Integrin α3β1 controls mRNA splicing that determines Cox-2 mRNA stability in breast cancer cells Free

Integrins are heterodimeric receptors on the cell surface that mediate adhesion to the extracellular matrix (ECM) (Hynes, 2002). Integrins expressed on tumor cells are involved in both inside-out and outside-in signaling events that regulate a variety of cell functions that promote cancer progression and metastasis, and several integrins are currently being exploited in the clinic as important targets for therapeutic intervention (Desgrosellier and Cheresh, 2010; Goodman and Picard, 2012). Indeed, in their roles as bidirectional signaling receptors at the interface of the tumor cell and its microenvironment, integrins regulate tumor-cell-mediated changes to the microenvironment that facilitate cancer progression, as well as tumor cell responses to such changes. Integrins can control gene expression that drives cancer progression and metastasis, and early studies have shown that integrin-dependent changes in transcription are a crucial point of gene regulation in both normal and tumor cells (Rossetti et al., 2002; Shi et al., 2004; Xiao et al., 1998). By contrast, regulatory roles for integrins in post-transcriptional gene regulation remain underexplored. This is an important question as it has become increasingly clear in recent years that post-transcriptional regulation of mRNA processing, through alternative splicing or generation of 3′-untranslated region (3′-UTR) variants, can have profound effects on the expression of genes involved in tumor growth, invasion, and metastasis (Dutertre et al., 2010; Lapuk et al., 2010). As these pathways involve components of the splicing or polyadenylation machinery that target multiple mRNA transcripts, it is likely that this regulation is important for controlling groups of genes that collectively drive tumorigenesis and cancer progression.

Post-transcriptional gene regulation occurs, in large part, through incorporation of mRNA sequences that confer susceptibility to pathways controlling mRNA turnover. For example, variations within the 3′-UTR can alter the content of AU-rich elements (AREs) that target mRNA for rapid decay through interactions with RNA-binding proteins (RBPs) that regulate mRNA stability (Danckwardt et al., 2008; Di Giammartino et al., 2011; Khabar, 2010). Similarly, alternative splicing or polyadenylation can lead to incorporation of sequences that render a transcript susceptible to nonsense-mediated mRNA decay (NMD), a surveillance mechanism that targets aberrant mRNAs containing premature termination codons (PTCs), as well as mRNAs with extended 3′-UTRs, for enhanced turnover (for reviews, see Hwang and Maquat, 2011; Nicholson et al., 2010). NMD is important for the elimination of mutant transcripts containing PTCs, including those responsible for many genetic disorders. However, in recent years, NMD has also been recognized as an important mechanism of post-transcriptional gene regulation in both normal and pathological processes (Hwang and Maquat, 2011; Nicholson et al., 2010). Indeed, NMD targets include tumor-promoting mRNAs, and suppression of NMD in tumor cells enhances resistance to microenvironmental stresses and promotes tumorigenesis (Gardner, 2010). NMD requires up-frameshift protein 1 (UPF1), an ATP-binding RNA helicase that functions as part of a multi-factor complex to recognize aberrant translation-termination which is triggered by a ribosome stalling at a PTC (Hwang and Maquat, 2011; Nicholson et al., 2010) or at a normal termination codon that precedes a lengthy 3′-UTR (Hogg and Goff, 2010). Although malignant progression has been associated with changes in both ARE-mediated mRNA stability and NMD (Gardner, 2010; Khabar, 2010), the mechanisms whereby signals from the extracellular microenvironment are transduced into tumor cells to control post-transcriptional gene regulation, including NMD, remain unclear.

The laminin-binding integrin α3β1 (hereafter referred to as α3β1) is expressed in a variety of epithelial cell types, and expression of α3β1 is enhanced or correlated with progression in many different cancers, including squamous cell carcinoma (SCC), prostate cancer and breast cancer (Janes and Watt, 2006; Morini et al., 2000; Pontes-Júnior et al., 2010). Preclinical studies have shown that α3β1 promotes malignant behavior of breast cancer cells in vivo and in vitro (Cagnet et al., 2013; Mitchell et al., 2010; Morini et al., 2000; Scales et al., 2013; Sugiura and Berditchevski, 1999; Wang et al., 2004), as well as skin tumorigenesis in vivo (Sachs et al., 2012), implicating this integrin as a potential therapeutic target to inhibit cancer progression and metastasis (Subbaram and DiPersio, 2011). Laminin-332, a major ECM ligand for α3β1, is often expressed highly in breast cancer cells, where it enhances motility (Carpenter et al., 2009; Carpenter et al., 2008). At least some functions of α3β1 in transformed or immortalized cells are attributable to the ability of α3β1 to regulate genes that stimulate invasive growth and/or angiogenesis, including matrix metalloproteinase-9 (MMP-9) (Iyer et al., 2005; Morini et al., 2000) and cyclooxygenase (Cox-2, also known as prostaglandin G/H synthase 2 in human) (Mitchell et al., 2010). In immortalized keratinocytes, α3β1-dependent-induction of MMP-9 occurs through a post-transcriptional mechanism of enhanced mRNA stability (Iyer et al., 2005).

In the current study, we performed genome-wide microarrays to identify a variety of genes that are regulated by α3β1 in MDA-MB-231 breast cancer cells. Using a short hairpin RNA (shRNA) to knock down the α3 integrin subunit, we showed that suppression of α3β1 modulates a number of genes that control several tumor-promoting pathways. Interestingly, the mRNA transcripts for many of these genes were differentially spliced, or otherwise alternatively processed, in α3β1-deficient cells. In the case of Cox-2 mRNA, we detected alternative exon usage (AEU) within the open reading frame and the 3′-UTR, including retention of an intron between exons 7 and 8 (intron 7) that has been shown previously to confer susceptibility to NMD (Gehring et al., 2005). Consistently, we showed that Cox-2 mRNA was more stable in α3β1-expressing cells and that siRNA-mediated knockdown of UPF1, or treatment with Ataluren (a nonsense suppressing agent), caused accumulation of Cox-2 mRNA in α3β1-deficient cells. RNA-mediated suppression of α3β1 similarly reduced Cox-2 mRNA levels and stimulated intron 7 inclusion in MCF7 cells, indicating that this regulation extends to other breast cancer cells. Our findings show that suppression of integrin α3β1 in breast cancer cells leads to reduced Cox-2 expression by promoting NMD-mediated turnover of Cox-2 mRNA, in part through the generation of NMD-susceptible Cox-2 mRNA variants. Our microarray analysis indicates that α3β1-mediated generation of alternative transcripts extends to many other genes, indicating a generally important role for this integrin in controlling post-transcriptional mRNA processing and stability.

We previously showed that stable, shRNA-mediated suppression of α3β1 in MDA-MB-231 breast cancer cells reduced expression of the Cox-2 gene and abrogated tumor growth in an orthotopic model (Mitchell et al., 2010). To determine whether acute suppression of α3β1 similarly reduces Cox-2 expression, and whether this regulation extends to other breast cancer lines, we transfected MDA-MB-231 cells or MCF7 cells each with two different siRNAs that target the α3 integrin subunit (hereafter referred to as α3). A luciferase-targeting control siRNA did not substantially alter α3 expression; however each siRNA targeting α3 significantly reduced α3 levels, whether assessed at the level of mRNA (Fig. 1A) or protein (Fig. 1C). Notably, flow cytometry revealed that treatment with α3-targeting siRNA efficiently reduced α3β1 expression on the cell surface in more than 50% of the MDA-MB-231 cell population, whereas the remaining proportion retained high α3β1 surface expression (data not shown; supplementary material Fig. S1). Despite this partial knockdown, in both cell lines, Cox-2 mRNA expression was decreased by α3-targeting siRNA compared with control siRNA (Fig. 1A,B, MDA-MB-231 cells; Fig. 1C,D, MCF7 cells; small differences between control-transfected cells and untreated cells were not statistically significant).

Previous studies have suggested that some α3β1 signaling functions might be independent of laminin binding (Zhang et al., 2003; Zhang et al., 1999). Therefore, we took two approaches to ask whether perturbing binding of α3β1 to laminin alters Cox-2 mRNA expression in MDA-MB-231 cells grown on a laminin-332-rich matrix. In the first approach, we suppressed endogenous α3 using the siRNA that targets the 3′-UTR of human α3 mRNA (siα3-UTR; Fig. 1A), then infected these cells with adenovirus that expresses mRNA encoding either wild-type α3 (α3wt) or a previously described point mutant of α3 (α3G163A) that does not bind to laminin-332 (Zhang et al., 1999). Importantly, these exogenous mRNAs lacked the 3′-UTR target sequences for siα3-UTR. Efficient suppression of endogenous α3, and subsequent restoration with each exogenous α3 variant, was confirmed by RT-PCR, western blot and flow cytometry (Fig. 2A, and supplementary material Fig. S1). Interestingly, RT-PCR showed that, although Cox-2 mRNA was significantly reduced in cells treated with α3-targeting siRNA, it was restored by transduction with either wild-type or mutant α3 to levels that were not significantly different from control cells (Fig. 2B). In the second approach, we observed that treatment of cells plated on laminin-332-rich matrix with the monoclonal antibody P1B5, which inhibits α3β1 binding to laminin (Symington and Carter, 1995), did not alter Cox-2 mRNA levels compared with treatment with control IgG (Fig. 2C). Thus, treatment of cells with agents known to perturb α3β1 binding to laminin does not appreciably alter Cox-2 mRNA expression, indicating that binding to laminin is not essential for this regulation. However, we cannot rule out the possibility that this signaling can also occur from sites of α3β1-mediated adhesion to its laminin ligands.

To determine whether the effects of suppressing α3β1 extend to other genes, we assessed global gene expression in MDA-MB-231 cells that were stably transduced with lentivirus encoding either control shRNA or shRNA targeting α3. As described previously, MDA-MB-231 cells transduced with control shRNA express abundant α3β1 on the cell surface, whereas those transduced with α3-targeting shRNA [α3-knockdown (KD) cells] show greatly reduced levels of α3β1, accompanied by reduced Cox-2 mRNA expression (Mitchell et al., 2010). For the current experiments, we utilized cells transduced with the more effective of two different α3-targeting shRNAs (i.e. shRNA#13; see supplementary material Fig. S2A), both of which suppressed α3 and reduced Cox-2 expression (supplementary material Fig. S2B). RNA was isolated from cells expressing either the control shRNA or the shRNA targeting α3, and microarray analysis was performed using the Affymetrix GeneChip® Human Exon 1.0 ST Array platform. Heat-map analysis of gene expression in α3-KD cells revealed distinct gene clusters that showed differential expression, either up or down, relative to control cells (Fig. 3A). In total, 1285 genes were differentially expressed (designated as DE) at least 1.5-fold upon suppression of α3β1, of which 270 genes were upregulated and 1015 genes were downregulated. Importantly, the Exon 1.0 ST Array platform contains multiple probes spanning each exon of all genes in the human genome, which allowed us to also identify α3β1-dependent inclusion of alternative mRNA sequences caused by differential splicing or other transcript processing mechanisms. We identified 147 genes that showed AEU in α3-KD cells, which we define as inclusion or exclusion of sequences within at least one exon (Fig. 3B). We found examples of genes with differences in all regions of the mRNA transcript, which were presumably due to differential splicing (i.e. within internal exons), alternative polyadenylation signals (i.e. within the 3′-UTR), or alternative transcription start sites (i.e. within the 5′-UTR). Of these genes, 96 also appeared on the list of DE genes (Fig. 3B), suggesting that a subset of AEU events might be associated with differences in transcription or mRNA stability.

Ingenuity® Pathway Analysis (IPA®) revealed that α3β1-dependent AEU and DE genes could be grouped according to their roles in a wide range of cellular functions (Fig. 3C). A combination of IPA and Qiagen pathway analysis (www.qiagen.com) revealed that genes showing α3β1-dependent AEU or DE fell into a number of pathways and implicated Cox-2 (gene symbol PTGS2) as an α3β1-dependent gene that is involved in several pathways (supplementary material Table S1). Cox-2 was the most significantly downregulated gene on the DE list (∼70-fold reduction in α3-KD cells), consistent with our previous report (Mitchell et al., 2010). Of note, Cox-2 also appeared on the list of α3β1-dependent AEU genes, suggesting that alternative mRNA processing might be a mechanism through which α3β1 regulates Cox-2 mRNA levels.

Building on our findings that integrin α3β1 promotes Cox-2 gene expression in MDA-MB-231 cells, we accessed the Oncomine^TM database to assess the relationship between expression of the integrin α3 gene and the Cox-2 gene in clinical breast cancer samples. Our analysis revealed that expression of the Cox-2 gene (PTGS2) showed positive correlation with that of the α3 gene (ITGA3) among samples within two different breast cancer cohorts (supplementary material Fig. S3A,B). Cox-2 and α3 both showed similar expression trends across different stages of breast carcinomas in similar cohorts (supplementary material Fig. S3C–F).

At least three splice variants of the Cox-2 mRNA transcript have been reported, as shown schematically in Fig. 4A (adapted from the UCSC genome browser). Little is known about the relative abundance and tissue distributions of these splice variants; however, they exhibit differences within the coding region, as well as in the 5′- and 3′-UTRs of the mRNA transcript (Fig. 4A). To compare the sequence content of Cox-2 mRNA in control cells with that of α3-KD cells, we used GeneSpring software to evaluate microarray data generated from probe sets specific for individual Cox-2 exons. Although Cox-2 mRNA levels were substantially reduced in α3β1-deficient cells (see above), we observed increased inclusion of internal sequences, as well as of sequences at both the 5′ and 3′ ends of the transcript (Fig. 4A). The 5′-end differences presumably reflect an alternative transcription start site. The 3′-UTR of Cox-2 mRNA, which is encoded within exon 10, harbors multiple AREs that are known to confer instability to Cox-2 mRNA upon binding to RNA-binding proteins that destabilize transcripts (Harper and Tyson-Capper, 2008). Interestingly, it has been reported that Cox-2 mRNA is also a target of NMD, which could be partly attributed to the presence of an extended 3′-UTR (Singh et al., 2008). We also observed that an intronic sequence between exons 7 and 8 was over-represented in Cox-2 mRNA of α3-KD cells compared with control cells (Fig. 4A, red bracket), indicating an alternative splicing event that retains the intron between exons 7 and 8 (intron 7) when α3β1 is suppressed. Retention of this 282-base-pair intron was previously reported to occur in Cox-2 mRNA isolated from the myometrium from women undergoing labor (Huang et al., 2003), although a function for this splice variant has not been described. Interestingly, our analysis of this intronic sequence revealed numerous in-frame PTCs (Fig. 4B), suggesting that it might confer instability to the Cox-2 mRNA by means of NMD.

To validate α3β1-dependent splicing of intron 7 from the Cox-2 mRNA, we performed real-time reverse transcription (RT)-PCR to amplify select regions of the Cox-2 mRNA transcript and assess their relative inclusion in control or α3-KD cells. The primer slope and efficiency analysis for each set of real-time PCR primers ensured that all primers had similar efficiency in the detection of differential intron or exon inclusion (supplementary material Table S2). Our microarray analysis indicated that exon 9 was among those exons that were represented within the Cox-2 mRNA to a similar extent in both control and α3-KD cells (Fig. 4A). Therefore, we normalized PCR signals for individual exons and introns to that for exon 9 to correct for the overall reduced levels of Cox-2 mRNA in the latter cells, consistent with normalization methods used in other studies of alternatively spliced genes (Kurokawa et al., 2010). This analysis confirmed that α3-KD cells showed a statistically significant increase in the retention of intron 7 compared with control cells (Fig. 5B). By contrast, exon 5 showed similar inclusion in control or α3-KD cells (Fig. 5A), consistent with the microarray results indicating that it was similarly represented in the Cox-2 mRNA from control and α3-KD cells (Fig. 4A). To control for off-target effects, we also observed retention of intron 7 in a distinct line of α3-KD MDA-MB-231 cells that were stably transduced with a different α3-targeting shRNA (supplementary material Fig. S2C). Moreover, we obtained similar results in both of the α3-KD lines when signals were normalized to exon 5 (data not shown). We were able to independently confirm that intron 7 was included in the Cox-2 mRNA from α3-KD cells by performing RT-PCR overamplification using primers from within exons 7 and 8 that span the intron, which detected a larger PCR product corresponding to the retained intron (Fig. 5E). This result was further corroborated by RNase protection assays, which showed that a probe spanning the junction of exon 7 and intron 7 was fully protected by RNA from α3-KD cells, indicating intron inclusion (supplementary material Fig. S4; bands corresponding to partial probe protection were also detected, most likely indicating NMD degradation products). By contrast, only the portion of the probe corresponding to exon 7 was protected by RNA from control cells (supplementary material Fig. S4), indicating Cox-2 mRNA that lacks intron 7. Interestingly, we also observed significantly increased inclusion in α3-KD cells of a sequence from within exon 10 (i.e. sub-region 10−6; Fig. 5D), possibly indicating alternative use of 3′-UTR sequences that encompass a number of AREs known to destabilize the Cox-2 mRNA (Harper and Tyson-Capper, 2008).

Taken together, the above findings indicate that suppression of α3β1 leads to the generation of Cox-2 mRNA variant(s) that could be susceptible to NMD or other mechanisms of mRNA degradation. As already discussed, inhibition of NMD can promote tumorigenesis (Gardner, 2010; Wang et al., 2011), and previous work from our lab showed that suppression of α3β1 in MDA-MB-231 cells decreases tumorigenesis in vivo. To assess the potential involvement of NMD in α3β1-mediated regulation of Cox-2 mRNA, we first analyzed control and α3-KD cells for phosphorylation status of the translation initiation factor eIF2α, which, when phosphorylated, inhibits NMD and leads to enhanced stability of mRNA transcripts that promote tumorigenesis (Wang et al., 2011). Levels of phosphorylated eIF2α were decreased in α3-KD cells compared with control cells (Fig. 6A), suggesting that the NMD pathway is enhanced when α3β1 is suppressed. It is noteworthy that this level of regulation is in addition to the α3β1-dependent generation of Cox-2 mRNA variants that contain NMD-target sequences (Fig. 4), which suggests that α3β1 might regulate mRNA turnover both by enhancing NMD itself and by controlling the generation of mRNA targets for this pathway. Interestingly, several other genes identified from our microarray analysis as being downregulated upon α3 suppression are known targets of NMD (supplementary material Table S3) (Mendell et al., 2004).

Because α3β1-deficient cells showed increased inclusion of sequences within the Cox-2 mRNA that might target it for NMD, we compared Cox-2 mRNA stability in control and α3-KD cells using an mRNA decay assay that we described previously (Iyer et al., 2005). For these experiments, we cultured cells in serum-free conditions, under which Cox-2 mRNA is relatively stable in MDA-MB-231 cells (Jang et al., 2000), which allowed us to focus on the effects of suppressing α3β1 on Cox-2 mRNA stability. These results revealed a higher rate of Cox-2 mRNA turnover in α3-KD cells than in control cells (Fig. 6B), in contrast with mRNAs for both β-actin (Fig. 6C) and GAPDH (Fig. 6D) which showed comparable turnover in control and α3-KD cells (a representative gel is shown in supplementary material Fig. S5). These results indicate that Cox-2 mRNA is less stable in α3-KD cells, possibly owing to increased susceptibility to NMD.

As mentioned previously, NMD has emerged as an important mechanism for regulating the expression of genes that drive tumor progression (Wang et al., 2011), but roles for integrins in this regulation are not known. Normal processing of mammalian pre-mRNAs involves formation of spliceosomes that mediate the removal of introns. If PTCs are present in an mRNA transcript owing to retention of an intron, or as a result of mutation, then the translational machinery pauses at the PTC and UPF1 is recruited as part of a NMD complex that targets the mRNA for degradation (Gardner, 2010; Lykke-Andersen et al., 2000). Thus, suppression of UPF1 effectively results in the accumulation of NMD target mRNAs (Wang et al., 2011). We reasoned that if suppression of α3β1 leads to reduced Cox-2 mRNA through a NMD pathway, then inhibition of UPF1 in α3-KD cells should lead to the accumulation of Cox-2 mRNA. To test this hypothesis, we treated control or α3-KD MDA-MB-231 cells with siRNA against UPF1, then assessed total levels of Cox-2 mRNA by RT-PCR. Western blot confirmed that treatment of cells with UPF1-targeting siRNA led to reduced levels of UPF1 protein, compared with treatment with the control siRNA against luciferase (Fig. 7A). Suppression of UPF1 in α3-KD cells led to a statistically significant increase in steady-state levels of Cox-2 mRNA over those seen in cells treated with control siRNA; however, GAPDH mRNA levels were not affected (Fig. 7B,C). However, we did not observe a statistically significant increase in Cox-2 mRNA in α3β1-expressing cells treated with UPF1 siRNA (Fig. 7B,C), indicating that the Cox-2 mRNA transcript was not susceptible to NMD in these cells. Enhanced NMD of Cox-2 mRNA in α3-KD cells was further supported by treating these cells with Ataluren (previously known as PTC124), a small-molecule agent that potently suppresses nonsense mutations by promoting ribosomal read-through of PTCs (Welch et al., 2007). Indeed, treatment of α3-KD cells with 5 µm Ataluren led to an accumulation of Cox-2 mRNA that was statistically significant by 48 hours (Fig. 7D,E). A similar accumulation of Cox-2 mRNA was seen with increasing dose of Ataluren in the α3-KD MDA-MB-231 cells generated with a distinct α3-targeting shRNA (supplementary material Fig. S2D). As predicted, RT-PCR using the intron-7-spanning primers detected the retained intron more readily in Ataluren-treated α3-KD cells than in control-treated α3-KD cells (Fig. 7F). Taken together, these results demonstrate an important role for NMD in the reduction of Cox-2 mRNA that occurs when α3β1 is suppressed in breast cancer cells.

There are many published examples where integrin-mediated adhesion to ECM regulates gene expression through changes in transcription. By contrast, roles for integrins in the post-transcriptional regulation of mRNA processing and stability are underexplored. In the current study, we have identified a novel role for the integrin α3β1 in regulating gene expression programs in breast cancer cells through post-transcriptional mechanisms of alternative exon usage, which represents a novel paradigm of integrin-mediated gene regulation. In the case of Cox-2, we demonstrate that suppression of α3β1 leads to retention of an intron in the Cox-2 mRNA transcript that is predicted to confer susceptibility to NMD through incorporation of PTCs (Fig. 8). These findings identify α3β1 as a potential therapeutic target to promote NMD-mediated suppression of Cox-2 mRNA, and possibly of other cancer-promoting mRNAs, in breast cancer cells.

It has long been known that NMD is an important constitutive pathway for rapid degradation of mutant transcripts, often associated with human diseases, that contain nonsense mutations (Bhuvanagiri et al., 2010). However, in recent years, NMD has emerged as an important mechanism for regulating nonmutant transcripts in response to a variety of cellular stresses, and it is becoming clear that suppression of NMD regulates genes that contribute to the ability of the cell to adapt to such stresses (Gardner, 2010). In particular, suppression of NMD in cancer cells appears to be important for the regulation of many genes that contribute to tumor progression and metastasis (Gardner, 2008; Gardner, 2010). Nevertheless, outside-in signaling that controls NMD in tumor cells, and the microenvironmental cues that trigger this signaling, have not been investigated extensively. Consistently, our current findings support a novel role for the integrin α3β1 as a suppressor of NMD in breast cancer cells because shRNA-mediated suppression of α3β1 led to decreased phosphorylation of eIF2α, indicative of enhanced NMD (Wang et al., 2011). Furthermore, some of the genes that were downregulated in α3β1-deficient cells (supplementary material Table S3) are known targets of UPF1-dependent NMD (Mendell et al., 2004), suggesting that α3β1 might inhibit NMD of multiple genes. In addition, suppression of α3β1 led to AEU events that are predicted to render mRNAs vulnerable to NMD. The latter regulation was exemplified by retention in the Cox-2 mRNA of a PTC-containing intron between exons 7 and 8, and possibly by AEU within the 3′-UTR encoded by exon 10 (Fig. 4), either of which could contribute to NMD-mediated Cox-2 mRNA turnover.

In addition to targeting mRNAs with PTC mutations that occur within the coding region, NMD can also target mRNAs with retained introns or extended 3′-UTRs (Hwang and Maquat, 2011; Nicholson et al., 2010; Roy and Irimia, 2008). As mentioned above, we also observed AEU in the 3′-UTR of the Cox-2 mRNA, where sub-regions within exon 10 showed altered inclusion when α3β1 was suppressed, which possibly indicated an alternative splicing event within the 3′-UTR that changes its length and contributed to NMD targeting of the Cox-2 mRNA. Although our results implicated NMD in the increased turnover of Cox-2 mRNA that occurs in α3-KD cells, we cannot rule out contributions by other mRNA degradation mechanisms. Indeed, the 3′-UTR of Cox-2 mRNA is rich in AREs that control mRNA stability in response to other stimuli (Harper and Tyson-Capper, 2008; Moore et al., 2011), suggesting that several mechanisms of mRNA degradation might converge to determine the overall stability of the Cox-2 mRNA transcript. Consistently, inhibition of NMD by suppressing UPF1 did not completely restore Cox-2 mRNA levels in α3-KD cells (see Fig. 7B).

Pro-tumorigenic roles for integrin α3β1 have been well documented in both skin carcinogenesis models and breast carcinoma cells (Mitchell et al., 2010; Sachs et al., 2012). In addition, α3β1 has important roles in normal developmental and post-developmental processes of tissue morphogenesis and repair, including cutaneous wound healing (Margadant et al., 2009). In the skin, expression of α3β1 in the epidermis is important for proper assembly and organization of the basement membrane, as well as for paracrine crosstalk to the vascular compartment that promotes wound angiogenesis (Mitchell et al., 2009), leaving open the intriguing possibility that the α3β1-dependent gene regulation that we describe here also extends to normal tissue remodeling processes. Indeed, we have shown that some functions of α3β1 in activated keratinocytes are due to the regulation of genes with known roles in ECM remodeling and induction of angiogenesis, such as MMP-9 (Iyer et al., 2005; Lamar et al., 2008) and MRP-3 (mitogen-regulated protein-3, also known as proliferin-3 or as prolactin family 2, subfamily c, member 4) (Mitchell et al., 2009). Interestingly, α3β1-dependent induction of MMP-9 in immortalized keratinocytes was due to enhanced mRNA stability (Iyer et al., 2005), possibly through the AREs that reside in the 3′-UTR (Akool et al., 2003). α3β1-dependent MMP-9 expression was also reported in breast cancer cells (Morini et al., 2000), although the mechanisms of this regulation remain unclear. Of note, neoplastic transformation can stabilize ARE-containing mRNAs (Brennan and Steitz, 2001), and enhanced mRNA stability of Cox-2 has been linked to human cancers (Dixon et al., 2001).

Our observation that regulation of Cox-2 mRNA by α3β1 did not appear to require binding to laminin is suggestive of a ligand-independent mechanism, as has been described for this integrin by others (Zhang et al., 2003; Zhang et al., 1999). However, we cannot exclude that the properties of α3β1 signaling functions from within cell–ECM adhesions might also be sufficient to induce Cox-2 expression. It is also not yet known whether the control of Cox-2 mRNA expression or AEU is specific to α3β1 or also extends to other integrins. α3-KD MDA-MB-231 cells express other integrins on their surface (Mitchell et al., 2010), and these are clearly not sufficient to maintain the AEU patterns that are observed in α3β1-expressing cells (Fig. 4). Nevertheless, we cannot rule out the possibility that certain other integrins can similarly regulate gene expression through AEU or that integrins other than α3β1 have more prominent roles in this regulation in other cancer cell types.

In summary, our findings support a model wherein suppression of α3β1 in breast cancer cells reduces Cox-2 gene expression by allowing inclusion of a PTC-containing intron (Fig. 8), and possibly alterations in the 3′-UTR, both of which might contribute to NMD-mediated mRNA decay. These findings have important implications for understanding how integrins, in particular α3β1, contribute to cancer progression through gene regulation that drives malignant cell behavior and tumor angiogenesis, and how this regulation might be exploited in therapeutic strategies to inhibit cancer progression. Indeed, both α3 and Cox-2 are expressed highly in human breast cancers (Morini et al., 2000; Singh-Ranger et al., 2008; Subbaram and DiPersio, 2011), and our data mining analysis revealed that both are correlated with disease progression in human breast cancer (supplementary material Fig. S1). In addition, our previous work demonstrated that suppression of α3β1 in MDA-MB-231 cells reduced orthotopic tumorigenesis and Matrigel invasion in vitro, both attributable in large part to reduced Cox-2 expression (Mitchell et al., 2010). Moreover, we have recently established a positive correlation between expression of α3 and Cox-2 in clinical samples of human invasive ductal carcinoma, supporting the clinical relevance of α3β1-dependent Cox-2 regulation (A. Aggarwal and C. M. DiPersio, personal communication). It is well known that Cox-2 plays important roles in driving tumor angiogenesis and breast cancer progression, and clinical investigations of breast and other cancers have indicated that treatment with Cox-2 inhibitors might improve survival in some cancer patients (Harris, 2009; Howe, 2007; Singh-Ranger et al., 2008). However, adverse side effects have been associated with direct inhibitors of Cox-2, and indirect suppression of Cox-2 through targeting of integrin α3β1, or its downstream effectors, might avoid these side effects (Subbaram and DiPersio, 2011). Therefore, further investigation of the pathways whereby α3β1 controls AEU and/or suppresses NMD might lead to new therapeutic strategies to block these pathways in tumor cells; thereby reducing gene expression of Cox-2 (and other pro-cancer genes) through enhanced NMD and providing protective effects in breast cancer patients.

MDA-MB-231 cells were obtained from the American Type Culture Collection (ATCC, Manassas, VA) and cultured as described previously (Mitchell et al., 2010). Stable α3-KD cells were generated using lentivirus expressing one of two distinct shRNAs (shRNA#13 and #14) that target the α3 integrin subunit (MISSION, Sigma, St Louis, MO), and control cells were generated using lentivirus expressing an shRNA that does not target any human genes, as described previously (Mitchell et al., 2010). To suppress UPF1 or α3 using siRNA, cells were cultured to 50–60% confluence, then transfected for 96 hours with siRNA that targets UPF1 (Dharmacon, Lafayette, CO), α3 (Sigma, St Louis, MO, USA) or luciferase as a control, according to the manufacturer's instructions. α3-siRNA targeted either coding sequences [siRNA (si)α3–78, 3−79] or the 3′-UTR (siα3-UTR) of the human α3 mRNA. Cells were harvested for RNA analysis, western blot or flow cytometry as described below. α3-siRNA sequences were as follows: siα3-UTR, 5′-CAGAUGUUGGGAGGAUACA-3′; siα3-78, 5′-GUGUACAUCUAUCACAGUA-3′; siα3-79, 5′-GAGAUCACCGUCCAUGGCA-3′.

3.5×10⁵ cells were suspended in serum-free medium with 10 µg/ml of either control IgG or P1B5 (Millipore Corporation, Billerica, MA) at 37°C for 30 minutes, then replated for 24 hours in complete medium onto laminin-332-rich matrix prepared from SCC-25 cells, as described previously (DiPersio et al., 2000). RNA was prepared and assayed for Cox-2 mRNA expression as described below.

Lentiviral transduction of MDA-MB-231 cells has been described previously (Mitchell et al., 2010) and was performed using the same amounts of virus for control and α3-targeting shRNAs. The mutant α3 subunit (G163A) was a kind gift from Yoshikazu Takada (UC Davis, CA), and contains an alanine substitution in the predicted 4-1 loop (residues 153–165), which abolishes α3β1 binding to laminin-332 (Zhang et al., 1999). Wild-type or mutant α3 was cloned into the pAd-CMV-V5-DEST vector using the Gateway cloning system (Invitrogen, Grand Island, NY). Parental MDA-MB-231 cells were seeded overnight onto laminin-332-rich matrix on six-well dishes at 50–60% confluence. Cells were treated with an siRNA targeting the 3′-UTR of α3, as described above, then infected the next day with adenovirus encoding either wild-type or mutant α3. Cells were transfected with a second round of siRNA on the following day, then harvested two days later for western blot, RNA analysis or flow cytometry. Flow cytometry was performed using the anti-α3 monoclonal antibody P1B5, as we have described previously (Mitchell et al., 2010). Cox-2 and GAPDH mRNA levels were measured by RT-PCR, as described below.

Total RNA was isolated using Trizol extraction followed by clean-up on RNeasy columns (Qiagen, Valencia, CA), which included a DNase step. RNA was checked for quality using a NanoDrop and Agilent Bioanalyzer. RNA (1 µg) was processed according to the standard Affymetrix Whole Transcript Sense Target labeling protocol that included a riboreduction step. The fragmented biotin-labeled cDNA was hybridized over 16 hours to Affymetrix Exon 1.0 ST arrays and scanned on an Affymetrix Scanner 3000 7G using Affymetrix GeneChip Command Console software. The resulting CEL files were analyzed for quality using Affymetrix Expression Console software, then imported into GeneSpring GX11.5 and analyzed for both differential expression and alternative splicing. For gene expression analysis, data were quantile-normalized using PLIER on the core level and baseline transformed to the median of the control samples. The probe sets were further filtered to exclude those in the bottom twentieth percentile across all samples. The resulting entity list was subjected to an unpaired t-test with Benjamini-Hochberg False Discovery rate correction and a 1.5-fold filter to identify transcripts that are differentially expressed between the conditions at a P-value <0.05. For determination of alternatively spliced transcripts, the summarized probesets were filtered using the data above background (DABG) algorithm and probeset P-value <0.1 to select for entities with 50% of core probesets marked ‘Present’ in at least 100% of samples in at least one of the two conditions. This list was then subjected to splicing analysis of variance (ANOVA) (P<0.05) with a Benjamini-Hochberg False Discovery rate correction. A splicing index of greater than 0.5 was used to identify transcripts that showed alternative splicing between two conditions being compared. The complete gene list was submitted to Gene Expression Omnibus under series GSE43097.

IPA® analysis was performed following import of lists of differentially expressed and alternatively spliced genes into Ingenuity Pathway Analysis Software (V8.8) (Ingenuity® Systems, www.ingenuity.com). Analysis of the network identified biological functions and/or diseases that were most associated with the genes in the network. A Fisher's exact test was used to calculate the probability that each biological function and/or disease assigned to that network was due to chance alone. Canonical pathways analysis was performed to identify pathways from the Ingenuity® Pathway Analysis library that were most associated with the genes in the network. The significance of the association of a given canonical pathway and the network of gene targets was measured by (1) by determining the ratio of the number of genes in the network that mapped to the canonical pathway divided by the total number of genes that map to the canonical pathway, and (2) using a Fisher's exact test to calculate a P-value predicting the probability that the biological function or disease assigned to that network is explained by chance alone.

Comparisons of α3 and Cox-2 expression among clinical breast cancer samples, as well as across tumors of different grade, were obtained using Oncomine^TM (Compendia Bioscience, Ann Arbor, MI). For these analyses, breast cancer cohorts were scanned for correlative expression between the ITGA3 (α3) and PTGS2 (Cox-2) genes, and expression of each gene was assessed among different disease stages where possible.

Total RNA was isolated from MDA-MB-231 cell derivatives using RNeasy mini-prep kit (Qiagen) according to the manufacturer's instructions. In some experiments, cells were treated with Ataluren at the indicated concentrations, or with DMSO as a vehicle control, for various time-points (24–72 hours) before RNA isolation. For these experiments, Cox-2 mRNA in cells treated with Ataluren was normalized to that in control cells treated with DMSO alone at each time-point (Fig. 7E) as DMSO itself could have a suppressive effect on Cox-2 mRNA (Hollebeeck et al., 2011). cDNA was synthesized from 1.0 µg RNA using First-Strand cDNA Synthesis kit (Promega, Madison, WI). Conventional PCR for Cox-2 or GAPDH was carried out using REDTaq ready mix (Sigma), as described previously (Lamar et al., 2008; Mitchell et al., 2010). Signals were quantified using a Fuji Film Bio Imager. PCR using primers from Cox-2 exons 7 and 8 that spanned intron 7 was performed at an increased cycle number, in order to facilitate detection of intron inclusion. Intron-7-spanning primers were as follows: forward, 5′-TGCCTGGTCTGATGATGTATGCCA-3′; reverse, 5′-TTTGAAAGGTGTCAGGCAGAAGGG-3′. Real-time PCR was performed using SYBR green master mix (Bio-Rad, Hercules, CA) in an iCycler iQ Multicolor Detection System (Bio-Rad), using β-actin mRNA as a control. GAPDH primer sequences have been published previously (Lamar et al., 2008). β-actin primers were as follows: RT-PCR – forward, 5′-TACCTCATGAAGATCCTCACC-3′; reverse, 5′-TTTCGTGGATGCCACAGGAC-3′; real-time PCR – forward, 5′-ACCAACTGGGACGACATGGAGAAA-3′; reverse, 5′-TAGCACAGCCTGGATAGCAACGTA-3′. Other conventional and real-time PCR primer sequences for Cox-2 specific exons are provided in supplementary material Table S4.

RNA was isolated as described above, and equal quantities were assayed by RNase protection using the Ribonuclease Protection Assay kit (Ambion, Grand Island, NY) according to the manufacturer's protocol. RNA probes were transcribed with biotin-16-UTP using MAXIscript in vitro transcription kit (Ambion). Probes were designed to hybridize to Cox-2 mRNA either completely within exon 7 as a control (326 nucleotides), or spanning the junction of exon 7 and intron 7 (309 nucleotides). Products were electrophoresed on a 7% denaturing urea gel, then transferred to Brightstar-Plus membranes (Ambion) using a semi-dry transfer apparatus (Bio-Rad) at 200 mAmps for 1 hour. Membranes were UV-crosslinked and assayed using the BrightStar Biodetect system for nonisotopic detection (Ambion).

Cells were lysed in Cell Lysis Buffer (Cell Signaling Technology, Beverly, MA), and protein concentrations determined using a Bicinchoninic Acid Protein Assay kit (Pierce, Rockford, IL). Equal protein was assayed by immunoblot using rabbit anti-sera against UPF1 (1∶1000 dilution) (Bethel Laboratories, Montgomery, TX), phosphorylated eIF2α, total eIF2α (Cell Signaling Technologies), α3 (DiPersio et al., 1995) or extracellular signal-regulated kinase (ERK) (Santa Cruz Biotechnology, Santa Cruz, CA) (1∶1000), followed by horseradish peroxidase-conjugated goat anti-rabbit IgG (1∶1000). Chemiluminescence was performed using SuperSignal kit (Pierce).

mRNA stability assays were performed as described previously (Iyer et al., 2005). Briefly, cells in serum-free Dulbecco's Modified Eagle's Medium were pre-treated with 10 µg/ml cycloheximide for 16 hours, then allowed to recover for 4 hours in serum-free DMEM before adding 10 µg/ml actinomycin D to inhibit transcription. The initial time-point was collected 30 minutes later, followed by collection at the time-points indicated. Cox-2, β-actin and GAPDH mRNA levels were measured by RT-PCR as described above.

We thank Yoshikazu Takada (UC Davis, CA) for providing the α3 laminin binding mutant. We also thank Whitney Longmate, Dara Missan and Kara Mitchell (Albany Medical College, NY) for excellent technical assistance, and Susan LaFlamme (Albany Medical College, NY) for critical reading of the manuscript.