A metabolic CRISPR/Cas9 screen identifies HO-1 as a potent IFN-I production inhibitor in response to RT. Tumor metabolism and related enzymes are intimately connected with the IFN-mediated immune response (23, 24). To identify key metabolic genes regulating IFN-I production during RT, we used a metabolic gene-KO library targeting 2,981 genes with 29,790 single-guide RNAs (sgRNA) for a systematic screen (25). First, the nasopharyngeal carcinoma (NPC) cell line HK1 was stably integrated with a reporter cassette containing mCherry driven by IFN-stimulatory response elements (ISREs) and the IFN-β promoter to visualize and quantify IFN-I production (Figure 1A). Then, the reporter cells were treated with RT or cGAMP, demonstrating the high specificity and efficiency of this system (Figure 1B). Moreover, as a negative control, KO of IRF3 almost abolished mCherry expression induced by irradiation (Figure 1B). Next, reporter-expressing cells were transduced with the metabolic CRISPR/Cas9-KO library, treated with RT, and sorted by flow cytometry on the basis of the highest and lowest 30% mCherry fluorescence signals for deep sequencing (Figure 1C and Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/JCI181044DS1).

A metabolic CRISPR/Cas9 screen identifies HO-1 as a potent IFN-I production inhibitor in response to RT. (A) Schematic overview of the mCherry reporter construct. mCherry expression is driven by ISREs followed by the IFN-β promoter. (B) Control HK1 cells and IRF3-KO HK1 cells were stimulated with radiation, cGAMP (10 μM), or IFN-β (100 ng/mL). mCherry reporter expression was further analyzed by flow cytometry. (C) Overview of the CRISPR screen. Reporter-expressing HK1 cells were transduced with the sgRNA library. After radiation, the cells were sorted by flow cytometer according to mCherry expression and divided into the highest 30% and the lowest 30% mCherry-expressing populations. Genomic DNA from the sorted cells was deep sequenced to reveal gRNA enrichment. (D) Distribution of the RRA score of the top hits enriched in the mCherry high expression group versus the low expression group. (E) Volcano plot illustrating the important candidates based on the comparison of high mCherry-expressing group versus the low mCherry-expressing group. P-adj, adjusted P value. (F–H) Reporter expression (F), HLAA (G), and CXCL10 (H) mRNA levels after knocking down the top 10 candidates in a post-RT CRISPR screen. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, by 1-way ANOVA (B and F–H). Data are shown as the mean ± SD (n = 3 biologically independent samples).

We used the robust rank aggregation (RRA) algorithm and identified several genes in the hyperresponsive population (Figure 1, D and E), among which MYC, SCAP, G6PD, and DAK were previously reported to inhibit IFN-I production and downstream signaling (26–29), supporting the validity of our screening strategy. We then validated the top 10 genes with a siRNA (Supplemental Figure 1B). HMOX1 (HO-1), which ranked as the top gene, had the most potent effect on inhibition of IFN-β production after RT, whereas the other genes, except MYC, showed weak effects (Figure 1, F–H, and Supplemental Figure 1C).

Meanwhile, HO-1 had higher expression in patients with NPC who had tumor recurrence after RT, while none of the other top 10 genes in the CRISPR screen were upregulated in the RNA-Seq analysis (Supplemental Figure 1D). Thus, we selected HO-1 for further investigation.

HO-1 inhibits RT-mediated IFN-I production independent of its enzymatic activity. To further explore the role of HO-1 in inhibiting IFN-I production under RT, we constructed HMOX1-KO HK1 and HeLa cell lines using CRISPR/Cas9 (Supplemental Figure 2A). After RT, compared with control cells, we observed increased IFN-β levels in HMOX1-KO cells (Figure 2A). Similar results were observed in the human prostate cancer, breast cancer, and fibrosarcoma cell lines DU145, MDM-MB-231, and HT-1080, respectively (Supplemental Figure 2B). Additionally, the mRNA levels of typical IFN-Is (IFNB1 and IFNA2) and IFN-stimulated genes (ISGs) (HLAA and CXCL10) were upregulated in HMOX1-KO cells after RT (Figure 2B).

HO-1 inhibits RT-mediated IFN-I production. (A) ELISA for IFN-β content in the supernatant of control or HMOX1-KO cells before and after RT. (B) Typical IFN-Is and ISGs mRNA levels of control or HMOX1-KO cells before and after RT. (C) Tumor growth of HMOX1-inducible knockdown HK1 tumors in HuHSC-NCG mice, following with or without RT (10 Gy) (n = 5 in each group). (D–F) RT-qPCR analysis for mRNA levels of typical IFN-I and ISG genes (D) and flow cytometric analysis of CD8+ T cell infiltration (E) and TNF-α and IFN-γ expression of CD8+ T cells (F) in the HK1 model (n = 5 in each group). APC, allophycocyanin. (G) Ifnb1, Ifna4, and Cxcl10 mRNA levels in BMDMs from Hmox1fl/fl and Hmox1fl/fl LyzCre/Cre mice. BMDMs were infected with HSV-1 or VSV. Data are shown as the mean ± SD (A, B, and D–G) and the mean ± SEM (C). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, by 1-way ANOVA (A, B, D, and E), 2-way ANOVA (C), and unpaired, 2-tailed, Student’s t test (F and G). n = 3 biologically independent experiments, unless otherwise stated. Dox, doxycycline.

Next, we sought to verify the in vivo effect of HO-1 on the regulation of IFN-I production and correlated RT efficacy. To this end, we constructed 2 cold tumor models (low CD8+ T cell infiltration) with the murine melanoma cell line B16 and the breast cancer cell line 4T1, and then a hot tumor model (high CD8+ T cell infiltration) with the colon adenocarcinoma cell line MC38. To eliminate confounding effects of stable HMOX1 KO on tumor progression, we established Hmox1-inducible knockdown cell lines based on a doxycycline-inducible shRNA expression system (Supplemental Figure 2C). We found that the induction of Hmox1 knockdown enhanced RT efficacy, leading to decreased tumor volumes in the B16, MC38, and 4T1 models (Supplemental Figure 2D). Moreover, knockdown of Hmox1 also promoted intratumoral IFN-I production in vivo (Supplemental Figure 2E). More important, we also validated our findings in a model based on a human cancer cell line and a reconstructed immune system. Transgenic HK1 cells with inducible HMOX1 knockdown were injected into HuHSC-NCG mice, which were constructed by implanting human hematopoietic stem cells after bone marrow ablation. Under RT, HMOX1 knockdown resulted not only in delayed tumor growth (Figure 2C) but also elevated intratumoral IFN-I and ISG expression (Figure 2D). CD8+T infiltration and function were also boosted after knocking down HMOX1 (Figure 2, E and F).

To further validate the role of HO-1 in other scenarios except for RT, we used virus infection as another essential IFN-I inducer and generated Hmox1fl/fl mice and Hmox1fl/fl LyzCre/Cre mice. HO-1–deficient bone marrow–derived macrophages (BMDMs) exhibited upregulated IFN-I (Ifnb1 and Ifna4) and ISG (Cxcl10) expression levels in response to both herpes simplex virus type 1 (HSV-1) and vesicular stomatitis virus (VSV) (Figure 2G and Supplemental Figure 2F). Besides, HO-1 deficiency enhanced the expression of canonical inflammation molecules (Tnfa and Il6) induced by dsDNA (HSV-1 and HT-DNA) in BMDMs (Supplemental Figure 2, G and H).

IFN-Is can be induced by various signaling pathways. Specifically, cytosolic RNA/DNA or LPS activates the retinoic acid–inducible gene I–like (RIG-I–like) receptors (RLR/mitochondrial antiviral signaling protein [MAVS]), cGAS/STING, or TLR/ TIR domain–containing adaptor inducing IFN-β (TLR/TRIF) pathways, respectively, to promote IFN-I production (30). Silencing key adaptor proteins (STING, MAVS, and TRIF) in the 3 pathways mentioned above revealed that HO-1 impaired STING-mediated, but not MAVS- or TRIF-mediated, IFN-β production upon irradiation (Supplemental Figure 3, A and B). To further confirm the involvement of cGAS/STING signaling, we examined the functional state of important molecules in the pathway after knocking out HMOX1 under RT, and found that HO-1 deficiency enhanced the phosphorylation of STING, TBK1, IRF3, and STAT1 (Figure 3A).

HO-1 inhibits the activity of cGAS and STING under RT independent of its enzymatic activity. (A) Immunoblot analysis of essential molecules in IFN-I signaling from control or HMOX1-KO cells before and after RT. (B) ELISA of cGAMP production of control or HMOX1-KO cells before and after RT. (C) ELISA of IFN-β production in the supernatant of control or HMOX1-KO cells with or without cGAMP stimulation. (D and E) Immunoblot analysis of the indicated proteins from control or HMOX1-KO cells with the indicated treatment. (F and G) ELISA of cGAMP (F) or IFN-β (G) production in HK1 cells treated with the indicated metabolites. (H and I) HMOX1-KO HK1 cells were stably transfected with WT HO-1 or HO-1H25A. ELISA of cGAMP (H) or IFN-β (I) production with or without the indicated stimulation. Data are shown as the mean ± SD. ***P < 0.001 and ****P < 0.0001, by 1-way ANOVA (B, C, and F–I). n = 3 biologically independent experiments. p-, phosphorylated.

The induction of IFN-Is by cGAS/STING signaling involves a chain of protein-protein interactions, the core of which includes cGAS, STING, and TANK-binding kinase 1 (TBK1). In this regard, we further explored the exact molecule inhibited by HO-1 in this pathway. Notably, HO-1 deficiency not only elevated cGAMP production under RT (Figure 3B and Supplemental Figure 3C) but also enhanced IFN-β production under cGAMP treatment (Figure 3C). cGAMP is generated by cGAS from ATP and GTP, with its activity being regulated by dsDNA (31). However, HMOX1 KO did not affect the amount of cytosolic dsDNA after RT or various types of chemotherapeutics (cisplatin, oxaliplatin, etoposide, or doxorubicin), indicating that HO-1 might directly influence cGAS function (Supplemental Figure 3, D and E). Moreover, HMOX1 KO markedly promoted cGAMP-mediated STING and TBK1 phosphorylation (Figure 3D and Supplemental Figure 3F), both of which were critical for activating IRF3 to mediate IFN-β transcription. To further clarify the molecule affected by HO-1, we used 5′ppp-RNA, a typical microbe-associated molecular pattern (MAMP) activating TBK1 via the RIG-I/MAVS pathway and found that HO-1 did not affect TBK1 activation in this scenario (Figure 3E), which suggested that STING might be another direct target of HO-1.

We subsequently investigated whether the effect of HO-1 depended on its enzymatic activity. We found that adding exogenous HO-1 metabolites (bilirubin, biliverdin, or CORM3) into the medium of HK1 cells did not affect cGAMP or IFN-β production after RT (Figure 3, F and G). Consistently, the enzymatically inactive HO-1 mutant (HO-1H25A) (32) showed no obvious difference compared with WT HO-1 in affecting cGAMP or IFN-β production under RT or cGAMP treatment (Figure 3, H and I, and Supplemental Figure 3G).

Taken together, these data suggest that HO-1 inhibited the function of both cGAS and STING in tumor cells under RT independent of its enzymatic activity.

RT induces HO-1 expression and promotes its cleavage. Notably, we observed that RT had a direct effect on HO-1, resulting in increased HO-1 expression and cleavage in all 5 tumor cell lines (Figure 4A and Supplemental Figure 4A); and the induction of HO-1 expression by RT was dose dependent below 10 Gy (Supplemental Figure 4B). Among the 5 tumor cell lines, we selected HK1 and MDM-MB-231, along with their corresponding normal epithelial cell lines (NP69 and MCF10A), for further investigation. RT also induced the expression and cleavage of HO-1 in NP69 and MCF10A, suggesting the universality of this phenomenon in both tumor and normal cell lines. Nevertheless, the effect of RT on HO-1 in the normal epithelial cells were weaker than those observed in the paired tumor cell lines (Supplemental Figure 4C).

RT induces HO-1 and promotes its cleavage. (A) Immunoblot analysis of HO-1 expression and truncation in the indicated cells before and after RT. (B and C) Immunoblot analysis of HO-1 expression and truncation in HK1 cells after RT (B) or IFN-β treatment (C) combined with or without NAC treatment. (D and E) Immunoblot analysis of Flag–HO-1 expression in HK1 cells before and after RT. (D) Flag tag was fused to N-terminus or C-terminus of HO-1, respectively. (E) Mutating S272-F276 of HO-1 individually or mutating all 5 amino acids between S272 and F276. (F) Subcellular distribution (ER and nucleus) of full-length HO-1, uncleavable HO-1 mutant, cleaved HO-1 (HO-1ΔTMS) in HK1 cells with or without RT. Calreticulin staining for the ER; DAPI staining for the nucleus (scale bars: 10 μm). FL, full-length. (G and H) Nuclear and cytoplasmic protein extraction experiment was performed to determine the cellular localization of exogenous HO-1 or its mutants before (G) and after (H) RT in HK1 cells. (I) Subcellular distribution of endogenous HO-1 was determined with immunofluorescence staining in HK1 cells stimulated with RT (scale bars: 10 μm). (J) Nuclear and cytoplasmic protein extraction experiment was performed to determine the cellular localization of endogenous HO-1 at the indicated time point of RT in HK1 cells. (K and L) HMOX1-KO HK1 cells were stably transfected with the indicated HO-1 mutants. With or without RT, cGAMP (K) and IFN-β (L) production was determined with ELISA. (M and N) HMOX1-KO HK1 cells were stably transfected with indicated HO-1 mutants. With or without cGAMP, STING activation was determined with immunoblot analysis (M), and IFN-β production was determined by ELISA (N). Representative data from 1 experiment are shown (n = 3 biologically independent experiments). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, by 1-way ANOVA (K, L, and N). Data are shown as the mean ± SD.

Then, we explored the mechanism by which RT upregulated HO-1 expression. Considering that RT induces abundant ROS production in tumor cells, we interrogated whether the upregulation of HO-1 was attributed to ROS. Interestingly, adding N-acetyl-d-cysteine (NAC) almost abolished HO-1 upregulation in the early phase of RT (0–6 hours after RT); however, HO-1 was still markedly upregulated in the late phase of RT (24–48 hours after RT) (Figure 4B). This prompted us to identify other HO-1–inducing factors besides ROS. To this end, we used pharmacological inhibitors to block several signaling pathways activated by RT (33, 34), including PI3K/AKT, MAPK, NF-κB, Jak/STAT1, and ATR, and found that only the STAT1 inhibitor fludarabine notably abrogated HO-1 upregulation in the late phase of RT (Supplemental Figure 4D). The Jak/STAT1 pathway was activated mainly by IFN signaling. As expected, IFN-β treatment also led to notable HO-1 upregulation (Figure 4C). To further verify the role of IFN-Is, we cotreated tumor cells with NAC and IFN-β and found that NAC did not modulate the effect of IFN-β on HO-1 upregulation (Figure 4C). Therefore, we speculated that HO-1 might exist as an inherent, inflammation-limiting feedback mechanism under RT.

In addition, after RT exposure, 2 immunoreactive bands were unexpectedly observed when using anti–HO-1 antibody, with 1 migrating at 28 kDa and the other migrating at 32 kDa (Figure 4A and Supplemental Figure 4A), implying that RT not only upregulated HO-1 expression, but also induced its cleavage. To identify the sites at which HO-1 was cleaved, a Flag tag was fused to either the N-terminus or C-terminus of HO-1 (N-Flag-HO-1/HO-1-Flag-C). After transfection and RT, immunoblotting showed 2 closely positioned bands when Flag was fused to the N-terminus, while only 1 band appeared when Flag was fused to the C-terminus (Figure 4D), suggesting that HO-1 was cleaved near the C-terminus. Consistently, previous studies have shown that HO-1 has multiple cleavage sites at the C-terminus, primarily between S272-F276 (32, 35). However, mutating any single site was not sufficient to eliminate the cleavage led by RT, and only mutating all amino acids between S272-F276 resulted in an uncleavable state (Figure 4E). Structurally, the S272-F276 segment is located within the transmembrane domain (266–288 aa) of HO-1, which is responsible for its ER location. We observed that whether HO-1 was truncated from S272 or F276, the distribution for either one was similar to the transmembrane domain fully truncated mutant. Specifically, cleaved HO-1 no longer resided on the ER; instead, it was mostly distributed in the nucleus, with a small portion found in the cytoplasm (Supplemental Figure 4E). To clarify the relationship between RT and the changes in HO-1 distribution, we further analyzed the distribution of 3 forms of HO-1, including WT, uncleavable, and cleaved (HO-1ΔTMS), before and after RT. Before RT, WT and uncleavable HO-1 were only located on the ER, whereas the cleaved counterpart was distributed both in the nucleus and the cytosol but not on the ER (Figure 4F). After RT, WT HO-1 appeared to have nuclear distribution and simultaneously displayed 3 distributions: nucleus, cytosol, and ER, while the distribution of uncleavable and cleaved HO-1 (HO-1ΔTMS) did not change and was consistent with that observed before RT (Figure 4F). These observations were further confirmed by nuclear and cytoplasmic protein extraction experiments (Figure 4, G and H). Consistently, endogenous nucleus-located HO-1 was observed only after RT in the truncated form (Figure 4, I and J, and Supplemental Figure 4, F and G). Together, these data indicated that RT-mediated HO-1 cleavage destroyed its transmembrane domain and led to redistribution.

Rhomboid serine proteases, γ-secretases, and signal peptide peptidase (SPP) are 3 major intramembrane protease families (36). By pretreatment with specific inhibitors [DCI, DAPT and (Z-LL)2-ketone], we discovered that only the SPP inhibitor suppressed the cleavage of HO-1 under RT (Supplemental Figure 4H), which was accompanied by increased cGAMP and IFN-β production (Supplemental Figure 4I). In line with this, both cleaved and uncleavable HO-1 inhibited IFN-β production under RT (Figure 4K). Previously, we had already identified that WT HO-1 inhibited both cGAS and STING activity. With a similar method, we found that cleaved HO-1 (HO-1ΔTMS) suppressed cGAMP production (Figure 4L), while uncleavable HO-1 inhibited IFN-β production and STING activation after cGAMP treatment, without affecting cGAMP production under RT (Figure 4, M and N). This indicated that cleaved HO-1 acted on cGAS and uncleavable, ER-locked HO-1 acted on STING.

To conclude, RT induced HO-1 expression and promoted its cleavage. Different forms (cleaved or WT) of HO-1 suppressed the function of cGAS or STING.

Cleaved HO-1 directly interacts with cGAS and inhibits its nuclear export. Next, we sought to elucidate the mechanism by which cleaved HO-1 regulates cGAS activity. Redistribution of cellular positions was a prominent change in cleaved HO-1 compared with the WT counterpart. Meanwhile, cGAS is distributed to both the nucleus and cytoplasm (37), so we first explored whether distribution is important for HO-1 in regulating cGAS. To this end, the nuclear location signal (NLS) peptide and nuclear export signal (NES) peptide were respectively fused to cleaved HO-1 (HO-1ΔTMS) to change its distribution (Supplemental Figure 5A). We found that increased nuclear distribution of cleaved HO-1 (HO-1ΔTMS) led to stronger inhibitory effects on cGAMP and IFN-β production under RT, while cytoplasm-located cleaved HO-1 (NES-HO-1ΔTMS) nearly abolished these effects (Figure 5, A and B). Furthermore, immunoprecipitation revealed that cleaved HO-1 fused with NLS (NLS-HO-1ΔTMS) showed strong interaction with cGAS under RT (Figure 5C). Additionally, as more cleaved HO-1 entered the nucleus, the nucleus-located cGAS increased after RT (Figure 5, D and E, and Supplemental Figure 5B). Unexpectedly, RT dramatically enhanced cGAS nuclear export (Figure 5, F and G, and Supplemental Figure 5, C and D). Cytoplasmic localization of cGAS is critical for its role as a DNA sensor and in cGAMP synthesis. In our scenario, when cGAS was confined to the nucleus by leptomycin B (LMB), a nuclear export inhibitor, production of both cGAMP and IFN-β was markedly suppressed under RT (Figure 5, H–J).

Cleaved HO-1 inhibits the nuclear export of cGAS. (A–E) HMOX1-KO HK1 cells were stably transfected with cleaved HO-1 (HO-1ΔTMS), exclusively nucleus-located cleaved HO-1 (NLS-HO-1ΔTMS), or exclusively cytoplasm-located cleaved HO-1 (NES-HO-1ΔTMS) individually. (A and B) ELISA of cGAMP (A) or IFN-β (B) production before and after RT. (C) The interaction of Flag-tagged HO-1 mutants and HA-tagged cGAS in HEK293T cells was analyzed by immunoprecipitation under RT. WCL, whole-cell lysate. (D, F, and H) Subcellular distribution (cytoplasm and nucleus) of cGAS was determined by immunofluorescence staining of HK1 cells with the indicated mutants or RT stimulation (scale bars: 10 μm). The percentages of cells (n = 200) in the nucleus, cytoplasm, or both the cytoplasm and nucleus were calculated. (E and G) The cytoplasmic and nuclear protein fractions were extracted for immunoblot analysis to determine the subcellular localization of cGAS in HK1 cells with the indicated mutants or RT stimulation. (I and J) ELISA of cGAMP (I) or IFN-β (J) production before and after RT (related to Figure 5H). Data are shown as the mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, by 1-way ANOVA (A and B) and unpaired, 2-tailed Student’s t test (I and J). All representative data from 1 experiment are shown (n = 3 biologically independent experiments). N, predominantly in the nucleus; C, predominately in the cytoplasm; C+N, evenly distributed in the nucleus and cytoplasm.

On the basis of the above findings, we speculated that cleaved HO-1 exerts its function by inhibiting cGAS nuclear export. As expected, in HK1 or HeLa cells, we found that before RT, HMOX1 KO did not influence cGAS nuclear translocation; after RT, HO-1 deficiency increased cGAS nuclear export (Figure 6, A and B, and Supplemental Figure 5, E and F). Consistently, cleaved HO-1 (HO-1ΔTMS) inhibited cGAS nuclear export, but uncleavable HO-1 had a faint effect compared with HO-1–deficient cells after RT (Figure 6, C and D). We previously showed that exogenous HO-1 interacted with cGAS (Figure 5C), and similarly, we found that in endogenous conditions, cleaved HO-1 entered the nucleus and directly interacted with cGAS, and the interaction was gradually enhanced within 12 hours after RT (Figure 6, E and F). To further elucidate the detailed interaction between cleaved HO-1 and cGAS, we generated truncated cGAS mutants by separating cGAS into the N-terminus (1–160 aa) and the C-terminus (161–522 aa) (38) (Figure 6G). Immunoprecipitation revealed that the C-terminus was required for the interaction between cGAS and cleaved HO-1 under RT (Figure 6H). The NES sequence of cGAS resided in the C-terminus (39). CRM1 is the main nuclear export receptor that binds to the NES on target proteins, including cGAS, to mediate their nuclear export. Notably, the interaction between cGAS and CRM1 was enhanced under RT, while reexpression of cleaved HO-1 (HO-1ΔTMS) impaired the interaction (Figure 6I). Collectively, these data suggest that RT induces HO-1 cleavage, which then translocates into the nucleus and disturbs the cGAS-CRM1 interaction to impair RT-induced nuclear export of cGAS.

Cleaved HO-1 directly interacts with cGAS in the nucleus. (A and C) Cytoplasmic and nuclear protein fractions were extracted for immunoblot analysis to determine the subcellular localization of cGAS in HK1 cells with the indicated cell lines and stimulation. (B and D) Subcellular distribution (cytoplasm and nucleus) of cGAS was determined with immunofluorescence staining of HK1 cells with the indicated cell lines and stimulation (scale bars: 10 μm). The percentages of cells (n = 200) in the nucleus, cytoplasm, or both the cytoplasm and nucleus were calculated (E) Confocal microscopy images of cGAS and HO-1 in HK1 cells before and after RT (scale bars: 10 μm). (F) The cytoplasmic and nuclear protein fractions of HK1 cells at the indicated RT time points were extracted for coimmunoprecipitation. (G and H) The interaction of HA-tagged full-length cGAS (aa 1–522), N-terminus of cGAS (aa 1–160), C-terminus of cGAS (aa 161–522), and Flag-tagged HO-1ΔTMS in HEK293T cells was analyzed by immunoprecipitation. (I) HMOX1-KO HK1 cells were stably transfected with cleaved HO-1 (HO-1ΔTMS). The interaction of endogenous cGAS and CRM1 in HK1 cells was analyzed by immunoprecipitation. All representative data from 1 experiment are shown (n = 3 biologically independent experiments).

HO-1 inhibits STING oligomerization and consecutive ER-to-Golgi translocation by direct interaction. We next investigated the mechanism by which HO-1 regulates STING activation. As demonstrated previously, ER-locked, uncleavable HO-1 selectively suppressed STING-mediated IFN-β production (Figure 4, M and N). Immunofluorescence indicated that, in the resting state, STING was completely colocalized with HO-1; after cGAMP treatment, a part of STING aggregated into puncta and lost colocalization with HO-1, while the remaining part still colocalized with HO-1 (Figure 7A). Since resting STING is located in the ER as a dimer, we next explored whether HO-1 exerts its effects by directly interacting with STING. Consistently, immunoprecipitation revealed that HO-1 interacted with STING with or without cGAMP, and the amount of STING bound to HO-1 was reduced after cGAMP addition (Figure 7B). cGAMP-activated STING is transported from the ER to the Golgi apparatus. In this regard, we found that knocking out HMOX1 further decreased ER-located STING and increased its distribution in the Golgi apparatus (Figure 7, C and D), which could be reversed by reexpression of HMOX1 (Figure 7, C and D). Therefore, we speculated that HO-1 played a role in restricting STING translocation. To confirm this, we analyzed the interaction between TBK1 and STING, because TBK1 binds mainly to ER-detached, Golgi-located, and oligomeric STING and acts as the key factor to turn on downstream signaling. As anticipated, the interaction between TBK1 and STING was enhanced after knocking out HMOX1 (Figure 7E).

HO-1 inhibits STING oligomerization and consecutive ER-to-Golgi translocation by direct interaction. (A) Confocal microscopy images of STING and HO-1 in HK1 cells with the indicated treatment. Pearson’s r value was used as a statistical measure to determine the extent of colocalization between HO-1 and STING. (B) The interaction of endogenous HO-1 and STING in HK1 cells was analyzed by immunoprecipitation with the indicated treatment. (C and D) Control, HMOX1-KO, and HMOX1-overexpressing HK1 cells were stained with anti-STING (C and D), anti-calreticulin (C), and anti-GM130 (D) antibodies. Pearson’s r value was used as a statistical measure to determine the extent of colocalization between STING and calreticulin or GM130. (E) The interaction of endogenous STING and TBK1 in HK1 cells was analyzed by immunoprecipitation with the indicated treatment. (F and G) STING polymerization in control and HMOX1-KO HK1 cells with the indicated treatments, followed by native PAGE and SDS-PAGE. (H) HEK293T cells were transfected with the indicated STING mutant plus vector or STING mutant plus HO-1, followed by confocal imaging. (I) HEK293T cells were cotransfected with plasmids expressing HO-1 and STING, or its mutants, followed by native PAGE and SDS-PAGE. (J) HK1 cells were stably transfected with doxycycline-induced (Dox) STING expression plasmids. After doxycycline treatment at the indicated dose, native PAGE for detection of STING polymers and SDS-PAGE were performed. (A) Imaging data were analyzed with Fuji software to reveal colocalization as white dots. (A, C, and D) Pearson’s correlation coefficient was quantified using ImageJ (NIH). n = 10 cells (quantified in a blinded manner). Data are shown as the mean ± SD. Scale bars: 10 μm. **P < 0.01 and ****P < 0.0001 by unpaired, 2-tailed Student’s t test (A) and 1-way ANOVA (C and D).

The formation of STING oligomers is the key step in recruiting the coat protein complex II (COPII), which facilitates the translocation of STING and its spatial redistribution. Therefore, we explored whether HO-1 influences STING oligomerization and found that, essentially, the way cGAMP triggered the STING signaling cascade was by inducing STING oligomerization. In this regard, we found that cGAMP treatment induced stronger STING oligomerization in HMOX1-KO cells (Figure 7F). Considering that translocated oligomeric STING would undergo degradation, pretreatment with brefeldin A (BFA) to inhibit ER-Golgi translocation was adopted to avoid potential confounding. When STING was restrained on the ER, KO of HMOX1 also enhanced its oligomerization (Figure 7G); notably, HO-1 also inhibited baseline STING oligomerization even without cGAMP (Figure 7F). To further verify that HO-1 inhibited STING on the ER, and to rule out the confounding effect of cGAMP uptake, we used 2 other cellular models, in which STING oligomerization and activation occurred automatically. In 1 model, we mutated R281 or R284 amino acids that resided in the polymerization interface of STING (20). After confirming autoactivation (Supplemental Figure 6, A and B), we found that HO-1 overexpression increased ER distribution and weakened the activation and polymerization of these 2 mutants (Figure 7, H and I, and Supplemental Figure 6C). In another model, we developed a doxycycline-induced STING expression system in cGAS-deficient cells, as elevated STING protein on the ER would be advantageous for STING polymerization. Autoactivation was induced when the doxycycline concentration reached 0.6–0.8 μg/mL (Supplemental Figure 6D). We discovered that HO-1 depletion also enhanced STING autoactivation and polymerization (Figure 7J).

Successful COPII vesicle formation is critical for STING to undergo ER-Golgi translocation (21). Here, we found that the interaction between STING and key members of the COPII complex (SAR1A, SEC24C) (40) was enhanced by cGAMP treatment, but impaired by HO-1 (Supplemental Figure 6, E and F). The curvature of the ER membrane is critical for recruiting SAR1 and highly affected by protein aggregation on the membrane. Using a GFP133 membrane curvature probe (21, 41) alongside ER-specific staining, we found that HO-1 deletion promoted ER membrane curvature under cGAMP treatment (Supplemental Figure 6G). Therefore, the presence of HO-1 decreased STING aggregation as well as impaired membrane curvature of the ER, thereby resulting in an unpolymerized, ER-locked state of STING.

Then, we sought to explore a detailed mechanism by which HO-1 inhibits STING oligomerization. First, by generating truncated STING mutants (N-terminus, 1–139 aa; C-terminus, 140–379 aa; the N-terminus contains the transmembrane segment and the C-terminus facilitates its aggregation; ref. 12), we found that the interaction between HO-1 and STING was not domain selective (Figure 8, A and B). Second, we performed molecular docking and molecular dynamics (MD) simulations. Considering that the basic unit of the STING oligomer was the STING dimer (42), we constructed a STING dimer using Alphafold2 (DeepMind), and the complexes of the STING dimer and HO-1 dimer were constructed with the HDOCK web server (http://hdock.phys.hust.edu.cn/). Through modeling and calculations, we found that HO-1 occupied the interface between 2 STING dimers (residues 273–280 in human STING were verified to be located at the tetramer interface of STING), which was essential for supporting STING tetramerization or hyperpolymerization (Figure 8C). Specifically, residues Thr222, Arg100, Arg100, Tyr97, and Gln212 in HO-1 formed a hydrogen bond with residues His16, Gln266, Gln273, Tyr274, and Glu340 in STING, respectively, and Arg113 in HO-1 formed a salt bridge with residue Glu340 in STING (Supplemental Table 1). To verify the effects of residues in the above docking model, Tyr97, Arg100, Arg113, Gln212, and Thr222 in HO-1 were mutated into alanine to disrupt the formation of a hydrogen bond or a salt bridge. Mutation of each residue variably weakened the binding between HO-1 and STING; and simultaneous mutation of all 5 residues almost abolished the binding (Figure 8D). Corresponding to the strength of each mutant’s binding to STING, STING oligomerization was stronger when the interaction was weaker (Figure 8E); and the variation of binding energy was consistent with the immunoprecipitation (Supplemental Table 2). In line with this finding, the binding energy between homogenous STING dimers was higher than that between HO-1 dimer and the homogenous STING dimer (HO-1 dimer+STING dimer vs. STING dimer+STING dimer: –280.23 kcal/mol vs. –243.68 kcal/mol), implying that the HO-1 dimer+STING dimer had more stable binding (Supplemental Table 3). STING tetramerization formed by the aggregation of 2 STING dimers is the first step in subsequent hyperpolymerization. This prompted us to analyze the effect of HO-1 on STING tetramerization. The binding modes between the STING tetramer and the HO-1 dimer after MD simulations are shown in Supplemental Figure 7A. In all 3 binding systems, the STING tetramer became unstable after binding to the HO-1 dimer according to the analysis of root mean square deviation (RMSD) and root mean square fluctuation (RMSF) (Supplemental Figure 7, B and C). The elevated radius of gyration (Rg) values further illustrated that the dimer-dimer interaction became less tight than the pure STING tetramer after HO-1 binding (Supplemental Figure 7D). Taken together, these data suggest that HO-1 inhibits STING polymerization on the ER and subsequent COPII-mediated translocation from the ER to the Golgi, ultimately impairing STING activation.

Molecular docking of HO-1 and STING. (A and B) The interaction of MYC-tagged full-length STING (aa 1–379), N-terminus of STING (aa 1–139), C-terminus of STING (aa 140–379) and Flag-tagged HO-1 in HEK293T cells was analyzed by immunoprecipitation. (C) View of binding modes between the STING dimer and the HO-1 dimer based on MD simulations. (D) The interaction of MYC-tagged full-length STING and Flag-tagged WT HO-1 or its mutants in HEK293T cells was analyzed by immunoprecipitation. (E) HEK293T cells were cotransfected with plasmids expressing STING and HO-1, or its mutants and stimulated or not with cGAMP, followed by native PAGE and SDS-PAGE. (B, D, and E) All representative data from 1 experiment are shown (n = 3 biologically independent experiments).

HO-1 inhibition enhances the efficacy and abscopal effect of RT in vivo. We subsequently sought to clarify whether inhibition of HO-1 improved the efficacy of RT in vivo. For a long time, HO-1 inhibitor screening was based on the extent of enzyme inhibition. However, here, we revealed that HO-1 exerted its effects via a nonenzymatic mode, and thus we first screened existing HO-1 inhibitors. Imidazole HO-1 inhibitors [including Zn-(II)-protoporphyrin IX and Tin-protoporphyrin IX] increased HO-1 expression and decreased IFN-β production under RT (Supplemental Figure 8, A and B), whereas a recently discovered inhibitor HO-1–IN-1 had an inhibitory effect on both in vitro and in vivo HO-1 expression and promoted IFN-β production (Supplemental Figure 8, A–C). More important, HO-1–IN-1 disrupted the endogenous interaction between cGAS and cleaved HO-1 as well as STING and full-length HO-1, leading to enhanced nuclear export, cGAMP production, and STING activation (Figure 9, A–C, and Supplemental Figure 8, D and E).

HO-1 inhibitor enhances the efficacy and abscopal effect of RT in vivo. (A) The cytoplasmic and nuclear protein fractions were extracted for immunoblot analysis to determine the subcellular localization of cGAS in MC38 cells treated as indicated. (B) ELISA of cGAMP production in MC38 cells treated as indicated. (C) Immunoblot analysis of STING and TBK1 phosphorylation in MC38 cells treated as indicated. (D–J) Effect of the HO-1 inhibitor combined with RT on tumor growth (D and E), mRNA levels of typical IFN-Is and ISGs (F and G), CD8+ T cell infiltration (H and I), and IFN-γ and TNF-α expression in CD8+ T cells (J) from B16 (D, F, and H) and MC38 (E, G, I, and J) tumors (n = 6 in each group). (K and L) Effect of the HO-1 inhibitor combined with RT on tumor growth (K), mRNA levels of typical IFN-Is (L) of HK1 tumors implanted into HuHSC-NCG mice (n = 5 in each group). (M and N) Schematic illustration and tumor growth of nonirradiated abscopal tumors and irradiated primary tumors with the indicated treatment (n = 6 in each group). Data are shown as the mean ± SD (B, F–I, J, and L) and the mean ± SEM (D, E, K, and N). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, by unpaired, 2-tailed Student’s t test (B, J, and L), 2-way ANOVA (D, E, K, and N), and 1-way ANOVA (F–I).

Next, we investigated the effect of HO-1–IN-1 in vivo, with or without regional RT, and found that HO-1–IN-1 combined with RT suppressed tumor growth compared with RT alone (Figure 9, D and E, and Supplemental Figure 8F). Further assays revealed that intratumoral IFN-I production, ISG levels (H2kb and Cxcl10), CD8+ T cell infiltration, and function (TNF-α and IFN-γ secretion) were elevated in the presence of HO-1–IN-1 (Figure 9, F–J, and Supplemental Figure 8, G–I). Consistently, in humanized mice, HO-1–IN-1 also further decreased the tumor volumes formed by HK1 cells and promoted intratumoral IFN-I production (Figure 9, K and L). Except for local control of irradiated sites, we also examined the effects of HO-1 inhibition on abscopal tumors (Figure 9M). RT combined with HO-1–IN-1 not only delayed substantial progression of the primary tumor, but also improved tumor control at the abscopal sites (Figure 9N).

To fully clarify the involvement of cGAS and STING, we generated cGas- or Sting-KO MC38 cells (Supplemental Figure 8J). We found that knocking out either cGas or Sting weakened the sensitizing effect of HO-1–IN-1 on RT, but still showed some effect (Supplemental Figure 8, K and L). Moreover, in a parallel experiment, anti-CD8–depleting antibody almost abolished the RT-sensitizing effect induced by HO-1 inhibition (Supplemental Figure 8M).

Taken together, these data demonstrate that inhibition of HO-1 has promising RT-sensitizing effects in multiple preclinical in vivo models.

High expression of HO-1 correlates with unfavorable RT prognosis. To determine the clinical significance of HO-1 expression in RT, we performed IHC staining of NPC tissues from 220 patients who had undergone RT. On the basis of the staining intensity, patients’ samples were categorized into a high HO-1 expression group (n = 116, 52.7%) and a low HO-1 expression group (n = 104, 47.3%) for further analysis (Figure 10A and Supplemental Table 4). By integrating the IHC results with our clinical data from Sun Yat-sen University Cancer Center (SYSUCC), we found that locoregional recurrence was positively correlated with high HO-1 expression in NPC (Figure 10B). Kaplan-Meier analysis revealed that high HO-1 expression was correlated with worse overall survival (OS) and disease-free survival (DFS) (Figure 10, C and D, and Supplemental Table 4). Consistently, HO-1 was associated with poor DFS in patients with NPC based on a published RNA-Seq dataset (43) (Figure 10E). Moreover, we also conducted survival analysis of patients from TCGA and Cbioportal. After stratifying patients according to those who received RT and those who did not, we found that for patients with esophageal carcinoma (ESCA) or glioblastoma (GBM), HO-1 expression did not correlate with progression-free survival (PFS) in patients who did not undergo RT, whereas its expression showed a strong correlation with worse PFS in patients who underwent RT (Supplemental Figure 9, A–D). Similarly, higher HO-1 expression was associated with worse OS in pediatric patients with brain cancer who received RT, but not in the non-RT cohort (Supplemental Figure 9, E and F). For patients with diffuse glioma, HO-1 expression was more deterministic of poor OS for patients who received RT compared with those who did not, as evidenced by higher HRs (Supplemental Figure 9, G and H).

High expression of HO-1 correlates with unfavorable RT prognosis. (A) Representative images of immunohistochemical staining for HO-1 protein expression, which is graded according to the staining intensity in 220 NPC tissues. Scale bars: 100 μm. (B) Correlations of the locoregional recurrence status with HO-1 expression detected by IHC. P value was determined by 2-tailed χ2 test. (C and D) Kaplan-Meier analysis of OS (C) and DFS (D) according to HO-1 expression. (E) Kaplan-Meier analysis of DFS based on HO-1 expression in the published bulk RNA-Seq dataset. (C–E) P values were determined using the log-rank test. (F) Proposed working model of HO-1. By an unbiased CRISPR screen, we identified HO-1 as an irradiation-related regulator of IFN-I production. Mechanistically, irradiation induced HO-1 expression and promoted its cleavage. Cleaved HO-1 underwent nuclear translocation, interacted with cGAS, inhibited its nuclear export upon radiation, and suppressed its enzymatic activity. ER-anchored full-length HO-1 disturbed STING polymerization and subsequent COPII-mediated ER-Golgi transportation, leading to impaired activation of downstream signaling.

In summary, these results suggest that high HO-1 expression is associated with an unfavorable prognosis and response to RT.