HDAC3 Inhibition Upregulates PD-L1 Expression in B-Cell Lymphomas and Augments the Efficacy of Anti–PD-L1 Therapy

The programmed cell-death protein 1 (PD-1)/programmed cell death-ligand 1 (PD-L1) axis has been characterized as a potent inhibitor of immune activation and an important mechanism underlying tumor immune escape (1, 2). PD-1 is generally highly expressed on tumor-infiltrating T cells. Interaction of PD-1 with its ligand PD-L1 on tumor cells or immune cells can inhibit the initial activation of T cells and suppress effector T-cell generation and functions, including cytokine production and cytotoxicity (3, 4). PD-L1 is expressed in various types of human cancer including Hodgkin lymphoma (5) and diffuse large B-cell lymphomas (DLBCL; ref. 6), and suppresses antitumor T-cell response (7). Blockade of the PD-L1/PD-1 axis can reactivate antitumor immunity and induce durable clinical response against a growing list of human cancers including B-cell lymphoma (1, 2, 8). In cancer patients, the clinical response to PD-L1/PD-1 immunotherapy correlates with tumor and host PD-L1 expression, along with other predictive biomarkers such as tumor-infiltrating CD8⁺ T cells and mutation burden (9, 10). Accumulating evidences have indicated that altered PD-L1 expression by small molecules can modify the efficacy of anti–PD-1/PD-L1 therapy in preclinical mouse model. For example, CDK4/6 inhibitor synergizes with anti–PD-1 antibody to elicit an enhanced therapeutic efficacy by increasing tumoral PD-L1 expression (11).

PD-L1 expression on tumor cells can be regulated at both transcriptional and posttranslational levels (7). The transcriptional control of PD-L1 has been extensively studied during the past several years. Many transcription factors, including IRF1, STAT1/3, and MYC, have been reported to bind to the promoter region of PD-L1 gene and regulate its transcription (12–15). Inflammatory signals modulate PD-L1 gene expression via these transcription factors. For example, IFNγ, a potent inducer of PD-L1 transcription, functions by activating the JAK/STAT/IRF1 pathway (13). These transcription factors directly bind to the transcriptional regulatory elements of PD-L1 gene and alter chromatin status via their associated epigenetic regulators. Histone deacetylases (HDAC) are ubiquitously expressed epigenetic regulators by removing acetyl groups from the N-acetyl lysine amino acid on the tail of histone to silence gene transcription (16). HDACs have been found to deacetylate nonhistone proteins that are associated with various functions such as gene expression (17). Various HDAC isoforms differ in their subcellular location and targets, which account for their different biological functions. Recently, pan-HDACs or class-specific HDAC inhibitors have been reported to induce PD-L1 expression in lung cancers and melanoma (18, 19), suggesting that HDACs are involved in the regulation of PD-L1. However, which HDAC isoform is crucial for the regulation of PD-L1 and the molecular mechanisms of action remain unknown.

HDAC inhibitors are well known to affect cancer cell viability and are already in use for the treatment of some subtypes of hematologic malignancies (20, 21). In addition to direct cytotoxicity, HDAC inhibitors can alter the immune landscape of tumor cells, through changes in expression of costimulatory and coinhibitory molecules, MHC and tumor antigens, as well as cytokine production by tumor cells (21–23). HDAC inhibitors can also affect the phenotypes of different immune cell subsets in the tumor environment and draining lymph node (24, 25). HDAC inhibitors have been proposed to potentially synergize with immunotherapies (21, 26). Nonselective HDAC inhibitors have been recently reported to augment the antitumor effect of anti–PD-1 antibody by increasing PD-L1 expression in syngeneic murine tumor models including lung cancers and melanoma (18, 27). We reasoned that HDAC inhibition may be a rational therapeutic strategy to be implemented in combination with PD-1 blockade for the treatment of B-cell lymphoma. However, nonselective HDAC inhibitors target multiple HDACs and may cause serious unfavorable toxicities such as thrombocytopenia, fatigue, and diarrhea, limiting their clinical application (28). It is thereby ideal to find out HDAC isoform–specific inhibitors which can be used to promote PD-L1 expression and the clinical response to PD-1 immunotherapy while avoiding the adverse events associated with pan-HDAC inhibition.

In this study, we report that, in B-cell lymphomas, HDAC3 is a key regulator of PD-L1 transcription via direct and indirect mechanisms, and HDAC3 inhibition augments the response to anti–PD-1 blockade in a syngeneic murine lymphoma model.

Nonselective HDAC inhibitors SAHA, LBH589 (panobinostat), valproic acid sodium salt (VPA), trichostatin A (TSA), and HDAC1/2-slective inhibitor romidepsin as well as HDAC3-selective inhibitor RGFP966 were purchased from Selleck Chemicals, and its chemical structure was described previously (29). Bromodomain inhibitor JQ1 and DNA methyltransferase inhibitor 5-Azacitidine (5-Aza) were also obtained from Selleck Chemicals. The BCL6 BTB inhibitor FX1 was provided by professor Ari Melnick (Weill Cornell Medical College).

The B-lymphoma cells lines A20, OCI-LY1, and TMD8 were obtained from the ATCC and authenticated via short tandem repeat profiling. OCI-LY1 cells were maintained in medium containing 90% Iscove and 10% FCS and supplemented with antibiotics. A20 and TMD8 cells were maintained in medium containing 90% RPMI and 10% FCS supplemented with antibiotics, l-glutamine, and 2-[4-(2-Hydroxyethyl)-1-piperazinyl]ethanesulfonic acid (HEPES) Potential contamination by mycoplasma was often checked using the Lookout Mycoplasma PCR Detection Kit purchased from Sigma-Aldrich. Scramble siRNA, HDAC3 siRNA (#1: 5-CCCGGTGTTGGACATATGAAA-3 and #2: 5-GACACCCAATGAAACCTCATCGCCT-3), and BCL6 siRNA (5-CCATTGTGAGAAGTGTAACCTGCAT-3) were purchased from Shanghai GenePharma Company. siRNAs were transiently electroporated into lymphoma cell lines using cell line nucleofector transfection kit (AMAXA) according to the manufacturer's instruction. To generate PD-L1 knockout (PD-L1^ko) cells, A20 cells were transiently transfected with a Cas9-single guide RNA (sgRNA) expression vector (pSpCas9(BB), Addgene) targeting PD-L1 or control guide RNAs. PD-L1 sgRNA, 5′-GTATGGCAGCAACGTCACGA-3′. Ten days after transfection, PD-L1^ko cells were sorting on PD-L1–negative cells by flow cytometer.

Total RNA was prepared with Trizol reagent (Invitrogen), and cDNA was synthesized using Superscript reverse transcriptase and random primers (Invitrogen). Quantitative real-time PCR (qPCR) was performed using the Power SYBR Green PCR master mix (Vazyme) on an ABI Prism sequence detection system (Applied Biosystems). Gene-specific primers are listed in Supplementary Table S1.

Single-cell suspensions were prepared from fresh mouse tumor tissues or cultured cells. Antibodies used for staining were listed in the Supplementary Data. Cell surface staining was performed in FACS buffer (PBS with 2% FBS, 2 mmol/L EDTA, and 0.05% sodium azide). All flow cytometry data were acquired on a Fortessa X-20 (BD Biosciences), and live cells were gated for analysis with FlowJo software (Tree Star).

Six- to 8-week-old wild-type BALB/c mice were obtained from the Shanghai SLAC Laboratory Animal Co. Ltd. Mice were maintained in a specific pathogen-free facility, and all animal experiments were performed in accordance to protocols approved by the Institutional Animal Care and Use Committee of Shanghai Jiao Tong University, School of Medicine. Mice were s.c. inoculated with 5 × 10⁶A20 cells. RGFP966 was injected by the i.p. route at a dose of 50 mg/kg beginning on day 7 after tumor implantation and continued daily for 21 days. Alternatively, 200 μg anti–PD-L1 blocking antibodies (BE101) from BioXCell were injected by i.p. for 3 times (once every 3 days) from day 7 after tumor implantation. Solvent (2% DMSO, 30% PEG300, and 5% Tween 80) was used in the treatment control group. Tumor size were monitored with a digital caliper every 3 days and expressed as volume (0.5 × length × width × height).

Analyses were performed using Prism 6.0 (GraphPad). Statistical significance was calculated using the two-tailed, unpaired Student t test or one-way ANOVA, as specified. P values < 0.05 were considered statistically significant.

Other Materials and Methods are provided as Supplementary Data.

Pan- and selective class I/IV HDAC inhibitors have been reported to increase PD-L1 expression in melanoma and lung cancer cell lines (18, 19). We hypothesized that HDAC inhibition could also upregulate PD-L1 expression in B-cell lymphoma. A20 lymphoma cells were treated with SAHA, a pan HDAC inhibitor, for 48 hours at doses ranging from 0.1 to 1 μmol/L that caused less than 30% cell death at 48 hours (Supplementary Fig. S1A). SAHA treatment significantly increased surface PD-L1 levels in a dose-dependent manner (Fig. 1A). Note that 1 μmol/L SAHA induced an approximately 3-fold increase in PD-L1 levels. The clinically achievable concentration of SAHA is 500 nmol/L to 4 μmol/L (30). Accordingly, we used 1 μmol/L SAHA for the following experiments. Of note, upregulation of surface PD-L1 levels started as early as 12 hours after SAHA treatment without obvious cell death (Fig. 1B; Supplementary Fig. S1B). Consistent with previous reports (31, 32), 1 μmol/L SAHA upregulated the expression of many costimulatory molecules such as CD80, CD86, MHC I, and MHC II in A20 cells (Supplementary Fig. S1C). Real-time PCR assay further showed that SAHA treatment increased the mRNA abundance of PD-L1, indicating that SAHA regulates PD-L1 expression at the transcriptional levels (Fig. 1C). Upregulation of surface expression of PD-L1 and its transcript was observed in two human DLBCL cell lines OCI-LY1 and TMD8 treated with SAHA (Fig. 1D; Supplementary Fig. S1D). Finally, both TSA and LBH589, two pan-HDAC inhibitors, also induced PD-L1 expression in A20 cells (Supplementary Fig. S1E). Collectively, these results demonstrate that HDAC inhibition can increase the transcription and surface expression of PD-L1 in B-cell lymphomas.

We next sought to understand how HDAC inhibition increases PD-L1 transcription in B-cell lymphoma cells. HDAC inhibitors are known to modulate global gene expression mainly via the alteration of acetylation “marks” on histones, which opens up chromatin structures and recruits “acetyllysine reader” to the acetylated sites, subsequently activating downstream target gene expression (16). HDAC inhibition has been reported to modulate gene transcription by enhancing acetylation of histone H3 at lysine 4 (H3K4ac), lysine 9 (H3K9ac), or lysine 27 (H3K27ac) of promoters. Accordingly, we performed chromatin immunoprecipitation (ChIP) assay using antibodies against H3K4ac, H3K9ac, and H3K27ac in A20 cells treated with SAHA for 9 hours followed by qPCR analysis using three different pairs of primers (Fig. 2A). SAHA treatment increased the acetylation of H3K4, H3K9, and H3K27 at the promoter region of PD-L1 gene, within an approximately 1.5 kb region 5′ to its transcription start site (Fig. 2A).

According to previous reports, the bromodomain and extraterminal protein BRD4 directly binds to acetylated lysine on histone tails to facilitate gene transcription by RNA polymerase II (33). IFNγ treatment enhanced the H3K27ac mark and BRD4 recruitment at the promoter region of PD-L1 in melanoma cells (34, 35). In addition, BRD4 inhibition or knockdown inhibited basal or IFNγ-induced PD-L1 transcription (13). We, therefore, tested whether BRD4 was involved in HDAC inhibition–mediated PD-L1 transcription. ChIP-qPCR assay showed that SAHA treatment increased BRD4 occupancy at the promoter region of PD-L1 gene in A20 cells (Fig. 2B). Furthermore, JQ1, a pharmacologic inhibitor of BRD4 that functions by dissociating BRD4 from the chromatin (33), dramatically blocked SAHA-induced PD-L1 upregulation in A20 cells (Fig. 2C; Supplementary Fig. S2), whereas JQ1 alone had only a marginal effect on basal PD-L1 expression (Supplementary Fig. S2). JQ1 treatment could also prevent SAHA-induced PD-L1 upregulation in OCI-LY1 and TMD8 cells (Fig. 2D). Collectively, these results demonstrate that HDAC inhibition upregulates PD-L1 transcription by increasing histone acetylation of the PD-L1 promoter.

We next investigated which HDAC isoform plays an important role in modulating PD-L1 transcription. We focused particularly on Class I HDACs because they are ubiquitously expressed by B-cell lymphomas and regulate gene transcription mainly by affecting the acetylation status of histone tails. Class I HDACs have four members including HDAC1, HDAC2, HDAC3, and HDAC8. HDAC1, HDAC2, and HDAC3 are well characterized and potential therapeutic targets for B-cell lymphomas. Both Class I HDAC inhibitor VPA and selective HDAC1/2 inhibitor Romidepsin dramatically increased acetylated histone H3 (H3ac) and acetylation of H3K27 in A20 cells 9 hours after treatment, whereas selective HDAC3 inhibitor RGFP966 resulted in a less profound, but observable increase in H3ac and H3K27ac (Fig. 3A). Moreover, VPA, romidepsin, or RGFP966 impaired cell survival and upregulated the surface levels of many costimulatory molecules such as CD80, CD86, and MHC I in A20 cells although to a different extent at 2 days of treatment (Supplementary Fig. S3A and S3B). Interestingly, treatment with either VPA or RGFP966 markedly enhanced surface PD-L1 levels in A20 cells and two human B-cell lymphoma cells at 2 days of treatment (Fig. 3B). In contrast, Romidepsin had no appreciable or weak effect on expression of PD-L1 in these cell lines (Fig. 3B). Similar to SAHA, RGFP966 treatment also increased mRNA abundance of PD-L1 in these cell lines (Fig. 3C). Moreover, siRNA-mediated depletion of HDAC3 upregulated surface PD-L1 levels in A20 cells (Fig. 3D and E). Finally, the mRNA expression levels of HDAC3 and PD-L1 were inversely correlated in 203 cases of human DLBCLs(Fig. 3F). These results indicate that HDAC3 is a critical HDAC isoform that inhibits PD-L1 expression in B-cell lymphoma.

Our aforementioned results showed that pan-HDAC inhibitors promoted PD-L1 transcription by increasing histone acetylation at the PD-L1 promoter (Fig. 2). We next investigated whether HDAC3 inhibition upregulated PD-L1 expression via the similar mechanism. ChIP-qPCR assay showed that RGFP966 treatment increased acetylation of H3K27 at the promoter region of PD-L1 (Fig. 4A). Accordingly, JQ1 treatment largely reversed RGFP966-enabled PD-L1 upregulation in A20 cells (Fig. 4B). These results indicate that HDAC3 inhibition induces histone acetylation to facilitate PD-L1 transcription.

We next sought to identify the mechanism by which HDAC3 regulates histone acetylation of the PD-L1 promoter. HDAC3 lacks DNA-binding domain and is unable to directly bind to DNA. HDAC3 is frequently associated with SMRT and NCOR corepressor complex that are recruited to the regulatory sequences of target genes by transcription factors (36). The transcription repressor BCL6 (B-Cell Lymphoma 6) is a master oncogene in B-cell lymphoma and modulates gene transcription through recruitment of SMRT/NCOR/HDAC3 repressor complex (37–39). Previous ChIP-seq studies demonstrated that BCL6 directly binds to the promoter of PD-L1 and is suggested to regulate its transcription. By analyzing our published ChIP-seq of human germinal center (GC)-derived lymphoma cell line OCI-LY1 (GEO number: GSE29282; ref. 39), we found that BCL6 and SMRT cobound to the promoter of PD-L1 gene in OCI-LY1 (Fig. 4C). ChIP-qPCR assay confirmed that HDAC3, SMRT, and BCL6 bound to the same locus at the promoter region of PD-L1 in these cells (Fig. 4C). HDAC3 was coimmunoprecipitated with anti-BCL6 antibodies, but not control IgG in A20 cells (Fig. 4D). Moreover, siRNA-mediated BCL6 knockdown significantly reduced the enrichment of HDAC3 at the PD-L1 promoter (Fig. 4E). Finally, BCL6 BTB inhibitor FX1, which disrupts the interaction between the BCL6 BTB domain and SMRT/HDAC3 complex, increased PD-L1 expression (Fig. 4F). Collectively, these results suggest a model in which HDAC3–SMRT corepressor complex was recruited to the promoter of PD-L1 by BCL6 to modulate histone acetylation.

DNA methyltransferase 1 (DNMT1) is required to maintain DNA methylation across the genome (40). 5-Aza mainly targets DNMT1 and DNMT1 knockdown upregulates hypermethylated endogenous retrovirus genes in cancer cells, subsequently causing an IFN response to activate immune-related genes including PD-L1 (41, 42). Interestingly, these 5-Aza–induced genes are not generally methylated at promoter regions (41, 42). Recently, it has been uncovered that HDAC3 inhibition led to degradation of DNMT1 in multiple myeloma (43). We reasoned that HDAC3 inhibition might reduce DNMT1 protein level, subsequently activating PD-L1 transcription in A20 cells. We observed that treatment with either SAHA or RGFP966 resulted in a decrease of DNMT1 protein abundance in a dose-dependent manner (Fig. 5A). Interestingly, DNMT1 protein levels did not change 12 hours after treatment (Fig. 5A), and PD-L1 started to increase as early as 12 hours (Fig. 5B), suggesting that DNMT1 degradation contributes to PD-L1 upregulation at late-time point. Consistent with this, 5-Aza increased PD-L1 expression 48 hours after treatment (Fig. 5B). Interestingly, the PD-L1 promoter was hypomethylated in A20 cells, and RGFP966 treatment did not cause noticeable changes in DNA methylation patterns in the PD-L1 promoter (Supplementary Fig. S4). Notably, simultaneous treatment with 5-Aza and RGFP966 caused a more striking upregulation of surface PD-L1 in A20 cells as compared with treatment with each single inhibitor (Fig. 5B). Taken together, these results indicate that HDAC3 inhibition induces DNMT1 degradation, subsequently activating PD-L1 transcription.

We next asked whether HDAC3 inhibition increases PD-L1 expression on tumor cells in syngeneic murine lymphoma model. To this end, mice were injected s.c. with A20 cells. Seven days later, mice were administrated with RGFP966 (50 mg/kg) for 9 consecutive days, and tumor tissues were then removed for flow cytometric analysis of PD-L1 expression. This dose of RGFP966 has been previously described (29). We observed that RGFP966 treatment led to a 2-fold increase in PD-L1 expression on tumor cells (Supplementary Fig. S5A). Because PD-L1 expression on host cells also affects the antitumor activity of anti–PD-L1 immunotherapy (44, 45), we investigated the effect of RGFP966 on PD-L1 expression on immune cells including dendritic cells (DC, CD11c⁺), myeloid-derived suppression cells (MDSC, CD11b⁺Gr1⁺), and macrophages (CD11b⁺F4/80⁺) in the tumor microenvironment and draining lymphoid nodes. An increase in the expression of PD-L1 was observed on tumor-infiltrating DCs, but not MDSCs and macrophages (Supplementary Fig. S5B). PD-L1 expression was not significantly altered in these immune cells in the draining lymphoid nodes (Supplementary Fig. S5C). Taken together, HDAC3 inhibition increased PD-L1 expression on tumor cells and on DCs in the tumor environment in vivo.

Recent preclinical studies have revealed that both tumor and host-derived PD-L1 can suppress antitumor immunity (44–46). To clarify the relative contributions of PD-L1 upregulation on tumor cells in inhibiting the therapeutic response to HDAC3 inhibition, we generated PD-L1^KO A20 cells (Supplementary Fig. S6A). PD-L1 deficiency did not affect cell growth in vitro, but significantly delayed tumor growth (Supplementary Fig. S6B and S6C). Moreover, RGFP966 treatment retarded PD-L1^KO, but not wild-type tumor growth, indicating that RGFP966-induced upregulation of PD-L1 represents an important mechanism underlying RGFP966 resistance. We next investigate whether HDAC3 inhibition might enhance the antitumor effect of anti–PD-L1 therapy in vivo. To test this hypothesis, A20 cells were implanted, and 7 days later, tumors were treated with RGFP966 (50 mg/kg), anti–PD-L1 alone (200 μg per mouse), or a combination of both agents (Fig. 6A). Treatment with RGFP966 alone did not affect tumor growth and failed to lead to tumor regression (0/6). Anti–PD-L1 antibodies alone effectively controlled tumor growth, and 2 of 10 tumors were finally cleared. Interestingly, RGFP966 in combination with anti–PD-L1 antibody markedly retarded tumor progression and resulted in 7 complete responses out of the 11 treated mice (Fig. 6B). As expected, SAHA treatment could increase the antitumor effect of anti–PD-L1 antibody in syngeneic murine lymphoma model (Supplementary Fig. S7). These results demonstrate that HDAC3 inhibitor synergized with anti–PD-L1 therapy to elicit an enhanced therapeutic efficacy.

The epigenetic regulation of PD-L1 in B-cell lymphoma remains to be elucidated. Here, we found that HDAC inhibitors can increase the mRNA abundance and surface levels of PD-L1 on a panel of murine and human B-cell lymphoma cell lines (Fig. 1). Mechanistically, HDAC inhibition led to a rapid increase in histone acetylation at the promoter region of PD-L1 and subsequent recruitment of BRD4 to drive its transcription (Fig. 2). Accordingly, BRD4 inhibitor JQ1 dramatically suppressed the upregulation of PD-L1 triggered by HDAC inhibition.

Using specific HDAC inhibitors and gene knockdown, we found that HDAC3, but not HDAC1/2, is the key HDAC isoform that is responsible for the regulation of PD-L1 transcription (Fig. 3). HDAC3 is likely to be the primary enzyme responsible for the deacetylase activity that is associated with SMRT-mediated repressive events. ChIP-seq further showed that BCL6 recruited HDAC3–SMRT corepressor complex to the PD-L1 promoter (Fig. 4). Therefore, BCL6–HDAC3–SMRT complex appears to locate to the PD-L1 promoter to suppress its transcription by directly deacetylating histones (Fig. 6C). In addition to BCL6, other transcription factors may be involved in the recruitment of SMRT–HDAC3 complex to the PD-L1 promoter. In addition to increased histone acetylation at the PD-L1 promoter, DNMT1 degradation may contribute to PD-L1 upregulation at late time point after HDAC3 inhibition (Fig. 6C). HDAC3 inhibition did not directly affect DNA methylation of the PD-L1 promoter (Supplementary Fig. S4), but may trigger an IFN response to activate PD-L1 expression via DNMT1 degradation.

Many types of B-cell lymphomas depend on HDAC3 for their survival and proliferation (47, 48). Targeting HDAC3 has been a promising therapy for B-cell lymphoma. However, HDAC3 inhibition–induced increase of PD-L1 could be one of the underlying mechanisms accounting for HDAC3 inhibitor resistance via evasion of immune surveillance checkpoints (Fig. 6C). This notion is supported by our observation that RGFP966 significantly retarded PD-L1–deficient, but not wild-type, tumor growth (Supplementary Fig. S6). RGFP966 failed to eradicate PD-L1–deficient lymphomas. In contrast, combination of RGFP966 and anti–PD-L1 antibody can completely eradicate the majority of wild-type lymphoma (Fig. 6B). Therefore, RGFP966 augments the therapeutic effect of PD-L1 blockade by increasing PD-L1 expression on tumor and DCs. Of note, HDAC3 is ubiquitously expressed in tumor cells and effector immune cells within tumor microenvironment. HDAC inhibitors have been reported to elevate T-cell chemokine expression to argument responses to PD-1 immunotherapy in lung cancer (49). In addition to PD-L1, HDAC3 inhibition may modulate immune-related genes in these cells to enhance anti–PD-L1 therapy. Therefore, PD-L1 upregulation represents an important, but not unique, mechanism by which RGFP966 augments the response to anti–PD-L1 therapy.

In summary, we identified HDAC3 as an important epigenetic regulator of PD-L1 in B-cell lymphoma and the molecular mechanisms by which HDAC3 suppresses PD-L1 transcription. Selective HDAC3 inhibitor facilitates the clinical response to PD-L1 immunotherapy while avoiding adverse events associated with pan-HDAC inhibition. Our study also provides the preclinical rationale for combination of selective HDAC3 inhibitors with PD-1/PD-L1 immunotherapy in the treatment of B-cell lymphomas.

No potential conflicts of interest were disclosed.

Conception and design: C. Huang

Development of methodology: S. Deng, Q. Hu, C. Peng, C. Huang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Deng, Q. Hu, H. Zhang, F. Yang, C. Peng

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Deng, Q. Hu, H. Zhang, F. Yang, C. Peng, C. Huang

Writing, review, and/or revision of the manuscript: C. Peng, C. Huang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Peng, C. Huang

Study supervision: C. Peng, C. Huang

We thank professor Ari Mlenick (Weill Cornell Medical College) for providing FX1. This work was supported by the National Natural Science Foundation of China (grant numbers: 31870872, to C. Huang; 31800755, to C. Peng) and the 1000-Youth Elite Program.

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.