signature=7c222ef2a797592b2588678ade940852,A microRNA gene expression signature predicts response to...

Identification and validation of the miRNA expression signature of response

We initiated these studies by identifying a signature of response to erlotinib using miRNA expression levels in NSCLC cell lines. In a previous publication, we separated a panel of NSCLC cell lines into sensitive and resistant to erlotinib treatment by measuring apoptosis after 48 h of treatment (Balko et al, 2006). Here, we measured the expression of 381 miRNA genes in the same resistant (A549, UKY-29, H460, and H1975) and sensitive (H1650, H3255, PC-9, and H358) cell lines using Taqman microRNA arrays.

Gene expression data were normalised and filtered, then a t-test was used to identify genes differentially expressed between erlotinib-sensitive and -resistant cell lines. A 13-gene miRNA signature was selected from 38 differentially expressed miRNA genes (P<0.1), because this signature was the most accurate of trained models in predicting response in validation data sets. The 13-gene signature miRNAs were clustered by relative expression levels and presented as a heatmap (Figure 1A). Eleven of the signature miRNAs are upregulated, and two miRNAs were found to be downregulated, in erlotinib-sensitive cells (S in Figure 1; H1650, H3255, H358, and PC9).

Figure 1

Hierarchical clustering of 13-gene microRNA signature. (A) Signature miRNA clustered by expression values and sample (lung cell lines) using GenePattern (Broad Institute) (B) Signature miRNA clustered by expression and sample (lung and colorectal tumours). Samples A–D are primary lung tumours (P) from patients treated with erlotinib. Samples U38–U40 are metastatic lung tumours (M) from patients treated with erlotinib and U41 is a metastatic colorectal tumour from a patient treated with the EGFR inhibitor, cetuximab.

The 13-gene signature was used to develop a predictive algorithm of response to erlotinib utilising our previously published DLDA (Balko et al, 2006). A leave-one-out cross-validation assay was used to internally validate the predictor using training cell line data (resistant lines=A549, H460, H1975, and UKY29; sensitive lines=H1650, H3255, H358, and PC9). H1975 cells failed this validation. H1975 cells are genetically different compared with the other resistant lines in that they contain an EGFR mutation correlating with sensitivity and a second mutation that confers resistance (Gendreau et al, 2007; Figure 1A).

Further validation of the predictive accuracy of the selected 13-gene signature was carried out. First, the signature was compared with randomly selected 13-gene predictors for association with sensitivity to EGFR inhibitors using external validation data. We randomly selected 10 000 13-gene predictors from all 378 probes to test the accuracy of the selected signature. Our selected signature was significantly better than randomly selected signatures at predicting response (P=0.049). Next, expression data from NSCLC and PDAC cell lines with previously described erlotinib sensitivities were tested using the predictive signature (Eberhard et al, 2005; Tzeng et al, 2007). Finally, expression data from seven NSCLC and one colorectal cancer (primary and metastatic tumours), paraffin-embedded tumour samples from patients treated with EGFR inhibitors were also tested. These data indicate that with further validation the miRNA expression signature of erlotinib response may have clinical utility (Table 1).

Interestingly, we found that the 13-gene signature could not only predict response of patients to erlotinib, but could also discriminate primary (P) from metastatic (M) tumours, suggesting that the biological phenotypes underlying the signature were associated with both resistance to EGFR inhibition and metastatic behaviour (Figure 1B).

Annotation of the 13-gene signature of response

To dissect the biological contribution of signature miRNAs to erlotinib sensitivity and potentially to metastasis, we searched for potential target mRNAs of the miRNA genes in our predictive signature using TargetScan, miRDB, and miRANDA. Targets identified by miRANDA/miSVR with scores less than −1.25 were imported into IPA and assessed for gene ontology and/or signalling pathway membership. The most significantly counted canonical pathway is indicated for each miRNA along with the P-value of enrichment in Table 2, and relevant targets are noted in Figure 2.

Figure 2

TGFβ response pathway. Highlighted are targets of the microRNAs that are members of the response signature.

Targets of the signature miRNAs were enriched for Wnt/β-catenin canonical signalling pathway (Table 2). Using highly enriched canonical pathways from each miRNA surveyed, networks of signalling events were constructed. The networks contain actual and inferred genes from the signature miRNA targets. The networks revealed that TGFβ1 was a hub in networks formed by mir-140, -636, -301a, -224, and -200c genes (Supplementary Figure 1), even though the TGFβ1 canonical signalling pathway was the top-enriched pathway in only the mir-636 network (Table 2). Signalling events important in EGFR biology were also evident, including MAPK and EGFR signalling pathways, but not as the most counted pathways.

Literature interrogation of signature miRNAs indicated that the expression cluster (Figure 1A) containing the mir-200 family, mir-205, mir-135b, and mir-141 target genes essential for the regulation of EMT (Burk et al, 2008; Park et al, 2008; Mongroo and Rustgi, 2010). Using these data, we decided to pursue the role of the 13-gene signature on TGFβ1-induced EMT and erlotinib sensitivity.

Biological characterisation of the signature members and EMT

Epithelial-to-mesenchymal transition regulator genes, such as E-cadherin and vimentin, are included in previously published signatures of response to EGFR inhibition (Yauch et al, 2005; Coldren et al, 2006; Thomson et al, 2005). Similarly, we identified ∼1500 genes with altered expression between erlotinib-sensitive and -resistant cells (Balko et al, 2006). Although EMT genes were not included in our 180-gene predictor of response, we re-evaluated the ∼1500 genes used to generate the 180-gene GEPR to identify other genes that may influence in EMT.

ZEB1/TCF8, a target of the mir-200 family of miRNA, was significantly downregulated in erlotinib-sensitive cells (P=1.51801E−17; Balko et al, 2006). E-cadherin, a target repressed by ZEB1, was correspondingly upregulated in the same cells consistent with the observations of others (Yauch et al, 2005; Coldren et al, 2006). Interestingly, we found that the mir-200 family of miRNA is upregulated in erlotinib-sensitive cells providing a possible explanation for ZEB1 loss in these cells (Figure 1A).

To validate the microarray expression data, A549 (EGFR wt), PC9 (EGFR delE746-A750), and H1650 (EGFR delE746-A750) were assayed for expression of ZEB1. Expression was measured by qPCR in each cell line (Figure 3A). ZEB1 expression in A549 cells (erlotinib resistant) is two- to three-fold higher than in either of the two EGFR mutant, erlotinib-sensitive cells (H1650 and PC9), supporting the microarray analysis. However, ZEB1 protein is upregulated in A549, while H1650 and PC9 cells poorly express ZEB1 protein despite varying levels of mRNA (Figure 3C). These data indicate that miRNA targeting ZEB1 message may inhibit translation rather than promote degradation of the message.

Figure 3

ZEB1 is expressed in erlotinib-resistant lung and pancreatic cancer cell lines. (A) ZEB1 mRNA expression in lung cancer cell lines: A549, H1650, and PC9. (B) ZEB1 mRNA expression in pancreatic cancer cell lines: MiaPaca, Panc-1, Aspc-1, and Bxpc-3. (C) ZEB1 protein expression in lung and pancreatic cancer cell lines. Calnexin serves as a loading control. (D) Expression of 13-gene miRNA signature in pancreatic cell lines, clustered by expression and sample as in Figure 1.

To understand whether sensitivity to erlotinib and expression of EMT genes is mediated by similar mechanisms in pancreatic cancers, as in NSCLC, ZEB1 protein expression was also evaluated in the PDAC cell lines. We hypothesised that ZEB1 gene expression would be reduced in the erlotinib-sensitive cell lines (Aspc-1 and Bxpc-3) compared with the erlotinib-resistant lines (MiaPaCa and Panc-1). ZEB1 mRNA was highly expressed in three of the four lines, but was not significantly expressed in Bxpc-3 cells (Figure 3B). ZEB1 protein expression correlated with gene expression data in three of four cell lines. Importantly, in erlotinib-sensitive pancreatic cancer lines, ZEB1 protein levels were reduced (Figure 3C). These data indicate that miRNAs can inhibit translation or promote message degradation, and ZEB1 mRNA expression alone is not a reliable surrogate for ZEB1 protein levels. Figure 3D illustrates expression levels of the 13 signature miRNA genes demonstrating that erlotinib-sensitive pancreatic cancer cell lines have high expression of the mir-200 family similar to the NSCLC-sensitive cells.

Together, these data indicate that signature miRNAs highly expressed in erlotinib-sensitive cells can control expression of target genes involved in EMT induction. We will specifically explore the role of mir-200c, a representative member of the highly expressed cluster of signature miRNAs, in lung cell lines to determine what signals can modulate expression of the miRNA, influence induction of EMT, and impact response to erlotinib.

Expression levels of mir-200 modulate both signature miRNA and EMT targets

We used erlotinib-resistant A549 lung adenocarcinoma cells for initial characterisation of the role of mir-200c in erlotinib sensitivity and EMT induction. A549 cells express low levels of mir-200c compared with erlotinib-sensitive NSCLC cells, such as PC9 (Figure 1A). We expressed mir-200c from a lentiviral expression vector in A549 cells and observed increased mir-200c message levels, no change in ZEB1 expression, and gene expression of E-cadherin relative to the parent cells (Figure 4A). We also investigated the expression of representative miRNA from each of the major expression clusters (Figure 1A) in A549 cells expressing ectopic mir-200c. Expression of mir-301a and -34c each increased, relative to parent A549 cells.

Figure 4

Ectopic expression of mir-200c alters expression of signature miRNA, EMT genes, and modulates sensitivity to erlotinib in A549 cells. (A) ZEB1, E-cadherin, mir-34c, mir-301a, and mir-140 levels were determined in A549 cells. The mir-200c precursor was expressed from a plasmid vector, and mir200c-expressing cells were selected with puromycin. Expression of the query genes was evaluated by qPCR in control and transfected cells. The data are the mean fold change, relative to control (A549 parent cells), of two independent experiments each evaluated in triplicate showing standard deviation (s.d.). (B) A549 and A549+mir200c-expressing cells were treated with erlotinib for 72 h in 0.1% serum-containing RPMI. Cell growth was measured by SRB assay in triplicate, averaged, and compared with 0 h controls. Cell growth at 72 h is plotted against erlotinib concentration.

A549 cells are insensitive to erlotinib at concentrations that kill EGFR mutant cells. We asked whether ectopic expression of mir-200c, resulting in increased expression of a subset of the signature miRNAs, would increase sensitivity to erlotinib. Ceppi et al (2010) previously showed that introduction of mir-200c into H1299 cells restored sensitivity to cetuximab, also an EGFR inhibitor. Sensitivity to erlotinib was tested using concentrations from 1 to 50 μ M in A549 and A549+mir200c cells. Ectopic expression of mir-200c modestly increased the sensitivity of A549 to erlotinib, similar to results observed by others (Ceppi et al, 2010; Figure 4B).

ZEB1 expression, induction of EMT, and erlotinib sensitivity are likely controlled by the combined activity of many factors, including the signature miRNA reported here. We furthered these observations by evaluating expression of mir-200c and proteins involved in the induction of EMT following treatment with TGFβ1, an EMT inducer, because several signature miRNAs target members of the TGFβ response pathway (Figure 2, Supplementary Figure 1).

TGFβ1 controls the expression of mir-200c

Ingenuity Pathway Analysis software revealed that TGFβ signalling pathway members are targets of signature miRNAs (Figure 2, Supplementary Figure 1). TGFβ1 can initiate EMT, and EMT is a hallmark of tumour invasion and metastasis. TGFβ1 signalling may also contribute to erlotinib response in lung cancers (Yao et al, 2010). Therefore, we asked whether the induction of EMT by TGFβ1 is mediated by differential regulation of signature miRNA expression, especially that of mir-200c.

Epithelial-to-mesenchymal transition is characterised by increased expression of mesenchymal markers, such as ZEB1 and N-cadherin, while losing epithelial markers such as E-cadherin. A549 cells demonstrate a mesenchymal phenotype, but express both Zeb1 and E-cadherin. Because EMT is a plastic state, allowing cells to vary expression of EMT-specific proteins, cells grown in two dimensions may not commit to either epithelial or mesenchymal states. The impact of TGFβ1 treatment in both A549 cells and mir-200c-expressing A549 cells was evaluated by expression of RNAs and proteins important for EMT (Adam et al, 2009; Gibbons et al, 2009).

In both A549 and A549+mir200c cells, TGFβ1 treatment changed cellular morphology (Figure 5A). TGFβ1 predictably reduced E-cadherin and increased ZEB1, N-cadherin, and Snail in A549 cells (Figure 5B). Unexpectedly, all miRNAs tested demonstrated increased expression relative to untreated cells (Figure 5C). Of the miRNAs tested, we expected mir-140 to be the only miRNA upregulated, as it does not obviously link to EMT induction and is poorly expressed in erlotinib-sensitive cells. In A549+mir200c cells after treatment, both ZEB1 and Snail proteins are poorly expressed, and the miRNAs tested are expressed at the same level as untreated parental A549 cells. Thus, TGFβ1 treatment counters the ectopic expression of mir-200c, but cannot fully induce EMT (poor Snail expression and decreased ZEB1) and does not alter response to erlotinib (Supplementary Figure 2). These data indicate that mir-200c and TGFβ1 treatment have independent effects on EMT induction.

Figure 5

TGFβ1 treatment has differential effects on EMT in A549 cells compared with ectopic mir-200c expression. (A–C) A549 or A549+mir200c cells were treated for 3 days with TGFβ1 (2 ng ml−1). (A) Cellular morphology is altered in the presence TGFβ1 in both cell lines. (B) Proteins involved in EMT are differentially expressed in the presence of TGFβ1 in A549 and A549+mir200c-expressing cells. ZEB1, E-cadherin, β-catenin, Snail, and pERK1/2 expression was determined by western analysis. α-Tubulin serves as a loading control. (C) Expression of EMT effector mRNA and a panel of signature miRNAs are differentially expressed in A549 and A549+mir200c expressing. Expression of ZEB1, E-cadherin, mir-200c, mir-34c, mir-301a, and mir-140 genes were evaluated by qPCR in control and mir-200c-expressing cells. The data are mean fold change, relative to control (untreated A549 cells), of two independent experiments each evaluated in triplicate. Bars show standard deviation (s.d.).

To evaluate the influence of TGFβ1 treatment on signature miRNA expression, the promoter region of each miRNA was evaluated for TGFβ1-responsive elements using ChipMAPPER (Marinescu et al, 2005). mir-200c, -141, -34c, and -301a promoters contain SMAD3/4 binding sites, indicating probable responsiveness to TGFβ1 signalling. Further experiments will be required to confirm this observation.

TGFβ1 or reduction of mir-200c levels accelerates wound healing

Beas2B cells, immortalised bronchial epithelial cells, have not undergone complete EMT and express high levels of both mir-200c and ZEB1. We utilised a wound-healing assay to evaluate the migration response, as an in vitro surrogate of metastasis, of Beas2B cells in the presence of TGFβ1 treatment and/or reduction in mir-200c expression.

Beas2B cells were transduced with a lentiviral vector expressing an shRNA) to reduce mir-200c expression. Short hairpin RNA-transduced and control Beas2B cells were grown to a monolayer and scratched. Both mock and transduced cells were pre-treated with TGFβ1 1 day prior to the scratch, and treatment continued throughout the experiment. Reduction in mir-200c expression accelerated wound healing compared with mock-transduced, parental Beas2B cells. The healing phenotype in both conditions is accelerated with TGFβ1 treatment (Figure 6). These data indicate that TGFβ1 and mir-200c participate in complementary pathways for migration in lung cells.

Figure 6

Treatment with TGFβ1 and abrogation of mir-200c accelerates wound healing in Beas2B cells. Beas2B cells were transduced with a mir-200c shRNA lentivirus or mock infected. Both cell lines (Beas2B or Beas2B-anti-mir200c) were pre-treated for 24 h with TGFβ1 (2 ng ml−1) then scratched and treatment was continued. The wound was measured in 10 fields of view each day for 4 days or until the wound closed to calculate standard deviation. Closure is expressed as percentage of day 0.