Tissue-resident memory CD8+ T cells amplify anti-tumor immunity by triggering antigen spreading through dendritic cells

Tissue-resident memory CD8+ T (Trm) cells mediate potent local innate and adaptive immune responses and play a central role against solid tumors. However, whether Trm cells cross-talk with dendritic cells (DCs) to support anti-tumor immunity remains unclear. Here we show that antigen-specific activation of skin Trm cells leads to maturation and migration to draining lymph nodes of cross-presenting dermal DCs. Tumor rejection mediated by Trm cells triggers the spread of cytotoxic CD8+ T cell responses against tumor-derived neo- and self-antigens via dermal DCs. These responses suppress the growth of intradermal tumors and disseminated melanoma lacking the Trm cell-targeted epitope. Moreover, analysis of RNA sequencing data from human melanoma tumors reveals that enrichment of a Trm cell gene signature associates with DC activation and improved survival. This work unveils the ability of Trm cells to amplify the breath of cytotoxic CD8+ T cell responses through DCs, thereby strengthening anti-tumor immunity.

C ytotoxic CD8 + T lymphocytes (CTL) play a pivotal role in providing effective antigen-specific immunity against tumors. Tumor-specific CTL responses are initiated at secondary lymphoid organs when naïve CD8 + T cells are activated by mature migratory dendritic cells (DCs) presenting tumor-derived antigens on MHC class I molecules 1,2 . Antigenspecific CD8 + T cells massively proliferate and differentiate into cytotoxic effector T (Teff) cells, which can then migrate to peripheral tissues and recognize cancer cells through their T-cell receptor (TCR). CTL destroy target tumor cells through mechanisms including release of granules containing perforin and granzymes and inducing FasL-mediated apoptosis. To achieve long-lasting anti-tumor immunity, it is necessary to establish memory CD8 + T-cell responses 3,4 . Classically, the circulating memory compartment consists of central-memory (Tcm) and effector-memory (Tem) CD8 + T cells 5 . Tcm cells circulate between secondary lymphoid organs and blood, whereas Tem cells circulate between blood and non-lymphoid tissues 5 . In contrast to these circulating subsets, tissue-resident memory CD8 + T (Trm) cells stably reside in lymphoid and non-lymphoid tissues where they provide potent local innate and adaptive immune responses 6 . The remarkable ability of Trm cells to mediate protective immunity has prompted the development of more potent vaccination strategies by eliciting Trm cells against infectious diseases 7,8 . Evidence supporting a central role of Trm cells in tumor immunosurveillance has recently emerged from animal models 9,10 . We and others have demonstrated that antigen-specific Trm cell responses mediate strong tissuerestricted immunity against cutaneous melanoma and other tumor models 9,[11][12][13] . However, the precise mechanisms by which Trm cells mediate enhanced anti-tumor immunity are poorly understood.
In human cancers, infiltration of CD103 + CD8 + T cells in solid tumors has been associated with longer survival in patients with breast, lung, endometrial, ovarian, cervical, urothelial and melanoma tumors [14][15][16][17][18][19][20][21][22][23] . Moreover, tumor-infiltration of Trm cells was recently associated with improved survival in melanoma patients that received an antibody blocking the inhibitory receptor PD-1 23 . T-cell residency across different tissues, including tumors, is defined by a distinctive gene expression program commanded by transcription factors, including Runx3, Blimp1, Hobit and Nur77 [24][25][26] . Tissue adaptation involves constitutive upregulation of CD69, CD49a and CD103, which sustain enhanced ability of Trm cells to become established in the tumor niche and better suited to fight tumors. CD69 is a C-type lectin expressed by Trm cells from most tissues that render these cells unresponsive to tissue egress signals such as sphingosine-1phosphate (S1P) by reducing the levels of S1P receptor 27 . CD49a, or α1β1 integrin, is an adhesion molecule that binds to the extracellular matrix proteins collagen and laminin 28 and distinguish Trm cells with higher cytotoxic potential in skin and melanoma tumors 29,30 . Trm cells confined to epithelial barriers also express CD103 (integrin αEβ7), which binds to E-cadherin and facilitates their interaction with epithelial cells 31 . In CD8 + T cells isolated from lung tumors, CD103 molecules have been shown to distribute preferentially near the immune synapse formed with the target tumor cell and to facilitate cytotoxic vesicle degranulation in an E-cadherin-dependent fashion 32 . Thus, tissue adaptation-related features and enhanced cytotoxicity may contribute to the superior protective potential of Trm cells in human solid cancers.
Complementary to their cytotoxic activity, Trm cells secrete large amounts of effector molecules, most prominently IFN-γ and TNF-α, which can activate other immune cells with anti-tumor potential. In the context of viral infections, IFN-γ produced by Trm cells triggers an innate-like alarm state characterized by the production of chemokines and antimicrobial molecules in the tissue and the recruitment of circulating memory CD8 + T cells 33,34 . In cancer, the presence of a tissue-resident gene signature is associated with higher density and enhanced cytotoxicity of CTLs infiltrating lung and breast tumors 16,35 . In viral models, Trm cell-derived IL-2 has been shown to promote upregulation of granzyme B in NK cells 36 . Additionally, antigenspecific activation of Trm cells leads to the production of TNF-α, which promotes rapid DC maturation and up-regulation of the lymph node homing chemokine receptor CCR7 36,37 . However, whether Trm cell activation in the tissue causes DC migration to draining lymph nodes and the subsequent initiation of protective CTL immune responses remains unknown. Particularly in the context of the tumor microenvironment, the innate-like capabilities of Trm cells and their potential cross-talk with DCs remain largely unexplored. We hypothesized that Trm cells cooperate with DCs to support anti-tumor immunity by initiating secondary T-cell responses against tumor-derived antigens.
Here, we demonstrate that skin Trm cell activation promotes maturation and trafficking to draining lymph nodes of migratory dermal DCs. Trm cell-mediated melanoma rejection leads to the spreading of circulating CTL responses against tumor-derived neo-and self-antigens that protects against intradermal tumors and disseminated melanoma lacking the Trm cell-targeted antigen. Transcriptional analysis of human melanoma tumors reveals that a Trm cell gene signature associates with DC activation and improved survival. This work highlights the ability of Trm cells to cross-talk with DCs and orchestrate the broadening of anti-tumor CTL immunity.

Results
Trm cells trigger maturation and migration of dermal DCs. We first aimed to study the potential interplay between Trm cells and migratory DCs in the skin. To this end, we generated ovalbumin (OVA)-specific skin Trm cells in mice using intradermal (i.d.) vaccination. This was followed by administration of an anti-CD8 antibody during the memory phase of the response (>4 weeks post vaccination), which efficiently deplete circulating CD8 + T cells ( Supplementary Fig. 1a, b), including circulating memory and effector OVA-specific CD8 + T cells in lymphoid and nonlymphoid tissues, as previously shown 11 . Then, depletionresistant Trm cells (Fig. 1a) were specifically activated by i.d. injection of the immunodominant OVA (257-264) peptide, readily producing IFN-γ and TNF-α within the first 6 h (Fig. 1a, b). Interestingly, we observed that skin XCR1 + conventional type 1 DCs (cDC1), also known as dermal DCs or DDCs (Fig. 1c) upregulated CD80, CD86, MHC class II and IL-12 molecules 24 h after Trm cell activation ( Fig. 1d-g). These data indicate that Trm cells induce maturation of dermal DCs, which are specialized in antigen cross-presentation and priming of CD8 + T cells 38 . Then, we analyzed the presence of skin migratory DCs in draining lymph nodes based on the expression of high levels of MHC class II, CD207 (langerin) and CCR7 (Fig. 2a). We observed a marked accumulation of different migratory DC subsets at 24 and 48 h after Trm cell activation, including dermal DCs, Langerhans cells (LCs) and CD11b + DCs (Fig. 2b). Among these subsets, dermal DCs displayed upregulated expression of the maturation marker CD86 (Fig. 2c). These results indicate that antigen-specific activation of Trm cells triggers maturation and migration to draining lymph nodes of skin-derived dermal DCs, revealing a cross-talk among these cells.
Trm cells spread CTL responses against tumor neo-antigens. Given the ability of Trm cells to activate migratory skin DCs and also mediate tumor-cell killing 11,36 , we hypothesized that ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12319-x antigen-specific Trm cell-mediated tumor rejection would lead to the generation of secondary responses against other tumorderived antigens, a phenomenon known as antigen spreading. To test this, mice bearing OVA-specific Trm cells were depleted from circulating CD8 + T cells and let during 8 weeks to replenish this compartment ( Supplementary Fig. 1c). This allows resetting the endogenous repertoire in terms of specificity while maintaining OVA-specific Trm cells in the skin. Then, mice were challenged i.d. with MC38 cells expressing the OVA (257-264) peptide (MC38-OTI), which were rejected by OVA-specific Trm cells. As controls, unvaccinated mice (no Trm) challenged with MC38-OTI and Trm cell-bearing mice challenged with parental MC38 were used. Secondary CD8 + T-cell responses raised against highly relevant neo-epitopes present in MC38 cell line 39 were analyzed 12 days later in tumor-draining lymph nodes (Fig. 3a). To this end, lymph node cells were ex vivo stimulated with neo-epitopes   carrying missense mutations MUT 1 (SIIVFNLL from Dpagt1 gene), MUT 2 (AQLANDVVL from Reps1 gene) and MUT 3 (ASMTNMELM from Adpgk gene) and the production of effector molecules was analyzed by intracellular staining and flow cytometry. In contrast to control groups, rejection of MC38-OTI mediated by OVA-specific Trm cells resulted in the expansion of CD8 + T cells specific to all neo-epitopes tested (Fig. 3b, c), detected as IFN-γ-producing CD8 + T cells, which also expressed high levels of CD44 (Fig. 3d). These results indicate that spreading of CD8 + T-cell responses to multiple antigens is triggered by Trm cell-mediated tumor rejection. Neo-epitope-specific CD8 + T cells displayed high expression of other effector molecules, such as TNF-α, granzyme B and IL-2 ( Fig. 3e, g), which is consistent with anti-tumor cytotoxic activity. Indeed, these mice were able to reject a re-challenge with MC38 cells, which express the neo-epitopes but cannot be recognized by OVA-specific Trm cells ( Fig. 3h-k).

Trm cells promote melanoma-antigen spreading through DCs.
To confirm these results in a relevant metastatic melanoma model, we used B16F10 cells, which are less immunogenic and express melanocyte-associated self-antigens, such as gp100. Favorably, responses against H-2 Kb-restricted gp100 (25)(26)(27)(28)(29)(30)(31)(32)(33) peptide can be tracked by transferring congenic TCR-transgenic CD8 + T cells from pmel-1 mice without the need to wait for the replenishment of the endogenous repertoire. Mice bearing OVAspecific Trm cells and devoid of circulating CD8 + T cells received i.v. transfer of pmel-1 CD8 + T cells (Fig. 4a). The following day, mice were challenged i.d. with B16F10 cells expressing the OVA (257-264) peptide (B16F10-OTI), which are rejected by OVAspecific Trm cells, as previously shown by us 11 . After 12 days, the generation of gp100-specific CTL responses was analyzed in the draining lymph nodes. Control groups were either left unchallenged (CTRL) or challenged with B16F10 parental cell line that do not activate OVA-specific Trm cells. As compared to control groups, only mice challenged with B16F10-OTI presented a significant expansion of gp100-specific CD8 + T cells (Fig. 4b, c), which produced IFN-γ after ex vivo peptide stimulation and displayed high expression of CD44 (Fig. 4d), indicating that they were efficiently primed. Since both B16F10-OTI and B16F10 cells express gp100 but only B16F10-OTI can be recognized by OVAspecific Trm cells, these results suggest that melanoma recognition by Trm cells triggers the spreading of CD8 + T-cell responses to tumor-derived antigens.
To explore whether cross-presenting dermal DCs mediate antigen spreading, we carried out similar experiments using Langerin-DTR mice, which allow the selective depletion of CD207/ langerin + dermal DCs and LCs from the skin after diphtheria toxin (DTx) administration [40][41][42] . Taking advantage of the relatively faster repopulation of dermal DCs (~2 weeks) derived from bone marrow precursors, in comparison to LCs (>4 weeks) that arise from slowly proliferating skin precursors, we performed the B16F10-OTI challenge in mice devoid of only LCs (single DTx administration two weeks before challenge) or depleted of both LCs and dermal DCs (continuous DTx administration starting one day before challenge) (Supplementary Fig. 2) 43,44 . Similar to wildtype mice, pmel-1 CD8 + T cells were clonally expanded following B16F10-OTI challenge in Langerin-DTR mice that were not treated with DTx (No DTx; DDC+/LC+) or received a single DTx dose (Single DTx; DDC+/LC-) (Fig. 4e, f). Interestingly, the expansion of pmel-1 CD8 + T cells was severely reduced in the case of mice depleted of both dermal DCs and LCs by continuous DTx administration (Continuous DTx; DDC-LC-), indicating that dermal DCs are necessary for antigen spreading induced by Trm cells. If dermal DCs directly present tumor-derived antigens to naïve CD8 + T cells in the lymph nodes remain to be determined.
Trm cell-induced CTL spreading suppresses melanoma growth. We next determined whether secondary CD8 + T-cell responses triggered by Trm cells were able to protect against B16F10 melanoma cells lacking OVA antigen and cannot be recognized by vaccination-induced Trm cells 11 . To this end, mice that rejected B16F10-OTI cells were injected 2-3 weeks later in the opposite flank with B16F10 melanoma cells (re-challenge) (Fig. 5a). Interestingly, these mice suppressed the growth of cutaneous tumors as compared to control mice that did not receive initial B16F10-OTI challenge (Fig. 5b,c), and therefore did not prime gp100-specific pmel-1 CD8 + T cells. In addition, no protection against B16F10 re-challenge was observed in mice that rejected B16F10-OTI melanoma but that did not receive transfer of pmel-1 CD8 + T cells (Fig. 5b, c), directly implicating the participation of primed pmel-1 CD8 + T cells in the anti-tumor effects observed. These results imply that Trm cell-mediated melanoma rejection triggers the spreading of CD8 + T-cell responses against melanoma-associated antigens, providing cross-protection against melanoma lacking Trm cell-targeted antigen. This can potentially be important to control highly heterogeneous tumors containing antigen-loss escape mutants.
To address the potential of Trm cell-induced gp100-specific CTL responses to protect against tumors disseminated in distant tissues, we repeated the previous experiment but substituted i.d. for i.v. B16F10 re-challenge to form disseminated pulmonary melanoma foci. Similar to i.d. re-challenge experiments, we observed that gp100-specific CTLs educated upon Trm cellmediated melanoma rejection suppressed the formation of melanoma foci, as compared to unchallenged or nontransferred controls (Fig. 5d). These data suggest that Trm cells can orchestrate the generation of systemic CTL responses, which have the potential to protect against metastatic tumors.
Trm cell-DC cross-talk in human melanoma. Finally, we set out to determine whether there is evidence of cross-talk between Trm cells and DCs in human cancer. Using previously described gene signatures for Trm cells, activated and immature DCs 35,45 , we Fig. 1 Skin Trm cell activation induces maturation of dermal DCs. C57BL/6 mice bearing OVA-specific skin Trm cells and depleted of circulating CD8 + T cells by administration of an anti-CD8 antibody were intradermally inoculated with control (CTRL) or OVA (257-264) (OVA) peptides to activate Trm cells. a, b OVA-specific CD45.1 + CD8 + T cells in the skin were analyzed 6 h later by intracellular cytokine staining and flow cytometry. a Representative pseudocolor dot-plot showing CD69 and CD103 expression. b Representative pseudocolor dot-plots and graph showing IFN-γ and TNF-α production by skin Trm cells. c-g DCs in the skin were analyzed 24 h after peptide stimulation. c Gating strategy used to identify skin DC subpopulations, including CD11b + DCs, Langerhans cells (LC) and dermal DCs (DDC) and representative histograms showing XCR1 expression in each subset. d-g Representative histograms (black: CTRL; red: OVA) and graphs showing of the expression of CD80, CD86, MHCII and IL-12 of each skin DC subset. For quantification, the geometric mean fluorescence intensity (MFI) was normalized relative to the average of the control group. Pooled data from two independent experiments, n = 7 mice in the group treated with control peptide and n = 8 mice in the group stimulated with OVA (257-264) peptide. Bars are the mean ± SEM. *p < 0.05, **p < 0.01, ***p < 0.001 by Mann-Whitney unpaired test analyzed tumor transcriptomic data of patients with cutaneous melanoma available within The Cancer Genome Atlas 46,47 . We found a striking correlation between Trm cell and activated DC signatures (r = 0.862), with a weaker correlation between Trm cells and immature DCs (r = 0.446; Fig. 6a). This is in keeping with our finding that Trm cells promote DC maturation in mice and suggests a similar process may occur in human melanoma. Whilst both Trm and activated DC enrichment correlated with better overall survival, this association was weaker for immature DCs (Fig. 6b). As these signatures are correlated, we carried out multivariable Cox regression analysis to estimate their individual contributions to patient survival, additionally correcting for total T-cell infiltrate using a previously published signature 48 and stage, showing Trm cell enrichment to remain a strong predictor of survival (Fig. 6c). These results suggest that that a similar Trm cell-DC cross-talk may occur in human melanoma.     Lymph node analysis Fig. 4 Trm cells promote melanoma-antigen spreading through dermal DCs. C57BL/6 and Langerin-DTR mice bearing OVA-specific skin Trm cells and depleted of circulating CD8 + T cells were intravenously transferred with gp100-specific CD90.1 + pmel-1 CD8 + T cells and one day later challenged intradermally with B16F10-OTI melanoma cells. Control groups were either unchallenged (CTRL) or challenged with or B16F10 melanoma cells that do not activate Trm cells. Analysis of gp100-specific pmel-1 CD8 + T cells was performed 12 days after tumor challenge in inguinal draining lymph nodes stimulated ex vivo with cognate gp100 (25)(26)(27)(28)(29)(30)(31)(32)(33) peptide to analyze IFN-γ production by intracellular cytokine staining. a Experimental scheme. b Representative pseudocolor dot-plots showing the frequency of gp100-specific pmel-1 CD8 + T cells and IFN-γ production. c Graph showing frequencies of gp100-specific pmel-1 CD8 + T cells. This reverse flow of information from adaptive to innate immune responses was initially described during anti-viral immune responses. A few reports have demonstrated that following Trm cell activation, a strong innate-like alarm state is induced in the tissue through the production of a plethora of effector molecules 33 . Among these, TNF-α has been shown to promote maturation of DCs 36 . However, whether Trm cellinduced DC maturation results in the generation of new CD8 + Tcell responses had not previously been addressed. The present study reveals that such cross-talk between Trm cells and DCs occurs in the context of anti-tumor immunity and, more importantly, that it results in the propagation of circulating antitumor CTL responses.
The results obtained in mouse models are supported by human data showing a strong correlation between Trm cell and activated DC gene signatures in tumors from melanoma patients. The broader anti-tumor CTL responses triggered by Trm cells can eventually underlie the association between Trm cell infiltration and higher density of CTLs observed in some human solid tumors 16,35 , as well as the superior predictive potential and better response to immunotherapy that Trm cells have in comparison to total CD8 + T-cell infiltration 16 . This mechanism may have broader implications because Trm cellinfiltration has been shown to predict better clinical outcome in other types of solid tumors [14][15][16][17][18][19][20][21][22][23] . On the other hand, crosspresenting migratory DCs are key players in the generation of anti-tumor T-cell immunity and their absence abolishes the rejection of immunogenic tumors and decreases the response to immune checkpoint blockade and adoptive T-cell therapy 1,[50][51][52] .
Emerging evidence indicates that effective anti-tumor immunity requires the coordinated action of tissue-resident and circulating T-cell compartments 12,53,54 . However, how these two compartments team-up to control tumors is poorly understood. It has been previously demonstrated that virus-specific Trm cells can recruit circulating bystander memory CD8 + T cells to the infection site after antigen recognition through the production of IFN-γ 33 . In tumor models, circulating Tcm cells have been shown to differentiate into Trm cells 12 . Here we show that Trm cells can increase the breadth of the circulating CD8 + T-cell repertoire.  Our findings suggest a novel mechanism by which resident and circulating T cells can collaborate to fight tumors. The mechanism described in this manuscript could be of particular relevance to control highly heterogeneous tumors, which represents a major challenge to oncological treatments and, in particular, immunotherapies. Indeed, tumor-cell exon sequencing has revealed that multiple regions inside the same tumor or different lesions in the same patient have divergent mutation patterns 55,56 and probably a differential expression of antigens. Consequently, Trm cells derived from different metastasis of the same patient have a high interlesional TCR diversity 57 . On the other hand, the adaptive immune system exerts a selective pressure on tumor cells, driving the survival of more resistant cancer cell subpopulations, a phenomenon known as immune editing 58 . As a result, cancer cell clones that do not express immune targeted antigens can escape immune control and form new tumors 58 . In this regard, Trm cell responses can drive the control of resistant clones, such as antigen-loss escape mutants, by broadening anti-tumor CTL responses against multiple tumorderived antigens, as shown here.
In summary, we propose that Trm cells represent a new orchestrator of anti-tumor immunity. Interestingly, it has been suggested that Trm cells are major targets of checkpoint blockade 57 and that checkpoint blockade promotes Trm cell formation in tumors 12 . Hence, we envision that the ability of Trm cells to increase the breath of anti-tumor T-cell immunity via DCs may play an important role in cancer immunotherapy. Accordingly, recent studies have evidenced the importance of the cross-talk between DCs and CD8 + T cells for effective cancer immunotherapy 59 . Moreover, a recent study has revealed that PD-1 blockade leads to the expansion of new tumor-reactive T-cell clones in patients with advanced skin cancer 60 . In consequence, the development of therapeutic approaches, such as vaccines, Tcell-based therapies and monoclonal antibodies, that boost the  Supplementary  Fig. 4.
Ex vivo intracellular cytokine staining. Inguinal lymph nodes were obtained 12 days after the tumor challenge and CD8 + T cells were stimulated ex vivo with the gp100 (25)(26)(27)(28)(29)(30)(31)(32)(33)  Tumor challenge. Mice were injected intradermally in the lower back skin close to the vaccination site with 50 μL of PBS containing 1 × 10 6 of tumor cells. Tumor growth was monitored by measuring perpendicular tumor diameters with calipers. Tumor volume was calculated using the following formula: V = (D x d 2 )/2 where V is the volume (mm 3 ), D is larger diameter (mm) and d is smaller diameter (mm). Mice were sacrificed when moribund or when the mean tumor diameter was ≥15 mm,) according to the approved ethical protocol. When indicated, mice were rechallenged with 1 × 10 6 B16F10 or MC38 cells in the contralateral site. For intravenous re-challenge 1 × 10 6 of B16F10 melanoma cells in 200 ul of PBS were inoculated through the tail vein. Mice were sacrificed two weeks later and lungs were obtained, washed in PBS and stored in 3 mL of Fekete´s solution. Lung foci quantification was performed taking pictures of the lungs on both sides (Canon EOS rebel T5) followed by quantification of dark melanoma foci.
RNA sequencing analysis. Upper quartile normalized RSEM expected RNA transcript counts and clinical data 47 from The Cancer Genome Atlas (TCGA) project were downloaded from the National Cancer Institute GDC PanCanAtlas project website (https://gdc.cancer.gov/about-data/publications/pancanatlas) and cutaneous melanoma cases (SKCM) filtered. Trm cell and DC gene signatures were previously described by Charoetntong et al. and Savas et al. 35,45 . A tumorinfiltrating T-cell signature was used as previously described by Danaher et al. 48 . Non-protein coding genes were removed from these signatures for consistency with TCGA data. For each signature, enrichment scores were calculated by taking the mean log 10 + 1 normalized expression of each gene, followed by z-score transformation. The correlation between Trm cell and DC gene signatures was evaluated by Pearson correlation.
Statistical analysis. Statistical analysis was performed using Graphpad Prism software (Graphpad Software Inc.). RNA sequencing and survival analyses were carried out in the R statistical programming environment. Mann-Whitney unpaired tests were performed between relevant groups. Statistical analyses for tumor growth was performed using two-way ANOVA Bonferroni post-hoc test. Error bars in figures indicate the mean plus SEM. Survival analysis by Cox regression was carried out with the "survival" package and Kaplan-Meier survival curves were drawn using the "survminer" package with patients grouped on the median value of each variable tested and with log-rank p values reported. Overall p value <0.05 was considered statistically significant; *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 and ****p ≤ 0.0001.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study are available from the authors on reasonable request.