A Novel Therapeutic Strategy Targeting the Mesenchymal Phenotype of Malignant Pleural Mesothelioma by Suppressing LSD1

This article is featured in Highlights of This Issue, p. 1

Malignant pleural mesothelioma (MPM) is an aggressive cancer with an extremely poor prognosis (1, 2) that currently has no curative treatments (3). Although the majority of patients with MPM are treated with standard chemotherapy, the tumor is largely unresponsive and less than 40% of patients experience any clinical benefits, with a median overall survival (OS) of just 12 months after diagnosis (4). Recently developed methods targeting genetic alterations have demonstrated dramatic clinical responses in patients with non–small cell lung cancer (5, 6); however, this approach has failed to demonstrate efficacy against MPM in clinical trials (7). This is largely due to the fact that common oncogenes are rarely mutated in MPM, whereas large-scale molecular studies have revealed frequent chromosomal abnormalities, copy-number loss, and gene mutations in tumor suppressor genes (TSG) such as BAP1, CDKN2A, CDKN2B, NF2, and TP53 (8, 9). Since TSGs are challenging to target (10), a different approach is required to identify novel therapeutic targets for MPM. Importantly, patients with MPM harbor mutations in histone modifiers, such as SETD2, SETDB1, SETD5, ASH1L, PRDM2, KDM2B, and KMT2D. Thus, epigenetic regulation, particularly histone methylation, could be a novel therapeutic target for patients with MPM.

Although patients with MPM are known to respond poorly to chemotherapy, the underlying reasons remain unclear. MPM is histologically classified into three subtypes: epithelioid, sarcomatoid, and biphasic (11), and studies have shown that patients with sarcomatoid MPM have a significantly shorter OS than patients with other subtypes and are less likely to benefit from systemic chemotherapy (12, 13). Large-scale molecular studies have revealed that MPM displays a high degree of intratumoral molecular diversity at both genetic and epigenetic levels (9, 14–16). Moreover, these studies identified molecular clusters characterized by different clinical outcomes that are only partly related to the histologic subtype. Interestingly, the clusters with high epithelial–mesenchymal transition (EMT)–related gene expression were associated with a significantly poorer prognosis, despite being histologically classified as epithelioid. Therefore, mesenchymal subsets may contribute toward a poor clinical outcome.

EMT is one of the most well known forms of cellular plasticity that is likely to contribute toward tumor heterogeneity (17) and increasing evidence has demonstrated the role of EMT in drug resistance (18). Previously, we showed that epigenetic modifications, particularly histone modifications, are involved in the molecular mechanisms underlying EMT (19–21). Therefore, mesenchymal subsets that involve epigenetic regulation should be considered as therapeutic targets to improve the outcome of patients with MPM.

Lysine-specific demethylase 1 (LSD1/KDM1) is a flavin adenine dinucleotide-dependent histone demethylase that removes histone H3 mono- and dimethylation at lysines 4 and 9 (22, 23). LSD1 overexpression has been documented in various types of cancer and has been correlated with E-cadherin inhibition associated with enhanced cell migration and invasion (24, 25). In addition, LSD1 is known to facilitate EMT and tumor progression by repressing epithelial marker expression (26) and has been reported to be essential in Snail-1-and Slug-mediated EMT (27, 28). Since LSD1 has many roles in the epigenetic regulation of EMT in breast, prostate, and lung cancer (25, 26, 29), it is considered a potential therapeutic target for preventing EMT. Although LSD1 is ubiquitously expressed in MPM, it remains unclear whether it plays an important role in EMT and could be a target for treating patients with MPM.

In this study, we show that suppressing LSD1 attenuates the mesenchymal phenotype and sensitizes MPM cells to cisplatin (CDDP) by orchestrating a positive feedback loop involving milk fat globulin protein E8 (MFGE8) and Snail, which regulates EMT and CDDP-induced apoptosis via the FAK–AKT–GSK3β pathway. Therefore, targeting the mesenchymal phenotype using LSD1 inhibitors in combination with CDDP should be considered as a novel therapeutic strategy for patients with MPM.

The mesothelioma cell lines ACC-MESO-1, ACC-MESO-4, Y-MESO-8A, Y-MESO-9, Y-MESO-14, Y-MESO-25, Y-MESO-29, and Y-MESO-72 were established from Japanese patients at the Aichi Cancer Center Research Institute and were identified using short tandem repeat analysis or a single-nucleotide polymorphism array. 293T human embryonic kidney cells, NCI-H28, MSTO-211H, NCI-H226, and NCI-H2452 cell lines were obtained from the ATCC. All cells were maintained in RPMI-1640 medium (Wako) supplemented with 10% FBS, penicillin (100 U/mL), and streptomycin (100 μg/mL) at 37°C in 5% CO₂. All cell lines were periodically tested for Mycoplasma contamination using a MycoAlert Mycoplasma Detection Kit (Cambrex), with negative results. Recombinant human MFGE8 and human MFG-E8 Quantikine ELISA kits were purchased from R&D Systems.

Lentiviral short hairpin RNA (shRNA) constructs were obtained from Sigma-Aldrich with the following target sequences: sh non-target (shNT), 5′-CCG GGC GCG ATA GCG CTA ATA ATT TCT CGA GAA ATT ATT AGC GCT ATC GCG CTT TTT-3′; shLSD1 #1, 5′-CCG GGC TAC ATC TTA CCT TAG TCA TCT CGA GAT GAC TAA GGT AAG ATG TAG CTT TTT G–3′; shLSD1 #2, 5′-CCG GGC TAC ATC TTA CCT TAG TCA TCT CGA GAT GAC TAA GGT AAG ATG TAG CTT TTT G–3′.

The human Snail-1 ORF with six serine to alanine mutations (30) was fused with a modified estrogen receptor (ER) and cloned into the pBABE-puro retroviral expression vector. The pLV-MFGE8 plasmid was purchased from VectorBuilder. Conditioned medium containing infective lentiviral or retroviral particles was generated by cotransfecting 293T human embryonic kidney cells (2 × 10⁶ cells) with 3 μg of lentiviral vector, 1 μg pHCMV-VSV-G, and 3 μg of pCMV ∂8.91 or Amphotropic using Lipofectamine Ltx (Invitrogen Corporation) according to the manufacturer's instructions. Supernatants were collected 48 hours after transfection and filtered using a 0.45 μmol/L membrane (Millipore. Cells were infected directly using 8 μg/mL polybrene.

Scratch wound-healing assays were performed in 6-well tissue culture plates. Briefly, confluent monolayers were disrupted by standardized wound scratching using a sterile 1,000-μL pipette tip, washed twice with medium, and allowed to grow for a further 18 to 24 hours. The cells were then observed using a Nikon Eclipse TS100 inverted microscope (Nikon) and photographed at 40× magnification. The wound gap was measured using ImageJ analysis software (NIH) and the results were calculated as follows:

Total RNA was extracted from each cell line using an RNeasy Mini kit (Qiagen) according to the manufacturer's instructions, and then 300 ng was reverse-transcribed into cDNA using a Revertra cDNA synthesis kit (Toyobo). qPCR was performed using Fast SYBR Green Master Mix (Applied Biosystems) with the following cycling conditions: denaturation at 95°C for 20 seconds, 40 cycles of amplification (denaturation at 95°C for 30 seconds, annealing and extension at 60°C for 30 seconds), and melting-curve analysis. qPCR was performed in triplicate with β-actin as an internal control using the following primer sequences: MFGE8, forward 5′-TCT GTG CGT GTG ACC TTC TTG-3′ and reverse 5′-CCA GGG GTT ATC GTC ATT GCT-3′; β-actin forward 5′-CTC TTC CAG CCT TCC TTC CT-3′ and reverse 5′-AGC ACT GTG TTG GCG TAC AG-3′.

Cell lysates were prepared using a lysis buffer (50 mmol/L Tris-HCl, pH 6.8, 2% SDS, 10% glycerol) supplemented with protease inhibitors (Roche) and a phosphatase inhibitor cocktail (Thermo Fisher Scientific). Protein concentration was measured using a DC Protein Assay Kit (Bio-Rad Laboratories). Lysates were solubilized in sample buffer solution with reducing reagent (6×) for SDS-PAGE (Nacalai Tesque). Equal amounts of whole-cell lysate were fractionated on a 4% to 20% polyacrylamide gradient-SDS gel, transferred to polyvinylidene difluoride membranes, and blocked in TBST and Tween (TBST; 20; 25 mmol/L Tris, pH 7.4, 136 mmol/L NaCl, 5 mmol/L KCl, and 0.1% Tween) containing 5% milk or 5% BSA. The blots were incubated with primary antibodies against LSD1 (1:1,000, catalog no. 2139, Cell Signaling Technology), GAPDH (1:3,000, catalog no. 016–25523), E-cadherin (1:1,000, catalog no. 610181, BD Biosciences), Zo-1 (1:1,000, catalog no. 5406, Cell Signaling Technology), fibronectin (1:1,000, catalog no. F3648, Sigma Chemicals), vimentin (1:1000, catalog no. 550513, BD Biosciences), phospho-AKT (1:1000, catalog no. 4060S, Cell Signaling Technology), AKT (1:1,000, catalog no. 4685, Cell Signaling Technology), phospho-FAK (Tyr397, 1:1,000, catalog no. 8556, Cell Signaling Technology), FAK (1:1,000, catalog no. 71433, Cell Signaling Technology), GSK-3-α/β (1:1,000, catalog no. 5676, Cell Signaling Technology), phospho–GSK-3-α/β (Ser21/9, 1:1,000, catalog no. 9331, Cell Signaling Technology), Histone H3 mono methyl K4 (1:500, catalog no. ab8895, Abcam plc), Histone H3 di methyl K4 (1:1,000, catalog no. ab7766, Abcam plc), Histone H3 tri methyl K4 (1:500, catalog no. ab8580, Abcam plc), Histone H3 (1:1,000, catalog no. ab8580, Abcam plc), Snail (1:1,000, catalog no. 3879, Cell Signaling Technology), integrin αV (1:1000, catalog no. 4711, Cell Signaling Technology), integrin β3, integrin β5 (1:1,000, catalog no. 13166, Cell Signaling Technology), MFGE-8 (1:1,000, catalog no. HPA002807, Novus Biologicals), and caspase-3 (1:200, catalog no. sc 7148, Santa Cruz Biotechnologies). After overnight incubation, the membranes were washed with TBST, incubated with horseradish peroxidase (HRP)–conjugated secondary antibodies (1:3000, GE Healthcare), and washed with TBST and Clarity Western ECL Substrate (Bio-Rad Laboratories). Images were analyzed using a lumino imager (LAS-4000 mini; Fuji Film).

Paraffin-fixed surgical samples were cut into 3 mm–thick sections, deparaffinized in xylene, and rehydrated using an ethanol solution. Epitope retrieval was carried out in an autoclave at 120°C for 15 minutes, endogenous peroxidase was blocked with 0.3% H₂O₂ for 30 minutes, and the sections were blocked with normal horse serum for 30 minutes. After incubation at 4°C overnight with rabbit anti-human MFGE8 polyclonal antibodies, the sections were washed, incubated at 4°C for 2 hours with HRP-linked secondary antibodies, and developed in liquid 3,3′-diaminobenzidine before counterstaining with hematoxylin.

Cells were seeded (2.0 × 10³) in 96-well microtiter plates with or without various concentrations of chemotherapeutic agents, including CDDP (Sigma-Aldrich), RN-1 (EMD-Millipore), and S2101 (EMD-Millipore), after shLSD1 or shNT infection or treatment with 4-hydroxy-tamoxifen (4-OHT), if applicable. After 72 hours, 10 mL of WST-8 (cell counting kit-8, Dojindo Laboratories) was added to each well, and the optical density was measured at 450 nm using a microplate reader (Bio-Rad Laboratories) 4 hours later. Results are expressed as a percentage of cell viability.

Cells infected with shLSD1 or shNT were treated for 48 hours with CDDP or vehicle control and then cells in the supernatant or adhering to the plates were collected, washed with PBS, pooled, and suspended in binding buffer. The cells were then stained using an Annexin V Apoptosis Detection Kit from the MEBCYTO Apoptosis Kit (MBL) according to the manufacturer's instructions. Flow cytometry analysis was performed using BD LSRFortessa and BD FACSDiva software (BD Biosciences). The percentage of apoptotic cells was determined by adding early-stage apoptotic cells [right lower quadrant; Annexin positive and propidium iodide (PI) negative] and late-stage apoptotic cells (right upper quadrant; Annexin positive and PI positive).

Cells were treated with CDDP for 48 hours following infection with shLSD1 and shNT or treatment with 4-OHT, if applicable. After the media had been removed, the cells were rinsed with ice-cold PBS and lysed with lysis buffer from a Human Phospho-Kinase Array (R&D Systems). Protein concentration was measured using a DC Protein Assay Kit (Bio-Rad Laboratories). Equal amounts of protein (200 μg/membrane) were incubated with antibody arrays overnight and then incubated with anti–phospho-tyrosine-HRP antibodies for 2 hours at 37°C. ECL (GE Healthcare) was used to detect the anti–phospho-kinase-HRP antibodies. Array spot signals were acquired using a lumino imager (LAS-4000 mini; Fuji Film) and analyzed using ImageJ.

RNA was isolated using an RNeasy Mini kit (Qiagen), and biotinylated cDNA was prepared from 100 ng total RNA using GeneChip WT PLUS Reagent Kit (Thermo Fisher Scientific) following the manufacturer's instructions. Following fragmentation, 2 μg of single-stranded cDNA was hybridized for 16 hours at 45°C on GeneChips Clariom S assay chips (Thermo Fisher Scientific). The arrays were washed and stained in the GeneChip Fluidics Station 450 (Thermo Fisher Scientific). Clariom S array was scanned using GeneChip Scanner 3000 7G (Thermo Fisher Scientific). Data files were generated and processed using the Affymetrix software and analyzed using the Transcriptome Analysis Console Software v.3.0 (Thermo Fisher Scientific). The threshold set for up- and downregulated genes was fold change > 2 and FDR corrected P < 0.05. Microarray data have been deposited in the NCBI Gene Expression Omnibus (GEO) database under accession code GSE 182154.

Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) was performed by Active Motif, Inc. The paired-end 42-bp sequencing reads (PE42) generated by Illumina sequencing (using NextSeq 500) were mapped to the genome using the BWA algorithm with default settings. Alignment information for each read was stored in the BAM format. Only the reads that passed Illumina's purity filter, aligned with no more than 2 mismatches, and mapped uniquely to the genome were used in the subsequent analysis. Genomic regions with high levels of transposition/tagging events were determined using the MACS2 peak calling algorithm (31). Since both the reads (tags) from paired-end sequencing represent transposition events, they were used for peak calling but treated as single independent reads. Peaks were identified using the MACS2 algorithm at a cut-offpoint of P value 1e-7. The DESeq2 algorithm (32) was used to identify the top differential peaks between the two groups. The program normalizes the counts between the samples using the “median or ratios” method and calculates log₂ fold change (Log₂FC), shrunken-Log₂FC, P value, P_adj (FDR) for each peak. Peaks of gain and loss were defined by P_adj, FDR < 0.1 and Log₂FC > 1/< (−1). Peaks within 10,000 bp upstream or downstream of a gene were counted as being associated with that gene. ATAC-seq data have been deposited in the NCBI GEO database under accession code GSE 182154.

Between June 2005 and June 2018, blood and surgical samples were collected from 32 patients with histologically confirmed MPM and 12 patients who had undergone surgery at Juntendo University Hospital, respectively. This study was approved by the Juntendo University Research Ethics Committee. Written informed consent was obtained from all patients enrolled in this study. Serum aliquots were immediately obtained from the blood samples and stored at −80°C until further analysis. Serum MFGE8 levels were measured using a commercial sandwich ELISA kit (R&D Systems). Surgical samples were fixed in 10% PBS and embedded for histologic analysis.

MFGE8 mRNA expression and clinical data were obtained from publicly available whole-exome sequencing data in The Cancer Genome Atlas (TCGA) via cBioPortal. mRNA expression data were analyzed using unpaired t tests, OS was calculated using the Kaplan–Meier method, and the log–rank test was used to assess the association between MFGE8 expression and OS using GraphPad Prism 6.03.

All statistical analyses were performed using GraphPad Prism 6.03 unless otherwise indicated. All data represent the mean ±SD of biological replicates. Mean values were compared using unpaired, two-tailed Student t tests or one-way ANOVA. P values of < 0.05 were considered statistically significant.

In this study, the histologic subclassification of mesothelioma cell lines was based on previous reports (33) and cell morphologies on 2D cultures. The sarcomatoid or epithelioid is morphologically defined by spindle-like or polygonal forms of cells, respectively. First, we measured the EMT-related gene expression of 12 MPM cell lines by Western blotting for EMT markers. Interestingly, some epithelioid and sarcomatoid MPM cell lines displayed hybrid epithelial and mesenchymal protein expression (Fig. 1A); therefore, we treated epithelioid MPM cells with TGFβ1 to investigate whether they retained reversibility of the EMT program. Unexpectedly, we observed no differences in morphology or protein expression (Supplementary Fig. S1A and S1B).

Since the majority of sarcomatoid MPM cell lines displayed strong Snail expression (Fig. 1A), we investigated whether Snail induced a mesenchymal phenotype in epithelioid MPM cell lines. A Snail-1 variant (Snail-1^6SA) was generated by substituting 6 targeted amino acids to confer constitutive activity and a fused ER response element to mediate exogenous 4-OHT regulation. Expression of Snail-1^6SA induced a dramatic change in the morphology of the epithelioid lines, ACC-MESO-4 and Y-MESO-25, with the cobblestone-like epithelial cells adopting an elongated fibroblast-like morphology (Fig. 1B; Supplementary Fig. S2A).

To confirm the EMT phenotype, we measured the expression of characteristic EMT molecular markers by Western blotting (Fig. 1C; Supplementary Fig. S2B), finding that the epithelial marker E-cadherin was significantly downregulated while the mesenchymal markers fibronectin and vimentin were upregulated. Wound-healing assays demonstrated that cellular migration was also dramatically enhanced in these cells (Fig. 1D; Supplementary Fig. S2C), which displayed significantly higher IC₅₀ values for CDDP than the control cells following 4-OHT exposure for 72 hours (Fig. 1E; Supplementary Fig. S2D). Together, these results indicate that Snail overexpression initiates EMT and renders them resistant to CDDP in MPM cells.

Snail-1 regulates the transcriptional repression of E-cadherin and other epithelial markers by binding to E-box consensus sequences, thereby recruiting chromatin modifiers such as LSD1, HDAC1, and HDAC2, and components of the Polycomb-2 complex (19). Since LSD1 has been reported as an essential effector of the Snail-1–dependent transcriptional repression of epithelial genes (27, 28), we investigated the expression of LSD1 in MPM cell lines. Western blotting revealed that LSD1 was highly expressed in all MPM cell lines (Fig. 2A); therefore, we evaluated whether LSD1 inhibition suppressed the mesenchymal phenotype induced by Snail-1 by treating ACC-MESO-4 ER-Snail-1^6SA cells with 4-OHT in the presence or absence of the LSD1 inhibitors, RN-1 and S2101. Interestingly, both RN-1 and S2101 abrogated the dramatic change in morphology (Fig. 2B). RN-1 attenuated Snail-1^6SA induced downregulation of epithelial markers and upregulation of mesenchymal markers (Fig. 2C); it also reduced the enhanced cellular migration induced by Snail-1^6SA (Fig. 2D). Therefore, LSD1 inhibition appears to suppress the initiation of EMT by Snail-1.

To investigate whether LSD1 contributed toward the mesenchymal phenotype of sarcomatoid MPM cells, we knocked down endogenous LSD1 using lentivirus-driven shRNA constructs (shLSD1). Inhibiting LSD1 expression using shLSD1 (#1 and #2) caused MPM cells to shift from an elongated fibroblast-like morphology to a cobblestone-like epithelial morphology (Fig. 3A; Supplementary Fig. S3A). Moreover, suppressing LSD1 increased E-cadherin expression and reduced fibronectin and vimentin expression (Fig. 3B and C, Supplementary Fig. S3B), whilst significantly reducing cell migration in the wound-healing assay (Fig. 3D; Supplementary Fig. S3C). Thus, suppressing LSD1 appears to induce an epithelial phenotype in sarcomatoid MPM cells.

To investigate whether attenuating the mesenchymal phenotype affected sensitivity to chemotherapy, we performed a chemosensitivity assay in ACC-MESO-1 and Y-MESO-8A cells infected with shLSD1 or shNT. Interestingly, suppressing LSD1 significantly reduced the IC₅₀ for CDDP compared with the shNT cells (Fig. 3E, Supplementary Fig. S3D). Furthermore, we performed a chemosensitivity assay in the ACC-MESO-1 cells treated with an LSD1 inhibitor, S2101, in combination with CDDP, showing synergistic drug effects (Fig. 3F). Western blot analysis following CDDP treatment showed elevated cleaved caspase-3 expression in the LSD1 knockdown cells compared with controls (Fig. 3G; Supplementary Fig. S3E). This was also confirmed by Annexin V staining following CDDP treatment; the LSD1 knockdown cells treated with CDDP showed a higher fraction of positive cells compared with the control cells. (Fig. 3H; Supplementary Fig. S3F). Furthermore, LSD1 inhibitor significantly reduced the IC₅₀ for cisplatin in ACC-MEOS-4 cells compared with those without LSD1 inhibitor (Supplementary Fig. S3G). Together, these findings suggest that LSD1 suppression sensitizes MPM cells to CDDP-induced apoptosis.

To elucidate the potential mechanism underlying CDDP sensitization following LSD1 suppression, we investigated histone methylation status using Western blotting. Surprisingly, LSD1 suppression did not affect global H3K4 nor H3K9 mono-, bi-, and trimethylation, despite its role in H3K4 and H3K9 methylation, suggesting that the mechanism of CDDP sensitization is limited to more specific targets (Fig. 4A).

Therefore, we measured the phosphorylation of multiple kinases in MPM cells using a human phospho-kinase array, with a focus on kinases that were inactivated in the epithelial phenotype and activated in the mesenchymal phenotype. Interestingly, the FAK were inactivated by knockdown of LSD1 in ACC-MESO-1 cells; Snail-1^6SA led to the activation of FAK in ACC-MESO-4 cells (Fig. 4B and C), suggesting that the FAK is likely to be involved in the mechanism of CDDP sensitization. The activation of FAK and downstream effectors was confirmed using Western blotting, which revealed that FAK and AKT phosphorylation were downregulated in ACC-MESO-1 and Y-MESO-8A cells infected with shLSD1 #1 following CDDP treatment but upregulated in ACC-MESO-4 ER-Snail-1^6SA cells exposed to 4-OHT following CDDP treatment (Fig. 4D; Supplementary Fig. S4A and S4B). Therefore, the FAK-AKT pathway may be involved in CDDP-induced apoptosis.

To investigate the potential mechanisms underlying FAK pathway activation, we performed transcriptome analysis using two different shRNA constructs and selected genes with statistically significant changes in expression (P < 0.05) representing more than two-fold up- or downregulation. We identified 335 downregulated and 244 upregulated transcripts between ACC-MESO-1 control and shLSD1-treated cells that were shared by both shLSD1#1 and shLSD1#2 (Supplementary Fig. S4C; Supplementary Table. S1).

To determine which of these transcripts were directly regulated by LSD1-mediated changes in histone methylation, we performed genome-wide ATAC-seq to assess global chromatin accessibility following knockdown of LSD1 by shLSD1 #1. LSD1 suppression dramatically altered overall chromatin accessibility in ACC-MESO-1 cells, increasing accessibility at 129 loci and decreasing accessibility at 346 loci (Supplementary Fig. S4C; Supplementary Table. S2). Overlapping the transcriptional and chromatin changes induced by LSD1 suppression indicated 13 upregulated and 10 downregulated genes as possible direct LSD1 targets (Fig. 4E; Table 1). We decided to focus on MFGE8, a secreted glycoprotein that can bind to cells expressing αVβ3 and αVβ5 integrin via its RGD motif and enhance the FAK pathway (34, 35). The chromatin region in the gene locus of MFGE8 is hardly accessible following LSD1 suppression (Fig. 4F). LSD1-mediated MFGE8 regulation in ACC-MESO-1 and Y-MESO-8A cells infected with shLSD1 was validated using mRNA quantification (Fig. 4G; Supplementary Fig. S4D). These findings indicate that LSD1 directly regulates MFGE8 expression, leading to the activation of the FAK-AKT pathway.

To assess the role of MFGE8 in MPM, we first investigated the expression of MFGE8, αVβ3, and αVβ5 integrins in MPM cell lines (Fig. 5A). The majority of MPM cell lines expressed αV integrin subunits; however, MFGE8 and integrin β3 expression, but not integrin β5 expression, tended to be higher in sarcomatoid MPM cell lines. We confirmed these results using RNA sequencing (RNA-seq) data from the TCGA (Supplementary Fig. S5A), finding that MFGE8 and integrin β3 expression were significantly higher in sarcomatoid/biphasic MPM than in epithelioid MPM, whereas integrin β5 expression did not differ.

Next, we investigated MFGE8 expression in human samples. Although there was no statistical difference in serum MFGE8 expression between sarcomatoid/biphasic and epithelioid MPM, MFGE8 was detected in the serum samples of all patients with MPM (Supplementary Fig. S5B). Subsequent IHC analysis revealed that sarcomatoid/biphasic MPM samples displayed more intense MFGE8 staining than epithelioid MPM (Fig. 5B). Therefore, we analyzed the relationship between MFGE8 expression and OS using the Kaplan–Meier method and log–rank test, finding that patients in the TCGA MPM cohort with high MFGE8 expression showed a significantly shorter OS than those with low MFGE8 expression (P = 0.0314; Supplementary Fig. S5C), indicating that MFGE8 could play an important role in MPM.

Since MFGE8 is required for Snail expression and is known to stimulate the acquisition of mesenchymal properties (36, 37), we investigated the effect of MFGE8 on the mesenchymal phenotype in MPM by overexpressing MFGE8 in ACC-MESO-1 and ACC-MESO-4 cells using pLV-MFGE8 and downregulating MFGE8 in ACC-MESO-1 and Y-MESO-8A cells using shLSD1. As expected, overexpression of MFGE8 reduced E-cadherin expression and increased vimentin expression in ACC-MESO-4 (Supplementary Fig. S5D). Moreover, Snail was upregulated by MFGE8 overexpression and downregulated by LSD1 knockdown (Fig. 5C and D; Supplementary Fig. S5E). Interestingly, Snail expression decreased alongside the downregulation of phospho-GSK3β, which is downstream of AKT and negatively regulates the stability and nuclear localization of Snail protein (30). Moreover, the restoration of MFGE8 using rhMFGE8 abolished phospho-GSK3β and Snail downregulation in ACC-MESO-1 cells infected with shLSD1, indicating that LSD1 regulates Snail expression via MFGE8 (Fig. 5E), which was also upregulated by Snail overexpression (Supplementary Fig. S5F) Furthermore, integrin β3 expression was reduced by suppressing LSD1 (Fig. 5D; Supplementary Fig. S5E) but enhanced by Snail overexpression (Supplementary Fig. S5G). Together, these results indicate that a positive feedback loop exists between MFGE8, Snail, and integrin β3 that is regulated by FAK-AKT-GSK3β signaling.

To investigate the mechanism underlying FAK pathway inactivation and apoptosis induced by LSD1 suppression, we blocked MFGE8-integrin binding using the RGD peptide in ACC-MESO-4 ER-Snail-1^6SA cells treated with 4-OHT. The RGD peptide downregulated phospho-FAK and AKT expression and enhanced cleaved caspase-3 expression following CDDP treatment (Fig. 6A), whereas pretreatment with hrMFG-E8 rescued the phospho-FAK and AKT downregulation caused by LSD1 suppression and abolished cleaved caspase-3 induction following CDDP treatment (Fig. 6B; Supplementary Fig. S6A). Moreover, LSD1 suppression failed to inactivate the FAK pathway and induce cleaved caspase-3 in ACC-MESO-1 and Y-MESO-8A cells overexpressing MFGE8 following CDDP treatment (Fig. 6C; Supplementary Fig. S6B). Therefore, LSD1 suppression appears to sensitize MPM cells to CDDP by downregulating MFGE8 to inactivate the FAK-AKT pathway.

The limitations in identifying new targets for genetic alterations in treating MPM have been well established; therefore, there is a critical need to develop new targets for alternative approaches that can be effective. We hypothesized that epigenetic regulation could contribute to the poor response to chemotherapy; and therefore, could be a novel therapeutic target. In this study, we demonstrated LSD1 inhibition in combination with CDDP as a novel therapeutic strategy for treating MPM.

Studies have demonstrated that MPM often displays a hybrid epithelial and mesenchymal phenotype with a combination of epithelial and mesenchymal morphologic and molecular features (11). In this study, we confirmed that all MPM cell lines expressed mesenchymal markers and that some also had hybrid epithelial/mesenchymal marker expression. In addition, we demonstrated the reversibility of the EMT program in epithelioid MPM cells which adopted a mesenchymal phenotype when induced by Snail, and in sarcomatoid MPM cells which switched to an epithelial phenotype when LSD1 was suppressed. To the best of our knowledge, this is the first study to show that both epithelioid and sarcomatoid MPM cells retain the reversibility of the EMT program.

Having found that suppressing LSD1 attenuated the mesenchymal phenotype and sensitized MPM cells to CDDP-induced apoptosis, we investigated the histone methylation status of MPM cells to understand how LSD1 regulates their mesenchymal phenotype and apoptosis. Although LSD1 is responsible for H3K4 and H3K9 methylation, no differences in their global methylation status were observed following LSD1 suppression, suggesting that their regulation is limited to more specific targets. Histone modifiers regulate several transcriptional pathways and are promising therapeutic targets (20, 21). However, clinical success has been limited, because the specific regulatory mechanism remains to be elucidated and it is complex involving several genetic pathways (38). Therefore, we tried identifying the LSD1 targets; however, identifying the direct targets of LSD1 proved challenging due to demethylation of both active (H3K4) and repressive (H3K9) chromatin marks. To this end, we utilized ATAC-seq to investigate genome-wide changes in chromatin accessibility following LSD1 suppression and overlapped these data with kinase phosphorylation profiles, finding that MFGE8 was the most likely to be a direct target of LSD1.

MFGE8 plays important roles in tumor growth, invasion, apoptosis, angiogenesis, and EMT (36, 39–41), and is expressed in several cancers; however, this is the first study to show that MFGE8 protein is expressed in serum and tumor tissue samples from patients with MPM. Interestingly, MFGE8 staining was more intense in sarcomatoid/biphasic MPM than in epithelioid type MPM, while MFGE8 mRNA expression was also significantly higher in sarcomatoid/biphasic MPM than epithelioid MPM in the TCGA cohort. Furthermore, the patients with MPM from the TCGA cohort with high MFGE8 expression displayed a significantly shorter OS than those with low MFGE8 expression. Considering that high MFGE8 expression is a poor prognosis marker in several cancers (42, 43), MFGE8 expression in IHC should be used as a biomarker for poor prognosis in patients with MPM. In addition, we observed no statistically significant differences in the MFGE8 expression between serum samples from sarcomatoid/biphasic and epithelioid MPM. This might be because of the limited sample number or the ubiquitous pattern of expression of the protein (44). Additional large-scale studies are required to confirm whether MFGE8 expression in serum is a possible biomarker for poor prognosis. MFGE8 inhibition is a new potential therapeutic strategy for patients with melanoma, bladder, ovarian, and triple-negative breast cancer (36, 40, 45). However, considering that it is difficult to target the secretory protein, it is reasonable to target LSD1 that regulates the MFGE8 expression.

Previously, we showed that pretreatment with an LSD1/2 inhibitor prevented the downregulation of epithelial marker genes and the upregulation of mesenchymal marker genes mediated by Snail-1 (19); however, it remains unclear how LSD1 regulates EMT-related gene expression. To elucidate the underlying mechanism, we utilized the fact that Snail protein stability and nuclear localization are regulated by GSK3β, a critical downstream element of the AKT pathway (30). As expected, LSD1 suppression decreased Snail expression and inactivated phospho-GSK3β, while MFGE8 restoration reversed these effects, indicating that LSD1 regulates Snail expression via MFGE8. Although MFGE8 is known to be required for Snail expression and induce EMT in melanoma (36), we revealed that MFGE8 regulates Snail expression via the FAK-AKT-GSK3β pathway and that LSD1 regulates mesenchymal gene expression by MFGE8 inducing Snail.

MPM is generally resistant to chemotherapy and recent large-scale molecular studies have shown that mesenchymal subsets are highly associated with poor prognosis in epithelioid, sarcomatoid, and biphasic MPM (9, 14–16). Since the majority of patients with MPM have mesenchymal marker gene expression, the mesenchymal component may explain the poor response to chemotherapy of all histologic MPM subtypes. Thus, it is reasonable to suppose that combining targeted therapies against mesenchymal properties with current chemotherapeutics in MPM could improve treatment efficacy. Indeed, we demonstrated that attenuating the mesenchymal phenotype sensitized MPM cells to CDDP-induced apoptosis via the FAK-AKT pathway, which is a potential therapeutic target for MPM (46, 47). However, a phase II clinical trial targeting FAK (ClinicalTrials.gov NCT01870609) was terminated due to a lack of efficacy, suggesting that there may be a narrow therapeutic window to directly target FAK. In this study, LSD1 directly regulated MFGE8 expression, which has been reported to regulate apoptosis via the integrin β3/FAK/AKT signaling pathway and induce resistance to drug-induced apoptosis (35, 48). In addition, restoring MFGE8 abolished CDDP sensitization, indicating that LSD1 regulates both the mesenchymal phenotype and CDDP-induced apoptosis via MFGE8 expression. Moreover, we identified that LSD1 orchestrates a positive feedback loop involving Snail, MFGE8, and integrin β3 that drives the mesenchymal phenotype and prevents apoptosis via the FAK-AKT pathway in MPM (Fig. 6D).

This study has some limitations. First, we showed that LSD1 suppression leads to sensitization of MPM cells to CDDP-induced apoptosis in vitro. However, it was difficult to establish a corresponding in vivo model of the MPM cells in which LSD1 suppression attenuates the mesenchymal phenotype. Second, there were more than 20 candidate LSD1 targets besides MFGE8 obtained by overlapping the transcriptional and chromatin changes following LSD1 suppression. However, in this study, we focused only on MFGE8 because it is involved in the FAK pathway. Considering that histone modifiers regulate multiple genes, other candidate genes may be involved in the mesenchymal phenotype and apoptosis. Finally, we identified the mechanism of LSD1 regulation of MFGE8 expression. However, further studies are required to elucidate the mechanism underlying MFGE8 regulation, because we did not observe a direct correlation between the expression levels of LSD1 and MFGE8.

In conclusion, targeting epigenetic regulation, including targeting histone modifiers, is emerging as a promising strategy for the treatment of cancer. However, its applications are limited; there have been impressive results in hematologic malignancies, but no substantial results in solid tumors (38). Therefore, there remain some significant challenges in the development of epigenetic therapy. We demonstrated that suppressing LSD1 attenuates the mesenchymal phenotype and sensitizes MPM cells to CDDP-induced apoptosis by orchestrating a positive feedback loop involving Snail, MFGE8, and integrin β3 regulation. These findings suggest that patients with MPM could be treated using LSD1 inhibitors, which are already being developed rapidly and are in clinical trials for acute myeloid leukemia/myelodysplastic syndrome (NCT02717884, NCT02273102). Our study revealed the mechanisms of LSD1 regulation of the mesenchymal phenotype and apoptosis in MPM, which shed light on epigenome-targeted therapy as a promising novel strategy for solid tumors. We strongly believe that targeting the mesenchymal phenotype using LSD1 inhibitors should be considered as a novel combination therapy with CDDP in patients with MPM.

K. Tajima reports grants from JSPS KAKENHI, and Institute for Environmental & Gender-Specific Medicine, Juntendo University during the conduct of the study; grants from Novartis, and Eli Lilly and Company, Japan outside the submitted work. F. Takahashi reports grants from AstraZeneca, Nippon Boehringer Ingelheim Co. Ltd., MSD, Novartis; and grants from Eli Lilly and Company, Japan outside the submitted work. T. Asao reports personal fees from AstraZeneca, Chugai Pharmaceutical Co. Ltd., Eli Lilly and Company, Japan, MSD, Nippon Boehringer Ingelheim Co. Ltd., Ono Pharmaceutical Co. Ltd., Taiho Pharmaceutical Co. Ltd.; and personal fees from Takeda Pharmaceutical Co. Ltd., outside the submitted work. R. Ko reports grants and personal fees from Nippon Boehringer Ingelheim Co. Ltd.; personal fees from AstraZeneca, Taiho Pharmaceutical Co. Ltd., Ono Pharmaceutical Co. Ltd., Chugai Pharmaceutical Co. Ltd., Novartis; and personal fees from MSD outside the submitted work. Y. Sekido reports grants from Eisai Co. Ltd., outside the submitted work. K. Takahashi reports grants from Astellas Pharma Inc., Bayer Yakuhin, Ltd., Chugai Pharmaceutical Co. Ltd., MSD K.K, Nippon Boehringer Ingelheim Co. Ltd., Novartis Pharma K.K, Ono Pharmaceutical Co. Ltd., Pfizer Inc., grants from Taiho Pharmaceutical Co. Ltd., and grants from Eli Lilly and Company, Japan K.K outside the submitted work. No disclosures were reported by the other authors.

A. Wirawan: Data curation, formal analysis, validation, investigation, methodology. K. Tajima: Conceptualization, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. F. Takahashi: Supervision. Y. Mitsuishi: Data curation, formal analysis, supervision, investigation. W. Winardi: Formal analysis, validation, investigation. M. Hidayat: Validation. D. Hayakawa: Validation. N. Matsumoto: Data curation, formal analysis, validation. K. Izumi: Data curation, formal analysis, validation. T. Asao: Resources. R. Ko: Resources. N. Shimada: Resources, formal analysis, validation. K. Takamochi: Resources. K. Suzuki: Resources. M. Abe: Resources. O. Hino: Supervision. Y. Sekido: Supervision, writing–review and editing. K. Takahashi: Resources, supervision, writing–review and editing.

This work was supported by JSPS KAKENHI grant number 16K09586 (to K. Tajima), a grant for cross-disciplinary collaboration, and a grant from the Institute for Environmental & Gender-Specific Medicine, Juntendo University (to K. Tajima). We would like to thank Dr. S. Maheswaran (MGH Cancer Center) for the helpful discussions and critical reading of this manuscript. We would like to thank Editage for editing and reviewing this manuscript for the English language.

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.