The lncRNA EPIC1 suppresses dsRNA-induced type I IFN signaling and is a therapeutic target to enhance TNBC response to PD-1 inhibition

Abstract

Increases in retroelement-derived double-stranded RNAs (dsRNAs) in various types of cancer cells facilitate the activation of antitumor immune responses. The long noncoding RNA EPIC1 interacts with the histone methyltransferase EZH2 and contributes to tumor immune evasion. Here, we found that EPIC1 in tumor cells suppressed cytoplasmic dsRNA accumulation, type I interferon (IFN) responses, and antitumor immunity. In various cancer cell lines, knockdown of EPIC1 stimulated the production of dsRNA from retroelements and an antiviral-like type I IFN response that activated immune cells. EPIC1 inhibited the expression of LINE, SINE, and LTR retroelements that were also repressed by EZH2, suggesting a potential role for the EPIC1-EZH2 interaction in regulating dsRNA production. In a humanized mouse model, in vivo delivery of EPIC1-targeting oligonucleotides enhanced dsRNA accumulation in breast cancer xenografts, reduced tumor growth, and increased the infiltration of T cells and inflammatory macrophages into tumors. Furthermore, EPIC1 knockdown improved the therapeutic efficacy of the PD-1 inhibitor pembrolizumab, a PD-1 inhibitor, in the humanized mouse model. Together, our findings establish EPIC1 as a key regulator of dsRNA-mediated type I IFN responses and highlight its potential as a therapeutic target to improve the efficacy of immunotherapy.

INTRODUCTION

Immunotherapy, which harnesses patients’ immune systems to fight against tumors, has shown remarkable success in cancer treatment (1). However, in many cancers, the overall patient eligibility for immunotherapy and response rate to the therapy remain low (2, 3). Adaptive immunotherapy resistance has also emerged as a major obstacle to achieve durable and long-lasting treatment outcomes (2, 4). There is an urgent need for identifying new, effective biomarkers and targets that can predict and overcome immunotherapy resistance in clinical settings (5).

Retroelements (REs) are abundant genomic components that make up more than half of the human genome. (6). REs are sorted into several subclasses, such as SINEs (Alu elements, for example), LINEs (L1, for example), retrotransposons [non-viral long terminal repeats (LTRs) and endogenous retroviruses (ERVs), for example] (6, 7). Many essential cellular functions such as gene regulation and maintenance of genome stability have been reported to rely on REs. Deregulated expression of double-stranded RNA (dsRNA) inducing REs can cause cellular stress that may eventually activate the innate immune response (8–11). Several reports have shown that RE-induced immune responses can cause both favorable and unfavorable cellular events (8–10, 12–15). On one hand, intrinsic cellular overexpression of REs can induce a hyperinflammatory response, which has been implicated as a major cause of many genetic disorders (10, 12–14). On the other hand, RE-derived dsRNA overexpression in tumor cells can facilitate activation of the antitumor immune response in multiple cancer types (9, 15–18). Because of the regulatory role of REs in the antitumor immune response, their dysregulation in cancer has been explored as a potential antitumor therapeutic target, especially when combined with immunotherapy (18–20). Several epigenetic regulators, such as the histone methyltransferase enhancer of zeste homolog 2 (EZH2) and DNA methyltransferase (DNMTs), have been characterized as master regulators of RE expression (19, 21–24). It is important to understand how these epigenetic regulators specifically recognize and regulate RE expression.

Multiple studies have highlighted the pivotal roles of noncoding RNAs in epigenetic regulation in cancer (25–27). Long noncoding RNAs (lncRNAs) can act as molecular scaffolds, bridging epigenetic regulators with their target loci to facilitate specific binding and regulation (25–28). Our previous studies indicated that epigenetically-induced lncRNA1 (EPIC1) can interact with EZH2 and the transcription factor MYC to promote tumorigenesis and enable various types of cancer cells to evade immune surveillance through suppressing antigen presentation and type II interferon signaling (26, 28). EPIC1 is an intergenic lncRNA-encoding gene located on chromosome 22. Compared with adjacent normal tissues, EPIC1 is frequently epigenetically activated by loss of DNA methylation at its promoter in multiple cancer types, including breast cancer, melanoma, and lung cancer.

Here, we further investigated the mechanism by which EPIC1 modulates immune function with a specific focus on its regulation of RE expression and dsRNA accumulation in tumor cells. We found that suppression of EPIC1 substantially induced immunogenic RE expression and cytoplasmic dsRNA accumulation, leading to the activation of type I interferon signaling. We observed that EPIC1-mediated suppression of type I interferon signaling contributed to an immunosuppressive tumor microenvironment (TME) in multiple types of cancer cells. Finally, using an hCD34⁺ humanized mouse model, we demonstrated that targeting EPIC1 sensitized triple-negative breast cancer (TNBC) xenograft tumors to pembrolizumab (anti-PD-1) treatment.

RESULTS

EPIC1 suppresses type I interferon signaling in multiple cancer cell lines

Our previous studies revealed that the lncRNA EPIC1 can increase the expression of cell cycle genes to promote tumor growth while suppressing antigen presentation genes to evade immune surveillance (26, 28). To further explore EPIC1’s modulation of the TME, we analyzed the post-EPIC1–knockdown (EPIC1-KD) transcriptomes of MCF7 and other human cancer cell lines (Hs 578T breast cancer cells and A2780 and A2780cis ovarian cancer cells). Analysis revealed that EPIC1 can suppress the expression of genes involved in type I interferon signaling in multiple cancer cell lines (Fig. 1A; fig. S1A; data file S1 and S2). Further analysis of TCGA patient samples revealed increased EPIC1 expression negatively correlates with type I interferon response and is associated with low T cell infiltration in multiple cancer types (Fig. 1B; fig. S1B). Consistently, in vitro knockdown of EPIC1 significantly increased the expression of type I interferon-stimulated genes (ISGs) such as DDX58 (RIG-I), IFIH1 (MDA5), IRF7, IFI27, IFI44, IFI44L, MX1, and CXCL10 (Fig. 1, C and D). On the other hand, overexpressing EPIC1 suppressed ISG expression (Fig. 1, C and D). Similar results were observed in cell lines with high endogenous EPIC1 expression (26), including breast cancer (T-47D and MDA-MB-468), melanoma (MEL-526), pancreatic (PANC1), and prostate cancer (PC3) cell lines (fig. S1, B to E), indicating that EPIC1 is commonly anti-correlated with type I interferon activation in multiple cancer types.

Fig. 1. EPIC1 suppresses type I interferon signaling.

(A) RNA-seq analysis in MCF7 cells after EPIC1 knockdown (KD, siEPIC1) or overexpression (OE). Volcano plots show the differently expressed genes in these cells compared to cells transfected with nontargeting siRNA (Control) or empty vector (VC, vector control). Bar graph shows the top 50 differentially expressed type I interferon–regulated genes. n = 1 RNA-seq sample per group. (B) Box plot shows the average expression of ISGs in TCGA patient samples with no, low, moderate, or high EPIC1 abundance across different cancer types. The upper statistics in the plot represent the correlation between EPIC1 abundance and ISG expression; the lower statistics in the plot represent the correlation between EPIC1 abundance and type I interferon score in EPIC1⁺ samples. The heatmap shows the relative average infiltration of immune cells across patients with no, low, moderate, or high EPIC1. LUSC, lung squamous cell carcinoma; LGG, lower grade glioma; BLCA, bladder urothelial carcinoma; OV, ovarian serous cystadenocarcinoma; HNSC, head and neck squamous cell carcinoma; CESC, cervical squamous cell carcinoma and endocervical adenocarcinoma; UCS, uterine carcinosarcoma; STAD, stomach adenocarcinoma; PRAD, prostate adenocarcinoma; SKCM, skin cutaneous melanoma; LUAD, lung adenocarcinoma. (C) qPCR analysis of the indicated transcripts 48 h after EPIC1-KD (siEPIC1) or overexpression (EPIC1-OE) in MCF7 cells. n = 3 biologically independent samples per group. (D) Immunoblotting and quantification of MDA5 and RIG-I in EPIC1-KD (pink) and EPIC1-OE (blue) MCF7 cells. β-actin is a loading control, n = 3 biologically independent samples per group. (E) qPCR analysis of IFNB1 expression and ELISA-based quantification of secreted IFN-β in MCF7 cells after 24 h of treatment with siEPIC1. P21 is a positive control (26), n = 3 or 5 biologically independent samples per group. (F) Schematic illustration of the experimental design of CM preparation and treatment. (G) Immunoblotting and qPCR analysis for the indicated proteins and transcripts in MCF7 after 24 h treatment with CM from MCF7 cells treated with two different siEPIC1s or siControl or recombinant IFN-β (rIFN-β). n = 2 (immunoblotting) or 3 (qPCR) biologically independent samples per group. (H) qPCR analysis of the indicated transcripts in MCF7 cells after 48 h of siRNA treatment with isotype control (IgG) or human type I interferon receptor-2–neutralizing antibody (IFNα/βR2). Data in (C to E), (G), and (H) are presented as mean ± SD, two-tailed Student’s t test.

Based on this observation, we hypothesized that EPIC1 knockdown may trigger autocrine and paracrine antitumor immune responses. Indeed, we observed a significant increase of interferon-beta (IFN-β) expression in EPIC1-KD cell lines at both the mRNA and extracellular protein levels (Fig. 1E; fig. S1F). Because type I interferon activation in cancer cells promotes antitumor immunity in the TME (29–32), we sought to investigate whether activated IFN-β induced ISGs through autocrine and/or paracrine signaling. We collected culture supernatants (conditioned media, CM) after EPIC1 knockdown. Untreated cancer cells were then exposed to the CM or recombinant IFN-β (rIFN-β) for 24 hours (Fig. 1F). Cells exposed to CM from EPIC1-KD cells (CM-siEPIC1) or rIFN-β exhibited markedly increased STAT1 phosphorylation, indicating activation by extracellular factors like IFN-β (Fig. 1G). In addition, these cells also showed increased expression of ISGs, including DDX58 (RIG-I), IFIH1 (MDA5), IRF7, IFI27, IFI44, IFI44L, MX1, and CXCL10, without changes in EPIC1 amounts (Fig. 1G; fig. S1G). To further confirm the dependency on autocrine and paracrine signaling, we blocked the type I interferon receptor with a neutralizing antibody against IFNα/βR2 and observed a reduction of ISG induction in EPIC1-KD cells (Fig. 1H). Collectively, these results suggest that EPIC1 suppression promotes an antiviral-like type I interferon response with potential antitumor effects.

EPIC1 suppresses type I interferon signaling by reducing the production of dsRNA

Studies suggest that the cytosolic accumulation of dsRNA triggers an antiviral-like type I interferon response in cancer cells, enhancing immune responses (15–18). Given the ability of EPIC1 to suppress type I interferon signaling across various cancer cell lines, we investigated whether it induces any change in dsRNA abundance. Immunofluorescence and flow cytometry analyses using a dsRNA-specific antibody (J2) (33) detected increased cytoplasmic dsRNA amounts in MCF7, MDA-MB-488, and MEL526 cells with EPIC1 knockdown. In contrast, EPIC1 overexpression showed the opposite effect (Fig. 2A; fig. S2, A to E). To confirm this was not a siRNA artifact, we used EPIC1-specific locked nucleic acid (LNA) and observed similar dsRNA activation (fig. S2F).

Fig. 2. EPIC1 reduces dsRNA biogenesis.

(A) Immunofluorescence and flow cytometry analysis of dsRNA in MCF7 cells 24 h after treatment with siEPIC1 or induction of EPIC1 overexpression. Scale bar, 20 μm. n = 6 or 3 biologically independent samples per group. Data are presented as mean ± SD, two-tailed Student’s t test. (B) Schematic illustrates the proposed EPIC1-mediated regulation of dsRNA-induced type I interferon responses through EZH2. (C) Schematic illustrates the experimental design of dsRIP in MCF7 cells. (D) Immunoprecipitation (IP) enrichment of REs captured by dsRIP in siEPIC1 and siEZH2 experiments (y-axis), compared to siControl (x-axis). The IP enrichment is derived from the difference between IP and input sample (see Materials and Methods). Colormap shows the density of REs. n = 1 dsRIP sample per group. (E) Differentially induced REs in cells treated with siEPIC1 or siEZH2. X-axis represents contribution score, and Y-axis represents enrichment score (see Materials and Methods). Sizes of the dots indicate the absolute expression amount in IP samples. Colors of the dots indicate IP enrichment. Dots with black edge represent the REs that were differentially induced by treatment. (F) Percentage of REs that were differentially induced by siEPIC1 or siEZH2. The percentages were calculated based on the RE annotation from RepeatMasker. P-values are derived from Chi-square test. (G) Venn diagrams showing the overlapping REs between siEPIC1 and siEZH2 treatments. The total number of overlapping REs and those in each sub-category [simple repeats, LTRs (including ERVs), LINEs, and SINEs] are indicated. P-values are derived from Fisher’s exact test. (H) Heatmap showing the IP enrichment of REs induced by siEPIC1 and/or siEZH2. Colors indicate the enrichment amount. Color bars represent the induction type (siEPIC1-induced, siEZH2-induced, or induced by both). Only REs with contribution scores higher than 0 in both siEPIC1 and siEZH2 are shown. (I) Distribution of dsRNA enrichment difference between siEPIC1- and siEZH2-induced LTR/ERVs and LINEs.

Endogenous dsRNAs can originate from various sources, including REs (LTRs, LINEs, and SINEs), RNA splicing errors, and mitochondrial damage (7–11, 34–36). Because our previously study showed that EPIC1 can interact with EZH2 (28), and EZH2 has been reported to inhibit RE expression (19, 37), we hypothesized that EPIC1-dependent regulation of dsRNAs involves EZH2 (Fig. 2B; fig. S2G). Indeed, knockdown of either EPIC1 or EZH2 in MCF7 cells resulted in increased amounts of dsRNA and type I interferon signaling (fig. S2, H to I). To identify which specific dsRNAs are regulated by EPIC1 and EZH2, we performed dsRNA immunoprecipitation sequencing (dsRIP-seq) on EPIC1- and EZH2-KD (siEPIC1 and siEZH2) MCF7 cells (Fig. 2C; data file S3). Only cytoplasmic fractions were used for dsRIP-seq, because cytoplasmic accumulation of dsRNAs may indicate homeostatic imbalance and become immunoreactive.

Among the 10,208 REs captured by dsRIP-seq, 6,140 (60.14%) and 6,216 (60.89%) of them showed higher enrichment in EPIC1- and EZH2-KD cells compared to control cells, respectively (Fig. 2D). This is consistent with the increased dsRNA accumulation detected by dsRNA-specific antibody (J2) (Fig. 2A; fig. S2, A to F, H). We further identified 1,323 and 1,236 differentially induced REs by EPIC1 and EZH2 knockdown, respectively (Fig. 2E; data file S4). LINEs, SINEs, LTRs (including both non-viral LTRs and ERVs), and simple repeats were the most enriched RE classes in both knockdown groups (fig. S2J). Specifically, siEPIC1 significantly induced 63.3% (157/248) of the dsRIP-captured LINEs and SINEs, as well as 51.3% (307/599) of the LTRs. In contrast, only 4.7% (670/14,255) of the simple repeats are induced by siEPIC1 treatment (Fig. 2F). On the other hand, siEZH2 significantly induced 66.1% (164/248) of the dsRIP-captured LINEs and SINEs and 54.8% (328/599) of the LTRs, whereas only 3.9% (561/14,255) of the simple repeats were induced in the same experiment (Fig. 2F).

Knocking down either EPIC1 or EZH2 induced 769 REs, suggesting substantial overlap in their regulatory effects (Fig. 2G). Among these overlapped dsRNAs, we found siEPIC1 shared more LTR (71.0% of total siEPIC1-induced LTRs), LINEs (85.5%), and SINEs (83.0%) with siEZH2, compared to the simple repeats (42.6%) (Fig. 2, G and H; fig. S2K). To determine if EPIC1 and EZH2 coregulate these shared REs, we further measured the dsRNA enrichment at 320,748 RE-harboring locations across the genome. We found that compared to locations with REs induced by either siEPIC1 or siEZH2 alone, locations harboring shared REs showed more consistency in dsRNA enrichment (Fig. 2I; fig. S2L). These results suggest that loss of either EPIC1 or EZH2 affects the transcription of dsRNAs from locations harboring shared REs, indicating a potential interaction between EPIC1 and EZH2 in suppressing dsRNA biogenesis.

EPIC1 knockdown stimulates dsRNA-mediated antitumor immune responses and suppresses tumor growth

Cytosolic dsRNAs are sensed by dsRNA sensors RIG-I and MDA5, which mediate an antiviral-like innate immune response by stimulating the aggregation of mitochondrial antiviral signaling protein (MAVS) (38–40). To determine whether EPIC1 attenuates type I interferon responses by reducing dsRNA biogenesis, we generated MAVS knockout MCF7 cells using CRISPR/CAS9. Flow cytometry analysis reveals increased dsRNA abundance in both control (sgControl/WT) and MAVS knockout (sgMAVS) cells upon EPIC1 knockdown (Fig. 3A and fig. S3A). However, compared to control cells, the siEPIC1-induced increase in IFN-β abundance, STAT1 phosphorylation, and the expression of ISGs and genes related to antigen processing and presentation was significantly diminished in MAVS-knockout cells (Fig. 3, B and C; fig. S3, B to D).

Fig. 3. EPIC1 suppresses the type I interferon response by inhibiting dsRNA production.

(A) Quantification of dsRNA by flow cytometry (circle and triangle symbols) at 24 h and ELISA-based quantification of IFN-β (grey bars) at 48 h after EPIC1 knockdown (siEPIC1) in MCF7 cells. n = 6 or 3 biologically independent samples per group. (B and C) Immunoblotting (B) and qPCR (C) analysis in sgControl and sgMAVS MCF7 cells after 24 h of treatment. (D) Flow cytometry analysis of THP-1 cells after 48 h treatment with CM from MCF7 cells after 48 h treatment with two different siEPIC1s or siControl or 24 h treatments with rIFN-β. n = 4 biologically independent samples per group. (E) Flow cytometry analysis of human primary CD8⁺ T cells transduced with gp100 antigen and cocultured with MEL-526 cells or 24 h treatment with rIFN-β. n = 6 biologically independent samples per group. (F) Schematic illustrates the experimental design of coculture experiment. Control and MAVS-knockout MCF7 cells were transfected with siEPIC1 or siControl, the CM was removed, filtered to remove siRNAs, and added back to the culture along with THP-1 cells. Flow cytometry analysis of THP-1 cells after co-culture with control (sgControl) and MAVs-knockout (sgMAVS) MCF7 cells treated with control siRNA or siEPIC1. n = 4 biologically independent samples per group. Data in (A) and (C to F) are presented as mean ± SD, two-tailed Student’s t test, the P values in D and E represent the minimum P value predicted for the two cell populations in the same experimental group.

Tumor type I interferon signaling plays important roles in macrophage and T cell activation in the TME (29–32, 41). To investigate the inhibitory effect of EPIC1 on type I interferon signaling in macrophages, THP-I monocytic cells were treated with CM from EPIC1-KD MCF7 or MDA-MB-468 TNBC cells (Fig. 1F). Treatment with CM from EPIC1-KD cells significantly increased the total CD169⁺ and CD169⁺ HLA-DR⁺ monocyte populations (Fig. 3D; fig. S3E). To further determine the effect of EPIC1 regulation of type I interferons on cytotoxic T cell activation, we performed hCD8⁺ T cell–cancer cell coculture assays (fig. S3F). Specifically, we cocultured hCD8⁺ T cells (HLA-A*02 restricted gp100 TCR-transduced, see Materials and Methods) with EPIC1-KD or EPIC-overexpressing MEL-526 cells (specifically expressing HLA-A*02 and gp100) (fig. S3F). EPIC1 knockdown in MEL-526 cells increased the activation of 4–1BB⁺ (CD137⁺) and 4–1BB⁺ PD-1⁻ (CD274⁻) cytotoxic T cell populations in the coculture system (Fig. 3E). Conversely, EPIC1-overexpressing MEL-526 cells significantly inhibited CD8⁺ T cell activation during coculture (fig. S3G). Taken together, our results demonstrate that EPIC1 suppresses the immune response in cancer cells, macrophages, and T cells by inhibiting type I interferon signaling.

To further determine whether the activation of CD169⁺ macrophages indeed depend on dsRNA-MAVS signaling, we cocultured THP-1 monocytes with EPIC1-KD MAVS knockout MCF7 cells (Fig. 3F). In contrast to the ability of EPIC1-KD control cells to expand the total-CD169⁺ and CD169⁺ MHC-II⁺ populations, EPIC1-KD MAVS knockout cells were ineffective in inducing CD169⁺ monocyte differentiation (Fig. 3F). These observations suggested that EPIC1-mediated suppression of type I interferon signaling in the TME depends on dsRNA-MAVS signaling.

We next sought to determine how targeting EPIC1 affects tumorigenesis in an in vivo setting. Using a human TNBC (MDA-MB-468) xenograft tumor model, we administered siRNA or LNA into mice through an established PMBOP-CP nanocarrier to specifically knock down EPIC1 in tumor cells (42) (Fig. 4A). Mice treated with EPIC1-targeting siRNA or LNA (siEPIC1 or LNA-EPIC1) exhibited a reduced tumor growth rate compared to controls (Fig. 4B; fig. S4A). Further histological analysis showed successful knockdown of EPIC1 in tumor tissues, as well as increased dsRNA levels. (Fig. 4C). These results show that targeting EPIC1 reduced tumor dsRNA expression in vivo.

Fig. 4. In vivo targeting of EPIC1 sensitizes TNBC xenografts to pembrolizumab.

(A) Schematic illustrates experimental design for intravenous (i.v.) nanoparticle delivery of EPIC1-targeting or control siRNA or LNA into mice bearing MDA-MB-468 xenograft tumors. The siRNAs or LNAs were delivered by i.v. PMBOP-CP nanomicelles (42). (B) Growth rate of MDA-MB-468 tumors in athymic nude mice treated with PBS, siControl, siEPIC1, LNA-Control, or LNA-EPIC1. Data are presented as mean ± SEM. n = 4 (LNA-Control) or 5 (all other groups) mice per group with one tumor per mouse. Arrows indicate the day of treatment. P values were calculated using a two-tailed Student’s t test on day 14. (C) In situ hybridization to detect EPIC1 using RNAScope and immunofluorescence analysis of dsRNA in MDA-MB-468 xenograft tumor sections from athymic nude mice. Bar graph shows the EPIC1 and dsRNA abundance in tumor sections. n = 3 or 4 tumors per treatment group and 3–4 sections of each tumor were used for quantification. Human CD44 (hCD44) is a cell surface marker of MDA-MB-468 cells used to distinguish human tumor cells from mouse stromal cells. Scale bar, 20 μm (left) and 10 μm (right). Data are presented as mean ± SD, two-tailed Student’s t test. (D) Schematic illustrates the experimental design for analysis of MDA-MB-468 xenograft tumors from humanized (HuCD34-NCG) mice after treatment with LNA-EPIC1 or LNA-Control and pembrolizumab (Pem) or IgG control delivered i.v. or intraperitoneally (i.p.) at the indicated time points. After recovery from primary tumor resection, some mice were inoculated again with MDA-MB-468 cells at day 42 and sacrificed for analysis at day 59 (Fig. S5, C to E). (E) Growth rate of MDA-MB-468 tumor in humanized mice after treatment. Data are presented as mean ± SEM. n = 5 mice per group with one tumor per mouse. Arrows indicate the day of treatment. P values were calculated using a two-tailed Student’s t test on day 14. (F) Immunofluorescence analysis of dsRNA in MDA-MB-468 tumors developed in humanized mice after treatment. Scale bar, 10 μm. Tumor growth in (B) and (E) were calculated by Day-N/Day-0 ratio.

EPIC1 knockdown enhances antitumor immune response and pembrolizumab efficacy

hCD34⁺ humanized mouse models are useful for testing the efficacy of immunotherapeutic drugs, such as the programmed cell death protein 1 (PD-1) inhibitor pembrolizumab (Keytruda), in TNBC (43). To further investigate how the regulation of dsRNA by EPIC1 modulates the efficacy of pembrolizumab immunotherapy in a human TNBC tumor model, we used HuCD34-NCG mice bearing MDA-MB-468 xenograft tumors (Fig. 4D). Preliminary analysis confirmed the successful engraftment of the human immune cells in HuCD34-NCG mice after 14 weeks of injection of human CD34⁺ stem cells (fig. S4B). Following the development of MDA-MB-468 tumors, we treated the humanized mice with a pembrolizumab biosimilar (Pem, an anti-human PD-1) or isotype control (IgG) antibodies with or without LNA-EPIC1. We also utilized the PMBOP-CP nanocarrier (42) to deliver EPIC1 targeting and non-targeting LNA intravenously to these humanized mice. MDA-MD-468 tumors, which have high endogenous EPIC1, exhibited a modest response to pembrolizumab or LNA-EPIC1 treatment compared to control (LNA-Control + IgG) (Fig. 4E). Conversely, EPIC1 knockdown showed increased dsRNA accumulation in tumor tissues and significantly sensitized human TNBC response to pembrolizumab treatment (Fig. 4, E and F).

To explore how knockdown of EPIC1 altered the tumor immune microenvironment and pembrolizumab response, we performed single-cell RNA sequencing (scRNA-seq) analysis on tumor-infiltrated hCD45⁺ cells collected from the four treatment groups: LNA-Control + IgG, LNA-Control + Pem, LNA-EPIC1 + IgG, and LNA-EPIC1 + Pem (Fig. 5A). In total, we identified 14 human cell clusters in the MDA-MB-468 tumors (Fig. 5, B and C). These clusters included human cytotoxic T lymphocytes (CTLs), T_reg cells, NK cells, B cells, and macrophages (Fig. 5, B and C; fig. S5, A and B). Our observations demonstrate that a broad spectrum of human immune cells infiltrated MDA-MD-468 tumors. A more in-depth look at the gene expression of CD8⁺ cells revealed that transcripts associated with cytotoxic T cell activation were increased in samples from tumors treated with LNA-EPIC1, pembrolizumab, or LNA-EPIC1 + pembrolizumab (Fig. 5D). This suggested a potential synergistic increase of human CTL–mediated immune response in the pembrolizumab and LNA-EPIC1 combination treatment groups. These observations were further validated by immunohistochemical and immunofluorescence analysis of tumor-infiltrating hCD8⁺ and hCD4⁺ lymphocytes (Fig. 5E). In addition to T cells, EPIC1 knockdown significantly increased macrophage infiltration (Fig. 5F). Moreover, the infiltrated macrophages displayed higher expression of inflammatory markers such as CXCL5, CXCL2, and CXCL3 (Fig. 5G). To further investigate how the immune system changes and reacts to tumor recurrence after treatment, we reintroduced the MDA-MB-468 tumor to HuCD34-NCG mice after recovery from primary tumor removal (Fig. 4D). Following tumor growth, we analyzed circulatory lymphocytes and TILs. The results showed that the population of peripheral B cells (CD45⁺ CD3⁻ CD19⁺) and tumor-infiltrating CD4⁺ T cells was increased in all mice previously treated with LNA-EPIC1 or Pem or combination (fig S5, C to E). Together, our data showed that targeting EPIC1 can activate the immune response and boost the therapeutic effect of pembrolizumab in the humanized mouse model.

Fig. 5. Inhibition of EPIC1 activates antitumor immune responses.

(A) Schematic illustrates the use of scRNA-seq to analyze the infiltration of human CD45⁺ cells (hCD45⁺) into TNBC xenograft tumors. To ensure the specificity hCD45⁺ cells, magnetic cell sorting was used to enrich hCD45⁺ before library preparation, and scRNA-seq reads were mapped onto the human transcriptome. Mice bearing MDA-MB-468 xenografts tumors were treated with EPIC1-specific LNA (LNA-E) or control LNA (LNA-C) in combination with pembrolizumab (Pem) or IgG control antibodies. (B) UMAP analysis of MDA-MB-468 tumor-infiltrating hCD45⁺ cells from all four treatment groups mentioned in (A) sorted cells into 14 clusters. Each cell is represented by a dot with a different color according to its assigned cluster. n = 2 (for LNA-E + Pem) or 1 (all other groups) tumors per group for a total of 5 tumors were used in scRNA-seq analysis. (C) Cell type annotation and canonical marker expression. The sizes of the dots represent the fraction of cells in the group that express the corresponding markers. The colors of the dots represent the average expression of markers. The bar plots on the side represent the number of cells in each annotated group. (D and E) Characterization and quantification of tumor-infiltrating T cells. Violin plots (D) show the gene expression patterns of tumor-infiltrating CD8⁺ T cell clusters in each treatment. The two LNA-E + Pem samples represent two different tumors used for scRNA-seq analysis . Immunofluorescences images (E) show hCD4⁺ and hCD8⁺ cells in tumor sections. Scale bar, 50 (main images) and 25 μm (magnified images). The box and whiskers plots represent the quantification of hCD4⁺ and hCD8⁺ cells in immunofluorescence images of tumor sections. Multiple regions of tumor sections from each group were used for imaging and quantification using ImageJ. Values were calculated by the ratio of sample/LNA-C + IgG, presented as mean ± SD, n = 2 (for LNA-E + Pem) or 1 (all other groups) tumors per group and 7–9 section from each tumor used for quantification. (F and G) Characterization of tumor-infiltrating macrophages. Quantification of total macrophages (F) and macrophages expressing the indicated inflammatory markers (G) in the TME identified using scRNA-seq analysis. Violin plots in (G) show the differences in gene expression between the LNA-EPIC1 and LNA-Control groups treated with IgG. The P value among the cell types calculated by one-end Fisher exact test. n = 1 tumor per group used for scRNA-seq analysis. (H) A schematic depiction of the EPIC1-EZH2 axis suppressing RE dsRNA–dependent activation type I interferon signaling. Targeted inhibition EPIC1 in cancer cells increases RE-derived dsRNA and activates MAVS-mediated type I interferon signaling, which promotes the activation of immune cells in the TME and can improve the response to anti-the anti-PD-1 drug pembrolizumab.

DISCUSSION

Immunotherapies that inhibit the receptor PD-1 or its ligand PD-L1 are effective in unleashing T cells to recognize and kill tumor cells in some cancer types. Despite the initial success, primary and secondary resistance to immunotherapy remains a major obstacle in improving the survival outcome. In the present study, we demonstrated that lncRNA EPIC1 suppressed type I interferon signaling in multiple cancer cell lines, including breast cancer, prostate cancer, pancreatic cancer, and melanoma. We showed that the regulation of type I interferon secretion by EPIC1 can modify cancer cells, macrophages, and T cells in the TME. We found that EPIC1-mediated regulation of type I interferon was driven by its ability to suppress the cellular accumulation of dsRNAs. Further integrative analysis using dsRIP-seq revealed that EPIC1- and EZH2-regulated dsRNAs included those produced from REs such as LINEs, SINEs, and LTRs. These RE family genes are known to be one of the primary sources of immunogenic dsRNA REs (44). Furthermore, we have also shown that EPIC1-regulated REs significantly overlap with EZH2-regulated REs. Previous study has demonstrated EZH2-mediated dsRNAs suppression is an important mechanism for tumor immune evasion (19). Our previous study showed that EPIC1 can interact with EZH2 and enhances its chromatin binding (28). Future studies are warranted to investigate if EPIC1-mediated regulation of REs/dsRNAs is mediated by its activation of EZH2.

We also demonstrated EPIC1 can be a potential therapeutic target in combination with immunotherapy. Coculture assays of T cells or monocytes with cancer cells showed that EPIC1 knockdown could significantly increase the therapeutic effect of pembrolizumab through antitumor T cell and macrophage activation. The combined therapeutic effect of anti-EPIC1 and pembrolizumab was further validated using in vivo hCD34⁺ humanized mouse models. The anti-EPIC1 and pembrolizumab–treated hCD34⁺ humanized mouse model provides insight into how the implanted human immune system in hCD34⁺ humanized mice interact with human TNBC breast cancer. The scRNA-seq data allowed us to determine whether the infiltrated CD45⁺ cells were of human or mouse origin by mapping the transcriptome of each CD45⁺ cell to the human genome. Our results demonstrated that hCD34⁺ mice indeed mobilized a complete human immune response to xenografted human tumors. The finding that in vivo knockdown of EPIC1 increased macrophage infiltration was also consistent with the in vitro coculture assay. These results suggest the dsRNA-induced type I interferon signaling that was suppressed by EPIC1 plays an important role in tumor macrophage infiltration.

One caveat of the hCD34⁺ NCG humanized model is that it is difficult to determine the difference between tumor antigen–driven and alloantigen-driven antitumor immune responses. To address this caveat, we reinoculated mice after anti-EPIC1 and pembrolizumab treatment with the same cancer cells originally used to establish tumors before treatment (Fig. 4D). Our data showed a trend that the immune response against the reinoculated cancer cells increased in the mice treated with anti-EPIC1 and pembrolizumab (fig. S5C–E). This observation suggests that after the first tumor challenge and therapy, the mice may have developed a tumor antigen–driven immune response to the xenografted cancer cells. Building on our prior work demonstrating EPIC1 overexpression in over 50% of solid tumors with minimal expression in normal tissues (26), future studies should explore the functional consequences of EPIC1-regulated dsRNAs in more cancer types.

In summary, our findings demonstrate that EPIC1 acts as a master regulator of the dsRNA-type I interferon signaling pathway across multiple cancer types, and we identified and characterized the source of immunogenetic dsRNAs suppressed by EPIC1 (Fig. 5H). These observations provide a previously unknown mechanism for EPIC1-mediated tumor immune evasion. Further, in vivo humanized mouse model study not only establishes EPIC1 as a potential biomarker for immunotherapy resistance in various cancers, but also open avenues for developing combination therapies targeting EPIC1 and immunotherapy.

MATERIALS AND METHODS

Cell culture and reagents

Human breast cancer cell lines (MDA-MB-231, MDA-MB-468, MCF7, and T-47D), a human melanoma cell line (MEL-526), a human pancreatic cancer cell line (PANC1), a human prostate cancer cell line (PC-3), and human embryonic kidney (HEK) 293T cells were purchased from the American Type Culture Collection (ATCC). The human ovarian cancer cell line (OVCAR-4) was purchased from NIH/NCI. The human ovarian cancer cell lines (A2780 and A2780cis) were purchased from the European Collection of Cell Cultures (ECACC). The human monocyte cell line (THP-1) was provided by Dr. Wen Xie (Department of Pharmaceutical Sciences, University of Pittsburgh). All cell lines were cultured according to the recommended guidelines and supplemented with 10% heat-inactivated fetal bovine serum (Gibco, #10438–026) and a 1X antibiotic and antimycotic (Gibco, #15140–122) mixture.

For in vitro RNA knockdown assays, two different siRNAs, one LNA for EPIC1 (26); one siRNA for EZH2 (28); and a negative control siRNA or control LNA (26, 28), respectively, are used. The sequences of the oligos used for knockdown studies are listed in table S1. For all cells, each siRNA or LNA was transfected individually using Lipofectamine RNAiMAX transfection reagent (Invitrogen, #13778150) at a concentration of 40 nM according to the manufacturer’s guidelines. Each of the treated cells was processed separately for further analysis. Lipofectamine 2000 (Invitrogen, #11668019) was used for plasmid transfection. Recombinant IFN-β (rIFN-β) was purchased from PeproTech (#300–02BC), used at a concentration of 50 to 100 pg/mL. Human IFNa/bR2 (#MAB4015) and isotype control antibodies (#MAB004) were purchased from R&D Systems and were used at a concentration of 10 μg/mL for naturalization.

RNA-seq analysis

The Splices Transcript Alignment to a Reference (STAR) (45) and RNA-seq by Expectation Maximization (RSEM) (46) are used to profile RNA-seq data including MCF7 EPIC1 overexpressing stable cells and control cells (GSE229721), MCF7 EPIC1 knockdown and control samples (GSE98538), and EPIC1 knockdown and control samples in Hs 578T breast cancer cells and A2780 and A2780cis ovarian cancer cells (GSE229722). Mapping and quantification are based on the human reference genome GRCh38. Quality control and mapping information are provided in data file S1. In total, 24,334,944 reads for EPIC1 knockdown, 17,714,539 reads for EPIC1 knockdown control, 20,481,855 reads for EPIC1 overexpression, 21,912,036 reads for EPIC1 overexpression control are uniquely aligned to the genome. Next, Cufflinks (v2.2.1) is used to conduct differential expression gene analysis between overexpression/knockdown and the corresponding control experiments. Finally, median ratio normalization (MRN) (47) and logarithmic transformation has been applied to normalize the estimated count, leading to the final processed expression matrix with a dimension of 4×60,468.

RNA isolation and quantitative real-time polymerase chain reaction (qPCR) analysis

Total RNA from cultured cells was isolated using TRIzol reagent (Invitrogen, # 15596018), according to the manufacturer’s instructions. Following RNA isolation, 1–2 μg of total RNAs was used to synthesize cDNAs using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, #4368813). qPCR analyses were performed using Power SYBR Green PCR Master Mix (Applied Biosystems, #4367659) on a QuantStudio 6 Pro Real-Time PCR System (Applied Biosystems). Relative mRNA expression was determined by calculating ΔΔCt values normalized to GAPDH. The primer sequences used for the qRT-PCR analysis are listed in table S2.

Immunoblotting

Total proteins from the specified cells were extracted using RIPA lysis buffer containing 1X protease and phosphatase inhibitor (MCE, HY-K0010). The protein concentration of each sample was measured using a BCA protein assay kit (Thermo Scientific, #23227) according to the manufacturer’s instructions. Equal amounts of protein samples were mixed with an appropriate volume of 5X protein sample buffer containing a reducing agent. After incubation at 98°C for 10 min, equal amounts of protein samples were separated by SDS-polyacrylamide gel electrophoresis on 4–12% gel and transferred to polyvinylidene difluoride (PVDF) membrane (Bio-Rad, #162–0177). The protein transferred membranes were immunoblotted with appropriate primary antibodies for overnight at 4°C, followed by appropriate horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 h at room temperature. Signals were visualized using chemiluminescence substrate (Thermo Scientific, #32106, 34577) and exposed to films using an AX700LE film processor (Alphatek) or the iBright CL1500 imaging system (Invitrogen). Band intensities were quantified using the ImageJ software and normalized to respective β-actin expression. The details of the antibodies used for immunoblotting are listed in table S3.

Enzyme-linked immunosorbent assay (ELISA)

For interferon-beta (IFN-β) ELISA, cell culture supernatants were used after treatment. After the removal of cell debris by centrifugation, ELISA was performed according to the manufacturer’s protocol. The human IFN-β ELISA kit was purchased from R&D Systems (#DY814–05). The optical density was determined using a Bio-Rad xMark Microplate Absorbance Spectrophotometer.

In vitro coculture assays

hCD8⁺ T cell co-culture experiments: MEL-526 cells after treatment or stable cells, as previously described, were trypsinized and used. GP100 TCR-transduced primary human CD8⁺ T cells were provided by Dr. Kammula’s laboratory at UPMC Hillman Cancer Center. MEL-526 cells (1 ×10⁵) and T cells (0.5 ×10⁵) were seeded in a ratio of 2:1 in 96 well U-bottom plates and incubated for 24 h. Cells then harvested and processed for flow cytometry analysis.

THP-1 co-culture experiments: cancer cells were transfected with siRNA or LNA for 36 h. The culture medium was then removed and passed through a 0.22 μM syringe filter to remove leftover siRNA or LNA-Lipofectamine and maintain the cancer cell secretome in the culture microenvironment. After washing, the filtered conditioned medium and THP-1 cells (at a ratio of 2:1 of cancer cells: THP-1 cells) were added to the respective cancer cells and maintained under the conditions mentioned in Fig. 3F.

In vivo xenograft models

Female athymic nude mice aged 5–6 weeks were purchased from Charles River Laboratories and used in the xenograft model. Female humanized mice (HuCD34-NCG) were obtained from Charles River Laboratories. HuCD34-NCG mice were developed by infusing human cord blood-derived CD34⁺ cells into ~14 weeks triple immunodeficient mice. Before obtaining the mice, peripheral blood samples were analyzed to confirm the rate of humanization by Charles River Laboratories (Fig. S4B). The data revealed that all the mice successfully developed immune cells (Fig. S4B). MDA-MB-468 cells cultured in vitro were trypsinized and washed twice with sterile PBS and then 5 × 10⁶ cells/mouse suspended in 100 μL of PBS:Matrigel (1:1) (Corning, #354234) were subcutaneously implanted into the right mammary fat pad. Cells were tested for mycoplasma contamination before injection, and all cells were confirmed to be free of mycoplasma. Under continuous monitoring, tumor dimensions were measured using a caliper, and tumor volumes were determined using the following formula: tumor volume = length × (width)² × 0.5. For the in vivo knockdown experiments, we used a PMBOP-CP (42) nanocarrier system to deliver siRNA or LNA targeting EPIC1 or control. Each mouse was administered intravenously with 200 μL of PMBOP-CP nanocarrier loaded with 30 μg siRNA or LNA once every five days for a total of three times. In addition, humanized mice were administered intraperitoneally with 10 mg/kg InVivoSIM anti-human PD-1 (Pembrolizumab Biosimilar, BioxCell, #SIM0010) or RecombiMAb human IgG4 (S228P) isotype control (BioxCell, #CP147) twice a week for a total of four times. The tumor growth rate was measured using the day-N/day-0 treatment ratio. After 14 days of treatment, all tumor-bearing athymic nude mice were sacrificed using the humane method and the tumors were collected. In humanized mice, tumor resections were performed using a survival surgical procedure after the treatment. The mice were monitored for recovery and pain maintenance following surgery. After complete recovery and the indicated time periods, MDA-MB-468 cells (5 × 10⁶ cells/mouse) mice were subcutaneously reimplanted into the left mammary fat pad of mice, as mentioned in the first injection. Following tumor development, mice were euthanized using the humane method and various specimens were collected for analysis.

Immunocytochemistry-Immunofluorescent (ICC-IF) analysis

For immunocytochemical analysis, the cells were cultured in 8-well chambered slides (Thermo Scientific, #154534) and treated as previously described. The cells were then washed with PBS (Gibco) and fixed with 4% formaldehyde for 15 min. After three washes with PBS, the cells were incubated with a blocking-permeabilization buffer (5% goat serum and 0.3% Triton X-100 in 1X PBS) for 1 h. The cells were incubated with anti-dsRNA (J2) antibody diluted (1:300) in antibody diluent buffer (1% BSA and 0.3% Triton X-100 in 1X PBS) overnight at 4°C in a humidified chamber. After three 5 min washing with PBS, the cells were incubated with fluorescence-conjugated anti-mouse secondary antibodies diluted in antibody diluent buffer for 1 h at room temperature in the dark. After three washes with PBS for 5 min each, the cells were mounted using the ProLong Gold Antifade Mountant with DAPI (Invitrogen, #P36935) and coverslips. Slides were imaged using KEYENCE BZ-X800 fluorescence microscope, analyzed using BZ-X800 Analyzer (KEYENCE). The details of the antibodies used are listed in table S3.

Immunohistochemistry-immunofluorescence (IHC-IF) analysis

Formalin-fixed paraffin-embedded (FFPE) tissue blocks were sliced into 5 μm-thick sections and prepared using standard sodium citrate antigen retrieval procedures. Tissue sections were incubated in a blocking buffer containing 5% goat serum diluted in PBST. Mouse on mouse blocking reagent (Vector Laboratories, #MKB-2213) was used with blocking solution depending on the secondary antibodies. Sections are then incubated with primary antibodies diluted in an antibody diluent buffer containing 1% BSA for overnight at 4°C in a humidified chamber. The sections were then washed thrice with PBST for 5 min each. For fluorescence-conjugated primary antibodies, the appropriate fluorescence-conjugated secondary antibodies were diluted in antibody diluent buffer and incubated for 1 h at room temperature in a humidified and dark chamber. Sections were then washed three times with PBST and once with PBS, mounted using ProLong Gold Antifade Mountant with DAPI, and coverslips. Images from the sections were obtained using KEYENCE BZ-X800 fluorescence microscope. For measurement of target expression, 3–9 regions from each tumor section were quantified using ImageJ software. The details of the antibodies used are listed in table S3.

Flow cytometry analysis

dsRNA flow cytometry analysis: Cells were collected by trypsinization after treatment and washed twice with PBS. Cells were fixed with 4% formaldehyde for 20 min at RT. After two washes with PBS, cells were permeabilized for 15 min at RT using 0.1% Triton X-100 in PBS. Cells then incubated with 1% BSA for at RT for 1 h followed by incubation with 2.5 μg/mL of anti-dsRNA (J2) antibody (CST, #76651) at RT for 1 h. After three washed cells were incubated with 2.2 μg/mL of Alexa Fluor 594 (AF594) or Alexa Fluor 488 (AF488) conjugated goat anti-mouse IgG H&L antibody (abcam, #ab150117) at RT for 1 h. Cells then washed three times with PBS and suspended in 0.5% BSA in PBS. Centrifugation steps were performed at 300 × g in 4°C. Cells were analyzed using MACSQuant analyzer (Miltenyi Biotec). Gating strategy given in fig. S2C.

MEL-526: hCD8⁺ T cells co-culture analysis: Cells were collected and washed twice with PBS and incubated with GHOST dye (1:1000 in PBS, Tonbobio, #13–0870) for 30 min at 4°C. Cells then washed twice with FACS buffer (PBS with 2% FBS) and incubated with fluorescent conjugated antibodies targeting hCD3, hCD4, hCD8, hCD279 (PD-1), and hCD134 (4–1BB) for 1 h at RT. After washing 2–3 times, cells were analyzed using Cytek Aurora flow cytometer (Cytek Biosciences). Gating strategy: singlet hCD3⁺ GHOST⁻ cell population were gated to separate live T cells and were further gated to obtain hCD8⁺ hCD4⁻ cells. These cells were used to detect hPD-1 and hCD134 expressions.

THP-1 cells analysis: Cells were collected, washed twice with PBS and incubated with GHOST dye for 30 min at 4°C. Cells then washed twice with FACS buffer and incubated with fluorescent conjugated antibodies targeting hCD11b, hHLA-DR, hCD169 for 1 h at RT. After 2–3 washes cells were analyzed using Cytek Aurora flow cytometer (Cytek Biosciences). Gating strategy: singlet hCD11b⁺ GHOST⁻ cell population were gated to separate live hCD11b⁺ cells and used to detect hHLA-DR and hCD169 expressions.

Analysis of TILs and blood immune cells analysis from HuCD34-NCG mice: Single-cell suspensions were prepared by mechanical disruption of tumors using scissors and incubating in tumor dissociation buffer (RPMI 1640 w/o serum supplemented with DNase I (Roche, #10104159001) 0.3 mg/mL and Liberase TL (Roche, #05401020001) 0.25 mg/mL) for 30 min at 37°C, and then dispersed through 70 μm cell strainers. Blood samples were collected and treated with ACK lysing buffer (Gibco #A1049201). After washing using PBS, all samples were stained with GHOST dye for 30 min at 4°C. After two washes with FACS buffer incubated with fluorescent conjugated antibodies targeting hCD45 (BV421), hCD3 (APC-Cy7), hCD4 (PE), hCD8 (AF700), hCD11b (AF594), hCD279 (FITC), and hCD137 (APC) or hCD19 (APC). After washing the cells were analyzed using Cytek Aurora flow cytometer (Cytek Biosciences). Gating strategy: singlet hCD45⁺ GHOST⁻ cell population were gated to separate live hCD45⁺ cells and were further gated to obtain hCD3⁺ hCD8⁺, hCD3⁺ hCD4⁺ and hCD3⁻ hCD19⁺ cells. All data were analyzed using FlowJo Software (10.9.0). Antibodies used for flow cytometry analysis are listed in table S3.

Cloning, shRNA construction, sgRNA construction, and viral transduction

Full-length EPIC1 was amplified from total RNAs of MCF7 or T-47D cells, construction of retroviral or lentiviral EPIC1 expression plasmids, and stable EPIC1 overexpressing MCF7, MEL-526 and MDA-MB-468 cells were established as previously described (26). Stable EPIC1 knockdown (shEPIC1) cells were established as described previously (26). Two shRNA targeting EPIC1 and one control shRNA were used and the oligo sequences are listed in table S4.

To achieve MAVS knockout using CRISPR technology, we designed three guide RNA (gRNA) for MAVS and one scrambled gRNA as a control and used the lentiCRISPRv2 vector (table S5). After preparation of lentiviral particles using HEK-293T cells, targeted cells were infected with the lentivirus packaged with Cas9 and single-guide RNA (sgRNA) expression plasmid encoding puromycin resistance (Addgene plasmid, #52961). Knockout efficiency was determined by immunoblotting. gMAVS_F3/R3 and gMAVS_F2/R2 displayed maximum knockout efficiencies of their target gene in MCF7 cells and were used for subsequent experiments.

BaseScope assay

The BaseScope assay is a refined version of RNAScope technology uses few as 1–3 Z pairs for in situ hybridization technology to detect RNA. BaseScope assays were performed according to the manufacturer’s protocol, using the BaseScope Detection Reagent Kit v2-RED (Advanced Cell Diagnostics (ACD) Hayward, #322900-USM) and EPIC1 probes from ACD (Ba-Hs-LOC284930–3zz-st-C1). Briefly, tumors were fixed in 10% neutral buffered formalin, and paraffin blocks were prepared using standard processing. 5 μm thick tumor sections were incubated at 60 °C for 1 h, followed by xylene treatment to remove paraffin. After paraffin removal, sections were dehydrated in 100% ethanol, followed by drying at 60°C for 5 min. Next, the tissue sections were pretreated with hydrogen peroxide at room temperature (RT) for 10 min, followed by performing target retrieval using an Oster^® Steamer at 100°C. Followed by dehydrating in 100% ethanol and dried at 60°C for 5 min, sections were incubated with RNAscope^® Protease IV at RT for 30 min. Next, the slides were incubated at 40°C with the EPIC1 probes for 2 h in in the HybEZ system (ACD). Then Signal amplification was performed using the following steps: AMP-1 for 30 min, AMP-2 for 30 min, AMP-3 for 15 min, AMP-4 for 30 min, AMP-5 for 30 min, AMP-6 for 15 min at 40°C in HybEZ followed by AMP-7 for 30 min, and AMP-8 for 15 min at RT. After each step, the slides were washed with 1X wash buffer (ACD) twice at RT for 2 min with occasional agitation. Chromogenic detection was performed using BaseScope Fast RED, followed by counterstaining with Gill’s hematoxylin (Sigma-Aldrich). All images were collected using a Zeiss Axiostar plus microscope. Images were quantified for EPIC1 expression using ImageJ software and normalized to their nuclear staining signals.

dsRNA immunoprecipitation and sequencing (dsRIP-seq)

MCF7 cells after siRNA transfection were collected by trypsinization. Cells then washed twice with cold PBS. A small portion of the cells were taken for qPCR analysis. Cell pellets then resuspended in 1 mL of nuclear isolation buffer (1:5 of nuclear isolation buffer contains 1.28 M sucrose, 40 mM Tris-HCl pH 7.5, 20 mM MgCl, 4% Triton X-100; 1:5 of 1X PBS; 3:5 of nuclease free water). Cell then incubated for 20 min at 4°C in rotating mixer. Then samples centrifuges at 4000 rpm for 10 min. Pellet-free supernatant were transferred to fresh tube, 5–10% of samples were saved as input. Anti-dsRNA (J2) or isotype control antibody (5 μg/mL) added to this fraction and incubated overnight at 4°C in rotating mixer. Dynabeads Protein G (50 μL/reaction, Invitrogen, #10004D) taken and washed twice with bead washing buffer (50 mM Tris pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% NP-40, 0.5% sodium deoxycholate, 1 mM PMSF, 1X protease inhibitor, 1X RNase inhibitor). Beads then resuspended in 50 μL of nuclear isolation buffer and added to cell fraction. This mixture incubated for 1–2 h at 4°C in a rotating mixer. After incubation centrifuged at 9000 rpm for 10–15 sec. Then using magnetic rack, supernatants were removed, and beads-pellets were washed using beads wash buffer four times. After washing, centrifuged at 9000 rpm for 10–15 sec to remove excess buffer. Bead pellets were resuspended in 1 mL of TRIzol reagent and proceeded to RNA isolation steps. GlycoBlue coprecipitant (Invitrogen, #AM9516) was used to enrich RNA content. Isotype control samples had very low RNA content and were not proceeded to sequencing. Total RNAs were then purified using RNA clean and concentrator kit-5 (Zymo, #R1016). NEBNext Ultra II Directional RNA Library Prep Kit for Illumina used for library preparation with rRNA depletion.

Quantification of Retroelements

To quantify retroelements (REs) in MCF7 cell lines after EPIC1 or EZH2 knockdown, dsRIP-sequencing reads (pair-end) were mapped to the human genome (build hg38) using bowtie2 (https://github.com/BenLangmead/bowtie2). Quantification is performed based on RepeatMasker annotation (build v4.0.5) (https://www.repeatmasker.org/species/hg.html) using RepEnrich2 (48) (https://github.com/nerettilab/RepEnrich2). Quality control and mapping information are provided in data file S3. The estimated count per RE gene for each sample was logarithmically transformed with a pseudo count of 1 being added. The IP enrichment of an RE gene in a specific condition (i.e., siEPIC1/siEZH2/siControl treatment) is defined as the fold change of its expression level in IP versus INPUT. To identify siEPIC1 or siEZH2-induced REs, we require an RE gene to meet the following criteria: (1) with expression level in IP sample higher than 5; (2) with IP enrichment higher than 2-fold; (3) with treatment enrichment score above the average; (4) with treatment contribution score higher than 0. Among them, the treatment enrichment score is defined as:

$$EnrichmentScore={IPenrichment}_{siTreatment}-{IPenrichment}_{siControl}$$

The treatment contribution score is defined as:

$$ContributionScore=\left|{IP}_{siTreatment}-{IP}_{siControl}\right|-\left|{INPUT}_{siTreatment}-{INPUT}_{siControl}\right|$$

Single-cell RNA sequencing (scRNA-seq) analysis

For single-cell analysis, five tumors from five individual humanized mice subjected to four different treatments were used. Single-cell suspensions were prepared by mechanical disruption of tumors using scissors and incubating in tumor dissociation buffer (RPMI 1640 w/o serum supplemented with DNase I (Roche, #10104159001) 0.3 mg/mL and Liberase TL (Roche, #05401020001) 0.25 mg/mL) for 30 min at 37°C, and then dispersed through 70 μm cell strainers. After washing, hCD45⁺ cells were isolated from tumor cell suspensions using human CD45 MicroBeads (Miltenyi Biotec, #130–045-801), according to the manufacturer’s instructions. Mouse cells were depleted from the hCD45⁻ fractions using the Mouse Cell Depletion Kit (Miltenyi Biotec, #130–104-694) according to the manufacturer’s instructions to isolate non-immune human cells. The isolated human CD45⁺ and human tumor cells from each tumor were labeled using individual hashtag antibodies (cell hashing). Anti-human hashtag antibodies #3, 6, 9, 12, 14 were used to label hCD45⁺ cells and anti-human hashtag antibodies #4, 7, 10, 13, 15 were used to label human tumor cells respectively. The details of the hashtag antibodies used in this experiment are listed in table S6. After cell hashing, the cells were counted, pooled, and proceeded to library preparation. Pooled cell suspensions were captured on a 10x genomics chromium controller using single-cell 5PrimeV2 chemistry (10x Genomics, Chromium Next GEM Single Cell 5’ Reagent Kits v2 - Dual Index). Gene expression (GEX), TCR and BCR libraries were dual indexed using SI-TT indices from the Dual Index Kit TT Set A, 96 rxns PN-1000215 (10x genomics). Antibody cell surface protein feature barcode containing hashtags dual indexed using SI-TT indices from the Dual Index Kit TN Set A, 96 rxns PN-1000250 (10x genomics). Libraries were sequenced using the Illumina PE150 technology. For the final analysis we considered only the hCD45⁺ group samples because of the low frequency of tumor cells in each group.

Single-cell data processing and analysis

The Cell Ranger v 7.0.1 was used to process FASTQ files using the GRCh38 database as the human reference genome. The generated UMI count gene-by-cell matrix was then demultiplexed and analyzed using the “Seurat” R package (v 4.3.0) (49). To filter out low-quality cells and cell doublets, cells with uniquely detected genes ranging from 200 to 2500 were selected. Low-quality or dying cells with more than 15% of reads that are mapped to the mitochondrial genome were excluded. Gene expression was then normalized and scaled prior to the principal component analysis (PCA) on the highly variable features. The first 20 PCs were used for clustering analysis by “FindClusters” function with a resolution 0.5. The clustering results were visualized by UMAP plots, followed by differentially expressed gene analysis using Wilcoxon rank sum test in “FindAllMarkers” function. Based on the differentially expressed genes identified for each cluster, we annotated the clusters with immune cell types according to the presence of canonic markers within the top 100 genes (table S7). The processed scRNA-seq data can be accessed through GSE229723.

Statistical analysis

For the correlation analysis, both Pearson and Spearman’s correlations were applied to avoid potential conflicts on the linearity assumption. For enrichment and exclusiveness, a pre-rank gene set enrichment analysis (50) was applied to assess the enrichment of specific features in single samples. For survival analysis, both the Cox Proportional Hazard model and log-rank test were used to compare prognosis between the groups. Statistical analyses were performed using Python (version 3.8.0) and R (version 4.1.1) software. One-end Fisher exact test used in scRNA-seq data. Student’s two-tailed t test was used to determining significance values between the various experimental groups in all other experimental results. P values less than 0.05 were used to find statistically significant differences. The statistical methods used are stated in the figure legends.

Supplementary Material

Figs. S1 to S5.

Tables S1 to S7.

Data files S1 to S4.

Data and materials availability:

The RNA-seq and scRNA-seq data used to support the present study have been deposited in the Gene Expression Omnibus under accession numbers GSE229720, GSE229721, GSE229722, and GSE229723. Raw and normalized expression data for TCGA samples were obtained from the NCI Genomic Data Commons Data Portal. Any code used in this manuscript will be made available upon request. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. Further information and requests for resources and reagents should be directed to the lead authors, M.Z. (miz45@pitt.edu) and D.Y. (dyang@pitt.edu).