Inhibition of LSD1 Attenuates Oral Cancer Development and Promotes Therapeutic Efficacy of Immune Checkpoint Blockade and YAP/TAZ Inhibition

Recent statistics show that oral cavity-related cancers account for 53,000 cases and 10,860 deaths in the United States annually (1, 2). The most common type of head and neck cancer is oral squamous cell carcinoma (OSCC). Treatment for OSCC frequently involves a combination of surgery, radiotherapy, and chemotherapy. Resistance to therapy complicates treatment effectiveness and the 5-year survival rate remains at ∼65% (3). In particular, epigenetic modifications can result in acquired resistance to treatment, leading to poor outcomes among patients with OSCC. However, the role of epigenetic mediators is not well understood, hindering further therapeutic development. One such aberrantly upregulated epigenetic regulator is Lysine-specific demethylase 1 (LSD1), which is encoded by the KDM1A gene. LSD1 is a flavin adenine dinucleotide—dependent amine oxidase. LSD1 also promotes cancer stem cells and chemoresistance (4, 5). We also showed that LSD1 inhibition attenuates OSCC and key signaling pathways, including EGFR signaling. EGFR is also implicated in OSCC progression and therapeutic resistance. The composition of the tumor microenvironment could attenuate immunosurveillance (6). Consistent with immune altering roles, ablation of LSD1 promotes antitumor immunity in combination with checkpoint blockade in melanoma (2).

LSD1 has also been shown to influence the Hippo signaling effector YAP (7). YAP is a transcriptional effector that has been shown to contribute to OSCC development, regulating pro-tumorigenic gene expression (8–10). While connections between LSD1 and pro-tumorigenic signaling effectors such as EGFR and YAP have been identified, the relationship between LSD1 and YAP in OSCC development and potential therapeutic application is not well understood. To gain insight into the roles of LSD1 in OSCC development, we examined LSD1 function using the tobacco smoke toxin, 4 Nitroquanolone-1-oxide (4NQO), model of OSCC. This model captures pathologic changes similar to those found in human OSCC (11–15).

Here, we evaluated the role of LSD1 in the genetic model in combination with 4NQO and determined if the topical application of the LSD1 inhibitor could attenuate its progression to OSCC. We evaluated the topical application of the LSD1 inhibitor in OSCC mice and further report that conditional deletion of Lsd1 in expanding oral epithelium reverses tumorigenic phenotypes in 4NQO treated mice. LSD1-inhibited epithelium revealed signaling networks impacted by LSD1 in tumor development, including YAP/TAZ and PD-L1 pathways, which play a critical role in OSCC. Combination therapies targeting these pathways showed improved targeting of 4NQO-induced OSCC, suggesting potential co-inhibition of these signals in OSCC therapy.

All the experiments were performed with prior approval of the Institutional Animal Care and Use Committee, Boston University Medical Campus, Boston University. C57BL/6J mice were fed with 4NQO (100 μg/mL in propylene glycol and water) in their drinking water for 16 weeks, followed by regular water for the remainder of the experiment using a well-established protocol (16). The progressive changes at different stages of OSCC were determined at specific time points. To evaluate if LSD1 expression progressively increases in our models as in human OSCC, we performed an immunostaining analysis of tongue OSCC. All time points are hereafter indicated as weeks post-4NQO treatment.

Lsd1-deficient mice were generated by crossing conditional floxed mice (loxP-Lsd1-loxP; ref. 17) with K14 promoter-driven tamoxifen-inducible Cre mice (K14CreERT, Jackson Laboratory, Stock No. 005107). Male and female mice were fed 4NQO in their drinking water. Given the early increase in LSD1 expression post-4NQO treatment (Supplementary Fig. S1), we treated mice at week 8 with tamoxifen to induce Lsd1 deletion. After 8 weeks of initiating 4NQO treatment, mice were randomized into two groups: half were treated with vehicle and half with tamoxifen. Tamoxifen was applied topically onto the tongue (75 mg/kg body weight, every 24 hours for 5 consecutive days, as per standard protocol; ref. 18) for tongue epithelium-specific deletion of Lsd1. K14 promoter-driven conditional Lsd1-floxed mice treated with vehicle were designated Lsd1^WT/WT, whereas tamoxifen-treated mice with deleted Lsd1 were designated Lsd1^−/− for homozygotes. The immunostaining and hematoxylin and eosin (H&E) staining were performed on n = 5–8/group and 4 sections/mouse.

RNA sequencing (RNA-seq) was performed with 100 ng of total RNA using BGISEQ-500 (Beijing Genomic Institution, https://www.genomics.cn/, BGI) and published in other studies (19, 20). Raw FASTQ sequencing reads were mapped against the reference genome of Mus musculus (mm10). Total read counts and normalized transcripts per million were obtained. Before, differential expression, batch effects or sample heterogeneity was tested using iSeqQC (21). Differential gene expression was tested between Lsd1^−/− and Lsd1^WT/WT samples using DESeq2 package in R/Bioconductor. To evaluate the biological processes implicated in LSD1 deleted tongue, functional annotations were performed on the differentially expressed genes using DAVID (22). Finally, to characterize the modulation of the immune cell after the LSD1 deletion, we used the deconvolution methods as used in the reported study (23). The gene signatures for 29 different immune cell types were obtained from the study (24) and supervised gene set enrichment analysis using GSEA-v4.0 (25) was performed.

The cell line authentication was performed by Genetica, Inc and routine Mycoplasma testing was performed using Mycoplasma Detection Kit from Southern Biotech (Catalog No. 13100–01). Low passage number HSC-3 and CAL-27 cells were grown in 100-mm³ plates for 48 to 72 hours or at confluency less than 90%. Next, proliferation assays were performed by plating HSC-3 and CAL-27 cells (20,000 cells per well, 6-replicates per treatment) treated for 48 hours with either vehicle control or with SP2509, then evaluated by CyQUANT Assay. Clonogenic survival was evaluated after 14 and 21 days (26). Most of the inhibitors used in our in vitro and in vivo studies are being used for clinical applications. SP2509 (MedChemExpress) is a reversible LSD1 enzyme activity inhibitor (27) and has similar properties to clinical compound SP-2577 (Seclidemstat; ref. 27). Verteporfin (Sellekchem, Inc) is a YAP inhibitor (28–30). EGFR kinase inhibitor Erolitinib (MedChemExpress; ref. 31) and TGFβ receptor type I (TGFβRI) kinase inhibitor Galunisertib (LY2157299; Sellekchem, Inc; ref. 32) are also currently being tested in clinical trials.

Two studies were performed simultaneously with the different combinations. The first study was performed with 40 total C57BL/6 male and female mice with 4NQO for 18 weeks. They were randomized into 4 groups (n = 10/condition) and started treatment with: (i) vehicle, (ii) SP2509, (iii) Verteporfin, and (iv) SP2509 + Verteporfin. The route of drug administration for SP2509 and Verteporfin in 25-μL corn oil, 5% DMSO were topical applications onto the tongue after brief anesthesia with local distribution. The naïve group (n = 5) was without treatment of 4NQO as a control. We optimized a dose of SP2509 to 30 mg/kg and Verteporfin 2.5 mg/kg three times a week for 5 weeks by local application in DMSO and corn oil.

In the second study, a total 60 C57BL/6 male and female mice treated with 4NQO for 18 weeks were randomized into different groups: (i) IgG isotype control antibody, (ii) SP2509, (iii) anti–PD-1 antibody, 4) SP2509+ anti–PD-1 antibody, 5) anti–PD-L11 antibody, and 6) SP2509+ anti–PD-L1 antibody. Anti-mouse PD-1 (CD279; Clone: 29F.1A12; Bio X Cell; ref. 33) and anti-mouse PD-l1 (B7-H1; Clone: 10F.9G2; Bio X Cell; ref. 34) were injected intraperitoneally in the respective groups at 10 mg/kg (100 μL buffer). SP2509 in 25-μL corn oil, 5% DMSO was applied topically onto the tongue.

Total RNA was extracted by the TRizol protocol according to the manufacturer's instructions (Qiagen). RT-qPCR analysis was performed using TaqMan gene expression assays from Life Technologies, according to the standard protocol. The gene expression was normalized using three housekeeping genes: RSP18, GAPDH, and β-actin.

Tongue sections were stained with H&E and evaluated for pathology by a Board Certified pathologist.

Pearson rho and Spearman rho correlation of KDM1A with other genes were evaluated using Xena UCSC genome browser (https://xena.ucsc.edu) to evaluate clinical databases such as the Therapeutically Applicable Research to Generate Effective Treatments (TARGET; 5968 samples). Pearson rho and Spearman's rank rho correlation were considered very high and positive for value 1, whereas the negative correlation was –1.

To gain insight into the dynamics of LSD1 in OSCC, we crossed Lsd1-floxed mice obtained from Stuart Orkin laboratory (Mass General Hospital, Boston; ref. 17) with mice expressing Cre recombinase from the Krt14 promoter tumor studies (35), allowing us to conditionally delete the Lsd1 gene in Krt14-expressing cells that expand in OSCC. We treated these animals and control animals with the tobacco carcinogen 4NQO (experimental outline is depicted in Fig. 1A), which induced OSCC that mirrors the progressive onset of the human disease (11–15). Lsd1^−/− mice sacrificed at 24 weeks post-4NQO exposure showed a total absence or less severe gross pathologic development of OSCC compared with control Lsd1^WT/WT mice (Fig. 1B). Gene expression changes identified by RT-qPCR demonstrated reduced expression of OSCC related genes that are known to be regulated by LSD1 (7), including reduced expression of Ccn1/Ctgf, Yap, Wwtr1/Taz, Egfr, as well as Lsd1 itself (Fig. 1C). The survival of Lsd1^−/− mice was reduced compared with Lsd1^WT/WT mice (Supplementary Fig. S2); however, this was not statistically signficant. The volume of tumors in the tongues of Lsd1^−/− mice was reduced compared with Lsd1^WT/WT mice (Fig. 1D) and H&E staining of tissues isolated from the tongues showed reduced OSCC development in Lsd1−/− mice at week 25 post-4NQO treatment compared with Lsd1^WT/WT mice (Fig. 1E). Immunostaining analysis showed that LSD1 expression was reduced, whereas H3K4me1, which is the substrate of LSD1 enzyme, was increased (Fig. 1F).

RNA-seq analysis of Lsd1^−/− mouse tongues subjected to 4NQO showed differential expression of gene encoding factors associated with several oncogenic signaling pathways compared with Lsd1^WT/WT mice tongues (Fig. 2A–C). Pathway analysis of these differentially expressed genes suggested broad effects on the tumor microenvironment with a reduction in EGFR activity, JNK signaling, IL6 production, inflammatory responses, innate immune responses, and proteolysis. A comparative analysis Lsd1^−/− gene expression with published signature from 29 different cell types (24) suggested that Lsd1^−/− mouse tongues exhibit elevated neutrophils, T-regulatory cells, memory cells, and activated dendritic cells, along with a reduction in M2 macrophages, T helper cells, gamma-delta T cells, and eosinophils. Overall, candidate-based and RNA-seq analysis further suggested that Lsd1 deletion could affect YAP/TAZ-mediated effects, immune regulation, and immune checkpoints.

Reversible inhibitors of Lsd1 have shown the ability to reduce the growth of several tumor cell models (27). Accordingly, we found that treatment of human OSCC HSC-3 cells with the LSD1 inhibitor SP2509 showed a dose-dependent reduction in proliferation (Fig. 3A) and clonogenic survival (Fig. 3B and C). Our observations suggested that LSD1 intersects with several signaling pathways in OSCC development (7), which prompted us to further explore these relationships for potential therapeutic applications. Specifically, we tested the effects of SP2509 together with inhibitors of YAP (Verteporfin), TGFβ (Galunasirtib), and EGFR (Erlotinib). Verteporfin and Erlotinib showed dose-dependent inhibition of HSC3 proliferation (Fig. 3D–E) and Verteporfin showed more efficiency. Galunasirtib did not inhibit proliferation (Fig. 3F). We initially evaluated the combination of these inhibitors using proliferation assays with human HSC3 cells and found the additive effects of SP2509 and Verteporfin at 10:1 or 5:1 ratios showed more than 90% inhibition of HSC3 cell proliferation (Fig. 3G).

Our in vitro observations prompted us to test the combination of these inhibitors in vivo. For this, we first titrated the dose of Verteporfin for the local application onto the mouse tongue to any adverse effects. Our study showed that 2.5 to 3 mg/kg of Verteporfin was safe and did not induce any adverse effects. SP2509 25 mg/kg + 2.5 mg/kg Verteporfin in DMSO and corn oil emulsion was therefore tested in combination for in vivo studies using the 4NQO-induced model of OSCC progression (experimental outline depicted in Fig. 4A). Topical application of SP2509 alone inhibited pathologic changes, including reduced papillomatous lesions, SCC, and inflammatory cells on the tongue, as well as reduced OSCC growth, but the combination with Verteporfin showed striking additive effects compared with a single drug alone (Fig. 4B–D). Analysis of lysates collected from tissues treated with these inhibitors showed reduced expression of Lsd1 (Fig. 4E).

To evaluate the relationship between LSD1 and PD-L1, analysis of gene expression data available from The Cancer Genome Atlas (TCGA) studies showed an inverse correlation between Lsd1 and PD-L1 (Fig. 5A). Consistent with these observations, we observed an increase in the levels of Pd-l1 in SP2509 and SP2509/Verteporfin-treated tissues (Fig. 5B). Further, Lsd^−/− mouse tongues also showed increased expression of Pd-l1 compared with Lsd^WT/WT mouse tongues (Fig. 5C). Finally, immunostaining analysis also showed that LSD1 pharmacologic inhibition (Fig. 5D) of LSD1 knockout (Fig. 5E) increased the levels of PD-L1 expression. Overall, above data indicates that Pd-L1 is repressed by LSD1 and that LSD11 deletion promotes the expression of PD-L1 (Fig. 5A–E). These observations suggested the possibility that Lsd1 inhibition may promote efficacy towards anti–PD-1 and anti–PD-L1 immunotherapies.

To evaluate potential additive effects of the Lsd1 inhibitor SP2509 with immune checkpoint inhibitors, we treated mice exposed to 4NQO with SP2509 alone and in combination with anti–PD-1 and anti–PD-L1 (experimental outline is depicted in Fig. 6A). SP2509 was topically applied onto the tongue of developing tumors, and the mice were simultaneously administered anti–PD-1 or anti–PD-L1 neutralizing antibodies by intraperitoneal injection, a common administration route for these antibodies (33, 34). Monitoring of tumors showed reduced growth with SP2509, anti–PD-1, or anti–PD-L1 treatment alone and demonstrated a more effective reduction of tumor growth with combined SP2509 and anti–PD-1 treatment (Fig 6B–D). The combination of SP2509 reduced gross pathologic and papillomatous lesions, normal basal epithelial morphology and inflammatory lesions, and Lsd1 expression (Fig. 6E). Immunostaining analysis also showed that LSD1 inhibition by SP2509 promoted the presence of CD8⁺ cells, suggesting elevated T-cell infiltration. Correlation analysis of LSD1/KDM1A in TCGA datasets (Supplementary Table S1) showed that Pearson correlation of LSD1 was moderate to high with EGFR (0.42), YAP (0.48), and CTGF (0.41). Spearman's rank rho was also moderate to the high of LSD1 with EGFR (0.43), YAP (0.51), and CTGF (0.31). Thus, we proposed a model from our studies that LSD1-driven feed-forward loop in OSCC (Fig. 6G).

Loss of tongue tumors by LSD1 deletion, the most striking finding, prompted us to evaluate gene expression changes and the potential therapeutic potential of pharmacologic inhibition of LSD1. LSD1 deletion reduced expression of OSCC-related genes, including Ctgf, Yap, Wwtr1/Taz, and Egfr. LSD1 has been shown to collaborate with YAP (7), a Hippo signaling effector that promotes phenotypes associated with OSCC (8) and other cancers (36, 37). Studies have further shown that both LSD1 and YAP localize to the NuRD chromatin remodeling complex and LSD1 is recruited to the chromatin by a YAP/TAZ-TEAD complex (38, 39). However, the regulatory relationship between LSD1 and YAP is unknown. Cross-talk between LSD1 and nuclear YAP/TAZ signals may affect common downstream transcriptional targets, such as CTGF and the PD-1/PD-L1 axis. PD-L1 expression directly protects tumors from cytolytic T cell killing (40), so the expression of PD-L1 within a tumor cell or tumor-infiltrating cells explains the therapeutic effect of PD-1/PD-L1 blocking antibodies (41–44). LSD1 has been reported to contribute to cancer initiation, progression, and relapse through multiple mechanisms (45–47). Our observations suggest that LSD1 has a specific role in reprogramming the immune microenvironment during OSCC. A previous study showed that LSD1 promotes the progression of a type of medulloblastoma (48), while the ablation of LSD1 promotes antitumor immunity in combination with checkpoint blockade in melanoma (2). Both tumor- and host-derived PD-L1 can play critical roles in immunosuppression (43) and our findings indicate that LSD1 deletion in K14 positive cells in 4NQO mouse could be direct or indirect PD-L1 expression on K14 positive and other cell types, and possibly not restricted to LSD1^−/− cells.

Consistent with our study, the histone demethylase KDM4A has been shown to promote antitumor immunity and sensitivity to PD-1 antibody in OSCC growth and metastasis (49). Epigenetic therapies have primarily focused on enzymes that regulate histone acetylation and DNA methylation, with several DNA methyltransferase and histone deacetylase inhibitors now in clinical use and approved by the FDA to treat leukemias and other cancers. Histone lysine demethylases remain a relatively untapped source of potential “druggable targets” and, thus, the LSD1 signaling network may offer opportunities for exploiting immunotherapy. Indeed, we found that SP2509 and Verteporfin promote the expression of PD-L1, which could provide a more sustained effect for immunotherapy and combination therapies. Ablation of LSD1 promotes PD-L1 expression, and antitumor immunity is shown in breast cancer (2). Thus, our study validates this observation in OSCC and further identifies that LSD1 inhibition promotes PD-L1 expression. Our data also suggests that LSD1 inhibition by SP2509 promotes CD8⁺ T-cell infiltration. Further, our analyses also showed that LSD1 correlates with EGFR, YAP/TAZ, and PD-L1 expression and likely participates in a network of signals induced by these genes in specific clinical cancers (as shown in Fig. 6F).

This study has translational significance by leveraging the fact that compounds targeting LSD1 regulated genes/pathways are extensively studied and are in clinical trials. SP2509 and SP-2577 (Seclidemstat) have similar properties (27) and Seclidemstat is in clinical trials in patients with relapsed or refractory Ewing sarcoma (NCT03600649), advanced solid tumors (NCT03895684), and gynecologic cancers (NCT04611139). Verteporfin (Visudyne) disrupts YAP/TAZ-TEAD interactions and is used in combination with other drugs (28–30). It is also being tested in clinical trials for various other indications (NCT04590664, NCT03033225). Our study showed the effect and mechanism for future application in OSCC. Finally, there is no ongoing clinical trial for OSCC testing LSD1 inhibitors or their combination with YAP/TAZ-TEAD interactions as pharmacologic inhibitors or PD-1/PD-L1 checkpoint antibodies, so our work offers translational directions. Thus, our studies have identified a basic mechanism of LSD1 in OSCC. Overall, our findings could have translational importance in future clinical studies for smoking-induced oral cancer.

G.J. Hanna reports grants from Bristol-Myers Squibb outside the submitted work. No disclosures were reported by the other authors.

T. Alhousami: Conceptualization, data curation, formal analysis. M. Diny: Conceptualization, data curation, formal analysis. F. Ali: Conceptualization, data curation, formal analysis. J. Shin: Conceptualization, data curation, formal analysis. G. Kumar: Conceptualization, resources, data curation, software, formal analysis, investigation. V. Kumar: Software, formal analysis, visualization. J.D. Campbell: Software, formal analysis. V. Noonan: Data curation, writing–review and editing. G.J. Hanna: Conceptualization, writing–review and editing. G.V. Denis: Writing–review and editing. S. Monti: Funding acquisition, writing–review and editing. M.A. Kukuruzinska: Resources, writing–review and editing. X. Varelas: Funding acquisition, writing–review and editing. M.V. Bais: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

The authors acknowledge M.V. Bais was supported by NIH/NIDCR grants R21DE026892 and R03DE025274. M.A. Kukuruzinska, S. Monti, and X. Varelas were supported by NIH/NIDCR grant R01DE030350. X. Varelas was funded by NIH/NHLBI R01HL124392 and an American Cancer Society - Ellison New England Research Scholar Grant (RSG-17–138–01-CSM). The authors thank CTSA grant U54-TR001012 for their support of pilot studies.

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.