment scores from GSEA for the LCMV genes signatures were plotted as a radar plot using R (3.5.1) package fmsb_0.6.3.

Statistical analysis

Statistics were performed using Graphpad v.8.0.2. The Mantel–Cox log-rank test was used for comparisons in the survival/tumor incidence analysis shown Figs. 1e,f,i and 6f and Extended Data Fig. 4e. Two-way analysis of variance (ANOVA) was used for comparing tumor growth curves, as in Fig. 1h and Extended Data Figs. 3e–h and 4e, and time course data as in Figs. 2f–h, 3e and 5e,f and Extended Data Fig. 5g. One-way ANOVA was used for the comparison among multiple (>2) groups, as in Fig. 1a; a two-tailed unpaired Student’s t-test was used for comparison between two groups (genotypes, patient subgroups or treatment), as in Figs. 2a,b,d, 5c and 6e and Extended Data Figs. 1b, 3d, 4c,d,g,i; whereas a paired Student’s t-test was specifically used for comparing two genotypes under the coadoptive transfer setting, such as in Fig. 3b–d,f and Extended Data Figs. 5d,e and 6b,d. The Wilcoxon test was used for Fig. 6i,k.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.