Introduction

Lung cancer is the most frequently diagnosed malignancy and the leading cause of cancer death worldwide [1, 2]. Platinum-based chemotherapy remains a common first-line treatment for non-small cell lung cancer (NSCLC) patients without treatable oncogenic alterations [3, 4]. Unfortunately, many patients show acquired chemoresistance and tumor relapse [5]. Cancer cells harboring a stem-like phenotype, referred to as cancer stem cells (CSCs), have been characterized as a subpopulation with a tumor-initiating ability, self-renewal, metastatic dissemination, and innate multidrug resistance [6]. The mechanism of innate drug resistance in CSCs has not yet been fully elucidated. Hedgehog, Notch, and Wnt/β-catenin signaling pathways that contribute to the maintenance of the CSC phenotype are also involved in chemoresistance, as well as in epithelial-to-mesenchymal transition (EMT) [7].

EMT is a process of cellular transformation from epithelial phenotypes to mesenchymal attributes. After the first report by Mani et al. [8] that cancer cells undergoing EMT could generate stem-like properties, EMT has been tightly linked to the induction of the CSC phenotype, chemoresistance, and metastasis for the past decade [9]. In general, higher invasiveness, stemness, and drug resistance of cancer cells are commonly associated with a mesenchymal state rather than with an epithelial state [10]. However, this concept of EMT-directed cancer cell behaviors has been challenged recently. First, as cancer cells transform from an epithelial state to a mesenchymal state, they may become resistant to some drugs and sensitive to others [11, 12]. Second, inhibition of some EMT inducers, such as Prrx1, enhances distal metastasis and CSC properties [13]. Third, breast cancer cells in a tumor or circulation exhibit dynamic EMT and show distinct drug responses and redox metabolism [14,15,16]. These diverse studies suggest that EMT may not always couple with metastasis, stemness, and chemoresistance during disease progression. A recent study showed that the Wnt3a-induced CSC phenotype could be achieved only when the cancer cells were at an initially mesenchymal state (low E-cadherin) [17], suggesting that the response of cancer cells to environmental stimuli might depend on their native EMT status. However, the cytotoxic effect of chemotherapy agents on CSCs in an epithelial state or a mesenchymal state is unclear.

In this study, we found that CSCs in an epithelial state were more resistant to conventional chemotherapy agents than CSCs in a mesenchymal state. Among 13 genes that were selectively upregulated in the epithelial-type CSCs, the MyoD family inhibitor domain-containing (MDFIC; also known as human I-mfa domain-containing, HIC) protein contributed to the chemoresistance and self-renewal of epithelial-type CSCs. The MDFIC p32 isoform promoted β-catenin stabilization, nuclear translocation, and transcriptional activity, leading to an increase in chemoresistance. In addition, in a tissue microarray consisting of tumor specimens from 125 NSCLC patients, high MDFIC protein expression was associated with poor disease-free survival and overall survival of the patients.

Results

The E-cadHCD133H subpopulation has higher proliferation and sphere formation abilities

To study cellular physiology of lung CSCs at different EMT statuses, we first sorted the E-cadherin-high (E-cadH) and E-cadherin-low (E-cadL) subpopulations from PC14 and A549 cell lines, which contain heterogeneous cancer cells with different EMT statuses (Supplementary Figs. S1 and S2). Next, the CD133-high (CD133H) and CD133-low (CD133L) cells were isolated from the E-cadH and E-cadL subpopulations of PC14 and A549 cell lines (Supplementary Fig. S3a and Fig. 1a). The E-cadHCD133H subpopulations had a lower proliferation rate and higher sphere formation ability than the E-cadLCD133H subpopulations (Supplementary Fig. S3b and Fig. 1b). Morphologically, the E-cadHCD133H subpopulations generated compact spheroids, while the E-cadLCD133H subpopulations formed aggregate-like spheroids in the suspension cultures (Fig. 1c). To evaluate the tumor initiation ability of the four subpopulations, we subcutaneously transplanted 100 cells of each subpopulation into nude mice. Only the E-cadHCD133H and E-cadLCD133H subpopulations were capable of generating tumors compared with the E-cadHCD133L and E-cadLCD133L subpopulations, p = 0.043 by Fisher’s exact test (Fig. 1d). However, the tumor initiation ability of CD133H cells was not changed by their native EMT status.

Fig. 1: Characterization of the epithelial- and mesenchymal-type CSCs.
figure 1

a The western blot images showed the expression patterns of CD133 and EMT biomarkers. b The sphere formation efficiency of the four subpopulations. Scale bar, 100 μm. Statistical analysis showed the sphere number per 100 seeded cells. c The bright-field and fluorescent images showed the different morphology of tumor spheres. Scale bar, 50 μm. d The tumor initiation ability of the four subpopulations of the PC14 cell line was evaluated using subcutaneously xenografted nude mice (n = 8). e The heat maps showed the cell viability of the four subpopulations of PC14 and A549 cell lines treated with different concentrations of chemotherapy agents for 72 h. f The tumor-bearing mice were treated with cisplatin for 2 weeks, and the tumor size was measured. g The E-cadHCD133H or E-cadLCD133H subpopulations were injected into the tail veins of the nude mice (n = 5). Representative images showed metastatic nodules (arrows) in the lungs of mice.

Full size image

The E-cadHCD133H subpopulation is more resistant to chemotherapy agents

Next, we treated the four subpopulations with conventional chemotherapy agents at different concentrations. The results showed that E-cadHCD133H subpopulations were more resistant to the drug treatments than other subpopulations (Fig. 1e). In a subcutaneous xenograft model, intraperitoneal injection of cisplatin suppressed the growth of E-cadHCD133L- and E-cadLCD133L-derived tumors. However, the E-cadHCD133H- and E-cadLCD133H-derived tumors continued growing during cisplatin treatment (Fig. 1f). To further assess the therapeutic effect of cisplatin on the metastasis of circulating tumor cells, E-cadHCD133H or E-cadLCD133H cells were intravenously injected into the lateral tail veins of the mice followed by cisplatin treatment immediately. Although cisplatin treatment reduced lung metastasis in both groups, more tumor nodules were found in the lungs of mice that received E-cadHCD133H cell transplantation (Fig. 1g). This result suggested that more circulating E-cadHCD133H might survive and completed distal metastasis in the lung. Notably, the E-cadherin expression patterns of the metastatic tumors derived from the four subpopulations were not changed after cisplatin treatment (Supplementary Fig. S3c). Together, these data indicated that the epithelial-type CSCs displayed higher chemoresistance than the mesenchymal-type CSCs.

Drug-resistant signature of the E-cadHCD133H subpopulation

From a gene expression microarray analysis of the four subpopulations, 205 genes that were differentially expressed in the E-cadHCD133H subpopulation (Supplementary Table S1). From gene ontology of gene set enrichment analysis (GSEA), the negative regulation of canonical Wnt signaling, EMT, apoptotic process, and cell mobility-related biological processes were enriched in the E-cadLCD133H subpopulation (Fig. 2a). However, genes associated with lung cell differentiation, lung epithelium development, and cell adhesion were enriched in the E-cadHCD133H subpopulation. In addition, the commonly enriched pathways based on the hallmark gene sets as defined by GSEA were determined. In the E-cadHCD133H subpopulation, the pathways of G2M checkpoint and DNA repair were enriched, while EMT and apoptosis pathways were enriched in the E-cadLCD133H subpopulation (Fig. 2b).

Fig. 2: Identification of the drug-resistant signature of the E-cadHCD133H subpopulation.
figure 2

a Gene ontology analysis for the E-cadHCD133H subpopulation of PC14. b The pathways enriched in the E-cadHCD133H and E-cadLCD133H subpopulations were identified by GSEA. c The genes that were concurrently altered in the E-cadHCD133H subpopulation and the drug-resistant-associated microarray datasets were defined as the drug-resistant signature (DRS). d The expression patterns of the DRS in the selected datasets. e The DRS scores of the patients enrolled in the GSE31210 dataset were correlated with tumor relapse. p < 0.001 by Pearson’s Chi-square test. f The Kaplan–Meier survival curve showed that high DRS scores were associated with the poor disease-free survival of the patients in GSE31210. Hazard ratio = 4.2078 [2.5741–6.8783]. g Two-way hierarchical clustering of the 32 genes that individually correlated with the tumor relapse of the patients in GSE31210.

Full size image

After filtering the genes with three chemoresistance-associated datasets, GSE14231, GSE12791, and GSE15709, 86 genes were putatively associated with chemoresistance (Supplementary Table S2). We therefore defined the expression pattern of these genes as the drug-resistant signature (DRS) (Fig. 2c, d). In a publicly available microarray dataset, GSE31210, which consisted of 226 patients with stage I and II lung cancer, high DRS scores were correlated with tumor relapse (Fig. 2e) and poor disease-free survival of the patients (Fig. 2f, Supplementary Tables S3 and S4). Among the 86 DRS genes, 32 genes were individually correlated with the tumor relapse of the patients (Supplementary Table S5). Indeed, the two-way hierarchical clustering of the 32 genes apparently classified the tumor recurrence of the patients (Fig. 2g).

MDFIC contributes to the drug resistance of the E-cadHCD133H subpopulation

The Cox univariate regression analysis showed that 13 out of the 32 tumor relapse-correlated genes potentially predicted the prognosis of the patients (Fig. 3a). MDFIC was at top of the ranked list of the ten genes that were significantly and selectively upregulated in the E-cadHCD133H subpopulation (Fig. 3b). Upregulation of MDFIC mRNA and protein levels in the E-cadHCD133H subpopulations of PC14 and A549 was confirmed by western blot and quantitative real-time PCR analyses (Fig. 3c, d). Protein translation of MDFIC can be initiated from an upstream GUG or an in-frame downstream AUG start codon, leading to the production of p40 and p32 isoforms, respectively [18]. The MDFIC-p40 isoform was mainly located in the nucleus, while the MDFIC p32 isoform was distributed in the cytoplasm (Fig. 3e). Ectopic expression of MDFIC-p32-myc in the E-cadHCD133L subpopulations, which expressed relatively lower levels of endogenous MDFIC and were sensitive to the chemotherapy agents, resulted in drug resistance to conventional chemotherapy agents compared with that of the vector-only control cells (Fig. 3f). However, overexpression of MDFIC-p40-myc did not alter the drug response of the cells to the drug treatments (Supplementary Fig. S4a). In contrast, knockdown of MDFIC sensitized the E-cadHCD133H subpopulations to the chemotherapy agents (Fig. 3g). In a subcutaneous xenograft mouse model, cisplatin treatment significantly inhibited the growth of MDFIC-knockdown tumors compared with that of scramble control tumors (Fig. 3h). In addition to the regulation of drug response, knockdown of MDFIC induced EMT but decreased sphere formation of the E-cadHCD133H subpopulation, while overexpression of MDFIC-p32-myc reversed EMT and increased sphere formation of the E-cadHCD133L subpopulation (Supplementary Fig. S5). These results suggested that the MDFIC p32 isoform contributed to drug resistance and self-renewal of epithelial-type CSCs, and it might also function as an EMT suppressor.

Fig. 3: MDFIC promotes chemoresistance of the E-cadHCD133H subpopulations of PC14 and A549 cell lines.
figure 3

a Cox’s univariate regression analysis of indicated genes was performed for the disease-free survival of the patients in GSE31210. b The fold changes of indicated genes between the E-cadHCD133H and the E-cadLCD133H subpopulations. Western blot (c) and qPCR (d) confirmed that MDFIC was selectively upregulated in the E-cadHCD133H subpopulations of PC14 and A549. e Subcellular localization of MDFIC p32 and p40 isoforms in the E-cadHCD133L subpopulation of PC14. Scale bar, 10 μm. f Ectopic expression of the MDFIC p32 isoform significantly induced the chemoresistance of the E-cadHCD133L subpopulations of PC14 and A549. g Knockdown of MDFIC by two clones of shRNAs sensitized the E-cadHCD133H subpopulations of PC14 and A549. h The mice bearing the tumors of MDFIC-knockdown E-cadHCD133H subpopulation of PC14 were treated with cisplatin for 2 weeks, and then, the tumor size was measured.

Full size image

High MDFIC expression in tumors is associated with poor prognosis of non-small cell lung cancer patients

Next, we evaluated the association of MDFIC with the prognosis of lung cancer patients in a tissue microarray comprising the clinical specimens of 125 NSCLC patients. The expression of MDFIC protein was evaluated by immunohistochemistry (IHC) using MDFIC antibody. The MDFIC staining signal was scored on a scale from 0 to 3 (Fig. 4a). The patients at the late stage had relatively higher MDFIC expression levels compared with the patients at the early stage (Fig. 4b). In addition, high MDFIC expression was significantly correlated with poor disease-free and overall survival of the patients (Fig. 4c, d and Table 1). Multivariate Cox regression analysis indicated that high MDFIC expression might be a good predictor for the poor outcome of the patients with NSCLC (Table 2). Similar results were also observed in a TCGA database, which consists of 740 lung cancer patients. High MDFIC mRNA levels were associated with poor disease-free and overall survival of the patients (Fig. 4e, f; Supplementary Table S6). Together, these data indicated that upregulation of MDFIC in tumors was correlated with poor prognosis of patients with NSCLC.

Fig. 4: High MDFIC expression correlates with poor overall and disease-free survival of NSCLC patients.
figure 4

a MDFIC expression was examined by IHC in the tissue microarray. Representative images illustrated the IHC scores of MDFIC expression. Scale bar, 100 μm. b Quantification of MDFIC expression by IHC of the NSCLC specimens. The number (n) of samples for each stage is indicated at the top of each column. Kaplan–Meier survival analysis of the disease-free survival (c) and overall (d) survival of the 125 NSCLC patients in the tissue microarray stratified by MDFIC protein levels. The Kaplan–Meier survival analysis showed an association of MDFIC mRNA expression and disease-free survival (e) and overall (f) survival of the 740 NSCLC patients in TCGA database.

Full size image
Table 1 The clinical characteristics of the 125 NSCLC patients with low and high MDFIC expression in the tissue microarray.
Full size table
Table 2 The multivariate Cox regression analysis of TNM stage factors and MDFIC expression for the disease-free survival and overall survival of 125 patients in the tissue microarray.
Full size table

The MDFIC p32 isoform interacts with the Axin-GSK-3−β-catenin complex and stabilizes β-catenin

To study the molecular mechanism of MDFIC-mediated drug resistance, the potential binding proteins of MDFIC were predicted by using STRING [19, 20]. The protein network suggested that MDFIC might interact with the proteins involved in the Hippo and Wnt signaling pathways, including β-catenin (CTNNB1) (Fig. 5a and Supplementary Table S7). Immunoprecipitation assays showed that MDFIC p32-GFP, but not MDFIC p40-GFP, interacted with β-catenin, GSK-3, and Axin (Supplementary Fig. S4b, c). In parallel, immunoprecipitation of the endogenous MDFIC in the E-cadHCD133H cells by using anti-MDFIC or anti-β-catenin antibody confirmed the interaction of MDFIC with the Axin/GSK-3/β-catenin complex (Fig. 5b, c). The ectopic expression of MDFIC p32-GFP increased the total β-catenin but decreased the phosphorylated β-catenin at serine 33/37 (S33/37) and threonine 41 (T41) residues, which are all involved in β-catenin degradation (Fig. 5d). Treatment with MG132, a specific proteasome inhibitor, increased the β-catenin level, which was similar to the effect of MDFIC p32 overexpression on β-catenin (Fig. 5e). These results indicated that cytosolic MDFIC might interact with the Axin/GSK-3/β-catenin complex and stabilize β-catenin.

Fig. 5: MDFIC interacts with the Axin/GSK-3/β-catenin complex and stabilizes β-catenin via inhibiting β-catenin phosphorylation at Ser33/37.
figure 5

a MDFIC-interacting proteins were predicted by STRING. b Immunoprecipitation of MDFIC was performed by using anti-MDFIC antibody in the cell lysate of E-cadHCD133H subpopulation of PC14 and A549. c Immunoprecipitation of β-catenin was performed by using anti-β-catenin antibody in the cell lysate of E-cadHCD133H subpopulation of PC14 and A549. d The pLEX-GFP or MDFIC-p32-GFP plasmid was ectopically expressed in E-cadHCD133L subpopulation of PC14 and A549, and the expression of GFP, total β-catenin, and phosphorylated β-catenin at Ser33, Ser37, and Thr41 (S33/37/T41) was examined by western blots. e The pLEX-MCS vector or MDFIC-p32-GFP plasmid was ectopically expressed in E-cadHCD133L subpopulation of PC14 and A549, and the cells were treated with vehicle or MG132 (10 μg/ml) for 6 h. The expression of β-catenin was examined by western blot.

Full size image

The MDFIC p32 isoform promotes the nuclear translocation of β-catenin, leading to drug resistance

To study whether MDFIC-mediated β-catenin stabilization can functionally contribute to drug resistance, we first analyzed nuclear localization of β-catenin in the four subpopulations. A high level of nuclear β-catenin was detected in the E-cadHCD133H subpopulations, which exhibited the highest drug resistance and MDFIC expression (Fig. 6a). The downstream target genes of β-catenin, Axin2, CD44, Cyclin D1, and Survivin, were upregulated in the E-cadHCD133H subpopulation compared with other subpopulations (Fig. 6b). The enforced expression of MDFIC p32 isoform increased the total and nuclear β-catenin (Fig. 6c and Supplementary Fig. S6a). Next, the transcriptional activity of β-catenin was evaluated by a TOP/FOP reporter assay. Overexpression of MDFIC p32 isoform significantly increased the luciferase activity, indicating the enhancement of the transcriptional activity of β-catenin (Supplementary Fig. S6b). Consistently, overexpression of MDFIC p32 isoform elevated the expression of downstream genes of β-catenin (Fig. 6d). Knockdown of β-catenin in the MDFIC p32 isoform-overexpressing E-cadHCD133L subpopulations increased the drug sensitivity of cells to the cisplatin and paclitaxol treatments compared to the scramble control group (Fig. 6e, f). In a xenograft animal study, overexpression of MDFIC-p32-myc significantly inhibited the therapeutic effect of cisplatin, while knockdown of β-catenin significantly sensitized the MDFIC-p32-myc-expressing tumors to cisplatin treatment (Fig. 6g). To confirm the role of MDFIC p32 isoform in nuclear translocation of β-catenin in clinical samples, IHC for the expression of MDFIC, and β-catenin was performed in the tumor samples from 25 NSCLC patients. The staining intensity of MDFIC and cytosolic and nuclear β-catenin was analyzed by using HistoQuest image analysis system. The results showed that the elevation of MDFIC levels was correlated with the increase of nuclear β-catenin staining signal (Fig. 6h). These results indicated that the MDFIC p32 isoform promoted the drug resistance of epithelial-type CSCs by enhancing the nuclear translocation and transcriptional activity of β-catenin.

Fig. 6: MDFIC contributes to the nuclear translocation and transcriptional activity of β-catenin.
figure 6

a The expression of the indicated proteins in the nuclear and cytoplasmic fractions of the four subpopulations of PC14 and A549. b The mRNA expression of the β-catenin-downstream genes in the four subpopulations of PC14 and A549. c The expression of the indicated proteins in the nuclear and cytoplasmic fractions of vector- or MDFIC-p32-myc-expressed of E-cadHCD133L subpopulation of PC14 and A549. d MDFIC-p32-myc was enforced to be expressed in the E-cadHCD133L subpopulation. The mRNA levels of the indicated genes were analyzed by qPCR. e The β-catenin was knocked down by two clones of shRNA in the stable clone of the MDFIC-p32-myc-expressed E-cadHCD133L subpopulation of PC14 and A549. The protein levels of β-catenin and MDFIC-p32-myc were analyzed by western blots. f Cell viability was analyzed after paclitaxol or cisplatin treatment for 72 h. g The four groups of tumor-bearing mice were treated with cisplatin for 4 weeks (n = 5). The tumor size was measured weekly. h The IHC staining signal of MDFIC and nuclear β-catenin was examined and plotted.

Full size image

Discussion

In the past decade, CSCs have been considered to display a mesenchymal phenotype, representing higher cell mobility, invasiveness, and drug resistance. However, a growing number of studies indicate that cancer cells with a stem-like phenotype may not always be in a mesenchymal state [13, 17, 21]. Here, we showed that CSCs (CD133H) could be isolated from both the E-cadH and E-cadL subpopulations of PC14 and A549 cell lines, suggesting heterogeneity of the EMT status of CSCs. The epithelial-type CSCs and mesenchymal-type CSCs had similar tumor-initiating and distal metastasis abilities. Nevertheless, the epithelial-type CSCs had higher self-renewal ability, tumor growth, and chemoresistance compared with the mesenchymal-type CSCs. Consistently, Tiran et al. reported that the CSCs isolated from a patient-derived lung adenocarcinoma cell culture showed a stronger association with the epithelial phenotype and were more resistant to cisplatin [21]. In triple-negative breast cancer, GSK-3 inhibitors selectively killed CD44+/CD24− cells with mesenchymal attributes and spared the cells with epithelial properties [16]. Pore et al. found that small-cell lung cancer with a mesenchymal phenotype (c-MET-high/E-cadherin-low) is associated with better survival and showed a trend toward lower circulating tumor cells [22]. In agreement with these reported data, our results revealed that epithelial-type CSCs might be a unique subpopulation driving chemoresistance, suggesting that targeting CSCs with epithelial attributes might be a potential strategy to prevent tumor relapse in NSCLC.

The heterogeneity of tumor is one of the reasons for drug resistance and relapse. Our study demonstrated the heterogeneous EMT and CSC phenotypes of cancer cells. It is worth to note that, in addition to CD133, more biomarkers for isolating CSCs in NSCLC have been identified in the past decade [23]. Although our data showed that the CD133H subpopulation harbored higher tumor-initiating ability compared with the CD133L subpopulations, we cannot exclude that the CD133L subpopulations may also contain tumor-initiating cells without CD133 expression. In addition, a little number of CD133H cells might escape from the FACS and contaminated the CD133L subpopulations. Thus, the cancer cell-derived tumors might remain heterogeneous.

Activation of canonical Wnt signaling by binding of Wnt ligands to Frizzled receptors leads to initiation of EMT, invasiveness, stemness, and drug resistance [24,25,26]. In the absence of Wnt ligands, β-catenin associates with the destruction complex consisting of adenomatous polyposis coli protein and Axin, which induce the phosphorylation of β-catenin at S45 and S33/37 by casein kinase 1 and GSK-3, respectively [27]. In epithelial cells, β-catenin binds to E-cadherin on the plasma membrane, while cytoplasmic β-catenin proteins associate with the destruction complex, leading to β-catenin degradation [28]. In our data, MDFIC interacted with the destruction complex and decreased the phosphorylation of β-catenin at Ser33/37, leading to the stabilization and nuclear translocation of β-catenin. Because MDFIC was selectively upregulated in the epithelial-type CSCs, the E-cadHCD133H subpopulation, the total β-catenin in the epithelial CSCs was dramatically higher than in the mesenchymal CSCs. Moreover, by comparing the epithelial-type CSCs and non-CSCs, nuclear β-catenin was significantly increased in the epithelial-type CSCs. These results confirmed that MDFIC contributes to β-catenin stabilization and nuclear translocation in epithelial CSCs, even in the absence of Wnt ligands.

MDFIC contains a cysteine-rich C-terminal domain, which shows a high degree of homology to the C-terminal domain of I-mfa [29, 30]. MDFIC interacts with several mammalian proteins and modulates their biological functions, although the regulatory mechanism of these MDFIC-mediated modulations is still unclear. Oakley et al. showed that MDFIC interacted with a glucocorticoid receptor and increased its phosphorylation at S211 and transcriptional activity [31]. Wang et al. reported that MDFIC bound to a regulatory histidine-rich region of cyclin T1 and a lysine and arginine-rich motif of cyclin T1 and T2 through its I-mfa domain [32]. Importantly, Kusano and Raab-Traub demonstrated that MDFIC directly bound to Axin through its I-mfa domain, and this interaction of MDFIC and Axin did not affect the binding of Axin to GSK-3 [33]. However, a Xenopus ortholog of MDFIC, XIC, inhibited the DNA binding of T cell factor 3 (TCF3) and the ability of β-catenin to activate the lymphoid enhancer factor/T cell factor (LEF/TCF) reporter constructs in early Xenopus embryos [34]. These studies indicate that MDFIC may be involved in the Wnt signaling pathway at different levels. Further research is required to understand the detailed mechanism of the MDFIC-regulated Wnt signaling pathway.

After nuclear translocation, β-catenin acts as a transcriptional coactivator with LEF/TCF transcription factors to stimulate the expression of Wnt downstream target genes [35]. Since the binding of MDFIC to Axin, mainly distributed in the cytoplasm, and LEF/TCF members, mainly localized in the nucleus, has been examined in other studies, the subcellular localization of MDFIC may ultimately affect its regulatory function in the Wnt signaling pathway [36]. The MDFIC p32 isoform lacks an N-terminal nuclear localization signaling and is distributed in the cytoplasm, while the MDFIC p40 isoform is mainly localized in the nucleus [30]. In our study, the MDFIC p32 isoform was associated with the destruction complex and contributed to β-catenin stabilization and nuclear translocation, while the MDFIC p40 isoform did not interact with β-catenin or other destruction complex members. Furthermore, our data showed that ectopic expression of the MDFIC p40 isoform did not alter the drug response of the cells to chemotherapy agents, suggesting that cytoplasmic MDFIC, mainly the MDFIC p32 isoform, is required for the β-catenin-mediated chemoresistance.

In our data, ectopic expression of MDFIC significantly increased the chemoresistance and sphere formation ability. Unexpectedly, we found that overexpression of MDFIC reversed EMT, while knockdown of MDFIC induced EMT, suggesting that MDFIC may function as an EMT inhibitor. These results suggest that the high level of MDFIC may maintain the epithelial phenotype of the E-cadHCD133H subpopulation, which exhibited higher chemoresistance and stemness. A recent study showed that E-cadherin, β-catenin, and SOX15 form a complex that binds to the promoter of caspase 3, leading to Twist1 cleavage and degradation in the selected epithelial-type subpopulation of the A549 cell line [17]. Accordingly, upregulation of MDFIC in the E-cadHCD133H subpopulation may contribute to the stabilization of the E-cadherin/β-catenin/SOX15 complex and subsequently maintain the epithelial phenotype of the cells by inhibiting Twist1 transcriptional activity. However, the role of SOX15 in cancer progression is still under debate [37,38,39]. Thus, future studies are required to explore the association of SOX15, β-catenin, and E-cadherin.

From an evolutionary perspective of neoplasia, EMT results in greater tumor initiation, metastatic potential, and resistance to therapies of cancer cells [40]. Many clinical studies indicate that high expression of mesenchymal or CSC biomarkers correlates with poor prognosis in lung cancer [41, 42]. However, many studies indicate that cancer cells with a stem-like phenotype may not always be in a mesenchymal state [13, 17, 21]. Here, we demonstrated that CSCs in the epithelial state were more resistant to chemotherapy agents. This phenomenon might be regulated through MDFIC-promoted β-catenin activity. Our study showed that targeting epithelial-type CSCs by inhibiting MDFIC and β-catenin activity may be a potential strategy to overcome chemoresistance in NSCLC.

Materials and methods

Human subjects

The tissue microarray (TMA-38AB), which was provided by Dr Michael Hsiao (Genomics Research Center, Academia Sinica, Taiwan) consists of tumor samples from 125 NSCLC patients enrolled in Kaohsiung Medical University Hospital from 1991 to 2007. All specimen collection procedures and analyses were approved by the Institutional Review Board of Kaohsiung Medical University Hospital (KMUHIRB-E(I)-20170048). Informed consent was obtained from all enrolled patients.

For studying the correlation between MDFIC and β-catenin nuclear translocation, the tumor samples from 25 NSCLC patients were collected from the Biobank (KMUH-BLOCK-108-0805) in Kaohsiung Medical University Hospital, Kaohsiung Taiwan with the approval of Institutional Review Board of Kaohsiung Medical University Hospital (KMUHIRB-E(I)-20190261). The histologic diagnosis of the specific type of lung cancer was made according to the recommendations of the World Health Organization classification, and the tumor size, local invasion, lymph node involvement, distal metastasis, and final disease stage were determined according to the 7th edition of the TNM staging system for lung cancer by the International Union Against Cancer and the American Joint Committee on Cancer [43]. The specimens were paraffin-embedded, sectioned, and used for IHC.

Cell lines and cell culture

For the maintenance of CSCs, the CD133H cells were cultured in ultralow adhesion culture dishes (Corning, New York, NY, USA) containing DMEM/F-12 with N2 supplement (Invitrogen), 20 ng/mL epidermal growth factor (EGF), and 20 ng/mL basic fibroblast growth factor (referred as stem-cell medium) (PeproTech, Rocky Hill, NJ, USA). The cell lines were kept under a humidified incubator containing 5% CO2 at 37 °C. Only early passage (<20 passage) CD133H cells were used in this study. All cells were routinely tested for mycoplasma contamination.

Fluorescence-activated cell sorting (FACS) and isolation of the subpopulations

For isolating the epithelial (E-cadH)- and mesenchymal (E-cadL)-type NSCLC cells, 1 × 106 PC14 and A549 cells were incubated with the PE-conjugated anti-E-cadherin antibody (BD Biosciences, San Jose, CA, USA) on ice for 40 min. After washing, the E-cadH (top 2%) and E-cadL (bottom 2%) cells were sorted by using a FACSAria cell sorter (BD Biosciences). Two rounds of FACS were performed to ensure the purity of the sorted subpopulations (>95%). For isolating the CD133H subpopulations, the E-cadH and E-cadL subpopulations were infected by a CD133 P1 promoter-driven GFP reporter lentivirus [44]. The GFP-positive cells were sorted and recovered in the stem-cell medium in ultralow adhesion culture dishes, which allowed the growth of tumor spheres. To evaluate the sphere formation efficiency of each subpopulation, 500 cells were cultured in six-well ultralow adhesion culture plates containing stem-cell medium. After 1 week, ten photos of nonoverlapping areas were randomly taken, and the number of tumor spheres was counted.

Gene expression microarray

Total RNA was isolated from the four subpopulations by using RNeasy Mini Kits (QIAGEN, Valencia, CA, USA). The Affymetrix cDNA microarray analysis was performed according to the user manual. The cDNA samples were purified and labeled with biotin-conjugated ribonucleotides using an IVT labeling kit (Affymetrix, Inc., Santa Clara, CA, USA). The labeled cDNA was subsequently fragmented and hybridized overnight with the Affymetrix human U133 2.0 plus arrays (Affymetrix). After hybridization, the chips were washed and stained. The chips were scanned using an Affymetrix Gene Chips canner 3000. The .DAT files were processed using the Affymetrix Gene Chip Operating System to generate the .CEL files. The raw intensities in the .CEL files were normalized by robust multichip analysis, and fold-change analysis was performed using GeneSpring GX software (Agilent Technologies, Santa Clara, CA, USA). The microarray data were uploaded, and the accession number is GSE93586 in the Gene Expression Omnibus database of National Center for Biotechnology Information.

Statistical analysis

All observations were confirmed by at least three independent experiments, and the data are presented as the mean ± SD. A two-tailed Student’s t test was used for the comparison between two groups. A 1-way ANOVA followed by Dunnett’s post hoc test was used for multiple comparisons. The variations within each group were considered in 1-way ANOVA. The similarity of variance between groups for each comparison was tested. The Fisher’s exact test was used for the tumor initiation animal study. The log-rank test was applied to evaluate the statistical significance of the Kaplan–Meier survival curve. Cox proportional hazards regression analysis was used to test the prognostic significance of the factors in the univariate and multivariate models. The association of DRS and tumor relapse was evaluated by Pearson’s Chi-square test. For all statistical tests, *p < 0.05 or **p < 0.01 was considered significant. Statistical analysis was performed using SPSS (Statistical Package for the Social Sciences, version 14.0) software.

For the primer sequences used in real-time PCR (Supplementary Table 8) and other detail information please see Supplementary information.