Type I IFNs promote cancer cell stemness by triggering the epigenetic regulator KDM1B

Cancer stem cells (CSCs) are a subpopulation of cancer cells endowed with high tumorigenic, chemoresistant and metastatic potential. Nongenetic mechanisms of acquired resistance are increasingly being discovered, but molecular insights into the evolutionary process of CSCs are limited. Here, we show that type I interferons (IFNs-I) function as molecular hubs of resistance during immunogenic chemotherapy, triggering the epigenetic regulator demethylase 1B (KDM1B) to promote an adaptive, yet reversible, transcriptional rewiring of cancer cells towards stemness and immune escape. Accordingly, KDM1B inhibition prevents the appearance of IFN-I-induced CSCs, both in vitro and in vivo. Notably, IFN-I-induced CSCs are heterogeneous in terms of multidrug resistance, plasticity, invasiveness and immunogenicity. Moreover, in breast cancer (BC) patients receiving anthracycline-based chemotherapy, KDM1B positively correlated with CSC signatures. Our study identifies an IFN-I → KDM1B axis as a potent engine of cancer cell reprogramming, supporting KDM1B targeting as an attractive adjunctive to immunogenic drugs to prevent CSC expansion and increase the long-term benefit of therapy.


Introduction
Despite the signi cant progress made over the last decades, the cure of cancer still remains one of the greatest medical challenges. Indeed, even neoplasms initially responding to conventional, targeted or immune-based therapies could acquire resistance and/or relapse into more aggressive and metastatic cell line and (ii) CD133 + CD44 + CD24 +low and CD133 + CD44 + CD24 +high in B16.F10 melanoma cell line (Supplementay Figure 1a). These results are in line with the intra-and inter-tumoral heterogeneity often ascribed to CSCs 28 . To assess whether this phenomenon was exclusive of the murine cancer model, we treated human U2OS osteosarcoma and MDA-MB-231 BC cell lines with recombinant human IFN-α2a and then analyzed the expression of standard human CSC markers. As expected, we detected IFN-CSC subpopulations in both U2OS (CD133 + CD44 + and CD44v6 + CD24 + cell subsets) and MDA-MB-231 (CD133 + CD44 + and CD44v6 + CD24 +low cell subsets) cell lines (Supplementary Figure 1b).
We then investigated whether IFN-CSC enrichment relied on the positive selection of pre-existing CSCs or on an active de novo induction of CSCs, a strategy cancer cells might deploy to evolve resistance. To address this question, we isolated MCA205 CD133 + and CD133 -(i.e., non-CSC) cell fractions by uorescence-activated cell sorting (FACS) and treated them with IFNs-I. Intriguingly, by ow cytometry, we found that IFN-I treatment led to a signi cant increase in the fraction of CD44H cells and the levels of the pluripotency transcription factor (TF) SRY (sex determining region Y)-box 2 (SOX2) in both the CD133 + and CD133subsets ( Figure 1b). In parallel quantitative RT-PCR (qRT-PCR) analyses of common stemrelated TFs and CSC markers, we found that exogenous IFNs-I signi cantly upregulate Kruppel-like factor 4 (Klf4), POU domain, class 5, transcription factor 1 (Pou5f1, best known as Oct3/4), Sox2 and nestin (Nes) in FACS-isolated CD133cells and Nanog homeobox (Nanog) in FACS-isolated CD133and CD133 + cells (Figure 1b). These results suggest the co-occurrence of a process of positive selection of rare, preexisting CSCs along with de novo CSC induction in response to exogenous IFNs-I.
We further analyzed the phenotypic and transcriptional pro les of IFN-CSCs, observing that IFN-I-treated epithelial cancer cells (AT3 and B16.F10) acquired a typical stem-like elongated morphology (Supplementary Figure 1c). Moreover, in distinct murine cancer cells, IFNs-I promoted the emergence of the side population (SP, a bona de CSC feature), which was signi cantly reduced by co-treatment with verapamil (VRP), a blocker of ATP-binding cassette (ABC) transporters ( Figure 1e). Notably, only spheres pre-exposed to IFNs-I retained a CSC-related phenotypical and transcriptional pro le when serially plated in standard CSC culture conditions ( Figure 1f).
These data collectively demonstrate that exogenous IFNs-I favor the appearance of putative CSCs in multiple murine and human cancer cell lines.
Since we have previously shown a key role for IFNs-I during bona de ICD 18 , we asked whether CSC subpopulations could also be enriched during immunogenic chemotherapy. We took advantage of a library of pre-validated MCA-derived clones de cient for cardinal elements of the IFN-I pathway, including: (1) Ifnar1, (2) stimulator of interferon response cGAMP interactor 1 (Sting1, best known as Sting), (3) toll-like receptor 3 (Tlr3), (4) toll-like receptor adaptor molecule 1 (Ticam1, best known as Trif), (5) interferon induced with helicase C domain 1 (I h1, best known as Mda5), and (6) mitochondrial antiviral-signaling protein (Mavs, also known as Ips-1) (Figure 2a) 18 . We exposed these clones to the ICD inducer OXP ("donor" dying cells), then co-cultured "donor" dying cells with untreated clones of the same genotype ("receiving" viable cells) for 24h, and nally analyzed "receiving" cells at phenotypic and transcriptional levels (Supplementary Figure 2a). Similar to what observed upon IFN-I treatment, wild-type (Wt) clones responding to OXP displayed a signi cant increase of the two CD44H and CD44L CSC subpopulations (Figure 2b), which we will refer to as "ICD-CSCs". Notably, ICD-CSC enrichment was impaired only in To explore the in vivo appearance of ICD-CSCs, we locally treated MCA205 tumors growing in syngeneic immunocompetent mice with DOX or CDDP and evaluated CSC enrichment in recollected xenografts 15 days post-treatment (i.e., when tumors start escaping growth control 18 ). Strikingly, we found a twofold increase of CD44H and NANOG + , upon DOX but not CDDP administration ( Figure 2g).
Altogether, these results demonstrate that IFNs-I, during immunogenic chemotherapy, promotes CSC enrichment, both in vitro and in vivo, and point to this effect as an adaptive response cancer cells may deploy to escape therapy control.
Horizontal transfer of nucleic acids from dying to viable cancer cells, upstream of IFN-I signaling, drives cancer stemness.
To dissect the molecular mechanisms underlying ICD-CSC enrichment, we co-cultured OXP-treated "donor" MCA205 cells with untreated "receiving" MCA205 cells alone or in combination with benzonase (BNZase), which degrades all nucleic acids, or RNase A, RNase H or DNase, which selectively degrade ssRNAs, dsRNAs or DNA. We observed differential effects in the two CD44H and CD44L ICD-CSC subsets, with BNZase preventing the enrichment of both CSC populations, while RNase A, RNase H and DNase signi cantly affecting only CD44L cells (Figure 3a). Accordingly, BNZase halved the proportion of ICD-CSCs in "receiving" AT3 and CT26 cells (Supplementary Figure 3a). The observation that only the depletion of all nucleic acids nulli es the enrichment of ICD-CSCs, again suggests that this phenomenon depends on an intact IFN-I signaling.
Altogether, these data indicate that ICD-CSC enrichment occurs through paracrine processes involving free and EV-mediated transfer of nucleic acids and stem-related mRNAs.
IFN-CSCs and ICD-CSCs exhibit heterogeneity of drug-response, tumorigenic and invasive potential, and immunogenicity.
We then analysed FACS-isolated CD44H and CD44L ICD-CSCs separately, and searched for hallmark CSC features like chemo-refractoriness, self-renewal ability, tumorigenic and metastatic potential, and the capability to escape immune control. We treated CD44H and CD44L cells with various ICD inducers, and found a diverse sensitivity to drugs, with only CD44H cells showing a higher therapeutic resistance than parental counterparts, both in vitro (Supplementary Figure 4a) and in vivo (Figure 4a), upon transplantation in immunocompetent mice. On the one hand, by monitoring the in vitro evolution of FACSisolated MCA205 ICD-CSCs, we demonstrated as both subsets were able to rapidly regenerate the phenotypic complexity of parental cells (Figure 4b). On the other hand, in vivo studies revealed that CD44H ICD-CSCs are signi cantly more tumorigenic and less immunogenic than CD44L ICD-CSCs. Indeed, although both subpopulations were able to generate tumors in immunocompromised NOD SCID gamma (NSG) mice, only CD44H ICD-CSCs developed neoplasms at the lowest doses and overcame immunosurveillance thus growing in immunocompetent hosts at the highest number of injected cells ( Figure 4c). Consistently, only half of the immunocompetent mice rejecting CD44H ICD-CSCs were vaccinated against viable parental cells ( Figure 4d). Conversely, CD44L ICD-CSCs or parental MCA205 cells endowed animals with 100% long-term protection against tumor re-challenge. Moreover, when intravenously injected into immunocompetent mice, CD44H (but not CD44L) ICD-CSCs developed lung metastases ( Figure 4e).
Since we identi ed CD44H ICD-CSCs as the MCA205 subpopulation mainly driving in vivo tumor aggressiveness and therapeutic resistance, we focused on this subset. To gain insights into the immunogenicity of ICD-CSCs, we analyzed the proliferation rate of isolated CD8 + H-2Kb/ovalbumin (OVA)-speci c OT-1 T cells previously primed with dendritic cells (DCs) that had taken up apoptotic OVAexpressing CD44H (CD44H-OVA) ICD-CSCs or parental cells, and then boosted with viable cells of the same type. In line with the immune privileged nature observed in vivo (Figure 4c,d), CD44H-OVA ICD-CSCs induced a signi cantly lower expansion of OT-1 CD8 T cells than parental conterparts ( Figure 5a) and resisted CD8-mediated killing ( Figure 5b). These data prompted us to hypothesize that CD44H ICD-CSCs could escape immune control by inducing CD8 T cell exhaustion. To pursue this hypothesis, we analyzed the expression of common IC ligands, nding the upregulation of PDL1, PDCD1LG2, CEA1 and LGALS9 ( Figure 5c). Consistently, CD8 + T tumor-in ltrating lymphocytes (TILs) isolated from MCA205-bearing mice 15 days after intratumoral injection of DOX (when CSC enrichment occurs), but not of CDDP (which does not enrich for CSCs), displayed a signi cant upregulation of the LGALS9 receptor IC Hepatitis A virus cellular receptor 2 (HAVCR2, best known as TIM-3) (Figure 5d). We extended the characterization of ICD-CSCs to AT3 cells (i.e., the CD24L cell subset), and con rmed their regenerative potential To further characterize ICD-CSC immunogenicity, we measured cytokine production through Luminex Multiplex Assay, observing a unique chemokine secretion pattern in CD44H MCA205 and CD24L AT3 ICD-CSCs as compared to their respective parental cells. This encompasses reduced levels of proin ammatory chemokines CCL2 and CCL5, which mediate in ammatory monocyte tra cking and DC-T cell interactions 30 , and enhanced capability to secrete CXCL1 and CXCL2 (the latter in CD24L AT3 cells), which promote chemoresistance and metastasis 31 ( Figure 5e). Notably, CD24L AT3 cells also showed higher levels of the regulatory T cell chemoattractant CCL22 32 than parental AT3 cells. Accordingly, when CD24L ICD-CSCs or parental AT3 cells were confronted with histocompatible splenocytes in ad hoc micro uidic devices 33 and then analyzed by videomicroscopy for their in vitro capability to recruit immune cells, only parental cells were able to attract and stably interact with splenocytes at as early as 24h ( Altogether, these results allowed us to make several key observations: adaptation of cancer cells to immunogenic chemotherapy enables cell selection and drives phenotype switching. Both phenomena actively contribute to intratumor heterogeneity as the collection of CSC subpopulations have differential therapeutic response, aggressiveness and immunogenicity.
Global chromatin remodeling downstream of IFNs-I.
We next inferred and reconstructed protein-protein interaction subnetworks and biological processes speci cally modulated in CD44H IFN-CSCs by using the clusterPro ler and enrichPlot R packages ( Figure  6e and Supplementary Figure 5a). Gene ontology (GO) analysis showed that most of the upregulated genes in CD44H cells (red module) have signi cant functional connections with cell growth promotion, stemness maintenance and tissue remodeling, with immune suppression (e.g., negative regulation of leukocyte activation), despite an intact response to IFN-I and IFN-II, with response to stress (e.g., positive regulation of response to DNA damage stimulus, regulation of autophagy and apoptosis), lipid metabolism, and, of note, with enhanced chromatin accessibility. Accordingly, we organized the downregulated genes (blue module) into 4 biological processes: cell growth arrest, cell differentiation, oxidative phosphorylation and protein dephosphorylation. These results provide clues about the modular re-organization of speci c pathways downstream of IFNs-I.
Of note, among the genes speci c of the CSC fraction (CD44H cells), we also identi ed multiple ISGs, including (but not limited to) I 27l2a, I 27l2b and the epigenetic regulator Kdm1b (Figure 6a,b). We were particularly intrigued by Kdm1b since chromatin remodeling plays a critical role in cancer evolution, cellular plasticity and immune escape 12,34,35,36 . At rst, we measured the enrichment of TF-binding motifs in the ATAC-seq study by using the HOMER motif software (Figure 6c and Supplementary Figure  5b). We observed signi cant differences between CSCs and parental cells, in particular we found enrichment of motifs for various TFs of the helix-turn-helix (HTH) superfamily (i.e., RFX, Rfx1, Rfx2, Rfx5 and X-box), the Homeobox basic helix-loop-helix (bHLH) member Pitx1:Ebox, the Rel homology domain (RHD) family member NFkB-p65, and the zinc nger (Zf) family member ZBTB in CD44H cells, suggesting their major role in the global chromatin remodeling in CSCs. Conversely, the Zf motifs CTCF, BORIS and NRSF, the transcriptional enhanced associate domain (TEA, TEAD) motifs (i.e., TEAD and TEAD1-4), the Rel homology domain (RHD)-basic leucine-zipper (bZIP) superfamily member NFAT-AP1, the ETS, RUNT, the interferon-sensitive response element (ISRE) and the CCAAT box-binding transcription factor (CTF) motifs were more accessible in parental cells. Thereafter, we explored the role of the ISG KDM1B in the induction of ICD-CSCs. To this aim, we added the KDM1B inhibitor tranylcypromine (TCP) to the "donor"-"receiving" in vitro co-culture and found a signi cant reduction of CD44H percentages in "receiving" cells ( Figure 6f). Of note, in vivo administration of TCP in MCA205 tumor bearing mice ( Figure   6g, left panel), prevented the enrichment of ICD-CSCs as well as the induced expression of TIM-3 on CD8 + TILs ( Figure 6g, central and right panels).
Overall, these data demonstrate that KDM1B, downstream of IFNs-I, edits the epigenome of cancer cells toward stemness, immune escape and therapy resistance ( Figure 6h).

IFN-I metagenes correlate with stemness in BC patients.
To investigate the clinical relevance of the IFN-I→KDM1B axis, we rst calculated the Spearman correlation between KDM1B, IFN-I-related metagenes, stem-related reprogramming factors and IC ligands in multiple publicly available transcriptomic data of BC patients responsive to anthracyclines 18, 37 . We  Table 1). We found a signi cant increase of CD44 + CD24 -/+low Allred scores in 15% of cases ( Figure 7f). Altogether, these results suggest the co-occurrence of IFN-I signature and CSC markers during anthracycline-based immunogenic chemotherapy.

Discussion
IFNs-I may either restrain or promote tumor growth depending on the duration and intensity of the transduced signaling, two features that jointly delineate the patterns of ISG expression, so-called "IFN signature" 21 , and shape the accessibility to chromatin, so-called "IFN-mediated epigenomic signature" 38, 39 . The leverage of transcriptional and epigenetic changes de nes cell responses to environmental hints and dictates the e cacy of natural and therapy-induced immunosurveillance 11,17,40,41 . Here, we provide preclinical and clinical evidence that the acute induction of IFNs-I, as during chemotherapy-induced ICD, favors the appearance of subpopulations of CSCs. This occurs via positive selection of CSCs coupled to a KDM1B-dependent de novo reprogramming of cancer cells toward a stem-like phenotype. Therefore, beyond stimulating antitumor immune response, IFNs-I can foster malignant progression leaving a detrimental "imprint" on cancer cells.
Our study sheds light on the debated and poorly investigated contribution of IFN-I signaling on tumor heterogeneity and CSC induction. On the one hand, we and others previously reported that IFNs-I hinder CSC generation/survival, showing that the abrogation of steady-state endogenous IFN-I signaling leads to the emergence of breast CSCs in HER2/neu transgenic mice and triple-negative BC 42,43 . On the other hand, acute production or exogenous administration of IFNs-I favored cancer stemness in mouse models of pancreatic cancer 44 and human BC and squamous carcinoma cell lines 45 . Nonetheless, in these studies the molecular mechanisms underlying IFN-I-CSC expansion have not been analyzed, and this phenomenon has been neither investigated in the context of ICD, nor associated with potential cancer cell reprogramming. In this respect, it appears of interest the evidence that the induction of the ISG IFI27 in ovarian carcinoma biopsies and cell lines drives EMT, cancer stemness, invasiveness and therapeutic resistance 46 . Whether IFI27 is involved in ICD-CSC expansion requires further investigations. Based on our results, we surmise that, when produced at chronic low levels, IFNs-I limit CSC proliferation and survival, restraining tumor growth. At odds, acute IFN-I production, as during immunogenic chemotherapy, favors the survival of pre-existing CSCs and cancer cell de-differentiation, potentially leading to therapy resistance/failure.
Here, we also found a certain degree of phenotypic and functional heterogeneity within IFN-CSCs, consistently with the current view of an adaptable, evolutive and dynamic nature of CSCs 47,48 . Moreover, we showed that speci c IFN-I-CSC subsets are characterized by high resistance to (immuno)chemotherapy and in vivo tumorigenicity, metastatic potential and low immunogenicity, in line with previous observations 49,50 . In our setting, CSC immunoprivilege encompasses a reduced capability of these cells to attract and stably interact with effector immune cells, in part due to decreased secretion of proin ammatory chemokines and enhanced capability to suppress T cell activation, and in part due to upregulated expression of IC ligands and cognate receptors. At the mechanistic level, IFN-I-related immune escape has been previously associated with the upregulation in cancer (stem) cell of (i) PD-L1 and LGALS9 23 , (ii) nitric oxide synthase 2 (NOS2), which favors the recruitment of regulatory cells 51 , and (iii) Serpinb9, which inhibits granzyme B activity and thus CD8 + T cell cytotoxicity 52 . Intriguingly, by integrating ATAC-seq and RNA-seq data, we found upregulation, in CD44H IFN-CSCs, of serpins and downregulation of Uba7, a tumor suppressor ISG which codes for a protein able to attract effector T cells 53 . Whether these factors play a major role in protecting CSC from immune attack remains to be established.
By co-culturing experiments, we demonstrated that ICD-CSC enrichment involves an autocrine/paracrine cancer cell-to-cancer cell circuitry centered on the IFN-I→IFNAR→KDM1B signaling pathway. We propose a model whereby CSC induction lies on the horizontal transfer of nucleic acids and possibly stem-related encoding mRNAs from cancer cells undergoing ICD to viable cancer cells. Notably, such intercellular communication can also occur via EVs, according to the role recently ascribed to EVs in conferring resistance and metastatic recurrence to anthracyclines 54 . In our model, once transferred from dying to viable cells, nucleic acids act as DAMPs leading to acute IFN-I production, which ultimaltely drives KDM1B-mediated cancer cell reprogramming, and thus therapy failure and tumor re-growth ( Figure 6h). Two results support this IFN-I-KDM1B centered model. First, ICD-CSC enrichment was abrogated by inhibiting the ISG KDM1B, by ablating IFNAR signaling, or by depleting the entire (but not single) spectrum of self-nucleic acids. This latter result also suggests the existence of a certain degree of redundancy in, and compensation between, nucleic acid-sensing pathways, ensuring IFN-I production also when speci c signals or sensors are depleted/missing. Second, by integrating ATAC-seq and RNAseq analyses, we revealed extensive transcriptional reprogramming in ICD-CSCs, manifested with increased expression of genes related to stemness, invasiveness and metastatization coupled to silencing of tumor suppressor and immune stimulating genes.
Although further con rmation in human models is required, we hypothesize that activation of the IFN-I signaling directly stimulates CSCs in tumors undergoing ICD. We thus surmise the existence of a mechanism similar to that underlying virus-induced cell transdifferentiation which leads to the upregulation of core pluripotency genes 25 . Supporting our hypothesis, IFNs-I were recently ascribed to have a role in chromatin remodeling and gene expression reprogramming 39,55 . Moreover, the expression of diverse KDMs has been correlated with "cold" TMEs in different tumor models, as also the use of epidrugs with the reinstatement of in ammation 56,57,58 . Recently, epidrug-related immune modulation was shown to co-occur with MYC suppression 59 . Moreover, our retrospective studies on BC patients that had received anthracycline based therapy showed signs of mutual correlation between KDM1B, IFN-Irelated, stem-associated and IC metagenes, as well as the enrichment of CSCs along with immunogenic treatments. Further validation on a larger cohort of patients with patient follow-up will be launched.
In conclusion, we demonstrated that acute IFNs-I elicit a protective but ephemeral anticancer response.    For testing the potential of induced cancer stem cells (CSCs) to growth under standard culture conditions already established for colorectal CSCs 60 , we used the CSC medium described in 60 . In more details DMEM/F12-based culture medium was supplemented with 4 mg/mL bovine serum albumin (BSA), 1X Penicillin-Streptomycin-Amphotericin B (PSF), 0.13% NaHCO3, 6 mM Hepes, 2 mM L-glutamine, 0.1 mg/mL apotransferrin, 0.4 units heparin sodium salt, 1.1% glucose, 25 µg/mL insulin, 6.3 ng/mL progesterone, 9.7 µg/mL putrescine dihydrochloride, and 5.2 ng/mL sodium selenite and supplemented with 20 ng/mL human epidermal growth factor (EGF), 10 ng/mL human basic broblast growth factor (bFGF), and 10 mM nicotinamide. CSCs were passaged once/twice a week at dilution 1:2 by mechanical dissociation, through a micropipette, and incubated in standard culture conditions in ultra-low attachment tissue culture asks.
Cyto uorometric analysis and cell sorting. To assess the expression of speci c surface markers on putative-induced CSCs, 1 x 10 5 murine and human tumor cells were cultured in 6-well plates in 2 mL of growth medium and treated 72h with puri ed mouse IFN-α/β or recombinant human Roferon-A ® (6000 U/mL) or with DOX (25 µM) or OXP (300 µM) alone or in combination with TCP (10 µM) for 48h. Cells were then collected, washed in Dulbecco's Phosphate-Buffered Saline (D-PBS) and stained with uorescently labeled mAbs directed against human/murine CD44, CD133 and/or CD24, or with puri ed- To evaluate how free nucleic acids contribute to the acquisition of CSC traits, 3 x 10 5 murine tumor cells were cultured in 6-well plates (2 mL of medium/well) and treated with 300 μM OXP for 24h ("donor" cells). Thereafter, "donor" cells were collected, washed from OXP and incubated at 37°C for up to 4h in 1,5 mL-eppendorf microtubes containing growth medium, supplemented or not, with 200 IU/mL BNZase, 10 IU/mL RNase A, 10 IU/mL RNase H or 100 IU/mL DNase. Next, such "donor cells" were cocultured with untreated live cells ("receiving" cells) for 24h in the presence or not of the indicated nucleases before cyto uorometric-mediated assessment of CSC surface markers on "receiving" cells.
Side-population (SP) assay. Semiquantitative RT-PCR. Total RNA extraction and genomic DNA removal were performed as above. RT-PCR reaction was carried out, as indicated before in the qRT-PCR section, in the presence of the same speci c primers and probes used for qRT-PCR. Ampli ed products were resolved by electrophoresis in 2% agarose gel and visualized by using SYBR Safe DNA gel staining.
Clonogenic assay. For assessing the clonogenicity of IFN-induced CSCs (referred to as IFN-CSCs), 1 x 10 3 IFN-α/β pre-treated cancer cells were seeded in 24-well plates between two layers consisting of 0.4% agarose for a nal volume of 500 μL of growth medium supplemented as in 60 (CSC medium). Cells were incubated under standard culture conditions for up to 15 days. Colonies were then xed/stained with 0.02% crystal violet (diluted in 20% methanol) and counted under an inverted microscope. Some of these spheres, prior to xation, were recovered, cultured in ultra-low attachment asks in CSC medium and analyzed for their morphology and transcriptional pro les.
Multidrug resistance assay. To determine CSC resistance to conventional chemotherapeutics, 5 x 10 3 parental and IFN-CSC MCA205 cells were seeded in 96-well plates (90 µL of medium/well) and either left untreated ("0") or treated with growing doses of OXP (3-30- T cell proliferation and cancer cell killing assays. MCA205-OVA were UV irradiated as in 63 and co-cultured with BM-derived DCs (differentiated as in 64 ) at a 2:1 ratio for 24h. DCs were then cultured at a 5:1 ratio with splenic puri ed CD8 + OT-1cells for 72h. Cross-primed CD8 + OT-1 cells were then labelled with 1 µM CFSE dye (Sigma Aldrich) for 10min at 37°C, and re-stimulated with live parental or CD44L MCA205-OVA cells at 1:5 ratio. Three days later, cells were recovered and analyzed by ow cytometry (CytoFLEX, Beckman Coulter) for CFSE levels on live gated CD8 + cells and PI levels on CD45cells. Data were analyzed by using FlowJo software v10.0.7.
Micro uidic co-culture system for tumor-immune interactions. Micro uidic devices were fabricated in polydimethylsiloxane following well-established replica molding procedures 65 . Prior to cell loading, each device was sterilized under UV light in a laminar ow hood for 30min and then lled with RPMI 1640 growth medium and incubated for 1h at 37°C to equilibrate the system. To follow chemical and physical contacts between tumor cells and immune cells, parental AT3 cells, or their ICD-induced CSC counterparts (referred to as ICD-CSCs), were co-cultured with histocompatible H-2Kb splenocytes from C57BL/6J mice.
Cells were loaded into the reservoirs of micro uidic devices as follows: the chambers were lled with 200 µl RPMI 1640 growth medium containing on one side 2 x 10 6 mouse splenocytes and on the other 5 CXCL10/IP-10; CXCL12/SDF-1α. The analysis was performed by using 50 µL of 2-fold diluted samples. and sequenced with PE42 sequencing on the NextSeq 500 sequencer (Illumina). Analysis of ATAC-seq data was performed as follows. Reads were aligned to the mouse genome (mm10) using the BWA algorithm. Duplicate reads were removed, only reads mapping as matched pairs and only uniquely mapped reads (mapping quality ≥1) were used for further analysis. Alignments were extended in silico at their 3'-ends to a length of 200 bp and assigned to 32-nt bins along the genome. The resulting histograms (genomic "signal maps") were stored in bigWig les. Peaks were identi ed using the MACS 2.1.0 algorithm at a cutoff of p-value 1e-7, without control le, and with the -nomodel option. Peaks that were on the ENCODE blacklist of known false ChIP-Seq peaks were removed. Signal maps and peak locations were used as input data to Active Motifs proprietary analysis program, which creates Excel tables containing detailed information on sample comparison, peak metrics, peak locations and gene annotations. A peak re-calling strategy was used to reduce false positives as previously done in 70 .
RNA-seq. To determine the overall transcriptional pro le, 2.5 x 10 5 parental MCA205 cells and their IFN-CSC counterparts were harvested, washed and cryopreserved in 1,5 mL eppendorf microtubes. As above, RNA-seq analysis was performed by Epigenetics Services Active Motif, Inc. Total RNA was isolated from cells by using the RNeasy Mini Kit. For each sample, 2 µg of total RNA was then used in Illumina's TruSeq Stranded mRNA Library kit. Libraries were sequenced on Illumina NextSeq 500 as paired-end 42-nt reads.
Sequence reads were analyzed with the STAR alignment -DESeq2 software pipeline described in the Data Explanation document.
Transcription factor motif discovery and network analysis. Motif enrichment analysis was performed using the HOMER software comparing the motifs enriched in the target set (i.e., the loci obtained from ATAC-Seq analysis) with those of reference (i.e., randomly selected background sequences). Only motif ratios ≥2 with P-value (Benjamini-Hochberg correction) ≤0.05 were considered biologically and statistically signi cant. The functional enrichment analysis was performed by using the clusterPro ler package as in 71 . Network visualizations were made by using the enrichPlot package.
Animals. Mice were maintained in speci c pathogen-free conditions in a temperature-controlled environment with 12h light -12h dark cycles and received food and water ad libitum. All the in vivo experimentations were in compliance with the EU Directive 63/2010 and included in an experimental protocol approved by the Institutional Animal Experimentation Committee (858/2015-PR). Six-to-7 weekold female C57Bl/6J, and NOD SCID gamma (NSG) mice were purchased from Jackson Laboratory, C57BL/6-Tg(TcraTcrb)1100Mjb/J OT1 mice were from Charles River, housed in the animal facility at the Istituto Superiore di Sanità and employed after an acclimatization period of 7 days. All experiments were randomized and blinded and sample sizes were calculated to detect a statistically signi cant effect.
Tumor models, vaccination and chemotherapy. To assess cancer cell tumorigenic capacity, 1 x 10 2 ; 1 x To in vivo uncover the ability of immunogenic chemotherapy to induce CSC appearance and to test IFN-CSC therapy response, 1 x 10 6 parental or IFN-CSC MCA205 were subcutaneously inoculated into the ank of C57Bl/6J mice and tumor growth was weekly monitored (as above described). When the tumor surface reached 35-45 mm 2 mice were randomized to control and treatment groups (10 mice/group) and injected with either CDDP (2.5 mg/kg) or DOX (2.9 mg/kg), both intratumorally in 50 μl of D-PBS, TCP (5 mg/kg) intraperitoneally every 3 days alone, a combination of DOX+TCP, or 50 μl of D-PBS intratumorally.