Identification of ATXN3 as a PD-L1–positive regulator through unbiased CRISPR screening. To identify the specific deubiquitinases that promote tumoral PD-L1 expression, we first designed a targeted library of all 96 mammalian deubiquitinase family members based on the optimized single-guide RNA (sgRNA) sequences and cloned each of them into a lentivirus-based lentiCRISPR v2 vector system, which coexpresses CRISPR-associated endonuclease 9 (Cas9) (Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/JCI167728DS1). The pooled CRISPR plasmids were further validated by sequencing for their equal representation and then transfected into Lenti-X packaging cells. The pooled viruses were titrated as reported (24); B16 melanoma cells, which express a high level of PD-L1, were infected with 0.3 multiplicity of infection of the pooled DUB-KO lentivirus library; and, 48 hours after transduction, cells were selected with 2 μg/mL of puromycin for 3–4 days. The infectivity of selected cancer cells was further validated by intracellular staining of Cas9 (Supplemental Figure 1, B and C). The puromycin-selected cells were then sorted for PD-L1–low and –high populations to identify positive and negative PD-L1 regulators, respectively (Supplemental Figure 1D). Genomic DNA was purified from the sorted cells, and the specific region carrying the guide sequences was amplified by a 1-step PCR for sequencing (Figure 1A). Guides enriched in either PD-L1–low or –high populations were analyzed using MAGeCK count and test functions. Genes were ranked using the MAGeCK radioreceptor assay enrichment score. Fold change in read counts of each sgRNA within the top genes was analyzed. Guides were selected for further validation based on P values (Supplemental Figure 1, E–H). A total of 4 deubiquitinases, ATXN3, USP30, USP32, and OTULIN, were enriched in PD-L1–low populations, which are potential PD-L1–positive regulators (Supplemental Figure 1, E and F, and Figure 1B). USP30 was recently shown to promote tumoral PD-L1 expression (25), providing confidence for our screening. On the other hand, while 4 deubiquitinases, USP6NL, USP18, USP27X, and OTUD3, were exclusively enriched in the PD-L1–high pool of B16 cells, only USP6NL reached statistical significance, implicating USP6NL as a negative PD-L1 regulator (Supplemental Figure 1, G and H, and Figure 1B). Importantly, further analysis of CRISPR KO B16 melanoma cells confirmed that targeted suppression of ATXN3 and USP30 dramatically reduced PD-L1 expression (Figure 1C). It has been reported that both USP7 and USP22 protect PD-L1 from ubiquitination-mediated degradation in cancers including gastric cancer, lung adenocarcinoma, and liver cancer (20-23), neither of which was enriched in our screening (Figure 1B). This is likely explained by regulation of PD-L1 by USP7 and USP22 in a tumor type–specific manner, since we have shown previously that CRISPR targeted USP22 deletion resulted in PD-L1 downregulation in human breast cancer cells but not in B16 cells (26). Nevertheless, our CRISPR screening identified ATXN3 as a previously unknown positive regulator in cancer cells.

Identification of ATXN3 as a PD-L1–positive regulator in cancer cells by unbiased CRISPR screening. (A) Schematic of the deubiquitinase CRISPR knockout screening workflow. (B) Guides enriched in PD-L1–low and PD-L1–high populations with their fold enrichment. The guide code of each gene for further validation is indicated. (C) Guide hits described were validated by flow using individual guide knockouts. (D) Western blotting validation of ATXN3 knockout and PD-L1 expression with specific sgRNAs in LLC1 cells. WT, cells transfected with empty vector; KO, ATXN3-knockout stable cell strains. (E and F) Representative flow cytometry plots and quantification by MFI of cell-surface PD-L1 in LLC1 cells. (G) Western blotting analysis of ATXN3 and PD-L1 expression in A549 cells with knockout of ATXN3. (H and I) Representative flow cytometry plots and quantification of cell-surface PD-L1 in A549 cells with knockout of ATXN3. (J and K) Cd274 and Atxn3 mRNA levels were analyzed by reverse transcription quantitative PCR (RT-qPCR) in LLC1 cells. (L and M) Cd274 and Atxn3 mRNA levels were analyzed by RT-qPCR in B16 cells. (N) Correlation of CD274 mRNA levels with ATXN3 mRNA levels in lung cancer patients (n = 22). (O) Correlation of CD274 with ATXN3 expression in multiple tumors based on TCGA data (n = 40). C, F, and I–M: 2-tailed unpaired t test; N: Pearson’s correlation analysis. *P < 0.05, **P < 0.01, ***P < 0.001.

ATXN3 promotes PD-L1 expression at transcriptional level in a broad spectrum of cancer cells. We then validated the potential role of ATXN3 in regulating tumoral PD-L1 expression in mouse and human lung cancer cells. CRISPR deletion of ATXN3 expression in Lewis lung carcinoma LLC1 cells resulted in a significant reduction in PD-L1 protein expression as detected by Western blotting and flow cytometry (Figure 1, D–F). Furthermore, CRISPR-mediated ATXN3 deletion dramatically reduced PD-L1 expression in mouse B16 melanoma, colon cancer MC38, and triple-negative breast cancer 4T1 cells (Supplemental Figure 2) and human lung small cell adenocarcinoma A549 cells (Figure 1, G–I), which was further confirmed by an alternative approach using shRNA-mediated knockdown of the ATXN3 gene (Supplemental Figure 3, A and B). In contrast, only a modest but statistically significant reduction in PD-L1 expression was detected in human colon cancer HCT116 cells by ATXN3-targeted knockdown (Supplemental Figure 3, C and D). This is likely due to HCT116 expressing low levels of ATXN3 endogenously (Supplemental Figure 3E). Interestingly, ATXN3 appears to positively regulate PD-L1 expression at the transcriptional level, since real-time reverse transcription PCR analysis confirmed that Atxn3 targeting dramatically inhibited Cd274 mRNA expression in both LLC1 and B16 mouse cancer cells (Figure 1, J–M). To support this conclusion, we further demonstrated that ATXN3 expression dramatically enhanced luciferase reporter activity under the control of an optimal 2 kb human CD274 promoter region (Supplemental Figure 4). These results indicate that ATXN3 is a positive regulator of PD-L1 gene expression in a variety of both human and mouse cancer cells at the transcriptional level.

To further validate our findings in samples from patients with cancer, we first assessed mRNA expression levels of ATXN3 and CD274 in 22 lung cancer patients and found a positive correlation between ATXN3 and CD274 expression (Figure 1N). Consistently, analysis of the The Cancer Genome Atlas (TCGA) database revealed a positive correlation between ATXN3 and CD274 expression in human lung adenocarcinoma (Figure 1O). Importantly, in addition to lung adenocarcinoma, the positive correlation between ATXN3 and CD274 expression was identified in 33 of 40 total types of human cancers, including bladder urothelial carcinoma, breast invasive carcinoma, cholangial carcinoma, colon adenocarcinoma, lymphoid neoplasm diffuse large B cell lymphoma, head and neck squamous cell carcinoma–HPV+, kidney chromophobe, kidney renal clear cell carcinoma, kidney renal papillary cell carcinoma, brain lower-grade glioma, liver hepatocellular carcinoma, mesothelioma, ovarian serous cystadenocarcinoma, pancreatic adenocarcinoma, pheochromocytoma and paraganglioma, prostate adenocarcinoma, rectal adenocarcinoma, sarcoma, skin cutaneous melanoma, stomach adenocarcinoma, testicular germ cell tumors, thyroid carcinoma, thymoma, uterine corpus endometrial carcinoma, and uveal melanoma (Figure 1O), indicating that ATXN3 is a potential positive regulator of PD-L1 transcription in a broad spectrum of human cancers.

ATXN3 is a positive regulator of IFN-γ–induced PD-L1 transcription in tumor cells. Tumoral PD-L1 expression is regulated by intrinsic oncogenic and adaptive signaling pathways. We first analyzed the possible involvement of ATXN3 in regulating PD-L1 expression induced by tumor microenvironmental cytokines, such as IFN-γ (27–29). As expected, treatment of human lung cancer A549 cells with IFN-γ resulted in a significant increase in PD-L1 expression. Targeted deletion of ATXN3 by CRISPR largely abrogated IFN-γ–induced PD-L1 expression (Figure 2, A and B). Downstream transcription factors such as IRF1 are involved in IFN-γ–induced PD-L1 expression (10, 28). Indeed, coimmunoprecipitation (co-IP) and Western blotting detected that ATXN3 interacts with IRF1 in transiently transfected HEK293T cells (Figure 2C). The endogenous interaction between IRF1 and ATXN3 was further confirmed in human lung cancer A549 cells (Figure 2D), indicating that ATXN3 may enhance IFN-γ–induced PD-L1 expression through IRF1.

ATXN3 potentiates IFN-γ–induced PD-L1 expression through stabilizing IRF1 and STAT3. (A and B) WT and ATXN3-KO cells were treated with IFN-γ (10 ng/mL) for 24 hours, and surface PD-L1 levels were analyzed. (C) ATXN3 interacts with IRF1 in transiently transfected HEK293T cells. (D) Interaction of endogenous ATXN3 and IRF1 in A549 cells. (E) HA-ubiquitin and FLAG-IRF1 expression plasmids were cotransfected with Myc-ATXN3 into HEK293T cells. IRF1 ubiquitination was determined by immunoprecipitation of IRF1 and immunoblotting with HA antibody. (F and G) FLAG-IRF1 was cotransfected with or without Myc-ATXN3 plasmids into HEK293T cells. The transfected cells were treated with cycloheximide (CHX) for different times. The protein levels of FLAG-IRF1 (top panel) and Myc-ATXN3 (middle panel) with β-actin control (bottom panel) were analyzed by Western blotting. Representative images (F) and quantification data from 3 independent experiments are shown (G). (H and I) Immunoblot analysis of IRF protein stability in WT and ATXN3-KO A549 cells as in F and G. (J) Interaction between ATXN3 and STAT3 in transfected HEK293T cells. (K) Endogenous interaction between ATXN3 and STAT3 in A549 cells. (L) The effect of ATXN3 on STAT3 ubiquitination was determined as in E. (M and N) The effects of ATXN3 on STAT3 protein stability were analyzed as in F and G. (O and P) Immunoblot analysis of STAT3 protein stability in WT and ATXN3-KO A549 cells as in H and I. (Q) The interaction between ATXN3 and STAT1 was tested in A549 cells. (R) ATXN3 enhances tumoral PD-L1 expression through protecting IRF1 and STAT3 from ubiquitination-induced protein degradation. B: Ordinary 1-way ANOVA; G, I, N, and P: 2-tailed unpaired t test; *P < 0.05, **P < 0.01, ***P < 0.001. WCL, whole-cell lysate.

To determine the functional consequence of the interaction between ATXN3 and IRF1, we analyzed the effect of gain of the deubiquitinase ATXN3 functions on IRF1 ubiquitination. The gradual ubiquitination of IRF1 was detected in anti–FLAG-IRF1 immunoprecipitants when IRF1 and HA-ubiquitin were both expressed. Further expression of ATXN3 dramatically inhibited IRF1 ubiquitination (Figure 2E), indicating that ATXN3 is a deubiquitinase of IRF1. Consequently, gain of ATXN3 expression increased expression of IRF1 and its protein stability (Figure 2, F and G). Conversely, targeted ATXN3 deletion facilitated IRF1 protein degradation (Figure 2, H and I). Collectively, these observations indicate that ATXN3 promotes IFN-γ–induced PD-L1 transcription at least partially through the stabilization of its transcriptional factor IRF1 in cancer cells.

In addition to IRF1, the transcription factor STAT3 has been identified as a downstream transcription factor for PD-L1 expression (5, 30), raising a possibility that ATXN3 may also promote PD-L1 expression through STAT3. In fact, STAT3 interaction with ATXN3 was detected in transiently transfected HEK293T cells as well as endogenously in human lung cancer A549 cells (Figure 2, J and K). Expression of ATXN3 largely diminished STAT3 ubiquitination (Figure 2L). Therefore, gain of ATXN3 function increased STAT3 protein expression and prolonged its half-life (Figure 2, M and N). Conversely, loss of ATXN3 resulted in reduced STAT3 expression due to the elevated protein degradation (Figure 2, O and P). In contrast, ATXN3 interaction with another member of the STAT family, STAT1, which promotes PD-L1 transcription (10, 31), was undetected in A549 cells (Figure 2Q). Notably, ectotrophic expression of both IRF1 and STAT3 largely restored IFN-γ–induced PD-L1 expression in ATXN3-null lung cancer cells. In contrast, coexpression of both IRF1 and STAT3 failed to rescue PD-L1 expression when cells were cultivated under hypoxic conditions (Supplemental Figure 5A). Therefore, our data indicate that ATXN3 enhances IFN-γ–induced tumoral PD-L1 expression by protecting IRF1 and STAT3 from ubiquitination-induced protein degradation (Figure 2R).

ATXN3 is a positive regulator for tumoral PD-L1 transcription under hypoxia conditions. In addition to IFN-γ, tumor microenvironmental factors, such as hypoxia, have been known to strongly upregulate PD-L1 expression (6–8). Consistent with previous reports, PD-L1 expression on the surface of A549 cells was significantly increased under hypoxic conditions. Importantly, this hypoxia-induced PD-L1 upregulation was largely, while not totally, diminished by ATXN3 targeted deletion (Figure 3, A and B), implying a possibility that ATXN3 enhances hypoxia-induced PD-L1 expression in tumor cells. Since both HIF-1α and HIF-2α have been identified as PD-L1 transcription factors in cancer cells in response to hypoxia (6–8), we then reasoned whether ATXN3 enhances tumoral PD-L1 transcription through stabilizing HIF family transcription factors. Indeed, co-IP and Western blotting detected ATXN3 interaction with HIF-2α, but not HIF-1α, in transiently transfected HEK293T cells (Figure 3, C and D), implying a possibility that ATXN3 enhances hypoxia-induced PD-L1 expression specifically through HIF-2α. To support this notion, we confirmed endogenous interaction of ATXN3 with HIF-2α in human lung cancer A549 cells (Figure 3E).

ATXN3 selectively functions as a HIF-2α deubiquitinase to promote tumoral PD-L1 transcription. (A and B) A549 cells were cultured under normoxia and hypoxia (hyp) (1% pO2) for 48 hours, and surface PD-L1 levels were analyzed by flow cytometry and quantification. (C) ATXN3 specifically interacts with HIF-2α. HA–HIF-2α expression plasmid was cotransfected with or without FLAG-ATXN3 into HEK293T cells. Their interactions were examined by co-IP with anti-FLAG antibodies and by Western blotting with anti-HA antibodies. (D) The interaction between ATXN3 and HIF-1α was tested in transfected HEK293T cells. (E) Endogenous interaction between ATXN3 and HIF-2α in A549 cells. (F) HA-ubiquitin, FLAG–HIF-2α, and Myc-ATXN3 plasmids were cotransfected into HEK293T cells. HIF-2α ubiquitination was determined by immunoprecipitation of HIF-2α with anti-FLAG antibodies and immunoblotting with anti-HA antibody. (G and H) HIF-2α was cotransfected with or without ATXN3 plasmids into HEK293T cells. The transfected cells were treated with CHX for different times. The protein levels of HIF-2α (top panel) and ATXN3 (middle panel) were analyzed by Western blotting. β-Actin was used as a loading control (bottom panel). (I and J) Immunoblot analysis of HIF-2α protein stability in WT and ATXN3-KO A549 cells. (K) ATXN3 enhances hypoxia-induced PD-L1 expression through protecting HIF-2α from ubiquitination-induced protein degradation. B: Ordinary 1-way ANOVA; H and J: 2-tailed unpaired t test. *P < 0.05, **P < 0.01, ***P < 0.001.

A deubiquitinase often inhibits the ubiquitination of and stabilizes its interacting partners to achieve its pathobiological functions (32). As expected, ATXN3 inhibited HIF-2α ubiquitination, which consequently resulted in elevated HIF-2α expression (Figure 3F). Results from further pulse-chase experiments confirmed that ATXN3 expression increased HIF-2α expression levels and prolonged its half-life (Figure 3, G and H). Conversely, ATXN3 targeted suppression resulted in reduced HIF-2α expression and facilitated its degradation (Figure 3, I and J). However, unlike IRF1 and STAT3 coexpression that fully rescued IFN-γ–induced PD-L1 expression (Supplemental Figure 5A), HIF-2α expression could not rescue ATXN3-KO cancer cell expression of PD-L1 when cultivated under hypoxia conditions, nor when IFN-γ was added, suggesting that additional genes are involved in the hypoxia/ATXN3/HIF-2α pathway to control hypoxia-induced PD-L1 expression on cancer cells. Collectively, our results indicate that ATXN3 enhances hypoxia-induced PD-L1 expression through protecting HIF-2α from ubiquitination-induced protein degradation (Figure 3K).

ATXN3 interacts with multiple transcription factors to promote tumoral PD-L1 expression. Apart from hypoxia and IFN-γ, several intrinsic oncogenic and tumor microenvironmental factor adaptive pathways are involved in regulating tumoral PD-L1. We then tested the possibility that ATXN3 promotes tumoral PD-L1 expression through downstream transcription factors, such as NF-κB (p65), c-MYC, and AP-1 (Supplemental Table 1). However, ATXN3 interaction with NF-κB (p65), a transcription factor responsible for inflammatory cytokines, such as TNF-α–induced PD-L1 expression, was not detected even when they were overexpressed (Supplemental Table 1). Similarly, the interaction of ATXN3 with c-MYC, a transcription factor responsible for metabolite-induced PD-L1 expression (4), was not detected in transiently transfected HEK293T cells (Supplemental Table 1).

It has been well documented that the AP-1 transcription factor directly promotes PD-L1 expression in cancer cells (33–35). Interestingly, co-IP and Western blot analysis detected the interaction of JunB, but not c-Jun, with ATXN3 in transiently transfected HEK293T cells (Supplemental Figure 6A and Supplemental Table 1), implying a possibility that ATXN3 may enhance AP-1–mediated PD-L1 gene transcription in cancer cells through JunB stabilization. Indeed, the endogenous JunB interaction with ATXN3 was confirmed (Supplemental Figure 6B and Supplemental Table 1), and ATXN3 expression largely abrogated JunB ubiquitination (Supplemental Figure 6C). Consequently, overexpression of ATXN3 dramatically increased JunB protein expression and prolonged JunB half-life (Supplemental Figure 6, D and E, and Supplemental Table 1), indicating that JunB stabilization by ATXN3 may be involved in PD-L1 upregulation in cancer cells. Collectively, our study reveals that ATXN3 is a positive regulator for PD-L1 transcription through stabilizing multiple transcription factors including HIF-2α, IFR1, STAT3, and JunB.

Suppression of ATXN3 enhances antitumor immunity. Tumor cells evade neoantigen-specific antitumor immunity through upregulating their cell-surface expression of checkpoint receptors including PD-L1 (36). Since ATNX3 promotes PD-L1 expression, we posed that ATXN3 suppression may enhance antitumor immunity in vivo. We then used the LLC1 Lewis lung carcinoma syngeneic tumor model to test whether targeted ATXN3 suppression enhances antitumor immunity in C57BL/6 mice. Indeed, upon LLC1 challenge, mice implanted with ATXN3-null LLC1 cells showed striking tumor rejection compared with those with WT LLC1 cells, with a dramatic reduction in both tumor volumes and weight (Figure 4, A–C). Analysis of the PD-L1 expression on CD45– tumor cells confirmed that ATXN3 deletion resulted in a dramatic reduction in PD-L1 expression (Figure 4D). As expected, tumoral ATXN3 deletion resulted in a significant increase in CD4+ and CD8+ T cell tumor infiltration (Figure 4, E and F). In contrast, the frequency of immunosuppressive FoxP3+ Tregs was significantly reduced in ATXN3-KO tumors (Figure 4G). Notably, the expression levels of immunosuppressive receptors including PD-1, PD-L1, and CTLA-4 on the surface of CD8+ T cells were all reduced (Figure 4, H–J). Further analysis of tumor-infiltrating CD8+ T cells revealed a less exhausted phenotype, with a significant reduction in the fraction of Tim3+, LAG3+, Blimp1+, and EOMES+ as well as annexin V+ CD8+ T cells (Figure 4, K–N and Q), but with a modest increase in T-bet+ cells (P = 0.1878) and CD44+CD8+ T cells (Figure 4, O and P). As a consequence, the production of tumor-infiltrating granzyme B and IFN-γ by CD8+ T cells in ATXN3-KO tumors was significantly increased (Figure 4, R and S). Further analysis of intratumoral myeloid cells saw comparable frequencies of Gr1hiCD11b+ myeloid-derived suppressor cells, CD11b+F4/80+ macrophages, and CD11c+MHC-IIhi dendritic cells between WT and ATXN3-KO tumors (Supplemental Figure 7, A and B). Furthermore, the expression levels of MHC-I, MHC-II, CD80, and CD86 on myeloid cells were comparable between WT and ATXN3-KO tumors (Supplemental Figure 7C). Similarly, both the frequency and the PD-L1 expression levels of the myeloid cells from tumor-draining lymph nodes were comparable in WT and ATXN3-KO LLC1 tumors (Supplemental Figure 7, D and E). These results indicate that ATXN3 suppression in tumor cells improves antitumor immune response through PD-L1–mediated suppression of CD8+ T cell immunity. Indeed, depletion of CD8+ T cells partially abolished the tumor growth inhibition caused by ATXN3 knockout (Figure 4T), suggesting that CD8+ T cells mediate the elevated antitumor immunity by suppressing ATXN3. Importantly, stable expression of PD-L1 on ATXN3-null LLC1 cells partially recovered the syngeneic tumor growth (Figure 4U), indicating that ATXN3 potentiates tumor evasive function through, at least in part, PD-L1–mediated suppression of antitumor immunity.

ATXN3 inhibition improves antitumor immunity partially through downregulating tumoral PD-L1 expression. (A–C) WT or ATXN3-KO LLC1 cells were injected subcutaneously into C57BL/6 mice (n = 10). Tumor growth curve (A), photograph (B), and weight (C) are shown. (D) MFI of surface PD-L1 on LLC1 tumors (n = 5). (E–G) Quantification of CD4+ (E) and CD8+ T cell (F) and Treg (G) percentages (n = 5–10). (H and I) Quantification of cell-surface PD-1 (H) and PD-L1 (I) MFI on CD8+ T cells (n = 5). (J and K) Quantification of cell-surface CTLA-4 MFI (J) and Tim3 percentage (K) in CD8+ T cells (n = 5). (L) MFI of cell-surface LAG3 and percentage in CD8+ T cells (n = 5). (M–O) Intracellular staining of Blimp1+ CD8+ T cell (M), EOMES+ CD8+ T cell (N), and T-bet+ CD8+ T cell (O) percentage in LLC1 tumors (n = 5-7). (P) Quantification of cell-surface CD44+ CD8+ T cell percentage from LLC1 tumors (n = 5–10). (Q) Apoptotic CD8+ T cells in the tumors were analyzed (n = 5–7). (R and S) Representative flow staining and quantification of intracellular cytokine staining of granzyme B+CD8+ and IFN-γ+CD8+ in CD45+ T cell populations from LLC1 tumors (n = 5). (T) Tumor growth curve and tumor photograph of C57BL/6 mice injected subcutaneously with WT and ATXN3-KO LLC1 cells with or without treatment of anti-CD8 depleting antibodies (n = 5). (U) Left: Tumor cell-surface PD-L1 expression. Right: Tumor growth of WT or ATXN3-KO LLC1 cells stably expressing PD-L1 (as shown in the left plot) in C57BL/6 mice (n = 5). A and C–S: 2-tailed unpaired t test; T and U: ordinary 1-way ANOVA. *P < 0.05, **P < 0.01,***P < 0.001.

Consistently, genetic ATXN3 suppression resulted in reduced PD-L1 expression and better B16 tumor rejection (Supplemental Figure 8, A and B). The elevated tumor rejection by ATXN3 suppression was largely diminished by CD8+ T cell depletion (Supplemental Figure 8A), confirming our initial conclusion that tumoral ATXN3 achieves its immune surveillance function in part through suppressing, either directly or indirectly, CD8+ T cell antitumor immunity. Flow cytometry analysis of CD45– tumor cells confirmed a dramatic reduction in their surface PD-L1 expression (Supplemental Figure 8B). Further analysis of CD45+ intratumoral immune cells detected a statistically significant increase in CD8+ but not CD4+ T cells (Supplemental Figure 8, C and D); however, the frequency of FoxP3+ Tregs was decreased (Supplemental Figure 8E) in ATXN3-KO tumors. Importantly, in addition to their increased frequency, CD8+ T cells in ATXN3-null tumors produced significantly higher levers of both IFN-γ and granzyme B (Supplemental Figure 8, F and G). Together with the fact that ATXN3 positively correlated with PD-L1 expression in more than 80% of human cancers, our study collectively shows that ATXN3 inhibition enhances antitumor immunity in a broad spectrum of cancers.

The finding that stable PD-L1 expression could not fully rescue the ATXN3-null syngeneic tumor growth suggests that ATXN3 executes its tumorigenic functions in part through PD-L1–independent mechanisms. However, targeted ATXN3 deletion did not affect LLC1 lung cancer cell growth and colony formation (Supplemental Figure 9, A and B), largely excluding the possibility that ATXN3 promotes LLC1 cancer cell growth. To further support this, we observed that the WT and ATXN3-KO LLC1 tumor growth was comparable in immune-compromised nude mice (Supplemental Figure 9C). In contrast, ATXN3 CRISPR deletion slightly reduced B16 (Supplemental Figure 9, D and E) but increased MC38 cell growth and colony formation in vitro (Supplemental Figure 9, F and G), implying that ATXN3 plays a diverse role in different cancer types. Importantly, targeted deletion of ATXN3 dramatically reduced PD-L1 expression in all types of tumor cells tested (Figure 1, E–H, and Supplemental Figure 2), regardless of whether their proliferation was altered or not. Therefore, our observations collectively support our conclusion that ATXN3-mediated PD-L1 expression is one of the critical mechanisms underlying its tumorigenic functions.

ATXN3 suppression improves checkpoint blockade antitumor immune therapy. Tumor cells evade antitumor immunity in part through PD-L1 expression to suppress PD-1+ T cell immune response to neoantigens. Therefore, blocking PD-1/PD-L1 binding with specific antibodies enhances antitumor immunity, which has achieved some clinical successes in treatment of human cancers (37, 38). However, checkpoint blockade immunotherapy often causes immune-related adverse events, such as autoimmune inflammatory responses in digestive system, heart, and kidney, which can be lethal (39, 40). Therefore, reducing systemic checkpoint blockade antibodies without impairing therapeutic efficacy has been considered as a future direction. Since ATXN3 suppression reduces tumor PD-L1 expression, we asked whether ATXN3 suppression in tumor cells improves anti–PD-1 therapeutic efficacy. We used a suboptimal dose, 25–100 μg per mouse, of anti–PD-1 antibody for only 3 times to treat pre-established WT and ATXN3-KO syngeneic LLC1 tumors in mice (Figure 5A). Surprisingly, when 25 μg anti–PD-1 antibody was used, we observed a more modest effect on suppressing WT tumor growth, but this dose still largely inhibited ATXN3-KO tumor growth (Figure 5B). Consistently, treatment of mice with ATXN3-null tumors with anti–PD-1 at the suboptimal dose of 50 μg nearly totally rejected the tumor (Figure 5, C–E), implying a synergistic effect of ATXN3 inhibition and anti–PD-1 therapy. Further analysis of tumor-infiltrated immune cells showed a further dramatic increase in both CD4+ and CD8+ T cells (Figure 5, F–H). Notably, the frequency of IFN-γ–producing CD8+ T cells in ATXN3-KO tumors was further increased by anti–PD-1 treatment in mice bearing WT but not ATXN3-null tumors (Figure 5, F and I). Similar results were obtained when mice with WT and ATXN3-KO tumors were treated with a higher dose of anti–PD-1 antibody (Figure 5J), implying that the dose of 50 μg per mouse is sufficient in this syngeneic model. A similar result was obtained when the B16 melanoma model was used (Supplemental Figure 8H), further supporting our conclusion that ATXN3 inhibition enhances checkpoint blockade therapy even with suboptimal anti–PD-1 treatment.

ATXN3 inhibition improves the preclinical efficacy of anti–PD-1 therapy. (A and B) Scheme representing the experimental procedure (A) and tumor growth curves (B) of C57BL/6 mice injected subcutaneously with WT or ATXN3-KO LLC1 cells and treated with PD-1 antibody (25 μg per mouse, once every 2 days, n = 5). (C–E) Tumor photograph (C), tumor growth curves (D), and tumor burdens (E) for C57BL/6 mice bearing LLC1 tumors treated with PD-1 antibody (50 μg per mouse, once every 2 days, n = 10). (F) Representative flow staining of CD4+ T cells, CD8+ T cells, and IFN-γ+CD8+ T cells in CD45+ T cell populations from LLC1 tumors (n = 10) as described in C and D. (G–I) Quantification of CD4+ T cell (F), CD8+ T cell (G), and IFN-γ+CD8+ T cell (H) percentage in CD45+ populations from LLC1 tumors (n = 10) as described in C and D. (J and K) C57BL/6 mice (6–8 weeks) were injected subcutaneously with WT or ATXN3-KO LLC1 cells and treated with PD-1 antibody (100 μg per mouse, once every 2 days, n = 5). Tumor growth curve was measured every 2 days (J), and mouse tumors were weighed at the end of the experiment (K). B, D, E, and G–K: Ordinary 1-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001.

Positive correlation of ATXN3 with PD-L1 levels in human cancers. Our data collectively documented that ATXN3 is a positive regulator for PD-L1 transcription and targeted ATXN3 inhibition enhances antitumor immunity. To further validate our findings in human cancers, we analyzed protein expression levels of ATXN3, PD-L1, IRF1, and HIF-2α in human lung adenocarcinoma tissue microarrays and melanoma tissue microarrays, which included 61 lung adenocarcinoma cases and 48 melanoma cases, respectively. Indeed, the protein expression levels of ATXN3, PD-L1, IRF1, and HIF-2α were all elevated in both lung cancer and melanoma compared with their adjacent normal tissues (Figure 6, A–C). Importantly, ATXN3 expression was positively correlated with PD-L1 as well as with the PD-L1 transcription factors HIF-2α and IRF1 (Figure 6, D and E). Therefore, our data support ATXN3 regulation of PD-L1 signaling in human cancer.

Elevated ATXN3 expression and its positive correlation to PD-L1 and its transcription factors in human lung cancer and melanoma. (A) Representative images from immunohistochemical staining of PD-L1, ATXN3, IRF1, and HIF-2α in human lung adenocarcinoma (LUAD) and melanoma patients. Scale bar: 50 μm. (B) PD-L1, ATXN3, IRF1, and HIF-2α protein levels in tumor tissues compared with normal tissues in LUAD patients (n = 61); “PD-L1%” means the percentage of PD-L1–positive area versus all tissue area. (C) PD-L1, ATXN3, IRF1, and HIF-2α protein levels in tumor tissues compared with normal tissues in patients with melanoma (n = 48). (D) Correlation analysis of ATXN3 expression with PD-L1, IRF1, and HIF-2α expression in LUAD patients (n = 61). (E) Correlation analysis of ATXN3 expression with PD-L1, IRF1, and HIF-2α expression in melanoma patients (n = 48). (F) ATXN3 is a positive regulator for PD-L1 transcription through stabilizing multiple transcription factors including HIF-2α, IFR1, STAT3, and JunB and enhances tumor evasion. B and C: 2-tailed unpaired t test; D and E: Pearson’s correlation analysis. **P < 0.01, ***P < 0.001.

Collectively, our study identified ATXN3 as a positive regulator of PD-L1 transcription in tumors through stabilizing a group of PD-L1 transcription factors including HIF-2α, IRF1, STAT3, and JunB in response to extracellular stimuli such as IFN-γ and hypoxia. This ATXN3-mediated PD-L1 upregulation enhances tumor evasion of antitumor immunity (Figure 6F). Therefore, targeted ATXN3 suppression enhances antitumor immunity and improves the preclinical efficacy of antitumor immune therapy.