Transcriptional Profiling Identifies Cyclin D1 as a Critical Downstream Effector of Mutant Epidermal Growth Factor Receptor Signaling

Development of anilinoquinazoline epidermal growth factor (EGF) receptor (EGFR) inhibitors has greatly affected the treatment of advanced non–small cell lung cancer (NSCLC). These compounds show particular promise in certain subsets of patients, including women, nonsmokers, younger patients, patients with adenocarcinoma histology, and Asian populations (1, 2). Somatic EGFR mutations have been identified in such patients at a high frequency and it seems that responsiveness to this class of agents strongly correlates with the presence of these EGFR mutants (3–7). Although it has been reported that these mutations mediate oncogenic effects by altering downstream signaling and antiapoptotic mechanisms (8), alterations in transcriptional pathways are not fully understood.

Despite the initial success of EGFR inhibitors, resistance emerges in the majority of patients over time. Recently, we and others (9, 10) identified a single base pair change leading to a threonine to methionine (T790M) amino acid alteration in the ATP-binding pocket of the EGFR, which leads to steric hindrance caused by the introduction of a bulkier methionine residue interfering with drug binding and demonstrating high-level resistance against gefitinib and erlotinib. Although we showed that an alternative, irreversible anilinoquinazoline EGFR inhibitor, CL-387,785, can overcome the resistance conferred by T790M (9, 11), it is predictable that novel resistance mutations would emerge over time, limiting the efficacy of these irreversible inhibitors (12, 13). The H1975 NSCLC cell line carries a double L858R/T790M mutation and is highly resistant to gefitinib, whereas prominent growth arrest and apoptosis result after treatment with CL-387,785 (10, 11). Based on these observations, we hypothesized that the study of genes that are differentially expressed in the “resistant” gefitinib-treated and the “sensitive” CL-387,785-treated H1975 cells may allow the identification of pivotal downstream target genes in EGFR-driven cancers, especially in NSCLC with EGFR-activating mutations, and provide a novel strategy to overcome and/or prevent the emergence of the resistance. To this end, we did a transcriptional profiling study of mutant EGFR target genes using H1975 cells and identified cyclin D1 as a critical downstream effector of mutant EGFR. These results suggest that attenuation of cyclin/cyclin-dependent kinase (CDK) pathways may be an alternative therapeutic target in EGFR-mutant NSCLC.

Reagents. Stock solutions for gefitinib, erlotinib, CL-387,785, and flavopiridol were prepared as previously described (11, 14).

Cell culture. Calu-1, A549, H1975, and H460 were obtained from the American Type Culture Collection (Rockville, MD) and were maintained in the manufacturer-specified growth medium. The HCC827 cell line was obtained from Dr. Pasi A. Jänne (Lowe Center for Thoracic Oncology and Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA) and was maintained as described previously (15, 16). H3255 cells were maintained as described previously (17).

Oligonucleotide array analysis. H1975 cells were grown to 60% confluence in RPMI containing 10% fetal bovine serum. Pentuplicate plates were incubated by the addition of either gefitinib (1 μmol/L) or CL-387,785 (1 μmol/L), or DMSO (0.01%) as a control for 6 and 24 hours. Cells were collected at the same time, and total cellular RNA was isolated using the RNeasy kit (Qiagen). RNA specimens were then processed and hybridized to Affymetrix HG-U133A microarrays and scanned. The expression value for each gene was calculated using Affymetrix GeneChip software and the robust multichip average method for signal extraction that is part of BioConductor (18).

Preprocessing, filtering, and statistical analysis. The raw expression data consisted of the “signal” units as obtained by application of the robust multichip average method to the “.CEL” files produced by the Affymetrix scanner (18, 19). Agglomerative hierarchical clustering was preliminarily done on the entire data set projected on the space of a reduced set of 2,307 probe sets/“genes” [filter: 0.2≤CV≤10;P-calls≥10%;avg(EXP) ≥50 in at least 10% of the samples]. We used Pearson correlation, as the distance measure, and a centroid-based agglomeration rule. The software package dChip was used for this analysis (20). Differential analysis was done on a set of 5,570 genes with median absolute deviation of the signal across chips above the 75th percentile. Genes differentially expressed with respect to the binary phenotype of interest were identified by computing their variance-thresholded t statistic (with the threshold set to the minimum variance observed within the replicate controls). Empirical, gene-specific P values were computed by permutation test. When both time points (6 and 24 hours) were pooled in the analysis, restricted permutations aimed at controlling for the potential confounding effect of the time points were carried out (21, 22). Because several thousand genes were tested, the nominal P values were corrected for multiple hypotheses testing by the false discovery rate (FDR) procedure (23). The analysis was carried out using GenePattern (24) and a set of ad hoc R scripts.

Gene Ontology (GO) annotation was done by testing (based on the hypergeometric distribution) for the overrepresentation of GO categories in each of the marker lists of interest. The software package GeneMerge was used for this purpose (25).

Real-time PCR assay. The mRNA levels of genes were measured by SYBR green real-time PCR. DNase-treated RNA was reverse transcribed and subsequently amplified using an ABI Prism 7700 Sequence Detector (Applied Biosystems) by the following variables: 50°C (2 minutes), 95°C (10 minutes) followed by 40 cycles of 95°C (15 seconds), and 60°C (60 seconds).

Primers for human cyclin D1 were as follows: forward primer, 5′-ACCTGAGGAGCCCCAACAA 3′; reverse primer, 5′-TCTGCTCCTGGCAGGCC-3′. Human cyclin D3: forward primer, 5′-TGGATGCTGGAGGTATGTG-3′; reverse primer, 5′-CGTGGTCGGTGTAGATGC-3′. Human G₀-G₁ switch gene 2 (G0S2): forward primer, 5′-CGCCGTGCCACTAAGGTC-3′; reverse primer, 5′-GCACACAGTCTCCATCAGGC-3′. Human cyclin G₂: forward primer, 5′-ATCGTTTCAAGGCGCACAG-3′; reverse primer, 5′-CAACCCCCCTCAGGTATCG-3′. Human dual-specificity phosphatase 6 (DUSP6): forward primer, 5′-CAGTGGTGCTCTACGACGAG-3′; reverse primer, 5′-GCAATGCAGGGAGAACTCGGC-3′. Human glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were used for internal control: forward primer, 5′-GAAGGTGAAGGTCGGAGTC-3′; reverse primer, 5′-GAAGATGGTGATGGGATTTC-3′.

EGFR-mutant constructs and transfections. EGFR-mutant constructs were generated as described previously (9) and used to generate stable HCC827 cell lines using FuGene6 transfection (Roche, Basel, Switzerland) followed by selection in 1 mg/mL G418.

Antibodies and Western blotting. Whole-cell lysates were prepared as previously described (17). To assess cyclin D1 expression and the phosphorylation level of the proteins, the cells were serum-starved for 24 hours and were then stimulated with 100 ng/mL EGF for 3 hours in the presence or absence of inhibitors. To monitor expression of RNA polymerase II and its phosphorylated (phospho-) forms (Ser² and Ser⁵), cells were plated at 1 × 10⁶ per 10-cm dish and treated with 300 nmol/L flavopiridol for the indicated times. Forty micrograms of proteins were separated on 8% or 12.5% SDS-polyacrylamide gels. The anti-cyclin D1 antibody was from EMD Biosciences (San Diego, CA). Total EGFR antibody and total RNA-Pol II (N20) were purchased from Santa Cruz Biotechnology (Santa Cruz, CA). Total extracellular signal-regulated kinase 1/2 (ERK1/2) antibody was purchased from BD Transduction Laboratories (Lexington, KY). Phospho-EGFR (pTyr1068), phospho-AKT (pS473), phospho-ERK1/2 (pT202/pY204), and total AKT antibodies were purchased from Cell Signaling Technology (Danvers, MA). Anti-phospho-RNA Pol II (Ser²) and anti-phospho-RNA Pol II (Ser⁵) were from Covance Research Products (Berkeley, CA). Antibodies were used according to the manufacturers' recommended conditions.

Growth inhibition assay. Growth inhibition was assessed by 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt (MTS) assay using CellTiter 96 AQueous One solution proliferation kit (Promega, Madison, WI) as previously described (11). Briefly, HCC827 stable cells were transferred to triplicate wells at 10,000 per well in 96-well flat-bottomed plates. The next day, the cells were incubated with various concentrations of inhibitors for 72 hours.

Cell cycle analysis. Cell cycle was assessed as previously described (17). Briefly, following fixation and treatment in 500 μg/mL RNase A, cells were resuspended in 69 μmol/L propidium iodide (1 mL) in 30 μmol/L sodium citrate. Cells were analyzed for DNA content by flow cytometry using the ModFit program (Verity Software House, Topsham, ME).

Apoptosis analysis. Cells were plated at 1 × 10⁵ per well in a six-well plate and treated with DMSO or flavopiridol. Apoptosis was assessed using an Annexin-V–FLUOS staining kit (Roche) as previously described (9, 11).

CL-387,785 induces G₁ arrest and apoptosis in gefitinib-resistant H1975 cells. Previously, we have shown that H1975 cells harboring a double L858R/T790M mutation are highly resistant to gefitinib, whereas the irreversible EGFR inhibitor, CL-387,785, overcomes the resistance. To confirm this observation, we did fluorescence-activated cell sorting analysis using propidium iodide. Apoptotic DNA fragmentation as well as cell cycle distribution was analyzed by propidium iodide staining of ethanol-permeabilized cells (17, 26). H1975 cells were treated for 6 and 24 hours with gefitinib (1 μmol/L) or CL-387,785 (1 μmol/L). Control cells were grown in the presence of 0.01% DMSO. After exposure to CL-387,785 for 24 hours, the cells showed a significantly higher G₁ peak consistent with G₁ arrest compared with DMSO- or gefitinib-treated cells. After 48 hours, CL-387,785–treated cells showed a dramatic increase in G₁ content together with increased sub-G₁ population consistent with induction of apoptosis (Fig. 1). These results suggest that CL-387,785 induces cell cycle arrest in the G₁ phase and subsequent apoptosis in H1975 cells.

Transcriptional profiling of gefitinib versus CL-387,785-treated H1975 cells. To identify the genes differentially expressed in gefitinib-treated and CL-387,785–treated H1975 cells, a transcriptional profiling study was done on total cellular RNA using Affymetrix HG-U133A chips. The raw expression data were preprocessed, rescaled, filtered, and analyzed as described in Materials and Methods. Experiments were carried out in pentuplicates, with two time points for each experiment (6 and 24 hours). Hierarchical clustering with a correlation-based, centroid-linkage algorithm was done for exploratory purposes. The visualization of the correspondingly sorted data set clearly shows the separation of the CL-387,785–treated samples from the rest, as well as the much weaker distinction of the gefitinib-treated samples. Formal differential analysis based on a variance-thresholded t statistic and permutation test confirms the hierarchical clustering results. This analysis clearly distinguished gefitinib and CL-387,785–treated samples (Fig. 2A). The genes altered by CL-387,785 treatment are shown as Table 1 (6 hours) and Supplementary Tables S1 and S2 (24 hours). The marker signatures of the differently treated samples was further examined by assessing the marker overlap between sets with the use of a strict permutation based FDR value of ≤0.01 (Supplementary Table S3). This analysis showed that 1,641 marker genes were down-regulated and 1,948 genes were up-regulated when all CL-387,785–treated samples were compared with the rest of the samples, whereas only three genes were down-regulated and one gene was up-regulated when gefitinib-treated samples were compared with the rest of the samples. These results illustrate the striking changes as a result of CL-387,785 treatment and also show that gefitinib has essentially no “off-target” effect on these cells and shows great selectivity of the drug at the concentration used for EGFR. Furthermore, the same comparison at 6 hours of treatment identified 615 down-regulated and 688 up-regulated genes distinguishing CL-387,785–treated samples from the rest, whereas at 24 hours, the corresponding analysis identified 1,923 down-regulated genes (477 of these overlapped between 6 and 24 hours, 77%) and 2,607 up-regulated genes (582 overlapping, 85%). This latter comparison is based on relatively small sample sizes in each group and therefore should be considered more representative than definitive.

Next, we were interested in identifying those genes manifesting a differential expression between DMSO/gefitinib and CL-387,785 at 6 hours to identify the earliest wave of transcriptional changes most likely highly enriched in direct targets of mutant EGFR. Because no significant difference was seen between the DMSO and gefitinib-treated samples, it was prudent to pool these samples to provide more power for our analysis. Then, we used very stringent criteria to select genes that are both statistically different in the two groups and made biological sense based on substantial fold changes. We used a variance-thresholded t statistic to rank genes. From the ranked list of genes, we excluded genes with a FDR > 0.01 (representing no association). We further narrowed this list down by limiting our analysis to genes differing at least 2-fold. The top genes thus obtained are shown in Table 1. This analysis identified only five known up-regulated genes by CL-387,785 in this set, cyclin G₂, DRE1, programmed cell death 4, aquaporin 3, and pellino homologue 1. We confirmed the up-regulation of cyclin G₂ by quantitative reverse transcription-PCR analysis on the RNA used for the array analysis (Fig. 2B). Of note is that cyclin G₂ is involved in negative cell cycle regulation (27, 28).

The analysis of the highly down-regulated genes identified 18 genes falling into three major clusters. First, many of the down-regulated genes code for growth factors, growth factor receptors, or proangiogenic molecules, such as transforming growth factor α (TGFα), interleukin 11, Cyr61 (also called CCN1), thromboplastin, EphA2, and vascular endothelial growth factor. Many of these factors are regulated by EGFR signaling corroborating the validity of our findings and also suggesting the presence of a positive feedback loop in EGFR signaling (29). A second group of genes comprises genes encoding DUSPs, such as DUSP4, DUSP5, and DUSP6. This finding suggests the presence of a strong negative feedback loop toward the mitogen-activated protein (MAP) kinase (MAPK) pathway induced by oncogenic EGFR signaling (30). The third group of genes includes activator protein-1 (AP-1) components, such as c-Jun and FOS-like antigen-1. Because the AP-1 complex is one of the major mediators of proliferative signals through MAPK and signal transducers and activators of transcription (STAT) signaling, this also seems very consistent with an EGFR signaling effect (31, 32). Last, one of the most highly suppressed genes is cyclin D1. This finding is of particular importance because cyclin D1 is known to play a major role in cell cycle progression and has previously also been described as a target of EGFR and ErbB2 signaling (33–36). Three of these genes, cyclin D1, G0S2, and DUSP6 were selected and confirmed by quantitative real-time PCR (Fig. 2B).

Next, we did the same analysis for the 24-hour gene set using the same selection criteria. Here, our analysis led to the identification of a much larger number of regulated genes (Supplementary Tables S1 and S2). Given the large number of genes, we next did GO annotation analysis of the marker genes identifying CL-387,785–treated samples to arrive at a comprehensive view of the gene changes observed. This analysis showed that the list of down-regulated markers in the CL-387,785–treated samples is significantly enriched with members of these five categories: nucleus, cell cycle, mitosis, DNA replication, and cell division (Bonferroni-corrected P < 1.0e⁻¹¹; Supplementary Table S4). This finding strongly suggests that the marked gene changes observed at 24 hours are a reflection of dramatic cell cycle and proliferation arrest. Next, we examined whether we might be able to identify a specific cyclin D1–mediated signature within this group. The D-type cyclins, D1, D2, and D3, associate with CDK4 and CDK6 and play a critical role early in the G₁ phase of the cell cycle. These complexes phosphorylate the retinoblastoma protein and inactivate its ability to act as a transcriptional repressor in a complex with E2F. The release of E2F leads to transcriptional induction of genes required for progression from G₁ to S phase, most notably cyclin E. Therefore, cyclin D1 down-regulation should lead to shutdown of E2F-mediated transcriptional activity. A previous study identified E2F-regulated gene changes using transcriptional profiling of E2F-inducible model systems. A comparison of the E2F-regulated genes identified through the study of Ishida et al. (37) to the gene set down-regulated in the CL-387,785–treated specimens very remarkably showed a perfect match between the two sets for the genes that were present on both the mouse arrays used in that study and the HG-U133A human arrays used in our current study (Table 2). These results further suggest that CL-387,785 induces G₁ cell cycle arrest (Fig. 1) mediated by repression of cyclin D1 and E2F followed by apoptosis.

Cyclin D1 is down-regulated by CL-387,785, but not by gefitinib or erlotinib, in H1975 cells. To confirm the findings of the oligonucleotide array analysis, the repression of cyclin D1 was confirmed by real-time quantitative PCR analysis (Fig. 2B). When H1975 cells were treated by CL-387,785, expression of the cyclin D1 gene on the RNA level was repressed to 31.4 ± 4.8% (mean ± SD) of gefitinib-treated or DMSO-treated control cells (Fig. 2B). In contrast, expression of the cyclin D3 gene was repressed only modestly to 66.9 ± 4.9% (Fig. 1C), suggesting that down-regulation of the cyclin D1 is more correlated with drug sensitivity. In addition, Western blotting analysis showed that cyclin D1 was down-regulated as early as 2 hours into CL-387,785 treatment, whereas it stayed unchanged in response to gefitinib (Fig. 2C). These results suggest that cyclin D1 is down-regulated by suppression of EGFR signaling at both mRNA and protein levels.

Down-regulation of cyclin D1 is correlated with sensitivity to EGFR tyrosine kinase inhibitors. Next, to ask whether down-regulation of cyclin D1 is directly correlated with sensitivity to EGFR inhibitors, we established isogenic cell lines that express EGFR-Del747-752 (HCC827/Del) or EGFR-Del747-752-T790M (HCC827/Del-TM) in a lung adenocarcinoma cell line, HCC827. HCC827 is heterozygous for the E746_A750 mutation in exon 19 of the EGFR gene and is very sensitive to gefitinib and erlotinib treatment (15, 16). We used Del747-752 and Del747-752-T790M constructs because they were found in a patient resistant to tyrosine kinase inhibitors (TKI; ref. 9). The EGFR constructs were engineered with a hemagglutinin tag at the COOH-terminal tail so that ectopic and endogenous EGFR could be distinguished (Fig. 3A). As shown in Fig. 3B, MTS assays showed that the introduction of T790M led to resistance to erlotinib. In this system, T790M receptors likely lead to resistance either by forming heterodimers with endogenous EGFR and/or through the formation of homodimers. In contrast, proliferation was still compromised in response to CL-387,785, possibly due to cell cycle arrest. To examine whether this growth inhibition was correlated with down-regulation of cyclin D1 as well as the phosphorylation of EGFR and its main downstream signaling effectors AKT and ERK1/2, we did Western blot analysis. In HCC827/Del cells, erlotinib inhibited cyclin D1 expression as well as phosphorylation of EGFR and its examined downstream signaling effectors (Fig. 3C,, left). In HCC827/Del-TM cells, erlotinib led to only modest inhibition of cyclin D1 expression and EGFR, AKT, and ERK1/2 phosphorylation, whereas CL-387,785 potently inhibited cyclin D1 expression and EGFR, AKT, and ERK1/2 phosphorylation in HCC827/Del-TM cells (Fig. 3C,, middle and right). Identical results were obtained when these cells were treated with gefitinib (data not shown). The down-regulation of cyclin D1 as well as phosphorylation of EGFR and its downstream targets correlated with the results of the growth inhibition assays done (Fig. 3B), suggesting that the efficacy of TKIs is at least in part associated with decreased cyclin D1 expression.

Cyclin D1 is up-regulated in EGFR-mutant NSCLC cell lines. The down-regulation of cyclin D1 in response to EGFR inhibition in H1975 cells suggests that the cyclin D1-CDK4/6 axis is critical in EGFR-mutant NSCLC. Consistent with this hypothesis, cell lines harboring EGFR mutations (HCC827 and H1975) had higher levels of cyclin D1 expression when compared with the cell lines expressing wild-type EGFR (A549 and H460; Fig. 4A). Interestingly, serum starvation led to down-regulation of cyclin D1 only in cells with mutant EGFR (Fig. 4A). In addition, EGF-induced cyclin D1 expression was inhibited by CL-387,785, which was correlated with inhibition of EGFR downstream molecules, including AKT and ERK (Fig. 4B).

EGFR-mutant lung cancer cells are routinely sensitive to the CDK inhibitor flavopiridol. The data in Figs. 3C and 4C suggest that cyclin D1 is a downstream target of mutant EGFR signaling. Therefore, attenuation of cyclin D1 function and/or expression may lead to proliferation arrest and apoptosis. To test this hypothesis, we treated EGFR-mutant cell lines with flavopiridol, a pan-CDK inhibitor that inhibits both cell cycle and transcriptional CDKs (37). Flavopiridol treatment resulted in dephosphorylation of the COOH-terminal domain of RNA Pol II at Ser² and Ser⁵ in all NSCLC cells tested, suggesting uniform drug uptake (Fig. 5A). The treatment of EGFR-mutant NSCLC cell lines with flavopiridol resulted in substantial apoptosis (Fig. 5B and C), whereas the same treatment induced only minimal apoptosis in A549 and Calu-1 cells, which express wild-type EGFR. These data suggest that the cyclin D1-CDK4/6 axis plays a pivotal role in EGFR-mutant NSCLC and also suggests that cyclin D1 may serve as an alternative target of therapy.

The reversible anilinoquinazoline EGFR TKIs erlotinib and gefitinib have shown dramatic responses in certain subsets of patients with NSCLC (38, 39). Recurrent, oncogenic mutations of the EGFR gene were recently identified in such patients at high frequency and it seems that responsiveness to this class of agents strongly correlates with the presence of such somatic EGFR mutations (3–5, 40, 41). These EGFR mutants have increased and prolonged tyrosine kinase activity in response to the ligand (EGF), are exquisitely sensitive to EGFR TKIs (3, 4), and are oncogenic in transgenic mouse models (42, 43). The immediate effects of EGFR activation occur through the activation of downstream signaling pathways, such as the activation of STAT/AKT/MAPK pathways (44). EGFR-mediated signaling is ultimately transduced to the nucleus via events such as STAT translocation and activation of downstream transcriptional effectors. The actual transcriptional signature of EGFR and, in particular, oncogenic, mutant-EGFR signaling remains poorly defined. Better understanding of such transcriptional changes might identify novel critical effectors of EGFR activation/blockade and provide new targets for therapeutic interventions.

Clinical resistance to such reversible inhibitors often develops through the emergence of the T790M secondary EGFR mutation (9, 10). However, irreversible anilinoquinazoline EGFR inhibitors, such as CL-387,785 and HKI-272, can overcome gefitinib/erlotinib resistance caused by the T790M mutation both in vitro and in mouse models (9, 11, 45, 46). Because novel resistance mechanisms against these irreversible EGFR inhibitors will likely occur in the long run (12, 13), the identification of new targets remains a very high priority.

Our goal was to identify such novel targets by examining the transcriptional signature of gefitinib versus CL-387,785–treated L858R/T790M double-mutant, gefitinib-resistant H1975 cells. We showed that only a very select number of genes were altered early, i.e., at 6 hours following EGFR blockade by CL-387,785. These genes included a number of known EGFR targets validating our results, such as vascular endothelial growth factor, AP-1 family members, and cyclin D1. This analysis also showed the presence of both negative as well as positive feedback loops in EGFR signaling, such as the EGFR blockade–induced down-regulation of EGF agonists such as TGFα and the down-regulation of a number of DUSPs that play a role in dampening MAPK activation. The late, i.e., 24-hour, signature included a much larger number of genes corresponding to a prominent proliferation arrest signature. This signature included suppression of E2F target genes. Taken together, the data imply that cyclin D1 down-regulation results in E2F inhibition with subsequent proliferation arrest and ultimate apoptosis. We confirmed cyclin D1 down-regulation at both the RNA and protein level and also showed that down-regulation of cyclin D1 was strongly correlated with sensitivity to EGFR inhibitors in isogenic HCC827 stable cell lines as well as H1975 cells. Our results indicate that cyclin D1 is down-regulated by suppression of EGFR signaling, resulting in G₁ arrest at the G₁ phase and subsequent apoptosis. Our results support that reduction of cyclin D1 expression may be a sensitive marker of TKI response. Of note, however, there was only mild growth inhibition of HCC827/Del-TM cells treated with <1 μmol/L CL387,785, despite the reduction in phospho-EGFR, cyclin D1, phospho-AKT, and phospho-ERK (Fig. 3B and C). These results suggest the presence of other prosurvival signals that contribute to resistance to EGFR inhibition. Further work will be required to identify these prosurvival factors. In addition, as shown in Fig. 4, it seems that cyclin D1 is overexpressed in lung cancer cells harboring mutant EGFR compared with cells with wild-type EGFR. It has been reported that mutant EGFR preferentially activates the AKT and STAT pathways compared with wild-type EGFR (8, 16). Indeed, STAT3 promotes uncontrolled growth and survival through dysregulation of gene expression, including cyclin D1 (47). In addition, the Ras/MAP/ERK kinase/ERK–dependent pathway is also implicated in the expression of the cyclin D1 gene (48, 49). Further work is required to clarify the mechanism by which cyclin D1 is up-regulated in mutant EGFR cells.

Progression of the early to mid G₁ phase is largely regulated by D-type cyclins, which associate with CDK4/6. We have shown that EGFR-mutant NSCLC cell lines express high levels of cyclin D1. Therefore, inhibition of the cyclin D1-CDK4/6 pathway may have therapeutic benefit in this subset of NSCLCs. Flavopiridol directly inhibits CDK4 and CDK6, and also reduces the transcription of cyclin D1 via CDK9 and CDK7 inhibition (50). Although cyclin D1 was depleted in response to flavopiridol in all of the cell lines studied, the reduction was more evident in H3255 and H1975 cells, which have higher baseline levels (data not shown). Over the 48-hour time period examined, flavopiridol induced substantial apoptosis in cell lines harboring EGFR mutation. A549 and Calu-1 cells are representative of a larger panel of EGFR wild-type NSCLC cell lines that initially undergo cell cycle arrest in response to flavopiridol, followed by limited apoptosis at later time points (i.e., 72 hours), most evident after exposure to concentrations >500 nmol/L (14). Interestingly, these cell lines both carry K-ras mutations. Such mutations have been noted to be mutually exclusive with EGFR mutations and lead to primary EGFR TKI resistance (51). It is possible that reduction of cyclin D1 in these cell lines arrests proliferation without affecting viability. Additional work will be required to determine whether K-ras mutant lung cancer cells are less dependent on cyclin D1 for survival.

Of note, early apoptosis in response to flavopiridol is not restricted to EGFR-mutant NSCLC cells (14). Therefore, analysis of additional cell lines will be necessary to confirm overall greater sensitivity of EGFR-mutant NSCLC cells to CDK inhibition. Flavopiridol has been shown to deplete a large variety of short half-life mRNAs, including those encoding antiapoptotic proteins, so that cell death may occur by mechanisms other than cyclin D1 depletion, both in EGFR mutant– and wild-type–expressing cells. Nonetheless, the transcriptional profiling done here, as well as the observed response to pharmacologic inhibition of cyclin D1–dependent kinase activity, suggests that suppression of this pathway will compromise the viability of EGFR-mutant NSCLC cells.

The identification of critical effectors of mutant EGFR, such as cyclin D1, should have immediate clinical implications. Importantly, the erlotinib/gefitinib–resistant H1975 cells remained sensitive to flavopiridol. Active inhibitors of the cyclin D1-CDK4/6 axis (52), alone or in combination with EGFR inhibitors, should be tested as a strategy to overcome or prevent resistance to EGFR TKI inhibitors in patients with EGFR-mutant NSCLCs.

Grant support: Specialized Programs of Research Excellence in Lung Cancer NIH grant PA20-CA090578-01A1 (D.G. Tenen); the Uehara Memorial Foundation (S. Kobayashi); a Career Development Award as part of the Dana-Farber/Harvard Cancer Center Specialized Programs of Research Excellence in Lung Cancer, NIH grant P20 CA90578 (T. Shimamura); NIH grant 1R01 CA116010 (M. Meyerson); and a Flight Attendant Medical Research Institute Young Clinical Scientist Award (B. Halmos).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank all members of the Tenen and Shapiro Laboratories for their helpful comments and technical advice.