Type I Interferon Regulates a Coordinated Gene Network to Enhance Cytotoxic T Cell–Mediated Tumor Killing

The cellular and noncellular components of the tumoral niche create a specific tumor microenvironment (TME), where cells continuously sense danger and damage signaling by extracellular and intracellular pattern recognition receptors (PRR) to balance the acts of the host immune system (1). One key event is the production of type I interferons (IFN) in response to the activation of specific PRRs, such as cyclic GMP–AMP synthase, which plays a central role in regulating TME (2). IFNs activate the JAK–STAT pathway, resulting in the induction of hundreds of IFN-stimulated genes (ISG; ref. 3), many of which remain uncharacterized (4). In general, it is thought that each of these ISGs may activate a pathway, and multiple activated pathways collectively contribute to the overall antitumor effect. However, little is known regarding which ISGs are key drivers to the development of the antitumor effect and how these ISGs might coordinate with one another.

Type I IFNs have been well established to orchestrate both innate and adaptive immune response in cancer (2, 5, 6). This is accomplished by mobilizing diverse cell types in the host immune system, including natural killer cells, dendritic cells, tumor-associated macrophages (TAM), B cells, and T cells (5–7). A critical step in gaining antitumor immunity is to increase the traffic of activated effector T cells, such as cytotoxic CD8⁺ T cells, to infiltrate into the tumor bed to recognize and kill target cancer cells (2, 7). While playing a central role in regulating T-cell priming and homing in the TME (2), excessive IFN signaling is also known to modulate the TME to disable the antitumor function of effector T cells (8). These complex biological responses likely result from pleiotropic actions of ISGs and various feedback control mechanisms, which appear to be highly dependent on specific tumor contexts (4, 9). Therefore, an important goal in enhancing antitumor immune response is to identify key ISG-mediated pathways that favor the development of an antitumor microenvironment, while minimizing the impact of those with adverse effects.

One such ISG, UBA7, encodes an ubiquitin-like modifier-activating enzyme (UBA7, also known as UBE1L) that catalyzes the covalent conjugation of the ubiquitin-like protein product of another ISG (ISG15) to additional ISGs as well as other cellular proteins in a process termed “protein ISGylation” (10). ISG15 is conjugated to a host of proteins in response to propagate the effects of IFN stimulation and is known to play a role in antiviral immunity, but the cellular consequence of UBA7-dependent ISGylation in the setting of cancer cells is not completely understood. Here, we uncover a regulatory network in which multiple UBA7-targeted ISGs act in a concerted fashion to activate key chemokines and attract cytotoxic T cells to the tumor bed. These findings reveal a synergistic ISG-mediated cellular program to achieve effective antitumor response in the host.

Type I IFNs have been well established to have immune-stimulatory functions against malignancies, including breast cancer. To identify ISGs that contribute to IFN's effect in preventing breast cancer progression, we examined the prognostic value of 81 individual ISGs showing correlated expression with ISG15 and another ISG, MX1, because both of them are highly induced by IFN (Supplementary Fig. S1A and S1B). We found that the expression of a subset of ISGs was significantly correlated with favorable relapse-free survival (RFS; Supplementary Fig. S1B; Supplementary Table S1). Interestingly, one of the top ISGs positively correlated with patient survival is UBA7 (Fig. 1A; Supplementary Fig. S1B). This is in contrast to the lack of positive prognostic value for MX1 (Supplementary Table S1), supporting the idea that not all ISGs positively contribute to antitumor immunity. Interestingly, ISG15, encoding the ubiquitin-like donor for UBA7-mediated ISGylation, was also known to be highly elevated in breast cancer (11). Unlike UBA7, however, the induction of ISG15 was negatively correlated with patient survival (Supplementary Table S1). This likely reflects the multifaceted function of ISG15, as free ISG15 is also responsible for attenuating IFN signaling by stabilizing the deISGylation protease USP18 (12). Importantly, the expression of UBA7 was associated with favorable RFS in triple-negative breast cancer (Fig. 1B). The association of UBA7 expression with patient survival was also validated in The Cancer Genome Atlas (TCGA) dataset of patients with breast cancer (Supplementary Fig. S1C). Collectively, these results strongly suggest that UBA7 may function as a critical tumor-suppressive ISG in breast cancer.

We next modeled the tumor-suppressive function of UBA7 in mice to establish its functional relevance. Previous studies have demonstrated that enhanced IFN signaling significantly changes the TME by increasing T-lymphocyte infiltration, thereby causing significant shrinkage of tumors in the mouse mammary tumor virus (MMTV)–polyomavirus middle tumor antigen (PyVmT) breast cancer model (13, 14). We therefore utilized this well-established mouse model to study the function of UBA7 in breast cancer. In line with a published RNA-sequencing (RNA-seq) dataset (15), we detected increased expression of both Uba7 and Isg15 in tumors compared with matched controls at week 10 (Fig. 1C), indicating augmented IFN signaling during tumor progression. Knockout (KO) of Uba7 in mice did not appear to affect normal mammary development (Supplementary Fig. S2). Although both wild-type (WT) and Uba7-deleted (KO) MMTV–PyVmT mice developed palpable tumors around week 8, we found that tumors grew faster in KO mice compared with WT mice from week 9 to the endpoint when the mice were euthanized, as indicated by the number of mice that developed tumors more than 0.5 cm (Fig. 1D and E). At week 10, for instance, 90% of KO mice developed large tumors more than 0.5 cm compared with 40% of WT mice (Fig. 1E), and the average volume of the largest tumors in KO mice was approximately 380 mm³ compared with approximately 100 mm³ in WT mice (Fig. 1F). Given that UBA7 is the activating enzyme for protein ISGylation, we detected both free ISG15 and ISG15-conjugated proteins in tumor lysates from WT mice at this stage, indicative of active protein ISGylation, whereas only free ISG15 was detected in tumor lysates from KO mice (Fig. 1G). This agrees with the idea that UBA7 is a major activating enzyme responsible for tumor-induced protein ISGylation. We also monitored total tumor burden on these animals by quantifying total tumor weight normalized to body weight at the endpoint and found that KO mice had more relatively large tumors (those with tumor burden more than 0.2) than WT mice (Fig. 1H). In addition, KO mice also showed increased incidence of lung metastasis (Fig. 1I and J). Collectively, our data on Uba7 KO mice revealed a tumor-suppressive function of protein ISGylation. To further support this finding, we examined USP18^C61A/C61A knock-in (KI) mice in which ISGylation was increased by inactivating the deconjugating enzyme USP18 (16) and found similar suppression to breast tumor growth (Supplementary Fig. S3A–S3C). Together, these results demonstrated that protein ISGylation was tumor-suppressive in the MMTV–PyVmT breast cancer model, which supports the role of UBA7 as a critical tumor-suppressive ISG in human breast cancer.

To understand how protein ISGylation suppressed tumor growth, we analyzed genes that were coexpressed with UBA7 from the TCGA breast cancer dataset. Excluding the expected association of classic ISGs, we found that >600 human genes showed coexpression with UBA7 (Supplementary Fig. S4A, R ≥ 0.30). Ingenuity pathway analysis revealed that the top five canonical pathways are all related to T cells (Supplementary Fig. S4B). Interestingly, the chemokine signaling pathway is also revealed by the analysis, and CXCR3, which is mostly expressed on activated T lymphocytes, showed the highest correlation with UBA7 expression among all examined chemokines and chemokine receptors (Supplementary Fig. S4C). Therefore, we hypothesized that the tumor-suppressive effect of UBA7 might be further reinforced by T lymphocytes in the TME.

To advance this hypothesis, we asked whether the observation made in patients with breast cancer could be mirrored in our MMTV–PyVmT breast cancer model. We first performed detailed analysis of T-cell populations in tumors derived from these mice and observed that CD3⁺CD4⁺ and CD3⁺CD8⁺ T cells were all significantly reduced in tumors from Uba7 KO mice compared with tumors from WT mice (Fig. 2A; Supplementary Fig. S5). In contrast, tumors from KO mice showed an increase in the percentage of CD11b⁺Gr1⁻F4/80⁺ TAMs. No significant difference was observed in the percentage of intratumoral CD11b⁺Gr1⁺ cells, which could be tumor-associated neutrophils and/or myeloid-derived suppressor cells (Fig. 2A). Considering the limited role of TAMs in regulating spontaneous PyVmT tumor growth, as established earlier (17, 18), we chose to focus our efforts on understanding protein ISGylation-regulated T-cell activity in the PyVmT tumor model. Staining for CD69 (a T-cell activation marker, which is also expressed on anergic T cells) identified similar percentages of CD3⁺CD4⁺ and CD3⁺CD8⁺ T cells expressing such marker within tumors from both WT and Uba7 KO groups (Supplementary Fig. S6A). Further analysis revealed that there was no significant difference in the percentage of effector CD44^hiCD62L^lo T cells within the CD3⁺CD4⁺ or CD3⁺CD8⁺ T-cell populations in tumors from WT and KO mice (Supplementary Fig. S6B). Additional analysis further suggested that intratumoral CD8⁺ T cells from Uba7 KO tumors were not compromised based either on cell proliferation or on the expression of the key cytotoxic effector molecule GZMB (Supplementary Fig. S7A and S7B). We interpreted these findings to indicate that UBA7 does not directly modulate T-cell proliferation or function, but rather does so through increasing their infiltration into the tumor bed.

Because the expression of both UBA7 and ISG15 has been detected in tumor cells as well as in their stroma (Supplementary Fig. S8A and S8B), we next asked whether the suppressive TME was due to induced protein ISGylation in tumors versus stromal cells. We first established mouse mammary epithelial cells (MEC) from Uba7 KO tumors. When Uba7 KO MECs were injected into WT and Uba7 KO mice, these cells developed tumors of similar sizes (Fig. 2B). This excludes the functional impairment of any tumor stromal cells due to lack of Uba7 expression in this tumor model. Conversely, restored expression of Uba7 rescued cellular protein ISGylation in MECs (Fig. 2C), which led to smaller tumors and reduced lung metastasis when these MECs were injected into the mouse mammary fat pad of WT mice (Fig. 2D and E). Notably, increased UBA7 expression did not affect cell growth when cultured in vitro (Supplementary Fig. S9). Meanwhile, knockdown of ISG15 in Uba7 KO MECs did not show further impact on tumor growth in vivo (Supplementary Fig. S10A and S10B), implying that intervening free ISG15 in Uba7 KO cells was insufficient to affect tumor growth. Furthermore, restored Uba7 expression greatly increased infiltration of CD3⁺CD8⁺ cytotoxic T lymphocytes in tumors derived from these cells (Fig. 2F), and CD8 deficiency abrogated the tumor-suppressive effect of Uba7-mediated ISGylation (Fig. 2G). In combination, these results strongly implicate a tumor cell–autonomous function of UBA7, suggesting that protein ISGylation may regulate T-lymphocyte activities through a paracrine effect within the TME.

Having established the tumor cell–autonomous effect, we prioritized our experimentation to understand the critical role of protein ISGylation in tumor cells. Considering the highest correlation between the expression of UBA7 and CXCR3 in patients with breast cancer and the established role of the CXCL9/10/11–CXCR3 axis in the regulation of T lymphocyte trafficking and antitumor immunity (2, 7, 19), we postulated that UBA7-mediated protein ISGylation might directly modulate the CXCL9/10/11–CXCR3 signaling axis within the TME. Therefore, we next examined whether protein ISGylation regulates the production of CXCR3 ligands. Given the documented upregulation of Toll-like receptor (TLR) signaling molecules (i.e., TLR3 and TLR4) in the MMTV–PyVmT breast cancer model (20) and their known functions in shaping anticancer immunity in human breast cancer (21, 22), we first focused on TLR4 signaling and found that TLR4 activation in combination with type I IFN priming greatly stimulated the expression of Cxcl9/10/11 (Supplementary Fig. S11). The observed synergy between IFN and TLR4 signaling prompted us to determine whether ISGylation directly contributed to the induction of CXCR3 ligands in our MEC model. Indeed, we found that UBA7-mediated ISGylation greatly enhanced the expression of Cxcl9/10/11 in response to TLR4 signaling in IFN-primed MECs (Fig. 3A). We further confirmed an ISGylation-dependent increase in CXCL9/10 protein secretion from mouse MECs after TLR4 activation in IFN-primed cells (Fig. 3B), although the CXCL11 protein was not detected under our conditions. We further showed that protein ISGylation also increased poly(I:C)–stimulated expression of Cxcl9/10 (Fig. 3C and D), indicating that ISGylation synergized with both TLR3 and TLR4 signaling to increase the expression of CXCR3 ligands. In contrast, we detected little effect of ISGylation on TNFα, IL1β, or poly(dA:dT)–stimulated expression of CXCR3 ligands under investigated conditions (Supplementary Fig. S12A–S12C). Consistent with the importance of UBA7-catalyzed ISGylation in TLR-stimulated expression of Cxcl9/10, we found that the expression of an enzyme-deficient form of UBA7 (C586A) failed to increase the expression of CXCR3 ligands (Fig. 3E and F). Furthermore, CXCR3 blockade abrogated the tumor-suppressive effect of ISGylation (Fig. 3G). Together, these results demonstrated a vital role for cancer-autonomous protein ISGylation in facilitating cytotoxic T cell–mediated killing in breast cancer through the increased expression of CXCR3 ligands, as illustrated (Fig. 3H).

With the established importance of the CXCR3 signaling axis in protein ISGylation–regulated tumor suppression, we next investigated the molecular mechanisms underlying ISGylation-facilitated CXCR3 ligand expression in response to TLR signaling in IFN-primed cells. Both TLR3 and TLR4 signaling are known to activate NFκB, MAPK, and TBK1/IKKϵ signaling to induce expression of many genes (23). In our experimental system, ISGylation induced by IFN priming did not seem to influence TLR4 activation–induced p38, JNK, or IRF3 phosphorylation (Supplementary Fig. S13A). Although we did observe ISGylation impaired ERK phosphorylation and increased the abundance of IKBα (an inhibitor of NFκB activation; Supplementary Fig. S13A), treatment with an ERK inhibitor, U0126, did not increase Cxcl9/10/11 expression (Supplementary Fig. S13B). These data imply that protein ISGylation may not activate TLR4 signaling through any of the cytoplasmic TLR4 proximal signaling cascades.

Although it remains unclear exactly how TLR activation licenses gene activation in IFN-primed cells, we decided to first focus on understanding how ISGylation contributes to the IFN-primed state, which is known to be critical for induction of CXCR3 ligands. In this regard, we noted that transcription factors STAT1 and STAT2 have been implicated in the regulation of CXCL9/10/11 expression. Although nuclear translocation is the primary mechanism for STAT activation in response to IFN signaling, we found that protein ISGylation had little influence on nuclear translocation of both STAT1 and STAT2 (Supplementary Fig. S14A). This directed us to investigate the possibility that ISGylation might regulate the function of STAT1 and STAT2 within the nucleus. Interestingly, both of these STATs are themselves ISGs and subject to protein ISGylation in response to IFN treatment (Supplementary Fig. S14B). Measuring the nuclear distribution of STAT1 by stochastic optical reconstruction microscopy, we observed the ISGylation-dependent redistribution of STAT1 to form clusters of various size and density in the nucleus (Supplementary Fig. S15A–S15D). We further observed increased interaction between nuclear STAT1 and STAT2 in the presence of ISGylation (Supplementary Fig. S15E). Importantly, ISGylation significantly increased STAT2 binding to IFN-stimulated response elements (ISRE) in Cxcl9/10 enhancers, and a similar trend was observed for STAT1 (Fig. 4A; Supplementary Fig. S15F). To determine the global impact of ISGylation on gene expression, we determined ISGylation-regulated gene expression by RNA-seq and found that the expression of a subset of IFN–TLR4 signaling–regulated genes was indeed enhanced by protein ISGylation (Supplementary Fig. S16A and S16B; Supplementary Table S2), some of which was further confirmed by RT-qPCR (Supplementary Fig. S16C). Using specific pharmaceutical inhibitors, we further found that the expression of CXCR3 ligands required coactivation of multiple signaling pathways, including the TBK1/IKKϵ and JAK–STAT pathways (Supplementary Fig. S17A and S17B). These data therefore attributed a subset of TLRs signaling–induced gene-expression events, including Cxcl9 and Cxcl10, to ISGylation-induced clustering of STAT1 and STAT2 in the nucleus of IFN-primed cells.

Given the substantial role of protein ISGylation in gene transcription, we further examined more detailed mechanisms. The observed clustering of both STAT1 and STAT2 prompted us to consider a role for ISGylation in mediating phase separation of the transcriptional machinery, a concept recently suggested to play critical roles in controlling compartmentalization and gene expression in the nucleus (24). Notably, promyelocytic leukemia (PML) is an ISG, and PML nuclear bodies are membrane-less organelles, which can be induced by IFN signaling (25). We therefore hypothesized that protein ISGylation–mediated complex formation of STAT1 and STAT2 might also show increased association with IFN-induced PML bodies. Indeed, type I IFN dramatically induced the formation of PML bodies in MECs, although protein ISGylation did not affect the number of PML bodies (Supplementary Fig. S18A–S18C). Importantly, we found that protein ISGylation facilitated condensation of STAT1 and STAT2, especially STAT2, around IFN-induced PML bodies by expansion microscopy (Fig. 4B–D). These observations suggest that ISGylation facilitates nuclear reprogramming.

To test the functional relevance of such IFN-induced nuclear compartmentalization, we took advantage of arsenic-mediated PML degradation, as reported earlier (26), and observed that short-term treatment with arsenic trioxide depleted cellular PML protein without affecting the levels of STAT1, STAT2, or ISGylation (Supplementary Fig. S19). Importantly, however, such treatment diminished ISGylation-dependent expression of Cxcl9/10/11 (Fig. 4E). However, Ccl2, whose expression is also regulated by IFN–TLR4 signaling, was not affected by either protein ISGylation or arsenic treatment (Fig. 4E), indicating that the specificity of ISGylation and PML controlled chemokine expression in our current biological context. These data thus provide evidence for STAT-dependent expression of CXCR3 ligands as a functional consequence of ISGylation-induced STAT clustering and redistribution in the nucleus. Importantly, these findings are corroborated by multiple clinical results, showing a significant association between the expression of both UBA7 and PML with CXCL9/10/11 expression and favorable overall survival (Supplementary Fig. S20A and S20B) in patients with breast cancer. Together, these results highlighted the importance of the cooperation between ISGylation and PML in facilitating expression of CXCR3 ligands, whose expressions are required for ISGylation-regulated tumor suppression.

Type I IFNs activate expression of hundreds of ISGs and regulate tumorigenesis and response to cancer therapy through both tumor cell–intrinsic and tumor cell–extrinsic mechanisms (5, 6). However, the roles of many individual ISGs in cancer have remained unclear. Therefore, understanding the functional interplay between ISGs may provide novel therapeutic interventions in treating human cancers. Here, we identified the ISGylation-activating enzyme UBA7 as a key ISG in regulating malignant progression. Interestingly, nearly all of the enzymes involved in protein ISGylation, including E1, E2, E3, and the deconjugating enzyme, are ISGs themselves, suggesting their tight link to IFN-regulated cellular functions. Protein ISGylation thus affords a unique angle to look into IFN-induced gene networks that may synergistically function in the development of antitumor activities. Our current work uncovers that the IFN-inducible protein ISGylation network has an immunostimulatory function that works by facilitating CXCR3 ligand expression and cytotoxic T-cell infiltration in orchestrating a tumor-suppressive microenvironment. It is worth noting that ISGylation-induced TME also had decreased TAMs. Although TAMs are known to have a limited role in regulating spontaneous and allograft PyVmT tumor growth (17, 18, 27), they do play important roles in regulating metastatic potential (18) and response to cancer therapies in PyVmT tumors (17, 27). Thus, future studies may be geared toward understanding the contribution of protein ISGylation to regulated TAMs during tumor metastasis and/or response to cancer therapies.

Several in vitro studies on human cancer cell lines have revealed both positive and negative roles of protein ISGylation in breast cancer (28, 29). For example, ISG15 and protein ISGylation have been implicated in promoting invasive properties of malignant breast cancer cell lines through modulation of cytoskeletal architecture (29). On the other hand, ISGylation of ΔNp63α appears to ablate the ability of ΔNp63α to promote anchorage-independent cell growth and tumor formation (28). Although these findings collectively suggest context-dependent effects of protein ISGylation in cancer, it is important to keep in mind that IFN responses are part of the adaptive and innate immune systems in the host, thus emphasizing the importance of studying the impact of protein ISGylation on cancer progression in immunocompetent hosts. Using immunocompetent hosts, our current work highlights the importance of protein ISGylation in orchestrating a tumor-suppressive microenvironment. This mechanism is well in line with our analysis of human cancer transcriptomes and associated clinical outcomes. Although it is possible that ISGylation may enhance or suppress tumorigenesis in different contexts, just like IFN treatment, our data provide direct evidence to support the function of UBA7 as a key tumor suppressor in breast cancer. This is further corroborated by frequent deletions of chromosome 3p21.3 where UBA7 resides in human cancer (30). Our findings suggest a mechanism underlying these genetic alterations in cancer.

We observed that the presence of protein ISGylation facilitates TLR3/4 signaling–activated CXCR3 ligand expression in our current model. TLR4 signaling is activated by a variety of ligands, including DAMPs (22, 31), and interestingly certain ISGs (e.g., IFI35, NMI) are known to function as DAMPs (32). In addition, women with breast cancer have higher relative abundance of Bacillus, Enterobacteriaceae, and Staphylococcus in their breast tissues (33), and TLR3 signaling–stimulated expressions of CXCR3 ligands have also been well characterized to play critical roles in regulating TME and antitumor immunity (19, 34). These prior observations are fully in line with the impact of loss-of-function TLR3 or TLR4 alleles in patients with breast cancer who relapse more quickly and show reduced overall survival after therapeutic intervention (21, 22), together indicating the importance of both TLRs in modulating anticancer immunity. Although we did not observe ISGylation-facilitated chemokine expression in the presence of poly(dA:dT) under investigated conditions, we cannot rule out the possibility that ISGylation may still contribute to signaling activation by other various endogenous stimuli, including some form of DNA released from damaged tissues, in the TME.

Our mechanistic studies have linked IFN priming to the dynamic regulation of PML bodies. PML is encoded by an ISG, and IFN signaling is known to induce the formation of PML bodies in the nucleus (25). PML has been implicated in the regulation of type I IFN–induced ISGs in cell lines, albeit with unknown mechanisms (35). Using superresolution microscopy, our current studies reveal that protein ISGylation enhances condensation of the critical transcription factors STAT1 and STAT2 and their association with PML bodies. We further observe that ISGylation-dependent chemokine expression requires the presence of PML nuclear bodies, together supporting a model in which PML nuclear bodies may contribute to the formation of a favorable nuclear environment for the expression of specific genes (36). It is entirely possible that microscopically detected PML bodies are indicative of IFN-induced condensation of both visible and invisible PML-containing complexes in the nucleus, which may underlie the IFN-induced and ISGylation-mediated gene-expression program. Our findings join the recent discovery that SUMOylation and the SUMO Interaction Motif in PML bodies contribute to multivalent interactions within PML bodies (25, 37), together suggesting that multiple post-transcriptional modifications are involved in modulating phase separation to account for the dynamic compositional control of PML bodies. It is important to point out that PML has been shown to function as a tumor suppressor in several cancer types, including breast cancer (38), but elevated PML expression has also been connected to the acquisition of aggressiveness and metastatic features in breast tumors (39). In addition, the importance of PML in regulating the immune response has been reported in a mouse model of prostate cancer (40). These seemingly contradictory observations might reflect the winning versus losing battles of the host immune system to different tumors.

The most important concept conveyed in this work is a cooperative gene network of ISGs, which is shaped by UBA7-mediated ISGylation to form a favorable nuclear environment for STAT1/2-mediated chemokine expression. These cancer cell–autonomous effects provide mechanistic insights into the regulation of cytotoxic T-cell infiltration into such TME when PRR-triggered type I IFN signaling is in synergy with activation of other inflammatory pathways, such as DAMP-induced TLR activation. Previous chromatin immunoprecipitation (ChIP) sequencing studies have shown binding of AP1, NFκB, STAT1/2, and IRF3 on enhancers of CXCR3 ligands in TLR-activated cells (41). Therefore, protein ISGylation–regulated clustering of STATs may facilitate their assembly on ISREs to form “enhanceosomes” (42), thereby inducing gene expression according to our working model (Supplementary Fig. S21). Importantly, by dissecting the functions of some key ISGs in this network, we suggest a potential strategy to enhance the ISGylation-induced anticancer effect through deactivating its negative regulator USP18. Such enhanced ISGylation may also help divert tumor-prone activities of ISG15 by reducing free ISG15 in the TME. This and other potential modulations of the ISG network may therefore amplify the antitumor effects of IFN-based cancer therapies, while limiting other cancer-promoting events.

All experimental protocols were approved by the Institutional Animal Care and Use Committee of the University of California, San Diego (UCSD). All mice were housed and bred at the vivarium of the Moores Cancer Center at UCSD. Uba7 KO mice were back-crossed to FVB mice over ten generations. CD8 KO mice with FVB background were kindly provided by Dr. Michael Karin (UCSD). USP18^C61A/C61Amice in C57BL/6 background were generated by Dr. Knobeloch Klaus-Peter (University of Freiburg). FVB mice expressing PyVmT under the control of the mouse MMTV-LTR were crossed with WT or Uba7 KO mice. Female mice heterozygous for the PyVmT transgene and homozygous-null or homozygous WT for Uba7 were studied. C57BL/6 mice expressing PyVmT under the control of the mouse MMTV-LTR were crossed with WT or USP18^C61A/C61A mice. Female mice heterozygous for the PyVmT transgene and WT or homozygous for USP18^C61A/C61A were studied. Female mice were palpated twice weekly to monitor the development of mammary tumors. Tumor-bearing mice were sacrificed for analysis at endpoint, which was 13 weeks (FVB) and 22 weeks (C57BL/6) of age after birth or whenever tumor burden reached 20%. Approximate tumor volumes were calculated by the formula 4/3 × 3.14 × (long diameter/2) × (short diameter/2)². For metastasis studies, mouse lungs were harvested at endpoint, and the number of lung metastases was determined by hematoxylin and eosin staining of ten serial lung sections per mouse. All studies were carried out following the NIH guidelines for the care and treatment of experimental laboratory rodents.

Single-cell suspensions were established from spontaneous mammary tumors arising in FVB MMTV–PyVmT females as previously described (43). Tumors were minced and incubated at 37°C for 2 hours in 5 mL of Ham's F-12K medium containing 1 mg/mL collagenase (Roche), 2 mg/mL soybean trypsin inhibitor (Sigma), and 2% BSA. After addition of fetal calf serum (FCS)–containing medium, the suspension was passed through a 70-mm nylon filter (Fisher Scientific). Single cells were pelleted by centrifugation and resuspended in PBS/2% BSA for immediate flow cytometric analysis. Stained cells were analyzed by a FACS Canto cytometer (BD Biosciences). Antibodies for flow cytometric analysis are listed in the Supplementary Table S3. For establishment of MEC cell lines, cells were cultured in DMEM/F-12 (1:1) medium mix containing 5% FCS, 2.5 mg/mL amphotericin B, 10 mg/mL penicillin–streptomycin, and MITO+ (BD Biosciences). Nonepithelial cells were removed by differential passaging, and epithelial origin of established cell lines was confirmed by CD326 (Biolegend) staining after five passages (CD326⁺ > 98%). The cell lines have been tested as Mycoplasma-free using Mycosensor PCR assay kit (Agilent Technologies), with the latest test in 2018. The cell-line authentication was not routinely performed. The cells were used within 2 months after thawing.

In vitro–cultured MECs (1 × 10⁶) resuspended in 1 mg/mL Matrigel (Corning) solution were injected into the abdominal mammary fat pads of female mice. Growing tumors were monitored once a week with a caliper. Approximate tumor volumes were calculated by the formula 4/3 × 3.14 × (long diameter/2) × (short diameter/2)².

Universal type I IFN was purchased from PBL Biomedical Laboratories. Cells were generally treated by IFN at 1,000 U/mL or as specified. Recombinant murine TNFα and IL1β were purchased from Peprotech. LPS was purchased from List Biological Laboratories, Inc. MEK/ERK inhibitor U0126 and TBK1/IKKϵ inhibitor MRT67307 were purchased from Selleckchem. Polyinosinic-polycytidylic acid sodium salt [poly(I:C)], Arsenic(III) Oxide and JAK inhibitor I were purchased from Sigma-Aldrich. Poly(dA:dT)/LyoVec was purchased from Invivogen. Antibodies used in Western blot analyses, Tubulin (Sigma-Aldrich), β-actin (Sigma-Aldrich), p-p38 (Cell Signaling Technology), p-JNK (Cell Signaling Technology), and IkBα (Biolegend), were purchased from the respective manufacturers. Rabbit anti-mouse ISG15 polyclonal antibodies have been described previously (44). Fc blocker (anti-CD16/32) was purchased from eBioscience. A full list of antibodies is shown in Supplementary Table S3.

FLAG-mUBA7 and mUBA7 C586A cDNA was cloned into pMSCV-puro (Clontech). All the constructs were confirmed by DNA sequencing. shRNA plasmids (pLKO.1 vectors) for mIsg15 (TRCN0000077333 and TRCN TRCN0000077334) and control were obtained from The RNAi Consortium.

The 293T cells have been purchased from the ATCC and authenticated by this organization with certificates and been tested as Mycoplasma-free using Mycosensor PCR assay kit (Agilent Technologies), with the latest test in 2018. HEK293T cells were cultured in DMEM (Invitrogen) with 10% bovine calf serum (HyClone), 2 mmol/L l-glutamine (Invitrogen), and penicillin–streptomycin (100 U/mL, Invitrogen) and were used within 2 months after thawing. Cells were transfected using polyethylenimine (PEI) reagent as previously reported (45). For Uba7 expression in Uba7 KO MECs, 293T cells were transfected with pMSCV-mUBA7 and pIK6.1MCV ecopac using PEI reagent. Retroviruses from the culture medium of these cells were infected to Uba7 KO MECs, and stable infected cells were selected by puromycin (1 μg/mL).

Total RNA from tumor tissue or MECs was isolated using Trizol reagent (ThermoFisher) according to the manufacturer's instructions. For RT-qPCR analyses, equal amounts of RNA were reverse-transcribed by First Strand cDNA Synthesis Kit (MCLAB), and the resulting cDNA templates were subjected to RT-qPCR using SYBR Green detection system (Kapa Biosystems) on the CFX96 thermal cycler (Bio-Rad). Primer sequences are listed in Supplementary Table S4.

ChIP was performed as described previously (46). A total of 1 μg anti-STAT2 (Cell Signaling Technology) and 2 μg anti-STAT1 (Santa Cruz Biotechnology) antibodies were used per ChIP reaction. ChIP data were expressed as a percentage of input and presented as the relative fold enrichment compared with mock control. Primers used for qPCR of ISRE regions of CXCL9/10 are listed in Supplementary Table S5.

For CXCL9/10/11 ELISA, MECs were seeded onto a 6-well plate, and supernatants were collected at different time points as indicated, spun down to remove debris, and analyzed according to the manufacturer's protocol (R&D Systems).

Cells were cultured on 18 × 18 mm-1.5 microscope cover glass (Fisher Scientific) in 12-well plates. After IFN treatment, cells were fixed in 3.7% formaldehyde, permeabilized with 0.1% Triton X-100, and blocked with 2 mg/mL BSA. Samples were incubated with primary antibodies against STAT1 (2 μg/mL), STAT2 (2 μg/mL), and PML (10 μg/mL) in blocking buffer at 4°C overnight. The procedure for sample gelation, digestion, and expansion was conducted as previously reported (47–49). Briefly, the samples were incubated in 0.01% acrylic acid N-hydroxysuccinimide ester (Santa Cruz Biotechnology) in PBS for 2.5 hours. A monomer solution [1x PBS, 2 mol/L NaCl, 25% sodium acrylate (w/w), 4% acrylamide (w/w), 0.04% (w/w) N,N′-methylenebisacrylamide] was mixed with 0.2% (w/w) ammonium persulfate (APS) and 0.2% (w/w) tetramethylethylenediamine (TEMED) to promote polymerization. The monomer solution mixed with the APS and TEMED was then added to samples to a depth of 2 mm and incubated for 2 hours at 37°C to allow for the completed polymerization. The size of the gel was measured. The polymerized gel was then immersed in 4 mL of 0.2 mg/mL proteinase K (Roche) in digestion buffer (50 mmol/L Tris, pH 8, 2 mmol/L CaCl₂, 3 mol/L NaCl, 0.5% Triton X-100, and 0.8 mol/L guanidine HCl) and incubated at 37°C for 3 hours. Digested gels were next placed in excess volumes of PBS for 2 hours. The gels were then stained with Alexa488- or Alexa568-conjugated secondary antibodies at room temperature overnight, followed by washing in10 mmol/L Tris buffer (pH 8.0), which achieved expansion ratio around 4 in each dimension. The expanded hydrogels were then placed onto poly-l-lysine–coated cover glass for confocal imaging.

The confocal data were analyzed by ImageJ/Fiji. For each z-stack image, the nucleus region was first identified by Gaussian blurring the DAPI channel with 1 μm interval followed by thresholding and binarization. Then, PML bodies in each nucleus were identified by Gaussian blurring the PML channel signals with interval of 140 nm, followed by thresholding and binarization. STAT1/STAT2 signals within and outside of the PML body regions were compared, with the condensation index defined as ratio of the mean intensity within PML bodies to that outside of PML bodies in nuclei. A total of 9 to 11 cells in each sample conditions were analyzed.

Experiments were repeated at least twice. Results are expressed as mean + SD or SEM as indicated. Statistical significance was evaluated by a paired or unpaired t test, or by one-way/two-way ANOVA as indicated in the figure legends, using P < 0.05 as indicative of statistical significance. Linear regression analyses in breast-invasive carcinoma were performed by RStudio using TCGA datasets. Data for ISG15-, MX1-, and UBA7-correlated genes were downloaded from cBioPotal (50, 51). The gene-expression data and survival information of 3554 patients were downloaded from the Gene Expression Omnibus (GEO) and analyzed as previously reported (52). Survival analysis was performed by drawing a Kaplan–Meier survival plot and computing a Cox proportional hazard regression as described previously (53). RNA-seq data analyzed in this study are available from GEO (GSE112532).

No potential conflicts of interest were disclosed.

Conception and design: J.-B. Fan, C. Burkart, D.-E. Zhang

Development of methodology: J.-B. Fan, S. Miyauchi, D.-E. Zhang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.-B. Fan, S. Miyauchi, H.-Z. Xu, D. Liu, H. Cheng, M. Yan, K.-P. Knobeloch, J.N. Rich, D.-E. Zhang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.-B. Fan, S. Miyauchi, H.-Z. Xu, L.J.Y. Kim, K.-i. Arimoto, M. Yan, Y. Zhou, B. Győrffy, D.-E. Zhang

Writing, review, and/or revision of the manuscript: J.-B. Fan, B. Győrffy, J.N. Rich, X.-D. Fu, D.-E. Zhang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.-E. Zhang

Study supervision: J.-B. Fan, H. Cang, D.-E. Zhang

We thank Dr. Sam Stoner for critical editing of this manuscript, and Dr. Judith Varner and all the members of the D.-E. Zhang laboratory for discussion, suggestions, and help. We gratefully acknowledge Drs. Ernest Borden and Deborah Lenschow for ISG15 antibodies, Dr. Lesley G. Ellies for C57BL/6 and FVB/N-Tg (MMTV–PyVmT) 634Mul/J mice, and Dr. Michael Karin for CD8 KO mice. This work was supported by funding from the NIH New Innovator Award to H. Cang (1-DP2-EB020400) and funding from NIH to D.-E. Zhang (R01CA177305).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.