miR-31 Displays Subtype Specificity in Lung Cancer

Lung cancer remains the deadliest cancer in both men and women in the United States and the world, projected to kill more than 130,000 individuals domestically and 1.8 million worldwide (1). This lethal disease encompasses many malignancies that originate from specific cellular sources within the lung and has historically been characterized histologically. The two major histologic subdivisions of lung cancer are non–small cell lung cancer (NSCLC), approximately 85% of cases, and small-cell lung cancer (SCLC), 15% of cases. NSCLC is further divided into multiple subtypes comprising lung adenocarcinoma (40%), squamous cell carcinoma (SQCC; 30%), and large cell lung carcinoma (LCLC; 15%). SCLC falls into a category of neuroendocrine lung tumors that also includes the rare subtypes of large-cell neuroendocrine carcinoma (LCNEC), typical carcinoids (TC), and atypical carcinoids (AC; ref. 2). Each of these subtypes are distinct not only histologically, but also for their cellular origins, oncogenic driver mutations, transcriptomic profiles, and epigenetic modifications (3, 4). Clinically, NSCLC and SCLC are treated as distinct diseases and thus have different treatment approaches. Surgical resection is common for stage I–II NSCLC, but not SCLC as most SCLCs have disseminated disease at time of diagnosis. SCLC is initially responsive to chemotherapy; however, it quickly recurs and becomes recalcitrant (5, 6). Large-cell neuroendocrine carcinoma is most commonly treated with the same regimen as SCLC (6). NSCLC is frequently not clinically divided for treatment; however, different histologic subtypes have been shown to respond differently to treatment (7). While some targeted therapies have been developed for patients with lung adenocarcinoma, namely those with activating mutations in EGFR or EML-ALK rearrangements, no targeted therapies are available for patients with any of the other aforementioned subtypes of lung cancer (3, 8, 9). This is largely due to the differing mutations reported within each histologic subtype. Approximately 75% of lung adenocarcinomas have genetic alterations that promote the RTK/RAS/RAF signaling pathway (10). The most commonly altered pathways in SQCC are inactivation of CDKN2A (72%), activation of PI3K/AKT (47%), squamous cell differentiation (44%), and oxidative damage (34%; ref. 11). SCLC is frequently driven by amplification of SOX2 or MYC (12) in combination with biallelic inactivation of TP53 and RB1 (13). Each histologic subtype of lung cancer is thus defined by a distinct set of mutations, and frequently even when such mutations affect synonymous cellular pathways, the precise genetic alterations between the subtypes remain distinct. This raises the question of whether there is a set of genes or any gene that may have cross-spectrum function within lung cancer.

Noncoding genes are an emerging area of growth in our understanding of lung cancer genetics and biology (14). One class of noncoding genes that has seen significant investigation are miRNAs, which have broadly been reported as having dysregulated expression in lung cancer, with many having tumor suppressive or oncogenic properties within lung cancer such as let-7 and miR-34 or miR-21, respectively (15, 16). Even still, many of these miRNAs have been reported as having disparate tumor-suppressive and oncogenic properties within different tissues of the body (17, 18). One such tissue-specific miRNA is miR-31, previously reported as having tumor-suppressive functions in multiple tissue types yet oncogenic in others (19). While miR-21 and other miRNA have been shown to aid oncogenesis in the lung, transgenic mice overexpressing miR-31 in the lower lung epithelia develop spontaneous lung adenomas and eventual adenocarcinomas, making this miRNA the most oncogenic miRNA characterized in lung cancer to date (20). Moreover, miR-31 has been demonstrated to possess clinical relevance in human lung adenocarcinoma with increased expression in patient tumors compared with normal lung and correlation with patient survival (20, 21). The role of this unique lung oncomiR in non-adenocarcinoma lung cancer (60% of lung cancer) is less clear, especially given miR-31's pro-oncogenic mechanism in lung adenocarcinoma is promotion of RAS/MAPK signaling, a pathway that is not commonly activated in non-adenocarcinoma human lung tumors (20, 22). We have determined that the overexpression of miR-31 is not common to all lung cancer, but instead unique to specific subtypes, and it displays pleiotropic effects on primary tumor growth and metastasis concordant with histologic subtype, while uniquely altering the cellular signaling of each subtype.

Lung cell lines were purchased from ATCC (Calu-1, H520, SK-MES, H1869, H661, H1915, H82, SW-1271, Beas-2B, NL20, 293T Phoenix-AMPHO) and the Japanese Collection of Research Bioresources (JCRB) Cell Bank; LK-2, EBC-1) and cultured in recommended conditions. HCC15 and HCC95 cells were provided by Dr. David Carbone of Ohio State University (Columbus, OH) and cultured in RPMI1640 supplemented with 10% FBS. H1650, H358, and H69 cells were provided by Dr. Philip Owens (UC Denver, Denver, CO) and cultured in RPMI1640 supplemented with 10% FBS. 16HBE, A549, PC9, H1437, H2030, H1299, and H2228 cells were provided by Dr. Christine Eischen (Thomas Jefferson University, Philadelphia, PA) and cultured in RPMI1640 supplemented with 10% FBS and 1× penicillin/streptomycin. Transfection of miRNA mimic, control mimic, inhibitor, or control inhibitor was done using DharmaFECT 1 (all from GE Healthcare Dharmacon). Plasmids were transfected using Lipofectamine 3000 (Thermo Fisher Scientific) as per the manufacturer's protocol. pBabe-miR-31 was purchased from Addgene. Additional miR-31 cloning primers, hsa-miR-31 Fwd HpaI CAAGTTAACACCTCCTGTGCCTAACTACATC, hsa-miR-31 XhoI Rev CAAACTCGAGAAGTAAGGAAGGTGAGAAAGGC. For both pBabe and pLXSN retroviral production, vectors were transfected into 293T Phoenix cells and virus was collected. For pLL3.7 lentiviral production, vectors plus packaging plasmids (psPAX2 and pCL-VSVG) were transfected into HEK293-T cells and virus was collected. Lung cells were infected with retrovirus or lentivirus containing media with 8 μg/mL polybrene. pBabe and pLL3.7-puro–infected cells were selected with puromycin (2 μg/mL) for 7 days, and stable pools were used for experiments. pLXSN-infected cells were selected with G418 (400 μg/mL) for 7 days. pLL3.7-GFP–infected cells were evaluated for GFP expression and only used if >90% of the cells were GFP positive. TaqMan qRT-PCR (see below) was used to verify miR-31 expression in all cells used for these studies.

Deidentified frozen lung samples were obtained from the UAB tissue collection and banking facility and the Vanderbilt University Medical Center Lung Biorepository, which banked samples following written informed consent. All samples were from surgical resections and were evaluated by a board-certified pathologist. All tumor samples used were determined to be >50% tumor by evaluation of hematoxylin and eosin (H&E)-stained sections. H&E-stained sections of each normal sample were evaluated and determined to lack any precancerous lesions. Deidentified lung paraffin sections (4 × 10-μm per block) were obtained from UAB Pathology archive formalin-fixed, paraffin-embedded (FFPE) tissue blocks and H&E sections were evaluated by board-certified pathologist (S. Harada). Clinical/pathologic features are listed in Supplementary Tables S1–S3.

For MTT assays, cells (5,000–10,000 per well) were placed in 96-well plates; the following day, cells were transfected with 50 nmol/L miR-31 mimic, inhibitor, or a corresponding control (all from GE Healthcare Dharmacon). The 0-hour time point was measured at the time of transfection. MTT assays were performed every 24 hours after transfection for 4 days, and absorbance was measured using a BioTek Epoch plate reader. Cell viability was assessed by Trypan blue dye exclusion. For clonogenic assays, cells (stable clones) infected with retrovirus encoding miR-31, or empty vector, were placed in 6-well plates at low density (500–2,000 cells per well). After 10–20 days, cells were fixed with methanol and stained with crystal violet. Colony growth was quantified by a plate reader (BioTek Epoch) that measures color intensity per surface area. For the soft agar assays, the retrovirus-infected cells were suspended in 0.6% agarose-containing media and seeded in 6-well dishes (3,000–5,000 cells per well) coated with 0.8% agarose. Following approximately 2–4 weeks in culture, colonies were counted.

RNA from cell lines was isolated using TRIzol (Thermo Fisher Scientific) according to the manufacturer's protocol. Ten nanograms of input RNA from frozen patient tissues was isolated using the miRNeasy Kit (Qiagen) and from FFPE tissues using the RecoverAll Total Nucleic Acid Kit (Thermo Fisher Scientific). cDNA synthesis for miRNA was generated using Taqman microRNA Reverse Transcriptase Kit (Thermo Fisher Scientific). TaqMan miRNA assays (Thermo Fisher Scientific) were used to quantify cDNA generated from miRNA. RNU6B was the endogenous control in human tissues. cDNA synthesis for mRNA was performed using iScript cDNA Synthesis Kit (Bio-Rad) for cell lines and SuperScript III cDNA Synthesis Kit (Invitrogen) for patient samples. SYBR Green (Qiagen) was used to quantify cDNA generated from mRNA. Levels of β-actin were used to normalize mRNA in human tissue. Assays were performed in triplicate. Primer sequences are listed in Supplementary Table S4.

For xenograft studies, male and female 5- to 6-week-old athymic nude (H520) or B6:129-Rag2tm1Fwall:R2G2 (all other cell lines) mice (Envigo) were injected subcutaneously in the flanks with 5 × 10⁶ H520, H661, H1915, H82 cells, 3 × 10⁶ H69, or 3.5 × 10⁶ Calu-1 cells infected with empty pBabe retrovirus or pBabe encoding miR-31 (H661), empty vector pLXSN or pLXSN encoding miR-31 (H520), or empty vector pLL3.7 or pLL3.7 encoding miR-31 (H1915, H82, H69, Calu-1). Tumor volume was calculated from measurements with electronic calipers. At time of sacrifice, tumors were removed, weighed, and half frozen and half formalin-fixed for further analysis.

miRNA expression profiles for lung SQCC and normal lung tissue samples were obtained from The Cancer Genome Atlas (TCGA) data portal (https://tcga-data.nci.nih.gov/tcga/; 2017). miRNA expression normalized reads per million (RPM) values were log₂ transformed prior to evaluation.

Hypergeometric analyses were performed as reported previously (23). RNA-sequencing files from lung adenocarcinoma, squamous cell carcinoma, large-cell neuroendocrine carcinoma, and small-cell lung carcinoma tumor studies were obtained from previously published datasets (10, 11, 13, 24). A total of 477 transcripts with conserved sites for miR-31–5p were obtained from TargetScan (25) and cross-referenced against RNA-sequencing profiles for each lung cancer subtype. Transcripts present in subtypes were ranked by expression and the lowest 200 transcripts for each subtype were evaluated for gene function by using the Database for Annotation, Visualization and Integrated Discovery (DAVID version 6.8; refs. 26, 27). Data were plotted by most significant enriched biological themes.

Lysates from H520, H661, and H82 (Vector control and miR-31 overexpression) cells were analyzed on the Tyrosine (PTK) and Serine/Threonine (STK) arrays using 15 μg (PTK) or 2 μg (STK) of protein per a standard kinomic protocol. Phosphorylation data was collected over multiple computer controlled pumping cycles and exposure times (10–200 ms) for approximately 144 substrates per chip. Raw image analysis was conducted using Evolve2 and comparative analysis upstream kinase prediction was done in BioNavigator v6.2 using scoring from Kinexus (www.phosphonet.ca). After kinetic reads, postwash captures were done at 10-, 20-, 50-, 100-, and 200-ms exposures. These values were integrated into a slope, multiplied by 100, and log₂ transformed. All arrays were quality controlled for acceptable signal (and compared with prior samples). Whole Chip comparative analysis (BioNavigator Upkin PTK v 15.0 STK v 8.0) was done between groups generating Kinase Statistic (KSTAT) and Specificity scores. Heatmap generated using the Broad Institute's Morpheus (https://software.broadinstitute.org/morpheus).

Kinase lists by UnitProt ID were uploaded to Webgestalt for gene set enrichment analysis using the Reactome pathway functional database (28). miR-31 target prediction analysis was performed by uploading predicted conserved miR-31 targets from TargetScan (25) to the Reactome database, which were then sorted by Reactome pathway (29).

Advanced Cell Diagnostics (ACD) performed miRNAscope ISH with custom hsa-miR-31 probe. IHC was performed by Vanderbilt University Medical Center TPSR: anti-p63 antibody [4A4] SKU: CM163A, TTF-1 Leica PA0364, SYP Leica PA0299, CSK ProteinTech 17720–1-AP, ATM Invitrogen MA5–32063, EPHB4 Cell Signaling Technology14960

Unless otherwise stated, values represent mean ± SEM, and one-tailed Student t test was used for comparisons. P values of less than 0.05 were considered significant.

All experiments involving mice were approved by the UAB Institutional Animal Care and Use Committee (21262) and followed all state and federal rules and regulations. All human samples used in experiments were deidentified and obtained from UAB Tissue Collection and Banking Facility and Vanderbilt University Medical Center Lung Biorepository, which banked samples following written informed consent. The UAB Institutional Review Board for Human Use (IRB) under protocol N160829001 determined that the research in this manuscript is not subject to FDA Regulations and is not Human Subjects Research.

We evaluated miR-31 expression in a cohort of patient lung tumors from archived tissues, clinical/pathologic details in Supplementary Table S1. As previously demonstrated (20), miR-31 is overexpressed in patient tumors with lung adenocarcinoma; however, we also observed significantly increased expression in SQCC, adenosquamous carcinoma (ADSQ), and LCNEC compared with normal lung tissue. When expression of miR-31 is compared with the mean of the normal lung tissue samples, 30 of 45 (67%) patients with lung adenocarcinoma, 27/32 (84%) patients with SQCC, 10/12 patients with (83%) ADSQ, and 10/16 (63%) LCNEC patient samples had increased miR-31 expression. If miR-31 expression is instead compared with the median of the normal lung tissue samples, these percentages increase further to 78%, 94%, 92%, 81%, respectively. Neither SCLC nor atypical carcinoids had increased miR-31 expression (Fig. 1A and B; Supplementary Figs. S1–S3). In addition to evaluating patient samples for miR-31 expression, a panel of human lung cancer cell lines was also evaluated. miR-31 was found overexpressed in the majority of lung adenocarcinoma (6/9), SQCC (6/8), LCLC cell lines (3/4), and (0/3) of the SCLC cell lines tested, reflecting what was observed in the patient tissues (Fig. 1C). These results were validated in another cohort of frozen lung tissue, which revealed miR-31 levels were significantly elevated in lung adenocarcinoma (12/15, 80% compared with mean normal) and SQCC (23/25, 92% compared with mean normal; Fig. 2A), as well as in data from TCGA (Supplementary Fig. S4A). We next evaluated another cohort of frozen matched adjacent uninvolved normal and tumor samples of SQCC, ADSQ, LCNEC, and carcinoids (typical and atypical). miR-31 was overexpressed in the tumors of SQCC, ADSQ, and LCNEC compared with nontumor adjacent tissue, but not in lung carcinoids (Fig. 2B). miR-31 expression was also evaluated across the cancer stages for each subtype. Whereas miR-31 has previously been reported to significantly increase expression in more advanced stage lung adenocarcinoma, similar trends were not observed in any of the other subtypes examined here (Supplementary Fig. S4B; ref. 20). High miR-31 expression has previously been shown to correlate with poor patient survival in both lung adenocarcinoma and SQCC TCGA data (20, 30). We thus examined our own patient data for survival correlation with miR-31 expression. Given the small size of the patient cohorts and the recency of the majority of the cases, we were not able to obtain statistically significant survival data; however, we did observe trends toward decreased survival with high miR-31 expression in all subtypes examined (Supplementary Fig. S4C and S4D).

To test the functional role of miR-31 in non-adenocarcinoma lung cancer, we inhibited and overexpressed miR-31 in multiple human cell lines representing each histologic subtype and performed several assays to evaluate in vitro growth including MTT assays, clonogenic assays, soft agar growth assays, and cell-cycle analysis. Interestingly, no differences in growth nor viability were detected when miR-31 levels were manipulated in any of the non-adenocarcinoma lung cell lines tested (Supplementary Fig. S5A–S5D). Given that not all the hallmarks of cancer can be evaluated in vitro and that miR-31 has been shown to spontaneously transform cells in vivo in lung adenocarcinoma (20), we performed subcutaneous xenograft assays to determine whether miR-31 affected in vivo tumor growth. Remarkably, we observed increased tumor volume and tumor weight with overexpression of miR-31 in SQCC cell lines Calu-1 and H520 along with similar trends in SQCC cell lines LK-2 and SK-MES (Fig. 3A and B; Supplementary Fig. S6A and S6B). The effect of miR-31 expression in LK-2 cells was partially obscured by sex differences in tumor volume, where males developed significantly larger tumors than females (Supplementary Fig. S6C). This was independent of miR-31 expression, and miR-31 promoted tumor growth in both sexes. This sex-dependent effect on tumor size was not observed in any other cell line. The observation of increased tumor size upon overexpression of miR-31 was not apparent in LCLC nor SCLC cell lines (Fig. 3C–F). Of note, however, in the H661 (LCLC) subcutaneous xenograft experiment, despite no statistical difference in tumor volume, we did observe a trend toward increased tumor weight, as well as a slight increase in size of visual appearance of the tumors (Fig. 3C).

While no differences in primary tumor size were measured in either LCLC cell line (H661, H1915), we did observe differences in spontaneous metastasis to the lung between control and miR-31–overexpressing cell lines. In H661 subcutaneous injected mice, we observed an increased incidence of spontaneous lung metastases in the miR-31 overexpression group (3/6) versus control (1/6; Fig. 4A). In addition to increased metastatic incidence, tumors overexpressing miR-31 yielded significantly more lung metastases in H661 cells, and a similar trend was observed in H1915 cells (Fig. 4A; Supplementary Fig. S6D). While both of these cell lines metastasized spontaneously to the lung, H82 (SCLC) cells readily metastasized to the liver, where we observed a statistically significant increase in spontaneous metastases from miR-31–overexpressing tumors (Fig. 4B). The H69 SCLC cell line tumors did not spontaneously metastasize. In SQCC, we observed a statistically significant increase in spontaneous lung metastases from Calu-1 miR-31 overexpression compared with vector control (Fig. 4C). We also observed an increased incidence of metastases in the miR-31 group (6/6) compared with control (3/6). H520 cells, which readily form subcutaneous tumors in athymic nude mice, did not spontaneously metastasize in athymic nude mice; therefore, we performed another H520 cell xenograft in R2G2 mice where these cells spontaneously metastasize to the liver. We did not observe a difference in spontaneous metastasis in this model, although these results may have been limited by the number of mice on experiment (Supplementary Fig. S6E). LK-2 cells spontaneously metastasize to both the lung and the liver from the flank but at extremely low incidence. When combined, we noted an increase in total metastatic incidence (2/6) control versus (3/5) when miR-31 was overexpressed. Overexpression of miR-31 before and after injection is shown in Supplementary Figs. S7A–S7F and S8A–SF, respectively.

We analyzed publicly available RNA-sequencing data for the most downregulated predicted targets of miR-31, finding that these genes clustered to different biological processes in each subtype, leading us to hypothesize that with varying expression of its targets within each subtype, miR-31 might affect disparate signaling cascades leading to the observed phenotypic differences (Fig. 5A). miR-31 has been reported to promote RAS/MAPK signaling in lung adenocarcinoma, therefore, we focused our attention on the additional subtypes of lung cancer (20, 22). To determine the effects of miR-31 on cell signaling for each subtype, we performed whole kinome tyrosine and serine/threonine analysis. We observed drastically different global alterations in kinase signaling among the subtypes (Fig. 5B). In H520 SQCC, miR-31 overexpression resulted in changes including increased activity of EPH-B1, -B4, -A8, BMX, TYK2, JAK2, and decreased activity of RAF1, with networks centering around JAK, Ephrin-A, and RAF1. When gene-set enrichment analysis (GSEA) was performed through secondary software, similar results were found with the top upregulated pathways again involved in EPH-Ephrin signaling. Interestingly, despite RAF1 being one of the kinases with most decreased activity, we still observed increases in total MAPK/ERK signaling (Fig. 5C and D; Supplementary Fig. S9A and S9B). The most downregulated pathways determined by GSEA in H520 cells upon overexpression of miR-31 were regulation of TP53 (Fig. 5C). In H661 LCLC, global downregulation of the cellular kinome was observed upon overexpression of miR-31 (Fig. 5B). miR-31 overexpression resulted in decreases in activity of FAK2, Met, CSK, PDGFR-alpha and beta, generating Pyk2 centric networks (Fig. 5E and F; Supplementary Fig. S10A and S10B). GSEA was performed on these kinases and we observed that the most upregulated signaling pathways involved cellular response to stress, senescence, and β-catenin–independent WNT signaling, while those pathways that showed downregulation were IGF1R, PI3K, and MAPK signaling. In H82 SCLC, miR-31 overexpression resulted in changes including increased activity of EPH -B2, -B1 and -A8, IKK-alpha and -beta, JAK2 and -3 and ATM with JAK and Ephrin-A centric networks (Fig. 5G and H; Supplementary Fig. S11A and S11B). Upon GSEA, we observe the most upregulated pathways following overexpression of miR-31 in SCLC to be G₂–M DNA damage checkpoint and cell-cycle checkpoints with the most downregulated pathways being constitutive signaling by aberrant PI3K in cancer and FCGR activation.

To validate the kinome data, we selected several kinases with increased and decreased levels of activity (Supplementary Figs. S9–S11) and performed IHC for these kinases on subcutaneous tumors for each subtype (Fig. 6). As indicated by the kinome data where we observed increased activity of EPHB4, we also observed increased protein expression of EPHB4 in the tumors from SQCC cell lines H520 and Calu-1–overexpressing miR-31 (Fig. 6A). In LCLC, CSK was identified as having decreased activity. IHC for CSK in LCLC cell lines H661 and H1915 revealed decreased expression of CSK in tumors overexpressing miR-31 (Fig. 6B). In SCLC, ATM was identified as a kinase with significantly increased activity resulting from miR-31 overexpression, which was also observed by IHC in the SCLC cell line H69 (Fig. 6C). H82 tumors were frozen at the time of collection and thus unable to be used for IHC analysis.

To determine the direct targets of miR-31, which might result in the signaling pathways altered, we input the list of conserved miR-31 targets from TargetScan into the Reactome Database and sorted the predicted targets by the pathways identified in the kinome data. The top upregulated pathways detected as altered in the kinome data from miR-31 overexpression in H520 SQCC cells were EPH-Ephrin–mediated repulsion of cells, EPH-Ephrin Signaling, EPHB-mediated forward signaling, and Ephrin signaling (Fig. 5C). We identified six conserved miR-31 targets associated with these pathways, as well as two additional nonconserved targets (Fig. 6D). The top upregulated pathways detected in the kinome data from miR-31 overexpression in H661 LCLC cells were cellular response to stress, oxidative stress–induced senescence, and cellular responses to external stimuli (Fig. 5E). We identified 18 conserved miR-31 targets associated with these pathways (Fig. 6E). The top upregulated pathways detected in the kinome data from miR-31 overexpression in H82 SCLC cells were G₂–M DNA damage checkpoint and cell-cycle checkpoints (Fig. 5G). We identified 8 conserved miR-31 targets associated with these pathways (Fig. 6F). Given that we observed increased EPH-Ephrin signaling in SQCC cell lines following overexpression of miR-31 and miRNAs typically act by suppressing the expression of their target genes, we chose to focus on targets involved in the negative regulation of EPH-Ephrin signaling. Surprisingly little is known about the negative regulation of this pathway, but two models have been proposed: endocytosis of the EPH:Ephrins and proteolytic cleavage of the EPH:Ephrins from the membrane (31). We therefore focused on target genes that might be involved in either of these mechanisms of EPH-Ephrin downregulation and identified AP2B1 and VAV3 as proteins reported to be involved in EPH-Ephrin endocytosis and ADAM10 and NCSTN as proteins with reported roles in EPH-Ephrin proteolytic cleavage (32–35). NCSTN is a component of the gamma-secretase complex, which has been reported to proteolytically cleave Ephrin-B (35). qRT-PCR of these genes in SQCC cell lines confirmed decreased expression upon overexpression of miR-31 (Fig. 6D). We hypothesize that downregulation of these negative regulators of EPH–Ephrin signaling may lead to EPH-Ephrin retention at the cell membrane and thus result in the increased EPH expression and signaling we observed.

Given that we see increased metastasis in LCLC cell lines following overexpression of miR-31, we chose to focus on miR-31 predicted targets within the cellular response to stress, oxidative stress induced senescence, and cellular responses to external stimuli pathways that might be tumor or metastasis suppressors. We identified E2F2, HIF1AN (FIH-1), EGLN1, EGLN3, BAP1, and UBN1 as potentially tumor or metastasis suppressors by literature search (36–40). qRT-PCR of these genes showed decreased expression following miR-31 overexpression in LCLC cell lines (Fig. 6E). Similarly, in SCLC cell lines, we observed increased metastasis following miR-31 overexpression, and thus focused on gene targets that have either been proposed to act as tumor or metastasis suppressors or have ambiguous roles in these processes. We identified TAOK1, CDK1, and YWHAE as genes related to these categories (41–43). qRT-PCR of these genes showed decreased expression following overexpression of miR-31 in SCLC cell lines (Fig. 6F). This data supports the hypothesis that downregulation of these direct target genes may lead to the global signaling changes observed and ultimately the progrowth and prometastatic phenotypes of miR-31 in these histologic subtypes of lung cancer.

To determine whether any of the direct targets proposed might be relevant clinically, we performed qRT-PCR for these targets in a cohort of frozen lung patient tissue samples. We observed significantly lower expression of AP2B1, ADAM10, and NCSTN in SQCC tumors compared with normal lung tissue (Fig. 7A–D). We did not observe a statistically significant difference in VAV3 expression (Fig. 7C). This indicated to us that not only do these targets decrease in expression following overexpression of miR-31 in vitro, these targets also appear to decrease in patients with overexpression of miR-31. These results are in line with previously published studies that have reported increased expression of EphB1 and EphB4 in NSCLC patient tissue compared with control lung, higher expression of EphB1 correlating to poorer patient survival, and EphB4 expression associated with lymph node metastasis (44–47). While we were unable to examine the proposed direct miR-31 targets in patients with LCNEC due to limited sample availability, it is important to note that the global signaling pathways identified in our study have also previously been reported in LCLC. We identified cellular response to stress and oxidative stress induced senescence as two of the most upregulated pathways following overexpression of miR-31 in LCLC, and increased oxidative stress has been shown to associate with more aggressive LCLC (48, 49). SCLC in particular has been shown to upregulate genes involved in the DNA damage response (DDR) such as CHEK1 and has been shown to be susceptible to ATR and CHK1 inhibitors, while lung adenocarcinoma is resistant to these inhibitors (50). Most notably we observed increased kinase activity of ATM, ATR, CHK1, and CHK2 following overexpression of miR-31 in our SCLC cell lines (Supplementary Fig. S11).

miRNAs have been an attractive area of interest as drug targets for a number of years, and there are a multitude of miRNA mimics and inhibitors now in phase I and II clinical trials. Some of which have promising early results, such as inhibition of miR-155 in cutaneous T-cell lymphoma and miR-16–based mimics in malignant pleural mesothelioma (51, 52). These promising results have highlighted the need to understand the function of additional miRNA and what tissues and diseases they may act as therapeutic targets in. The function of miR-31 has been shown to be tissue specific in a multitude of cancer types, previously reported as tumor suppressive in glioblastoma, prostate, ovarian, liver, and bladder cancer while oncogenic in colorectal, pancreatic, and lung adenocarcinoma (53–59). Despite these tissue specific roles, its function within subtypes of the same cancer has not been studied. miR-31 was previously reported to have increased expression in patient lung adenocarcinoma tumors and human lung adenocarcinoma cell lines when compared with adjacent noninvolved lung and nontransformed human lung cells, respectively (20, 21). Lung adenocarcinoma only comprises 40% of lung cancers, however, and the role of miR-31 across the additional subtypes has not been well characterized. Initial observations have suggested that in addition to its role in lung adenocarcinoma, miR-31 may both correlate with survival in lung SQCC (30) and be elevated in the three patient tumor samples analyzed (21), indicating that this miRNA is overexpressed in two of the most common histologic subtypes of lung cancer (30). Despite these findings, expression of miR-31 has not been fully investigated outside of lung adenocarcinoma for its oncogenic properties across the histologic spectrum of lung cancer. Here, we demonstrated that not only is miR-31 overexpressed in lung adenocarcinoma, it is overexpressed in patient SQCC, ADSQ, and LCNEC compared with normal lung tissue, yet not in SCLC nor lung carcinoids, prompting us to further study its function in each of these subtypes.

In lung adenocarcinoma, miR-31 has been shown to both spontaneously transform mouse epithelial lung cells and promote human adenocarcinoma tumor growth (20). Here, we determined that miR-31 also promotes human lung squamous cell carcinoma tumor growth, but not large-cell nor small-cell lung carcinoma tumor growth. Furthermore, miR-31 has been reported as elevated in advanced stage lung adenocarcinoma, its expression able to predict the presence of lymph node metastases, and to promote in vitro transwell migration in H23 lung adenocarcinoma cells (22). There has been no reported role in metastasis, however, for miR-31 in any other histologic subtype; therefore, we carefully examined all xenograft mice for the presence of spontaneous metastases and demonstrated that miR-31 promotes spontaneous metastasis in SQCC, LCLC, and SCLC. This is the first report to demonstrate disparate functionality of a miRNA in cancer types of the same tissue, as here we show that miR-31 can promote tumor growth in SQCC but not LCLC nor SCLC and yet can promote metastasis in LCLC and SCLC.

Broadly speaking, changes in cell signaling networks are what ultimately lead to cancer cell growth and metastasis. Each histologic subtype of lung cancer is characterized by the unique activation of various cell signaling pathways (60). While miR-31 overexpression has been reported to enhance oncogenic RAS signaling in lung adenocarcinoma, its contribution to the cellular signaling of non-adenocarcinoma lung cancer has not been established (20, 22). We determined that miR-31 regulates disparate cellular signaling pathways in each histologic subtype, and these pathways correlate with some of the most frequently altered pathways in each subtype. In SQCC, we found that miR-31 most strongly upregulates EPH-Ephrin and PI3K/MAPK signaling while downregulating TP53. Activation of PI3K/AKT signaling is one the commonly altered pathways in SQCC, activated in 47% of patients (11). EPH-Ephrin signaling has not been well studied in lung cancer; however, in one study, Eph mutations were reported in as many as 20% of patients with NSCLC (61). Notably, EphA2 has been reported to have increased expression and promote invasion and survival specifically in lung SQCC (62, 63) as well as promote acquired resistance to EGFR TKIs (64). Our kinome data indicated increased kinase activity specifically of EphB1, EphA8, EphB4, EphB2, EphB3, and EphA7 upon overexpression of miR-31 in H520 SQCC cells. EphB1, EphB3, and EphB4 have all previously been reported to have increased expression in lung cancer, as well as promote lung cancer cell proliferation and migration (44, 46, 65). Most notably, overexpression of EphB3 promoted both in vitro cell migration and in vivo tumor growth specifically in H520 SQCC cells (65). Thus, we conclude that miR-31 most likely promotes primary tumor growth and metastasis in SQCC by a combination of activation of EPH-Ephrin signaling, PI3K/MAPK signaling, and suppression of TP53 regulation. In H661 LCLC, the most upregulated signaling pathways upon overexpression of miR-31 involved cellular response to stress and oxidative stress induced senescence, while those pathways that showed downregulation were IGF1R, PI3K, and MAPK signaling. This is in direct contrast to the pathways activated in SQCC and what we previously observed in lung adenocarcinoma LUAD; however, these are pathways that have specifically been reported as associated with increased aggressiveness in LCLC (48, 49). In SCLC, we observed miR-31 overexpression resulted in JAK and Ephrin-A centric networks. The upregulation of Ephrin-A signaling observed here mirrors what we see in SQCC; however, our secondary GSEA did not indicate Ephrin-A signaling in SCLC to the extent it did in SQCC. Instead, upon GSEA, we observe the most upregulated pathways following overexpression of miR-31 in SCLC to be G₂–M DNA damage checkpoint and cell-cycle checkpoints with the most downregulated pathways being constitutive signaling by aberrant PI3K in cancer and FCGR activation. Given that nearly 100% of SCLC cases show biallelic loss of TP53 and RB1 and SCLC specifically has been shown to upregulate genes involved in DDR, it is interesting that the most upregulated pathways in SCLC following alteration of miR-31 are those involved in cell-cycle checkpoints and DNA repair. While there was some overlap observed between each subtype tested, miR-31 clearly altered specific signaling pathways in a subtype dependent manner.

In summary, we have determined that miR-31, a miRNA with reported pleiotropic function across cancer tissue types, not only displays tissue-specific function but functions disparately within cancer types of the same tissue. miR-31 is overexpressed in patient tumors of lung adenocarcinoma, SQCC, and LCNEC. miR-31 promotes primary tumor growth in lung adenocarcinoma and SQCC, while promoting metastasis in LCLC and SCLC. miR-31 alters distinct cellular signaling pathways within each histologic subtype that likely cause the phenotypic differences. There are few genes that display cross-spectrum function in human lung cancer. Although we suggest the mechanism(s) of oncogenesis for miR-31 vary greatly across histologic subtypes, it is now clear this gene may have broad clinical importance and biological function that might be actionable therapeutically in lung malignancy.

M.L. Davenport reports grants from NIH and grants from NIH during the conduct of the study. J.B. Echols reports grants from NIH during the conduct of the study. C. Yates reports grants from NCI and grants from NIMHD during the conduct of the study, personal fees from Riptide Biosciences, and personal fees from QED Therapuetics outside the submitted work; in addition. C. Yates served as a consultant for Riptide Biosciences and QED Therapeutics and owns stock in Riptide Biosciences. M.D. Edmonds reports grants from NIH (UL1TR003096) and grants from NIH (U54CA118948) during the conduct of the study; grants from Breast Cancer Research Foundation of Alabama and grants from METAvivor outside the submitted work. No disclosures were reported by the other authors.

M.L. Davenport: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. J.B. Echols: Formal analysis, investigation, writing–original draft. A.D. Silva: Investigation, writing–original draft. J.C. Anderson: Data curation, software, investigation, writing–original draft. P. Owens: Conceptualization, writing–original draft. C. Yates: Conceptualization, resources, writing–original draft, writing–review and editing. Q. Wei: Resources, writing–original draft. S. Harada: Resources, investigation, visualization, writing–original draft, writing–review and editing. D.R. Hurst: Conceptualization, formal analysis, writing–original draft, writing–review and editing. M.D. Edmonds: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.

We thank Vulcan informatics for data mining assistance and Dr. George Netto (UAB Pathology) for access to patient samples. This work was supported by T32 Predoctoral Fellowship NIH 5T32HL134640–02 (to M.L. Davenport), T32 Predoctoral Fellowship NIH 5T32GM008111–32 (to M.L. Davenport), NIH UL1TR003096 (to M.D. Edmonds), and NIH U54CA118948 (to M.D. Edmonds, C. Yates).

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.