Abstract

   Despite the success of immunotherapy, overcoming immunoresistance in
   cancer remains challenging. We identified a unique niche of
   tumor-associated macrophages (TAMs), coexpressing T cell immunoglobulin
   and mucin domain–containing 3 (TIM3) and V-domain immunoglobulin
   suppressor of T cell activation (VISTA), that dominated human and mouse
   tumors resistant to most of the currently used immunotherapies.
   TIM3^+VISTA^+ TAMs were sustained by IL-4–enriching tumors with low
   (neo)antigenic and T cell–depleted features. TIM3^+VISTA^+ TAMs showed
   an anti-inflammatory and protumorigenic phenotype coupled with
   inability to sense type I interferon (IFN). This was established with
   cancer cells succumbing to immunogenic cell death (ICD). Dying cancer
   cells not only triggered autocrine type I IFNs but also exposed
   HMGB1/VISTA that engaged TIM3/VISTA on TAMs to suppress paracrine
   IFN-responses. Accordingly, TIM3/VISTA blockade synergized with
   paclitaxel, an ICD-inducing chemotherapy, to repolarize TIM3^+VISTA^+
   TAMs to proinflammatory TAMs that killed cancer cells via tumor
   necrosis factor–related apoptosis-inducing ligand (TRAIL) signaling. We
   propose targeting TIM3^+VISTA^+ TAMs to overcome immunoresistant
   tumors.
     __________________________________________________________________

   Immunoresistant tumors enrich TIM3^+VISTA^+macrophages, and their
   therapeutic blunting causes macrophage-driven tumor suppression.

INTRODUCTION

   Immune checkpoint signaling, involving interactions between inhibitory
   receptors and their ligands, is used by tumors to suppress anticancer
   immunity ([58]1). Accordingly, immune checkpoint blockers (ICBs)
   targeting “classical” inhibitory receptors on T cells [such as
   programmed cell death protein 1 (PD1)] or their ligands [such as
   programmed cell death protein ligand 1 (PD-L1)] have revolutionized
   oncology ([59]2). Regrettably, several tumor types remain unresponsive
   to PD(L)1 blockade ([60]3, [61]4). These immunoresistant tumors
   typically show low antigenicity and sparse T cells but enrichment of
   tumor-associated macrophages (TAMs) ([62]5). Some of the most dominant
   immunoresistance mechanisms include defects in antigen presentation or
   availability, TAM-based immunosuppression, dysregulated interferon-γ
   (IFN-γ) signaling, or enrichment of alternative inhibitory receptors
   ([63]4, [64]6). Currently, substantial research is devoted to
   overcoming these barriers; however, this pursuit has also revealed some
   challenges ([65]3, [66]4). For instance, in some cases, coblockade of
   alternative inhibitory receptors along with PD1 has shown promise,
   e.g., coblockade of lymphocyte-activation gene 3 (LAG3) and PD1 ([67]7,
   [68]8). However, in other cases, blockade of T cell immunoglobulin and
   mucin domain–containing 3 (TIM3), T cell immunoreceptor with
   immunoglobulin and immunoreceptor tyrosine-based inhibitory motif
   domains (TIGIT), or V-domain immunoglobulin suppressor of T cell
   activation (VISTA) has had mixed or disappointing early clinical
   performance ([69]8–[70]10).

   Most studies targeting inhibitory receptors concentrate on T cells
   ([71]11, [72]12). While, previously, this association was considered to
   be exclusive, it has recently been extended to TAMs ([73]13, [74]14).
   However, the immunobiology of TAM-specific inhibitory receptors
   enriched in immunoresistant tumors, and their therapeutic or clinical
   impact remains unexplored. This requires urgent attention because high
   enrichment of TAMs can facilitate overrepresentation of this inhibitory
   receptor signaling ([75]15). However, it is not clear which classical
   or alternative inhibitory receptors are preferentially exploited by
   TAMs to support immune subversion. In addition, it is essential to
   understand whether these TAMs are conserved across both human and mouse
   contexts ([76]16).

   Here, using reverse translational approaches, we identified a niche of
   TIM3^+VISTA^+ TAMs that dominated the human and mouse immunoresistant
   (CD8^+ T cell sparse) tumors. Both TIM3 and VISTA were instrumental to
   sustain an anti-inflammatory and protumorigenic TAM phenotype and to
   blunt type I IFN signaling. Accordingly, TIM3/VISTA blockade
   preferentially synergized with chemotherapy that induced type I IFN
   signaling via immunogenic cell death (ICD). This combinatorial regimen
   helped replace the anti-inflammatory footprint of TIM3^+VISTA^+ TAMs
   with a proinflammatory TAM phenotype. These TAMs drove anticancer
   cytotoxicity via tumor necrosis factor (TNF)–related apoptosis-inducing
   ligand (TRAIL) signaling, thereby blunting the immunoresistant tumors.
   This therapeutic synergism is peculiar because it triggered TAM
   autonomous control of tumor growth, independent of CD8^+ T cells or
   dendritic cells (DCs). These results are important because, in the
   current literature, the therapeutic efficacy of TIM3/VISTA blockade,
   irrespective of their myeloid or T cell expression, always converges on
   CD8^+ T cell–driven tumor control ([77]17–[78]19). Last, genetic
   footprint of TIM3^+VISTA^+ TAMs exhibited pan-cancer negative
   prognostic and predictive impact. Thus, targeting TIM3^+VISTA^+ TAMs is
   a conserved strategy to overcome low antigenic tumors.

RESULTS

HAVCR2^+VSIR^+TAMs specifically associate with immunoresistant human tumors

   First, we set to identify the dominant inhibitory receptor(s) enriched
   in the myeloid cells or macrophages (MΦ) associated with
   ICB-nonresponsive human tumors. We assembled a comprehensive
   computational framework ([79]20–[80]22) integrating existing patient
   cohorts with single-cell RNA sequencing (scRNA-seq) data. This entailed
   supervised exploratory analyses across 114 patients, 11 tumor types,
   and 36 distinct scRNA-seq cohorts, as well as unsupervised validation
   analyses across ~300 human scRNA-seq cohorts, >15 million single cells,
   and >60 organ/disease contexts including cancer ([81]Fig. 1A).

Fig. 1. HAVCR2^+VSIR^+ TAMs are specifically associated with immunoresistant
human tumors.

   [82]Fig. 1.
   [83]Open in a new tab

   (A) Overview of our computational framework for single-cell
   transcriptome mapping based on existing datasets with patient-derived
   tumor tissue (created with [84]BioRender.com). (B and C) Dot plots of
   scRNA-seq data from patients with skin cutaneous melanoma (n = 32)
   responding or not responding to PD1/CTLA4 blockade. Size of dots
   represents number of cells, and color intensity represents expression
   levels of indicated genes in single-cell profiles of (B) T cells and
   (C) myeloid cells. (D) Density plots of scRNA-seq data from 14
   treatment-naive patients across four different types of cancer (lung,
   endometrial, and colorectal adenocarcinoma and clear cell renal cell
   carcinoma) showing HAVCR2 and VSIR expression in indicated immune cell
   types. T[reg], regulatory T cell; T[em], effector memory T cell. (E)
   Dot plots of 33 scRNA-seq datasets including eight cancer types
   (non–small cell lung, head and neck, colorectal, stomach, breast, and
   pancreas cancer, hepatocellular carcinoma, and lymphoma) (n = 45). Size
   of dots represents number of cells, and color intensity represents
   expression levels of indicated genes in single-cell profiles in
   indicated immune cell types. mreg, mature regulatory. (F) Density plots
   of scRNA-seq data from 23 patients with primary colorectal cancer (CRC)
   [4 microsatellite instable (MSI) high and 19 microsatellite stable
   (MSS)] including 10 matched normal adjacent tissues, showing HAVCR2 and
   VSIR expression in SPP1^+ TAMs and S100A8/9^HIGH TAMs. (G) Aggregated
   scores for representative pathway terms based on a broader differential
   pathway enrichment analysis between HAVCR2^+VSIR^+TAMs versus
   HAVCR2^NEGVSIR^NEGTAMs in dataset from (F). Full differential pathway
   enrichment analysis in fig. S1 (C and D). AP1, activating protein 1.

   We compared the expression of nine major inhibitory receptor–coding
   genes (NT5E, VSIR, ADORA2A, CTLA4, LAG3, HAVCR2, ENTPD1, PDCD1, and
   TIGIT) ([85]20, [86]23) and a broad TAM marker (CSF1R), in T cells
   versus myeloid cells from patients with melanoma ([87]24). These
   patients were either responsive or nonresponsive to
   anti-PD1/anti–cytotoxic T-lymphocyte associated protein 4 (CTLA4) ICBs.
   T cells from ICB-nonresponsive patients showed increased expression of
   multiple inhibitory receptor genes ([88]Fig. 1B). However, myeloid
   cells showed enrichment of specifically two inhibitory receptor genes:
   HAVCR2 (codes for TIM3) and VSIR (codes for VISTA), along with CSF1R
   ([89]Fig. 1C). This coexpression of HAVCR2 and VSIR was specific for
   TAMs and was not seen in other T or B cell subsets and DCs (fig. S1A).

   To verify whether TAM-specific coenrichment of HAVCR2/VSIR was
   generalizable to other tumors, we used a multicancer and multi-immune
   cell scRNA-seq dataset ([90]Fig. 1A) ([91]25). A strong HAVCR2^+VSIR^+
   phenotype was associated mainly with TAMs ([92]Fig. 1D). This was not
   evident in DCs (fig. S1B), CD4^+/CD8^+ T cells, B cells, monocytes, or
   natural killer (NK) cells ([93]Fig. 1D and fig. S1B). While
   interesting, in above analyses, the subpopulation diversity of TAMs,
   DCs, or monocytes was not captured. To ameliorate this, we used a
   larger multicancer and multimyeloid cell scRNA-seq meta-dataset
   ([94]Fig. 1A) ([95]26). Here, the HAVCR2^HIGHVSIR^HIGH phenotype was
   dominantly associated with CSF1R^HIGHTAMs, rather than other subsets of
   TAMs, DCs, or monocytes ([96]Fig. 1E).

   Above results called for more “controlled” analyses on how
   HAVCR2^+VSIR^+TAMs associate with immunogenic versus nonimmunogenic
   tumors. To address this in a singular cancer type, we used a scRNA-seq
   dataset of patients with colorectal cancer (CRC) ([97]Fig. 1A). This is
   because CRC tumors exhibit either immunogenic [i.e., microsatellite
   instable (MSI)] or nonimmunogenic [i.e., microsatellite stable (MSS)]
   pathologies ([98]27). Data of normal adjacent tissue–associated MΦs
   were also interrogated. HAVCR2^+VSIR^+ phenotype was strongly
   associated with anti-inflammatory SPP1^+TAMs in MSS-CRC ([99]Fig. 1F),
   rather than proinflammatory (S100A8/9^HIGH) TAMs or MSI-CRC ([100]Fig.
   1F) ([101]28). A differential pathway enrichment analysis between
   HAVCR2^+VSIR^+ and HAVCR2^NEGVSIR^NEGTAMs showed that
   HAVCR2^+VSIR^+TAMs are enriched for phagocytosis, pattern recognition
   receptor (PRR) signaling relevant for danger sensing (e.g., nucleic
   acids), and anti-inflammatory signaling ([102]Fig. 1G and fig. S1, C
   and D). These comprehensive single-cell analyses indicated that the
   HAVCR2^+VSIR^+TAM niche is preferentially enriched in immunoresistant
   human tumors.

HAVCR2 and VSIR comark a distinct anti-inflammatory MΦ subset in humans

   Next, we wanted to verify whether HAVCR2 and VSIR mark a distinct
   subset of MΦs/TAMs and how they associate with different MΦ subsets or
   polarization markers. Therefore, we performed an unbiased
   bioinformatics analysis across 2,489,382 MΦs from 1679 scRNA-seq
   studies of human origin (pan-disease or pan-tissue) with 26 markers of
   different MΦ subsets or polarization states ([103]Fig. 2A)
   ([104]29–[105]34). This analysis revealed that HAVCR2^+VSIR^+MΦs are
   indeed a distinct subset, and they did not overlap with many
   well-established markers of other MΦ subsets such as IL10, ARG1, LYVE1,
   FOLR2, CCL2, and CR1 ([106]Fig. 2A). Instead, the HAVCR2^+VSIR^+MΦs
   were coclustered with primarily anti-inflammatory polarization markers,
   e.g., CSF1R, CD163, MRC1, TREM2, and SPP1. Notably, ~31% of the MΦs
   coexpressed HAVCR2 and VSIR (fig. S1E).

Fig. 2. TIM3^+VISTA^+ TAM signature shows prognostic and predictive impact in
patients with cancer.

   [107]Fig. 2.
   [108]Open in a new tab

   (A) Coexpression of indicated genes’ correlation (Jaccard’s index)
   among themselves in scRNA-seq profiles of 2,489,382 MΦs from 1679
   scRNA-seq studies of human origin (pan-disease or pan-tissue). (B)
   Automated cell ontology analyses run on an existing database of ~300
   scRNA-seq human datasets interrogated using a genetic signature of
   HAVCR2^+VSIR^+TAMs derived from [109]Fig. 1F. Final datasets were
   selected as “hits” based on a high Jaccard’s index threshold of 0.7.
   (C) Overview of the high-powered bulk tumor transcriptome and
   high-resolution scRNA-seq mapping pipeline that was used for prognostic
   and predictive validation of our biomarker (created with
   [110]BioRender.com). (D) Radar plot showing the correlation of
   immune-deconvoluted MΦ fractions with the scRNA-seq–validated
   HAVCR2^+VSIR^+ TAM signature in TCGA datasets consisting of six
   pan-cancer immune landscape classes (C1, n = 2067; C2, n = 2424, C3, n
   = 2349; C4, n = 1142, C5, n = 387; C6, n = 180) (for more details, see
   fig. S2, A and B, and table S1). (E) Hazard ratios (HRs) for the impact
   of HAVCR2^+VSIR^+ TAM signature on overall survival (OS) of patients
   with cancer in TCGA datasets consisting of six pan-cancer immune
   landscape classes (C1, n = 2067; C2, n = 2424, C3, n = 2349; C4, n =
   1142, C5, n = 387; C6, n = 180) (for more details, see fig. S2, A and
   B, and table S1). (F and G) Kaplan-Meier curve of patients with cancer
   spanning five cancer types (skin cutaneous melanoma, bladder cancer,
   kidney cancer, glioblastoma, and stomach adenocarcinoma), where tumors
   were transcriptome-profiled before anti-PD(L)1 and/or anti-CTLA4
   antibody treatments. Kaplan-Meier curve shows (F) OS (n = 152) and (G)
   progression-free survival (PFS) (n = 129) of HAVCR2^+VSIR^+TAM
   signature^HIGH versus HAVCR2^+VSIR^+TAM signature^LOW patients
   (statistical autocutoff for expression; log-rank test for P value).

   Last, we validated the above results in an unbiased analysis. Here, we
   used an automated cell ontology analysis that used above
   HAVCR2^+VSIR^+TAM signature ([111]Fig. 1G) as a guide to find similar
   immune populations in a massive database of ~300 publicly available
   scRNA-seq datasets spanning >60 normal or diseased tissues ([112]35).
   Notably, the alignment between HAVCR2^+VSIR^+TAM signature and MΦs was
   a dominant characteristic of multiple human tumors (especially of
   epithelial origin), rather than other healthy/diseased tissues
   ([113]Fig. 2B). Together, this emphasized that coexpression of HAVCR2
   and VSIR marks a distinct MΦ or TAM subset.

TIM3^+VISTA^+ TAM signature shows prognostic and predictive impact in
patients with cancer

   It was essential to verify how TIM3^+VISTA^+ TAMs embed in the human
   pan-cancer immune landscape and how they prognostically or predictively
   associate with patient survival. For this, we pursued a high-powered
   bulk tumor transcriptome mapping ([114]Fig. 2C).

   The Cancer Genome Atlas (TCGA) consortium has established six
   pan-cancer immune landscape classes (C1 to C6; for class labels, see
   radar plot in [115]Fig. 2D) (fig. S2, A and B, and table S1) ([116]36).
   We analyzed the correlation of immune-deconvoluted MΦ fractions, from
   above classes, with our scRNA-seq–validated HAVCR2^+VSIR^+TAM signature
   in 8549 patients across 30 cancer types. The more immunoresistant C4/C5
   tumors showed the highest correlation between HAVCR2^+VSIR^+TAM
   signature and MΦs, followed by mixed inflammatory C3 tumors ([117]Fig.
   2D). In terms of prognostic impact, HAVCR2^+VSIR^+TAM signature
   associated with increased hazard ratio (HR > 1) implying shorter
   overall survival (OS) in various pan-cancer immune landscape classes,
   with the only exception being the immunogenic IFN-γ–dominant tumors
   ([118]Fig. 2E). Next, we interrogated an integrated multicancer dataset
   spanning five cancer types in patients, whose tumors were
   transcriptome-profiled relative to PD(L)1 and/or CTLA4 blockade
   ([119]Fig. 2C). OS and progression-free survival (PFS) of
   HAVCR2^+VSIR^+TAM signature^HIGH patients were significantly shorter
   than HAVCR2^+VSIR^+TAM signature^LOW patients ([120]Fig. 2, F and G).
   This highlighted a pan-cancer association of the HAVCR2^+VSIR^+TAM
   signature with nonimmunogenic human tumors and its ability to predict
   shorter patient survival in prognostic and ICB response–predictive
   settings.

TIM3^+VISTA^+ TAMs are dominantly enriched in an immunoresistant murine tumor

   To reverse translate above results, we needed unbiased selection of
   immunogenic versus nonimmunogenic murine tumor models. Hence, we
   analyzed existing tumor transcriptomes of 11 commonly used
   (subcutaneous) murine models in immuno-oncology ([121]37) for genetic
   signatures of proimmunogenic IFN-γ signaling versus MΦs ([122]21,
   [123]37). Through this, we selected the following representative tumor
   models: nonimmunogenic Lewis lung carcinoma (LLC) (MΦ^HIGHIFN-γ
   signaling^LOW) versus immunogenic MC38 (MΦ^LOWIFN-γ signaling^HIGH)
   (fig. S3A). In vivo analyses showed that LLC tumors indeed had a
   CD11b^+F4/80^+TAMs^HIGHCD4^+/CD8^+ T cells^LOW phenotype and higher
   anti-inflammatory M2-like TAMs [Colony stimulating factor 1 receptor
   (CSF1R)^HIGHmajor histocompatibility complex II (MHCII)^LOW and
   CD206^HIGHMHCII^LOW] compared to MC38 tumors ([124]Fig. 3A; fig. S3, B
   and D; and flow cytometry gating in the Supplementary Materials).
   Accordingly, LLC tumors were resistant to PD1/PD-L1 blockade ([125]Fig.
   3B), whereas MC38 tumors were significantly susceptible ([126]Fig. 3C).
   When we analyzed the TAMs compartment in LLC and MC38 tumors, via an
   unsupervised computational approach, we found TIM3^+VISTA^+CSF1R^+TAMs
   in both contexts (blue populations) with visually higher accumulation
   in LLC tumors ([127]Fig. 3, D to G). The TIM3^+VISTA^+CSF1R^+TAMs
   showed tumor type–dependent heterogeneity in terms of number of
   clusters ([128]Fig. 3, D and F, and fig. S3, E and F). The LLC tumors
   highly enriched for the immunosuppressive clusters of these TAMs, i.e.,
   CD206^HIGH and CD206^HIGHMHCII^+ ([129]Fig. 3D). MC38 and LLC tumors
   primarily shared the partially activated subset, i.e., MHCII^+.
   Immunoresistant LLC tumors enriched significantly more TIM3^+VISTA^+
   TAMs than MC38 tumors ([130]Fig. 3, H and I). Confirmatively, the
   TIM3^+VISTA^+ TAMs in LLC tumors showed substantial expression of
   anti-inflammatory markers such as CSF1R and/or CD206 ([131]Fig. 3I),
   and this was higher than MC38 tumors ([132]Fig. 3, J and K). Last,
   above patterns were TIM3/VISTA-specific because for PD1, which was also
   reported on TAMs ([133]15), the trend was significantly opposite (fig.
   S3G). Notably, while TIM3^+VISTA^+phenotype was the most proficient in
   TAMs, it was also recoverable from bone marrow–derived MΦs (BMDMs)
   (fig. S3H) or J774 murine MΦ-like cells (fig. S3I), thus emphasizing
   its broad origins. In conclusion, TIM3^+VISTA^+ TAMs exhibit a
   cross-species enrichment in nonimmunogenic tumors.

Fig. 3. TIM3^+VISTA^+ TAMs have an anti-inflammatory phenotype and dominantly
enrich in an immunoresistant murine tumor.

   [134]Fig. 3.
   [135]Open in a new tab

   (A) Basal immunophenotyping of LLC and MC38 tumors, showing percentages
   of CD8^+ T cells, CD4^+ T cells, and TAMs (n = 5). See fig. S3B for the
   violin plots and statistics. (B and C) Tumor volume curve of (B) LLC
   and (C) MC38 tumor–bearing mice treated with anti-PD1/PD-L1/CTLA4
   antibody or phosphate-buffered saline (PBS) [n = 4; Kruskal-Wallis test
   corrected for false discovery rate (FDR)]. (D to G) KDE plots showing
   TAM populations in LLC and MC38 tumors from flow cytometry data of
   CD45^+ cell fraction obtained from LLC and MC38 tumors. KDE plots were
   made on the basis of the Uniform Manifold Approximation and Projections
   (UMAPs) shown in fig. S3 (E and F). (H) Percentage of TIM3^+VISTA^+
   TAMs in LLC and MC38 tumors (n = 9; Mann-Whitney test). (I) Different
   markers used in [136]Fig. 2 (D to G) for LLC and MC38 tumor samples.
   Size of dots represents fraction of cells and color intensity
   represents mean expression levels in each group. (J and K) CD45^+ cell
   fraction from LLC and MC38 tumors. (J) Ratio of M1-like
   (MHCII^HIGHCSF1R^LOW) TAMs to M2-like (CSF1R^HIGHMHCII^LOW) TAMs and
   (K) ratio of M1-like (MHCII^HIGHCD206^LOW) TAMs to M2-like
   (CD206^HIGHMHCII^LOW) TAMs of TIM3^+VISTA^+ TAMs (n = 6; Mann-Whitney
   test). (L) UMAP of scRNA-seq data from LLC tumors. (M) scRNA-seq LLC
   tumor tissue dataset, showing Vsir and Havcr2 in different immune cell
   subsets. (N) Percentage of TIM3^+VISTA^+ on CD8^+ and CD4^+ T cells,
   TAMs, and DCs—conventional DC 1 (cDC1) and cDC2, monocytic DC (moDC),
   and plasmacytoid DC (pDC)—in LLC tumors (n = 4; Kruskal-Wallis test
   corrected for FDR, comparison to TAMs). (O) pHrodo Green^+
   TIM3^+VISTA^+ CD11b^+ F4/80^+ or TIM3^−VISTA^− CD11b^+ F4/80^+ in
   cocultures of BMDMs or TAMs from LLC or MC38 tumors, cocultured with
   pHrodo labeled E. coli bioparticles (n = 4; Kruskal-Wallis test
   corrected for FDR). N number represents number of biological repeats or
   number of independent animals used.

TIM3^+VISTA^+ TAMs are hyperphagocytic with an anti-inflammatory phenotype

   We wondered whether the phenotypic human-to-mouse conservation of
   TIM3^+VISTA^+ TAMs was also recapitulated on molecular levels. For
   this, we reanalyzed our previously published scRNA-seq data for LLC
   tumor–derived CD45^+ cells ([137]Fig. 3L) ([138]38). Notably, a strong
   Havcr2^+Vsir^+ phenotype was specific for TAMs [particularly for Imm
   TAM-1 subset, also showing high expression of Ccr2, Mafb, and Csf1r
   ([139]38)] ([140]Fig. 3M). As opposed to this, none of the DC subsets,
   other myeloid cells, CD4^+/CD8^+ T cells, B cells, or NK cells showed a
   proficient Havcr2^+Vsir^+ phenotype ([141]Fig. 3M and fig. S4A). This
   was also confirmed via in vivo tumor immunophenotyping for
   TIM3^+VISTA^+ status ([142]Fig. 3N) and TIM3^+ or VISTA^+ alone (fig.
   S4B). Together, this proved that TIM3^+VISTA^+ phenotype is specific
   for TAMs in both mouse and human tumors.

   Next, we did a differential pathway enrichment analysis between
   Havcr2^+Vsir^+ versus Havcr2^NEGVsir^NEGTAMs from LLC tumors.
   Havcr2^+Vsir^+TAMs enriched for pathways connected to anti-inflammatory
   activity [e.g., IL-4/transforming growth factor–β (TGFβ) production,
   fibrinolysis, complement activation, and vascular endothelial growth
   factor receptor signaling], efferocytosis (e.g., recognition of
   apoptotic cell), and proinflammatory signaling (e.g., Toll-like
   receptor or TLR signaling) (fig. S4C). For both human and mouse
   TIM3^+VISTA^+ TAMs, phagocytotic/efferocytotic pathways were
   redundantly enriched. Since heightened efferocytotic activity is
   crucial modulator of MΦ–cancer cell interface ([143]39), we
   investigated this by exposing TAMs and BMDMs (MΦ) to pHrodo-labeled
   Escherichia coli particles. The TIM3^+VISTA^+ TAMs/MΦ were
   hyperphagocytic since they phagocytosed E. coli particles ~60 to 70%
   higher than TIM3^LOWVISTA^LOWTAMs/MΦ ([144]Fig. 3O). Together, this
   confirmed cross-species conservation of molecular characteristics of
   TIM3^+VISTA^+ TAMs.

TIM3^+VISTA^+ TAMs blunt ICD in nonimmunogenic murine tumors

   We pursued therapeutic blockade of TIM3 or VISTA via antibodies to
   blunt the TIM3^+VISTA^+ TAMs. Unexpectedly, both LLC (fig. S5A) and
   MC38 (fig. S5B) tumors were resistant to TIM3 or VISTA blockade. Even
   TIM3/VISTA and PD1 coblockade ([145]9) either did not affect the growth
   of LLC tumors (fig. S5C) or did not outperform PD1 blockade for MC38
   tumors (fig. S5D). Thus, despite being immunoresistant and strongly
   enriching TIM3^+VISTA^+ TAMs, LLC tumors were resistant to unimodal
   TIM3 or VISTA blockade. We wondered whether targeting other
   TAM-relevant pathways was more impactful than TIM3/VISTA blockade.
   Hence, on the basis of differentially enriched pathways between
   HAVCR2^+VSIR^+ versus HAVCR2^NEGVSIR^NEGTAMs (fig. S1, C and D), we
   tried in vivo targeting of several well-established TAM-relevant
   targets, i.e., signal regulatory protein α (SIRPα), CD40, complement
   receptor type 1 and 2 (CR1/2), FcγRII/III, IL-10, CCL2, IL-1β or TGFβ.
   LLC tumors were completely resistant to blockade (SIRPα, CD40, CR1/2,
   FcγRII/III, IL-10, CCL2, IL-1β, and TGFβ) or agonism (CD40) of these
   TAM-relevant targets (fig. S5E). This highlighted the extreme
   challenges of targeting immunoresistant tumors.

   This made us wonder whether proinflammatory repolarization of
   TIM3^+VISTA^+ TAMs offered better therapeutic opportunities. Above
   scRNA-seq data had indicated the affinity of TIM3^+VISTA^+ TAMs for
   efferocytosis and danger signaling. ICD is a therapeutic cell death
   subroutine, which is well established to propagate danger signaling and
   create an immunogenic efferocytotic interface with myeloid cells
   ([146]40, [147]41). Hence, we pursued combinatorial treatment with a
   widely used ICD-inducing chemotherapy, i.e., paclitaxel (PTX) ([148]40,
   [149]42). We took along a well-established non-ICD–inducing
   chemotherapy as negative control, i.e., cisplatin (CDDP) ([150]43). In
   vitro, both PTX- and CDDP-treated LLC/MC38 cancer cells underwent cell
   death (fig. S5, F to I) with apoptotic features (fig. S5, J and K) and
   passively secreted danger signals such as extracellular adenosine
   5′-triphosphate (ATP) (fig. S6, A and B) or high mobility group box 1
   (HMGB1) (fig. S6C). However, only PTX-treated cancer cells showed
   ICD-specific markers ([151]44), i.e., surface heat shock protein 90
   (HSP90)^HIGHCD47^LOW (fig. S6, D and E) and surface
   calreticulin^HIGHCD47^LOW (fig. S6, F and G). Ex vivo TIM3^+VISTA^+
   TAMs established a more proficient efferocytotic interface with
   PTX-treated cancer cells (fig. S7, A and B). TIM3 or VISTA blockade
   significantly reduced efferocytosis of dying cancer cells by MΦs (fig.
   S7, C to F), thereby validating the direct proefferocytotic
   contribution of TIM3/VISTA.

   Unexpectedly, in vivo LLC and MC38 tumors were susceptible to CDDP
   ([152]Fig. 4, A and B) but resistant to PTX ([153]Fig. 4, C and D). We
   specifically chose the same concentration for CDDP and PTX to allow us
   to compare both chemotherapies with each other. In LLC tumors, TIM3 or
   VISTA blockade synergized with only PTX to cause significant reduction
   in tumor growth ([154]Fig. 4, A and C). This synergism was absent in
   MC38 tumors ([155]Fig. 4, B and D). In both models, TIM3 or VISTA
   blockade was not beneficial in combination with CDDP ([156]Fig. 4, A
   and B).

Fig. 4. Antitumor synergism between TIM3/VISTA blockade and ICD depends on
MΦs.

   [157]Fig. 4.
   [158]Open in a new tab

   (A and B) Tumor volume curve of (A) LLC tumors and (B) MC38 tumors
   treated with CDDP or PBS with anti-TIM3/VISTA antibody (LLC, n = 8;
   MC38, n = 4; Kruskal-Wallis test corrected for FDR). (C and D) Tumor
   volume curve of (C) LLC tumors and (D) MC38 tumors treated with PTX or
   PBS with anti-TIM3/VISTA antibody (LLC, n = 10; MC38, n = 4;
   Kruskal-Wallis test corrected for FDR). (E to G) CD45^+ cell fraction
   obtained from (C). (E) Percentage of TIM3^+VISTA^+ TAMs (n = 8;
   Kruskal-Wallis test corrected for FDR). (F) Ratio of M1-like
   (MHCII^HIGHCSF1R^LOW) TAMs to M2-like (CSF1R^HIGHMHCII^LOW) TAMs (n =
   8; Kruskal-Wallis test corrected for FDR). (G) Percentage of CD8^+ T
   cells (n = 8; Kruskal-Wallis test corrected for FDR). (H and I) Tumor
   volume curve of LLC tumors treated with PTX or PBS and/or
   anti-TIM3/VISTA antibody (H) combined with clodronate liposomes (n = 9;
   Kruskal-Wallis test corrected for FDR, comparison to PBS) and (I)
   combined with anti-CD8 antibody (n = 8; Kruskal-Wallis test corrected
   for FDR, comparison to PBS). (J and K) Lysate from CD45^− cell fraction
   obtained from (C). (J) Protein levels of different analytes.
   Incalculable values are depicted as gray (n = 4). (K) Pro–caspase 9,
   cleaved caspase 9, and actin Western blot. Ratio of quantification of
   pixel intensity is represented in a violin plot (n = 4; Kruskal-Wallis
   test corrected for FDR.) (L) Percentages of cancer cell death after
   coincubation with TAMs with different inhibitors [n = 4; two-way
   analysis of variance (ANOVA) corrected for FDR]. (M) Overview of a
   CRISPR-Cas9–mediated Vsir and Havcr2 knockout in CSF1R^+ cells (MΦs)
   (created with [159]BioRender.com). WT, wild-type. (N to P) LLC
   tumor–bearing TIM3^−VISTA^− MΦ or control knockout MΦ mice, treated
   with PTX or PBS. (N) Percentages of CSF1R^+ and CSF1R^− TIM3^+VISTA^+
   TAMs (n = 5; Kruskal-Wallis test corrected for FDR). (O) Percentages of
   TIM3^+VISTA^+ TAMs (n = 5; Kruskal-Wallis test corrected for FDR). (P)
   Tumor volume curve (control knockout PTX, n = 8; others, n = 6;
   Kruskal-Wallis test corrected for FDR).

   Notably, combination of PTX with TIM3/VISTA blockade significantly
   reduced the TIM3^+VISTA^+CSF1R^+TAMs ([160]Fig. 4E). This combination
   skewed these TAMs toward a more proinflammatory M1-like
   (MHCII^HIGHCSF1R^LOW) phenotype as compared to anti-inflammatory
   M2-like (MHCII^LOWCSF1R^HIGH) ([161]Fig. 4F). Above switch from anti-
   to proinflammatory phenotype was TIM3/VISTA-driven because in LLC-MΦ
   cocultures in vitro, TIM3 or VISTA blockade together with PTX-treated
   LLC cells caused increased enrichment of activated MHCII^HIGHMΦs (fig.
   S8, A and B). Notably, the triple combination of PTX treatment with
   TIM3 and VISTA coblockade was not significantly different from doublet
   combinations (fig. S8C). Thus, TIM3/VISTA blockade synergizes with
   ICD-inducing chemotherapy to repolarize TIM3^+VISTA^+ TAMs toward
   proinflammatory phenotype, which associated with blunting of tumors.

Antitumor synergism between TIM3/VISTA blockade and ICD depends on TAMs

   Blunted LLC tumors did not experience increased infiltration of
   CD4^+/CD8^+ T cells ([162]Fig. 4G and fig. S9A). An extended
   phenotyping of T cells further highlighted that regressing LLC tumors
   failed to enrich any of the typical antitumor T cell subsets such as
   proliferating Ki67^+CD8^+/CD4^+ T cells (fig. S9, B and C), effector
   memory CD127^+CD62L^−CD8^+/CD4^+ T cells (fig. S9, D and E), effector
   IFN-γ^+ (fig. S9, F and G) or IL-2^+ (fig. S9, H and I) CD8^+/CD4^+ T
   cells, or cytotoxic CD107a^+CD8^+ T cells (fig. S9J). Similarly, the
   regressing LLC tumors did not enrich any of the different DC subsets
   (fig. S10, A to C) or their matured/activated states (fig. S10, D to
   F). This indicated that the synergism of TIM3/VISTA blockade with PTX
   primarily associated with activation of TAMs.

   To validate this, we depleted MΦs via clodronate liposomes and CD8^+ T
   cells via anti-CD8 antibodies ([163]45). Clodronate liposomes were
   applied at a titrated dose that depleted TAMs and splenic MΦs (fig.
   S11, A and B) but did not affect DCs (fig. S11, A and B). Clodronate
   liposome–based MΦ depletion already substantially affected tumor growth
   compared to the phosphate-buffered saline (PBS) control ([164]Fig. 4H).
   MΦ depletion completely reshaped the chemotherapy susceptibility of LLC
   tumors. PTX treatment by itself reduced the tumor growth now ([165]Fig.
   4H), such that the synergism of PTX with TIM3 or VISTA blockade
   disappeared ([166]Fig. 4H). To further investigate this direct
   modulation of PTX resistance, we sorted TIM3^+VISTA^+ TAMs and
   TIM3^NEGVISTA^NEGTAMs from LLC tumors. These TAMs were cocultured with
   untreated (UT) LLC cancer cells, followed by PTX treatment. Presence of
   TIM3^+VISTA^+ TAMs mediated resistance to PTX, since under,
   specifically, this condition, we did not observe LLC cell death (fig.
   S11C). In contrast, the depletion of CD8^+ T cells (fig. S11, D and E)
   did not significantly disrupt synergism of PTX and TIM3/VISTA blockade
   ([167]Fig. 4I).

   Next, we wished to molecularly confirm these patterns. Hence, we used
   an antibody array to analyze protein levels of multiple
   cytokines/chemokines or other immune factors ([168]43) in CD45^+ cells
   from LLC tumors. Under all conditions, there was no enrichment of
   typically adaptive immune cytokines/factors (fig. S12A). However, a
   very specific enrichment of cytokines/chemokines/factors relevant for
   myeloid cells or TAMs was seen in UT and PTX-treated LLC tumors
   ([169]Fig. 4J). These were almost completely depleted in LLC tumors
   suppressed by treatment with PTX and TIM3/VISTA blockade ([170]Fig.
   4J). Thus, the therapeutic synergism of TIM3/VISTA blockade and
   ICD-inducing chemotherapy is completely dependent on only TAMs.

TIM3^+VISTA^+ TAMs autonomously enforce anticancer cytotoxicity via TRAIL
signaling

   Next, we were curious about the exact molecular pathway used by
   TIM3^+VISTA^+ TAMs to suppress LLC tumors in absence of antitumor T
   cells. LLC tumors suppressed by our combinatorial regimen showed a
   significant increase in apoptosis (marked by cleaved caspase 9)
   ([171]Fig. 4K). In ex vivo cocultures of LLC tumor–derived TAMs and LLC
   cells, blockade of TIM3 or VISTA potentiated proapoptotic stress in
   LLCs (fig. S12B). All this suggested that, perhaps, TAMs were directly
   killing cancer cells.

   To validate whether, and how, TAMs were directly killing the cancer
   cells, we isolated TAMs from LLC tumors that were either treated with
   PBS or PTX and blocked major MΦ-associated cytotoxicity pathways
   ([172]16, [173]46) in presence of UT LLC cancer cells ex vivo
   ([174]Fig. 4L). Specifically, we blocked the respiratory burst through
   various agents: apocynin (blocks nicotinamide adenine dinucleotide
   phosphate oxidase complex–based H[2]O[2] production) ([175]46),
   catalase (degrades H[2]O[2]) ([176]47), and N^G-monomethyl-l-arginine,
   monoacetate salt (L-NMMA) (blocks nitric oxide synthase–based NO
   production) ([177]48, [178]49); and we blocked various effectors of the
   extrinsic cell death pathway via antibodies, i.e., TNF ([179]49), TRAIL
   ([180]16), and Fas ligand (FasL) ([181]50). These cocultures confirmed
   that PTX-treated tumor-derived TAMs, in presence of TIM3/VISTA
   blockade, caused significant LLC cancer cell death ([182]Fig. 4L). Of
   all the different pathway ablations, this TAM-driven cancer cell death
   was specifically disrupted only by the blockade of TRAIL ([183]Fig.
   4L). This suggested that proinflammatory activation of TIM3^+VISTA^+
   TAMs created TAM-driven anticancer cytotoxicity via TRAIL signaling.

TAM-specific genetic ablation of TIM3 and VISTA synergizes with ICD-inducing
chemotherapy to blunt nonimmunogenic tumors

   Since TIM3^+VISTA^+ TAMs were connected with CSF1R^+TAMs ([184]Fig.
   3I), we wondered whether the synergism of PTX with TIM3/VISTA blockade
   was extendable to antibody-based CSF1R blockade. However, in vivo
   analyses did not show substantial tumor reduction upon combining PTX
   with CSF1R blockade (fig. S12C). Notably, while the CSF1R blockade
   significantly depleted CSF1R^+TAMs (fig. S12D), it failed to deplete
   TIM3^+VISTA^+CSF1R^+TAMs (fig. S12E). Inefficiency of CSF1R blockade in
   depleting particular TAM subsets in vivo has been reported previously
   ([185]36). These discrepant results demanded a stronger
   TIM3/VISTA-specific genetic ablation approach directed toward
   CSF1R^+TAMs.

   To accomplish this specifically in CSF1R^+MΦs in vivo, we first used
   lentiviral particles simultaneously expressing guide RNAs (gRNAs)
   against both Havcr2 and Vsir to transduce hematopoietic stem/progenitor
   cells (HSPCs) from Csf1r:Cre-ERT/R26:LSL-Cas9 mice (donor mice; CD90.1
   congenic strain), carrying a (conditional) tamoxifen-inducible Cas9 in
   CSF1R^+ cells ([186]Fig. 4M). In this way, Havcr2 (fig. S12F) and Vsir
   (fig. S12G) were simultaneously knocked down by at least 60% in the
   MΦs. Transduced HSPCs were transplanted into lethally irradiated
   recipients to create recipient mice with tamoxifen-inducible
   Havcr2^_Vsir^_ phenotype in CSF1R^+MΦs (i.e., Havcr2^_Vsir^_∆MΦ
   phenotype) ([187]Fig. 4M). Control mice were set up with nontargeting
   control gRNA. Both these mice groups were either left UT (PBS) or
   treated with PTX ([188]Fig. 4M). Confirmatively, all these mice had
   similar levels of circulating CD90.1^+ (donor) immune cells (fig.
   S12H), and TIM3^+VISTA^+CSF1R^+ MΦs were almost completely depleted in
   the TIM3^−VISTA^−∆MΦ mice ([189]Fig. 4N), highlighting the knockout
   efficiency of the model. In addition, there was a reduction of >60% in
   enrichment of TIM3^+VISTA^+ TAMs ([190]Fig. 4O), and this reduction in
   TIM3^+VISTA^+ phenotype was specific for TAMs (fig. S12I) rather than
   tumor-derived DCs or CD4^+/CD8^+ T cells (fig. S12, J to L),
   highlighting the knockout specificity of the used model.

   Using these mice, we observed that Havcr2^_Vsir^_ MΦs already resulted
   in a modest, albeit significant, reduction in LLC tumor growth compared
   to control mice ([191]Fig. 4P). However, the most significant reduction
   of LLC tumor growth was achieved upon PTX treatment, thereby confirming
   synergism between this chemotherapy and MΦ-specific Havcr2/Vsir
   ablation ([192]Fig. 4P). Accordingly, PTX treatment and MΦ-specific
   Havcr2/Vsir ablation resulted in the highest enrichment of
   proinflammatory MHCII^HIGHCSF1R^LOWTAMs (fig. S12M). There was no
   increase in infiltration of CD8^+/CD4^+ T cells or IFN-γ^+CD8^+ T cells
   (fig. S12, N to P). This confirmed that TAM-specific TIM3/VISTA
   blockade is necessary for maximal tumor blunting by ICD-inducing
   chemotherapy.

Low tumoral antigenicity drives IL-4–driven accumulation of TIM3^+VISTA^+
TAMs

   It was essential to uncover the mechanism behind preferential
   enrichment of TIM3^+VISTA^+ TAMs in LLC tumors. Hence, we pursued
   differential gene expression (DGE) analysis of bulk transcriptomes of
   LLC versus MC38 tumors (fig. S3A). LLC tumors enriched for
   immunosuppressive genes linked to prostaglandin E2 (PGE2) signaling
   (Ptgs1 and Cox5a), IL-1 signaling (Il18rap and Il1r1), TGFβ signaling
   (Tgfbr3), anti-inflammatory TAMs (Mrc1, Ccr2, Cd33, and Gas2), or
   IL-4/IL-13 pathway (Il4ra) ([193]Fig. 5A). Contrastingly, genes
   relevant for immunogenic pathways including IFN-stimulated gene (ISG)
   response (Rsad2, Oasla, Oas2, Oasl1/2, Mx2, Stat1/2, Ifi44, and Cxcl9),
   cytotoxic lymphocyte activity (Gzmf, Gzmd, and Gzmb), and
   tumor-relevant antigens (Magea10 and Magea1) were enriched in MC38
   tumors ([194]Fig. 5A).

Fig. 5. TIM3 and VISTA inhibit TAM-intrinsic ISG responses, thereby blunting
ICD.

   [195]Fig. 5.
   [196]Open in a new tab

   (A) Volcano plot of gene expression alterations between MC38 (red) and
   LLC tumors (green). (B) Mean fluorescence intensity (MFI) of TIM3 on
   VISTA^+ BMDMs (MΦ) stimulated with different cytokines
   (IL-4/IL-13/IL-18, n = 3; others, n = 4; Kruskal-Wallis test corrected
   for FDR). (C to F) IL-4 and IL-13 secretion in tumor lysate. Ratio of
   absolute values with blank was taken and ratio to total TAMs. (C and D)
   IL-4 (C) and IL-13 (D) production in LLC and MC38 tumors (n = 4;
   Mann-Whitney test). (E and F) IL-4 (E) and IL-13 (F) production in LLC
   and LLC-OVA tumors (n = 4; Mann-Whitney test). (G) Percentages of CD8^+
   T cells, CD4^+ T cells, and TAMs (n = 4). See fig. S13D for the violin
   plot and statistics. (H) Percentage of TIM3^+VISTA^+CSF1R^+ TAMs in LLC
   and LLC-OVA tumors (n = 5, Mann-Whitney test). (I) Tumor volume curves
   of LLC-OVA tumors (n = 4; Kruskal-Wallis test corrected for FDR). (J)
   IFN-β secretion of LLC cancer cells alone and their cocultures with
   BMDMs, with or without blockade with indicated antibodies (Abs) (n = 4;
   Kruskal-Wallis test corrected for FDR). (K) Quantitative polymerase
   chain reaction (qPCR) of ISG expression (n = 3). (L and M) IFN/ISG
   response in J774 MΦs with blockade of isotype or (L) anti-TIM3 and (M)
   anti-VISTA antibody and their cocultures with LLC [(M) PTX, n = 4;
   others, n = 3; Kruskal-Wallis test corrected for FDR]. (N to Q) LLC
   alone and their cocultures with MΦ with blockade of indicated
   antibodies. (N and O) IFN-β secretion in (N) MΦ blocked with
   isotype/anti-TIM3 antibody or (O) isotype/anti-VISTA antibody (n = 3;
   Kruskal-Wallis test). (P and Q) IFN/ISG response in J774 MΦs (P)
   blocked with isotype/anti-TIM3 antibody or (Q) isotype/anti-VISTA
   antibody [TIM3 LLC untreated (UT), n = 3; others, n = 4; VISTA LLC PTX,
   n = 4; others, n = 3; Kruskal-Wallis test corrected for FDR]. (R) Tumor
   volume curves of LLC tumors in ifnar1^−/− mice (n = 8; Kruskal-Wallis
   test corrected for FDR).

   We wondered whether one or more of the promyeloid factors that featured
   prominently in LLC tumors caused accumulation of TIM3^+VISTA^+ TAMs.
   Hence, we tested the ability of above LLC tumor–specific
   cytokines/factors to differentiate murine BMDMs. We incubated these MΦs
   with TGFβ, PGE2, IL-1β, IL-18, IL-4, IL-13, and an anti-inflammatory
   differentiation cocktail for MΦs (IL-4 + IL-13), in vitro ([197]Fig.
   5B). The most potent inducer of TIM3^+VISTA^+phenotype was the cytokine
   cocktail, primarily led by IL-4 ([198]Fig. 5B). This aligned with the
   strong anti-inflammatory (fig. S1, C and D) and IL-4 signaling
   character (fig. S4C) of Havcr2^+Vsir^+TAMs in LLC tumors. Therefore, we
   also looked at the in vivo enrichment of IL-4 or IL-13 relative to
   total TAMs (to account for the difference in TAMs between LLC and MC38
   tumors). We observed that TAMs in LLC tumors were significantly more
   exposed to IL-4 ([199]Fig. 5C) or IL-13 ([200]Fig. 5D) compared to the
   MC38 tumors. Since IL-4 was the most potent inducer of
   TIM3^+VISTA^+phenotype, we neutralized IL-4 via a depleting antibody in
   vivo, in both MC38 and LLC tumors (fig. S13, A and B). MC38 tumors were
   not affected upon neutralizing IL-4. In contrast, growth of LLC tumors
   were significantly reduced by neutralizing IL-4 (fig. S13B), such that
   IL-4 neutralization also significantly reduced the amount of
   TIM3^+VISTA^+ TAMs (fig. S13C).

   Tumoral IL-4 is a sign of type 2 anti-inflammatory responses ([201]16,
   [202]51). These responses dominate immunoresistant tumors; however,
   their dominance is more limited in antigenic/immunogenic tumors
   ([203]52). Since MC38 tumors showed higher antigenicity/immunogenicity
   than LLC tumors, we wondered whether the lack of strong antigens within
   LLC tumors ultimately facilitated IL-4 enrichment, thereby sustaining
   TIM3^+VISTA^+ TAMs. To ameliorate this, we genetically manipulated the
   LLC cells to express the well-established model antigen, ovalbumin
   (OVA). Notably, simply expressing OVA-antigen in LLCs converted the
   resulting in vivo tumor milieu into a more MC38-like, i.e., TAMs in
   LLC-OVA tumors were exposed to lower IL-4/IL-13 levels ([204]Fig. 5, E
   and F), and the immune landscape was more immunogenic
   (TAMs^LOWCD4^+/CD8^+ T cells^HIGH) ([205]Fig. 5G and fig. S13D).
   Accordingly, LLC-OVA tumors enriched significantly less TIM3^+VISTA^+
   TAMs ([206]Fig. 5H) and even those that they did enrich had more
   proinflammatory orientation (CSF1R^LOWMHCII^HIGH) (fig. S13E).
   Accordingly, the therapeutic synergism between PTX and TIM3/VISTA
   blockade disappeared in LLC-OVA tumors ([207]Fig. 5I). This established
   that low antigenic tumors facilitate an IL-4–driven exaggerated
   accumulation of TIM3^+VISTA^+ TAMs.

TIM3 and VISTA inhibit TAM-intrinsic ISG responses, thereby blunting ICD

   The mechanistic basis behind synergism of ICD-inducing PTX with
   TIM3/VISTA blockade was still unclear. However, one observation in
   above DGE analyses provided us with a clue. While the ISG response was
   enriched in MC38 tumors, a gene coding for type I IFN receptor (Ifnar1)
   was enriched in LLC tumors, indicating a misalignment between type I
   IFN sensing and downstream ISG response. This needed attention because
   ICD relies on type I IFNs to exert antitumor efficacy ([208]53).

   Accordingly, only LLC cells undergoing ICD via PTX autonomously
   secreted type I IFN (IFN-β) ([209]Fig. 5J). This was not significantly
   potentiated upon cocultures with MΦs and TIM3/VISTA blockade, thus
   emphasizing LLCs as primary IFN-β producers ([210]Fig. 5J). Next, we
   checked levels of downstream ISGs (Rsad2, Mx1, Irf7, and Cxcl10), and,
   unexpectedly, no clear ISG response was detected when PTX-treated LLCs
   were cocultured with MΦs ([211]Fig. 5K). This implied that IFN-β failed
   to activate ISG response. TIM3 or VISTA blockade successfully
   reinstated the ISG response ([212]Fig. 5K). In a coculture setting, it
   is hard to confirm whether ISG response is intrinsic to cancer cells or
   MΦs. To overcome this, we repeated above analyses with J774 MΦs
   expressing a luciferase-based genetic reporter linked to downstream ISG
   response ([213]21). With this, we successfully reproduced the above
   patterns ([214]Fig. 5, L and M) and confirmed that TIM3/VISTA
   blockade–driven ISG response was intrinsic to MΦs.

   ICD-associated type I IFN production is driven by dying cell-derived
   extracellular nucleic acids ([215]43, [216]53). To verify this, we
   repeated above experiments while degrading nucleic acids via
   deoxyribonuclease (DNase)/ribonuclease (RNase) ([217]43). Degradation
   of extracellular nucleic acids strongly ablated production of IFN-β
   ([218]Fig. 5, N and O) and MΦ-intrinsic ISG responses ([219]Fig. 5, P
   and Q). Thus, although ICD triggered autocrine type I IFNs via sensing
   of extracellular nucleic acids, the paracrine ISG response in
   interacting MΦs was blunted by TIM3/VISTA signaling.

   This suggested that TIM3/VISTA blockade synergized with PTX because it
   reestablished IFN-β–induced ISGs in TIM3^+VISTA^+ TAMs. Since host type
   I IFN receptors would be the key to this, we pursued in vivo validation
   with Ifnar1^−/−mice. Therapeutic synergism between PTX and TIM3/VISTA
   blockade was completely lost in Ifnar1^−/−mice ([220]Fig. 5R).
   Accordingly, this combination failed to proinflammatorily activate
   TIM3^+VISTA^+ TAMs (fig. S13F). Thus, TIM3/VISTA signaling blunts
   MΦ-associated ISGs, thereby disrupting ICD-inducing chemotherapy’s
   ability to initiate type I IFN–driven antitumor immunity.

Cancer cell–associated HMGB1 and VISTA engage TIM3 and VISTA on TAM to
suppress ISGs

   TIM3- and VISTA-based blunting of macrophagic ISGs cannot function
   without ligand-based engagement of TIM3/VISTA by cancer cells. Compared
   to UT LLC cells, LLC cells treated with PTX proficiently exposed
   multiple ligands for both TIM3, i.e., surface Ceacam1 ([221]Fig. 6A),
   surface Galectin-9 ([222]Fig. 6B), and extracellular HMGB1 ([223]Fig.
   6C) ([224]54), as well as VISTA, i.e., surface P-selectin glycoprotein
   ligand 1 (PSGL1) ([225]Fig. 6D) and surface VISTA ([226]Fig. 6E)
   ([227]55). Notably, VISTA has an exceptional receptor system, i.e.,
   VISTA on MΦs engages in homophilic intermolecular interactions with
   VISTA on target cells (such as cancer cells) ([228]56). To understand
   which of these ligands was responsible for engaging TIM3/VISTA on MΦs
   to blunt ISGs, we first used CRISPR-Cas9 to individually knockout the
   genes encoding for each of these ligands (fig. S13, G to K). Next, we
   treated these cells with PTX and cocultured them with J774 MΦs
   expressing the ISG reporter system. This revealed that only Hmgb1^−/−
   LLCs ([229]Fig. 6F) or Vsir^−/− LLCs ([230]Fig. 6G) successfully
   disrupted the J774 MΦ-intrinsic ISGs facilitated by the TIM3 or VISTA
   blockade, respectively. This suggested that TIM3/VISTA blockade
   synergized with PTX because of cancer cell–associated HMGB1/VISTA. To
   validate this, we pursued in vivo experiments with Hmgb1^−/− or
   Vsir^−/− LLCs. The synergism between PTX and TIM3/VISTA blockade was
   ablated in both Hmgb1^−/− ([231]Fig. 6H) and Vsir^−/− ([232]Fig. 6I)
   LLC tumors. Thus, dying cancer cells exposed HMGB1 and VISTA that
   engaged TIM3 and VISTA on MΦs to blunt latter’s ISG responses.

Fig. 6. Cancer cell–associated HMGB1 and VISTA engage TIM3 and VISTA on MΦs
to suppress ISG responses.

   [233]Fig. 6.
   [234]Open in a new tab

   (A and B) MFI of (A) Ceacam1 and (B) Galectin-9 on LLC cells UT or
   treated for 48 hours with 100 μM PTX (n = 4; Mann-Whitney test). (C)
   Western blot showing extracellular released HMGB1 in LLC UT or treated
   for 48 hours with PTX supernatant. (D and E) MFI of (D) PSGL1 and (E)
   VISTA on LLC cells UT or treated for 48 hours with PTX (n = 4;
   Mann-Whitney test). (F and G) IFN/ISG response in J774 reporter MΦs
   with blockade with isotype antibodies or (F) anti-TIM3 antibody and
   their cocultures with CRISPR-Cas knockout LLC cancer cells for
   Adeno-associated virus serotype 1 (AAV1) empty (control), LGALS9
   (Galectin-9), HMGB1 (HMGB1), and CEACAM1 (Ceacam1), UT or treated with
   either PTX and (G) anti-VISTA antibody and their cocultures with
   CRISPR-Cas knockout LLC cancer cells for AAV1 empty (control), VSIR
   (VISTA), and SELPLG (PSGL1), UT or treated with either PTX. Values are
   fold-changed to UT (n = 4; Kruskal-Wallis test corrected for FDR). (H)
   Tumor volume curves of LLC HMGB1^−/− tumors treated with PTX or PBS
   and/or combination with anti-TIM3 antibody (n = 8; Kruskal-Wallis test
   corrected for FDR). (I) Tumor volume curves of LLC VSIR^−/− tumors
   treated with PTX or PBS and/or combination with anti-VISTA antibody (n
   = 8; Kruskal-Wallis test corrected for FDR). N number represents the
   number of biological repeats or number of independent animals used.

Targeting TIM3^+VISTA^+ TAM impedes tumor growth in immunoresistant and
neoantigen^LOW melanoma

   It was necessary to validate above (major) features underlying
   TIM3^+VISTA^+ TAM enrichment and their immunotherapeutic targeting in
   physiologically superior orthotopic-like tumor models. Hence, we
   pursued validation with three subcutaneous melanoma models, i.e.,
   Braf^V600ECdkn2a^−/−Pten^−/−YUMM1.7 (neoantigen^LOW), YUMM1.7
   expressing the OVA-antigen, and neoantigen^HIGHYUMMER1.7 ([235]57,
   [236]58).

   In line with these antigenicity trends, YUMM1.7 melanoma showed a
   stronger TAMs^HIGHCD4^+/CD8^+ T cells^LOW phenotype ([237]Fig. 7A and
   fig. S14A) than YUMM1.7-OVA ([238]Fig. 7B and fig. S14B) or YUMMER1.7
   ([239]Fig. 7C and fig. S14C) melanomas in vivo. Confirmatively, YUMM1.7
   melanomas harbored much higher TIM3^+VISTA^+CSF1R^+TAMs ([240]Fig. 7D)
   and higher IL-4/IL-13 exposure toward TAMs (fig. S14, D and E) than
   YUMM1.7-OVA or YUMMER1.7 melanomas. No differences in TIM3 and VISTA
   expression on the level of CD8^+ T cells were observed between the
   models (fig. S14, F to H). Together, these distinct immune landscapes
   predicted that YUMM1.7 melanoma are resistant to PD1 blockade
   ([241]Fig. 7E), whereas YUMM1.7-OVA ([242]Fig. 7F) and YUMMER1.7
   ([243]Fig. 7G) melanomas were significantly susceptible.

Fig. 7. Targeting TIM3^+VISTA^+ TAM blunts immunoresistant and low neoantigen
melanoma.

   [244]Fig. 7.
   [245]Open in a new tab

   (A to C) Basal immunophenotyping of (A) YUMM1.7, (B) YUMM1.7-OVA, and
   (C) YUMMER1.7 tumors showing percentages of CD8^+ T cells, CD4^+ T
   cells, and TAMs (n = 6). See fig. S14 (A to C) for the violin plots and
   statistics corresponding to these pie charts. (D) Flow cytometry
   analysis of CD45^+ cell fraction obtained from subcutaneous YUMM1.7,
   YUMM1.7-OVA, and YUMMER1.7 tumors. Percentage of
   TIM3^+VISTA^+CSF1R^+TAMs (YUMM1.7, n = 5; others, n = 6; Mann-Whitney
   test). (E to G) Tumor volume curve of (E) YUMM1.7, (F) YUMM1.7-OVA, and
   (G) YUMMER1.7 tumor–bearing mice treated with anti-PD1 antibody (n = 4;
   Kruskal-Wallis test corrected for FDR). (H to J) Tumor volume curve of
   (H) YUMM1.7, (I) YUMM1.7-OVA, and (J) YUMMER1.7 tumor–bearing mice
   treated with PTX or PBS and/or combination with anti-TIM3 or anti-VISTA
   antibody (YUMM1.7, n = 8; others, n = 4; Kruskal-Wallis test corrected
   for FDR). (K to P) Flow cytometry analysis of CD45^+ cell fraction
   obtained from subcutaneous YUMM1.7, YUMM1.7-OVA, and YUMMER1.7 tumors
   treated with PTX or PBS and/or combination with anti-TIM3 or anti-VISTA
   antibody. (K to M) Percentage of TIM3^+VISTA^+CSF1R^+ TAMs, YUMM1.7
   (K), YUMM1.7-OVA (L), and YUMMER1.7 (M) subcutaneous tumors (YUMM1.7, n
   = 8; others, n = 4; Kruskal-Wallis test corrected for FDR). (N to P)
   Percentage of CD3^+CD8^+ T cells of live cells in YUMM1.7 (N),
   YUMM1.7-OVA (O), and YUMMER1.7 (P) subcutaneous tumors (YUMM1.7, n = 8;
   others, n = 4; Kruskal-Wallis test corrected for FDR). N number
   represents the number of biological repeats or number of independent
   animals used.

   Next, therapeutic testing confirmed that, while all melanoma models
   were resistant to PTX monotherapy, the combination of PTX and
   TIM3/VISTA blockade synergized to significantly blunt the growth of
   only immunoresistant melanoma YUMM1.7 ([246]Fig. 7H) but not
   immunosusceptible counterparts YUMM1.7-OVA ([247]Fig. 7I) or YUMMER1.7
   ([248]Fig. 7J). Above therapeutic synergism was accompanied by
   significant reduction in TIM3^+VISTA^+ TAMs only in YUMM1.7 melanoma
   ([249]Fig. 7, K to M). Last, for CD8^+ T cells, we observed no
   statistically significant differences across all three melanoma models
   ([250]Fig. 7, N to P). Together, we preclinically validated a
   generalizable association of TIM3^+VISTA^+ TAMs with immunoresistant or
   low neoantigenic tumors and the ability of ICD-inducing therapy and
   TIM3/VISTA blockade to specifically overcome these tumors ([251]Fig.
   8).

Fig. 8. Study overview.

   [252]Fig. 8.
   [253]Open in a new tab

   Schematic overview of our study’s results and translational as well as
   clinical value (created with [254]BioRender.com).

DISCUSSION

   We characterized a human-to-mouse conserved, TAM subset, i.e.,
   TIM3^+VISTA^+ TAMs. This TAM niche was specific to nonimmunogenic
   tumors with low neoantigenic burden and resistance toward PD(L)1
   blockade ([255]Fig. 8). This enrichment was mediated by IL-4 signaling.
   However, just enrichment of these TAMs was not enough to
   therapeutically target them in vivo. This mandated engagement of these
   TAMs via ICD induced by chemotherapy ([256]Fig. 8). A “two-signal
   threshold model for immune inhibition” ([257]59) seemed to explain the
   engagement between ICD and TIM3^+VISTA^+ TAMs. ICD-driven autocrine
   type I IFNs provided the “activation signal,” while, simultaneously,
   the HMGB1 and VISTA exposed by dying/stressed cells provided the
   “inhibitory signal.” Consequently, HMGB1/VISTA-based engagement of
   TIM3/VISTA on TAMs imposed an inhibitory threshold on the ability of
   these TAMs to mount an ISG response to ICD-derived type I IFNs. This
   disrupted ICD’s ability to activate these TAMs ([258]Fig. 8). However,
   the blockade of TIM3/VISTA released this inhibition of ISGs and, hence,
   ICD-inducing PTX synergized with TIM3/VISTA blockade to reduce growth
   of immunoresistant tumors. Mechanistically, this combination caused
   repolarization of TIM3^+VISTA^+ TAMs toward a proinflammatory, ISG
   response^+ phenotype that orchestrated TAM-driven cytotoxicity against
   cancer cells via TRAIL signaling ([259]Fig. 8). Notably, the failure of
   TIM3/VISTA blockade alone or with PD1 coblockade in regressing murine
   tumors, irrespective of their immunogenic profiles, recapitulated a
   similar failure of these regimen in recent multicancer clinical trials
   for both TIM3 and VISTA ([260]9, [261]10).

   Our study reverse-translated human characteristics of TIM3^+VISTA^+
   TAMs extracted from large-scale single-cell mapping into representative
   murine tumors. Thereafter, using mechanistic and translational
   approaches, we uncovered a multimodal immunotherapeutic strategy to
   target these TAMs in vivo, thereby blunting immunoresistant tumors.
   Treatment of low (neo)antigenic and nonimmunogenic tumors is a major
   unmet clinical need since these tumors do not respond to currently
   approved ICBs ([262]23, [263]50, [264]60, [265]61). Thus, our study has
   major implications for providing an immunotherapeutic solution not only
   tailored to overcome immunoresistant tumors but also tailored to blunt
   their most dominant immunoresistant compartment, i.e., TAMs. In
   addition, we successfully “translated” the TIM3^+VISTA^+ TAM signature
   as pan-cancer biomarker with prognostic and immunotherapy-predictive
   utility in clinical cohorts.

   We also demonstrated that the TIM3^+VISTA^+ coexpression phenotype is
   specific to TAMs in both human and mouse and not applicable to other
   immune cells including various subsets of DCs or T cells. For TIM3,
   various studies have proposed that almost all of its myeloid protumor
   activity is driven by TIM3^+DCs [especially conventional DC 1(cDC1)],
   independent of TIM3^+TAMs, particularly in the settings of mammary
   carcinoma or MC38/MC38-OVA tumors ([266]18, [267]19, [268]62). In these
   studies, DC-specific deletion of TIM3 was essential for tumor control
   via CD8^+ T cell–driven cytotoxicity. We believe that we can exclude a
   role for this DC–CD8^+ T cell axis in our setting due to following
   observations: (i) Antitumor synergism between PTX and TIM3/VISTA
   blockade did not activate DC maturation or CD8^+ T cell cytotoxicity in
   vivo; (ii) HAVCR2^+VSIR^+ (human/mouse) or TIM3^+VISTA^+ (mouse)
   phenotype was not strongly observed in tumoral DCs and was negligible
   in cDC1 cells; (iii) a clodronate liposomes dosage that selectively
   depleted MΦs, but not DCs, completely disrupted synergism of PTX and
   TIM3/VISTA blockade; (iv) selective and conditional genetic codeletion
   of TIM3 and VISTA in CSF1R^+TAMs was sufficient to render the tumor
   susceptible to PTX monotherapy in vivo; and (v) CD8^+ T cells were
   dispensable for tumor control. Last, TIM3 was shown to directly
   suppress type I IFN production from DCs ([269]18, [270]62). However, in
   our setting, TIM3/VISTA primarily disrupted the MΦ-intrinsic ISG
   responses. Although this suggests lack of TIM3/VISTA^+DCs involvement
   in our low antigenic settings, we cannot exclude their role in other
   contexts with sufficient antigenicity or a proinflammatory milieu
   ([271]18, [272]62). Similarly, while association of VISTA with myeloid
   cells, including TAMs, has been acknowledged, this was still from the
   perspective of MΦ/DC-driven inhibition of CD8^+ T cells ([273]55,
   [274]63). Moreover, the mechanism of VISTA-driven TAM suppression has
   not been sufficiently demonstrated. Our study fills these gaps in
   current literature.

   In conclusion, we show VISTA and TIM3 as redundant suppressors of
   TAM-associated ISG responses and TAMs-autonomous antitumor cytotoxicity
   (independent of CD8^+ T cells), together with a cross-species map of
   HAVCR2^+VSIR^+TAM features. These are all observations that have not
   been reported in earlier studies. This offers exciting
   immunotherapy-tailoring opportunities for TAM-enriched tumors that are
   insufficiently susceptible to CD8^+ T cell–centric approaches.

MATERIALS AND METHODS

Experimental design

   The primary objective of this study was to determine the antitumor
   response and gain mechanistic insights into combination therapy of
   chemotherapy and immune checkpoint inhibitors in different mouse models
   consisting of immunogenic and nonimmunogenic tumors. Primary outcomes
   for in vivo therapy studies included tumor growth measurements and
   weight loss measurements. In vitro mechanistic studies used
   enzyme-linked immunosorbent assay (ELISA), IFN/ISG response via J774
   reporter MΦs, quantitative polymerase chain reaction (qPCR), Western
   blot analysis, and flow cytometry analysis. This study used wild-type
   LLC, MC38, LLC-OVA, YUMM1.7, YUMM1.7-OVA, and YUMMER1.7 tumors in
   C57BL/6J wild-type or ifnar1^−/− mice to evaluate and confirm the
   effects of combination therapy of chemotherapy (PTX and CDDP) and
   immune checkpoint inhibitors (anti-TIM3 and anti-VISTA). The role of T
   cells and MΦs was studied by depletion of CD8^+ T cells via anti-CD8
   antibody and by depletion of MΦs via clodronate liposomes carefully
   dosed so they only deplete MΦs. Because we specially wanted to see the
   effect of depletion of our identified MΦ subset, we used hemopoietic
   stem cells from LSL-Cas9/Csf1r:Cre-ERT mice where CRISPR-Cas9–mediated
   Vsir and Havcr2 knockout in CSF1R^+ cells (MΦs) via in vivo genome
   editing was done. For all experiments, the number of animals in each
   group and P values are reported in the figure legends. The reported
   replicates refer to biological replicates. Tumor-bearing mice were
   randomized into different treatment groups, and weight loss
   measurements and tumor measurements were monitored blindly.

Experimental models

Mouse models

   Wild-type C57BL/6J mice were obtained from the KU Leuven breeding
   facility. The Ifnar1^−/− mice (B6.129S2-Ifnar1tm1Agt/Mmjax) were a gift
   from the laboratory of R. Vandenbroucke (VIB-Ugent). MΦ-specific TIM3
   VISTA knockout mice were generated by intercrossing the
   tamoxifen-inducible Csf1r:Cre-ERT (a gift of J. W. Pollard from the
   University of Edinburgh, Scotland) with LoxP-STOP-LoxP Cas9 mice
   [B6J.129(B6N)- Gt(ROSA) 26Sortm1(CAG-Cas9*,-EGFP)Fezh/J], purchased
   from the Jackson Laboratory. Acute deletion of TIM3 VISTA in MΦs was
   obtained by daily intraperitoneal injection of tamoxifen (0.05 mg/g of
   body weight) for 5 days after bone marrow reconstitution. Two days
   after the tamoxifen cycle, tumor cells were injected. Control mice were
   treated with tamoxifen according to the same protocol. All subcutaneous
   tumor experiments were done using 7- to 12-week-old female/male mice,
   maintained in the conventional mouse facility.

Cell lines

   LLC, LLC-OVA, MC38, YUMM1.7, and YUMM1.7-OVA cancer cells were a gift
   from the laboratory of M. Mazzone (VIB–KU Leuven). Overexpression of
   OVA in LLC and YUMM1.7 cancer cells were generated as previously
   described ([275]64). YUMMER1.7D4 cancer cells were bought from Merck
   Millipore. J774-DUAL cells, a murine MΦ-like cell containing a nuclear
   factor κB (NF-κB)–secreted embryonic alkaline phosphatase (SEAP) and
   IFN regulatory factor (IRF)–lucia reporter, were bought from InvivoGen.
   CRISPR-Cas9 knockout LLC cancer cells for CEACAM1, HMGB1, Galectin-9,
   PSGL1, and VISTA were generated by the laboratory of D. Daelemans (Rega
   Institute–KU Leuven).

   All cell lines were cultured at 37°C under 5% CO[2] split when 90%
   confluency was reached through enzymatic dissociation (Trypsin). MC38,
   LLC, and LLC-OVA cells were maintained in Dulbecco’s modified Eagle’s
   medium (DMEM) containing 2 mM l-glutamine, sodium bicarbonate (3.7
   g/liter), glucose (4.5 g/liter), and 1.0 mM sodium pyruvate with 10%
   heat-inactivated fetal bovine serum (FBS; 30 min at 56°C), penicillin
   (100 U/ml), and streptomycin (100 μg/ml). YUMM1.7, YUMM1.7-OVA, and
   YUMMER1.7D4 were maintained in F12 DMEM containing 2 mM l-glutamine,
   sodium bicarbonate (3.7 g/liter), glucose (4.5 g/liter), and 1.0 mM
   sodium pyruvate with 10% heat-inactivated FBS (30 min at 56°C),
   penicillin (100 U/ml), and streptomycin (100 μg/ml). J774-DUAL cells
   were maintained according to the manufacturer’s protocol. All cell
   lines were tested for Mycoplasma every month.

CRISPR-Cas9–mediated gene knockout in the LLC cell line

   Knockouts of Ceacam1, Hmgb1, Lgals9, Selplg, and Vsir genes were
   generated using four different guides per gene [single gRNA (sgRNA)
   target sequences are in the Supplementary Materials]. Guide sequences
   were cloned into the pLentiCRISPRv2 plasmid according to the standard
   cloning protocol. For lentiviral particle production, human embryonic
   kidney (HEK) 293T cells (received from J. Moffat, Donnelly Centre,
   University of Toronto, Canada) were plated in 40-ml supplemented DMEM
   in T150 (TPP) flasks at 45% confluency and incubated overnight.
   Twenty-four hours later, the cells were cotransfected using X-TremeGENE
   9 with the pLentiCRISPR plasmids and the lentiviral packaging plasmids
   pMD2.G and psPAX2 to generate lentiviral particles coated with the
   vesicular stomatitis virus glycoprotein protein and incubated
   overnight. Twenty-four hours after transfection, the medium was changed
   to DMEM supplemented with serum-free bovine serum albumin (BSA) growth
   medium [DMEM + BSA (1.1 g/100 ml) and gentamicin (20 μg/ml)]. The
   supernatant containing lentiviral particles was harvested 72 hours
   after transfection and stored at −80°C. Cells were transduced with
   lentiviruses expressing a pool of the four sgRNAs and then selected
   with puromycin for 3 days. Knockout phenotype was confirmed by Western
   blotting.

Method details

Cell death induction

   For cell death induction, cells were seeded in the appropriate dish as
   such that they reached an 80% confluency the next day. Apoptosis was
   induced by adding 100 μM CDDP (Sigma-Aldrich, #p4394) or 100 μM PTX
   (Sigma-Aldrich, #T1912). In case of coincubation experiments, cells
   were scraped and collected after 24 hours of cell death induction,
   washed in PBS, counted, and cocultured.

Thiazolyl blue tetrazolium bromide assay

   LLC and MC38 cells were seeded in a 96-well plate at a density of 5000
   cells per well 24 hours before cell death induction. Cell survival was
   obtained using the thiazolyl blue tetrazolium bromide (MTT) reagent
   (Abcam, #ab197010). Absorbance was read on a microplate reader (BioTek)
   at 490 nm at 24 and 48 hours after cell death induction. For all
   values, fold change to UT cancer cells was taken and multiplied with
   100 to obtain a percentage.

J774 MΦ reporter assay

   J774-DUAL reporter cells were seeded in a 96-well plate at a density of
   30,000 cells per well in a 96-well plate. After 24 hours, cells were
   pretreated with isotype (10 μg/ml) or blocking antibodies for 4 hours.
   UT and treated cancer cells were cocultured with the J774-DUAL reporter
   cells in a 1:1 ratio. In the case of neutralizing nucleic acids, UT and
   treated cancer cells were preincubated for 1 hour before with 3350
   U/10^6 cancer cell DNase I (Invitrogen, #18047-019) and 3.350 mg/10^6
   cancer cell RNase A (Thermo Fisher Scientific, #EN0531). Thereafter,
   cancer cells were cocultured with the J774-DUAL reporter cells in a 1:1
   ratio. To measure the IFN (IRF)/ISG response after 48 hours, another 80
   μl of medium was derived from the above cancer cell–J774 coculture in a
   white opaque-bottom 96-well plate. Here, 50 μl of Quanti-LUC substrate
   (Invivogen, #rep-qlc2) was added, and bioluminescence was directly
   measured with 100 ms of signal integration, with the microplate reader.
   All data derived from this reporter assay were minimum-maximum
   normalized for percentage scaling to account for interassay baseline
   variability.

Western blot

   For intracellular proteins, 800,000 cells were seeded in 10-cm dishes
   24 hours before cell death induction. Cells were scraped and collected
   at 6, 24, or 48 hours after cell death induction. Cells were
   centrifuged at 1500 rpm for 5 min, and the pellet was resuspended in
   100 μl of NP-40 lysis buffer (Thermo Fisher Scientific, #FNN0021) with
   protease (Thermo Fisher Scientific, #A32953) and phosphatase inhibitors
   (Thermo Fisher Scientific, #A32957). For secreted proteins, cells were
   2,500,000 cells seeded in 15-cm dishes, and medium was collected 24
   hours after cell death induction. Floating cells were removed by
   centrifugation. The supernatant was concentrated using Amicon Ultra-15
   centrifuge filter units (Merck, #UFC901024). A bicinchoninic acid assay
   (BCA) for colorimetric quantification of total protein was done with a
   BCA protein assay kit (Thermo Fisher Scientific, #23227) and a protein
   mixture of 90 μg was loaded onto the gel. The following proteins were
   detected: caspase 9 and cleaved caspase 9 (clone C9), VISTA (D5L5T),
   PSGL-1 (clone HECA-452), Ceacam-1 (clone E-12), Galectin-9 (clone
   D9R4A), HMGB1 (clone EPR3507), and β-actin (clones AC-74 and AC-15).
   The primary antibodies were diluted in 5% BSA + Tris-buffered saline
   with 0.1% Tween 20 detergent (TBST) with a dilution factor of 1:1000.
   As secondary antibody, we used the anti-rabbit antibody labeled with
   horseradish peroxidase (HRP) (Cell Signaling Technology, #7074S),
   anti-mouse antibody labeled with HRP (Cell Signaling Technology,
   #7076S), or anti-rat antibody labeled with HRP (Cell Signaling
   Technology, #7077S) diluted in 5% BSA + TBST with a dilution factor of
   1:2000. The antibodies tethering to the proteins on the membranes were
   detected using the enhanced chemiluminescence substrate, and the
   resulting precipitate was detected with the ChemiDoc MP imaging system.
   The precision plus protein dual color standards (Bio-Rad, #161037) was
   used as a ladder.

ATP secretion

   Medium of LLC and MC38 cells after cell death induction was harvested
   at 6, 24, and 48 hours. The presence of ATP in the medium was analyzed
   using the ATP assay system (Promega, #FF2000). A fold change to UT
   cells was taken to obtain ATP release.

Enzyme-linked immunosorbent assay

   BMDM cocultures with LLC (UT and treated) cancer cells were made as
   described above and used to perform an IFN-β ELISA (Invivogen
   #luex-mifnbv2). The IFN-β ELISA was performed according to the
   manufacturer’s protocol.

   Subcutaneous wild-type LLC/MC38 or LLC-OVA tumors were treated with PTX
   or PBS in combination with anti–TIM3 antibody (α-TIM3) or anti–VISTA
   antibody (α-VISTA) as described in the “Mouse experiments” section.
   Tumors were taken on day 24 after tumor cell injection, and whole-tumor
   lysate was made IL-4 ELISA (BioLegend, #431101) and IL-13 ELISA (Thermo
   Fisher Scientific, #431101) was done with whole-tumor lysate according
   to the manufacturer’s protocol. Absolute values were used to take a
   ratio with the blank. Thereafter, a ratio of IL-4/IL-13 to total TAMs
   was taken.

Flow cytometry–based detection of calreticulin, CD47, and HSP90

   A total of 200,000 LLC and MC38 cells were plated in 12-well plates in
   1 ml of DMEM and incubated for 24 hours. Cell death was induced for 48
   hours. Then, the cells were collected, washed with PBS, and transferred
   to 5 ml of flow cytometry tubes. The cells were resuspended in 50 μl of
   fluorescence-activated cell sorting (FACS) buffer (0.5% BSA and PBS
   solution) and 1:100 anti-calreticulin primary antibody (clone B44) or
   1:100 anti-HSP90 primary antibody (clone AC88). After 30-min incubation
   in the dark on ice, cells were washed with 1 ml of FACS buffer and
   centrifuged, and the supernatant was discarded. Then, cells were again
   resuspended in 50 μl of FACS buffer with 1:500 of goat anti-rabbit
   Alexa Fluor 488 (polyclonal) for calreticulin or 1:500 goat anti-Mouse
   secondary antibody for HSP90 (polyclonal), anti-CD47 (clone miap301)
   antibody, and Fixable Viability Dye eFluor 780. After 30-min incubation
   on ice in the dark, cells were fixed with Cytofix (BD Biosciences,
   #554655).

Flow cytometry

   Before staining procedure, Fc receptors of all samples were blocked
   using TruStain FcX (BioLegend, #101320) for 15 min. Cells were further
   stained with the indicated antibodies listed in table S2, diluted in
   0.5% BSA for 1 hour, and fixed with Cytofix (BD Biosciences, #554655).
   In case of intracellular markers, cells were fixed and permeabilized
   with the Cytofix/Cytoperm Kit (BD Biosciences, #554714). For the
   staining of transcription factors, the true nuclear transcription
   factor buffer (BioLegend, #42441) set was used. After fixation, cells
   were maintained in 0.5% BSA. For intracellular cytokine staining, the
   cells were stimulated with Dynabeads Mouse T activator CD3/CD28 (Thermo
   Fisher Scientific, #11456D). After 1 hour at 37°C and 5% CO[2], 2 μl of
   brefeldin A (Thermo Fisher Scientific, #00-4506-51) was added. Cells
   were then placed at 37°C and 5% CO[2] for 4 hours, transferred to 4°C
   overnight, and then stained for intracellular cytokines. Flow cytometry
   was performed on the Attune NxT (Thermo Fisher Scientific), FACSCanto
   (BD Biosciences), or the ID7000 (SONY). Cell doublets were excluded on
   the basis of forward scatter area (FSC-A)/forward scatter height
   (FSC-H). Flow cytometry data were analyzed using FlowJo. Gating
   strategies and unstained and fluorescent minus one controls are shown
   in the Supplementary Materials.

Murine splenocyte isolation

   Splenocytes were obtained from the spleen from wild-type C57BL/6J mice.
   Spleens were minced and filtered through a 70-μm cell strainer. Cells
   were incubated in red blood cell lysis buffer for 5 min and
   centrifuged.

Murine tumor-infiltrating leukocytes and TAM isolation

   Tumors were isolated at day 24 after tumor injection. A single-cell
   suspension was made, using the tumor dissociation kit (Miltenyi,
   #130-096-730). Tumor-infiltrating leukocytes (TILs) were isolated
   through magnetic bead separation via CD45 (Miltenyi, #130-110-618). For
   TAM isolation, anti-F4/80 microbeads (Miltenyi, #130-110-443) were
   used. Isolated TILs and TAMs were maintained in RPMI 1640 supplemented
   with penicillin (100 μg/ml), streptomycin (100 μg/liter), 2.5% Hepes
   (pH 7.5), and 10% heat-inactivated FBS. The following
   fluorochrome-conjugated antibody clones were used: FOXP3 (clone MF-14),
   GATA3 (clone TWAJ), CD62L (clone MEL-14), TCF1/7 (clone S33-966),
   CD107a (clone 1D4B), TOX (clone REA473), Tbet (clone O4-46), CD3 (clone
   17A2), CD8a (clone 53-6.7), Ki67 (clone 11F6), CD45 (clone 30-F11),
   Eomes (clone Dan11mag), CD127 (clone SB/199), PD1 (clone RMP1-30), CD4
   (clone GK1.5), TIM3 (clone 5D12/TIM-3), IL-2 (clone JES6-5H4), TNFα
   (clone MP6-XT22), IFN-γ (clone XMG1.2), PDL1 (clone 10F.9G2), PD1
   (clone 29F.1A12), Siglec H (clone 551), CD11c (clone N418), XCR1 (clone
   ZET), CD172A (clone P84), MHCII (clone M5/114.15.2), PD-L1 (clone
   10F.9G2), PD1 (clone RMP1-30), CD86 (clone GL1), F4/80 (clone
   T45-2342), CSF1R (clone AFS98), CD11b (clone ICRF44), TIM3 (clone
   B8.2C12), VISTA (clone MH5A), and CD206 (clone C068C2).

Cancer cell–TAM cocultures

   LLC cancer cells were cocultured with LLC tumor–derived TAMs
   (preincubated with isotype blocking antibodies or anti-TIM3/anti-VISTA
   blocking antibodies) in a 1:5 ratio. After 24, 48, and 72 hours,
   cocultures were scraped to collect the cells, centrifuged, and washed
   with PBS. Single-cell suspension was stained with fluorescently labeled
   antibodies diluted in FACS buffer (0.5% BSA and PBS solution) for 1
   hour on ice and then washed twice with staining buffer (BioLegend,
   #420201). Thereafter, cells were washed once in 1× annexin V binding
   buffer (BioLegend, #422201) and stained with fluorescently labeled
   annexin V antibody diluted in 1× annexin V binding buffer. After 15 min
   of staining at RT, samples were washed once, then resuspended with 1×
   annexin V binding buffer, and analyzed immediately at the flow
   cytometer. The following fluorochrome-conjugated antibody clones were
   used before annexin V staining: F4/80 (clone T45-2342), CD11b (clone
   ICRF44), Fixable Viability Dye eFluor 780, and annexin V.

Quantitative polymerase chain reaction

   The RNA of cancer cell–TAM cocultures was extracted using the PureLink
   RNA Mini Kit (Thermo Fisher Scientific, #12183025). Using the
   QuantiTect Reverse Transcription kit (QIAGEN, #205313), cDNA was
   synthesized from RNA. The qPCR was performed on the StepOnePlus
   Real-Time PCR system (Applied Biosystems) using SYBRgreen (HighQu,
   #QPD0150) with the following primers: Irf7, Rsad2, Mx1, Cxcl10, and
   Actin (all primers were ordered at Integrated DNA Technologies,
   sequences available inthe Supplementary Materials). Fold change was
   determined using the 2^∆∆CT method compared to the house keeping gene,
   Actin, and UT TAMs.

Antibody array

   At day 24, tumors were collected from PBS, PTX, PTX + TIM3, and PTX +
   VISTA mice. CD45^+ fractions were used to prepare tumor lysate with
   NP-40 lysis buffer, and protein concentration was measured. A protein
   concentration of 200 μg was used for a mouse cytokine array panel A
   (R&D Systems, #ARY006). Cytokine array was performed according to the
   manufacturer’s protocol. Arrays were read on the ChemiDoc (Bio-Rad).
   Dot intensities were determined using Image Lab (Bio-Rad). From all
   values, the background was subtracted.

Efferocytosis assay

   A total of 30,000 BMDMs were plated in white clear–bottom 96-well plate
   (Corning, #3610) per well. BMDMs were either pretreated with 20 μM
   cytochalasin D for 3 hours as a negative control or pretreated with
   isotype (10 μg/ml) or blocking antibodies for 4 hours. UT or dying
   cancer cells were collected 24 hours after cell death induction. Cells
   were stained with pHrodo Green (20 μg/ml; Thermo Fisher Scientific,
   #[276]P35373) for 1 hour and washed with FBS. Cells were added to BMDMs
   in a 1:1 ratio. Plates were kept in 37°C under 5% CO[2]. Fluorescence
   at an excitation of 490 nm and an emission of 520 nm were measured at
   24 and 48 hours after coincubation on a microplate reader (BioTek).
   Efferocytosis index was calculated with the following formula:
   (fluorescence intensity 37°C − fluorescence intensity cytochalasin D) ×
   (fluorescence intensity cancer cells 37°C / MTT percentage). MTT assay
   was performed as described above. In the case of efferocytosis with
   pHrodo Green E. coli BioParticles (Thermo Fisher Scientific,
   #[277]P35366), a concentration of 200 ug/ml was used.

TIM3^+VISTA^+ TAM and TIM3^−VISTA^− TAM cocultures with PTX

   Murine TILs were collected from LLC tumor–bearing mice as described
   above. Single-cell suspension was stained with F4/80 (clone T45-2342),
   Fixable Viability Aqua Zombi, TIM3 (clone B8.2C12), and VISTA (clone
   MH5A). Samples were sorted on the BD influx cell sorter to obtain
   TIM3^+VISTA^+ TAMs and TIM3^−VISTA^− TAMs (see gating strategy). From
   each sample, 100,000 TIM3^+VISTA^+ TAMs and TIM3^−VISTA^− TAMs were
   plated in a duplicate 12-well plate. The next day, 20,000 LLC cancer
   cells were added to all the TAMs and also plated alone. After 48 h of
   coculture, half of the samples were treated with 100 μM PTX, and the
   rest was left UT. Cocultures were scraped to collect after 48 hours of
   treatment, centrifuged, and washed with PBS. Single-cell suspension was
   stained with fluorescently labeled antibodies diluted in FACS buffer
   and analyzed immediately at the flow cytometer. The following
   fluorochrome-conjugated antibody clones were used: CD11b (clone
   ICRF44), Fixable Viability Aqua Zombi, and Ki67 (clone 11F6).

TAM cytotoxicity assay

   Murine TAMs were collected from LLC tumor–bearing mice treated with PBS
   or PTX as described above. A total of 30,000 TAMs were plated in a
   96-well plate per well. After 24 hours, TAMs were preincubated with
   either one of the following inhibitors for 4 hours: 100 μM apocynin,
   catalase (1000 U/ml), 100 μM L-NMMA, 10 μM anti-TNFα, 10 μM anti-FasL,
   or 10 μM of anti-TRAIL. After 4 hours, 30,000 UT LLC cancer cells were
   added to the TAMs with either anti-TIM3 (10 μg/ml) or anti-VISTA (10
   μg/ml) or no blocking antibody. After 48 hours of coculture, all the
   samples were stained with fluorescently labeled antibodies diluted in
   FACS buffer and analyzed immediately at the flow cytometer. The
   following fluorochrome-conjugated antibody clones were used: CD11b
   (clone ICRF44), Fixable Viability Aqua Zombi, and Ki67 (clone 11F6).

Murine BMDM generation

   Bone marrow was isolated from wild-type C57BL/6J or Ifnar1^−/− mice.
   Both the femur and tibia were flushed using PBS, and the cell
   suspension was centrifuged for 5 min at 1500 rpm. The pellet was
   resuspended in red blood cell lysis buffer (Merck Life Science),
   incubated for 5 min, and centrifuged. Cells were resuspended in RPMI
   1640 supplemented with penicillin (100 U/ml), streptomycin (100
   μg/liter), 2.5% Hepes (pH 7.5), and 10% heat-inactivated FBS. Bone
   marrow–derived cells were differentiated into MΦs by adding macrophage
   colony-stimulating factor (25 ng/ml; Peprotech, #315-02) into the
   medium for 6 days. Differentiation of MΦs was confirmed by flow
   cytometry; the following fluorochrome-conjugated antibody clones were
   used: CD11b (clone M1/70) and F4/80 (clone BM8).

In vitro and in vivo targeting of Havcr2 and Vsir

   To validate the knockout of TIM3 and VISTA, lentiCRISPRv2 vectors
   expressing the Cas9 along with a dual gRNA targeting Havcr2 (TIM3)
   locus (GTTACACTCTATCTACACCT) and Vsir (VISTA) (GGGTGATAGAGAAGTTACCG) or
   a nontargeting control gRNA (GAACAGTCGCGTTTGCGACT) were used. The
   lentiCRISPRv2 was a gift from F. Zhang (Addgene, plasmid no. 52961)
   ([278]65). The gRNAs were cloned as previously described by Shang et
   al. ([279]66). Lineage-negative HSPCs from LSL-Cas9/CSF1RcreERT
   transgenic mice were enriched with the mouse hematopoietic progenitor
   enrichment kit (STEMCELL Technologies). A total of 1 × 10^6 cells/ml
   were prestimulated for 5 hours with stem span serum-free medium
   (STEMCELL Technologies) supplemented with IL-3 (20 ng/ml), stem cell
   factor (SCF) (100 ng/ml), thrombopoietin (TPO) (100 ng/ml), and
   FMS-like tyrosine kinase 3 ligand (FLT-3L) (100 ng/ml) (Peprotech) and
   transduced with purified lentiviral vectors in the presence of
   polybrene for 48 hours. After adding virus, cells were centrifuged for
   1 hour at 1000g at RT. Cells were monitored over time, and a second
   spin infection was performed. Seven hours later, 1 × 10^6 cells were
   injected via tail vein in lethally irradiated C57BL/6-recipient mice. A
   fraction of transduced HSPCs were either collected for FACS analysis or
   differentiated into BMDMs.

Cell culture and transfection

   HEK293T were cultured in DMEM (Gibco) supplemented with 10%
   heat-inactivated FBS (Gibco), 2 mM l-glutamine (Life Technologies), and
   penicillin (100 U/ml)/streptomycin (100 μg/ml) (Life Technologies).
   Cells were maintained in a humidified incubator at 37°C and 5% CO[2].
   To generate lentiviral particles, HEK293T cells were seeded at 9 × 10^6
   cells in 20 ml in 15-cm dishes and transfected the following day. Two
   hours before transfection, medium was changed to Iscove’s modified
   Dulbecco’s medium supplemented with 10% FBS (Gibco), 2 mM l-glutamine
   (Life Technologies), and penicillin (100 U/ml)/streptomycin (100 μg/ml)
   (Life Technologies).

   Transfection was done by the calcium phosphate method with 32 μg of
   transfer vector carrying the respective sgRNAs for TIM3 and VISTA or
   the nontargeting gRNA as control. Each transfection also included 7 μg
   of a plasmid encoding PMD2.G and 16.25 μg of a plasmid encoding
   packaging proteins. Virus was collected 24 and 48 hours after
   transfection, filtered, and concentrated via ultracentrifugation.

In vivo experiments

Mouse experiments

   Seven- to 12-week-old female/male C57BL/6J mice or Ifnar1^−/− mice were
   subcutaneously injected with 1 × 10^6 LLC, LLC-OVA, MC38, YUMM1.7,
   YUMM1.7-OVA, YUMMER1.7, LLC HMGB1^−/−, or LLC VISTA^−/− cells. In the
   case of chemotherapy, mice were treated with CDDP (8 mg/kg) or PTX (8
   mg/kg) on days 10, 13, and 17 via intraperitoneal injections. When
   applicable, mice were treated or cotreated with 250 μg of anti-PDL1,
   anti-CTLA4, anti-CSF1R, anti-PD1, anti-TIM3, anti-VISTA, anti-CCL2,
   anti-SIRPα, anti-CR1/2, anti-TGFβ, anti-FCγRII/III, anti–IL-10, CD40
   agonism, or anti–IL-1β on days 11, 14, and 18 via intraperitoneal
   injections. For CD8 depletion experiments, mice were given 200 μg of
   anti-CD8, intraperitoneally 1 day before tumor inoculation and, from
   then on, every other day. For IL-4 depletion experiments, mice were
   given 500 μg of anti–IL-4, intraperitoneally 2 days before tumor
   inoculation and, from then on, every other day. For MΦ depletion
   experiments, 200 μl of clodronate liposomes was given 1 day before
   tumor inoculation and every 2/3 days subsequently. Mice were monitored
   and weighed every other day, and tumor volume was determined by height
   × width × length. Please see table S2 for further reference to the used
   clones.

Bone marrow transplantation

   To generate TIM3 VISTA bone marrow chimera, 7- to 8-week-old
   C57BL/6J-recipient mice were lethally irradiated with 9.2 Gy using the
   Small Animal Radiation Research Platform (XSTRAHL). Subsequently, 1 ×
   10^6 bone marrow cells from the appropriate genotype were injected
   intravenously via the tail vein. Tumor experiments were initiated 6
   weeks after bone marrow reconstitution. Red and white blood cell count
   was determined using a hematocytometer on peripheral blood, collected
   in heparin with capillary pipettes by retro-orbital bleeding.

Bioinformatics analyses

Transcriptomics analyses for murine subcutaneous tumors

   Affymetrix Mouse Exon 1.0 ST Array data associated with [280]GSE85509
   was downloaded from the Gene Expression Omnibus ([281]67). This study
   includes a range of subcutaneous tumors based on B16-F10, TC1, CT26,
   MC38, LL2/LLC, RENCA, 4 T1, TRAMPC1, EL4, P815, or PAN02 murine cancer
   cell lines implanted in syngeneic mice backgrounds ([282]37). Genetic
   signatures of proimmunogenic IFN-γ signaling versus MΦs were used to
   classify these tumors ([283]20, [284]21, [285]23–[286]27, [287]35,
   [288]37). For the differential analysis between LLC and MC38,
   corresponding CEL files were downloaded and normalized with the Robust
   Microarray Analysis (RMA) method, using oligo 1.60.0 ([289]68) in R
   4.2.0 ([290]69). Probe expression was subsequently summarized to the
   gene level using the associated Affymetrix NetAffx CSV file
   (MoEx-1_0-st-v1.na28.mm9.transcript.csv). Statistical significance of
   differences between LLC and MC38 was assessed with limma 3.52.2
   ([291]70), using t tests with empirical Bayesian shrinkage of the
   variance. P values were corrected for multiple testing following the
   Benjamini-Hochberg method. Results were represented as volcano plots. A
   select set of immune-relevant genes was highlighted with text labels.

Transcriptomic analysis in TCGA

   Bulk transcriptomics count data from the TCGA Toil-recompute project
   ([292]71) were downloaded from UCSC Xena ([293]72). Survival
   information was obtained from the same data hub. Immune
   classifications, delineating distinct immune subtypes in TCGA (C1 to
   C6; see Results for more details), were obtained from the supplementary
   files associated with the “The immune landscape of cancer” study
   ([294]73) (table S1), which lists calls for 11,081 samples.
   Precalculated immune deconvolutions for TCGA were downloaded from TIMER
   2.0 ([295]74), which contains a total of 11,071 samples. In this study,
   we only used the quanTIseq ([296]75) calls. The dataset was subset to
   only include primary tumors, recurrent tumors, metastatic tumors, and
   “additional new primary” or “additional metastatic” samples, resulting
   in a data matrix containing 9611 samples with RNA-seq, of which 8549
   contained survival information and MΦ quantification.

   Spearman correlations between a TIM3-VISTA metagene and the
   corresponding quanTIseq-deconvoluted MΦ population were calculated per
   immune subtype (C1 to C6) with SciPy 1.6.2 ([297]76) and visualized as
   radar plot, using Matplotlib 3.5.2 ([298]77). Samples for which no
   immune subtype or deconvoluted fraction was available were ignored.

Survival analysis in TCGA

   The TCGA dataset described above was used for a prognostic analysis, in
   which the effect of the expression of the TIM3-VISTA metagene (mean
   expression of all genes in signature) was associated with the OS, using
   a Cox regression as implemented in lifelines 0.26.3 ([299]78) with
   default settings. Overall, our analysis included 8516 patients with
   survival and metagene expression spanning 30 cancer types [Adenoid
   Cystic Carcinoma (ACC), 76; Bladder Urothelial Carcinoma (BLCA), 395;
   Breast Invasive Carcinoma (BRCA), 1085; Cervical Squamous Cell
   Carcinoma and Endocervical Adenocarcinoma (CESC), 301;
   Cholangiocarcinoma (CHOL), 35; Colon Adenocarcinoma (COAD), 280;
   Esophageal Carcinoma (ESCA), 171; Glioblastoma Multiforme (GBM), 157;
   Head and Neck Squamous Cell Carcinoma (HNSC), 513; Kidney Chromophobe
   (KICH), 65; Kidney Renal Clear Cell Carcinoma (KIRC), 513; Kidney Renal
   Papillary Cell Carcinoma (KIRP), 277; Liver Hepatocellular Carcinoma
   (LIHC), 361; Low Grade Glioma (LGG), 519; Lung Adenocarcinoma (LUAD),
   446; Lung Squamous Cell Carcinoma (LUSC), 447; Mesothelioma (MESO), 82;
   Ovarian Cancer (OV), 272; Pancreatic Adenocarcinoma (PAAD), 152;
   Pheochromocytoma and Paraganglioma (PCPG), 180; Prostate Adenocarcinoma
   (PRAD), 403; Rectum Adenocarcinoma (READ), 87; Sarcoma (SARC), 226;
   Skin Cutaneous Melanoma (SKCM), 104; Stomach Adenocarcinoma (STAD),
   383; Testicular Germ Cell Tumors (TGCT), 136; Thyroid Carcinoma (THCA),
   507; Uterine Corpus Endometrial Carcinoma (UCEC), 177; Uterine
   Carcinosarcoma (UCS), 57; Uveal Melanoma (UVM), 79]. We noticed an
   imbalance in SKCM toward the primary tumor samples (104 of 469 SKCM
   used) as primarily metastatic samples did not have a definite immune
   subtype assigned by Thorsson and co-workers ([300]73). However, this
   only affected a small proportion of the total pan cancer dataset. HRs
   were represented as a forest plot, using Matplotlib 3.5.2.

Analysis of scRNA-seq datasets

   Single-cell datasets covering patients with melanoma responding (or
   not) to immunotherapy (PD1/CTLA4 blockade) ([301]24), pan-immune
   dataset from four different cancers ([302]25), pan-myeloid dataset from
   eight different cancers ([303]26), and CRC dataset with patients with
   MSI and MSS CRC ([304]27) were accessed using a standardized and
   uniformized workflow of BBrowser_v_3 ([305]35), using built-in
   qualitative filters. The preexisting scRNA-seq dataset of subcutaneous
   LLC tumors ([306]38) was uploaded into the same workflow and further
   analyzed to maintain uniformity. This workflow was used for generating
   density plots or dot plots, as applicable. Cellular annotations either
   preexisted on the level of author-derived annotations or were based on
   automated annotations available within the above workflow. We used the
   HAVCR2^+VSIR^+TAMs within the CRC scRNA-seq dataset to perform DGE
   analyses and codependent pathway enrichment analyses (REACTOME for
   human and Gene Ontology Biological Process for mouse), using the Venice
   nonparametric analyses approach
   ([307]https://github.com/bioturing/signac). This was specifically done
   by comparing scRNA-seq profiles of HAVCR2^+VSIR^+TAMs (HAVCR2 and VSIR
   expression > 0) versus HAVCR2^_VSIR^_TAMs (HAVCR2 and VSIR expression =
   0) against each other. Notably, in some datasets, an alternative name
   for VSIR was C10orf54. The DGE analyses also allowed us to extract a
   signature for HAVCR2^+VSIR^+TAMs, which included all positively
   enriched genes in favor of this subset above the statistical cutoff of
   −log[10](P value) = 100. With this cutoff, we delineated the following
   signature for HAVCR2^+VSIR^+TAMs (some major TAM genes are indicated in
   bold): C10orf54, TYROBP, HAVCR2, B2M, GRN, LAPTM5, CD74, HLA-C, PSAP,
   NPC2, TMEM176B, VAMP8, HLA-A, HLA-DMB, HLA-DPA1, GPX1, CD68, HLA-DMA,
   TMSB4X, OAZ1, HLA-E, CFL1, ARPC1B, FCGRT, ITM2B, C1QC, PFN1, HLA-B,
   HLA-DPB1, HLA-DRA, CST3, ARHGDIB, CYBA, TMSB10, CTSB, HLA-DQB1, ITGB2,
   ACTG1, CAPG, ARPC3, S100A11, RPL15, CTSS, C1QB, ATP6V0B, CTSA, RPS2,
   YBX1, CTSC, PTMA, SRP14, AIF1, C1QA, PPT1, CTSH, CALM2, CSF1R, YWHAH,
   RPL28, CAPZB, ACTB, TYMP, CLIC1, ATP6V0E1, CTSD, RPL13A, HEXB, SERF2,
   EEF1A1, FCER1G, ACP5, TMEM176A, RPS19, PLD3, CORO1B, RPSA, RPL27A,
   PYCARD, SYNGR2, RPS3, BST2, NACA, PRDX1, LAIR1, PPIA, and ERP29. The
   full scRNA-seq profile of HAVCR2^+VSIR^+TAMs (HAVCR2 and VSIR
   expression > 0) was used to drive the automated Cell Ontology analyses
   (i.e., Cell Search algorithm) to interrogate the 300 scRNA-seq datasets
   within the curated BioTuring database as available on 17 February 2021
   ([308]35). The hits were filtered at a Jaccard’s index threshold of
   0.7. This analyses delineated the following dataset hits for MΦ node:
   basal cell carcinoma ([309]GSE123814), CRC ([310]GSE144735), liver
   cancer ([311]GSE125449), lung (COVID-19; [312]GSE145926), lung cancer
   (E-MTAB-6149), lung cancer ([313]GSE140819), pancreatic cancer [Genome
   Sequence Archive (GSA) GSA: CRA001160], and testis (normal;
   [314]GSE134144); and the following datasets are for monocyte node:
   fetal liver (E-MTAB-7407), lung (PRJEB31843), lung fibrosis
   ([315]GSE128033), lung fibrosis ([316]GSE121611), lung fibrosis
   ([317]GSE135893), lung cancer ([318]GSE131907), Scleroderma-associated
   interstitial lung disease (SSc-ILD) ([319]GSE128169), melanoma
   ([320]GSE123139), nasopharyngeal cancer ([321]GSE150430), peripheral
   blood mononuclear cells (sepsis; SCP548), peripheral blood mononuclear
   cells (SC2018), respiratory tract (EGAS00001004082), and spleen
   (PRJEB31843).

Immuno-oncology clinical trials analyses

   Patient survival analyses in immuno-oncology clinical trials was
   carried out using the above HAVCR2^+VSIR^+TAMs signature, i.e.,
   C10orf54, TYROBP, HAVCR2, B2M, GRN, LAPTM5, CD74, HLA-C, PSAP, NPC2,
   TMEM176B, VAMP8, HLA-A, HLA-DMB, HLA-DPA1, GPX1, CD68, HLA-DMA, TMSB4X,
   OAZ1, HLA-E, CFL1, ARPC1B, FCGRT, ITM2B, C1QC, PFN1, HLA-B, HLA-DPB1,
   HLA-DRA, CST3, ARHGDIB, CYBA, TMSB10, CTSB, HLA-DQB1, ITGB2, ACTG1,
   CAPG, ARPC3, S100A11, RPL15, CTSS, C1QB, ATP6V0B, CTSA, RPS2, YBX1,
   CTSC, PTMA, SRP14, AIF1, C1QA, PPT1, CTSH, CALM2, CSF1R, YWHAH, RPL28,
   CAPZB, ACTB, TYMP, CLIC1, ATP6V0E1, CTSD, RPL13A, HEXB, SERF2, EEF1A1,
   FCER1G, ACP5, TMEM176A, RPS19, PLD3, CORO1B, RPSA, RPL27A, PYCARD,
   SYNGR2, RPS3, BST2, NACA, PRDX1, LAIR1, PPIA, and ERP29. We accessed
   tumor transcriptomic data from patients with cancer with
   pre-/on-treatment samples profiled relative to anti-PD1, anti–PD-L1, or
   anti-CTLA4 immunotherapy using an existing computational workflow
   [KMPlot ([322]79, [323]80)] along with OS or PFS values for these
   patients (follow-up threshold = 36 months and survival time = days).
   These data originated from the following cohorts for OS: GIDE2019,
   [324]GSE121810, [325]GSE136961, [326]GSE140901, [327]GSE165252,
   [328]GSE165278, [329]GSE176307, [330]GSE183924, [331]GSE78220,
   [332]GSE91061, LIU2019, MARIATHASAN2018, and VANALLEN2015. For PFS,
   these data originated from the following: GIDE2019, [333]GSE121810,
   [334]GSE136961, [335]GSE140901, [336]GSE165252, [337]GSE176307,
   [338]GSE183924, [339]GSE93157, LIU2019, and VANALLEN2015. Here, 100%
   coverage of the above TAM signature was prioritized over accommodation
   of all patients from within these cohorts, thereby resulting in the
   analyses of 129 patients for PFS and 152 patients for OS. This is
   because not all cohorts captured all the genes relevant for our
   signature due to differences in profiling and processing methods across
   the studies. Final analyses entailed creation of Kaplan-Meier curves,
   at autoselect best cutoff (13,505.31 gene expression value, within the
   range of 3406 to 42,051 for OS, and 12,834.6 gene expression value,
   within the range of 3406 to 29,188 for PFS) with univariate Cox
   regression and log-rank P value analyses.

Flow cytometry analysis via Pytometry

   Our existing FACS data was reanalyzed through Pytometry 0.1.3. Values
   were arcsinh transformed and gated in accordance with our analysis via
   FlowJo (see the Supplementary Materials for gating strategies) to
   select subtypes. The principal components analysis space was built with
   five components, while the number of closest neighbors was defined as
   15. Neighbor calculation and Leiden clustering (Leidenalg 0.9.1) was
   performed with ScanPy 1.9.3 and projected as Uniform Manifold
   Approximation and Projection (UMAP) with Umap-learn 0.5.3. Kernel
   Density Estimate (KDE) plots were generated using Seaborn 0.12.2 and
   Matplotlib 3.7.1.

Correlation analysis

   Single-cell transcriptomic profiles of 2,489,382 MΦs were integrated
   and annotated from 1679 scRNA-seq studies of human origin (pan-disease
   or pan-tissue) and analyzed using the Talk2Data Database v.4.0.
   Talk2Data’s coexpression function was used to calculate a dotplot
   (scale mode: relative) of the top most correlated (Jaccard’s index)
   genes among HAVCR2, VSIR, SPP1, MRC1, ARG1, CD163, CSF1R, TREM2, TEK,
   IL10, TIMP1, PDGFB, TGIF1, ITGAX, IRF3, STAT1, IL1B, FOLR2, LYVE1, TNF,
   CCL2, MMP2, MMP9, IL6, CXCL10, and CR1 in 2,489,382 cells in MΦs. The
   Jaccard’s index between two genes g1 and g2 is calculated as follows:
   Jaccard’s index (g1, g2) = (number of cells that express both g1 and
   g2) / (number of cells that express at least one of the two genes).

Statistical analysis

   The statistical details of all the analyses are reported in the figure
   legends, figures, and/or Materials and Methods, including statistical
   analysis performed, statistical significance thresholds/values, and, in
   most cases, the counts/number of data points. All statistical analyses
   were performed using GraphPad Prism software. Statistical significance
   was calculated by two-tailed unpaired Mann-Whitney test on two
   experimental conditions and Kruskal-Wallis test with multiple
   comparisons correction for false discovery rate (FDR) with two-stage
   step-up method of Benjamini, Krieger, and Yekutieli when more than two
   experimental groups were compared. Multiple comparison was always done
   to PBS/UT or to the indicated controls in the graphs. Gene signatures
   were estimated by considering the average expression of all the genes
   within that signature, unless otherwise mentioned. Details about used
   software for analysis can be found in the Supplementary Materials.

Study approval

   Mouse experiments were approved by the animal ethics committee at KU
   Leuven (projects P114/2019 and p195/2020) following the European
   directive 2010/63/EU as amended by the Regulation (European Union)
   2019/1010 and the Flemish government decree of 17 February 2017.

Acknowledgments