Immune surveillance of brain metastatic cancer cells is mediated by IFITM1

Abstract

Brain metastasis, most commonly originating from lung cancer, increases cancer morbidity and mortality. Although metastatic colonization is the rate‐limiting and most complex step of the metastatic cascade, the underlying mechanisms are poorly understood. Here, in vivo genome‐wide CRISPR‐Cas9 screening revealed that loss of interferon‐induced transmembrane protein 1 (IFITM1) promotes brain colonization of human lung cancer cells. Incipient brain metastatic cancer cells with high expression of IFITM1 secrete microglia‐activating complement component 3 and enhance the cytolytic activity of CD8⁺ T cells by increasing the expression and membrane localization of major histocompatibility complex class I. After activation, microglia (of the innate immune system) and cytotoxic CD8⁺ T lymphocytes (of the adaptive immune system) were found to jointly eliminate cancer cells by releasing interferon‐gamma and inducing phagocytosis and T‐cell‐mediated killing. In human cancer clinical trials, immune checkpoint blockade therapy response was significantly correlated with IFITM1 expression, and IFITM1 enhanced the brain metastasis suppression efficacy of PD‐1 blockade in mice. Our results exemplify a novel mechanism through which metastatic cancer cells overcome the innate and adaptive immune responses to colonize the brain, and suggest that a combination therapy increasing IFITM1 expression in metastatic cells with PD‐1 blockade may be a promising strategy to reduce metastasis.

Interferon‐induced transmembrane protein 1 reduces brain metastasis of lung cancer cells by activating both innate and adaptive immune responses.

Introduction

Brain metastases, or secondary brain cancers, have caused more and more cancer morbidity and mortality because of recent progress in cancer diagnosis and prolonged survival after cancer treatment (Fidler, 2015; Valiente et al, 2018; Achrol et al, 2019; Zeng et al, 2019; Soffietti et al, 2020). In adults, 40–50% of brain metastases originate from lung cancer, with the next highest percentage arising from breast cancer (15–20%), melanoma (5–10%), and colon cancer (5–10%; Sundermeyer et al, 2005; Fidler, 2015; Goncalves et al, 2016; McArthur et al, 2017; Achrol et al, 2019; Zeng et al, 2019). Non‐small‐cell lung cancer is the most common cancer to metastasize to brain, and lung cancer patients with brain metastases have a median survival of 6 months (Steeg et al, 2011; Valiente et al, 2014, 2018; Fidler, 2015; Chen et al, 2016; Quail & Joyce, 2017; Achrol et al, 2019). Many studies have suggested that most cancer cells that cross the blood–brain barrier (BBB) die, and brain colonization by cancer cells is the most complex and rate‐limiting step of the metastatic cascade (Sevenich et al, 2014; Valiente et al, 2014; Chen et al, 2016; Priego et al, 2018). Metastasis‐initiating cells, which eventually adapt to the high selection pressure imposed by the brain microenvironment, successfully colonize the brain and produce clinical metastases. Although significant progress has been made in studying metastatic colonization (Kienast et al, 2010; Steeg et al, 2011; Giancotti, 2013; Quail & Joyce, 2013, 2017; Sevenich et al, 2014; Valiente et al, 2014; Wculek & Malanchi, 2015; Chen et al, 2016; Malladi et al, 2016; Sanger Mouse Genetics Project et al, 2017; Priego et al, 2018; Rodrigues et al, 2019; Zeng et al, 2019; Klein, 2020; Shih et al, 2020), colonization of cancer cells in the brain parenchyma and interactions between brain metastatic cancer cells and the brain microenvironment, especially the brain immune microenvironment, are still poorly understood.

The brain has a unique immune microenvironment, which is maintained by the BBB (Valiente et al, 2014; Fidler, 2015; Achrol et al, 2019). However, brain metastases disrupt the BBB, and the consequent blood–tumor barrier (BTB) is more permeable (Morikawa et al, 2015; Valiente et al, 2018; Achrol et al, 2019). The BTB and the brain tissue surrounding tumor jointly regulate lymphocyte recruitment to brain metastatic microenvironment (Joyce & Pollard, 2009; Taggart et al, 2018; Soffietti et al, 2020). Indeed, brain metastases were well infiltrated by CD8⁺ T cells, and accumulation of tumor‐infiltrating T lymphocytes (TILs) in brain metastases was found to positively correlated with the high abundance of microglia present (Friebel et al, 2020; Klemm et al, 2020; Croft et al, 2021). Accumulating evidence shows that dynamic interplay between brain metastatic cancer cells and the surrounding immune microenvironment is crucial to brain metastatic colonization (Quail & Joyce, 2017; Schulz et al, 2020; Klemm et al, 2021). Moreover, the immune microenvironment in brain metastases also mediates treatment response to immunotherapy directed against brain metastases (Soffietti et al, 2020; Niesel et al, 2021; Ott et al, 2021). Therefore, a deeper understanding of the immune microenvironment in brain metastases has recently been increasingly recognized as a resource to improve immunotherapy of brain metastases.

Microglia serve as the only resident immune cells of healthy brain parenchyma and are known to be important for immune response within brain (Steeg et al, 2011; Fidler, 2015; Achrol et al, 2019; Schulz et al, 2020). Microglia of the innate immune system have the ability to detect and kill cancer cells and to protect brain against metastatic colonization (Louie et al, 2013; Fidler, 2015; Achrol et al, 2019; Ott et al, 2021). In the brain metastatic microenvironment, microglia also function in antigen presentation (Quail & Joyce, 2017; Ott et al, 2021). Previous studies showed that activated microglia robustly expressed major histocompatibility complex (MHC) and costimulatory molecules, which are involved in antigen presentation and the activation of the adaptive immune system, implying that the brain metastatic microenvironment is able to regulate the adaptive immune response (Yang et al, 2010; Schulz et al, 2020; Klemm et al, 2021; Niesel et al, 2021). Although microglia play multiple roles in brain metastases, there have been a limited number of studies into its function in brain metastatic colonization, especially in microglia–brain metastatic cells communication in this process, because of a difficulty in distinguishing microglia from peripheral macrophages by immune staining and a shortage of appropriate mouse models (Bowman et al, 2016; Quail & Joyce, 2017; Ott et al, 2021). Recently, Wnt/β‐catenin, SDF1/CXCR4, and CCL2/CCR2 signaling pathways were reported to be important for microglia–brain metastatic cells communication (Pukrop et al, 2010; You et al, 2019). Despite these findings, the underlying mechanisms of microglia–brain metastatic cells communication in brain metastatic colonization remain largely unknown, and a better understanding of the mechanisms is critical for targeted therapy and immunotherapy for brain metastases.

As is widely known, brain metastatic tumors are infiltrated by CD8⁺ T cells, which are the most potent mediator of the anti‐tumor immune response in brain metastases and also the main component of T cell dysfunction (Ott et al, 2021). Brain metastasis‐infiltrating CD8⁺ T cells are independent of tumor stage and positively correlated with increased microglial density (Friebel et al, 2020; Klemm et al, 2020; Croft et al, 2021). Importantly, high density of CD8⁺ T cells in brain metastases is associated with improved survival prognosis in lung cancer, breast cancer, colorectal cancer, and melanoma (Joyce & Pollard, 2009; Valiente et al, 2018; Achrol et al, 2019; Soffietti et al, 2020). Interestingly, the majority of brain metastasis‐infiltrating CD8⁺ T cells expressed the immunoinhibitory receptor PD‐1, which mediated inhibitory signaling in CD8⁺ T cell exhaustion (Niesel et al, 2021; Ott et al, 2021).

At present, the standard of care for brain metastases is medication, surgery, radiotherapy, chemotherapy, and some combinations of these. Recently, Immune checkpoint blockade (ICB) therapies targeting surface proteins PD‐1 and CTLA4 and Anti‐PD‐L1 have shown impressive clinical efficacy in a subset of patients with lung cancer and melanoma brain metastases through potentiating the anti‐tumor immune response of brain metastasis‐infiltrating CD8⁺ T cells (Taggart et al, 2018; Valiente et al, 2018; Achrol et al, 2019; Soffietti et al, 2020). For instance, 33% of the NSCLC patients and 22% of the melanoma patients benefited from the pembrolizumab treatment (a fully humanized anti‐PD‐1 monoclonal antibody blocking the interaction between PD‐1 and its ligand PD‐L1), and combined ICB therapy (nivolumab, an anti‐PD‐1 antibody, and ipilimumab, an anti‐CTLA4 antibody) achieved an overall response rate of ~ 50% in the melanoma patients (Gao et al, 2016b; Achrol et al, 2019). However, ~ 50% of treated cancer patients still do not respond to ICB therapy, and there is a hope that an extensive understanding of the tumor immune microenvironment would be translated into more personalized and effective immunotherapies (Achrol et al, 2019; Ott et al, 2021). Therefore, there is an urgent need to study underlying mechanisms and identify combination treatments with ICB therapy to increase the response rates.

Here, we performed in vivo pooled genome‐wide loss‐of‐function genetic screens on a panel of human lung cancer cells and identified IFITM1 as a commonly found repressor of brain metastatic colonization. In addition, we analyzed the expression of IFITM1 in clinically annotated tissue microarrays (TMAs) and publicly available microarray databases and found that low expression of IFITM1 in tumor tissues predicts poor survival of lung cancer patients. We also uncovered novel and essential roles for IFITM1 in both innate immune recognition (via upregulation of complement component 3) and adaptive immune recognition (via increases in the expression and membrane localization of major histocompatibility complex class I) and elimination of potentially metastatic lung cancer cells in the brain parenchyma. Moreover, the expression of complement C3 mRNA was positively correlated with IFITM1 mRNA expression in human lung cancer samples, and high expression of IFITM1 and MHC‐I prolonged the survival of lung cancer patients.

Finally, we showed that combination therapy with an IFITM1‐overexpressing oncolytic virus and immune checkpoint blockade (ICB) may block brain metastasis in animal experiments, and we found that the ICB therapy response was significantly positively correlated with IFITM1 expression in human cancer clinical trials (Ribas et al, 2017; Bommareddy et al, 2018; Twumasi‐Boateng et al, 2018).

Results

In vivo functional genome‐wide CRISPR‐Cas9 knockout screens identify IFITM1 as a commonly found repressor of brain metastasis

To identify metastasis suppressor genes in metastatic different organ colonization on a large scale, we applied genome‐wide CRISPR knockout libraries (GeCKO v2; Sanjana et al, 2014) to the functional genetic screening platform (Fig 1A; Gao et al, 2012, 2014, 2016a). Six large‐scale screens were performed using six non‐small‐cell lung cancer cell lines harboring the N‐Ras Q61K/p53del mutation (H1299), epidermal growth factor receptor activating mutation (EGFR Δexon19, H1650, PC9, and HCC827), K‐Ras G12S mutation (A549), or K‐Ras G12C mutation (H2030). In five of the six screened cell lines (except for H2030), sgRNA targeting interferon‐induced transmembrane protein 1 (IFITM1) was recovered from four brain metastases arising in four mice (two brain metastases of H1299 cells and two brain metastases of PC9), and four nonbrain metastases of A549 cells, H1650 cells, and HCC827 cells (Fig 1B and Appendix Tables S1–S4). Although sgRNA targeting IFITM1 was also identified from bone metastatic lesions (Fig 1B), this study focused only on lung cancer brain colonization. IFITM1 knockout was confirmed at both the protein and DNA levels (Fig 1C–G and Appendix Fig S1A). These results suggest that IFITM1 is a general mediator of lung cancer metastasis to the brain, functioning independently of the specific oncogenic background.

Figure 1. In vivo CRISPR screening reveals that IFITM1 loss causes lung cancer brain colonization.

• A
  Schematic of the loss‐of‐function screening strategy based on CRISPR‐Cas9 knockout technology to identify brain metastasis suppressor genes.
• B
  Venn diagram showing the overlap among candidate genes identified in five independent screens.
• C–G
  Representative screening results. Representative images of BALB/c nude mice xenografted with control H1299 lung cancer cells or H1299 lung cancer cells transduced with the human GeCKO v2 library (pool 1 containing sgIFITM1; 1 × 10⁶) (C). The arrow indicates a brain lesion (visible to the naked eye) formed by H1299 cells from the pool 1 mouse (D). Images of metastatic H1299 cells with tdTomato and luciferase reporter expression isolated from the brain lesion (E). Expression of IFITM1 in these metastatic H1299 cells (F) (three independent experiments). Frequency of indel mutations in these metastatic H1299 cells; the wild‐type sequence and genomic position are shown at the top (G).
• H, I
  Bioluminescence imaging and quantification of brain metastases (H) and overall survival (I) of BALB/c nude mice xenografted with control or IFITM1‐knockout H1299 cells (populations of knockout cells) by intracardiac injection (1 × 10⁶ cells). #, Experimental endpoint (I). (The n‐values denote the number of mice per group, and four independent western blot experiments were performed).
• J
  Bioluminescence imaging and quantification of brain metastases and overall survivals of BALB/c nude mice xenografted with control or IFITM1‐overexpressing H460 cells by intracardiac injection (1 × 10⁵ cells; The n‐values denote the number of mice per group, and four independent western blot experiments were performed).
• K
  Representative immunohistochemical images of IFITM1 expression in brain metastatic lesions (41, and 16 for the control, and IFITM1 groups, respectively) of BALB/c nude mice xenografted with control or IFITM1‐overexpressing H460 cells by intracardiac injection (1 × 10⁵ cells).
• L
  Staining and quantification of brain metastatic lesions formed by intracardiac injection of LLC‐BrM cells into C57BL/6 mice (1 × 10⁶ cells). Yellow dotted lines, brain metastatic lesions; white arrows, solitary tumor cells (one datapoint means one section: 17 sections for day 1 control, 17 sections for day 1 IFITM1, 16 sections for day 7 control, 17 sections for day 7 IFITM1, 12 sections for day 20 control, and 10 sections for day 20 IFITM1; day 1, 3 mice; day 7, 4 mice; and day 20, 3 mice).
• M
  Representative immunohistochemical images and staining intensities of IFITM1 expression in clinical lung cancer samples from 321 primary tumors (Primary) and 36 brain metastatic lesions (Br mets).
• N
  Kaplan–Meier survival curves of lung cancer patients in the TCGA dataset (n = 415) stratified by IFITM1 mRNA expression. HR, hazard ratio.

Data information: The data are presented as the mean ± s.e.m. values. P‐values were determined by unpaired two‐way ANOVA with uncorrected Fisher's LSD test (H, quantification of brain metastases in J, and L), the log‐rank test (I, survival in J, and N), or an unpaired two‐tailed Student's t‐test with Welch's correction (M). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Source data are available online for this figure.

IFITM1 has been suggested to localize at the cell membrane, with the N‐terminus located in the cytoplasm and the C‐terminus in the extracellular space (Appendix Fig S1B; Bailey et al, 2014; Weston et al, 2014), and to protect cells against viral infection (Brass et al, 2009; Bailey et al, 2014). Consistent with the screening results, targeting IFITM1 with two distinct sgRNAs enabled H1299 human lung cancer cells (populations of knockout cells) to metastasize to the brain (Fig 1H and Appendix Fig S1C); in addition, mice injected with IFITM1‐knockout cells had significantly shorter overall survival times than mice injected with control cells (Fig 1I). By contrast, overexpression of IFITM1 reduced the capacity of H460 human lung cancer cells (Fig 1J and K), brain metastatic Lewis lung carcinoma cells (LLC‐BrM; Fig 1L and Appendix Fig S1D), and mouse Lewis lung carcinoma (LLC) cells (Appendix Fig S1E; populations of IFITM1‐overexpressing cells) to metastasize to the brain in both immunodeficient mice and immunocompetent mice. Moreover, neither knockout nor overexpression of IFITM1 modified the ability of cancer cells to proliferate, invade, and express epithelial–mesenchymal transition (EMT) or stemness‐related genes, and form tumor spheres in vitro (Appendix Fig S1F–M). Moreover, overexpression of IFITM1 did not affect extravasation of lung cancer cells in the brain (Appendix Fig S1N).

Low expression of IFITM1 predicts poor survival of cancer patients

To extend our findings in cell lines to clinical samples, we analyzed IFITM1 expression in clinically annotated tissue microarrays (TMAs). The TMAs comprised samples from 321 primary lung tumors and 36 brain metastatic lesions. Although IFITM1 was highly expressed in primary tumors, brain metastatic lesions showed lower expression levels (Fig 1M and Appendix Fig S2A and B). Similar results were obtained in human lung cancer datasets (Appendix Fig S2C and D). Furthermore, low expression of IFITM1 mRNA in primary tumors correlated with poor overall survival in lung cancer patients in large datasets (Fig 1N and Appendix Fig S2E and F). However, expression of IFITM2 or IFITM3 mRNA did not correlate with overall survival (P > 0.05, not significant), and neither knockout nor overexpression of IFITM3 regulated brain metastasis and overall survival as effectively as did IFITM1 in vivo (Appendix Fig S2G and H).

Microglia but not macrophages exert anti‐brain metastasis effects through IFITM1

Since cancer cells were directly delivered into the left ventricle, and modulation of IFITM1 did not affect extravasation of lung cancer cells in the brain (Appendix Fig S1N), we sought to investigate IFITM1 function during metastatic brain colonization, which occurs after extravasation.

Immunofluorescence imaging of brain sections revealed cancer cells surrounded by microglia on day 1 after intracardiac injection but by very few peripheral macrophages (Fig 2A); however, IFITM1 overexpression had only a marginal effect on microglial recruitment (Fig 2A). Notably, treatment with PLX5622, BLZ945, or minocycline, specific inhibitors of microglial activation (Bellver‐Landete et al, 2019; Benbenishty et al, 2019; Wies Mancini et al, 2019; Willis et al, 2020), rescued the capacity of IFITM1‐overexpressing cancer cells to colonize the brain (Fig 2B, and Appendix Fig S3A and B). Moreover, the anti‐brain metastasis effect of microglia was mediated by IFITM1 (Appendix Fig S3C–E). We would like to point out that PLX5622 selectively induces the death of microglia in vivo, with minimal effects on peripheral macrophages at 1,200 PPM (Appendix Fig S3F–I; Bellver‐Landete et al, 2019; Guldner et al, 2020; Willis et al, 2020). Interestingly, peripheral macrophage depletion using clodronate liposomes (Appendix Fig S3J–L) had no effect on the capacity of IFITM1‐knockout or IFITM1‐overexpressing cancer cells to colonize the brain (Appendix Fig S3M and N). Our results are consistent with the findings from previous studies, which demonstrated that the identity and function of microglia are distinct from peripheral macrophages (Bowman et al, 2016).

Figure 2. IFITM1 induces the anti‐brain metastasis effect of microglia and activated microglia exert anti‐brain metastasis effects via phagocytosis and IFNγ secretion.

1. Representative immunofluorescence images and quantification of microglial (TMEM119⁺) or macrophage (TMEM119⁻/F4/80⁺) density per microscopic field in the brain on day 1 or 7 after the intracardiac injection of control or IFITM1‐overexpressing LLC‐BrM cells (tdTomato, 1 × 10⁶) into C57BL/6 mice (90, 90, and 120 fields for normal, day 1, and day 7, respectively; normal, 3 mice; day 1, 3 mice; and day 7, 4 mice). Normal brain tissue was used as the control.
2. Bioluminescence imaging and quantification of brain metastases on day 12 in BALB/c nude mice allografted with control or IFITM1‐overexpressing LLC‐BrM cells by intracardiac injection (1 × 10⁵ cells). PLX5622 (1,200 PPM) was administered in chow from day −3 to day 12 (The n‐values denote the number of mice per group).
3. Flow cytometric analysis of violet‐labeled LLC‐BrM cells phagocytosed by microglia in the brains of LLC‐BrM cell‐allografted BALB/c nude mice on day 1 after intracardiac injection (1 × 10⁶ control or IFITM1‐knockout cells). Representative flow cytometry pseudocolor density plots and the gating strategy are shown, and the bar graph shows the percentage of violet + microglia (The n‐values denote the number of mice per group).
4. Representative image and flow cytometric results of DDAO‐labeled LLC‐BrM cells (1 × 10⁵) phagocytosed by primary mouse microglia (1 × 10⁵, from 16 C57BL/6 mice), primary mouse BMDMs (1 × 10⁵, from 3 C57BL/6 mice), or primary mouse BrM‐BMDMs (1 × 10⁵, from 17 C57BL/6 mice) in contact coculture for 1 h. BMDMs, primary bone‐marrow‐derived macrophages. BrM‐BMDMs, primary bone‐marrow‐derived macrophages from the brain metastatic lesions of LLC‐BrM cell‐isografted C57BL/6 mice on day 20 after the intracardiac injection of LLC‐BrM cells (1 × 10⁵ cells; at least four independent experiments).
5. LLC‐BrM cells (tdTomato) were surrounded by activated microglia (Iba1⁺, green) and IFNγ (white) in the brain on day 1 or 7 after intracardiac injection of control or IFITM1‐overexpressing cells (1 × 10⁶) into C57BL/6 mice. IFNγ intensities in the fields enclosed in the yellow dotted lines (ROIs; 90, 90, and 120 fields for normal, day 1, and day 7, respectively) are shown. The ROI was set as the entire area of the metastatic lesion for day 7 and as the area of single metastatic cell for day 1. ImageJ was used to quantify the immunofluorescence intensity in the ROI. Normal brain tissue was used as the control. ROI, region of interest.
6. ELISA results of mouse IFNγ in brain homogenates at the indicated time points after intracardiac injection of LLC‐BrM cells (1 × 10⁶) into C57BL/6 mice (3 mice per time point).
7. Results of ELISA for mouse IFNγ in the brain on day 1 after the intracardiac injection of control or IFITM1‐knockout LLC‐BrM cells (1 × 10⁶) into BALB/c nude mice. PLX5622 (1,200 PPM) was administered in chow from day −3 to day 1. BLZ945 (200 mg/kg) was administered by oral gavage from day −2 to day 1. Minocycline (33 mg/kg) was administered intraperitoneally on days −1, 0, and 1. Clodronate liposomes (100 μl/10 g) were administered intraperitoneally on day −1. Normal brain tissue was used as the control (The n‐values denote the number of mice per group).
8. Flow cytometric apoptosis analysis of control and IFITM1‐knockout LLC‐BrM cells (6 × 10⁴) in contact coculture with primary mouse microglia (6 × 10⁴, from 16 C57BL/6 mice), primary mouse BMDMs (6 × 10⁴, from 3 C57BL/6 mice), or primary mouse BrM‐BMDMs (6 × 10⁴, from 15 C57BL/6 mice) for 6 h. BMDMs, primary bone marrow‐derived macrophages. BrM‐BMDMs, primary bone marrow‐derived macrophages from the brain metastatic lesions of LLC‐BrM cell‐isografted C57BL/6 mice on day 20 after the intracardiac injection of LLC‐BrM cells (1 × 10⁵ cells; at least three independent experiments).

Data information: The data are presented as the mean ± s.e.m. values. P‐values were determined by unpaired two‐way (A, D, E, G, and H) or one‐way (B, C, and F) ANOVA with uncorrected Fisher's LSD test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and n.s., not significant.

Source data are available online for this figure.

Activated microglia eliminate brain metastatic cancer cells by inducing phagocytosis and releasing IFNγ

To examine the effect of microglia on brain metastatic cancer cells, in vivo experiments and in vitro coculture systems were applied (Appendix Fig S4A–C). qPCR analysis of the expression of candidate genes indicative of microglial activation states suggested that IFITM1‐overexpressing cancer cells promoted the activation of microglia in vitro and in vivo, and IFITM1 knockout attenuated the activation of microglia in vitro (Appendix Fig S4D and E).

Activated microglia exert their antimetastatic effects via phagocytosis and secretion of cytotoxic factors, such as IFNγ, TNFα, and nitric oxide (Fidler, 2015; Wolf et al, 2017). IFITM1 knockout reduced the phagocytosis of cancer cells by microglia in vivo (Fig 2C); in contrast, IFITM1 overexpression enhanced the phagocytosis of cancer cells by primary microglia under in vitro contact coculture conditions at approximately 1 or 1.5 h but not the phagocytosis of cancer cells by primary bone‐marrow‐derived macrophages (BMDMs) or primary bone‐marrow‐derived macrophages from brain metastatic lesions (BrM‐BMDMs; Fig 2D and Appendix Fig S4F–I).

Furthermore, numerous inflammatory cytokines, including interferon‐gamma (IFNγ), were upregulated in the brain of the mice xenografted with cancer cells on day 1 after the intracardiac injection (Appendix Fig S4J). Intriguingly, the results of the mouse cytokine array analysis of conditioned medium from H460 human lung cancer cells and BV‐2 cells (immortalized murine microglia) cultured in a direct contact coculture system were similar to the in vivo results (Appendix Fig S4J and K), implying that microglia are the early effector cells to metastatic cancer cells in the brain, consistent with previous reports (Steeg et al, 2011; Sevenich et al, 2014; Valiente et al, 2014, 2018; Fidler, 2015; Quail & Joyce, 2017; Priego et al, 2018; Achrol et al, 2019). Immunofluorescence imaging (Fig 2E) and ELISAs (Fig 2F and Appendix Fig S4L–N) confirmed that only IFNγ accumulated in the brain on day 1 after intracardiac injection. In addition, IFITM1 knockout decreased the IFNγ level in the brain on day 1 after intracardiac injection (Fig 2G sgIFITM1‐1 or sgIFITM1‐2 vs. sgControl); however, this effect was abolished by PLX5622, BLZ945, or minocycline treatment, and macrophage depletion had no impact on this effect (Fig 2G). Overexpression of IFITM1 increased the IFNγ level in the brain on day 1 after the intracardiac injection (Fig 2E, day 1). Interestingly, blocking antibodies specific for IFNγ (Appendix Fig S4O), but not blocking antibodies specific for TNFα (Appendix Fig S4P) or treatment with an inducible nitric oxide synthase (iNOS) inhibitor (L‐NMMA), prevented the enhanced apoptosis of cancer cells induced by IFITM1 under noncontact coculture conditions (Appendix Fig S4Q and R).

Although the microglia‐induced apoptosis of cancer cells was enhanced by IFITM1, the BMDM‐ or BrM‐BMDM‐induced apoptosis of cancer cells was not affected by IFITM1 under noncontact coculture conditions (Appendix Fig S4S). Furthermore, IFITM1 overexpression promoted cancer cell apoptosis caused by primary microglia under contact coculture conditions at 6 h (anti‐cancer function mediated by phagocytosis and IFNγ) but not that caused by BMDMs, BrM‐BMDMs, or primary astrocytes (Appendix Fig S4T and U). In contrast, IFITM1 knockout repressed the apoptosis caused by primary microglia but not that caused by BMDMs or BrM‐BMDMs (Fig 2H and Appendix Fig S4V).

Complement C3 secreted by brain metastatic cancer cells activates microglia

For molecular characterization of the mechanism by which brain metastatic cancer cells activate primary microglia, in vitro coculture systems were applied (Appendix Fig S4B and C), and TNFα was used as a marker of microglial activation since IFNγ expression was very low in vitro (Appendix Fig S5A and B). Fluorescence‐activated cell sorting (FACS) analyses were performed to detect apoptotic cancer cells in contact or noncontact coculture with primary microglia. Cancer cells activated microglia to the same extent under contact and noncontact coculture conditions, but microglia were activated to a lesser extent by conditioned medium (Fig 3A and Appendix Fig S5C), indicating that one or more molecules in the medium continually secreted by cancer cells probably mediated this function. The expression of iNOS was not markedly changed by the conditioned medium (Appendix S5D). The ability of conditioned medium to activate microglia was decreased when the preheating temperature was increased from 37 to 72°C, indicating that the molecule(s) were protein(s) (Fig 3B). Ultrafiltration experiments showed that the protein(s) were larger than 100 kDa (Fig 3C and Appendix Fig S5E and F).

Figure 3. Complement C3 mediates the IFITM1‐induced anti‐brain metastasis effect of microglia.

1. ELISA results of mouse TNFα in supernatants obtained from the indicated coculture systems at the indicated time points. Primary mouse microglia, 6 × 10⁴ cells from 16 BALB/c mice. H460, 6 × 10⁴ cells. CM, conditioned medium (from 6 × 10⁴ H460 cells; one independent experiment). We would like to point out that TNFα, instead of IFNγ, was used as a marker of microglial activation in vitro (Fig 3A–C and H–I, Appendix Fig S5O and P) because IFNγ expression was very low in vitro (Appendix Fig S5A and B).
2. ELISA results of mouse TNFα in the supernatants of primary mouse microglia (6 × 10⁴, from 16 BALB/c mice) stimulated with preheated conditioned medium from H460 cells (1 × 10⁴). CM, conditioned medium (two independent experiments).
3. ELISA results of mouse TNFα in the supernatants of primary mouse microglia (6 × 10⁴, from 16 BALB/c mice) stimulated with conditioned medium or molecular weight fractions of the conditioned medium from H460 cells (1 × 10⁴). CM, conditioned medium (four independent experiments).
4. Venn diagram showing the overlap among candidate genes identified by mass spectrometry of the > 100 kDa fraction of conditioned medium from H460, H1299, H2030, and H292 cells (1 × 10⁶ cells for each condition).
5. ELISA results of mouse IFNγ in the brain on day 1 after intracardiac injection of the indicated LLC‐BrM cells (control, C3 knockout, IFITM1 overexpression, or IFITM1 overexpression in combination with C3 knockout, 1 × 10⁶) into C57BL/6 mice. Normal brain tissue was used as the control (The n‐values denote the number of mice per group).
6. ELISA results of mouse IFNγ in the brain on day 1 after intracardiac injection of LLC‐BrM cells transduced with the indicated plasmids (1 × 10⁶ cells) into C57BL/6 mice. Normal brain tissue was used as the control (The n‐values denote the number of mice per group).
7. Control and IFITM1‐overexpressing LLC‐BrM cells (1 × 10⁶, tdTomato) were surrounded by activated microglia (left, green, Iba1⁺) and complement C3 (left, white) in the brain on day 1 after intracardiac injection into C57BL/6 mice (normal, 3 mice; day 1, 3 mice; and day 7, 4 mice). Complement C3 intensities in the fields enclosed in the yellow dotted lines (ROIs) are shown (right; 90, 90, and 120 fields for normal, day 1, and day 7, respectively). The ROI was set as the area of a single metastatic cell. ImageJ was used to quantify the immunofluorescence intensity in the ROI. Normal brain tissue was used as the control. ROI, region of interest.
8. Results of ELISA for mouse TNFα in the culture supernatant of primary mouse microglia (6 × 10⁴, from 16 C57BL/6 mice), primary mouse BMDMs (6 × 10⁴, from 3 C57BL/6 mice), or primary mouse BrM‐BMDMs (6 × 10⁴, from 37 C57BL/6 mice) stimulated with conditioned medium from control LLC‐BrM cells (1 × 10⁴) and IFITM1‐knockout LLC‐BrM cells (1 × 10⁴). BMDMs, primary bone‐marrow‐derived macrophages. BrM‐BMDMs, primary bone‐marrow‐derived macrophages from the brain metastatic lesions of LLC‐BrM cell‐isografted C57BL/6 mice on day 20 after the intracardiac injection of LLC‐BrM cells (1 × 10⁵ cells). CM, conditioned medium (three independent experiments).
9. Results of ELISA for mouse TNFα in the culture supernatant of primary mouse microglia (6 × 10⁴, from 16 C57BL/6 mice), primary mouse BMDMs (6 × 10⁴, from 3 C57BL/6 mice), or primary mouse BrM‐BMDMs (6 × 10⁴, from 37 C57BL/6 mice) stimulated with DMEM‐HG and conditioned medium from control LLC‐BrM cells (1 × 10⁴) and C3‐overexpressing LLC‐BrM cells (1 × 10⁴). BMDMs, primary bone‐marrow‐derived macrophages. BrM‐BMDMs, primary bone‐marrow‐derived macrophages from the brain metastatic lesions of LLC‐BrM cell‐isografted C57BL/6 mice on day 20 after the intracardiac injection of LLC‐BrM cells (1 × 10⁵ cells). CM, conditioned medium (three independent experiments).
10. Flow cytometric analysis of complement C3a receptor expression on the cell membrane of primary mouse microglia (from 16 C57BL/6 mice) or primary mouse BMDMs (from 3 C57BL/6 mice). Representative flow cytometric histograms are shown, and the bar graph shows the averaged MFI values. BMDMs, primary bone‐marrow‐derived macrophages. MFI, mean fluorescence intensity (four independent experiments).
11. Box plot showing complement C3a receptor 1 mRNA expression in human microglia and BMDMs from healthy individuals. The boxes indicate the 25^th to 75^th percentiles. The white horizontal lines within the boxes indicate the median levels, and the whiskers indicate the minimum and maximum values (in the style of Tukey). The RNA‐seq count data were downloaded from https://joycelab.shinyapps.io/braintime/ (the n‐values denote the number of clinical samples per group, and one sample obtained from one patient was performed for one RNA‐seq).
12. Flow cytometric analysis of apoptotic LLC‐BrM cells (6 × 10⁴) in contact coculture with primary mouse microglia (6 × 10⁴, from 16 C57BL/6 mice), which were activated by conditioned medium from control or C3‐knockdown H460 cells (1 × 10⁶). CM, conditioned medium (three independent experiments).
13. Bioluminescence imaging and quantification of brain metastases on day 12 in C57BL/6 mice isografted with control LLC‐BrM cells (Ctrl/sgCtrl), IFITM1‐overexpressing LLC‐BrM cells (IFITM1/sgCtrl), and LLC‐BrM cells with IFITM1 overexpression in combination with C3 knockout (IFITM1/sgC3‐1 or ‐3) by intracardiac injection (1 × 10⁵ cells; The n‐values denote the number of mice per group).
14. Bioluminescence imaging and quantification of brain metastases on day 12 in BALB/c nude mice allografted with control LLC‐BrM cells (sgCtrl/Ctrl), IFITM1‐knockout LLC‐BrM cells (sgIFITM1‐1 or −2/Ctrl), or LLC‐BrM cells with IFITM1 knockout in combination with C3 overexpression (sgIFITM1‐1 or −2/C3) by intracardiac injection (1 × 10⁵ cells; The n‐values denote the number of mice per group).
15. Box plot showing the correlation of IFITM1 and complement C3 mRNA expression in lung cancer patients in the TCGA dataset. The boxes indicate the 25^th to 75^th percentiles. The white horizontal lines within the boxes indicate the median levels, and the whiskers indicate the minimum and maximum values (in the style of Tukey; The n‐values denote the number of clinical samples per group, and one sample obtained from one patient was performed for one RNA‐seq).

Data information: The data are presented as the mean ± s.e.m. values. P‐values were determined by unpaired one‐way (C, E, F, and L–N) or two‐way (G–I) ANOVA with uncorrected Fisher's LSD test, an unpaired two‐tailed Student's t‐test with Welch's correction (J), or the Wilcoxon rank‐sum test (K and O). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and n.s., not significant.

Source data are available online for this figure.

Subsequent mass spectrometric analysis revealed complement component 3 (C3) as the only secreted protein that was larger than 100 kDa and could potentially activate microglia (Fig 3D and Appendix Table S5; Reis et al, 2018). Remarkably, both activation of microglia in the brain (Fig 3E, sgC3‐1 or sgC3‐3 vs. control and Appendix Fig S5G, shC3‐3 or shC3‐6 vs. control) and microglial activation induced by conditioned medium in vitro (Appendix Fig S5H and I) were reduced by complement C3 knockdown or knockout but were enhanced by overexpression of complement C3 both in vivo (Fig 3F and Appendix Fig S5J and K, C3 vs. control) and in vitro (Appendix Fig S5L). As expected, both the protein (Fig 3G, day 1) and mRNA (Appendix Fig S5M and N) expression of complement C3 were upregulated by overexpression of IFITM1 under basal and IFNγ‐stimulated conditions, which induced stronger activating effects on microglia (Fig 3E and Appendix Fig S5G, IFITM1 vs. Control), and the effects on microglial activation were reversed by complement C3 depletion both in vivo (Fig 3E, IFITM1/sgC3‐1 or IFITM1/sgC3‐3 vs. control and Appendix Fig S5G, IFITM1/shC3‐3 or IFITM1/shC3‐6 vs. IFITM1) and in vitro (Appendix Fig S5O). However, neither IFITM1 nor complement C3 affected the conditioned medium‐induced activation of BMDMs or BrM‐BMDMs (Fig 3H and I, and Appendix Fig S5P) due to the significantly lower expression of complement C3 receptor (C3aR) in BMDMs than in microglia (Fig 3J and K).

Additionally, microglia‐induced apoptosis of cancer cells was reduced when microglia were activated by the conditioned medium from complement C3‐knockdown cells (Fig 3L and Appendix Fig S5Q and R) but was elevated when microglia were activated by the conditioned medium from complement C3‐overexpressing cells (Appendix Fig S5S). Interestingly, knockout of complement C3 in cancer cells prevented the anti‐brain metastasis effect of IFITM1 in vivo (Fig 3M and Appendix Fig S5T), and overexpression of complement C3 rescued the effect of IFITM1 knockout on brain metastasis (Fig 3N and Appendix Fig S5U). Moreover, the expression of complement C3 mRNA was positively correlated with IFITM1 mRNA expression in human lung cancer samples (Fig 3O and Appendix Fig S5V and W).

These studies support a model in which incipient brain metastatic cancer cells with high expression of IFITM1 are able to activate microglia by secreting complement C3. Then, activated microglia eliminate cancer cells via IFNγ release and phagocytosis, and IFNγ reciprocally induces IFITM1 expression in cancer cells.

IFITM1 promotes the anti‐brain metastatic activity of CD8⁺ T cells

As indicated above, the effect of IFITM1 on overall survival was stronger in immunocompetent C57BL/6 mice (Appendix Fig S1E) than in immunodeficient BALB/c nude mice (athymic with a greatly reduced number of T cells, Fig 1J), and brain metastases were infiltrated with T cells in patients (Berghoff et al, 2016; Quail & Joyce, 2017; Priego et al, 2018; Achrol et al, 2019; Friebel et al, 2020; Klemm et al, 2020). Thus, we postulated that IFITM1 may regulate the anti‐brain metastatic activity of T cells. In fact, overexpression of IFITM1 augmented the efficacy of adoptive T cell immunotherapy in suppressing brain metastasis and extending overall survival times by injecting primary CD8⁺ and not CD4⁺ T cells into immunodeficient mice either before the establishment of brain metastases (Fig 4A and B and Appendix Fig S6A and B) or after this event (Fig 4C and Appendix Fig S6C). Conversely, depletion of CD8⁺ T cells (Fig 4D) reversed the inhibition of brain metastasis caused by IFITM1 overexpression (Fig 4E). Notably, IFITM1 overexpression enhanced the ability of primary microglia and CD8⁺ T cells to induce apoptosis in vitro (Fig 4F and Appendix Fig S6D), and microglia and CD8⁺ T cells were essential for the anti‐brain metastasis effect of IFITM1 in vivo (Fig 4G). As prophylactic vaccination and radiotherapy are commonly used to enhance the effect of immunotherapies preclinically and clinically, we used prophylactic vaccination with cancer cell lysates to mimic the initial immune recognition (Figs 4E and G, and 6F and Appendix Fig S7H and I).

Figure 4. IFITM1 enhances the anti‐metastatic activity of CD8⁺ T cells.

1. Schematic of efficacy of adoptive CD8⁺ T cell immunotherapy in suppressing brain metastasis (top far left), representative western blots of IFITM1 expression in control or IFITM1‐overexpressing LLC‐BrM cells (bottom far left), bioluminescence imaging (middle left), quantification of brain metastases (middle right), and overall survivals (far right) of BALB/c nude mice allografted with control or IFITM1‐overexpressing LLC‐BrM cells by intracardiac injection (1 × 10⁵ cells). Primary mouse naïve CD8⁺ T cells from 12 C57BL/6 mice or not were injected intravenously on days −1, 6, and 13 (1 × 10⁴ cells at each time point; The n‐values denote the number of mice per group, and four independent western blot experiments were performed).
2. Schematic of efficacy of adoptive CD8⁺ or CD4⁺ T cell immunotherapy in suppressing brain metastasis (far left), bioluminescence imaging (middle left), quantification of brain metastases (middle right), and overall survivals (far right) of BALB/c nude mice allografted with control or IFITM1‐overexpressing LLC‐BrM cells by intracardiac injection (1 × 10⁵ cells). Primary mouse naïve CD8⁺ or CD4⁺ T cells from 12 C57BL/6 mice were injected intravenously on days −1, 6, and 13 (1 × 10⁴ cells at each time point; The n‐values denote the number of mice per group).
3. Schematic of efficacy of adoptive CD8⁺ T cell immunotherapy in suppressing brain metastasis after establishment of brain metastases (far left), bioluminescence imaging (middle left), quantification of brain metastases (middle right), and overall survivals (far right) of BALB/c nude mice allografted with control or IFITM1‐overexpressing LLC‐BrM cells by intracardiac injection (1 × 10⁵ cells) was evaluated. Primary mouse naïve CD8⁺ T cells from 12 C57BL/6 mice (3 × 10⁵ cells at each time point) were injected intravenously on days 4, 11, 18, 25, and 32 (The n‐values denote the number of mice per group).
4. Flow cytometric analysis of CD8⁺ T cells in spleens/LNs and blood of the C57BL/6 mice on days 1 and 13 after intravenous injection of IgG or an anti‐CD8α antibody (250 μg/mouse) on day 0. LN, lymph node (Three mice per group).
5. Bioluminescence imaging and quantification of brain metastases in C57BL/6 mice isografted with control or IFITM1‐overexpressing LLC‐BrM cells by intracardiac injection (3 × 10⁵ cells). Mice were systemically primed boosted with homologous LLC‐BrM cell lysates (1 × 10⁶ cells for day −21 and 1 × 10⁵ cells for day −7) before injection of cancer cells, and IgG or an anti‐CD8α antibody (250 μg/mouse) was injected intravenously on days 3 and 5 (The n‐values denote the number of mice per group). Note that under anti‐CD8α antibody treatment condition, IFITM1 overexpression still inhibited the capacity of cancer cells to colonize the brain (P < 0.05).
6. Flow cytometric apoptosis analysis of control or IFITM1‐overexpressing LLC‐BrM cells (6 × 10⁴) in noncontact coculture with primary mouse microglia (1.8 × 10⁴, from 16 C57BL/6 mice) or in contact coculture with primary mouse naïve CD8⁺ T cells (3 × 10⁵, from 12 C57BL/6 mice) alone or in combination for 48 h (four independent experiments).
7. Bioluminescence imaging and quantification of brain metastases on day 20 in C57BL/6 mice isografted with control or IFITM1‐overexpressing LLC‐BrM cells by intracardiac injection (3 × 10⁵ cells). Mice were systemically primed boosted with homologous LLC‐BrM cell lysates (1 × 10⁶ cells for day −21 and 1 × 10⁵ cells for day −7) before injection of cancer cells. Then, IgG or an anti‐CD8α antibody (2 μg/mouse) was injected intravenously on day 3 alone or in combination with minocycline treatment (33 mg/kg) on day 1. Minocycline was administered intraperitoneally (The n‐values denote the number of mice per group).

Data information: The data are presented as the mean ± s.e.m. values. P‐values were determined by unpaired two‐way (quantification of brain metastases in A–C and D–F) or one‐way (G) ANOVA with uncorrected Fisher's LSD test or the log‐rank test (survival in A–C). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and n.s., not significant.

Source data are available online for this figure.

Figure 6. IFITM1 synergizes with anti‐PD‐1 immunotherapy and is crucial to the synergistic anti‐brain metastasis effect of combination therapy with an oncolytic virus and a PD‐1 blocking antibody.

1. GSEA revealed an association between IFITM1 expression and enrichment of IFNγ signaling pathway genes in clinical lung cancer samples from the TCGA dataset (n = 415). GSEA, gene set enrichment analysis. NES, normalized enrichment score. FDR, false discovery rate.
2. Flow cytometric analysis of PD‐L1 expression on the cell membrane of control or IFITM1‐overexpressing H460 cells (6 × 10⁴) stimulated with IFNγ (1,000 U/ml) at the indicated time points. Representative flow cytometric histograms are shown (three independent experiments).
3. Box plot showing the correlation of IFITM1 mRNA expression and immune checkpoint blockade therapy response in a melanoma clinical trial. The boxes indicate the 25^th to 75^th percentiles. The white horizontal lines within the boxes indicate the median levels, and the whiskers indicate the minimum and maximum values (in the style of Tukey; The n‐values denote the number of clinical samples per group, and one sample obtained from one patient was performed for one RNA‐seq).
4. Box plot showing the correlation of IFITM1 mRNA expression and anti‐PD‐1 immunotherapy response in a metastatic gastric cancer clinical trial. The boxes indicate the 25^th to 75^th percentiles. The white horizontal lines within the boxes indicate the median levels, and the whiskers indicate the minimum and maximum values (in the style of Tukey; The n‐values denote the number of clinical samples per group, and one sample obtained from one patient was performed for one RNA‐seq).
5. Bioluminescence imaging, quantification of brain metastases, and overall survival of BALB/c nude mice allografted with control or IFITM1‐overexpressing LLC‐BrM cells by intracardiac injection (1 × 10⁵ cells). Primary mouse naïve CD8⁺ T cells from 12 C57BL/6 mice (1 × 10⁴ cells at each time point) were injected intravenously on days −1, 6, and 13, and IgG or an anti‐PD‐1 antibody (100 μg/mouse) was injected intravenously on days 3, 6, and 9 (The n‐values denote the number of mice per group).
6. Bioluminescence imaging, quantification of brain metastases, and overall survival of C57BL/6 mice isografted with control or IFITM1‐overexpressing LLC‐BrM cells by intracardiac injection (3 × 10⁵ cells). Mice were systemically primed boosted with homologous LLC‐BrM cell lysates (1 × 10⁶ cells for day ‐21 and 1 × 10⁵ cells for day ‐7) before injection of cancer cells, and IgG or an anti‐PD‐1 antibody (100 μg/mouse) was then injected intravenously on days 3, 6, and 9 (The n‐values denote the number of mice per group).
7. Bioluminescence imaging and quantification of brain metastases in BALB/c nude mice allografted with control or IFITM1‐knockout LLC‐BrM cells by intracardiac injection (1 × 10⁵ cells). Mice were adoptively transferred with primary mouse naïve CD8⁺ T cell (1 × 10⁴ cells at each time point, from 12 C57BL/6 mice) on days ‐1, 6, and 13. IgG (100 μg/mouse, days 3, 6, and 9), an anti‐PD‐1 antibody (100 μg/mouse, days 3, 6, and 9), and/or UV‐irradiated oncolytic virus (10⁸ pfu/mouse, days 0, 2, and 4) was injected intravenously alone or in combination as indicated (The n‐values denote the number of mice per group).
8. Bioluminescence imaging and quantification of brain metastases in C57BL/6 mice isografted with control or IFITM1‐knockout LLC‐BrM cells by intracardiac injection (3 × 10⁵ cells). IgG (30 μg/mouse, days 3, 6, and 9), an anti‐PD‐1 antibody (30 μg/mouse, days 3, 6, and 9), or UV‐irradiated oncolytic virus (106 pfu/mouse, days 0, 2, and 4) was injected intravenously alone or in combination as indicated (The n‐values denote the number of mice per group).

Data information: The data are presented as the mean ± s.e.m. values. P‐values were determined by the Wilcoxon rank‐sum test (C and D), unpaired two‐way (quantification of brain metastases in E and F) or one‐way (quantification of brain metastases in G and H) ANOVA with uncorrected Fisher's LSD test, or the log‐rank test (survival in E and F). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and n.s., not significant.

IFITM1 enhances the cytolytic activity of CD8⁺ T cells by increasing the expression and membrane localization of MHC‐I

To examine the molecular basis of IFITM1 function, we performed a coimmunoprecipitation (co‐IP) assay of cell membrane proteins (Fig 5A and B) followed by mass spectrometry analysis. We found for the first time that major histocompatibility complex class I (MHC‐I) was associated with IFITM1 in the cell membrane fraction after IFNγ stimulation (Appendix Table S6). This association was confirmed by immunostaining and co‐IP (Fig 5C and D). Cell surface expression of MHC‐I was decreased by knockout of IFITM1 both in vivo (Fig 5E and F and Appendix Fig S6E) and in vitro (Fig 5G) and, in contrast, was increased by overexpression of IFITM1 (Fig 5H–J, and Appendix Fig S6F and G) although not by overexpression of IFITM2 or IFITM3 (Appendix Fig S6F). However, MHC‐I stability on the cell membrane was not altered by IFITM1 (Appendix Fig S6H). The above results suggest that the interaction between IFITM1 and MHC‐I contributes only to the membrane localization of MHC‐I. Surprisingly, IFITM1 (Fig 5K and L, and Appendix Fig S6I–K) also modulated MHC‐I mRNA expression, but IFITM2 and IFITM3 did not (Appendix Fig S6K). IFITM1 marginally influenced the mRNA expression of MHC‐II, B2M, and genes encoding proteins involved in endogenous antigen processing (TAP1, TAP2, TAPBP, ERAP1, and ERAP2) and both the mRNA and protein expression of CD47 (Appendix Fig S6L and M).

Figure 5. IFITM1 increases the expression and membrane localization of MHC‐I to promote the cytolytic activity of CD8⁺ T cells.

1. Immunoblot analysis of IFITM1 expression on the cell membrane, in the cytoplasm, and in the nucleus of H1299 cells with or without IFNγ stimulation (1,000 U/ml) for 48 h (four independent experiments).
2. Immunoprecipitation of endogenous membrane IFITM1 of H1299 cells with or without IFNγ stimulation (1,000 U/ml) for 48 h (three independent experiments).
3. Immunofluorescence staining of H1299 cells with anti‐IFITM1 and anti‐MHC‐I antibodies. H1299 cells (6 × 10⁴) were stimulated with 1,000 U/ml IFNγ for 48 h and were then treated with the anti‐MHC‐I antibody for 1 h. The colocalization percentages of IFITM1 (red) and MHC‐I (green) are shown (100 cells for each condition).
4. Co‐IP of endogenous IFITM1 and MHC‐I proteins in H1299 cells (1 × 10⁶) after IFNγ stimulation (1,000 U/ml) for 48 h (three independent experiments).
5. Flow cytometric analysis of MHC‐I expression on the cell membrane of LLC‐BrM cells isolated from the brains of LLC‐BrM cell‐isografted C57BL/6 mice on day 1 after intracardiac injection of control or IFITM1‐knockout LLC‐BrM cells (3 × 10⁵). Representative flow cytometric histograms are shown, and the bar graph shows the averaged MFI values. MFI, mean fluorescence intensity (The n‐values denote the number of mice per group).
6. Representative immunohistochemical images and staining intensities of MHC‐I expression in the brain metastatic lesions in LLC‐BrM cell‐isografted C57BL/6 mice on day 20 after intracardiac injection of control or IFITM1‐knockout LLC‐BrM cells (3 × 10⁵, Cas9/blue). Mice were systemically primed boosted with homologous LLC‐BrM cell lysates (1 × 10⁶ cells for day −21 and 1 × 10⁵ cells for day −7) before injection of cancer cells, and IgG or an anti‐CD8α antibody (250 μg/mouse) was then injected intravenously on days 3 and 5 (The n‐values denote the number of mice per group. One datapoint means one brain metastatic lesions; 114, 51, and 106 brain metastatic lesions for the sgControl, sgIFITM1‐1, and sgIFITM1‐2 groups, respectively).
7. Immunoblot analysis of MHC‐I expression in control or IFITM1‐knockout H1299 cells (6 × 10⁴) after IFNγ stimulation (1,000 U/ml) for 48 h (four independent experiments).
8. Flow cytometric analysis of MHC‐I expression on the cell membrane of control or IFITM1‐overexpressing H1299 cells (6 × 10⁴) after IFNγ stimulation (1,000 U/ml) for 48 h (six independent experiments).
9. Immunoblot analysis of MHC‐I expression in the cytoplasm, on the cell membrane, and in the nucleus of control and IFITM1‐overexpressing H1299 cells (6 × 10⁴; three independent experiments).
10. Bar plot showing the ratio of MHC‐I expression on the cell membrane to MHC‐I expression in the cytoplasm in the cells described in (I) (three independent experiments).
11. qPCR analysis of MHC‐I mRNA expression in control and IFITM1‐overexpressing H1299 cells (6 × 10⁴) with or without IFNγ stimulation (1,000 U/ml) for 24 h (three independent experiments).
12. qPCR analysis of MHC‐I mRNA expression in control and IFITM1‐knockdown H1299 cells (6 × 10⁴) with or without IFNγ stimulation (1,000 U/ml) for 24 h (three independent experiments).
13. Cleaved caspase 3 (green)‐positive LLC‐BrM cells (tdTomato) were surrounded by CD8⁺ T cells (white) in the brain metastatic lesions of LLC‐BrM cell‐isografted C57BL/6 mice on day 20 after intracardiac injection of control or IFITM1‐overexpressing LLC‐BrM cells (1 × 10⁶ cells). The positively stained areas of single CD8⁺ T cells, clustered CD8⁺ T cells, cleaved caspase 3⁺ cancer cells, and tdTomato⁺ cancer cells in the ROI were calculated by ImageJ. The ratios of the areas between the clustered and single CD8⁺ T cells are shown as the numbers of CD8⁺ T cells in the ROI, and the ratios of the areas between cleaved caspase 3⁺ cancer cells and tdTomato⁺ cancer cells are shown as the percentages of cleaved caspase 3⁺ cancer cells in the ROI (90 fields per group, 3 mice per group). Yellow dotted lines (ROIs), brain metastatic lesions; ROI, region of interest.
14. Flow cytometric apoptosis analysis of control or IFITM1‐overexpressing LLC‐BrM cells (6 × 10⁴) in contact coculture with primary mouse naïve CD8⁺ or CD4⁺ T cells (3 × 10⁵, from 12 C57BL/6 mice) in medium supplemented with IgG or an anti‐MHC‐I antibody (3 μg/ml) for 48 h (at least three independent experiments).
15. Kaplan–Meier survival curves of lung cancer patients in the TCGA dataset stratified by IFITM1 and MHC‐I mRNA expression. HR, hazard ratio.

Data information: The data are presented as the mean ± s.e.m. values. P‐values were determined by an unpaired two‐tailed Student's t‐test with Welch's correction (C, J, and M), unpaired one‐way (E and F) or two‐way (K, L, and N) ANOVA with uncorrected Fisher's LSD test, or the log‐rank test (O). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and n.s., not significant.

Source data are available online for this figure.

MHC‐I molecules present peptides derived from intracellular proteins to CD8⁺ T cells, hence converting them into cytotoxic T lymphocytes (CTLs), which are critical immune effectors to eliminate cancer cells (Schreiber et al, 2011). Overexpression of IFITM1 increased cancer cell apoptosis induced by CD8⁺ T cells both in vivo (Fig 5M) and in vitro (Fig 5N and Appendix Fig S6N and O) but did not increase that induced by CD4⁺ T cells (Fig 5N and Appendix Fig S6N). In addition, an MHC‐I blocking antibody impeded this increase in apoptosis (Fig 5N and Appendix Fig S6N). Moreover, overexpression of IFITM2 or IFITM3 did not affect primary CD8⁺ T‐cell‐induced apoptosis (Appendix Fig S6O). In addition, coexpression of IFITM1 and MHC‐I mRNA was associated with overall survival in lung cancer patients in large datasets (Fig 5O and Appendix Fig S6P and Q).

Collectively, these studies support a model in which microglia (brain‐resident macrophages) are activated by complement C3 secreted by brain metastatic cancer cells with high IFITM1 expression and, after activation, subsequently, destroy cancer cells via IFNγ secretion and phagocytosis. IFNγ also reciprocally induces IFITM1 expression in cancer cells. Moreover, effector CD8⁺ T cells could be attracted by activated microglia and cancer cells (Spranger et al, 2015; van der Woude et al, 2017) and recognize antigens presented by MHC‐I molecules, whose expression is upregulated by IFITM1, and consequently, promoting cancer cell apoptosis (Appendix Fig S6R and S).

IFITM1 synergizes with PD‐1 immune checkpoint blockade therapy

Gene set enrichment analysis (GSEA) of expression datasets further revealed significant enrichment of IFNγ signaling pathway genes in lung cancer patients with high expression of IFITM1 (Fig 6A, and Appendix Fig S7A and B). Furthermore, the expression of PD‐L1, an IFNγ signaling pathway gene, was modulated by IFITM1 (Fig 6B and Appendix Fig S7C), but PD‐L2 expression was not (Appendix Fig S7D and E). Therefore, IFNγ induced the expression of IFITM1 and PD‐L1, and IFITM1 inhibited the expression of PD‐L1. As an inhibitor, IFITM1 might exert a stronger effect on PD‐L1 than the inducer IFNγ. Notably, immune checkpoint blockade (ICB) therapy response was significantly positively correlated with IFITM1 expression in human cancer clinical trials (Fig 6C and D, and Appendix Fig S7F). IFITM1 overexpression combined with PD‐1 blocking antibody treatment improved the efficacy of adoptive CD8⁺ T cell immunotherapy in suppressing brain metastasis and extending overall survival times in immunodeficient mice (Fig 6E and Appendix Fig S7G). Similar synergistic effects of IFITM1 overexpression and PD‐1 blocking antibody treatment on brain metastasis and overall survival times were observed in immunocompetent mice (Fig 6F). In addition, IFITM1 overexpression combined with PD‐1 blocking antibody treatment increased apoptosis of cancer cells induced by primary CD8⁺ T cells under in vitro contact coculture conditions (Appendix Fig S7H and I).

Virus‐induced expression of IFITM1 mediates the synergistic anti‐brain metastasis effect of combination therapy with an oncolytic virus and a PD‐1 blocking antibody

Combination therapy with an oncolytic virus (OV) and a PD‐1 blocking antibody greatly increased the response rates in patients with advanced melanoma (Ribas et al, 2017). Oncolytic viruses function more through stimulation of anti‐cancer immunity than through killing of virus‐infected cancer cells (Bommareddy et al, 2018; Russell & Barber, 2018). Additionally, IFITM1 is highly inducible by viral infections (Bailey et al, 2014; Weston et al, 2014). We thus speculated that IFITM1 may play a crucial role in combination therapy with an oncolytic virus and a PD‐1 blocking antibody.

Ultraviolet (UV)‐irradiated oncolytic viruses, which lack oncolytic activity but still stimulate IFITM1 expression in cancer cells, were used in our experiments (Appendix Fig S7J). Remarkably, IFITM1 knockout completely suppressed the synergistic effects of the UV‐irradiated oncolytic virus/PD‐1 blocking antibody combination therapy on the efficacy of adoptive CD8⁺ T cell immunotherapy in suppressing brain metastasis in immunodeficient mice (Fig 6G and Appendix Fig S7K) and on the inhibition of brain metastasis in immunocompetent mice (Fig 6H).

Discussion

Increasing evidence points to the role of immunoediting in metastatic colonization and suggests that immune evasion is critical for metastatic relapse (Mohme et al, 2017). However, the underlying mechanisms remain incompletely understood. As shown in the study, we found that loss of IFITM1 induced several human lung cancer cell lines (harboring the N‐Ras Q61K/p53del mutation‐H1299, or EGFR Δexon19 PC9) to colonize the brain in T‐cell‐deficient nude mice (only microglia exert antimetastatic function through IFITM1), suggesting that the effect is independent of a specific oncogenic background. We later found that loss of IFITM1 exerted a similar prometastatic effect on syngeneic models. Cancer cells with high expression of IFITM1 not only activated microglia but also enhanced the cytolytic activity of CD8⁺ T cells. We, therefore, presume that IFITM1 suppresses brain metastasis in nude mice by activating microglia, whereas it exerts this effect in immunocompetent mice by activating both microglia and CD8⁺ T cells. Moreover, CD8⁺ T cells may exert stronger effects on the antimetastatic function of IFITM1 overexpression than microglia (Fig 4E). In addition, we analyzed the expression of IFITM1 in clinically annotated tissue microarrays (TMAs) and publicly available microarray databases of lung cancer samples and found that low expression of IFITM1 in tumor tissues predicts poor survival of lung cancer patients.

Mechanistically, our studies indicate that incipient brain metastatic cancer cells activated microglia via complement C3, and IFITM1 enhanced the cytolytic activity of CD8⁺ T cells by increasing the expression and membrane localization of MHC‐I. Once activated, microglia (of the innate immune system) and cytotoxic CD8⁺ T lymphocytes (of the adaptive immune system) jointly eliminated cancer cells via IFNγ release, phagocytosis induction, and T cell killing. Moreover, the analysis of the expression of complement C3 and MHC‐I in publicly available microarray databases of lung cancer samples indicated that the expression of complement C3 mRNA was positively correlated with IFITM1 mRNA expression, and low expression of IFITM1 and MHC‐I in tumor tissues predicts poor survival of lung cancer patients.

Interestingly, IFITM1 did not affect macrophage functions either in vivo or in vitro due to the significantly lower expression of complement C3 receptor in peripheral macrophages compared with brain‐resident microglia in both humans and mice. Microglia, the resident macrophages (terminally differentiated cells of the myeloid lineage) of the central nervous system, develop from erythromyeloid precursors in the yolk sac (Schulz et al, 2012), are the first and main element of the innate immune system encountered in the brain, and constitute as much as 10–15% of all cells in the brain. They constantly scan the environment and are activated by neoplastic tumors and infectious agents (Fidler, 2015; Quail & Joyce, 2017; Wolf et al, 2017; Schulz et al, 2020). Single‐cell RNA sequencing of healthy mouse brain tissue showed very little to no expression of C3aR1 in cell types other than microglia (Zhang et al, 2014; He et al, 2018; Vanlandewijck et al, 2018; Harder et al, 2020). Recently, Adrienne Boire and colleagues showed that complement C3 promotes the ability of breast and lung cancer cells to colonize the leptomeninges (Boire et al, 2017). We demonstrated instead that IFITM1 inhibited the capacity of lung cancer cells to colonize the brain parenchyma. Consistent with diverging roles for C3 in brain metastasis and leptomeningeal metastasis, Boire and colleagues noticed that C3 expression was much lower in brain metastatic derivatives (BrM) than in leptomeningeal metastatic derivatives (LeptoM) of the PC9 and LLC lung cancer cell lines (Fig 2B and Appendix Fig S2E in Boire et al, 2017). We did not directly address this issue because we did not detect any leptomeningeal metastases in our lung cancer metastasis models.

IFNγ can have dual effects on antitumor responses. The relative balance of the opposing effects induced by IFNγ determines the overall functional outcome. Hence, the molecule and mechanism underlying context‐dependent IFNγ function will be important for developing therapeutic strategies to manipulate IFNγ activity to promote health and suppress disease (Minn & Wherry, 2016; Ivashkiv, 2018). Undoubtedly, research findings on IFITM1, which had effects on only antimetastatic responses, are of not only theoretical but also clinical significance. They have the potential to be harnessed to develop new therapeutic strategies to improve host defense and augment responses to immune checkpoint blockade therapies in cancer patients.

Immune checkpoint blockade therapy has considerably improved the treatment of cancer patients; however, the majority of patients still do not respond to this therapy (Gao et al, 2016b; Taggart et al, 2018). Dysfunction of MHC‐I antigen presentation and deficiency of the IFNγ signaling pathway have been reported to be the molecular mechanisms of resistance to PD‐1 blockade therapy (Minn & Wherry, 2016; McGranahan et al, 2017; Ivashkiv, 2018). We showed that IFITM1 not only increased expression and cell membrane localization of MHC‐I to enhance antigen presentation on cancer cells but also correlated with IFNγ signaling pathway activity. Therefore, combined overexpression of IFITM1 and treatment with a PD‐1 blocking antibody overcame these resistance mechanisms and elicited a synergistic anti‐brain metastasis effect (Fig 6E and F, and Appendix Fig S7G; Popovic et al, 2018; Galon & Bruni, 2019). Moreover, the correlation analysis indicated that the immune checkpoint blockade (ICB) therapy response was significantly positively correlated with IFITM1 expression in human cancer clinical trials.

Imlygic, a genetically modified oncolytic virus, has recently been approved by the FDA for the treatment of melanoma patients (Kaufman et al, 2015; Russell & Barber, 2018; Samson et al, 2018) and functions more through stimulation of anti‐cancer immunity than through killing of virus‐infected cancer cells (Bommareddy et al, 2018; Russell & Barber, 2018; Samson et al, 2018). Combination therapy with Imlygic and a PD‐1 blocking antibody strongly increased the response rate and complete response rate in patients with advanced melanoma to as high as 62 and 33%, respectively, because of the induction of initial anti‐cancer immune responses by the oncolytic virus and the maintenance of these responses by the PD‐1 blocking antibody (Ribas et al, 2017). IFITM1 is also highly inducible by viral infection and IFNs and thus mediates cellular resistance to viral infection by inhibiting viral entry and replication (Bailey et al, 2014; Weston et al, 2014). Furthermore, we demonstrated that virus‐induced expression of IFITM1 played a critical role in the synergistic anti‐brain metastasis effect of oncolytic virus therapy combined with the PD‐1 blocking antibody (Fig 6G and H and Appendix Fig S7K). Moreover, we demonstrated that IFITM1 functioned independently of the oncolytic activity of the oncolytic virus–UV‐irradiated oncolytic virus, which cannot replicate, was used in our experiments, but the virus still stimulated IFITM1 expression in cancer cells.

Taken together, our data provide new insights into complex reciprocal interactions between brain metastatic cancer cells and immune cells during metastatic colonization in the brain. Our findings may illuminate a unified mechanism of brain metastasis and suggest that combination therapy with an IFITM1‐overexpressing oncolytic virus and immune checkpoint blockade may block brain metastasis (Ribas et al, 2017; Bommareddy et al, 2018; Twumasi‐Boateng et al, 2018).

Our study had four shortcomings. First, the theoretical redundancy of the screening for metastatic brain colonization was not sufficient to guarantee an unbiased analysis. Since the original purpose of in vivo genome‐wide CRISPR‐Cas9 screen was to identify the suppressors of metastatic colonization, an intracardiac injection model was used, and the theoretical redundancy of the screenings was 12.23–103.47 times the number of sgRNAs (Appendix Table S1). However, the theoretical redundancy of the screens for metastatic brain colonization decreased to 0.61–5.17 times the number of sgRNAs because a small population of cancer cells colonized the brain after intracardiac injection. Therefore, reducing the number of sgRNAs in the sub‐pool and finding other animal models to increase the population of cancer cells colonizing the brain will improve the theoretical redundancy of screens for metastatic brain colonization.

Second, in our study, PLX5622, BLZ945, or minocycline was used to deplete microglia or inhibit microglial activation. PLX5622, a CSF1R inhibitor, was developed and used to deplete microglia with little effect on macrophages. Although the results of published studies and our results indicated that PLX5622 had little effect on macrophages (Appendix Fig S3G; Bellver‐Landete et al, 2019; Benbenishty et al, 2019; Willis et al, 2020), Lei's study (Lei et al, 2020) suggested that PLX5622 was not a specific microglial inhibitor, and that PLX5622 might affect bone‐marrow‐derived macrophages. Therefore, other specific microglia inhibitors or specific genetic microglial deletion models, such as a temporally segregated TAM and DT administration strategy employed to exclusively deplete CNS myeloids in Cx3cr1^CreERT2iDTR mice, will aid in further verification of the microglial effects (Guldner et al, 2020).

Third, we found that incipient brain metastatic cancer cells with high expression of IFITM1 activated microglia by upregulating complement component 3. However, whether IFITM1‐induced activation of microglia regulates CD8⁺ T cells was not studied in this study, which still needs further exploration.

Finally, we found that IFNγ induced the expression of IFITM1 and PD‐L1, and IFITM1 inhibited the expression of PD‐L1. As an inhibitor, IFITM1 exerted a stronger effect on PD‐L1 than the inducer IFNγ. However, the potential mechanism is unclear and will be interesting to explore in the future.

Materials and Methods

In vivo genome‐wide CRISPR‐Cas9 screen in metastatic colonization

The human GeCKO v2 CRISPR knockout pooled library was a gift from Feng Zhang (Addgene 1000000049; Sanjana et al, 2014). The library consists of 121,411 sgRNAs targeting 19,050 genes (6 sgRNAs per gene) and 1,288 microRNAs (4 sgRNAs per microRNA), and also includes 2,000 nontarget control sgRNAs. The library was subdivided into four subpools.

H1299 (N‐Ras Q61K/p53del), A549 (K‐Ras G12S), PC9 (EGFR Δexon19), H1650 (EGFR Δexon19), HCC827 (EGFR Δexon19), and H2030 (K‐Ras G12C) human lung cancer cells stably expressing both tdTomato‐luciferase and Cas9 (H1299 and A549) or only Cas9 (PC9, H1650, HCC827, and H2030) were infected independently with the four subpools at a multiplicity of infection (MOI) of 3:1 and selected with puromycin for 2 weeks to confirm genome editing. Then, the cells were inoculated into BALB/c nude mice by intracardiac injection (1 × 10⁶ cells/100 μl of PBS for H1299 and A549 cells, 2 × 10⁵ cells/100 μl of PBS for PC9, H1650, HCC827, and H2030 cells).

Between 42 and 180 days after injection, macroscopic metastatic lesions (visible to the naked eye) were dissected and minced to isolate cancer cells, and the screening experiments ended after 180 days. The clonogenic cancer cells were expanded in selective medium and genomic DNA was extracted. The resident sequences in the genomic DNA encoding the CRISPR sgRNAs were amplified with primers (Appendix Table S7), the PCR products were cloned into the TA cloning vector, and at least 20 independent clones were analyzed by Sanger sequencing. In addition, the genomic DNA fragments containing the sgRNA target sites were amplified by PCR, the PCR products were cloned into the TA cloning vector, and approximately six independent clones were analyzed by Sanger sequencing to assess the on‐target mutagenic efficiency of the sgRNAs.

Patient samples

LC817a, LC814a, LC1505, LC1504, LC10013a, and GL861 clinical lung cancer tissue microarrays (TMAs) were purchased from US Biomax. The TMAs comprised 321 primary lung tumors and 36 brain metastatic lesions.

Immunohistochemical staining for human IFITM1 was performed on paraffin‐embedded TMAs with a VECTASTAIN ABC HRP Kit (PK6100, Vector Labs) and DAB (D5905, Sigma). Immunohistochemical staining was performed with a primary antibody specific for IFITM1 (HPA004810, Sigma) followed by biotinylated goat anti‐rabbit IgG (BA‐1000, Vector Labs). The IFITM1 staining intensity (H‐score) was calculated with the formula 3 × percentage of strongly stained cells + 2 × percentage of moderately stained cells + percentage of weakly stained cells and ranged between 0 and 300. The representative images of strongly, moderately, and weakly stained cells are shown in Appendix Fig S2A. The expression of IFITM1 was evaluated by three researchers including two professional pathologists in a double‐blinded manner.

The mRNA expression level of IFITM1 was determined by qPCR. Total RNA was extracted using a RNeasy FFPE (formalin‐fixed, paraffin‐embedded tissue sections) Kit (73504, QIAGEN) and reverse transcribed with a ReverTra Ace qPCR RT Kit (FSQ‐101, TOYOBO). An amount of cDNA corresponding to a 1/30 dilution of starting RNA was used for one reaction. qPCR was performed with 2 × SYBR Green qPCR Master Mix (Low ROX; B21702, Bimake). All quantities were normalized to those of endogenous β‐actin. Experiments were performed with an Applied Biosystems 7500/7500 Fast instrument. Primers used to amplify IFITM1 are listed in Appendix Table S7.

Cell lines

The H1299 (N‐Ras Q61K/p53del), H460 (K‐Ras Q61H), and A549 (K‐Ras G12S) human lung cancer cell lines and LLC cell line (originating from a Lewis lung carcinoma in a C57BL/6 mouse) were originally obtained from the American Type Culture Collection (ATCC). LLC cells harbor a heterozygous K‐Ras G12V mutation (Li et al, 2017). The H1650 (EGFR Δexon19), HCC827 (EGFR Δexon19), and H292 (CDKN2A del) human lung cancer cell lines were originally obtained from ATCC. The PC9 (EGFR Δexon19) human lung cancer cell line was originally obtained from the European Collection of Authenticated Cell Cultures (ECACC). The H1650, HCC827, H292, and PC9 cell lines were kind gifts from Dr. Fan Zhang (Tongji University, China). The H2030 (K‐Ras G12C) human lung cancer cell line and 4T1 cell line (originating from a mammary carcinoma in a BALB/c mouse, CDKN2Adel/Kit A942S/p53 P31X; Zeitouni et al, 2017) were obtained from ATCC. The BV‐2 cell line (originating from a C57BL/6 mouse) was purchased from Shanghai SUER Biological Technology and originally obtained from the Interlab Cell Line Collection (ICLC). The 293FT cell line was purchased from Life Technologies. Brain metastatic LLC cells (LLC‐BrM cells) were obtained from brain lesions through three rounds of in vivo selection in C57BL/6 mice.

H1299, H460, H1650, HCC827, H292, and H2030 cells were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS, 04‐001‐1ACS, BI), 2 mM L‐glutamine (21051‐024, Gibco), and 100 U/ml penicillin / 0.1 mg/ml streptomycin (P/S, C0222, Beyotime Biotechnology). A549, LLC, LLC‐BrM, PC9, 4T1, BV‐2, and 293FT cells were cultured in DMEM‐HG supplemented with 10% FBS, 2 mM L‐glutamine, and P/S.

For bioluminescent tracking, cell lines were infected with a lentiviral vector encoding tdTomato and firefly luciferase, and tdTomato‐positive cells were isolated by FACS.

All human cell lines were authenticated by short tandem repeat (STR) analysis provided by Bio‐Research Innovation Center Suzhou, SIBCB, CAS, and were routinely tested for mycoplasma contamination.

Primary cells

Primary mouse microglia and astrocytes were isolated from the brains of newborn BALB/c or C57BL/6 mice according to the method of McCarthy and Giulian (McCarthy & de Vellis, 1980). Microglia were cultured in DMEM‐HG supplemented with 10% heat‐inactivated FBS, 10% horse serum (26050‐088, Gibco), 2 mM L‐glutamine, and P/S; astrocytes were cultured in DMEM‐HG supplemented with 10% heat‐inactivated FBS, 2 mM L‐glutamine, and P/S.

For determination of purity, microglia and astrocytes were grown on glass coverslips, fixed with 4% formaldehyde, and permeabilized with 0.1% Triton X‐100. Immunofluorescence staining was performed with anti‐Iba‐1 (microglia) or anti‐GFAP (astrocytes) antibodies followed by Alexa Fluor™ 488 goat anti‐rabbit IgG (H + L; A11034, Invitrogen). Coverslips were mounted using ProLong™ Gold Antifade Mountant with DAPI (P36931, Thermo Fisher Scientific).

To prepare primary mouse bone‐marrow‐derived macrophages (BMDMs), bone marrow cells were isolated from the femurs and tibias of 6‐ to 8‐week‐old C57BL/6 mice. After the red blood cell lysis procedure, the bone marrow cells were cultured in DMEM‐HG containing 10% heat‐inactivated FBS and supplemented with 25 ng/ml murine M‐CSF (315‐02‐100, PeproTech) for 7 days. The medium was replenished on days 3, 5, and 6. Then, the adherent macrophages (BMDMs) were collected and assessed by FACS using CD11b and F4/80 surface molecules. The average purity of BMDMs was approximately 97%.

To prepare primary mouse bone‐marrow‐derived macrophages from brain metastatic lesions (BrM‐BMDMs), LLC‐BrM cells (1 × 10⁵ cells) stably expressing tdTomato–luciferase were injected into the left ventricle of C57BL/6 mice. Twenty days after injection, brain metastatic lesions were collected and mechanically minced and digested. Myelin and debris were depleted by Percoll gradient centrifugation. BrM‐BMDMs (tdTomato⁻) were sorted using CD45⁺CD11b⁺Ly6C⁻Ly6G⁻CD49d⁺ surface markers (Bowman et al, 2016).

Primary mouse naïve CD8⁺ T cells and primary mouse naïve CD4⁺ T cells were isolated from the spleen and lymph nodes of 6‐week‐old BALB/c or C57BL/6 mice using a MagniSort™ Mouse CD8 Naïve T Cell Enrichment Kit (8804‐6825‐74, Thermo Fisher Scientific) or a MagniSort™ Mouse CD4 Naïve T Cell Enrichment Kit (8804‐6824‐74, Thermo Fisher Scientific).

To prepare activated CD8⁺ T cells, CD8⁺ T cells were harvested using an EasySep™ Mouse CD8⁺ T cell Isolation Kit (19853, Stem cell) on day 21 after C57BL/6 or BALB/c mice were systemically primed (day 0) and boosted (day 14) with LLC‐BrM cell lysates and H460 cells, respectively (1 × 10⁶ cells for day 0 and 1 × 10⁵ cells for day 14; Nucl‐Cyto‐Mem Preparation Kit, P1201, Applygen).

CD8⁺ T cells and CD4⁺ T cells were cultured in RPMI 1640 medium supplemented with 10% heat‐inactivated FBS, 2 mM L‐glutamine, P/S, and murine IL‐2 (402‐ML‐100, R&D). All cells tested negative for mycoplasma contamination.

Mice

Mice were housed under specific pathogen‐free (SPF) conditions in the animal facility of Tongji University. All animal experiments were approved by the Institutional Animal Care and Use Committee of Tongji University (approval number: TJBA01020105). BALB/c nude, C57BL/6, and BALB/c mice were purchased from the Shanghai SLAC Laboratory Animal Center (Shanghai, China).

H1299, H460, A549, H1650, HCC827, PC9, and H2030 cells were xenografted into 5‐ to 7‐week‐old male and female BALB/c nude mice. LLC‐BrM cells were isografted or allografted into 5‐ to 7‐week‐old male and female syngeneic C57BL/6 mice or BALB/c nude mice. LLC cells were isografted into 5‐ to 7‐week‐old male and female syngeneic C57BL/6 mice. 4T1 cells were isografted into 5‐ to 7‐week‐old female syngeneic BALB/c mice. The ratios of male‐to‐female mice are approximately 1:1.

Bioluminescence imaging

Mice were anesthetized and injected retro‐orbitally with 1.5 mg of D‐luciferin (LUCK‐1G, Gold Biotechnology) at the indicated times. Animals were imaged in a NightOWL II LB 983 chamber (Berthold Technologies, Bad Wildbad, Germany) within 5 min after D‐luciferin injection, and data were recorded using Indigo™ software (Berthold Technologies, Bad Wildbad, Germany). To measure metastasis, photon flux was calculated in a circular region of interest (ROI) encompassing the head of the mouse.

Brain metastasis

Cancer cells, stably expressing tdTomato–luciferase, were harvested by trypsinization, washed twice in PBS, and resuspended at 1 × 10⁶ (H1299‐Cas9 cells transduced with sgControl, sgIFITM1‐1, sgIFITM1‐2, sgIFITM3‐1, or sgIFITM3‐2), 1 × 10⁶ (LLC cells transduced with control or IFITM1), 3 × 10⁵ (LLC‐BrM cells transduced with control or IFITM1), 3 × 10⁵ (LLC‐BrM‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2), 1 × 10⁵ (LLC‐BrM cells transduced with control or IFITM1), 1 × 10⁵ (LLC‐BrM‐Cas9 cells transduced with control/sgControl, IFITM1/sgControl, IFITM1/sgC3‐1, IFITM1/sgC3‐3, sgControl/Control, sgIFITM1‐1/Control, sgIFITM1‐2/Control, sgIFITM1‐1/C3, or sgIFITM1‐2/C3), 1 × 10⁵ (LLC‐BrM‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2), 1 × 10⁵ (H460 cells transduced with control, IFITM1, IFITM3, or Tet‐On IFITM1), 1 × 10⁵ (H460‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2), or 3 × 10⁴ (LLC‐BrM‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2) cells in 100 μl of PBS, were injected into the left ventricle of mice. Bioluminescence imaging was used to verify successful injection and to monitor metastatic outgrowth.

In vivo microglia inactivation studies

To block microglial activation in vivo, PLX5622, BLZ945, or minocycline, specific inhibitors of microglia activation (Pyonteck et al, 2013; Prabhakara et al, 2018; Benbenishty et al, 2019), was used.

To block microglial activation in the brain metastasis experiments, PLX5622 (Biochempartner) was administered in AIN‐76A standard chow (1,200 PPM, Dyets) from day −3 to day 12, BLZ945 (200 mg/kg) was administered by oral gavage once daily from day −2 to day 5 and from day 10 to day 16, or minocycline (33 mg/kg) was administered intraperitoneally once every other day from day −1 to day 12.

To block microglial activation in the ELISA experiments, which were used to detect mouse IFNγ in the brain on day 1 after intracardiac injection of LLC‐BrM cells (1 × 10⁶) into BALB/c nude mice, PLX5622 (Biochempartner) was administered in AIN‐76A standard chow (1,200 PPM, Dyets) from day −3 to day 1; BLZ945 (200 mg/kg) was administered by oral gavage on days −2, −1, 0, and 1; or minocycline (33 mg/kg) was administered intraperitoneally on days −1, 0, and 1.

To confirm that PLX5622 selectively eliminated microglia with little effect on bone‐marrow‐derived macrophages after treatment for 15 days, 3 mice per treatment group were sacrificed and perfused with PBS through the left ventricle on day 12. The brains were collected, mechanically minced and digested, and subjected to FACS analysis to assess bone‐marrow‐derived macrophages (CD45⁺CD11b⁺Ly6C⁻Ly6G⁻CD49d⁺) and microglia (CD45⁺CD11b⁺Ly6C⁻Ly6G⁻CD49d⁻) in the brain.

In vivo peripheral macrophage depletion studies

To deplete peripheral macrophages in vivo, clodronate liposomes were used.

To confirm that clodronate liposomes selectively eliminated peripheral macrophages without inducing toxicity in microglia after treatment for 12 days, LLC‐BrM cells (1 × 10⁵ cells) were injected into the left ventricle of C57BL/6 mice on day 0. PBS liposomes or clodronate liposomes (100 μl/10 g) stained with DiD perchlorate (DiIC18(5), 40758ES25, Yeasen) were administered intraperitoneally on days −1, 3, 7, and 11. To obtain fluorescence images, one mouse per treatment group was sacrificed and perfused with PBS through the left ventricle on day 12. The spleen and brain were fixed in 4% PFA for 36 h at 4°C, dehydrated in 30% sucrose for 48 h, and embedded in OCT compound. Nonconsecutive sections (10 μm thick) were prepared using a Leica microtome and mounted using ProLong™ Gold Antifade Mountant with DAPI (P36931, Thermo Fisher Scientific). Simultaneously, three mice per treatment group were sacrificed and perfused with PBS through the left ventricle on day 12. The spleen and brain were collected, mechanically minced and digested, and subjected to FACS analysis to assess macrophages (CD11b^lowF4/80⁺) in the spleen and microglia (CD45^intCD11b⁺) in the brain.

To deplete peripheral macrophages in the brain metastasis experiments, PBS liposomes or clodronate liposomes (100 μl/10 g) were administered intraperitoneally on days −1, 3, 7, and 11.

To deplete peripheral macrophages in the ELISA experiments, which were used to detect mouse IFNγ in the brain on day 1 after the intracardiac injection of LLC‐BrM cells (1 × 10⁶) into BALB/c nude mice, PBS liposomes or clodronate liposomes (100 μl/10 g) were administered intraperitoneally on day −1.

Adoptive T cell transfer

The preparation of primary mouse T cells was described earlier in the Primary Cells section. In T cell adoptive transfer experiments, primary mouse naïve CD8⁺ T cells or naïve CD4⁺ T cells were injected intravenously at the indicated doses and time points.

In vivo CD8⁺ T cell depletion studies

Depletion of CD8⁺ T cells in vivo was confirmed through flow cytometric analysis of CD8⁺ T cells in spleens/lymph nodes and blood of C57BL/6 mice on days 1 and 13 after intravenous injection of rat IgG2b isotype control or an anti‐CD8α antibody (250 μg/mouse) on day 0.

To deplete CD8⁺ T cells in brain metastasis experiments, the anti‐CD8α antibody or rat IgG2b isotype control was injected intravenously at the indicated doses and time points.

Immune checkpoint blockade therapy

An anti‐PD‐1 antibody or rat IgG2a isotype control was injected intravenously at the indicated doses and time points.

Combination therapy with an oncolytic virus and immune checkpoint blockade

UV‐irradiated (365 nm for 48 h) oncolytic virus (Oncorine, Shanghai Sunway Biotech), an anti‐PD‐1 antibody, or rat IgG2a isotype control, alone or in combination as indicated, was injected intravenously at the indicated doses and time points.

Extravasation into the brain

LLC‐BrM cells (labeled with tdTomato and firefly luciferase) transduced with control or IFITM1 (1 × 10⁶ cells in 100 μl of PBS) were injected into the left ventricle of C57BL/6 mice. One day after injection, whole brains were collected, mechanically minced and digested, and subjected to FACS analysis to assess the presence of control and IFITM1‐overexpressing LLC‐BrM cells in the brain.

Immunostaining

For immunofluorescence staining of mouse brain sections, mice were sacrificed and perfused with PBS through the left ventricle. Brains were fixed in 4% PFA overnight at 4°C, dehydrated in 30% sucrose for 48 h, and embedded in OCT compound. Nonconsecutive sections (10 μm thick) were sliced with a Leica microtome, and immunofluorescence staining was performed with an anti‐tdTomato antibody by using a Tyramide Signal Amplification Kit (B40931, Thermo Fisher Scientific) to visualize tumor cells, an anti‐Iba‐1 antibody or an anti‐TMEM119 antibody to visualize microglia, an anti‐F4/80 antibody to visualize microglia and macrophages, an anti‐CD8α antibody to visualize CD8⁺ T cells, an anti‐IFNγ antibody, an anti‐C3 antibody, and an anti‐cleaved caspase 3 antibody. Immunoreactions were detected with fluorescently labeled secondary antibodies (Thermo Fisher Scientific). Sections were mounted using ProLong™ Gold Antifade Mountant with DAPI (P36931, Thermo Fisher Scientific). At least, 10 random fields were counted in every section at 400× magnification (three sections per brain). For day 7 in Figs 2E and 3G, and day 20 in Fig 5M, we set the ROI as the entire area of the metastatic lesion. For day 1 in Figs 2E and 3G, we set the ROI as the area of a single metastatic cell. ImageJ was used to quantify the immunofluorescence intensity in the ROI. In Fig 2A, macrophages are TMEM119⁻ F4/80⁺ cells (green⁻ blue⁺ cells). Microglia are TMEM119⁺ F4/80⁻ cells (green⁺ blue⁻ cells) and TMEM119⁺ F4/80⁺ cells (green⁺ blue⁺ cells). The green⁻ blue⁺ cells (macrophages) had to be calculated by using the blue fluorescent signal (macrophages and microglia) to deduct the green fluorescent signal (microglia) because green⁻ blue⁺ cells (macrophages) cannot be quantified directly. The microscopic field was an area of 62.5 μm × 62.5 μm around a single metastatic cell (day 1) or a metastatic lesion (day 7). In Fig 5M, the positively stained areas of single CD8⁺ T cells, clustered CD8⁺ T cells, cleaved caspase 3⁺ cancer cells, and tdTomato⁺ cancer cells in the ROI were calculated with ImageJ. The ratios of the areas between clustered and single CD8⁺ T cells are shown as the numbers of CD8⁺ T cells in the ROI. The ratios of the areas between cleaved caspase 3⁺ cancer cells and tdTomato⁺ cancer cells are shown as the percentages of cleaved caspase 3⁺ cancer cells in the ROI.

To assess MHC‐I expression in brain metastatic lesions, C57BL/6 mice were systemically primed (day −21) and boosted (day −7) with homologous LLC‐BrM cell lysates (1 × 10⁶ cells for day −21 and 1 × 10⁵ cells for day −7) before injection of LLC‐BrM‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2 (3 × 10⁵ cells in 100 μl of PBS) into the left ventricle on day 0. Then, an anti‐CD8α antibody or rat IgG2b isotype control was injected intravenously at a dose of 250 μg/mouse (in 100 μl of PBS) on days 3 and 5. Mice were sacrificed on day 20. Immunohistochemical staining for mouse MHC‐I was performed on paraffin‐embedded brain sections with the VECTASTAIN ABC HRP Kit (PK6100, Vector Labs) and DAB (D5905, Sigma). Immunohistochemical staining was performed with a primary antibody specific for MHC‐I followed by biotinylated goat anti‐rat IgG. Cas9‐Flag immunohistochemical staining was performed on brain sections with a VECTASTAIN ABC‐AP Kit (AK‐5000, Vector Labs) and a BCIP/NBT Substrate Kit (SK‐5400, Vector Labs). Immunohistochemical staining was performed with a primary antibody specific for Flag followed by biotinylated goat anti‐rabbit IgG and a Tyramide Signal Amplification Kit (B40931, Thermo Fisher Scientific). The MHC‐I staining intensity (H‐score) was calculated with the formula 3 × percentage of strongly stained cells + 2 × percentage of moderately stained cells + percentage of weakly stained cells and ranged between 0 and 300. All metastatic lesions were counted in every section at 400× magnification. The representative images of strongly, moderately, and weakly stained cells are shown in Appendix Fig S6E.

For assessment of IFITM1 and MHC‐I colocalization, H1299 cells cultured on coverslips were stimulated with 1,000 U/ml IFNγ for 48 h in medium and were then incubated with an anti‐MHC‐I antibody on ice for 1 h in ice‐cold PBS/BSA. Cells were fixed with 4% PFA for 10 min at 4°C and permeabilized with 0.2% Triton X‐100 for 20 min at room temperature. Cells were incubated first with anti‐IFITM1 and anti‐MHC‐I antibodies overnight at 4°C in blocking solution and then with Alexa Fluor™ 488 goat anti‐rabbit IgG (H + L) and Alexa Fluor™ 568 goat anti‐mouse IgG (H + L) secondary antibodies for 1 h at room temperature in blocking solution to detect the anti‐MHC‐I and anti‐IFITM1 primary antibodies. Coverslips were mounted using ProLong™ Gold Antifade Mountant with DAPI (P36931, Thermo Fisher Scientific) and imaged with a Nikon A1R confocal microscope. One hundred cells were analyzed for each sample.

Mouse cytokine array analysis

For mouse cytokine array analysis of brain homogenates, 4 T1 cells (1 × 10⁶ cells in 100 μl of PBS) were injected into the left ventricle of BALB/c mice. Two days after injection, whole brains were collected and were mechanically minced and homogenized in ice‐cold PBS supplemented with protease inhibitors at a 10:1 ratio (PBS volume: tissue weight). Triton X‐100 was then added to a final concentration of 1%. After one freeze–thaw cycle at −70°C, the homogenates were centrifuged at 10,000 g for 5 min to remove cellular debris.

For mouse cytokine array analysis of conditioned medium, cell culture supernatants from cocultures of BV‐2 (3 × 10⁵) and H460 (6 × 10⁴) cells were centrifuged at 8,609 g for 5 min to remove cellular debris. Protein concentrations were measured with an Enhanced BCA Protein Assay Kit (P0010, Beyotime).

The homogenates and cell culture supernatants were analyzed using Mouse Cytokine Array Panel A (40 different cytokines, ARY006, R&D) according to the manufacturer's instructions.

ELISA

To determine the levels of mouse IFNα, IFNβ, IFNγ, IFNλ2/3, and TNFα in the brain, C57BL/6 or BALB/c nude mice were injected in the left ventricle with LLC‐BrM cells (1 × 10⁶ in 100 μl of PBS); H460 cells (1 × 10⁶ in 100 μl of PBS); LLC‐BrM‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2 (1 × 10⁶ in 100 μl of PBS); LLC‐BrM‐Cas9 cells transduced with control, sgC3‐1, sgC3‐3, IFITM1, IFITM1 and sgC3‐1 together, or IFITM1 and sgC3‐3 together (1 × 10⁶ in 100 μl of PBS); LLC‐BrM cells transduced with control or C3 (1 × 10⁶ in 100 μl of PBS); H460 cells transduced with control, shC3‐3, shC3‐6, IFITM1, IFITM1 and shC3‐3 together, or IFITM1 and shC3‐6 together (1 × 10⁶ in 100 μl of PBS); and H460 cells transduced with control or C3 (1 × 10⁶ in 100 μl of PBS). Mice were sacrificed at the indicated times. Whole brains were collected and mechanically minced and homogenized in ice‐cold PBS supplemented with protease inhibitors at a 10:1 ratio (PBS volume: tissue weight). Triton X‐100 was then added to a final concentration of 1%. The homogenates were centrifuged at 4,000 g for 10 min to remove cellular debris.

To measure the levels of mouse TNFα or IFNγ in conditioned media, cell culture supernatants were centrifuged at 10,000 g for 5 min to remove cellular debris.

The brain homogenates and conditioned medium samples were analyzed with a VeriKine™ Mouse Interferon Alpha ELISA Kit (421201, PBL Assay Science), VeriKine™ Mouse Interferon Beta ELISA Kit (424001, PBL Assay Science), Mouse IFN‐gamma Quantikine ELISA Kit (MIF00, R&D), Mouse IL‐28 Platinum (Interferon Lambda 2/3) ELISA Kit (BMS6028, eBioscience), and Mouse TNFα ELISA Kit (EMC102a.96, NeoBioscience) according to the manufacturer's instructions.

Conditioned medium

To obtain conditioned medium, H460 cells were transduced with control, IFITM1, IFITM1 and shC3‐3 together, IFITM1 and shC3‐6 together, or C3; H460 cells with shControl, shC3‐3, or shC3‐6; LLC‐BrM cells with shControl, shC3‐2, shC3‐3, Control, IFITM1, or C3; and LLC‐BrM‐Cas9 cells with sgControl, sgIFITM1‐1, or sgIFITM1‐2. The transduced cells were seeded in 10 cm dishes (1 × 10⁶ cells/dish) and subsequently allowed to adhere and grow for 24 h. Then, the cells were incubated for 48 h in serum‐free medium to produce conditioned medium. The supernatants were collected and passed through 0.45 μm syringe filter units (SLHV033RB, Millipore).

To obtain fractions of conditioned medium containing molecules of different molecular weights, spin filtration was performed according to the manufacturer's instructions (Millipore). Conditioned medium was applied to the 100 K filter (UFC510096, Millipore) and centrifuged at 14,000 g for 5 min, the filtrate was applied to a 50 K filter (UFC505096, Millipore) and centrifuged at 14,000 g for 5 min, and this filtrate was further applied to a 10 K filter (UFC501096, Millipore) and centrifuged at 14,000 g for 5 min. The resulting supernatants at each step were collected by reverse filtration into a new vial.

Conditioned medium and conditioned medium fractions containing molecules of different molecular weights were heated at the indicated temperatures for 10 min for further analysis.

Mass spectrometry

The > 100 kDa fraction of conditioned medium from H460, H1299, H2030, or H292 cells (1 × 10⁶ cells for each cell line) were separated on a NuPAGE 10% Bis‐Tris Gel (NP0341BOX, Invitrogen) and visualized using EZBlue™ Gel Staining Reagent (G1041‐500ML, Sigma). The entire gel lanes were excised and subjected to ESI‐LC–MS/MS analysis.

To identify proteins associated with endogenous membrane IFITM1, H1299 cells (4 × 10⁷) were treated with 1,000 U/ml human IFNγ (300‐02, Peprotech) at 37°C. After 48 h, the cell membrane was prepared using a Nucl‐Cyto‐Mem Preparation Kit (P1201, Applygen). A total of 13.95 mg of cell membrane proteins was incubated with 30 μg of an anti‐IFITM1 mouse monoclonal antibody (mAb) or 30 μg of mouse IgG isotype control for 2 h at 4°C and was then incubated with 40 μl of Protein G Plus Agarose (22851, Pierce) overnight. Immunoprecipitates were separated on a NuPAGE 10% Bis‐Tris Gel (NP0341BOX, Invitrogen) and visualized using EZBlue™ Gel Staining Reagent (G1041‐500ML, Sigma). The entire gel lanes were excised (excluding the IgG bands) and subjected to ESI‐LC–MS/MS analysis.

In vitro microglia, BMDMs, BrM‐BMDMs, astrocyte, and T cell killing assay

For contact coculture of microglia, BMDMs, BrM‐BMDMs, or astrocytes with cancer cells, primary mouse microglial, BMDMs, BrM‐BMDMs, or astrocyte cells were seeded at 6 × 10⁴, 6 × 10⁴, 6 × 10⁴, or 6 × 10⁴ cells/well in six‐well plates and subsequently allowed to adhere and grow for 48 h. Then, H460 cells transduced with control or IFITM1; H1299‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2; LLC‐BrM cells transduced with Control or IFITM1; and LLC‐BrM‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2 were added to each well (6 × 10⁴ cells/well), and maintained in contact coculture for 6 h.

For conditioned medium stimulation before contact coculture of microglia with cancer cells, primary mouse microglia (6 × 10⁴ cells/well) were cultured in DMEM‐HG supplemented with 10% heat‐inactivated FBS, 10% heat‐inactivated horse serum (06050‐114, Gibco), 2 mM L‐glutamine, and P/S for 48 h and were subsequently stimulated with conditioned medium from the indicated cells for 12 h. Then, H1299 or LLC‐BrM cells (6 × 10⁴) were added to each well and were maintained in contact coculture in DMEM‐HG supplemented with 10% heat‐inactivated FBS, 10% heat‐inactivated horse serum, 2 mM L‐glutamine, and P/S for 6 h.

For noncontact coculture of microglia, BMDMs, or BrM‐BMDMs with cancer cells, primary mouse microglia, BMDMs, or BrM‐BMDMs (1.8 × 10⁴ microglia/well for LLC‐BrM cells, or 6 × 10⁴ microglia/well for H460 cells) were cultured on the bottom of a six‐well plate for 48 h. LLC‐BrM or H460 cells transduced with control or IFITM1 (6 × 10⁴ cells/transwell insert) were cultured on the membrane of transwell cell culture inserts (0.4 μm, 140640, Thermo Fisher Scientific) in medium supplemented with rat IgG isotype control (30 μg/ml), anti‐IFNγ antibodies (10 μg/ml), anti‐TNFα antibodies (30 μg/ml), or L‐NMMA (iNOS inhibitor, 10 μM, M7033, Sigma) for 48 h.

For contact coculture of T cells with cancer cells, LLC‐BrM cells transduced with control or IFITM1; H460 cells transduced with control or IFITM1; H1299‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2; or H1299‐Cas9‐sgIFITM1 cells transduced with control, IFITM1, IFITM2, or IFITM3 were cocultured with naïve CD8⁺ T cells, naïve CD4⁺ T cells, or activated CD8⁺ T cells (3 × 10⁵) in medium supplemented with murine IL‐2 (30 U/ml, 402‐ML‐100, R&D) for 48 h.

For coculture of microglia and T cells with cancer cells, primary mouse microglia (1.8 × 10⁴ microglia/well for LLC‐BrM cells, or 6 × 10⁴ microglia/well for H460 cells) were cultured on the bottom of a six‐well plate for 48 h. Subsequently, LLC‐BrM or H460 cells transduced with control or IFITM1 (6 × 10⁴ cells/transwell insert) were cultured on the membrane of transwell cell culture inserts (0.4 μm, 140640, Thermo Fisher Scientific). After 6 h, naïve CD8⁺ T cells (3 × 10⁵ cells/transwell insert) were added into transwell cell culture inserts and were cocultured in medium supplemented with murine IL‐2 for 48 h.

To assess the synergistic effect of IFITM1 and immune checkpoint blockade therapy in vitro, LLC‐BrM or H460 cells transduced with control or IFITM1 were seeded at 6 × 10⁴ cells/well in six‐well plates and subsequently allowed to adhere and grow for 24 h. Then, activated CD8⁺ T cells (3 × 10⁵ cells/well) were added to each well in medium supplemented with an anti‐PD‐1 antibody or rat IgG2a isotype control for 48 h of contact coculture. To prepare activated CD8⁺ T cells, CD8⁺ T cells were harvested using the EasySep™ Mouse CD8⁺ T Cell Isolation Kit (19853, Stem cell) on day 21 after C57BL/6 or BALB/c mice were systemically primed (day 0) and boosted (day 14) with LLC‐BrM cell lysates (1 × 10⁶ cells for day 0 and 1 × 10⁵ cells for day 14; Nucl‐Cyto‐Mem Preparation Kit, P1201, Applygen) and H460 cells (1 × 10⁶ cells for day 0 and 1 × 10⁵ cells for day 14), respectively.

Cancer cells cultured in the above coculture systems were analyzed using an Annexin V‐Alexa Fluor 647/7‐AAD Apoptosis Detection Kit (40309ES60, YEASEN) according to the manufacturer's instructions. Flow cytometry experiments were performed using the FACSVerse flow cytometer (BD Biosciences), and flow cytometry data were analyzed with FlowJo software.

Phagocytosis assay

For the in vivo microglia phagocytosis assay, violet‐labeled (C34571, Thermo Fisher Scientific) LLC‐BrM‐Cas9 cells transduced with sgControl, sgIFITM1‐1, or sgIFITM1‐2 (1 × 10⁶ cells in 100 μl of PBS) were injected into the left ventricle of BALB/c nude mice. After 24 h, whole brains were collected and were mechanically minced and digested. Myelin and debris were depleted by Percoll gradient centrifugation. The violet‐labeled LLC‐BrM cells phagocytosed by microglia were identified by double‐positive staining for CD11b and violet using flow cytometric analysis.

For the in vitro microglia phagocytosis assay, LLC‐BrM cells were transiently transfected with control or mouse IFITM1 plasmid and H1299 cells were transiently transfected with control or human IFITM1 plasmid using MACSfectin™ reagent (130‐098‐411, Miltenyi Biotec). After 48 h, these LLC‐BrM and H1299 cells (1 × 10⁵) were stained with DDAO (C34564, Thermo Fisher Scientific) and subsequently cocultured with CFSE‐stained (C34554, Thermo Fisher Scientific) primary mouse microglia (1 × 10⁵), violet‐stained (C34571, Thermo Fisher Scientific) primary mouse BMDMs (1 × 10⁵), or violet‐stained (C34571, Thermo Fisher Scientific) primary mouse BrM‐BMDMs (1 × 10⁵), and analyzed by flow cytometry after 1 (LLC‐BrM) or 1.5 h (H1299). Flow cytometry experiments were performed using the CytoFLEX LX flow cytometer (Beckman Counter) or the FACSVerse flow cytometer (BD Biosciences), and flow cytometry data were analyzed with FlowJo software.

Co‐IP assay

To confirm the interaction between endogenous membrane IFITM1 and MHC‐I, H1299 cells were lysed in 1% Triton X‐100/TBS (10 mM Tris–HCl pH 7.4, 0.9% NaCl, 0.02% KCl) supplemented with Na₃VO₄, phosphatase inhibitor cocktail (B15001, Bimake), and protease inhibitors (539134, Merck). To immmunoprecipitate IFITM1, total lysates containing 1 mg protein were incubated with 7.5 μg of anti‐IFITM1 mouse monoclonal antibody for 2 h at 4°C and were then incubated with 10 μl of Protein G Plus Agarose (22851, Pierce) overnight. Immunoprecipitates and total lysates were subjected to immunoblotting with the indicated antibodies.

Analysis of protein expression

For FACS analysis, the single‐cell suspension was incubated first with the indicated primary antibodies at 4°C for 30 min, subsequently incubated with the secondary antibodies at room temperature for 30 min, and then analyzed by flow cytometry to evaluate cell‐membrane‐localized proteins. The single‐cell suspension was fixed in 1% PFA at 37°C for 30 min, subsequently treated with 0.1% TritonX‐100 at room temperature for 5 min, and then analyzed by flow cytometry using the indicated primary and secondary antibodies to evaluate total protein.

To examine the cell membrane dynamics of MHC‐I, H1299‐sgIFITM1 cells transduced with control or IFITM1 were treated with cycloheximide (300 μg/ml, A8244, APExBIO) for the indicated times and were then subjected to flow cytometric analysis using the anti‐MHC‐I mouse monoclonal antibody.

Flow cytometry experiments were performed using the FACSVerse flow cytometer (BD Biosciences), and flow cytometry data were analyzed with FlowJo software.

For immunoblotting, cells were lysed in RIPA buffer (50 mM Tris–HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% TritonX‐100, 1% sodium deoxycholate, and 0.1% SDS) supplemented with 1 mM Na₃VO₄, phosphatase inhibitor cocktail (B15001, Bimake), and protease inhibitors (539134, Merck) unless otherwise noted. Protein concentrations were measured with the Enhanced BCA Protein Assay Kit (P0010, Beyotime).

For immunoblotting of IFITM1, cells were lysed in 1% Triton X‐100/TBS (10 mM Tris–HCl pH 7.4, 0.9% NaCl, and 0.02% KCl) supplemented with Na₃VO₄, phosphatase inhibitor cocktail, and protease inhibitors. Protein concentrations were measured with the Enhanced BCA Protein Assay Kit.

Analysis of mRNA expression

Total RNA was extracted using an RNAprep Pure Cell/Bacteria Kit (DP430, TIANGEN) and reverse transcribed with the ReverTra Ace qPCR RT Kit (FSQ‐101, TOYOBO). An amount of cDNA corresponding to approximately 10 ng of starting RNA was used for one reaction. qPCR was performed with a TaqMan Gene Expression Assay (Applied Biosystems) or 2 × SYBR Green qPCR Master Mix (Low ROX; B21702, Bimake). All quantities were normalized to endogenous β‐actin. Experiments were performed on the Applied Biosystems 7500/7500 Fast instrument. Primers used to amplify genes and TaqMan gene expression assays are listed in Appendix Table S7.

T7 endonuclease I assay

Genomic DNA was extracted from H1299 cells using EasyPure® Genomic DNA Kit (EE101‐02, Transgen Biotech). The genomic region flanking the CRISPR target site was PCR amplified. The PCR products were denatured and re‐annealed. Subsequently, hybridized PCR products were treated with T7 endonuclease I (M0302L, NEB), and the digestions were analyzed by agarose gel electrophoresis. The cleavage bands were quantified using ImageJ. Primers used to amplify mouse IFITM1 are listed in Appendix Table S7.

Matrigel invasion assay

H1299‐Cas9 cells transduced with sgControl or sgIFITM1 (1 × 10⁵ cells) were cultured on the membrane of transwell cell culture inserts coated with 100 μg/well Matrigel (356237, Corning) in serum‐free medium. After a 6‐h incubation in wells containing SFM + 10% FBS, the cells and gel in the upper compartment of the insert were removed by wiping the upper side of the membrane with a cotton swab. Then, the inserts were fixed in 4% PFA for 10 min at room temperature and stained with crystal violet (C0121, Beyotime). Eight random fields were counted for every insert at 400× magnification (3 inserts per group).

CCK‐8 assay

H1299‐Cas9 cells transduced with sgControl or sgIFITM1 (1 × 10³ cells) were seeded in 96‐well plates and cultured for 6 days, or H460 cells transduced with control or IFITM1 (1 × 10³) were seeded in 96‐well plates and cultured for 5 days. At the indicated time points, a CCK‐8 solution was added to each well and incubated for 1 h at 37°C. Absorbance was measured at 450 nm using a spectrophotometer (Infinite M200 Pro, TECAN).

Tumor sphere formation assay

H1299‐Cas9 cells transduced with sgControl or sgIFITM1 (300 cells) were seeded in 24‐well ultralow attachment plates (Corning) and cultured for 7 days in serum‐free MEGM (Lonza) supplemented with 1:50 B27 (Thermo Fisher Scientific), 20 ng/ml EGF (BD Biosciences), 20 ng/ml bFGF (BD Biosciences), and 4 μg/ml heparin (Sigma). H460 cells transduced with control or IFITM1 (300 cells) were seeded in 24‐well ultralow attachment plates (Corning) and cultured for 7 days in serum‐free MEGM (Lonza) supplemented with 1:50 B27 (Thermo Fisher Scientific), 20 ng/ml EGF (BD Biosciences), 20 ng/ml bFGF (BD Biosciences), and 4 μg/ml heparin (Sigma). All tumor spheres in each well were counted.

Luciferase reporter assay

For the luciferase reporter assay to detect the activity of STAT1 or NF‐κB, H1299 cells were transiently cotransfected with the STAT1 or NF‐κB firefly luciferase reporter construct and the Renilla luciferase construct. Forty‐eight hours after transfection, the relative transcriptional activity was measured using a Dual Luciferase Assay Kit (E2940, Promega).

The human IFNγ reporter gene construct (pSTAT1‐Luc) was purchased from Shanghai Model Organisms Center, and the human TNFα reporter gene construct (pNF‐κB‐Luc) was generously provided by Dr. Ping Wang (Tongji University, China). pGL3 Control (E1741) and pRL Renilla luciferase control reporter vectors (E2261) were purchased from Promega.

Plasmids

cDNAs encoding full‐length human IFITM1, IFITM2, IFITM3, C3, and mouse IFITM1 and C3 were cloned and sequenced. Flag‐IFITM1, Flag‐IFITM2, and Flag‐IFITM3 were subcloned into the pBabe‐EGFP or pQCXIP (Clontech) retroviral vectors and verified by sequencing. C3‐Flag (human) was subcloned into the pQCXIP or pQCXIN (Clontech) retroviral vectors and verified by sequencing. C3‐Flag (mouse) was subcloned into the pBabe‐hygro retroviral vector. The lentiviral vector used for doxycycline‐inducible expression of IFITM1 was generated by subcloning the corresponding cDNA into the pCW vector (Gao et al, 2016a).

Constructs encoding shRNAs against human C3 (#3, TRCN0000057140 and #6, TRCN0000429898) and mouse C3 (#2, TRCN0000334476 and #3, TRCN0000334557) were generated by cloning the corresponding short hairpin RNA sequences into the pLKO.1 vector. siRNA SMARTpools targeting human IFITM1 (L‐019543‐00‐0005) and nontargeting control siRNA (D‐001810‐10‐05) were purchased from Dharmacon.

To construct the sgRNA expression vector, each 20 bp guide sequence was cloned into the lentiGuide‐Puro vector, which was a gift from Feng Zhang (Addgene plasmid #52963; http://n2t.net/addgene:52963; RRID: Addgene_52963); except for the guide sequence for human IFITM1 #2, which was generated using an online tool (http://crispr.mit.edu/). All guide sequences were taken from the human or mouse GeCKO v2 CRISPR knockout library and are listed in Appendix Table S8.

Populations of overexpression, knockdown, or knockout cells were used, and the efficiencies of overexpression, knockdown, or knockout were verified by qPCR or western blotting.

Antibodies

All the antibodies and their application were listed in Appendix Table S9.

Acquisition of gene expression and clinical data

For the lung cancer dataset GSE4573, the MAS5‐calculated signal intensity matrices and clinical information were downloaded from the Gene Expression Omnibus (GEO) website. The gene expression values for each patient were log2 scaled and transformed to z‐scores. The R package “JetSet” was used to select the best probe to represent the gene.

The RNA‐seq data matrices and clinical information for the TCGA cancer datasets lung adenocarcinoma (LUAD) and small cell lung cancer (SCLC, EGAS00001000925; George et al, 2015) were downloaded from cBioPortal (https://github.com/cBioPortal/datahub/tree/master/public, http://www.cbioportal.org/data_sets.jsp). Clinical information for each patient with an available survival time was collected for further analysis in this study. The gene expression values for each patient were log2 scaled and transformed to z‐scores.

For C3AR1 mRNA expression in healthy human microglia and BMDMs, RNA‐seq count data were downloaded from https://joycelab.shinyapps.io/braintime/ (Klemm et al, 2020).

IFITM1 mRNA expression in the primary tumors and brain metastatic lesions of lung and breast cancer patients

For lung cancer patients, publicly available microarray databases were downloaded from the National Center for Biotechnology Information (NCBI) GEO database. The cases of lung adenocarcinoma (LUAD) in the total sample were used since all brain metastatic lesions originate from primary lung adenocarcinoma. The redundant cases were excluded, and the average results for technical replicates were used.

We collected 18 and 8 lung cancer datasets from the GPL570 platform (Affymetrix Human Genome U133 Plus 2.0 Array) and GPL96 platform (Affymetrix Human Genome U133A Array), respectively.

For each platform, gene expression summarization was performed by normalizing the data in each raw CEL file with MAS5 algorithm in the R statistical environment (www.r‐project.org) using the affy Bioconductor library.

Details about the datasets can be found in Appendix Tables S10 and S11.

Survival analysis

A univariate Cox proportional hazards regression model (implemented with the “coxph” function in the “survival” package in R) was used to calculate the regression coefficient of each candidate gene with the overall survival time and status of patients in the LUAD dataset.

To validate the performance of the features (IFITM1 and MHC‐I) as molecular markers for prognosis, we defined the risk score as follows. When only one gene was used as the feature, the gene expression level was treated as a risk score. When IFITM1 and MHC‐I were considered as a comprehensive feature, the risk score was defined as a linear combination of their gene expression levels weighted by their estimated regression coefficients.

$${\sum }_{i=1}^n{\text{Value}}_i\times {\text{Weight}}_i$$

where value is the gene expression level, weight is the regression coefficient calculated as described above, and i is the component gene of the feature.

To confirm the suitable risk score cutoff for each cancer type, patients were assigned to a low‐risk group and a high‐risk group based on their risk score (the cutoff value: 0.9052606 for Fig 1N, 1.682972 for Appendix Fig S2E, and 1.2328035 for Appendix Fig S2F). We evaluated statistical significance by the log‐rank test and Kaplan–Meier analysis (implemented with the “survdiff” function in the “survival” package in R). The results were plotted using Kaplan–Meier survival curves.

Correlation analysis

We classified patients into two groups based on their IFITM1 expression level (high‐ and low‐expression groups). The cutoff was the same value calculated as described above. The correlation between IFITM1 and C3 was assessed by Wilcoxon rank‐sum test in R.

Gene set enrichment analysis (GSEA)

GSEA was performed using the GSEA platform (GSEA v3.0). The IFNγ pathway gene list was obtained from a published study (Gao et al, 2016b). The patients were classified by their IFITM1 expression level (high and low expression groups) as performed for the survival analysis. The default values were used, except that the metric for ranking genes was set to “t test” and the permutation type was set to “gene set”.

Response to immune checkpoint blockade therapy

The clinical and IFITM1 expression data for melanoma patients treated with an anti‐PD‐1 antibody (nivolumab or pembrolizumab) alone or in combination with an anti‐CTLA‐4 antibody (ipilimumab), for metastatic gastric cancer patients treated with an anti‐PD‐1 antibody (pembrolizumab), and for urothelial bladder cancer patients treated with an anti‐PD‐L1 antibody (atezolizumab) were obtained from published studies (Kim et al, 2018; Mariathasan et al, 2018; Gide et al, 2019). The IFITM1 expression values were log2 transformed and normalized using the R package “preprocessCore:normalize.quantiles”. The patients were classified as responders and nonresponders.

For the melanoma and metastatic gastric cancer patients, patients achieving a complete response (CR) or a partial response (PR) were considered responders, and patients with stable disease (SD) or progressive disease (PD) were considered nonresponders. Statistical analysis was performed with the Wilcoxon rank‐sum test using R.

For the urothelial bladder cancer patients, patients with CR, PR, and SD were considered responders, and patients with PD were considered nonresponders. The patients were divided into the low (< median, n = 149) and high (> median, n = 149) IFITM1 expression groups. Statistical analysis was performed with Pearson's chi‐squared test using GraphPad Prism (version 7.04).

Statistical analysis

Data analyses and figure plotting were performed in GraphPad Prism (version 7.04) or in the R statistical environment (www.r‐project.org). The numbers of replicates and independent experiments are listed in the figure legends. Values are presented as the mean ± s.e.m. values unless stated otherwise. The box plots show the distribution of the data. The boxes indicate the 25^th to 75^th percentiles. The horizontal lines within the boxes indicate the median levels, the lower whiskers indicate the smallest observation greater than or equal to the lower hinge − 1.5 × IQR, and the upper whiskers indicate the largest observation less than or equal to the upper hinge + 1.5 × IQR. The IQR is the interquartile range, or the distance between the first and third quartiles. A P‐value of < 0.05 was considered to indicate statistical significance for all analyses. No statistical methods were used to predetermined sample sizes, and the experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessments.

Author contributions

Hua Gao: Conceptualization; supervision; funding acquisition; writing – original draft; project administration; writing – review and editing. Xiaofei She: Data curation; formal analysis; investigation; writing – original draft; writing – review and editing. Shijun Shen: Data curation; software; formal analysis; writing – original draft. Guang Chen: Data curation; formal analysis; investigation. Yaqun Gao: Data curation; formal analysis; investigation. Junxian Ma: Formal analysis; investigation. Yaohui Gao: Data curation; investigation. Yingdi Liu: Data curation; investigation. Guoli Gao: Investigation. Yan Zhao: Investigation. Chunyan Wang: Investigation. Cizhong Jiang: Software; formal analysis; supervision; funding acquisition; writing – original draft. Ping Wang: Funding acquisition. Huanlong Qin: Funding acquisition.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Source Data for Appendix

Source Data for Figure 1

Source Data for Figure 2

Source Data for Figure 3

Source Data for Figure 4

Source Data for Figure 5

The EMBO Journal (2023) 42: e111112

Data availability

The mass spectrometry proteomics data produced in this study are available in the following databases:

The data of the proteins identified in the > 100 kDa fraction of conditioned media from all four cell lines: PRIDE PXD038129 (http://www.ebi.ac.uk/pride/archive/projects/PXD038129).

The data of the coimmunoprecipitated proteins associated with the endogenous membrane IFITM1 of H1299 cells after IFNγ stimulation: PRIDE PXD038130 (http://www.ebi.ac.uk/pride/archive/projects/PXD038130).