Muscleblind-like 2 controls the hypoxia response of cancer cells.

Hypoxia is a hallmark of solid cancers, supporting proliferation, angiogenesis, and escape from apoptosis. There is still limited understanding of how cancer cells adapt to hypoxic conditions and survive. We analyzed transcriptome changes of human lung and breast cancer cells under chronic hypoxia. Hypoxia induced highly concordant changes in transcript abundance, but divergent splicing responses, underlining the cell type-specificity of alternative splicing programs. While RNA-binding proteins were predominantly reduced, hypoxia specifically induced muscleblind-like protein 2 (MBNL2). Strikingly, MBNL2 induction was critical for hypoxia adaptation by controlling the transcript abundance of hypoxia response genes, such as vascular endothelial growth factor A (VEGFA) MBNL2 depletion reduced the proliferation and migration of cancer cells, demonstrating an important role of MBNL2 as cancer driver. Hypoxia control is specific for MBNL2 and not shared by its paralog MBNL1. Thus, our study revealed MBNL2 as central mediator of cancer cell responses to hypoxia, regulating the expression and alternative splicing of hypoxia-induced genes.

INTRODUCTION

Hypoxia is a common feature of solid tumors as the oxygen supply to the rapidly proliferating cancer cells is inadequate. Two types of tumor hypoxia occur as a result of structurally and functionally abnormal tumor vasculature. “Chronic” hypoxia arises in cancer cells that are too far away from the blood vessels to be supplied with sufficient oxygen by diffusion. “Acute” hypoxia is the result of temporarily limited perfusion by immature tumor vessels (Vaupel et al. 2004). Such hypoxic regions are spatially and temporally dynamic due to tumor angiogenesis and the reversibility of acute hypoxia, resulting in severe intra- and intertumoral heterogeneity of cellular oxygen levels. The oxygen content in hypoxic areas of solid tumors is typically <1.4% and may reach complete anoxia (Vaupel et al. 2007).

Hypoxia promotes virtually every step in tumor progression contributing to metabolic reprogramming, angiogenesis, extracellular matrix remodeling, epithelial to mesenchymal transition, cancer stem cell maintenance, immune evasion, and cell death resistance (Schito and Semenza 2016). Thus, tumor hypoxia signifies a more aggressive phenotype resulting from increased growth and metastasis. Moreover, tumor hypoxia counteracts radio-, chemo-, and immunotherapy, highlighting the importance of developing effective hypoxia-targeted therapies (Graham and Unger 2018). Well studied is the role of the HIF (hypoxia inducible factor) transcription factors in the adaptation to hypoxia. HIFs are heterodimers consisting of an oxygen-labile α subunit and a constitutively stable β subunit. Under hypoxia, HIFα subunits are stabilized, allowing for the transcriptional activation of several hundred target genes (Samanta and Semenza 2018).

Independent from transcriptional adaptation, hypoxia triggers a widespread post-transcriptional response by altering alternative splicing (AS) decisions, translation efficiency, and mRNA stability. For example, at low glucose conditions, several mRNAs of HIF target genes are stabilized under hypoxia (Carraway et al. 2017; Fry et al. 2017). While canonical protein synthesis is reduced due to the limited ATP supply during hypoxia, alternative translation pathways are activated, to sustain translation of proteins essential for hypoxia adaptation (for review, see Chee et al. 2019). Remarkably, ribosomal profiling of hypoxic primary hepatocytes showed that changes in translation even precede the transcriptional response (Hettiarachchi et al. 2019). In addition, hypoxia-driven AS events diversify the cellular response to hypoxia, both in primary and cancerous cells (Weigand et al. 2012; Hu et al. 2014; Brady et al. 2017; Han et al. 2017; Bowler et al. 2018). The current knowledge of the factors that control these post-transcriptional pathways is still very limited, but crucial in understanding the tumor-promoting effects of hypoxia.

In line with the central role of RNA-binding proteins (RBPs) in post-transcriptional gene control, aberrant RBP activities have been identified as potent cancer drivers (for reviews, see Anczuków and Krainer 2016; Pereira et al. 2017). Just recently, MBNL proteins were found to affect tumorigenesis (Perron et al. 2018; Tabaglio et al. 2018). The MBNL family consists of the three closely related paralogs, MBNL1, MBNL2, and MBNL3 (for review, see Konieczny et al. 2014). MBNL1 and MBNL2 are ubiquitously expressed, whereas MBNL3 expression is more restricted, with highest levels found in the placenta (Fardaei et al. 2002). All three MBNL paralogs encode two amino-terminal tandem zinc finger (ZnF) domains. These bind to RNAs containing two or more clustered 5′-YGCY-3′ motifs (Lambert et al. 2014). The ability of MBNLs to bind RNA is further altered by the structural context of the RNA targets (deLorimier et al. 2017; Taylor et al. 2018). Binding of MBNLs activates or represses mRNA alternative splicing depending on MBNL binding location (Charizanis et al. 2012; Wang et al. 2012), with MBNL1 being the most potent splicing regulator among the three paralogs (Sznajder et al. 2016). Additionally, MBNL proteins are proposed to regulate mRNA stability (Masuda et al. 2012) and localization (Adereth et al. 2005; Wang et al. 2012), alternative polyadenylation (Batra et al. 2014), and miRNA biogenesis (Rau et al. 2011). All three MBNL proteins undergo autoregulatory feedback loops and can functionally compensate each other in some cases (Konieczny et al. 2018).

In order to identify RBPs central to the post-transcriptional response of cancer cells during hypoxia, we obtained a comparative set of hypoxia-driven changes in gene and isoform expression as well as changes in RBP levels in two cancer types. We focused on lung and breast cancer, since these are the most common cause for cancer-related deaths in males and females, respectively (Bray et al. 2018; Siegel et al. 2019). Our data show highly correlated changes in transcript abundance, but strongly divergent changes in AS. Strikingly, chronic hypoxia specifically induces MBNL2, while overall RBP levels are reduced. We demonstrate a major role of MBNL2 in promoting hypoxia-dependent gene expression changes, both at the level of transcript abundance and AS. Accordingly, knockdown of MBNL2 in cancer cells leads to reduced proliferation and migration, emphasizing MBNL2 as a driver of cancer cell function in hypoxia.

RESULTS

Chronic hypoxia induces similar changes in transcript abundance, but divergent alternative splicing programs

We chose A549 lung and MCF-7 breast cancer cells to compare transcriptomic changes in response to chronic hypoxia, as both cell lines have no deletions or mutations within the HIF pathway, the two major families of splicing factors, the SR proteins (serin/arginine rich) and hnRNPs (heterogeneous nuclear ribonucleoproteins) or core spliceosomal genes, such as mutations in SF3B1 (splicing factor 3B1), which are common in cancer (Cerami et al. 2012; Gao et al. 2013; Seiler et al. 2018). First, we identified the lowest oxygen concentration that would allow continued survival under hypoxic stress. We incubated A549 and MCF-7 cells for 48 h at two different oxygen levels, 0.5% and 0.2% O2. Cell viability after hypoxia treatment was compared to normoxic (21% O2) controls using crystal violet staining. While incubation at 0.5% O2 showed no difference in cell viability between normoxic and hypoxic cells (Fig. 1A), lower oxygen levels of 0.2% significantly reduced the viability of both cell types (Supplemental Fig. 1). Further, VEGFA mRNA levels were quantified to control for the induction of hypoxia response genes after 48 h at 0.5% O2. VEGFA is a widely expressed, primary target of HIF transcription factors and thus, a commonly used marker for the hypoxia response of cells (Forsythe et al. 1996; Blancher et al. 2000). RT-qPCR confirmed a 3.6-fold or 7.2-fold increase in VEGFA mRNA levels in A549 and MCF-7 cells, respectively (Fig. 1B). Thus, incubation at 0.5% O2 for 48 h robustly induces a hypoxia response in both cell types without causing cell death and was therefore used for all further experiments.

FIGURE 1.

Chronic hypoxia induces highly correlated changes in transcript abundance in breast and lung cancer cells. (A) Comparison of the cell viability of A549 and MCF-7 cells cultivated for 48 h under normoxic (21% O2) or hypoxic (0.5% O2) conditions. n = 3. (B) RT-qPCR quantification of VEGFA mRNA levels in normoxic and hypoxic A549 and MCF-7 cells. Values are normalized to the housekeeping gene RPLP0. n = 3 (A549) and n = 4 (MCF-7). (C) Venn diagram showing the overlap of differentially expressed genes in A549 and MCF-7 cells in response to hypoxia. P < 9.52 × 10−124 (Fisher's exact test, one-tailed). The number in brackets refers to genes regulated in the same direction. (D) Scatter plot comparing the log2-transformed fold changes (hypoxia/normoxia) of 1224 shared differentially expressed genes in A549 and MCF-7 cells. Pearson correlation coefficient and associated P-value are given. (E) RT-qPCR quantification of the mRNA levels of hypoxia-induced genes in normoxic and hypoxic A549 and MCF-7 cells. Values are normalized to the housekeeping gene RPLP0. n = 5. (**) P < 0.01, (*) P < 0.05.

For the transcriptome-wide comparison of hypoxia responses, total RNA was extracted from normoxic and hypoxic A549 and MCF-7 cells, depleted from ribosomal RNA and subjected to deep sequencing (n = 2). Approximately 100 million reads were obtained for each sample. We detected expression of 18,214 genes in A549 and 18,880 genes in MCF-7 cells, with 15,684 genes shared between both cell types (transcripts per million [TPM] > 1 in at least one of the samples for each cell type). From these, differential expression analysis identified 2490 genes (16%) in A549 and 4503 genes (29%) in MCF-7, which changed significantly in their abundance (>1.5-fold, adjusted P < 0.05, Supplemental Table 1). Gene ontology (GO) analyses of the differentially expressed genes confirmed the expected induction of hypoxia-associated processes like metabolic reprogramming and extracellular matrix remodeling. Down-regulated genes were enriched for functions in DNA replication and ribosome biogenesis, reflecting lower energy consumption in response to hypoxia (Supplemental Fig. 2).

Comparison of the differentially expressed genes in the two cell types showed an overlap of 1224 altered genes, which corresponds to almost half of all regulated genes in A549 (49%) and more than a quarter (27%) in MCF-7 cells (Fig. 1C). These 1224 genes showed a highly correlated response to hypoxia, with >90% regulated in the same direction in both cell types (Fig. 1D). As expected, VEGFA mRNA levels were elevated in both cell types, with a 4.5-fold or 6.5-fold increase in A549 and MCF-7 cells, respectively. For further verification, we quantified the induction of three other known hypoxia response genes (DDIT4, LDHA, PLOD2) and confirmed their expected increase in both cell types (Fig. 1E).

Accompanying the changes in transcript abundance, we detected global changes in AS events (absolute delta percent spliced-in [|ΔPSI|] ≥ 10%, false discovery rate [FDR] < 0.05, Supplemental Table 2). Among the detected splicing events, cassette exons (CE) were preferentially altered in response to chronic hypoxia in both, A549 and MCF-7 cells (Fig. 2A). However, the preferred directionality of the AS events was slightly different between the two cell types. While CE were predominantly skipped in response to hypoxia in MCF-7 cells, A549 cells showed a slight preference for more exon inclusion. In contrast, usage of alternative 5′ and 3′ splice sites as well as complete intron retention was increased in hypoxic MCF-7 cells, but decreased in hypoxic A549 cells (Fig. 2B). Moreover, although the overall number of differentially spliced genes was similar to the number of differentially expressed genes, only 199 shared AS events were detected (Fig. 2C). This corresponds to just 9% or 5% of the AS events detected in A549 and MCF-7 cells, respectively. Moreover, of these 199 AS events only 123 (62%) were regulated in the same direction (Fig. 2D). For verification, we chose eight and seven CE from A549 and MCF-7, respectively. Cassette exons were selected from genes important for tumorigenesis. Comparison of exon inclusion levels with total mRNA levels under normoxia and hypoxia confirmed the expected changes for 10 of the 15 selected events (Supplemental Fig. 3; Supplemental Tables 3, 4). As an example, the reduced inclusion in response to hypoxia of CENPE exon 17 is shown for both cell types in Figure 2E.

FIGURE 2.

Hypoxia-dependent alternative splicing programs are highly cell type-specific. (A) Percentage of significantly altered splicing events (|ΔPSI| ≥ 10%, FDR < 0.05) in A549 and MCF-7 cells. Different event types are color coded. (CE) Cassette exon, (MXE) mutually exclusive exons, (A5SS) alternative 5′ splice site, (A3SS) alternative 3′ splice site, (RI) retained intron. (B) Percentage of inclusion and skipping for different event types in A549 and MCF-7 cells. (C) Venn diagram showing the overlap of alternative splicing events in A549 and MCF-7 cells in response to hypoxia. Not significant (Fisher's exact test, one-tailed). The number in brackets refers to AS events regulated in the same direction. (D) Scatter plot comparing the ΔPSI of shared alternative splicing events in A549 and MCF-7 cells. Pearson correlation coefficient and associated P-value are given. (E) RT-PCR of CENPE exon 17 after hypoxia treatment in A549 and MCF-7 cells. The short (118 bp) and long (192 bp) PCR products correspond to exon 17 exclusion and inclusion, respectively. n = 2. (N) Normoxia, (H) hypoxia, (M) size marker.

In summary, while chronic hypoxia triggers a highly correlated change in transcript abundance within the two cancer types tested, the observed AS changes are substantially different. In addition to the fact that only a small fraction of AS events are common to both cell types, AS events are also frequently changed in the opposite direction.

MBNL2 is induced by chronic hypoxia and facilitates cancer cell adaptation

Several RBPs and in particular major splicing factors, such as SR proteins and hnRNPs, are associated with cancer progression and metastasis (Anczuków and Krainer 2016). Thus, we were interested in changes of the RBPome in hypoxic A549 and MCF-7 cells. At the transcript level, RBPs are preferentially reduced, most noticeable for core spliceosomal proteins and hnRNPs (Supplemental Table 5).

SR proteins are known to change their phosphorylation status under hypoxia due to the differential expression of SR protein kinases (Jakubauskiene et al. 2015; Bowler et al. 2018). Thus, we focused on the hypoxia response of hnRNPs, which has not been analyzed so far. We could show reduced mRNA levels of all six tested hnRNPs, hnRNP A1, C, DL, E2, L, and M, under hypoxia in MCF-7 cells (Fig. 3A). In A549 cells, hnRNP A1 and M mRNA levels were significantly reduced under hypoxia, while the other tested hnRNPs showed only a trend toward mRNA level reduction. Since changes in mRNA levels do not necessarily reflect altered protein levels and vice versa, we also examined the protein levels of hnRNPs under hypoxia. Western blot analyses confirmed reduced protein levels of all six hnRNPs in both cell types (Fig. 3B). Mechanistically, these may result from overall reduced transcript levels or a shift of AS toward NMD-sensitive isoforms as shown for hnRNP DL in MCF-7 cells (Supplemental Fig. 4; Kemmerer et al. 2018). For some hnRNPs in A549 cells, however, our analysis did not reveal any changes in transcript abundance or AS events, indicating further regulatory mechanisms, such as lower translation efficiency or protein stability. In addition to splicing factors, like SR proteins and hnRNPs, changes in core spliceosomal proteins will affect splicing changes. Thus, we examined the mRNA levels of four core spliceosomal proteins (LSM4, SF3B4, SNRPC, and SNRPD1) and, similar to hnRNPs, found a more pronounced reduction of core spliceosomal factor mRNA levels in MCF-7 cells (Supplemental Fig. 5). This finding is in line with the increased exon skipping and intron retention specifically detected in hypoxic MCF-7 cells, suggesting an overall lower splicing efficiency of hypoxic MCF-7 compared to hypoxic A549 cells.

FIGURE 3.

Chronic hypoxia reduces hnRNP mRNA and protein levels. (A) RT-qPCR quantification of hnRNP A1, C, DL, E2, L, and M mRNA levels in normoxic and hypoxic A549 and MCF-7 cells. Values are normalized to the housekeeping gene RPLP0. n = 3–5. (B) Western blot of hnRNPs A1, C, DL, E2, L, and M in normoxic and hypoxic A549 and MCF-7 cells. Anti-hnRNP antibodies were used to visualize the respective protein levels. n = 4.

Among the few elevated RBPs under hypoxia, MBNL2 caught our interest as it has recently been shown to be elevated in ccRCC (clear cell renal cell carcinoma) and HCC (hepatocellular carcinoma) patient samples (Lee et al. 2016; Perron et al. 2018). We verified increased mRNA and protein levels of MBNL2 in A549 and MCF-7 cells (Fig. 4A,B; Supplemental Fig. 6). Interestingly, its paralogs MBNL1 and MBNL3 were unaffected or even decreased at mRNA and protein level (Fig. 4A,B; Supplemental Fig. 7), suggesting a specific role of MBNL2 in hypoxia adaptation. Since MBNL2 was described as an oncogene in ccRCC (Perron et al. 2018), but as a tumor suppressor in HCC (Lee et al. 2016), we investigated the influence of MBNL2 expression on the cell viability of hypoxic A549 and MCF-7 cells. We performed an siRNA-mediated, transient knockdown of MBNL2 in both cell types and quantified cell viability in the presence and absence of the chemotherapeutic drug cisplatin using crystal violet staining. Cisplatin concentrations were chosen to induce only low levels of cell death under normoxic conditions, allowing for the detection of changes in cisplatin sensitivity upon hypoxia and/or knockdown of MBNL2 (Supplemental Fig. 8A). MBNL2 knockdown induced cell death in hypoxic A549, both with or without cisplatin-treatment (Fig. 4C; Supplemental Table 6). Moreover, knockdown of MBNL2 had a stronger effect on A549 cell viability under hypoxia compared to normoxia in all tested conditions (Supplemental Fig. 8B). In addition, migration of hypoxic A549 cells was reduced after knockdown of MBNL2 (Fig. 4D,E). In hypoxic MCF-7 cells, MBNL2 knockdown increased cisplatin-induced cell death at 20 µM (Fig. 4C). Further, knockdown of MBNL2 reduced cell viability of hypoxic, but not normoxic MCF-7 cells in the absence of cisplatin (Supplemental Fig. 8B). Together, these results suggest that MBNL2 induction supports the adaptation of breast and lung cancer cells to hypoxia.

FIGURE 4.

MBNL2 is specifically induced by chronic hypoxia and contributes to the adaptation of cancer cells. (A) RT-qPCR quantification of the mRNA levels of MBNL1 and MBNL2 in normoxic and hypoxic A549 and MCF-7 cells. Values are normalized to the housekeeping gene RPLP0. n = 4. (B) Western blot of MBNL1 and MBNL2 in normoxic and hypoxic A549 and MCF-7 cells. Anti-MBNL1 and anti-MBNL2 antibodies were used to visualize the respective protein levels. Total lane protein is shown as loading control. n = 2. (N) Normoxia, (H) Hypoxia. (C) Cell viability of hypoxic A549 and MCF-7 cells after knockdown of MBNL2. Cells were additionally treated with the indicated concentrations of cisplatin for 24 h. Absorption was normalized to normoxic control cells with the respective cisplatin concentration. n = 3. (D) Migration assay of hypoxic A549 cells after knockdown of MBNL2. As positive control, cells were stimulated with 100 ng/µL hEGF. n = 3. (E) Representative pictures used to quantify the numbers of migrated A549 cells in D. (**) P < 0.01, (*) P < 0.05.

MBNL2 controls the transcript abundance of hypoxia response genes

To gain insight into the molecular mechanism by which MBNL2 facilitates the adaptation of cancer cells to hypoxia, we performed a transcriptomic analysis after the knockdown of MBNL2 in hypoxic MCF-7 cells (Fig. 5A; Supplemental Fig. 9). Total RNA was extracted from hypoxic MCF-7 cells transfected with an siRNA targeting MBNL2 (siMBNL2) or a nonsilencing control siRNA (siCTRL), poly(A)-selected and subjected to deep sequencing (n = 2). Approximately 80 million reads were obtained for each sample. A total of 4370 (27%) genes were differentially expressed after knockdown of MBNL2 (>1.5-fold, adjusted P < 0.05, Supplemental Table 7). GO analysis of these genes showed an enrichment in hypoxia response genes, including genes related to glucose metabolism, emphasizing a contribution of MBNL2 to hypoxia adaptation (Supplemental Fig. 10).

FIGURE 5.

MBNL2 controls the hypoxia response of breast and lung cancer cells by regulating transcript abundance. (A) Western blot of MBNL2 after siRNA-mediated knockdown in hypoxic A549 and MCF-7 cells. Anti-MBNL2 was used to verify the respective knockdown. Total lane protein is shown as loading control. n = 3–4. (B) Venn diagram showing the overlap of differentially expressed genes after knockdown of MBNL2 and in response to hypoxia in MCF-7. P < 1.61 × 10−37 (Fisher's exact test, one-tailed). The number in brackets refers to genes regulated in the opposite direction. (C) Scatter plot comparing the fold changes of shared differentially expressed genes after knockdown of MBNL2 and in response to hypoxia in MCF-7. (D) RT-qPCR quantification of the mRNA levels of HIF target genes in normoxic and hypoxic A549 and MCF-7 cells and after knockdown of MBNL2. Values are normalized to the housekeeping gene RPLP0. n = 3–6. (E) Western blot of MBNL1 and MBNL2 after siRNA-mediated knockdown of MBNL1 in hypoxic MCF-7 cells. Anti-MBNL1 and anti-MBNL2 were used to verify the respective protein levels. Total lane protein is shown as loading control. n = 3. (F) RT-qPCR quantification of the mRNA levels of MBNL1, MBNL2, and HIF target genes. n = 3. (**) P < 0.01. (*) P < 0.05.

Comparison of the differentially expressed genes after MBNL2 knockdown with the hypoxia response of MCF-7 cells showed 1528 shared regulated genes (Fig. 5B; Supplemental Table 7). Thus, about a third of all hypoxia-regulated genes in MCF-7 cells are affected by MBNL2 levels. These shared genes were preferentially changed in the opposite direction (71%) (Fig. 5C), supporting a role of MBNL2 induction in the regulation of hypoxia response genes.

Interestingly, a similar proportion of shared genes was found when comparing only to genes that were regulated consistently in hypoxic A549 and MCF-7 cells (351 out of 1,109 genes, 32%; Supplemental Fig. 11A; Supplemental Table 7), suggesting a cell type-independent function of MBNL2 in controlling hypoxia adaptation. Noticeably, these shared differentially expressed genes were mostly increased in response to hypoxia and decreased after knockdown of MBNL2, including several HIF target genes (Supplemental Fig. 11B).

This suggests that one function of MBNL2 induction might be to further increase the mRNA levels of HIF target genes. To test this hypothesis, we verified by RT-qPCR that the mRNA levels of HIF targets were reduced after the knockdown of MBNL2. Indeed, knockdown of MBNL2 significantly attenuated the hypoxia-dependent induction of the HIF target genes ALDOC, ENO2, ITGA5, LOX, and VEGFA in MCF-7 cells (Fig. 5D). Likewise, a reduced induction of ALDOC, ITGA5, and VEGFA after MBNL2 knockdown was detected in A549 cells (Fig. 5D).

In order to investigate whether our findings from the cell lines are relevant in cancer patients, we used mRNA level data from patients with breast invasive carcinoma (BRCA; 1088 patients) and lung adenocarcinoma (LUAD; 533 patients) of The Cancer Genome Atlas (TCGA) PanCancer Atlas (Hoadley et al. 2018). We detected significant positive correlations of MBNL2 mRNA levels with the mRNA levels of the hypoxia response genes VEGFA, ITGA5, and LOX in BRCA patients and LOX in LUAD patients (Supplemental Fig. 12). Further, MBNL2 mRNA levels were positively correlated with HIF1A and HIF2A mRNA levels, suggesting that MBNL2 induction might also depend on hypoxia signaling in cancer patients (Supplemental Fig. 12). More globally, when comparing hypoxia response genes, that is, genes that were significantly induced in A549 and MCF-7 in response to hypoxia (Fig. 1D), to all other genes in the TCGA data sets, we found a significant trend toward positive correlation with MBNL2 mRNA levels (Supplemental Fig. 13; Supplemental Table 8). Together, these observations support that MBNL2 controls the transcript abundance of several HIF target genes under hypoxia, with potential implications in human cancer patients.

In order to test whether the paralog MBNL1 is also able to control the transcript abundance of HIF target genes, we performed an siRNA-mediated knockdown of MBNL1 under hypoxia (Fig. 5E). Interestingly, knockdown of MBNL1 did not lead to a reduction of HIF target mRNA levels, but to an increase (Fig. 5F). This increase is most likely the result of an associated increase in MBNL2 levels (Fig. 5E), due to cross-regulation between the two paralogs (Konieczny et al. 2018). This underlines the specific role of MBNL2 in the hypoxia response of cancer cells.

MBNL2 binds to the VEGFA 3′UTR and controls the level of secreted VEGF-A protein

It has been suggested that MBNL2 stabilizes mRNAs from correlation of MBNL2 mRNA levels with putative target mRNA levels across a panel of different cell types and tissues (Ray et al. 2013). However, knockdown studies showed no global decrease in target mRNA levels, but suggested involvement of MBNL2 in mRNA localization, in particular of secreted proteins, thereby promoting efficient protein secretion (Wang et al. 2012). Thus, we tested whether MBNL2 affects the stability of the VEGFA mRNA and/or the secretion of VEGF-A protein.

A direct interaction of MBNL2 with the VEGFA mRNA was confirmed by MBNL2 immunoprecipitation experiments (Perron et al. 2018). Since CLIP-seq data mostly showed binding of MBNL2 to 3′UTRs (Charizanis et al. 2012), we first tested whether MBNL2 is able to directly recognize the 3′UTR of VEGFA. Thus, we examined the 3′UTR of VEGFA for possible MBNL2 binding sites, that is, two clustered 5′-YGCY-3′ motifs (Lambert et al. 2014). We identified three potential binding sites in the 3′UTR of the VEGFA mRNA (Supplemental Fig. 14). These potential binding sites, plus 40–60 nt flanking sequences, were cloned in place of the natural MBNL binding site of the Atp2a1 minigene, an established test system for MBNL binding, based on the differential inclusion of Atp2a1 alternative exon 22 (Cywoniuk et al. 2017). In this system, exon 22 inclusion levels are a measure of MBNL binding activity (Supplemental Fig. 15). Alternative exon 22 splicing is analyzed after plasmid-based expression of MBNLs in a MEF (mouse embryonic fibroblast) cell line with a genetic knockout of both, Mbnl1 and Mbnl2. This allows for individual testing of the binding behavior of different MBNL paralogs and their various isoforms.

The expression pattern of the different MBNL isoforms depends on tissue type and developmental stage (Wang et al. 2012). The different isoforms show distinct characteristics with respect to their cellular localization, RNA binding affinity and splicing regulation (Sznajder et al. 2016). MBNL2 in particular contains two alternatively spliced exons in its coding sequence giving rise to four protein isoforms ranging from 38 to 41 kDa (Sznajder et al. 2016). MBNL2 isoform 40 (MBNL2-40) was chosen for binding site assessment in this study, because it is the most efficient splicing regulator of the four MBNL2 protein isoforms (Sznajder et al. 2016) and the most abundant isoform detected in MCF-7 cells (Fig. 4B). MBNL1 isoform 42 was chosen for comparison because MBNL1-42 is the corresponding isoform to MBNL2-40, containing a similar sequence composition encoded by the same alternative exons (Sznajder et al. 2016) and is also abundantly expressed in MCF-7 cells (Fig. 4B).

The naturally encoded MBNL binding site present within the Atp2a1 minigene served as a positive control and led to an increased inclusion of exon 22 after expression of MBNL2-40, while inclusion was reduced when this binding site was deleted from the minigene (Supplemental Fig. 16). The inclusion of exon 22 was restored after MBNL2-40 expression and insertion of VEGFA binding sites 1 or 3 into the minigene, less so with binding site 2 (Supplemental Fig. 16). Equal overexpression efficiencies (Supplemental Fig. 17) allowed us to compare the splicing activity and thus, the VEGFA 3′UTR binding capacity, of MBNL2-40 and MBNL1-42. Comparing the ΔPSI values for the positive control, which contains the high affinity Atp2a1 binding site, showed a higher splicing regulatory activity of MBNL1-42, compared to MBNL2-40 (Fig. 6A). This observation is consistent with previous experiments showing that MBNL1 isoforms are more active splicing regulators than the corresponding MBNL2 isoforms (Sznajder et al. 2016). Strikingly, using the VEGFA binding site 1, MBNL2-40 was similarly efficient in restoring exon 22 inclusion as in the positive control, suggesting that MBNL2-40 could bind to the VEGFA 3′UTR with high affinity. In contrast, MBNL1-42 was not able to restore exon 22 inclusion with any of the three VEGFA binding sites to the level of the positive control, indicating that it might bind to the VEGFA 3′UTR with lower affinity (Fig. 6A).

FIGURE 6.

MBNL2 binds to the 3′UTR of VEGFA and controls secreted VEGF-A protein levels. (A) RT-PCR quantification showing different Atp2a1 exon 22 (E22) inclusion levels depending on inserted binding sites (BS) from the VEGFA 3′UTR and MBNL2 isoform 40 (MBNL2-40) or MBNL1 isoform 42 (MBNL1-42) expression in Mbnl1 and 2 double knockout MEF cells. Shown are ΔPSI values with respect to the negative control in which the binding site is deleted (see also Supplemental Fig. 15). Statistical significance is given in reference to the inclusion level observed with the natural Atp2a1 binding site. n = 3. (B) Relative amounts of MBNL1 and MBNL2 bound to the VEGFA 3′UTR RNA and calculated KD values. RNA carrying (CUG)20 repeats was used as positive control. n = 3. (C) Half-life of VEGFA mRNA. MCF-7 cells were treated with 5 µg/mL actinomycin D (ActD). Thereafter, total RNA was isolated at 2 h intervals and VEGFA mRNA levels were quantified by RT-qPCR. VEGFA mRNA level was normalized to the housekeeping gene RPLP0. n = 3. (D) Quantification of secreted VEGF-A protein levels by ELISA. Secreted VEGF-A levels were determined in the supernatants of normoxic and hypoxic A549 and MCF-7 cells and after knockdown of MBNL2. n = 4. (**) P < 0.01. (*) P < 0.05.

In order to examine the binding affinities of MBNL2 and MBNL1 to the VEGFA 3′UTR directly, we quantified the interaction between the two MBNL paralogs and the three RNA binding sites from the VEGFA 3′UTR in vitro. MBNL2 showed an 11-fold higher binding affinity for VEGFA binding site 1 than MBNL1 (Fig. 6B; Supplemental Fig. 18), underlining the results from the Atp2a1 splicing assay (Fig. 6A). Thus, MBNL2 is able to directly recognize the 3′UTR of VEGFA and has a higher binding affinity to binding site 1 than its paralog MBNL1.

Next, we tested whether MBNL2 stabilizes the mRNA of VEGFA. Thus, we measured changes in mRNA half-life after knockdown of MBNL2 in hypoxic MCF-7 cells. Actinomycin D assays showed no change in VEGFA mRNA half-life in response to MBNL2 knockdown (Fig. 6C). We then tested if MBNL2 affects VEGF-A protein levels and quantified secreted VEGF-A by ELISA (Fig. 6D). Knockdown of MBNL2 completely suppressed hypoxia-dependent induction of secreted VEGF-A protein in both, A549 and MCF-7 cells. Thus, regulation at protein level is more pronounced than regulation at mRNA level. In sum, MBNL2 directly recognizes the 3′UTR of VEGFA. Our results indicate that it does not affect the stability of VEGFA mRNA, but possibly affects secretion of the VEGF-A protein.

MBNL2 controls hypoxia-dependent AS

Along with the changes in transcript abundance, we also detected 2074 MBNL2-dependent changes in AS events (|ΔPSI| ≥ 10%, FDR < 0.05, Supplemental Table 9). For verification, we chose six CE sensitive to MBNL2 knockdown. Cassette exons were selected from genes important for tumorigenesis. Comparison of exon inclusion levels with total mRNA levels with and without knockdown of MBNL2 in hypoxic MCF-7 cells, confirmed the expected changes for five of the six selected events (Supplemental Fig. 19; Supplemental Table 10).

Comparison with the splicing changes during hypoxia showed 393 shared events, corresponding to 19% or 9% of the events detected after knockdown of MBNL2 or hypoxia treatment of MCF-7 cells, respectively (Fig. 7A). As for the differential expression described above, these 393 AS events were preferentially regulated in the opposite direction (89%), suggesting that MBNL2 induction contributes to hypoxia-dependent alternative splicing (Fig. 7B). The majority of these were exon skipping events in which the inclusion of CE is reduced in response to hypoxia. These are increased again after MBNL2 knockdown, indicating a repressive role of MBNL2. This is exemplified by PIGN exon 12, which inclusion is reduced in response to hypoxia. Inclusion was sustained after simultaneous knockdown of MBNL2 (Fig. 7C), demonstrating that modulation of MBNL2 levels is sufficient to reverse hypoxia-induced splicing changes in MCF-7 cells. PIGN catalyzes one step in the synthesis of GPI (glycosylphosphatidylinositol)-anchors of membrane proteins. The alternatively spliced exon 12 is located in the catalytic domain and may thus affect the enzymatic function of PIGN. Interestingly, a reduction of PIGN function through alternative splicing has been shown to cause genetic instability and thus promote leukemic progression in acute myeloid leukemia (Teye et al. 2017). Moreover, PIGN levels are altered by alternative splicing of another CE in endothelial cells in response to hypoxia (Weigand et al. 2012), pointing to a more widespread function of PIGN in hypoxia adaptation.

FIGURE 7.

MBNL2 controls the hypoxia-dependent alternative splicing of CE. (A) Venn diagram showing the overlap of AS changes after knockdown of MBNL2 and in response to hypoxia in MCF-7. Not significant (Fisher's exact test, one-tailed). The number in brackets refers to AS events regulated in the opposite direction. (B) Scatter plot comparing the fold changes of shared AS events after knockdown of MBNL2 and in response to hypoxia in MCF-7. Pearson correlation coefficient and associated P-value are given. (C) RT-PCR of PIGN exon 12 in normoxic and hypoxic MCF-7 cells and after knockdown of MBNL2. The short isoform (100 bp) corresponds to exon 2 exclusion, the long isoform (160 bp) to exon 12 inclusion. n = 2. (M) Size marker.

In sum, MBNL2 promotes the response of cancer cells to hypoxia by controlling the mRNA levels of HIF target genes. Thus, the expression of genes that determine hypoxia-driven angiogenic, metastatic and metabolic reprogramming of cancer cells is collectively regulated by MBNL2. In addition, MBNL2 contributes to hypoxia-dependent AS, in particular by reducing the inclusion of CE.

DISCUSSION

Hypoxia is an important modulator of tumorigenesis and metastasis. It is a common characteristic of a tumor's microenvironment and alters gene expression in affected cells. Cellular adaptation to hypoxia enables to sustain proliferation, escape apoptosis and initiate angiogenesis, ultimately promoting tumorigenesis. So far, proteins that are targeted in cancer therapies mostly belong to the HIF signaling pathway (for review, see Wilson and Hay 2011). However, while earlier studies focused mainly on transcriptional changes induced by HIF transcription factors, post-transcriptional regulation has recently been identified as a critical factor in the adaptation and survival of cancer cells during hypoxia (Oltean and Bates 2014; Gallego-Paez et al. 2017; Khabar 2017; Escobar-Hoyos et al. 2019; Robichaud et al. 2019). Such post-transcriptional regulation is controlled by RBPs, rendering them new potential targets for cancer therapeutics.

Comparing the hypoxic responses of breast and lung cancers, the most common cause of cancer-related deaths in males and females, respectively, we observed a specific increase in MBNL2 levels, while other splicing factors are generally reduced. MBNL proteins have just recently been shown to be potent regulators of tumor progression. MBNL1 contributes to colorectal carcinogenesis via interference with Dicer1 recruitment to miRNA precursors (Tang et al. 2015). MBNL2 is increased in clear cell renal cell carcinoma (ccRCC) and can act as an oncogenic driver (Perron et al. 2018). MBNL3 was found to promote hepatocarcinogenesis (Yuan et al. 2017). In contrast, MBNL1 and MBNL2 were also described to suppress metastasis (Fish et al. 2016; Tang et al. 2019; Zhang et al. 2019). For MBNL1 was proposed that these contrasting roles might depend on the levels of specific isoforms (Tabaglio et al. 2018).

The observed increase of MBNL2 in renal carcinoma is particularly interesting as ccRCC usually involves loss of the Von Hippel–Lindau tumor suppressor (VHL) protein (Meléndez-Rodríguez et al. 2018). Under normoxic conditions, VHL targets hydroxylated HIFα subunits for polyubiquitination and subsequent degradation by the proteasome (Gossage et al. 2015). Thus, the HIF pathway is constitutively activated and might be responsible for induction of MBNL2. In addition, elevated MBNL2 mRNA levels were detected in PyMT mouse breast cancer cells after chronic and intermittent hypoxia (Chen et al. 2018). Together with our study, these data suggest that MBNL2 activation in hypoxia is widespread in cancer and modulates tumorigenesis and metastasis.

Our study shows MBNL2-dependent VEGFA expression in breast and lung cancer cells. In accordance, VEGFA mRNA levels along with four other hypoxia response genes were strongly correlated with MBNL2 mRNA levels in ccRCC tumor samples (Perron et al. 2018). This suggests that MBNL2 universally controls the hypoxia response of various cancer types. Under hypoxia, VEGFA expression is induced by HIF1, resulting in neoangiogenesis, promoting proliferation and cell migration, and ultimately tumor growth (Yancopoulos 2010). Accordingly, it is therapeutically targeted using monoclonal antibodies and tyrosine kinase inhibitors in various cancers. In addition to the transcriptional control, VEGF-A levels are tightly controlled also at the post-transcriptional level (for review, see Arcondeguy et al. 2013). The translation of VEGF-A is regulated by two internal ribosome entry sites (IRESs), two alternative CUG initiation codons and a GAIT element (interferon-γ-activated translation inhibitor) in its 3′UTR (Yao et al. 2013; Arif et al. 2018). The stability of the VEGFA mRNA is controlled by several AU-rich elements (AREs) and a CA-rich element (CARE) that is bound by miRNAs under normoxia (Jafarifar et al. 2011). Under hypoxic conditions, nuclear hnRNP L is translocated into the cytoplasm, competes with miRNA binding and thus stabilizes the VEGFA mRNA. The three MBNL2 binding sites in the VEGFA 3′UTR proposed in this study do not overlap with the previously identified binding sites of RBPs or miRNAs. Thus, we have identified an additional, independent layer in VEGFA regulation by MBNL2.

Binding assays showed that MBNL2-40 is most likely to bind the VEGFA 3′UTR binding site 1 (Fig. 6A,B). Mechanistically, we could rule out a stabilizing effect of MBNL2 on the VEGFA mRNA in actinomycin D assays. Instead, we found that MBNL2 knockdown suppressed VEGF-A protein secretion. Accordingly, “SRP-dependent cotranslational targeting of proteins to the membrane” was the most significant GO term of reduced genes after MBNL2 knockdown in MCF-7 cells (see Supplemental Fig. 10). In line with our findings, MBNLs have been shown to regulate mRNA localization, particularly of secreted proteins (Du et al. 2010; Wang et al. 2012). An mRNA-stabilizing role of MBNL2 has not yet been experimentally confirmed. However, we cannot exclude the possibility that MBNL2-dependent stabilization of specific mRNAs contributes to its role in hypoxia adaptation. Interestingly, MBNL1-dependent destabilization of mRNA targets was shown (Masuda et al. 2012). MBNL2 induction in hypoxia could lead to displacement of MBNL1 and thus to mRNA stabilization of target mRNAs other than VEGFA.

Notably, the knockdown of MBNL1 showed an opposite effect to the knockdown of MBNL2 during hypoxia. Since all studies so far have found a broad overlap of the binding specificities of MBNL1 and MBNL2, it is difficult to rationalize how the function of the two proteins can be separated. One possibility might be that slight changes in affinity for certain mRNAs, as suggested by our binding affinity assay, could be sufficient to promote preferential association with one paralog in vivo. Further possibilities could be different cellular localization, different association with target mRNAs, or differential interaction with other proteins during hypoxia. Alternatively, the two paralogs could in principle both stimulate the translation of secreted proteins, but with different efficiencies. Thus, a change in the ratio of the two paralogs can lead to seemingly opposite effects, depending on whether the more active paralog is induced or reduced. Such a competitive mode of action has been demonstrated for MBNLs in splicing regulation (Taylor et al. 2018).

In sum, MBNL2 governs the expression of the therapeutic target VEGFA in various cancers, indicating a universal, cell type-independent control. Moreover, since its paralogs are not induced by hypoxia, MBNL2 plays a unique role in hypoxia adaptation, crucial for the survival of cancer cells. Remarkably, TCGA data set analyses revealed a predominant positive correlation of MBNL2 expression with hypoxia response genes in breast and lung cancer samples, suggesting that MBNL2 affects hypoxia adaptation also in cancer patients.

In addition to changes in transcript abundance, we found that MBNL2 also controls hypoxia-dependent changes in alternative splicing. Preferably, the inclusion of CE reduced under hypoxia was restored by the knockdown of MBNL2, suggesting a mostly repressive function of MBNL2 in the splicing response to hypoxia. This is not limited to hypoxia-regulated AS, as increased CE inclusion was the predominant splicing event observed in hypoxic MCF-7 cells after knockdown of MBNL2. In contrast, previous studies found equal numbers of induced and reduced CE after MBNL protein depletion (Du et al. 2010; Charizanis et al. 2012). It is still unclear how exactly MBNL proteins control alternative splicing. Thus, the observed suppressive effect of MBNL2 may be linked to the widespread changes observed within the splicing machinery in response to chronic hypoxia.

Widespread reduction in splicing-associated proteins has been demonstrated as a general stress response of cancer cells after different types of chemotherapy (Anufrieva et al. 2018). In addition, altered SR protein activity was reported in hypoxic prostate cancer (Bowler et al. 2018) and HeLa cells (Jakubauskiene et al. 2015). Together with the widespread reduction of hnRNPs identified here, we suggest global effects of hypoxia on the splicing machinery. Interestingly, this change does not lead to a global retention of introns, as previously observed in response to acute hypoxia (up to 24 h) (Brady et al. 2017; Han et al. 2017). Thus, short-term hypoxia could lead to a block in pre-mRNA splicing that is reminiscent of other stress responses, such as heat shock (Biamonti and Caceres 2009; Shalgi et al. 2014). During adaptation, this is then translated into a cell type-specific splicing program to enable continued cell survival under prolonged hypoxia.

In summary, the post-transcriptional response to hypoxia is a highly complex and multilayered process that superimposes and fine-tunes transcriptional regulation by HIF proteins. We took a step toward a deeper understanding of these processes and the crosstalk between HIFs and RBPs, which is necessary to open up new approaches in cancer therapy. To this end, we achieved a comparable data set of hypoxia-driven changes in RBP levels and alternative splicing in two unrelated cancer types. This approach enabled us to reveal a new layer in the regulation of hypoxia adaptation by MBNL2. We identified novel MBNL2 targets and demonstrated MBNL2 control of VEGFA expression at mRNA and protein level. This function in hypoxia adaptation is specific for MBNL2 and is not shared by its paralog MBNL1. Moreover, hypoxia regulation by MBNL2 seems to be universally functional in cancer cells. MBNL2, its targets and thus the complete MBNL2 signaling pathway have the potential to initialize new attempts in cancer research.

MATERIALS AND METHODS

Cell culture and chronic hypoxia treatment

A549 (DSMZ no. ACC-107) and MCF-7 cells (DSMZ no. ACC-115) were cultured in T75 flasks in DMEM (Sigma-Aldrich) or RPMI-1640 medium (Sigma-Aldrich), respectively, supplemented with 10% FBS (Thermo Fisher Scientific), 1 mM sodium pyruvate (Thermo Fisher Scientific) and Penicillin Streptomycin (Pen Strep, Thermo Fisher Scientific).

For hypoxia treatment, 1 × 105 A549 cells or 2 × 105 MCF-7 cells were seeded in 12-well plates and incubated under normoxic conditions (37°C, 5% CO2, 21% O2). Twenty-four hours after seeding, cells were exposed to hypoxia (37°C, 5% CO2, 0.5% O2, Eppendorf Galaxy 48 R incubator). RNA samples were prepared 48 h later. Incubation of the cells under hypoxia (0.5% oxygen) for 48 h mimics chronic hypoxia.

Transfection

For reverse transfection of siRNA, 1 × 105 A549 or 2 × 105 MCF-7 were transfected after the manufacturer's protocol using Lipofectamine RNAiMAX (Thermo Fisher Scientific). For subsequent RNA or protein isolation, cells were transfected in a 12-well format with 200 pmol of a nonsilencing control siRNA (siCTRL: 5′-UUCUCCGAACGUGUCACGU[dT][dT]-3′), an siRNA targeting MBNL2 (siMBNL2: 5′-CACCGUAACCGUUUGUAUG[dT][dT]-3′) (Paul et al. 2006) or an siRNA targeting MBNL1 (siMBNL1: 5′-AAGGACGAGGTCATTAGCCAT[dT][dT]-3′) (Qiagen). Samples were prepared or assays performed 48–72 h post-transfection.

RNA isolation

Total RNA from A549 and MCF-7 cells was isolated either using TRIzol reagent (Invitrogen), followed by TURBO DNase (Thermo Fisher Scientific) treatment or by using the miRNeasy Mini kit (Qiagen), including the optional on-column DNA digestion with the RNase-Free DNase Set (Qiagen). After isolation, 500 ng RNA were quality checked on a 1% agarose gel.

RT-PCR and RT-qPCR

RT-PCR analysis and RT-qPCR quantification was performed as in (Weigand et al. 2012). Oligonucleotide sequences are listed in Supplemental Table 11. All PCR products were verified by sequencing.

RNA-sequencing

Total RNA from normoxic (21% O2) and hypoxic (48 h, 0.5% O2) A549 and MCF-7 cells (n = 2) was used to generate rRNA-depleted (RiboZero, Illumina) strand-specific cDNA libraries (TruSeq stranded mRNA library preparation kit, Illumina). Libraries were prepared by the Core Unit Systems Medicine (University of Würzburg) according to the manufacturer's protocol omitting the poly(A) enrichment step. All libraries were sequenced on a NextSeq 500 (Illumina), obtaining 75 nt single-end reads. Approximately 100 million reads were obtained for each library.

For the MBNL2 knockdown experiments (n = 2), total RNA from hypoxic (48 h, 0.5% O2) MCF-7 cells transfected either with a nonsilencing control siRNA (siCTRL) or an siRNA targeting MBNL2 (siMBNL2) was used to generate poly(A)-enriched strand-specific cDNA libraries (mRNA selection, TruSeq stranded mRNA library preparation kit, Illumina). Libraries were prepared by the Core Unit Systems Medicine (University of Würzburg) according to the manufacturer's protocol. Libraries were sequenced on a NextSeq 500 (Illumina), obtaining 75 nt single-end reads (Core Unit Systems Medicine, University of Würzburg). Approximately 80 million reads were obtained for each library.

RNA-seq data analyses

Reads were mapped to the human genome (GRCh38/hg38 assembly) using the splice-aware aligner STAR version 2.4.5a (Dobin et al. 2013). The htseq-count script from the HTSeq python package version 0.6.1 (Anders et al. 2015) was used to count reads within genes annotated in GENCODE version 24. Differential expression analysis between hypoxic and normoxic conditions, as well as siMBNL2 and siCTRL, was performed using the R/Bioconductor package DESeq2 (Love et al. 2014). Differentially expressed genes were defined based on an adjusted P < 0.05 and an absolute fold change >1.5. Alternative splicing events were detected using rMATS version 3.2.5 (Shen et al. 2014), with default parameters, considering only reads mapped to splice junctions. A cutoff was adopted at a false discovery rate (FDR) < 0.05 and an absolute change in “percent spliced in” (|ΔPSI|) ≥ 10%. Further, we required at least 10 and five reads (average over replicates) supporting the junctions of inclusion and skipping events, respectively, and an average inclusion level ≥10% in either one or the other condition, to minimize false positives. For both differentially expressed genes and alternative splicing events, we kept only genes with a minimum expression level of >1 transcript per million (TPM) in at least one of the four samples obtained for each cell type/condition. Gene Ontology (GO) enrichment analysis was performed using the enrichGO function in clusterProfiler package (version 3.6.0) (Yu et al. 2012) in R (version 3.4.3) applying a cutoff of 0.05 for P-value and Q-value.

Gene expression correlation

Gene expression data from patients with breast invasive carcinoma (BRCA, 1088 patients) and lung adenocarcinoma (LUAD, 533 patients) of The Cancer Genome Atlas (TCGA) PanCancer Atlas (Hoadley et al. 2018) were downloaded from the Genomics Data Commons (GDC) Data Portal (https://portal.gdc.cancer.gov/) and cBioPortal (http://www.cbioportal.org/). In order to test for an association between MBNL2 mRNA levels and the expression of other genes, Pearson's product-moment correlation was calculated on log2-transformed fragments per kilobase of transcript per million mapped reads (FPKM) values. Hypoxia response genes were the 556 genes that were consistently induced in A549 and MCF-7 cells in response to hypoxia (Fig. 1D). Analyses were performed for all genes with a mean expression of FPKM > 1 over all samples per cohort, that is, 14,922 genes in BRCA and 15,590 genes in LUAD, including 455 and 469 hypoxia response genes, respectively. Using two different cutoffs, we detected 6814 genes with a significant Pearson correlation coefficient (PCC) > 0.15 and 2174 genes with PCC > 0.3 in the BRCA data set as well as 5433 genes with PCC > 0.15 and 3150 genes with PCC > 0.3 in the LUAD data set (P < 10 × 10−6, Benjamini–Hochberg correction).

Fluorescence detection and Western blot analysis

For Western blot analyses, A549 and MCF-7 cells were lysed in lysis buffer [137 mM NaCl, 10% glycerol, 20 mM Tris-HCl pH 8.0, 2 mM EDTA pH 8.0, 1% Igepal, 5 µL protease inhibitor cocktail (Sigma-Aldrich)] for 20 min on ice. After centrifugation (15 min at 17,000g, 4°C) the protein content of the samples was determined in three technical replicates according to the Bradford method. A total of 10–20 µg protein extract was loaded onto 10% SDS polyacrylamide gels and blotted onto PVDF membranes (Bio-Rad). Before blotting, total lane protein (loading control) was visualized using 2,2,2-trichloroethanol (Ladner et al. 2004). Membranes were blocked with 2% skim milk. MBNL1 antibodies MB1a(4A8) (Holt et al. 2007) was a kind gift from Wolfson Centre for Inherited Neuromuscular Disease (CIND). Horseradish peroxidase-conjugated anti-mouse or anti-rabbit IgG (Jackson ImmunoResearch) were used as secondary antibodies. Blots were developed with the ECL system (Bio-Rad) or Amersham ECL select (GE Healthcare) for weaker signals. Images were detected using the ChemiDoc Imaging System (Bio-Rad). Membranes were reused for a maximum of three times.

For MBNL2 splicing assays, protein lysate was prepared with MEF 12KO cells harvested after 48 h post-transfection with 750 ng of GFP-MBNL1-42, GFP-MBNL2-40, or GFP expressing vector and 200 ng of VEGFA BS3 minigene. The cells were resuspended in 1× RIPA buffer supplemented with Halt Protease Inhibitor Cocktail (Thermo Fisher) and incubated on ice with vortexing at 10 min interval. The lysate was then incubated at −80°C O/N followed by 20 min centrifugation at 4°C and 15,000g. The concentration of proteins was measured with BCA protein assay kit (Abcam). A total of 30 µg of proteins was mixed with loading buffer but not denatured at high temperature and separated on 10% SDS polyacrylamide gels. GFP-MBNL fusion proteins and control GFP were detected at 488 nm excitation, protein marker was visualized at 635 nm excitation at Amersham Typhoon RGB laser scanner. Subsequently, the samples were transferred to a nitrocellulose membrane (Protran BA 85, Whatman) using a wet transfer apparatus (1 h, 100V, 4°C). The membrane was blocked for 1 h with 5% skim milk in PBST buffer [phosphate-buffered saline (PBS), 0.1% Tween-20] and incubated with a primary antibody against α-Tubulin (1:2000, Abcam, ab52866). Anti-rabbit (1:2500, Agrisera, AS09602) secondary antibody was conjugated with horseradish peroxidase and detected using Immobilon Forte Western HRP substrate (Millipore).

Antibodies are listed in Supplemental Table 12.

VEGF-A detection by enzyme-linked immunosorbent assay (ELISA)

Human VEGF-A ELISA Kit (Sigma-Aldrich) was used to quantify secreted human VEGF-A protein in cell culture supernatants from normoxic and hypoxic (48 h, 0.5% O2) A549 and MCF-7 cells transfected either with a nonsilencing control siRNA (siCTRL) or an siRNA targeting MBNL2 (siMBNL2). Cell culture supernatants were diluted 1:2 in Sample Diluent Buffer B and protein detection was performed according to the manufacturer's protocol.

Cell viability assay

A549 and MCF-7 cells were transfected as described above and incubated under normoxic or hypoxic conditions (48 h, 0.5% O2). Twenty-four hours later, cisplatin was added at final concentrations of 10 or 20 µM. After another 24 h under normoxia/hypoxia the cells were fixed with 0.5% formaldehyde in PBS and stained with 0.5% crystal violet in PBS. After three washing steps with PBS the cells were incubated with 33% acetic acid. Samples were transferred to a 96 well plate. Absorption at 570 nm was measured in a TECAN infinite M 200 Pro plate reader. Absorption was normalized to normoxic control cells with the respective cisplatin concentration.

Transwell migration assay

A549 cells were transfected as described above. 24 h later, cells were serum-starved (0% FBS) and exposed to hypoxia (48 h, 0.5% O2). Cells were washed in PBS and harvested using 40 µL Trypsin-EDTA (0.05% trypsin, 0.02% EDTA in PBS). The harvested cells were collected and resuspended in starvation medium. We used 24-well transwell chambers (Thincert cell culture inserts, pore diameter 8 µm, translucent PET membrane, Greiner). A total of 5 × 104 cells in 100 µL starvation medium were seeded per insert, which was placed into a lower well containing 700 µL DMEM medium supplemented with 10% FBS (Thermo Fisher Scientific), 1 mM sodium pyruvate (Thermo Fisher Scientific) and Pen Strep (Thermo Fisher Scientific) without or with 100 ng/µL hEGF as a positive control or starvation medium (0% FBS) as a negative control. After 5 h under hypoxic conditions, the experiment was stopped. Cells from the insert were removed with a cotton swab. Migrated cells were washed with PBS and fixed with methanol. Migrated cells were stained with crystal violet (0.5% in PBS) and the membranes were mounted using Pertex (Histolab). Quantification of migrated cells was done by counting six fields per well from random sites of the transwell. No cells were detectable in the negative control.

mRNA decay assay

For mRNA decay experiments, MCF-7 cells transfected with either a nonsilencing siRNA (siCTRL) or an siRNA targeting MBNL2 (siMBNL2) were incubated under hypoxic conditions (48 h, 0.5% O2). After 48 h cells were treated with 5 µg/mL actinomycin D (Sigma) for 30 min. Total RNA was extracted at the indicated time points and mRNA levels were quantified by RT-qPCR. MRNA levels were normalized to RPLP0 mRNA levels and plotted against time.

Splicing minigene assays

VEGFA splicing minigenes based on the ATP2A1 minigene were prepared as previously described (Taylor et al. 2018). Briefly, three sequences derived from the VEGFA 3′UTR, carrying potential MBNL binding sites were obtained in two PCR steps introducing NotI and SalI restriction sites.

Mouse Embryonic Fibroblasts (MEFs) with MBNL1 and MBNL2 double knockout (12KO) were a gift from Maury Swanson (Lee et al. 2013). The growth conditions of MEFs 12KO were previously described (Cywoniuk et al. 2017). Prior to transfection, the cells were plated in 12-well plates and transfected at 50%–60% confluence with Lipofectamine 3000 (Invitrogen) according to the manufacturer's protocol. Cotransfection was conducted with 200 ng of a minigene and 750 ng of GFP-MBNL1-42, GFP-MBNL2-40, or GFP expression vectors. GFP-MBNL constructs were previously described (Sznajder et al. 2016; Konieczny et al. 2017). Cells were harvested 42 h after transfection.

Total RNA isolation was conducted using TRI reagent (Sigma-Aldrich) following the manufacturer's protocol. cDNA was synthesized using the SuperScript III Reverse Transcriptase kit (Invitrogen). Splicing analysis was performed with GoTaq DNA polymerase (Promega). Oligonucleotide sequences are listed in Supplemental Table 11.

Preparation of radiolabeled RNA fragments

The templates for transcription reactions for VEGFA RNA fragments derived from pre-mRNAs were obtained in PCR using specific oligonucleotide sets (_TF/TR) having a promoter for polymerase T7. The PCR products were then purified prior to transcription reaction according to the manufacturer's protocol (A&A Biotechnology). The transcription reaction and 5′-end radiolabeling were performed as previously described with slight changes (Sznajder et al. 2016). Briefly, transcription reaction was performed in 40 µL composed of 10 µL of DNA template, 1 mM NTPs (Invitrogen), 3 mM guanosine (Sigma-Aldrich), 20 U T7 RNA Polymerase (Promega), 1× T7 transcription buffer (Promega), 40 U Rnasin Plus RNase Inhibitor (Promega). Purification of transcript was conducted on a denaturing 6% polyacrylamide gel (19:1 acrylamid:bisacrylamid) and through ethanol precipitation. For radio-labeling, 2 pmol of transcript was incubated with 2 pmol of [γ-32P] ATP, 1 U Rnasin Plus, 10 U OptiKinase (Affymetrix), 1× reaction buffer (Affymetrix) and ddH20 up to 10 µL, at 37°C for 30 min. Labeled RNA was subsequently run on a denaturing 8% polyacrylamide gel (19:1) in 0.5× TBE, at 100 V for 1 h. A band of RNA was visualized on IP through FLA-5100 (FujiFilm), cut out followed by ethanol precipitation and resuspended in 20 µL ddH2O. (CUG)20 was a kind gift from Włodzimierz J. Krzyżosiak.

Recombinant GST and (His)6-tagged MBNL1 and MBNL2

GST fusions with MBNL1 and MBNL2 proteins encoded by exons 1-4 of the MBNL1 and MBNL2 with a (His)6-tag at the carboxyl terminus were prepared and purified as previously described (Sznajder et al. 2016). The protein concentration was measured using both Bradford assay and Sypro Ruby staining using 10% SDS-PAGE.

Quantification of MBNL–RNA interaction

A biochemical assay based on double-membrane filtration analysis was performed as previously described with slight changes (Sznajder et al. 2016). Briefly, the reaction was performed in a 30 µL volume. 5′-labeled transcripts (0.05 nM) were first subjected to a denaturation-renaturation step. Subsequently, the RNAs were incubated with the indicated concentrations of MBNL1 and MBNL2 (ranging from 0 to 1250 nM) in buffer B containing 250 mM NaCl, 15 mM KCl, 50 mM, Tris-HCl pH 8.0, 0.05% Tween-20, 1 mM MgCl2 at 37°C for 15 min. An amount of 25 µL samples were loaded onto a filter binding apparatus with nitrocellulose (Protran BA 85, Whatman) and nylon (Hybond N+, Amersham) membranes previously wetted in buffer B. The signal from membranes was visualized after IP through a FLA-5100 and quantified using the Multi Gauge software (FujiFilm). The KD values of the MBNL–RNA complexes were calculated based on the signal of free RNA and MBNL–RNA complexes from three experimental replicates with GraphPad Prism using the one site specific binding curve equation

Statistical analysis

All bar graphs are reported as mean values ± standard deviation. Statistical analysis was done using Student's t-test, two-tailed, paired, or Fisher's exact test, one-tailed, as indicated. Pearson's product-moment correlation was used to estimate the relationships between variables shown in scatter plots by using the cor.test function in R (version 3.4.3). The comparative enrichment of hypoxia-regulated AS events was tested with a two-sample test for equality of proportions, by using the prop.test function in R (version 3.4.3).

DATA DEPOSITION

The RNA-seq data of A549, MCF-7, and siMBNL2 experiments have been deposited with the Gene Expression Omnibus database (GEO; www.ncbi.nlm.nih.gov/geo/) and are available under accession numbers GSE131378 and GSE136231.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.