CD4+ T cell-induced inflammatory cell death controls immune-evasive tumours

Most clinically applied cancer immunotherapies rely on the ability of CD8+ cytolytic T cells to directly recognize and kill tumour cells1–3. These strategies are limited by the emergence of major histocompatibility complex (MHC)-deficient tumour cells and the formation of an immunosuppressive tumour microenvironment4–6. The ability of CD4+ effector cells to contribute to antitumour immunity independently of CD8+ T cells is increasingly recognized, but strategies to unleash their full potential remain to be identified7–10. Here, we describe a mechanism whereby a small number of CD4+ T cells is sufficient to eradicate MHC-deficient tumours that escape direct CD8+ T cell targeting. The CD4+ effector T cells preferentially cluster at tumour invasive margins where they interact with MHC-II+CD11c+ antigen-presenting cells. We show that T helper type 1 cell-directed CD4+ T cells and innate immune stimulation reprogramme the tumour-associated myeloid cell network towards interferon-activated antigen-presenting and iNOS-expressing tumouricidal effector phenotypes. Together, CD4+ T cells and tumouricidal myeloid cells orchestrate the induction of remote inflammatory cell death that indirectly eradicates interferon-unresponsive and MHC-deficient tumours. These results warrant the clinical exploitation of this ability of CD4+ T cells and innate immune stimulators in a strategy to complement the direct cytolytic activity of CD8+ T cells and natural killer cells and advance cancer immunotherapies.

Adoptive cell transfer (ACT) studies using tumour-infiltrating lymphocytes from patients that are expanded ex vivo before their reinfusion provided initial proof-of-principle for the clinical efficacy of T cell immunotherapy 11 . The recent success of the immune checkpoint blockade (ICB) with monoclonal antibodies targeting the immunoregulatory receptors CTLA4 and PD1 led to the clinical breakthrough of T cell-directed immunotherapies 12 . The efficacy of ICB is mainly attributed to reactivation of CD8 + T cells that specifically recognize tumour antigens in the form of processed peptide epitopes presented by major histocompatibility complex class I (MHC-I) molecules on tumour cells. Both antigen presentation and MHC expression are upregulated by interferons (IFNs). Following antigen recognition, CD8 + T cells release cytolytic granules and IFNγ that initiate cell death. Despite its clinical efficacy, ICB is limited by the emergence of MHC-deficient and IFN-unresponsive tumour cell clones that escape recognition and destruction by CD8 + cytolytic T cells 4,5 .
There is emerging evidence that CD4 + T cells can also contribute to antitumour immunity, independent of their role as helpers and regulators of CD8 + cytolytic T cells 13 . A subset of CD4 + T cells develops cytolytic effector functions towards MHC-II-expressing tumour cells 14,15 . In addition, CD4 + T cells were shown capable of eradicating tumour cells that do not express MHC-II by mobilizing myeloid cells, which are specialized to process and present peptide epitopes on their MHC-II molecules [16][17][18] . The therapeutic potential of this indirect CD4 + T cell effector mechanism has, however, remained unclear. Moreover, the Article spatiotemporal dynamics and the mechanism of action of CD4 + T cells within the tumour tissue have not been fully explained.

Clinical relevance of immune evasion
Tumour cells can evade recognition and destruction by CD8 + T cells through MHC-I downregulation 19 . We reassessed the clinical relevance of this immune evasion mechanism in skin metastases of patients with melanoma by immunohistochemistry. Our results show very low MHC-I expression on melanoma cells in seven out of 20 samples that was associated with the absence of tumour-infiltrating CD8 + T cells ( Fig. 1a and Extended Data Fig. 1a,b). Expression of MHC-II was mostly restricted to stromal and immune cells at the invasive margins in 15 out of 20 samples, often in association with infiltrating CD8 + T cells. Only two out of 20 samples that were densely infiltrated with CD8 + T cells showed high expression levels of MHC-II on melanoma cells, whereas three out of 20 samples that lacked CD8 + T cell infiltrates barely expressed MHC-II ( Fig. 1a and Extended Data Fig. 1c).
We independently analysed the expression of MHC molecules in single-cell RNA-sequencing (scRNA-seq) data obtained from a different cohort of 20 patients with treatment-naïve melanoma metastases ( Fig. 1b-e) 20 . Transcriptional MHC-I downregulation in melanoma cells was evident in four out of 20 samples ( Fig. 1c and Extended Data Fig. 1d). MHC-II expression was largely absent in melanoma cells and restricted to antigen-presenting immune cells ( Fig. 1d and Extended Data Fig. 1e-g). Transcriptional downregulation of MHC-I on melanoma cells was associated with poor response to ICB (Fig. 1e). In aggregate, these results indicate that MHC-I downregulation in tumour cells is a frequent phenomenon during tumour evolution that favours immune escape. melanoma cells. f, Structure of recombinant adenovirus Ad-PT. g, Experimental protocol for ACT immunotherapy of established tumours in mice (Cy, C, cyclophosphamide; V, Ad-PT; T, TCRtg Pmel-1 CD8 + or TRP-1 CD4 + T cells; I, innate stimuli, polyI:C and CpG) and time point for flow cytometric analyses. h,i, Kaplan-Meier survival curves of mice bearing established B16 melanomas and treated either with CD4 ACT or CD8 ACT (h) or with the indicated components of the CD4 ACT protocol (i). NT, non-treated; CR, complete responders. i, **P = 0.0084. j,k, Immune cell composition (n = 2 biologically independent samples) ( j) and phenotype of endogenous and transferred (VT, CVTI, right columns) CD4 + T cells (k) in tumours treated as indicated (mean ± s.e.m. from n = 4 biologically independent samples). l-o, Graphical representation of the immune cell interaction phenotypes (left) and Kaplan-Meier survival curves (right) of mice bearing established Ciita-KO (l), Trp1-KO (m), CRISPR-ctrl (n) or Jak1-KO (o) melanomas and treated as indicated. o, ****P < 0.0001. p, Immune cell composition of tumours treated as indicated (mean ± s.e.m. from four biologically independent samples). q, Structure of recombinant adenovirus Ad-OVA. r, Graphical representation of the immune cell interaction phenotype of ovalbumin-expressing HCmel12 Jak1-KO cells (left) and Kaplan-Meier survival curves (right) of mice bearing established melanomas treated as indicated. ****P < 0.0001. Survival was statistically compared using a log-rank Mantel-Cox test. NS, not significant.

Establishment of a CD4 ACT model
To experimentally investigate how CD4 + T cell effector functions could be therapeutically directed against MHC-deficient tumours that evade CD8 + T cell immunity, we used enhanced green fluorescent protein (eGFP + ) TRP-1 CD4 + TCRtg T cells and Venus + Pmel-1 CD8 + TCRtg T cells for ACT immunotherapies 21,22 . Vaccination with the recombinant adenovirus Ad-PT encoding a Pmel/gp100-TRP-1 fusion protein capable of stimulating both CD4 + and CD8 + TCRtg T cells (Fig. 1f) allowed us to directly compare their antitumour efficacy and their mechanism of action under identical experimental conditions using the ACT therapy protocol established in our previous work 23 (Fig. 1g). This protocol includes chemotherapeutic preconditioning with cyclophosphamide, a procedure also used in clinical ACT approaches, and adjuvant injections of the immunostimulatory oligonucleotides polyI:C and CpG that activate innate immunity through TLR3 and TLR9, respectively 24 . Similar oligonucleotides are currently explored in early phase clinical trials 25,26 .
Initial experiments demonstrated that adoptively transferred TRP-1 CD4 + T cells expanded much less efficiently in lymph nodes, peripheral blood and spleens when compared with adoptively transferred Pmel-1 CD8 + T cells (Extended Data Fig. 2a,b). This observation is in line with the previously described intrinsic difference in the proliferative capacity between CD4 + and CD8 + T cells 27 . Despite their relatively poor in vivo expansion, adoptively transferred TRP-1 CD4 + T cells eradicated established B16 melanomas as efficiently as Pmel-1 CD8 + T cells (Fig. 1h and Extended Data Fig. 2c,d).
Cyclophosphamide pretreatment and innate immune stimulation were required in our CD4 ACT protocol to eradicate established tumours ( Fig. 1i and Extended Data Fig. 2e,f), similar to our findings for CD8 ACT 24 . Flow cytometric analyses of tumour-infiltrating CD45 + cells at day 7 after adoptive T cell transfer revealed a comparatively small subpopulation of adoptively transferred TRP-1 CD4 + T cells representing only 1% of immune cells in treated mice ( Fig. 1j and Extended Data Fig. 2g,h). The combination of cyclophosphamide pretreatment and adjuvant innate immune stimulation strongly promoted the differentiation of transferred and endogenous CD4 + T cells towards a T helper (T H ) type 1 (T H 1)-directed phenotype and prevented the accumulation of regulatory T cells ( Fig. 1k and Extended Data Fig. 2i). The adoptive transfer of T cells and the injection of innate immune stimuli independently increased the myeloid immune infiltrate that consisted predominantly of monocytes and macrophages. The full ACT protocol further increased the accumulation of monocytes (Fig. 1j).

Indirect tumour recognition
TRP-1 CD4 + T cells can eradicate B16 melanomas through direct recognition and cytolytic destruction 14 . However, most human melanoma cells do not express MHC-II molecules 10 (Extended Data Fig. 1c,e). We therefore investigated the ability of TRP-1 CD4 + T cells to control MHC-II-deficient tumour cells by disrupting the Ciita gene encoding the MHC-II transactivator in HCmel12 mouse melanoma cells using CRISPR-Cas9 gene editing (Extended Data Fig. 2j). As controls, we also generated HCmel12 Trp1-knockout (Trp1-KO) cells that lack expression of the CD4 + T cell target antigen (Extended Data Fig. 2k). In vitro experiments confirmed that TRP-1 CD4 + T cells can directly respond to MHC-II-expressing HCmel12 cells in an antigen-specific manner (Extended Data Fig. 2l), but are more efficiently activated indirectly by MHC-II+ dendritic cells pulsed with HCmel12 cell lysates (Extended Data Fig. 2m). Of note, 40% of TRP-1 CD4 + T cells isolated from CD4 ACT-treated mice produced IFNγ following stimulation with tumour lysate-pulsed dendritic cells, confirming their T H 1 phenotype. Subsequent in vivo experiments showed that TRP-1 CD4 + T cells were able to eradicate established MHC-II-deficient, but not TRP-1-deficient HCmel12 melanomas (Fig. 1l,m and Extended Data Fig. 2n-p). Treatment of tumours consisting of HCmel12 CRISPR-control (ctrl) and HCmel12 Trp1-KO mixtures demonstrated that TRP-1 CD4 + T cells also exerted substantial bystander killing activity, but could not prevent the outgrowth of HCmel12 Trp1-KO cells in all mice (Extended Data Fig. 2q,r). Furthermore, antibody-mediated depletion experiments confirmed that the treatment efficacy of TRP-1 CD4 + T cells did not require the presence of CD8 + T cells (Extended Data Fig. 2s,t). Thus, TRP-1 CD4 + T cells can indirectly recognize and kill MHC-II-deficient tumour cells in the absence of CD8 + T cells.

Eradication of immune-evasive tumours
In subsequent experiments, we took advantage of the unique properties of HCmel12 melanoma cells that constitutively lack MHC-I and MHC-II expression unless exposed to IFNs. CRISPR-Cas9-mediated disruption of the Jak1 gene encoding for a central mediator of the IFN signalling pathway leads to IFN-unresponsive and completely MHC-deficient tumour cells (Extended Data Fig. 3a). Robust in vivo growth of HCmel12 Jak1-KO melanoma cells required antibody-mediated depletion of natural killer (NK) cells before tumour inoculation, in line with the known ability of NK cells to directly recognize and kill MHC-I-deficient cells by cytolysis. This experimental setting allowed us to investigate the capacity of adoptively transferred TRP-1 CD4 + T cells to indirectly recognize and kill IFN-unresponsive, MHC-deficient tumour cells independent from their ability to directly target and lyse MHC-II-expressing tumour cells and to provide help for the cytolytic activity of CD8 + T and NK cells. Our results demonstrate that adoptively transferred TRP-1 CD4 + T cells can indirectly eradicate established tumours that evade CD8 + T cell control in the absence of NK cells (Fig. 1n,o and Extended Data Fig. 3b,c).
Subsequent flow cytometric analyses revealed that adoptively transferred Pmel-1 CD8 + T cells represented a much larger subpopulation of tumour-infiltrating immune cells when compared to adoptively transferred TRP-1 CD4 + T cells, both in HCmel12 CRISPR-ctrl and HCmel12 Jak1-KO tumours, consistent with their differential in vivo expansion dynamics. Nevertheless, both CD4 and CD8 ACT therapies substantially increased the number of myeloid cells in HCmel12 CRISPR-ctrl and HCmel12 Jak1-KO tumours ( Fig. 1p and Extended Data Fig. 3d,e). The increased immune cell infiltrate was less pronounced in MHC-deficient HCmel12 Jak1-KO tumours when compared to HCmel12 CRISPR-ctrl tumours, in line with our observation in patient samples ( Fig. 1a and Extended Data Fig. 1a,b).
Using ovalbumin as a second tumour antigen model (Fig. 1q), we recapitulated the different in vivo expansion dynamics of adoptively transferred ovalbumin-specific OT-II TCRtg CD4 + T cells and OT-I TCRtg CD8 + T cells (Extended Data Fig. 3f,g). Again, ovalbumin-specific OT-II TCRtg CD4 + T cells were able to eradicate ovalbumin-expressing MHC-deficient HCmel12 Jak1-KO tumours, whereas ovalbumin-specific OT-I TCRtg CD8 + T cells were ineffective ( Fig. 1r and Extended Data Fig. 3h,i). Taken together, these results confirmed that a few CD4 + effector T cells can indirectly eradicate established IFN-unresponsive, MHC-deficient tumours that evade CD8 + T cell immunity independent of NK cells.

Intratumoural CD4 + T cell dynamics
We proposed that CD4 + effector T cells are efficiently activated in tumour tissues by antigen-presenting immune cells that constitutively express MHC-II, whereas CD8 + T cells require MHC-I-restricted antigen presentation by tumour cells. The expression pattern of MHC molecules in tumour tissues should therefore also determine the spatial distribution and migratory behaviour of adoptively transferred CD4 + and CD8 + T cells. To address these hypotheses, we generated amelanotic (Tyr-KO) HCmel12 CRISPR-ctrl and HCmel12 Jak1-KO cells that express tag-blue fluorescent protein (tagBFP) for immunofluorescence microscopy imaging (Extended Data Fig. 4a). Confocal microscopy showed only very few adoptively transferred eGFP + TRP-1 CD4 + T cells in local clusters at the invasive margin of established amelanotic tagBFP-labelled

Article
HCmel12 CRISPR-ctrl tumours. By contrast, large numbers of Venus + Pmel-1 CD8 + T cells infiltrated both the invasive margin and the tumour centre (Extended Data Fig. 4b,d), consistent with the flow cytometry results (Fig. 1j,p). TRP-1 CD4 + T cells also locally clustered at the invasive margin of MHC-deficient HCmel12 Jak1-KO tumours, whereas Pmel-1 CD8 + T cells only infiltrated the invasive margin but not the tumour centre (Extended Data Fig. 4c,e). These results showed that the distribution of CD8 + T cells in tumour tissues of our preclinical model depends
Nature | Vol 618 | 29 June 2023 | 1037 on MHC-I expression, reminiscent of the spatial distribution of CD8 + T cells in human melanomas (Extended Data Fig. 1b). Intravital two-photon microscopy at the invasive margin confirmed the differential intratumoural localization of adoptively transferred CD4 + and CD8 + T cells and revealed substantial differences in their migratory behaviour. In HCmel12 CRISPR-ctrl tumours, CD4 + T cells arrested both in the stromal and the tumoural compartment within the invasive margin, whereas CD8 + T cells remained highly motile in the stroma and only arrested in the tumoural compartment ( Fig. 2a-c and Supplementary Videos 1 and 2). In MHC-deficient HCmel12 Jak1-KO tumours, CD4 + T cells showed no changes in their migratory behaviour in the stroma and a slightly increased motility in the tumoural compartment of the invasive margin. By contrast, CD8 + T cells failed to arrest at all and were always highly motile in MHC-deficient HCmel12 Jak1-KO melanomas . Together, these observations indicated that CD4 + T cells can interact with MHC-II + immune and tumour cells at the invasive margin, whereas CD8 + T cells predominantly interact with MHC-I + tumour cells.
We confirmed the fundamental difference in the spatial distribution and migratory behaviour of CD4 + and CD8 + T cells in tumour tissues using amelanotic (Tyr-KO) MHC-deficient HCmel12 Jak1-KO tumours that express tagBFP-Ova. Very few adoptively transferred ovalbumin-specific dsRed + OT-II CD4 + TCRtg T cells clustered locally within the tumour invasive margin and arrested both in the stromal and the tumoural compartment. By contrast, large numbers of ovalbumin-specific Venus + OT-I CD8 + TCRtg T cells infiltrated only the tumour invasive margin but not the tumour centre and were always highly motile (Extended Data Fig. 4f,g). These observations confirmed that CD4 + T cells preferentially interact with MHC-II + antigen-presenting cells at the invasive tumour margin, whereas CD8 + T cells require MHC-I expression on tumour cells to exert their effector functions in vivo.

MHC-II-restricted antigen recognition
A likely interaction partner for CD4 + T cells are dendritic cells due to their ability to efficiently ingest and process tumour antigens for MHC-II-dependent antigen presentation 28-30 . To visualize antigen-specific interactions between TRP-1 CD4 + T cells and MHC-II + CD11c + antigen-presenting cells, we further generated amelanotic (Tyr-KO) HCmel12 Trp1-KO cells expressing tagBFP, injected them into opposite legs of CD11c-Venus transgenic mice that harbour fluorescent antigen-presenting immune cells 31 and treated established tumours with adoptively transferred eGFP + TRP-1 CD4 + T cells ( Fig. 2g and Extended Data Fig. 5a,b). Confocal microscopy revealed local accumulations of eGFP + TRP-1 CD4 + T cells in association with MHC-II-expressing CD11c-Venus + immune cells within tumour invasive margins in HCmel12 CRISPR-ctrl but not in HCmel12 Trp1-KO tumours (Extended Data Fig. 5c-f). Tumour cells surrounding clusters of CD11c-Venus + immune cells with CD4 + T cells upregulated the expression of MHC-II exclusively in mice bearing HCmel12 CRISPR-ctrl tumours, consistent with the notion that CD4 + T cells were activated locally and secreted IFNγ in an antigen-dependent manner.

Fig. 3 | CD4 + effector T cells and innate immune stimulation promote the recruitment and IFN-dependent activation of monocytes to eradicate established tumours. a,
Experimental protocol for scRNA-seq analyses of tumour-infiltrating CD11b + Ly6G − cells. b, Visualization and dimensionality reduction of single-cell transcriptomes from CD4 ACT-treated and non-treated (NT) mice using UMAP. c,d, UMAP plots with cell types assigned using SingleR (c) and z-score for the hallmark IFN gamma response gene set (MSigDB) of each cell (d). e, RNA velocity projected on UMAP plots for monocytes and macrophages (Mono-macs) of CD4 ACT-treated tumours. Arrows point towards the predicted course of cell maturation dynamics. Arrow sizes indicate the strength of predicted directionality. f, Immune cell composition of HCmel12 tumours treated as indicated (mean ± s.e.m. from n = 6 biologically independent samples). g,j, Experimental treatment protocol to investigate the impact of innate stimuli (g) or IFNγ-blockade ( j) on myeloid cell activation and tumour control. h,k, Percentage of intratumoural iNOS + mono-macs and neutrophils (mean ± s.e.m. from n = 5-7 biologically independent samples). h, ****P < 0.0001, *P = 0.0111, ***P = 0.0005; k, ****P < 0.0001, ***P = 0.0002). i,l, Kaplan-Meier survival curves of mice bearing established HCmel12 CRISPRctrl tumours, treated as indicated (i, ****P < 0.0001; l, **P = 0.0053). Means between groups were statistically compared using a one-way ANOVA with Tukey post hoc. Survival was statistically compared using log-rank Mantel-Cox test. NT, non-treated; C, cyclophosphamide; V, Ad-PT; T, TCRtg TRP-1 CD4 + T cells; I, innate stimuli, polyI:C and CpG; CR, complete responders.

Article
Intravital two-photon microscopy demonstrated that eGFP + TRP-1 CD4 + T cells preferentially arrested and engaged in long-lasting close interactions with CD11c-Venus + immune cells only in HCmel12 CRISPR-ctrl tumours, but not in HCmel12 Trp1-KO tumours (Fig. 2h,i,Extended Data Fig. 5g,h and Supplementary Video 5). Antibodymediated blockade of MHC-II-restricted antigen presentation abrogated the interaction between CD4 + T cells and CD11c-Venus + immune cells, confirming the specificity of our findings ( Fig. 2j-l and Supplementary Video 6). These observations demonstrate that CD4 + effector T cells cluster with CD11c-Venus + immune cells at the tumour invasive margin where they maintain prolonged antigen-specific and MHC-II-restricted interactions that enable them to eradicate MHC-deficient tumours. Of note, CD4 + effector T cells also clustered with MHC-II + dendritic antigen-presenting cells and macrophages in human melanomas (Extended Data Fig. 6).

Recruitment of IFN-activated monocytes
Next, we investigated how a comparatively small subpopulation of CD4 + effector T cells can eradicate large established tumours. We proposed that adoptively transferred CD4 + T cells and injections of synthetic nucleic acids direct a strong T H 1-associated pathogen defence mechanism that also engages mononuclear phagocytes towards tumour destruction. To explore this hypothesis, we performed scRNA-seq analyses of sorted CD11b + Ly6G-tumour-infiltrating mononuclear phagocytes from mice treated with our CD4 ACT protocol and from non-treated controls. Dimensionality reduction, visualization using uniform manifold approximation and projection (UMAP) and cell type annotation using SingleR showed a clear separation between the monocyte-macrophage clusters derived from CD4 ACT-treated and non-treated tumours ( Fig. 3a-c). Differential gene expression and gene set enrichment analyses revealed a strong activation of IFN-response genes on therapy ( Fig. 3d and Extended Data Fig. 7a,b).
Unsupervised Leiden clustering of the monocyte-macrophage lineage for CD4 ACT-treated and non-treated groups separated four and seven cell states, respectively. The four cell states in non-treated mice express marker genes characteristic for immature monocytes (NT0), monocyte-derived dendritic cells (NT1), monocyte-macrophage effector cells (NT2) and mature monocytes (NT3). The seven cell states in CD4 ACT-treated mice represent IFN-activated counterparts of the intratumoural monocyte-macrophage network found in non-treated controls (Extended Data Fig. 7c,d). Computation of the

Fig. 4 | CD4 + effector T cells cooperate with activated iNOS-expressing tumouricidal monocytes and macrophages to orchestrate remote inflammatory cell death of MHC-deficient and IFN-unresponsive tumours.
a, Graphical representation of interaction phenotypes of indicated HCmel12 variants. b, Experimental treatment protocol to study the impact of chemical iNOS inhibition using L-NIL on CD4 ACT-mediated tumour control. c, Kaplan-Meier survival graphs of mice bearing established HCmel12 Ciita-KO melanomas (left) or HCmel12 Jak1-KO melanomas (right) and treated as indicated (NT, nontreated; CR, complete responders; **P = 0.0033). Survival was statistically compared using a log-rank Mantel-Cox test. Means between groups were statistically compared using a one-way ANOVA with Tukey post hoc. d, Experimental protocol to assess the ability of the inflammatory mediators TNF, IFNγ and the nitric oxide donor SNAP to induce melanoma cell death. e,f, Percentage of cell death in mouse (e) and human (f) melanoma cells treated as indicated (mean ± s.e.m. from 2-3 technical replicates). g, Graphical summary of inflammatory cell death induction of MHC-deficient and IFN-unresponsive tumours by CD4 + T cells in cooperation with iNOS-expressing myeloid cells.
RNA velocity and pseudotime inference revealed a dynamic development of Ly6c-hi inflammatory immature monocytes towards phenotypes of IFN-activated monocyte-derived dendritic cells (ACT1, antigen-presentation phenotype), monocyte-macrophage effectors (ACT2a-c, tumouricidal phenotype) and Ly6c-lo mature monocytes (ACT3a,b, patrolling phenotype) ( Fig. 3e and Extended Data Fig. 7e,f). Flow cytometric analyses of tumour-infiltrating immune cells at days 2, 5 and 8 after adoptive CD4 + T cell transfer confirmed the dynamic recruitment of Ly6c-hi immature monocytes into the tumour microenvironment on CD4 ACT therapy and their differentiation into iNOS-expressing tumouricidal mononuclear phagocytes ( Fig. 3f and Extended Data Fig. 7g,h). Together, the flow cytometric and transcriptomic analyses indicated that CD4 + T cells and innate immune stimuli reprogramme the myeloid network in treated tumours through the recruitment of immature monocytes that acquire IFN-activated cellular states and dynamically differentiate towards MHC-II antigen-presenting and iNOS-expressing tumouricidal effector phenotypes.
Our initial data showed that CD4 + effector T cells and innate immune stimulation independently promoted the recruitment of immature monocytes into the tumour microenvironment ( Fig. 1j and Extended Data Fig. 2h). Next, we asked whether CD4 + T cells and innate immune stimuli synergized on a quantitative or qualitative level for the acquisition of tumouricidal monocyte effector functions. Omitting innate stimuli from our combined ACT therapy regimen reduced the recruitment of neutrophils, but not of immature Ly6C-hi monocytes (Extended Data Fig. 8a,b). However, both CD4 + T cells and innate immune stimuli were indispensable for full iNOS induction in the recruited monocytes (Fig. 3g,h and Extended Data Fig. 8b). Functionally, the synergism of the combined therapy was required locally in tumour tissues for the eradication of established tumours, leading to a striking increase in tumour-free survival ( Fig. 3i and Extended Data Fig. 8c-e). We proposed that the release of IFNγ was responsible for the CD4 + T cell-driven qualitative enhancement of tumouricidal monocyte effector functions on the molecular level. In support of this hypothesis, we found that antibody-mediated neutralization of IFNγ did not influence the absolute number of tumour-infiltrating monocytes and neutrophils, but significantly reduced the frequency of iNOS-expressing monocytes (Fig. 3j,k and Extended Data Fig. 8f,g). IFNγ was essential to eradicate established tumours ( Fig. 3l and Extended Data Fig. 8h).

Inflammatory tumour cell death
CD4 + T cell-derived IFNγ can either act on tumour cells or through IFN-dependent activation of iNOS-expressing tumouricidal myeloid cells 18,[32][33][34] (Fig. 4a). Treatment with the highly specific iNOS inhibitor (N6-(1-iminoethyl)-l-lysine, L-NIL) abrogated the ability of CD4 ACT treatment to control IFN-unresponsive, MHC-deficient HCmel12 Jak1-KO tumours, but had no effect on CD4 ACT treatment of IFN-responsive, MHC-deficient HCmel12 Ciita-KO tumours (Fig. 4b,c and Extended Data Fig. 9a,b). Transient antibody-mediated depletion of CCR2 + monocytes impaired the efficacy of CD4 ACT therapy to a greater extent than the depletion of Ly6G + neutrophils, supporting a predominant role of iNOS-expressing monocytes and macrophages for tumour eradication (Extended Data Fig. 9c-e). In aggregate, these results indicated that the ability of adoptively transferred CD4 + T cells to indirectly eradicate IFN-unresponsive, MHC-deficient tumour cells involved the remote action of nitric oxide released by IFN-activated tumouricidal myeloid cells.
Our results raised the question as to how myeloid cell-derived nitric oxide contributes to death of IFN-unresponsive tumour cells. Recent data that explained the cytokine driven immunopathology in patients with COVID-19 revealed an inflammatory mode of cell death driven by the concerted action of IFNγ, tumour necrosis factor (TNF) and nitric oxide 35 . In experiments inspired by these observations, we found that the nitric oxide donor S-nitroso-N-acetylpenicillamine (SNAP) effectively induced death of both IFN-responsive HCmel12 Ciita-KO and IFN-unresponsive HCmel12 Jak1-KO melanoma cells in vitro (Fig. 4d,e and Extended Data Fig. 9f,g). IFNγ and TNF were not effective alone and only induced death of IFN-responsive HCmel12 Ciita-KO melanoma cells when used in combination. The ability of the inflammatory mediators to induce cell death was fully recapitulated in a panel of IFN-responsive and IFN-unresponsive human melanoma cell lines ( Fig. 4f and Extended Data Fig. 9h). These in vitro results demonstrated that IFNγ sensitizes IFN-responsive melanoma cells towards TNF-induced cell death and suggested that myeloid cell-derived nitric oxide contributes to efficient inflammatory cell death of IFN-unresponsive melanoma cells (Fig. 4g).

Discussion
CD4 + T cells have primarily been perceived as helper cells for the activation of CD8 + effector T cells 36 , which kill tumour cells by direct cytolysis. In our work, we showed that CD4 + effector T cells are also able to independently eradicate established tumours as efficiently as CD8 + cytolytic T cells. Using intravital microscopy, we found that CD4 + and CD8 + effector T cells differ fundamentally in their mode and their site of action in tumour tissues. Only very few CD4 + effector T cells, representing 1% of tumour-infiltrating immune cells, locate at the tumour invasive margins where they interact with CD11c + MHC-II + antigen-presenting immune cells and indirectly eliminate tumours. By contrast, much larger numbers of CD8 + cytolytic T cells infiltrate the tumour centre where they directly target and kill MHC-I-expressing tumour cells.
Further innate immune stimulation boosted the T H 1-directed differentiation of CD4 + T cells, increased the recruitment of immature monocytes into the tumour microenvironment and supported their IFN-dependent activation and differentiation towards antigen-presenting and iNOS-expressing tumouricidal effector phenotypes. Together, CD4 + T cells and IFN-activated mononuclear phagocytes initiate an indirect inflammatory tumour cell death process that acts from the 'outside-in' and can be abrogated by neutralizing IFNγ. This unique mode of action operates in parallel to and independent of the direct cytolytic activities of NK cells 37 and enables the eradication of MHC-deficient and IFN-unresponsive tumours that evade direct recognition and destruction by CD8 + cytolytic T cells (Extended Data Fig. 10).
The ability of lymphocytes to cooperate with mononuclear phagocytes for immune defence was first observed in experiments with bacterial infections by investigators trying to understand immune resistance of mice to pathogens and tumours more than 50 years ago 38,39 . We faithfully recapitulate in our experimental model the cellular and molecular mechanisms underlying this cooperation. MHC-II-dependent antigen presentation by myeloid cells, dynamic IFN-dependent activation and differentiation of the monocyte-macrophage and dendritic cell lineages 40 , and the induction of remote cell death through inflammatory mediators may represent shared immune defence mechanisms that critically contribute to the control of tumours 33,34,41 and pathogens 35,42-46 .
Our results have important implications for the future of patient care, as ACT immunotherapies using CD4 + T cells have already been successfully implemented in clinical studies. ACT with NY-ESO-1 specific CD4 + T cells were used to successfully treat metastatic melanoma 47 and ACT with CD4 + T cells genetically engineered to express an MHC-II-restricted T cell receptor specifically recognizing the cancer germline antigen MAGE-A3 demonstrated clinical efficacy 48 . Our results strongly support the clinical development of ACT-based therapeutic strategies that include appropriate activating stimuli for myeloid cells to unleash the full potential of CD4 + T cell effector functions against immune-evasive tumours.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, Article acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-06199-x. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Patient biopsies
Skin metastases of 20 patients with melanoma (clinical stage III-IV), obtained during routine histopathological diagnostic procedures at the Department of Dermatology of the University Medical Centre Magdeburg, were analysed by immunohistochemistry for the expression of MHC-I (1:100),  and CD8 (undiluted), in addition to the melanoma markers MART-1 (undiluted), gp100 (undiluted), S-100 (undiluted) and Sox10 (undiluted) using the automated Ventana BenchMark platform and standard protocols. These studies were performed in the context of routine clinical workup and were approved by the ethics committee of the Otto-von-Guericke University Hospital Magdeburg (approval number 162/20).
Cell suspensions derived from immunotherapy naive tumour biopsies of 20 melanoma metastases in skin (n = 5), subcutis (n = 4) and lymph nodes (n = 11) from 19 patients (clinical stage IIIB-IV) were analysed for their messenger RNA expression profile in single cells. Methods for tumour dissociation, library construction, scRNA-seq data acquisition and analysis were described previously 20 . This study was approved by the UZ Leuven Medical Ethical Committee and written consent obtained from all patients. The immune cells were distinguished from other tumour microenvironment cells by high immune signature score and low copy number variation score. Next, the cells were reclustered and annotated using SingleR. The MHC-I and MHC-II gene signature scores were measured using the AUCell R package 49 .

MILAN (mIHC)
Multiplex immunofluorescent images were generated by sequential immunostaining and antibody removal according to the published MILAN protocol 50 from human melanoma biopsies as described previously 50 . (1 µg ml −1 ) for staining keratinocytes, effector T cells, MHC-II expressing myeloid cell subsets, melanoma cells and vessels respectively, is shown. Image analysis was performed as described previously 51 . Briefly, stains were visually evaluated for quality by an experienced pathologist. Flat field correction was performed using a custom implementation of the methodology 52 . Consecutive staining rounds were registered using a previously described algorithm 53 . Tissue autofluorescence was subtracted using a baseline image stained only with a secondary antibody.

Mice
Mice were housed in an ambient temperature-and humidity-controlled environment on a 12-h light/dark cycle to mimic natural conditions. Wild type C57BL/6J mice were purchased from Janvier or Charles River. The T cell receptor-transgenic Pmel-1 (B6.Cg-Thy1a/ Cy Tg(TcraTcrb)8Rest/J), TRP-1 (B6.Cg-Rag1tm1Mom Tyrp1B-w Tg (Tcra,Tcrb) 9Rest/J), OT-I (C57BL/6-Tg(TcraTcrb)1100Mjb/J) and OT-II (B6.Cg-Tg(TcraTcrb)425Cbn/J) mice, and the fluorescent B6-eGFP (C57BL/6-Tg (UBC-GFP) 30Scha/J) and CD11c-eYFP (B6. Cg-Tg (Itgax-Venus) 1Mnz) mice were purchased from Jackson Laboratories. Pmel-1-Venus mice were generated by crossing CAG-Venus mice with Pmel-1 mice. TRP-1-eGFP mice were generated by crossing B6-eGFP mice into the TRP-1-deficient Rag1-KO background of TRP-1 mice. OT-I-Venus mice were generated by crossing CAG-Venus mice with OT-I mice. OT-II-dsRed were generated by crossing OT-II mice with hCD2-dsRed mice (kindly provided by C. Halin). All transgenic mice were bred in house. Age matched cohorts of tumour developing mice were randomly allocated to the different experimental groups. All animal experiments were conducted with male mice on the C57BL/6 background under specific pathogen-free conditions in individually ventilated cages according to the institutional and national guidelines for the care and use of laboratory animals with approval by the Ethics Committee of the Office for Veterinary Affairs of the State of Saxony-Anhalt, Germany (permit licence numbers 42502-2-1393 Uni MD, 42502-2-1586 Uni MD, 42502-2-1615 Uni MD, 42502-2-1672 Uni MD) in accordance with legislation of both the European Union (Council Directive 499 2010/63/EU) and the Federal Republic of Germany (according to §8, section 1 TierSchG and TierSchVersV).

Cell lines and cell culture
The mouse melanoma cell line HCmel12 was established from a primary melanoma in the Hgf-Cd4k R24C mouse model by serial transplantation in our laboratory as described previously 54 . The mouse melanoma cell line B16 and the human melanoma cell lines A375 and SKmel28 were purchased from ATCC. The human melanoma cell lines MaMel04 and MaMel102 were kindly provided by D. Schadendorf. All cell lines were cultured in complete Roswell Park Memorial Institute (RPMI) medium consisting of RPMI 1640 medium (Life Technologies) supplemented with 10% foetal calf serum (Biochrome), 2 mM l-glutamine, 10 mM non-essential amino acids, 1 mM HEPES (all form Life Technologies), 20 µM 2-mercoptoethanol (Sigma), 100 IU ml −1 penicillin and 100 µg ml −1 streptomycin (Invitrogen) in a humidified incubator with 5% CO 2 . The cell lines were routinely screened for mycoplasma contamination and were authenticated by the commercial provider or by short tandem repeat fingerprinting.

In vitro cell death assays
For the measurements of cell death in mouse and human melanoma cell lines, cells were first seeded in 96-well plates in complete RPMI medium. Inflammatory mediators were added after 24 h (10 U ml −1 recombinant mouse IFNγ (Peprotech); 1,000 U ml −1 recombinant mouse TNF (Peprotech); 100 U ml −1 animal-free recombinant human IFNγ (Peprotech); 1,000 U ml −1 recombinant human TNF (Peprotech) and 100 µM SNAP (Cayman Chemicals)). After 24 h, floating and adherent cells were gathered and stained using the fluorescein isothiocyanate (FITC) Annexin V Apoptosis Detection Kit I (BD Pharmingen) and analysed using the Attune NxT acoustic focusing flow cytometer (ThermoFisher).

Adenovirus generation and expansion
To generate the adenoviral vaccine Ad-PT, a fusion construct was generated consisting of the first 150 base pairs of the human PMEL complementary DNA (coding for amino acids 1-50 of the human PMEL/gp100 protein including the CD8 + T cell epitope KVPRNQDWL) and 1,404 base pairs of the mouse Trp1 cDNA (coding for amino acids 51-518 including the CD4 + T cell epitope SGHNCGTCRPGWRGAACNQKILTVR) followed by sequences coding for a T2A viral self-cleaving peptide and the yellow fluorescent marker protein eYFP. This vaccine construct was cloned into the pShuttle vector (termed pShuttle-PT-YFP). A recombinant adenovirus vector with this sequence was then generated by a recombineering technique in Escherichia coli strain SW102 using bacmid pAdZ5-CV5-E3 + . The E1 region of this bacmid is replaced by a selection/counter-selection cassette called ampicillin, LacZ, SacB or the ALS cassette. Next, E. coli with this bacmid were electroporated with the PT-YFP transgene with homology arms flanking the ALS cassette obtained by PCR amplification using pShuttle-PT-YFP as a template. Positive colonies were isolated after antibiotic selection on LB-sucrose plates. Ad-PT and Ad-OVA were expanded using the 911 human embryonic retinoblast cell line. A confluent monolayer of the cells in T175 cell culture flasks was infected with Ad-PT or Ad-OVA at MOI 1. The cytopathic effects were observed at around 36 h of incubation at 37 °C. Next, cells were scraped, freeze-thawed three times Article and the lysates were cleared by centrifuging at the speed of 7,000g for 45 min. The crude virus was then titrated by the TCID 50 method according to standard protocols.

CRISPR-Cas9-mediated genetic cell engineering
To generate Ciita-KO, Trp1-KO, Jak1-KO and Tyr-KO HCmel12 variants, HCmel12 melanoma cells were used that can be readily genetically modified using CRISPR-Cas9-mediated gene editing 55 . Cells were seeded into a 12-well plate at a density of 5 × 10 5 cells per well and cotransfected with 1.6 µg pX330-sgRNA and 0.4 µg plasmid expressing GFP (pRp-GFP) using Fugene HD transfection reagent (Promega) according to the manufacturer's instructions. GFP positive cells were single-cell sorted using a FACSAria III Cell Sorter (BD) to generate polyclonal and monoclonal populations per targeted gene. HCmel12 cells were mock transfected with pX330 plasmid without single-guide RNA and the polyclonal cell line was used as a CRISPR-control in all performed experiments. Genomic DNA from cultured knockout variants was extracted using the NucleoSpin Tissue kit (Macherey-Nagel) according to the manufacturer's protocol. A two-step PCR protocol was performed to generate targeted PCR amplicons for next-generation sequencing. In the first PCR, specific primers for the target gene with more adapter sequences complementary to the barcoding primers were used to amplify the genomic region of interest with Phusion HD polymerase (New England Biolabs). In a second PCR, adapter-specific universal primers containing barcode sequences and the Illumina adapter sequences P5 and P7 were used (Illumina barcodes D501-508 and D701-D712). Next-generation sequencing was performed with MiSeq Gene and Small Genome Sequencer (Illumina) according to manufacturer's standard protocols with a single-end read and 300 cycles (MiSeq Reagent Kit v.2 300 cycle). For the detection of insertions or deletions, the web-based program Outknocker (http://www.outknocker.org/) was used as previously described 56 . FASTQ files were imported, and the sequence of the target gene amplicons was used as reference sequence for alignment.

Western blot
Melanoma cells were lysed using the M-PER mammalian protein reagent (Fermentas) with protease inhibitors (Thermo Scientific). The protein concentration was spectrophotometrically measured by a Bradford-based assay using Pierce BCA protein assay kit (Thermo Scientific) according to manufacturer's protocol. Laemmli buffer was added and lysates were boiled at 95 °C for 5 min. Then, 10 µg of protein was loaded and separated according to size by SDS-PAGE gel electrophoresis on a 3% stacking and 10% polyacrylamide gel. Proteins were transferred to polyvinyl difluoride membranes with a 0.2 µm pore size (GE Healthcare) by means of wet blotting for 1 h. Unspecific binding was blocked with 5% skimmed milk in PBS with Tween for 1 h. Blots were stained with a goat polyclonal Trp1 antibody (1:1,000, Novus Biologicals) overnight at 4 °C. Next, the blots were incubated with anti-goat IgG HRP (1:2,000, Santa Cruz) for 1 h at room temperature. Horseradish peroxidase conjugated β-actin (200 µg ml −1 ) was used as loading control. Bound antibody was detected by SignalFire ECL reagent (Cell Signaling Technology) and chemiluminescence was visualized using an Octoplus QPLEX imager (NH DyeAgnostics).

Retroviral transduction
To generate tagBFP, mCherry and OVA-tagBFP-expressing cell lines, retroviruses were produced by transfecting human embryonic kidney 293T cells with the retroviral packaging constructs pCMV-gag-pol and pMD.2G (expressing VSVg) and the retroviral plasmids pRp-tagBFP, pRp-mCherry and pRp-OVA-tagBFP, respectively, according to standard protocols. Retrovirus-containing supernatant was used to transduce the target cell lines and antibiotic selection of transduced cells was started 48 h after transduction using 10 µg ml −1 Puromycin.

Tumour transplantation experiments
For tumour inoculation, a total of 2 × 10 5 cells were injected intracutaneously (i.c.) into the shaved flanks or hindlegs of mice with a 30G (0.3 × 13 mm) injection needle (BD). Tumour development was monitored by inspection and palpation. Tumour sizes were measured three times weekly and presented as mean diameter. Mice were euthanized when tumours exceeded 15 mm mean diameter or when mice showed signs of sickness in adherence with the local ethical regulations. All animal experiments were performed in groups of four to six mice and repeated independently at least twice.
ACT therapy protocol ACT therapy was performed as previously described 23,24 . In brief, when transplanted melanoma cell lines reached a mean diameter of 3-5 mm, mice were preconditioned for ACT by a single intraperitoneal (i.p.) injection of 2 mg (roughly 100 mg kg −1 ) of cyclophosphamide in 100 µl of PBS 1 day before intravenous (i.v.) delivery of splenocytes isolated from TCR-transgenic Pmel-1 and/or TRP-1 donor mice harbouring naïve Pmel-1/gp100-specific CD8 + T cells and/or naïve TRP-1-specific CD4 + T cells (in 100 µl of PBS). Unless otherwise indicated, we transferred splenocytes containing 5 × 10 5 antigen-specific T cells. The adoptively transferred T cells were stimulated in vivo by a single i.p. injection of 2.5 × 10 8 PFU of the recombinant adenoviral vaccine Ad-PT in 100 µl of PBS. On day 3, 6 and 9 after T cell transfer, tumours were injected with 50 µg of CpG 1826 (MWG Biotech) and 50 µg of polyinosinic:polycytidylic acid (polyI:C, Invivogen) diluted in 100 µl of distilled water. Seven days after T cell transfer, blood was taken routinely from the Vena facialis to confirm successful expansion of transferred T cells by flow cytometry.   100). Intracellular staining was carried out using a Fixation/Permeabilization Solution Kit (BD or Biolegend). Single-cell suspensions from tumours were first stained with antibodies against-cell-surface antigens, then fixed and permeabilized, followed by intracellular staining. Dead cell exclusion was performed using 7-aminoactinomycin or propidium iodide. All data were acquired with an Attune NxT acoustic focusing flow cytometer (ThermoFisher). Gating and subsequent analyses were performed using FlowJo v. 10.8.1 for Windows (Tree Star,Inc.). Fluorescence-activated cell sorting was performed using an Aria III (BD Biosciences).

Quantification of tumour-infiltrating immune cells
To quantify the abundance of immune cell subpopulations in tumour tissues, 2,000 cells of interest per biological sample were concatenated to a single FCS file. The t-distributed stochastic neighbor embedding (t-SNE) plots were generated in FlowJo using the opt-SNE learning configuration 57 . The vantage-point tree K-nearest-neighbours algorithm and the Barnes-Hut gradient algorithm were set to 1,000 iterations, 30 perplexity and 840 learning rate. Immune cell subpopulations were annotated on the basis of heatmaps for characteristic marker combinations and their percentage in the tumour was calculated.

Analysis of tumour cell MHC expression and antigen recognition by CD4 + T cells
To quantify the expression of MHC molecules, tumour cells were pretreated with 100 U ml −1 recombinant murine IFNγ (Peprotech) for 72 h and then analysed by flow cytometry. To assess antigen recognition by CD4 + T cells, TRP-1 TCRtg mice were immunized with Ad-PT and subsequently injected with 50 µg of CpG and 50 µg of polyI:C i.c. 3 and 6 days after immunization. TRP-1 CD4 + T cells were isolated from the spleen and purified by two rounds of magnetic cell sorting (Miltenyi). Direct antigen recognition was determined by coculturing purified CD4 + T cells with IFNγ pretreated HCmel12 cells. Antigen recognition in proxy was assessed by initially generating bone marrow-derived dendritic cells with recombinant GM-CSF and IL-4 (Peprotech) as previously described. After 1 week, differentiated bone marrow-derived dendritic cells were then pulsed overnight with HCmel12 lysate, before coculture with purified CD4 + T cells. For both direct and myeloid cell-dependent antigen-recognition assays, the production of IFNγ from the CD4 + T cells was measured 16 h after coculture by intracellular cytokine staining using flow cytometry according to standard protocols.

Calculations of absolute immune cell counts in tumour tissues
Tumours were excised with tweezers and scissors, then weighed using the Entris 224-1S analytical balance (Sartorius). Single-cell suspensions were created mechanically using 5-ml syringe plungers (BD) and 70 µm cell strainers (Greiner). After immunostaining, cells were suspended in a defined volume and analysed on the Attune NxT acoustic focusing flow cytometer that uses a unique volumetric sample and sheath fluid delivery system allowing for accurate measurements of the number of cells analysed in a defined sample volume. The total number of viable CD45 + immune cells in an individual tumour can then be derived by multiplying the number of CD45 + immune cells counted in a defined sample volume with the total volume of the respective single-cell suspension. Division of this total number by the total weight of the tumour yields the absolute immune cell count per mg tumour tissue. The absolute count of various immune cell subpopulations was calculated from their relative percentage in viable CD45 + immune cells.

Immunofluorescence microscopy
Tumours were harvested on day 5 after ACT and fixed in 4% paraformaldehyde for 24 h, then dehydrated in 20% sucrose before embedding in optimal cutting temperature freezing media (Sakura Finetek). Next, 6 µm sections were cut on a CM305S cryostat (Leica), adhered to Superfrost Plus slides (VWR) and stored at −20 °C until further use. When thawed, slides were either fixed with ice-cold acetone and stained with rat anti-mouse I-A/I-E (1:50) and anti-rat IgG-Alexa Fluor 594 (1:100) or directly mounted with Vectashield Antifade Mounting Medium (Vector Laboratories). Images were acquired on an Axio Imager.M2 with a Colibri 7 LED illumination system (Zeiss) and analysed with ImageJ v.1.52i (http://imageJ.nij.gov/ij).

Intravital two-photon microscopy
Mice were anaesthetized with 100 mg kg −1 ketamine and 10 mg kg −1 xylazine i.p., complemented by 3 mg kg −1 acepromazine s.c. after the onset of anaesthesia. The animals were placed and fixed to a heated stage. Transparent Vidisic carbomer gel was applied to moisten the eyes during anaesthesia. The hind leg was fixed in an elevated position and the skin covering the melanoma was detached using surgical scissors and forceps. One drop of transparent Vidisic carbomer gel was used on the exposed site as mounting medium. Two component STD putty (3M ESPE) placed on both sides of the leg was used create a level surface using a 24 × 60 mm cover slip, which was gently pressed on the putty in a way that the coverslip made slight contact with the exposed site, without exerting pressure on the tumour. After complete polymerization of the putty, the mice were transferred onto a 37 °C heating plate under the two-photon microscope.
Imaging was performed using distilled water or transparent Vidisic carbomer gel as immersion liquid with a W Plan-Apochromat ×20/1.0 DIC VIS-IR objective mounted to a Zeiss LSM 700 upright microscope with the ZEN software environment (v.2.1, Zeiss), or a LaVision TrimScope mounted to an Olympus BX50WI fluorescence microscope stand and a XLUMPlanFl ×20/0.95 objective. Excitation on the LSM700 setup was performed with Mai Tai DeepSee (tuned to 800 nm) and Insight X3 (tuned to 980 nm) Ti:Sa oscillators (both from Spectra-Physics). Fluorescence signals were read out on a long-pass dichroic mirror detector cascade as follows: dsRed, 980 nm excitation and 555 nm dichroic transmission with a 587/45 nm bandpass filter; Venus, 980 nm excitation and 520 nm dichroic transmission with a 534/30 nm bandpass filter; second-harmonic generation, 800 nm excitation and 445 nm dichroic deflection unfiltered; tagBFP, 800 nm excitation and 490 nm dichroic deflection with a 485 nm short-pass filter; and eGFP, 980 nm excitation, 520 nm dichroic deflection and 490 nm dichroic transmission with a 525/50 nm bandpass filter. Excitation on the TrimScope setup was performed with a Chamaeleon Ultra II Ti:Sa oscillator tuned to 880 nm. Fluorescence signals were read out with a double split detector array with a 495 nm main dichroic mirror and 445 and 520 nm secondary dichroic mirrors (all long-pass) as follows: second-harmonic generation, 495 nm and 445 nm dichroic deflection unfiltered; tagBFP, 495 nm dichroic deflection and 445 nm dichroic transmission with a 494/20 nm bandpass filter; eGFP, 495 nm dichroic transmission and 520 nm dichroic deflection with a 514/30 nm bandpass filter; and Venus, 495 nm and 520 nm dichroic transmission with a 542/27 nm bandpass filter. Non-descanned photomultiplier tubes (for second-harmonic generation, dsRed and Venus in all setups, and for eGFP and tagBFP in the TrimScope setup) and high sensitivity detectors (for tagBFP and eGFP in the Zeiss setup) were used for signal collection.
Typically, three to four representative fields of view of 353 µm 2 size in x-, y-and a z-range of 48 to 60 µm with 4 µm step sizes were chosen for data acquisition. Z-stacks were captured in 60-90 s intervals and individual video length was 15-30 min. Data analysis was performed with the Bitplane Imaris software (v.8.3 to 9.7). T cells were identified using the Imaris spot function. Tumour area was identified using the surface function with low surface detail. CD11c-Venus cells were identified using the surface function with high detail. T cell speed was calculated using the Imaris software. Cells were considered arrested when speed was less than 2 µm min −1 . Contact duration was measured as the time that the distance between the centre of mass of a T cell to the closest CD11c cell surface was less than 8 µm. Snapshot images of 3D rendering and Article tracking were cropped, arranged and animated for time series using ImageJ v.1.52i (http://imageJ.nij.gov/ij).

Cell preparation for scRNA-seq
Three individual tumours per group were harvested and processed into single suspensions. CD45 + cells were enriched using a positive selection kit (Miltenyi). Next, individual samples were hashtagged with unique TotalSeq-B hashtag antibodies B0301-B0310 (1:300, Biolegend) and subsequently stained with fluorescently labelled antibodies. CD45 + CD11b + Ly6Gcells were sorted with an Aria III fluorescence-activated cell sorter (BD). Isolated cells were loaded onto one lane of a 10X Chromium microfluidics controller. cDNA of hashtag and gene expression libraries were amplified, and indices added by means of PCR. Sequencing was performed on an Illumina Novaseq on two lanes of a S1 cartridge with 150 bp read length in paired end mode. Reading depth was calculated to obtain roughly 50,000 reads per cell for the gene expression library and 5,000 reads per cell for the hashtag library.

scRNA-seq data processing and hashtag-demultiplexing
The scRNA-seq data generated using 10X Genomics Chromium technology were aligned and quantified using the Cell Ranger Single-Cell Software Suite against the mm10 mouse reference genome. The raw, unfiltered data generated from Cell Ranger were used for downstream analyses. Quality control was performed on cells on the basis of the three metrics: total unique molecular identifier (UMI) count, number of detected genes and proportion of mitochondrial gene count per cell. Specifically, cells with less than 1,000 UMIs, 1,000 detected genes and more than 25% mitochondrial UMIs were filtered out. To remove potential doublets, cells with UMI count above 40,000 were removed. Subsequently, we demultiplexed the samples tagged with distinct hashtag-oligonucleotides using Solo 58 . After quality control, we normalized raw counts by their size factors using scran 59 and subsequently performed log 2 transformation. The logarithmized and normalized count matrix was used for the downstream analyses.
Dimensionality reduction, unsupervised clustering and differential gene expression analyses Analysis of normalized data was performed using the scanpy Python package 60 . Initially, the 4,000 most highly variable genes were selected for subsequent analysis using scanpy.pp.highly_variable_genes with the parameter 'n_top_genes=4000'. Next, a principal component analysis was performed with 50 components using scanpy.tl.pca with the parameters 'n_comps=50, use_highly_variable=True, svd_solver='arpack''. Subsequently, dimensionality reduction was performed using UMAP with scanpy.tl.umap. Single cells were automatically assigned using R package SingleR 61 , with transcriptomes from the Immunological Genome Project as a reference. Clustering of single cells by their expression profiles was conducted by using the Leiden algorithm running scanpy.tl.leiden with the parameter 'resolution=1.0'. Clusters with fewer than 20 cells were removed from further analysis. Differential gene expression was performed between cells classified as macrophages and monocytes from non-treated and CD4 ACT-treated mice using a hurdle model implemented in the R package MAST. Subsequent gene set enrichment analysis was performed using gene set enrichment analysis in preranked mode using the log 2 fold change as a ranking metric. The IFN score was derived by calculating a z-score for all genes from the MSigDB gene set 'HALLMARK_INTERFERON_GAMMA_RESPONSE' for each cell.

RNA velocity
For RNA velocity, count matrices of spliced and unspliced RNA abundances were generated using the velocyto workflow for 10X chromium samples, with the genome annotation file supplied by 10X Genomics for the mm10 genome and a repeat annotation file retrieved from the UCSC genome browser. Subsequent analyses were performed using scVelo 62 . The count matrices were loaded into the scanpy environment, merged with the previously generated anndata objects and normalized using scvelo.pp.filter_and_normalize. Next, moments for velocity estimation were calculated, gene-specific velocities were estimated and the velocity graphs were computed. Furthermore, a partition-based graph abstraction was generated with velocity-directed edges.

Statistical methods
Statistical analyses and number of samples (n) are given in each figure legend. Mann-Whitney U-tests, unpaired two tail t-tests, analysis of variance (ANOVA) and log-rank tests were performed in Graphpad Prism (v.8).

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

Data availability
The raw sequencing mouse scRNA-seq data are available at the NCBI GEO under the accession GSE230427 without restrictions. The normalized and logarithmized count matrix used for the subsequent analyses is also available at the NCBI GEO under the accession GSE230427 without restrictions. Human scRNA-seq data used in this study are available at the European Genome-Phenome Archive with the identifier EGAS00001006488, available for non-commercial research purposes on reasonable request and subject to review of a project proposal that will be evaluated by the VIB-UZL Data Access Committee. Source data are provided with this paper.

transferred T-cells and quantitation of OT-I CD8+ and OT-II CD4+
TCRtg T-cell expansion in peripheral blood 7 days after ACT (mean ± SEM from n = 9 biologically independent samples, ****p < 0.0001 using a two-tailed paired t-test). h, Experimental protocol and (i) individual tumour growth curves of mice bearing established HCmel12-OVA Jak1-KO tumours, treated as indicated. Means between groups were statistically compared using a twotailed paired t-test. Last updated by author(s): Apr 26, 2023 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend,

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy The raw sequencing mouse scRNA-seq data are available at the NCBI GEO under the accession GSE230427 without restrictions. The normalized and logarithmised count matrix used for the subsequent analyses is also available at the NCBI GEO under the accession GSE230427 without restrictions. Human scRNAseq data used in this study are available at the European Genome-Phenome Archive (EGA) with the identifier EGAS00001006488, available for non-commercial research purposes upon reasonable request and subject to review of a project proposal that will be evaluated by the VIB-UZL Data Access Committee.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
The determination of sample sizes for animal experiments was based on our experience with success of tumour engraftment and efficacy of therapeutic intervention in order to adhere to the 3R guidelines of the local Ethics Committee of the Office for Veterinary Affairs. Tumour treatment experiments involved 4-6 mice per group and were performed at least twice. Analyses of tumour immune cell infiltrates and intravital microscopy experiments involved a minimum of 3 mice per group. This yielded consistently reproducible and statistically significant results. Similarly, group sizes for in vitro experiments were determined based on prior knowledge of variation.
Data exclusions No data was excluded from analysis.

Replication
Experiments were reliably reproduced and the number of experiments performed stated in methods and legends. Culminated and pooled data are shown where possible. Where representative data is shown, relevant experiments were repeated successfully at least twice with the exact number of repeats indicated in each case. Most experiments were repeated at least twice if not three or more times to verify that experimental findings were reproducible.
Randomization For in vivo tumour treatment experiments, mice were randomized into different groups when the tumours reached between 3-5 mm in diameter.

Blinding
Blinding was not performed in this study. The experimental observations presented would be consistent irrespective of blinding and therefore blinding was not relevant in this study.
Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response. The mouse HCmel12 cell line and all variants were generated in the Tüting Laboratory. The mouse B16 melanoma cell line was purchased from ATCC.
The human melanoma cell lines MaMel04 and MaMel102 were kindly provided by Dirk Schadendorf. The human melanoma cell lines, Skmel28 and A375, and HEK293T cells were purchased from ATCC. The 911 human embryonic retinoblast cell line was obtained from Crucell. Authentication B16, Skmel28, A375 and HEK293T cells were originally obtained from ATCC respectively and were therefore authenticated by nature portfolio | reporting summary

March 2021
Methodology Sample preparation Blood samples were resuspended in red cell lysis buffer and incubated for 15 minutes at room temperature. The samples were then centrifuged at 350g for 5 minutes and the supernatant was discarded. This process was repeated. FC-Blocking was performed by incubation of the samples with anti-CD16/32 (1:300) for 10 minutes at 4°C. After washing the samples, the cells were then stained with antibodies for 15 minutes at 4°C. The samples were subsequently washed and resuspended in FACS buffer prior to analysis.
Tumours, spleens and lymph nodes were homogenised through a 70 μm strainer to generate single cell suspensions. The samples were then centrifuged at 350g for 5 minutes and the supernatant was discarded. The samples were then resuspended in red cell lysis buffer and incubated for 5 minutes at room temperature. The samples were then centrifuged again and the supernatant was discarded prior to FC-Blocking of the samples with anti-CD16/32 (1:300) for 10 minutes at 4°C . After washing the samples, the cells were then stained with antibodies for 15 minutes at 4°C. The samples were subsequently washed and resuspended in FACS buffer prior to analysis.
Tick this box to confirm that a figure exemplifying the gating strategy is provided in the Supplementary Information.