Tumor MHCII immunity requires in situ antigen presentation by cancer-associated broblasts

Dimitra Kerdidani BSRC Fleming Emmanouil Aerakis BSRC Fleming Kleio Verrou BSRC Fleming Petros Stamoulis BSRC Fleming Katerina Goudevenou BSRC Fleming Alejandro Prados B.S.R.C. Alexander Fleming https://orcid.org/0000-0002-8610-7559 Christos Tzaferis BSRC Fleming Ioannis Vamvakaris Sotiria Chest Hospital Evangelos Kaniaris Sotiria Chest Hospital Konstantinos Vachlas BSRC Fleming Evangelos Sepsas BSRC Fleming Konstantinos Potaris BSRC Fleming Anastasios Koutsopoulos University of Crete Maria Tsoumakidou (  tsoumakidou@ eming.gr ) Alexander Fleming Biomedical Sciences Research Center https://orcid.org/0000-0001-7959-4793


Introduction
The series of immunological events that takes place between tumors and tumor draining lymph nodes forms a cyclic trajectory that is being referred to as the cancer-immunity cycle (1). In the rst step of these events tumor antigens are carried to the tumor draining lymph nodes (LNs) and partly transferred to resident dendritic cells (DCs) (2). In LNs, migratory and resident DCs present their antigenic cargo to antigen-inexperienced (naïve) T cells, which become differentiated effector cells that egress from LNs and enter tumors. In tumors, CD8 cells exhibit direct killing activity against cancer cells, but they are seriously dependent on CD4 T cells for function and transition to memory cells (3)(4)(5)(6). Although our current understanding of the functional space in the cancer-immunity cycle is that cancer antigen presentation primarily occurs in lymph nodes, the contribution of in situ cancer antigen presentation has not been be ruled out (7,8).
Very few studies have directly addressed the role of peripheral antigen presentation in T cell responses (7,(9)(10)(11)(12). In cancer three lines of evidence support that the TCRs are stimulated in situ within solid tumors. First, the CD4 TCR repertoire is regionally shaped by the local neoantigen load (13). Second, stem-like CD8 T cells reside in dense MHCII expressing cell niches within tumors (14). Third, right ank tumours that differ only in one MHCII neoantigen with left ank tumours are in ltrated by higher numbers of neoantigen-speci c CD4 + T cells (15). DCs are scarce and immature within solid tumors and are generally considered to exert their primary effects in lymph nodes (2,7,8,16,17). Because structural tissue cells greatly outnumber professional antigen presenting cells, express immune genes (18) and can be induced to present antigens (12,19), we hypothesized that they are required for local antigen presentation and antitumor immunity.
Fibroblasts have been long considered as immunosuppressive cells (20). Thus, the recently identi ed MHCII antigen presenting cancer-associated broblasts (apCAFs) in pancreatic adenocarcinoma (PDAC) and breast carcinoma (BC) were presumed to induce immune tolerance and tumor escape (21)(22)(23). Here we show in murine models of lung cancer that broblast-speci c targeted ablation of MHCII impairs local immunity, accelerating tumor growth. In primary human lung tumors CD4 T cells accumulate in apCAF dense spots. Human apCAFs directly primed adjacent effector CD4 T cell via the TCR and at the same time protected T cells from apoptosis via the c1q receptor c1qbp. These ndings reposition apCAFs as immunostimulatory cells and propose a new functional node in the cancer-immunity cycle: CD4 T cells must receive a second wave of antigen presentation after their egression from draining lymph nodes within tumors for effective MHCII immunity to occur.

Antigen-presenting CAFs (apCAFs) prime adjacent CD4 T cells within human lung tumors
We undertook a pilot analysis of MHCII expression in human lung adenocarcinomas and squamous cell carcinomas by immunohistochemistry. We noticed abundant niches of elongated MHCII + broblast-like cells inside the tumor bed (Supplementary Figure 1). To elaborate on the broblastic identity of these MHCII + cells we enzymatically dispersed primary human tumors and analyzed known mesenchymal markers by FACS (Figure 1a). First, FACS impressively underestimated the frequencies of broblasts compared to imaging. This discrepancy likely re ected differences in detachment e ciency of different cell types following tissue digestion, with broblasts being more adherent to the extracellular matrix than immune cells and thus more di cult to disconnect. MHCII + CAFs (apCAFs) were phenotypically similar to their MHCIIcounterparts and largely co-expressed FAP, PDGFRa and Podoplanin. Vimentin was expressed at variable levels, while aSMA was lowly expressed (Figure 1a). 2D visualization in t-SNE plots (input all mesenchymal markers) con rmed that apCAFs do not cluster separately from MHCII -CAFs. To determine the spatial relationship between apCAFs and CD4 T cells we analyzed whole tumor tissue sections by immuno uorescence staining. apCAFs were identi ed as cells that co-expressed MHCII and FAP. Quantitative analysis showed higher numbers of intratumoral MHCII + FAP + relatively to MHCII + FAPcells, con rming the relative abundance of apCAFs among antigen presenting cells. T cells accumulated in regions with dense apCAFs and were identi ed closer to apCAFs rather than FAP -MHCII + cells ( Figure  1b). We also found a signi cant correlation between numbers of CD4 T cells and apCAFs in different regions of the same tumor section (Figure 1b). This suggests that apCAFs create functional spots within lung tumors that sustain CD4 T cell populations and thus MHCII immunity.
We interrogated whether we could identify MHCII broblasts in healthy lungs. To rule out the possibility of contamination by non-broblastic cell populations we gated broblasts as FS high Lin -FAP + cells. MHCII was expressed by a subset of normal lung broblasts, but it was more frequently detected in CAFs, suggesting that the tumor microenvironment drives and/or sustains apCAFs (Figures 1c).
Spatial heterogeneity of T cell clones suggests regional activation by neighboring antigen-presenting cells that present their cognate antigens (13). We reasoned that apCAFs directly presented cancer MHCII peptides to adjacent CD4 T cells. To test this, we co-cultured primary human CD4 T cells and FAP + CAFs, puri ed from the same lung tumour fragment and assessed phosphorylation of the key TCR/CD28 signaling node, mTOR (Figure 1d). Primary cell phenotypes are dramatically altered once they are isolated and cultured. To avoid this bias, we used freshly FACS sorted primary tumor broblasts and freshly sorted autologous primary tumor in ltrating T cells for all our co-cultures. We used the bulk of CAFs rather than MHCII + CAFs to avoid blocking TCR recognition and did not supplement the culture medium with exogenous cytokines. Approximately 1 in 10 intratumoral T cells responded to direct CAF contact by mTOR phosphorylation and CD44 up-regulation and these effects were abolished with MHCII blocking (Figure 1d). Thus, human apCAFs acquire exogenous peptides within tumors in vivo and can directly prime adjacent CD4 T cells ex vivo. Figure 1. MHCII broblasts form CD4 T cell priming spots within human lung tumors. a) Representative FACS plots of a digested human lung tumor and expression of mesenchymal speci c genes. Cumulative data (n=5-9 patients). b) Representative whole-slide imaging for panMHCII, FAP and CD4 and cellular spatial relationship map from a lung cancer patient. After acquiring XY coordinates, XY location of MHCII+FAP+ and MHCII+FAP-cells were overlaid with CD4 cell density contour. Distance between MHCII+FAP+ or MHCII+FAP-and the closest CD4 cell. Correlations between MHCII+FAP+ densities with CD4+ cell densities (cells per 25x10 4 μm 2 regions in whole slide images, representative of n=3 patients). c) FACS plots showing MHCII in FS hi Lin -FAP + PDGFRa + broblasts of paired digested human lung tumors and tumor-free lungs (n=7 patients). d) CD4 T cells and broblasts were sorted from the same tumor fragment and co-cultured with panMHCII blocking antibody or isotype control. Representative FACS plots and cumulative data on phospho-mTOR and CD44 (n=4 patients). b, c, d)*P<0.05, **P<0.01, Unpaired or paired t-test.
apCAFs are detected in tumors of IFNγ su cient mice To assess whether apCAFs were present in murine tissues we analyzed three models of orthotopic lung cancer. We inoculated one commercially available (LLC mCherry ) or an autochthonous cancer cell line (CULA zsGreen ) in the left lung lobe of syngeneic mice or injected melanoma cells (B16F10 mCherry ) in the tail vein of mice. Tumors were digested and subjected to FACS analysis. Intratumoral FS high CD45 -EPCAM -CD31 -IAb(MHCII) + non-cancer cells co-expressed podoplanin and PDGFRa, but not aSMA (Figure 2a).
Murine FAP antibodies showed nonspeci c and unreliable staining and were thus excluded from the analysis. Similar to human apCAFs, 2D visualization in t-SNE plots indicated that apCAFs are admixed with MHCII -CAFs. Immuno uorescence staining detected MHCII + Podoplanin + cells within the tumor bed ( Figure 2a). Similar to what we had observed in humans, MHCII was more frequently expressed by CAFs relatively to healthy lung broblasts (Figure 2b). To determine whether broblasts depended on IFNγ for MHCII expression we inoculated LLC cells in the lungs of IL12 p35-/-, IL12p40-/-, IFNγR-/-and IFNγ-/mice. MHCII expression was decreased in CAFs of IL12 de cient mice and severely impaired in those of IFNγ/IFNγR de cient versus wild type mice (Figure 2b).
To probe the contribution of the tumor microenvironment in sustaining MHCII in broblasts, we developed an in vitro model system where we sorted PDGFRa + Podoplanin + broblasts from LLC mCherry tumors and cultured them in vitro. Although by day 3 all cultured CAFs had lost MHCII, exposure to fresh lung tumor, but not to healthy lung homogenates, restored MHCII expression (Figure 2c). It should be noted that MHCII induction was observed upon 3D, but not 2D culture conditions, consistent with previous observations that 2D cultures induce MHCII loss in PDAC broblasts and differentiation to myo broblasts (21). Notably, neutralizing IFNγ in tumor homogenates restrained MHCII induction in primary CAFs ( Figure  2c). Representative FACS plots of IAb in FS hi Lin -Podoplanin + PDGFRa + broblasts of a CULA lung tumor and a healthy lung of wild type Bl6 mice. Down, left. Cumulative data on IAb expression of LLC, CULA and B16 lung tumors and healthy lungs of wild type Bl6 mice (n=6-8 per group). Down, right. Cumulative data on IAb expression of LLC lung tumors of cytokine-de cient mice (n=4-7 per group). c) FS hi Lin -Podoplanin + PDGFRa + broblasts were isolated from pooled LLC lung tumors and cultured in 3D, as indicated. (n=4 experiments). b, c)*P<0.05, **P<0.01, Unpaired or paired t-test.
Human and murine apCAF transcriptomes reveal conserved immune phenotypes and epithelial signature To characterize human antigen-presenting broblasts we leveraged a public single-cell RNA sequencing (scRNAseq) dataset of primary human lung tumor, adjacent unaffected lung and computational analyses (24). There is no standardized computational method for characterizing gene expression levels in a binary form of "high" and "low". Furthermore, gene expression levels correlate variably with corresponding protein levels. We hypothesized that we would be able to identify antigen presenting broblasts in single broblast RNA sequencing datasets if we used cDCs as a metric for physiologically relevant MHCII gene expression. Re-clustering of 1284 single-cells annotated as either broblasts (798) or cross-presenting cDCs (486) (E-MTAB-6149 and E-MTAB-6653, n=3 patients) via k-means clustering (input all MHCII genes, k=3 clusters supported by Silhouette scoring), separated broblasts into 2 clusters that shared (MHCII + ) or did not share (MHCII -) MHCII expression features with cDCs. Similar to FACS analysis, dimensionality reduction (input 1785 expressed genes) showed that MHCII + broblasts are predominantly admixed with MHCIIbroblasts and do not form a separate cluster (Figure 3a). Differential expression analysis (DEA) revealed 115 deregulated genes in MHCII + versus MHCIIbroblasts, 67 of which were upregulated ( Figure  3b). Among the top upregulated were signature genes of PDAC apCAFs (CD74, SLPI)(21), the IL6 gene (known as a prototype in ammatory CAF gene) (20), components of the complement pathway (CFD, C1QA, C1QB). Among the top down regulated genes was ACTA2 (encoding the myo broblastic CAF gene aSMA) (20), a number of collagen genes (COL3A1, COL1A1, COL1A2) and other secreted extracellular matrix (ECM) proteins (SPARC, POSTN, BGN). Comparing pathway activities between MHCII + and MHCIIbroblasts revealed enrichment in upregulated genes that are involved in immunological processes and down-regulated genes involved in ECM organization (Figure 3b). Although MHCII + broblasts were interspersed among MHCIIbroblast in t-SNE plots, it was striking that SFTC (encoding for surfactant protein C) was on the top of the list of up-regulated genes, raising the possibility that MHCII + broblasts arise from alveolar type II cells, which constitutively express SFTC and MHCII (25). Three other genes known to be expressed by epithelial cells, CLU (clusterin) and the antimicrobial proteins lysozyme (LYZ) and SLPI, were amongst the 11 top up-regulated genes. Altogether our computational analysis of the transcriptome of MHCII + broblasts depicts i) prominent immune functions, ii) a pro le closer to in ammatory rather than myo broblastic CAFs (IL6 high, a-SMA low, ECM proteins low) and iii) strong cues of an alveolar epithelial origin.
To investigate whether the human lung MCHII broblast signature was conserved across mice, we sorted MHCII + and MHCII -CAFs and performed bulk RNA sequencing ( Figure 3c). Differential expression analysis (input all expressed 15675 genes, adj.P<0.05, |Log2FC|>3) revealed a large number of deregulated genes (486), almost all of which (483) were up-regulated in MHCII + CAFs ( Figure 3d). As in humans, among the top upregulated genes were the 2 MHCII CAF signature genes CD74 and SLPI. Pathway analysis was in concordance with the immune-related pathways that were identi ed in our human dataset ( Figure 3d). Intriguingly, the alveolar type II speci c genes SFTPs and many alveolar epithelial cell genes were again found increased (CLAUDINs, KRTs, SLC34A, NAPSA, LYZ). Alveolar type II cells are long-lived cells that self-renew locally (26). To further support the hypothesis that MHCII broblasts originate from alveolar epithelial cells, we used alveolar cell type signature gene sets (curated from cluster markers, Supplementary Methods) and used them as inputs for GSEA. GSEA comparing MHCII + versus MHCIIbroblasts of mice and humans demonstrated that MHCII + broblasts were signi cantly enriched in genes that characterized alveolar epithelial cells in both species (Figure 3e). To align the gene signatures expressed across MHCII broblasts in mice and humans, we coclustered genes based on their transcriptional pro le in each species, using genes that were conserved and variable across MHCII broblasts in the human dataset ( Figure 3f). Our analysis denoted conservation of the MHCII broblast pro le across the two species. Altogether these results indicate that the process of epithelial-to-mesenchymal transition regulates the in situ differentiation of alveolar type II cells to a subset of broblasts with antigen presenting functions. a) Fibroblast and dendritic cell clusters in scRNAseq datasets (E-MTAB-6149, -6653) from lung tumors and adjacent lungs were re-clustered to identify MHCII broblasts (n=3 patients). b) Heat map of differentially expressed genes (adj.P-values<0.05) between MHCII+ and MHCII-broblasts. GO enrichment analysis was performed using DAVID. Network visualization was conducted by Cytoscape's EnrichmentMap. Nodes represent GO biological processes and edges connect functional terms with overlapping genes. Clusters of related nodes are circled and given labels re ecting broad biological processes. GO terms with an FDR < 0.1 are shown. c) MDS representation of bulk RNAseq data of sorted MHCII + and MHCIIbroblasts from pooled mouse LLC lung tumors (n=3 experiments). d) Differential expression between MHCII + and MHCII -Deregulated genes with adj.P-values<0.05 are indicated. GO enrichment analysis using upregulated genes with |Log2FC|>3 as input. e) GSEA enrichment plots for alveolar epithelial gene sets. f) Homology analysis of gene expression across mouse and human MHCII broblasts.

Deletion of MHCII in broblasts has a localized impact on tumor in ltrating CD4 T cells
We asked whether cancer antigen presentation by CAFs was a bystander, suppressor or driver of tumor progression in vivo. We set up pilot experiments starting with three murine lines expressing Cre recombinase (Cre) driven off Twist (27), collagene VI(28), and collagen 1a2 (Col1a2) promoters (29) ). Signature genes of Th1, Th2, Th17, Th9 and Tregs, cytotoxicity or exhaustion marker genes were not different between the two groups (Supplementary Figure 3). Likewise, pathway activities of downregulated genes showed signi cant enrichment in macromolecular metabolic processes. TCR activation is known to play a fundamental role in T cell metabolism (31). Taken together, the results described above suggest that antigen presentation by MHCII broblasts is a critical driver of MHCII immunity in vivo.
Lymph node broblastic reticular cells (FRCs) have contradictory roles in immune responses, such as inhibition of T cell proliferation via nitric oxide (32) and T cell reprogramming towards memory T cells via IL6 (33). Albeit FRCs in LNs acquire transcriptional programs typically associated with tumor escape (34), we considered the possibility that immune evasion in Col1a2 Cre + IAb fl/fl mice resulted from MHCII deletion in FRCs. We assessed MHCII expression in LN FRCs of Col1a2 CreER + IAb fl/fl versus Col1a2 CreER -IAb fl/fl mice and found a trend toward decreased MHCII + FRCs (Supplementary Figure 4). However, there were no gross differences in nodal CD4 + , CD8 + T cells and B cells (Supplementary Figure 4). We questioned whether antigen presenting CAFs could sustain MHCI and MHCII immunity in the absence of continuous T cell migration from LNs. Therefore, we treated mice with a S1PR antagonist FTY720, which blocks egress of T cells from LNs, for 5 consecutive days, starting on day 7 after OVA-LLC transplantation 30 . CD4 and CD8 T cells were again found decreased, while ova peptide-MHCI tetramers stained less of CD8 T cells in lung OVA-LLC tumors of FTY720-treated MHCII conditional knockout mice ( Figure 4d). These results suggest that targeting MHCII in broblasts facilitates immune escape by acting locally and that T cell immunity in Col1a2 Cre + IAb fl/fl mice is compromised by MHCII loss in CAFs rather than FRCs. Effector T cells have less stringent and incompletely understood activation requirements compared to naive T cells (35). Since we could barely detect any of the classical co-stimulatory molecules (CD80, CD86, CD40) in MHCII broblasts we mined their single cell transcriptome looking for other immunerelated molecules with ligand-receptor functions. Human MHCII + broblasts overexpressed TYROPB and CD47 which ligate to SIRP receptors, CD52 and ANXA1 which bind to SIGLEC10 and ALX/FPR2, respectively, and C1q which binds to cC1qR/calreticulin and gC1qR/C1qbp ( Figure 1I). None, but the receptor of the globular head of C1q, i.e. gC1qr/C1qbp, was found to be expressed on the surface of primary human TIL CD4 cells (Figure 5a). C1qbp is located in most cell types in the mitochondria where it is responsible to maintain transcription of mitochondrial proteins. In the cytoplasm it acts as a regulator of RNA stability and plays a critical role in mRNA splicing. The role of c1q as an extracellular signaling molecule and c1qbp as a plasma membrane receptor on T cells has been rarely reported (36-38) and is largely unexplored. We set out to investigate whether MHCII broblast-derived c1q ampli ed TCR signaling on CD4 T cells via membrane c1qbp. First, we con rmed increased c1q gene expression by MHCII + versus MHCIIlung tumor broblasts in distinct patients ( Figure 5a). Then, we co-cultured human effector CD4 TILs and CAFs in the presence of C1qbp blocking antibodies. Surprisingly, we did not observe any effect on TCR signaling, but a decrease in the numbers of viable T cells (Figure 5b). This lead us hypothesize that c1q is an extrinsic pro-survival signal for CD4 T cells. To test this, we cultured stimulated peripheral blood CD4 T cells for 1 h under serum starvation conditions. Puri ed c1q acted directly on T cells to almost completely rescue them from apoptosis ( Figure 5c). Depleting MHCII broblasts from the CAF-TIL CD4 co-cultures abrogated the apoptotic effect of aC1qbp, indicating that ac1qbp blocks MHCII broblast-derived c1q (Figure 5d). We consider unlikely that these results have been confounded by an autocrine effect of T cell intrinsic c1q because we found lower c1q gene expression levels in TIL CD4 than MHCII + CAFs and aCD3/aCD28 stimulation did not upregulate CD4 T cell-intrinsic c1q (Supplementary Figure 5). Based on these ndings, we examined whether c1qbp could function as pro-survival receptor in vivo. OVA-speci c OTII T cells were transduced with a c1qbp versus empty lentivirus and adoptively transferred in LLC-OVA lung tumor bearing mice. An increased number of C1qbp-OTII cells were identi ed 3d post adoptive transfer within lung tumors and this was accompanied by a decrease in tumor burden (Figure 5e). Altogether these data suggest that CD4 T cells depend on c1qbp for their survival within tumors and point to MHCII broblasts as a potent c1q source. (IF) for c1qbp shows surface expression in intratumoral CD4 T cells. qPCR for c1qb in sorted MHCII + versus MHCII -FS hi Lin -FAP + CAFs in human lung tumors (n=4 patients). b) CD4 T cells and CAFs were sorted from the same tumor fragment and co-cultured with ac1qbp blocking antibody or isotype control. Abs .numbers of live T cells were assessed by FACS (n=5 patients). c) Peripheral blood (PB) CD4 T cells from a healthy donor were sorted and activated with aCD3/aCD28 beads prior to undergoing serum starvation with or w/o puri ed c1q. Representative FACS plots of 2 experiments. d) CD4 T cells, total CAFs and MHCII -CAFs were sorted from the same tumor fragment. T-CAFs were co-cultured with ac1qbp blocking antibody or isotype control. T cell apoptosis was assessed using annexin-V and PI staining. Representative FACS plots and cumulative data are shown (n=3 patients). e) C1qbp overexpressing OTII T cells were adoptively transferred in LLC mcherryOva lung tumor bearing mice. Representative FACS plots of LLC mcherryOva cells and OTII cells in digested lung tumors. Cumulative data (n=4 per group, pooled from 2 experiments). a-e) Unpaired or paired t-test. P-value<0.05 Discussion A functional dichotomy is currently thought to exist between draining lymph nodes and tumors, whereby T cells are primed in lymph nodes and exert effector functions in tumors. Here we show that antitumor effector T cells require secondary in situ TCR stimulation by a subset of CAFs for effective immunity to occur. Single cell/bulk RNA sequencing approaches revealed a conserved apCAF transcriptome signature among lung tumors of mice and humans that aligned to that of PDAC apCAFs and was enriched for immune-related and alveolar epithelial genes, suggesting that apCAFs arise from epithelial cells through the process of epithelial-to-mesenchymal transition. Fibroblast-speci c targeted ablation of MHCII induced a lymph node-independent impairment in local immunity, accelerating tumor growth. In primary human lung tumors CD4 T cells preferentially homed dense MHCII broblast areas. Primary human broblasts directly activated the TCRs of tumor in ltrating CD4 T cells and at the same time protected them from apoptosis via the c1q receptor c1qbp, which unveiled a previously unrecognized function of c1q/c1qbp as an anti-apoptotic ligand/receptor pair for CD4 T cells. Our studies introduce nonhematopoietic cells as key cancer antigen-presenting cells and suggest that the nal outcome of initial CD4 T cell priming within tumor draining lymph nodes depends on de novo antigen presentation within tumors.
In MHCII IHC tumor slides MHCII broblastic cells were not identi ed within macroscopically visible tertiary lymphoid structures (TLS) structures, suggesting that intratumoral broblastic antigen presenting niches are distinct to TLS. It is likely that that TLS are mainly sites of B cell somatic hypermutation and plasma cell generation (39)(40)(41), while intratumoral antigen presenting broblasts form immunological niches for T effector cells. MHCII broblasts did not express classical costimulatory molecules but were still able to activate the TCR/CD28 pathway in effector T cells and enhance T cell metabolism. We consider two possibilities that may explain this contradiction: i) Costimulatory and co-inhibitory receptors of effector T cells display great diversity and promiscuity. Thus MHCII broblasts mediate effector T cell activation via nonclassical costimulatory molecules (35). ii) Antigen recognition su ces to activate effector T cells (42).
Our results provide the rst, to the best of our knowledge, demonstration that a complement component is an antiapoptotic factor. C1q is known to regulate basic metabolic processes of T cells (36-38) but has been disregarded as an extracellular signal. This is also related to the fact that the receptor of its globular head, C1qbp, has a very short intra-cytoplasmic tail and must complex with other membrane molecules to trigger intracellular signaling (43,44). The outcome of T cell therapy is highly dependent on T cell survival within tumors (45,46). We showed that C1qbp overexpression increases the numbers of adoptively transferred tumor speci c CD4 T cells within tumors. Conceivably, it is likely that the increased e cacy of C1qbp-OTII cells is not only the result of apCAF-derived c1q. Nonetheless our results showing that i) apCAFs express high levels of c1q; ii) they are a key antigen presenting cells within lung tumors iii) c1q acts directly on T cells to rescue them from apoptosis; iv) TIL CD4 cells reside within apCAF spots, altogether suggest an important role for apCAF-derived c1q as a pro-survival factor for TIL CD4 cells.
The following arguments indicate that lung apCAFs derive from alveolar epithelial type II (ATII): i) in both human and mouse datasets the most up-regulated genes were surfactant proteins, which are highly speci c for ATII cells, ii) constitutive MHCII expression is a hallmark of ATII cells (25), iii) ATII cells can transit to broblasts though the process of epithelial-to-mesenchymal transition (47, 48) iv) PDAC apCAFs also express epithelial speci c genes (21). Lung apCAFs should be considered bona de broblasts rather than ATII cells because of their characteristic broblastic morphology, their ability to easily grow in culture (primary alveolar epithelial cells are notoriously di cult to grow), the absence of surface EPCAM, the co-expression of multiple signature mesenchymal genes (49). We have rejected the possibility that apCAFs originate from cancer cells. In the human dataset tumor cells clustered away from broblasts, while in the mouse data set cancer cells were brightly uorescent and negatively selected. Furthermore, we never detected uorescence in murine broblasts that were cultured up to passage 5.
The apCAFs that were recently identi ed in PDAC and BC were presumed to induce T cell anergy or differentiation to T regulatory cells (21)(22)(23). This presumption opposed the increase in T regulatory cells that had been observed upon broblast deletion in PDAC (50). In our models MHCII deletion in broblasts did not alter CD4 T cell subsets. However, it caused a profound decrease in the metabolic rate of tumor in ltrating CD4 T cells. This nding ts perfectly well with the pivotal role of T cell metabolism in antigenspeci c responses (31). Further supporting the idea that non-hematopoietic cells can act as stimulating rather than tolerizing antigen presenting cells, MHCII intestinal epithelial cells activate CD4 T cells in graft-versus-host disease and diverse types of non-hematopoietic lung antigen-presenting cells prime T resident memory (TRM) cells (12,19). IFNγ drives MHCII expression in epithelial cells (19). Our results also demonstrate that the tumor microenvironment controls MHCII in broblasts via IFNγ. That said, positive clinical responses to immune checkpoint blockade are associated with increased IFNγ and MHCII expression (51)(52)(53). Thus, on the basis of our models, another mechanism behind the success of checkpoint inhibitors could be their ability to sustain IFNγ-dependent apCAF spots which in turn activate the unleashed effector T cells within tumors.
Current immunotherapeutic strategies largely focus on immune cells. Our works point to mesenchymal cells as key partners in cancer immunotherapies. We envisage two possible scenarios for the immunotherapeutic exploitation of our ndings: in vivo delivery of tumor antigens to apCAFs and ampli cation of the antigen presenting broblastic spots by cellular re-programming of resident epithelial cells to transition to broblasts in situ. Conceivably, the above strategies should be combined with IFNγinducing therapies, such as checkpoint inhibitors, that will sustain MHCII expression within tumors.

Methods
Human samples. Paired human lung tumour and macroscopically healthy lung resection specimens were obtained from patients with lung adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) at Sotiria Chest Hospital. The study was approved by the ethical committee of the Hospital and informed consent was obtained from all patients. LUADs and LUSCs were clinically scored and staged according to Functional ex-vivo assays. For human immunological assays, CAFs and CD4 + T cells were harvested from the same tumour fragment and dispersed into single-cell suspensions as stated above. CAFs were sorted as FSC-A High CD45 -CD31 -EPCAM -FAP + PDGFRa + cells. To preserve functional TCRs, intratumoral CD4 + T cells were sorted as FSC-A Low SSC-A Low CD45 + CD14 -CD15 -CD16 -CD19 -CD34 -CD36 -CD56 -CD123 -CD8 -CTLA-4cells. CAFs were co-cultured overnight with CD4 + T cells at a 1:1 ratio in the presence of pan-HLA antibody (5μg/ml, Biolegend, Cat. No.361702) or GC1q R antibody (10μg/ml, Abcam, Cat. No. ab24733) or isotypic controls (mouse IgG2a (Biolegend, Cat. No.361702) and mouse IgG1 (Abcam, Cat. No. ab170190) respectively). Cells were co-cultured in complete RPMI (Gibco) medium.
Classical Ficoll-Pague (StemCell, Cat. No. 07861) density gradient centrifugation was followed for PBMCs isolation from human peripheral blood. Untouched CD4 + T cells were negatively selected from In vivo studies. Gender and age-matched, over 8-week-old mice were used for all studies. All mice were housed under standard special pathogen-free conditions at BSRC Alexander Fleming. All animal procedures were approved by the Veterinary Administration Bureau, Prefecture of Athens, Greece under compliance to the national law and the EU Directives and performed in accordance with the guidance of the Institutional Animal Care and Use Committee of BSRC Al. Fleming.
For the lung cancer model, mice were anesthetized via i.p. injection with xylazine and ketamine. Cancer cells (2 × 10 5 ) resuspended in 50ul DMEM (Gibco) and enriched with 20% growth-factor reduced extracellular matrix (Matrigel, BD Biosciences) were intrapleurally injected in the lung parenchyma of mice using a 29G needle (BD Biosciences). For the metastatic cancer model, mice were injected i.v.
Monoclonal antibody aCD4/GK1.5 (ATCC/ TIB-207) was administered intraperitonealy (i.p.) 2 days before tumor implantation and continuing 3 times per week for the duration of the study (150ug/mouse). RNA extraction and qPCR. For assessment of human C1q expression, total RNA was extracted from apCAFs, nonapCAFs, blood and intratumoral CD4 + T cells that were isolated as stated above, using Single Cell RNA Puri cation Kit (Norgen Biotek) according to manufacturer's instructions. Superscript II reversetranscriptase (Thermo Fisher Scienti c) was used for cDNA synthesis and SYBR Green (Thermo Fisher Scienti c) for qPCR performed on the CFX96 Touch™ Real-Time PCR Detection System from Bio-Rad.
Transcript levels of C1q were determined relative to U6 reference gene, using the ΔΔCt method. The following primers sets were used: human C1qb Fw: TAAAAGGAGAGAAAGGGCTTCCAGGG and Rv: Image acquisition and analysis. Images were acquired using a TCS SP8X confocal system (Leica). We selected a uorophore panel which allowed for simultaneous visualization of three targets and a nuclear stain (DAPI). During acquisition uorophores were excited with 405nm (UV Laser), 488 nm (Argon), 594 nm and 647nm (White Light Laser). For images shown in Figure 1 analysis was performed using Leica LASX. Tile scanning was performed in slides stained for pan-MHCII, CD4, FAP and nuclei were stained using DAPI. Autostitching using 10% overlap was followed by analysis. Imaris 9.6 was used for subsequent image manipulations. After creating a colocalization channel between pan-MHCII and FAP all channels were used to de ne "primary objects" (surfaces -spots) used to analyze the image (distances, cell number, size, XY positioning etc.). Shortest distance calculation and object identi cation tools were used for data acquisition and image analysis. Cell density analysis was performed by identifying each "object " respectively in elds of 250.000 μm^2. Data exported from Imaris 9.6 including XY location of CD4 + , MHCII + FAP + and MHCII + FAPobjects were used for density plot creation using custom R scripts. Fibroblasts of middle and high groups were considered as MHCII + broblasts (206 cells), while those of low group as MHCIIbroblasts. Gene IDs that had zero expression in all broblasts were removed from the analysis (16877 IDs were kept). 15487 out of them were successfully translated into gene names using the biomaRt R package. After the characterization of MHCII + and MHCIIbroblasts, genes with non-zero expression values in more than 25% of each group were kept (1699 genes for apCAFs and 1388 genes for non-apCAFs). The union of the both lists was considered as the list of the genes that are expressed in at least one group. This resulted in 1785 genes.
Differential expression and pathway analysis of human datasets. To identify the differential expressed genes between the MHCII + and MHCIIwe used the FindAllMarkers function of the Seurat R package, giving as input non-scaled read values of the two clusters. We kept cells that had at least 10  Gene set enrichment analysis of CAFs for alveolar genes. Gene set enrichment analysis was performed with GSEA (v4.1.0) software. The normalized count table produced by DESeq2 was used for the GSEA according to software recommendations on the standard GSEA run. For human CAFs, the gene set included marker genes of an alveolar cluster that was identi ed in the same dataset (24). For murine CAFs, we used a gene set that included marker genes of a murine alveolar cluster that was identi ed in another dataset (54). The rest of the analysis was performed with the default thresholds.
Cross-species homology analysis. Upregulated genes in MHCII + versus MHCIIbroblasts were identi ed independently for each species, as above, and normalized by dividing the average expression of the gene plus a regularization constant (10e-4) by the average of the cluster averages plus a regularization constant (10e-4). After selecting genes with conserved gene symbols, the normalized expression matrices were log-normalized and their correlation calculated by Pearson correlation distance.
Statistics and reproducibility. For the G-squared tests of independence the g2Test_univariate function of Rfast R package was used. The p-value was calculated only for the upper tail through the pchisq function of the same package. The Pearson correlation values were calculated with cor base R function.
Visualization. PCA was performed using the prcomp base R function, with centering but no scaling. MDS plot were drawn in R with ggplot2. Box plots were generated using the ggplot2 R package and default parameters. Violin plots were generated using the geom_violin function of ggplot2 R package, with default parameters. MA plots and GO terms dot plot were generated using the geom_point function of ggplot2 R package. Plots were generated using the ggplot2 with dropping. Heatmaps were generated using the pheatmap function of pheatmap R package.
Declarations Figure 1 MHCII broblasts form CD4 T cell priming spots within human lung tumors. a) Representative FACS plots of a digested human lung tumor and expression of mesenchymal speci c genes. Cumulative data (n=5-9 patients). b) Representative whole-slide imaging for panMHCII, FAP and CD4 and cellular spatial relationship map from a lung cancer patient. After acquiring XY coordinates, XY location of MHCII+FAP+ and MHCII+FAP-cells were overlaid with CD4 cell density contour. Distance between MHCII+FAP+ or MHCII+FAP-and the closest CD4 cell. Correlations between MHCII+FAP+ densities with CD4+ cell densities (cells per 25x104μm2 regions in whole slide images, representative of n=3 patients    . c) Differential expression between puri ed intratumoral CD4 T cells from Col1a2 CreER+I-Abfl/fl versus Col1a2 CreER-I-Abfl/fl mice (n=7). Deregulated genes with FDR<0.01 are indicated. GO enrichment analysis was performed using DAVID. Network visualization was conducted by Cytoscape's EnrichmentMap. Nodes represent GO biological processes and edges connect functional terms with overlapping genes. Clusters of related nodes are circled and given labels re ecting broad biological processes. GO terms with an FDR < 0.1 are shown. d) As in A, under FTY720 treatment (n=6-7 per group, pooled from 2 experiments). a, b, d) Mann-Whitney U test. P-value<0.05