Research ArticleCancer

Resistance to dual blockade of the kinases PI3K and mTOR in KRAS-mutant colorectal cancer models results in combined sensitivity to inhibition of the receptor tyrosine kinase EGFR

See allHide authors and affiliations

Science Signaling  11 Nov 2014:
Vol. 7, Issue 351, pp. ra107
DOI: 10.1126/scisignal.2005516


Targeted blockade of aberrantly activated signaling pathways is an attractive therapeutic strategy for solid tumors, but drug resistance is common. KRAS is a frequently mutated gene in human cancer but remains a challenging clinical target. Inhibitors against KRAS signaling mediators, namely, PI3K (phosphatidylinositol 3-kinase) and mTOR (mechanistic target of rapamycin), have limited clinical efficacy as single agents in KRAS-mutant colorectal cancer (CRC). We investigated potential bypass mechanisms to PI3K/mTOR inhibition in KRAS-mutant CRC. Using genetically engineered mouse model cells that had acquired resistance to the dual PI3K/mTOR small-molecule inhibitor PF-04691502, we determined with chemical library screens that inhibitors of the ERBB [epidermal growth factor receptor (EGFR)] family restored the sensitivity to PF-04691502. Although EGFR inhibitors alone have limited efficacy in reducing KRAS-mutant tumors, we found that PF-04691502 induced the abundance, phosphorylation, and activity of EGFR, ERBB2, and ERBB3 through activation of FOXO3a (forkhead box O 3a), a transcription factor inhibited by the PI3K to AKT pathway. PF-04691502 also induced a stem cell–like gene expression signature. KRAS-mutant patient-derived xenografts from mice treated with PF-04691502 had a similar gene expression signature and exhibited increased EGFR activation, suggesting that this drug-induced resistance mechanism may occur in patients. Combination therapy with dacomitinib (a pan-ERBB inhibitor) restored sensitivity to PF-04691502 in drug-resistant cells in culture and induced tumor regression in drug-resistant allografts in mice. Our findings suggest that combining PI3K/mTOR and EGFR inhibitors may improve therapeutic outcome in patients with KRAS-mutant CRC.


The development of targeted inhibitors to selectively block oncogenic driver signaling has become an increasingly large area of focus for oncology drug discovery (1). Several selective inhibitors, particularly against kinases and receptor tyrosine kinases (RTKs), have shown promising initial efficacy (26); however, with few exceptions, the duration of response is limited, drug resistance rapidly emerges, and patients develop fatally progressive disease (7, 8). This underscores the difficulty of successfully treating an adept, heterogeneous disease with a single targeted therapy, and highlights the fact that monotherapy is often not a tractable long-term therapeutic approach. Thus, alternative paradigms are needed to effectively mitigate acquired or intrinsic drug resistance, increase treatment benefit, and extend overall survival.

Resistance mechanisms can include alterations in the drug target itself, the pathway in which the target signals, or a parallel pathway that can alleviate the pressure on the cell due to blockade of the target (8, 9). Pathway alterations can be the consequence of a fixed somatic change such as a mutation or amplification, a dynamic change such as a temporary rewiring of signaling in the face of drug pressure, or a relief of a negative regulatory mechanism resulting from the drug pressure itself (10). Aspects of the cellular machinery may be altered, which lead to circumvention of drug pressure, including enhanced drug efflux pumps (11) or metabolic rewiring (12). Alternatively, drug resistance may be mediated by epigenetic reprogramming (1315), epithelial-to-mesenchymal transition (EMT) (1618), or emergence of a less differentiated, progenitor cell type (19, 20). These underlying themes provide a starting point for characterizing, understanding, and anticipating the emergence of resistance to a given targeted inhibitor.

Among the well-characterized resistance mechanisms are those that occur within the pathway that the target “driver” signals, or in a parallel pathway that can compensate for inhibition of the target pathway. Because true oncogenic drivers often elicit a strong tumor dependence on the pathway that the driver controls, leading to so-called pathway addiction (2123), it is not surprising that resistance often occurs as a result of a reestablishment of the signal on which the tumor was originally highly addicted. One such oncogenic driver that elicits a pathway addiction is mutant KRAS, which encodes the guanosine triphosphatase KRAS (24, 25). Despite decades of research, robust targeted therapies to directly affect mutant KRAS activity are lacking. As an alternative, several attempts have been made to target pathways known to be operative in the context of mutant KRAS. KRAS is directly implicated in the simplified linear RAS–RAF–MEK (mitogen-activated or extracellular signal–regulated protein kinase kinase)–ERK (extracellular signal–regulated kinase) signaling axis, as well as the PI3K (phosphatidylinositol 3-kinase)–mTOR (mechanistic target of rapamycin) signaling axis (26). The PI3K/mTOR pathway is one of the well-characterized and intensely studied in cancer biology, given its frequency of deregulation and its role in pathogenesis in several tumor types (27). Accordingly, there has been a tremendous effort to develop targeted inhibitors to numerous nodes within this pathway, with several in preclinical development or active clinical trials (28, 29).

Given the generally short-lived therapeutic success of recent targeted inhibitors as single agents in the clinic, and the interest in targeting the PI3K/mTOR pathway in several tumor types including those harboring KRAS mutations, we sought to characterize mechanisms of resistance to dual PI3K/mTOR inhibition that could arise in the RAS mutant setting using preclinical models of KRAS-mutant colorectal cancer (CRC) and the inhibitor PF-04691502, which has potent and selective activity against all class I PI3Ks as well as mTOR and exhibits broad antitumor activity (30).


Kras-mutant cells acquired resistance to dual PI3K/mTOR inhibition

Cell lines from genetically engineered mouse models (GEMMs) of CRC harboring mutations in Apc, Kras, and p53 (AKP) were generated and characterized as previously described (31). AKP cell lines displayed marked dependence on Kras for survival and growth, as well as a baseline sensitivity to the dual PI3K/mTOR-targeted inhibitor PF-04691502 (31), thus making them suitable for generating and characterizing acquired resistance. Drug-naïve (parental) AKP cells were subjected to an in vitro “step-up” treatment regimen for 3 months to generate resistance to PF-04691502 (Fig. 1A); the resulting PF-04691502–resistant cultures are referred to herein as 502R cells. Resistance was confirmed by comparing IC50 (50% maximal inhibition) cell viability curves. 502R cells displayed a mean 17.4-fold increase in IC50 relative to parental (Fig. 1B). 502R cells acquired marked changes in cell morphology and proliferation patterns: Parental cells appeared uniform and epithelial, whereas 502R cells grew in mesh-like networks of spindled, elongated cells (Fig. 1C). Whereas parental and 502R cells displayed comparable proliferation kinetics in the absence of PF-04691502, only 502R cells displayed decreased but steady growth over 4 days in the presence of 1 μM PF-04691502 (Fig. 1D). Short-term treatment of parental and resistant cells with PF-04691502 showed dose-dependent decreases in the abundance of phosphorylated (p) AKT (Ser473), S6RP (Ser235–236), and eukaryotic translation initiation factor 4E–binding protein 1 (4EBP1) (Fig. 1E), ruling out the possibility of lack of target modulation in resistant cells due to drug efflux pumps or other mechanisms involving drug exposure, metabolism, potency, or selectivity. In addition, the abundance of pERK, a MAPK (mitogen-activated protein kinase) pathway marker, was minimally affected in either cell line in response to this short-term treatment (Fig. 1E). Together, these data indicate that the step-up drug treatment regimen induced acquired drug resistance and phenotypic changes in 502R cells relative to parental cells, and that, despite resistance, acute pathway-specific modulation was maintained in 502R cells in response to the drug.

Fig. 1 Generation of PI3K/mTOR inhibitor resistance.

(A) Schematic representation of step-up drug treatment regimen. An Apc, Kras, p53 mutant (AKP) GEMM CRC parental cell line (AKP T90) was treated with increasing concentrations of the PI3K/mTOR inhibitor PF-04691502, starting at the initial IC50, and arriving at greater than the initial IC90. (B) Resistance to PF-04691502 was confirmed by comparing the cell viability IC50s of parental and drug-treated cells after 3 months of step-up drug treatment. Data are means ± SE from three experiments. (C) Phase-contrast images of parental and resistant cells to display changes in morphology. Scale bars, 100 μm. (D) Proliferation kinetics of parental and resistant cells in the presence of PF-04691502 or vehicle [dimethyl sulfoxide (DMSO)]. CTG, CellTiter-Glo. Data are means ± SE from three experiments. (E) Western blots of lysates from cells after acute 3-hour treatments with PF-04691502 (PF502) to confirm pathway modulation. Blots are representative of two experiments.

PF-04691502–resistant cells exhibit dynamic signaling

To gain an understanding of the potential dynamic mechanisms of drug resistance to PF-04691502, we compared phospho-signaling in parental (drug-naïve) and 502R cells with continued drug exposure (502R “drug on”) as well as drug exposure followed by 48 hours of drug withdrawal (502R “drug off”) using phospho-RTK arrays. The abundances of phosphorylated ERBB family members epidermal growth factor receptor (EGFR), ERBB2 (also known as HER2), and ERBB3 (also known as HER3) were increased in the resistant cells cultured under continuous exposure to the drug compared to parental cells, and these phosphorylated markers in resistant cells returned to parental levels when the drug was withdrawn (Fig. 2A). To examine this at the gene expression level, the abundances of transcripts encoding ERBBs, as well as IGF1R (insulin-like growth factor receptor 1) and INSR (insulin receptor), RTKs shown previously to be dynamically regulated by AKT pathway inhibition (32), were assessed in the presence or absence of PF-04691502. The amounts of EGFR, ERBB2, ERBB3, INSR, and IGF1R mRNA were all increased in 502R cells cultured in PF-04691502, with the expression of EGFR, ERBB3, and IGF1R reaching significance [consistent with the increase in the phosphorylated fraction in the phospho-RTK array (Fig. 2A)] and the expression of EGFR showing the greatest increase (Fig. 2B). The abundance of some phosphorylated RTKs decreased in 502R cells in response to PF-04691502 withdrawal (Fig. 2A), and a coordinate decrease was also observed at the mRNA level for all RTKs assessed, with EGFR, ERBB3, IGF1R, and INSR reaching significance (Fig. 2B), suggesting that the induction of RTKs in resistant cells was a dynamic response to drug pressure. Further, these data suggest that there may be an enrichment of cells with higher basal EGFR expression in resistant cultures, or that resistant cells have an increased capacity to transcriptionally up-regulate RTKs. Together, these findings indicate a potential role for the induction of ERBB family members and other RTKs in providing compensatory signaling in the presence of the PI3K/mTOR inhibitor, and that these changes were tightly regulated by drug pressure.

Fig. 2 Characterization of dynamic signaling in PF-04691502–resistant cells.

(A) Phospho-RTK arrays comparing untreated parental cells (Par) with 502R cells treated with PF-04691502 continuously (drug on) or after a 48-hour withdrawal (drug off). (B) Quantitative real-time polymerase chain reaction (qRT-PCR) to assess the abundance of a panel of mRNAs encoding RTKs in parental and 502R cells treated as in (A). Data are means ± SE from three experiments. (C) Western blots comparing the effects of siFOXO3a transfection in parental and 502R cells compared with cells transfected with nontargeting control siRNA (siNT). After siRNA treatment, cells were then treated with DMSO or PF-04691502 for 48 hours. Blots are representative of three experiments.

FOXO3a mediates dynamic ERBB family induction in PF-04691502–resistant cells

Given the dynamic induction of ERBBs in 502R cells in response to PF-04691502, we sought to determine the mechanism of this regulation. FOXO (forkhead box O) transcription factors are the target of feedback suppression of several RTKs (32). In addition, FOXO3a has a role in the induction of metastatic potential in response to PI3K or AKT inhibition (33). AKT phosphorylates FOXO proteins, resulting in their sequestration from the nucleus and, hence, inhibition of its transcriptional activity (34). Given this role of AKT in negatively regulating FOXO proteins, and the recent finding that FOXO3a plays a key role in regulating the transcription of a panel of genes encoding RTKs in response to an AKT-specific inhibitor (32), we assessed the role of FOXO3a in 502R cells using RNA interference technology. 502R cells displayed a mean ~1.5-fold increase in basal FOXO3a at both the mRNA and protein levels compared to parental cells (fig. S1, A and B). In both parental and resistant cells, siFOXO3a decreased Foxo3a expression (fig. S1A). One readout of FOXO regulation is its phosphorylation at Thr32, which is indicative of its inactive, cytoplasm-localized state (34). As expected, the abundance of pFOXO3a decreased upon PF-04691502 exposure (fig. S1B). In response to small interfering RNA (siRNA) against Foxo3a (siFOXO3a) under basal conditions, the abundance of pAKT (Ser473), a marker of PI3K/mTOR pathway activity, did not decrease in parental or resistant cells (Fig. 2C); however, in response to PF-04691502 exposure, the abundance of pAKT decreased substantially in parental and 502R cells regardless of siFOXO3a transfection (Fig. 2C, lane 1 against lanes 3 and 4, and lane 5 against lanes 7 and 8). The phosphorylation of ERK, a marker of MAPK pathway activity, was nearly ablated in parental cells but maintained in 502R cells in the presence of the drug (Fig. 2C, lanes 5 and 6 against lanes 7 and 8, respectively), suggesting that this pathway became activated in response to PI3K/mTOR pathway inhibition. Consistent with the maintained abundance of pERK in 502R cells in the presence of the drug, there was a coordinate increase in the abundance of EGFR (Fig. 2C, lane 5 against lane 7). However, the drug-induced increases in EGFR and pERK abundance were abrogated by knockdown of FOXO3a in 502R cells (Fig. 2C, lane 7 against lane 8). Together, these findings suggest that a potential mechanism of acquired resistance to PF-04691502 in KRAS-mutant mouse CRC cells is sustained MAPK signaling triggered by a FOXO3a-mediated increase in the abundance of EGFR and downstream activation of ERK.

EGFR is a mechanism of resistance to PF-04691502

Given the potential contribution of EGFR induction as a mechanism of resistance to PF-04691502, we profiled its abundance in individual subclones derived from the pooled 502R population cultured under continuous drug exposure. Indeed, compared to parental cells, both total EGFR and pEGFR abundances were markedly increased in all 12 subclones (Fig. 3A). A survey of four representative subclones indicated decreased sensitivity to PF-04691502 relative to parental cells (Fig. 3B), confirming that these 502R subclone cultures retained resistance.

Fig. 3 EGFR induction is observed across several PF-04691502–resistant subclones.

(A) Western blots to assess EGFR abundance and phosphorylation in GEMM AKP parental cells, the 502R pool, and 12 subclones from the 502R pool. Blots are representative of two experiments. (B) Cell viability curves in response to PF-04691502 in parental, 502R pool, and four representative 502R subclones. Data are means ± SE from three experiments. (C) Western blotting for EGFR and pERK in GEMM AKP parental cells transduced with a low or high amount of an MSCV-EGFR virus construct. Blots are representative of two experiments. (D) Cell viability in response to PF-04691502 in parental and MSCV-EGFR transduced cells. Data are means ± SE from three experiments. (E and F) FACS analysis of EGFR positivity in untreated (E) or PF-04691502–treated (0.5 μM for 7 days) (F) parental cells. Data are representative of three experiments.

To determine whether increased EGFR abundance was sufficient to cause resistance to PF-04691502, we engineered parental cells to stably express varying amounts of EGFR using a retroviral construct (MSCV-EGFR). Parental cells transduced with MSCV-EGFR displayed an increase in both total EGFR and pEGFR, as well as increased abundance of the downstream marker, pERK (Fig. 3C). These increases correlated with a minor increase in resistance to PF-04691502, with the “high EGFR”–expressing line displaying a slightly greater shift in IC50 (Fig. 3D). However, this ectopic expression was not sufficient to reach the levels of resistance seen in the 502R pool or the resistant subclones, in which IC50s range from 0.64 to 0.93 μM (Fig. 3B), suggesting that additional changes likely occurred during the process of acquired resistance in response to continued drug pressure.

To test the hypothesis that an EGFR-expressing subpopulation is increased in parental cells over time in response to PF-04691502, we treated parental cells with 0.5 μM PF-04691502, a dose comparable to the IC90 for 7 days, stained the cells with a fluorescein isothiocyanate (FITC)–labeled antibody detecting an external portion of EGFR, and subjected the cells to FACS (fluorescence-activated cell sorting) analysis. In untreated parental cells, 2.1% of cells were positive for EGFR, whereas 30.2% of PF-04691502–treated parental cells were positive for EGFR (Fig. 3, E and F), suggesting that continuous drug treatment increased the abundance of EGFR. Together, these data indicate that even acute PI3K/mTOR inhibition induces EGFR and that increased EGFR may be a mechanism of resistance in our model system, but that additional alterations may further drive resistance to PF-04691502.

PF-04691502–resistant cells are sensitive to ERBB inhibition

To rapidly identify potential vulnerabilities that arise in the resistant context, we subjected parental and 502R cells to a focused small-molecule library screen targeting known or putative druggable kinase targets, with 502R cells cultured in the continued presence of PF-04691502 (fig. S2). Compounds that inhibited viability by ≥2-fold in 502R cells compared to parental cells were considered for follow-up. Among such hits were compounds that targeted EGFR or ERBB2, members of the ERBB family. This was initially somewhat unexpected given that these cells harbor a KRAS mutation, and clinical responses to EGFR inhibitors in KRAS-mutant CRC patients are poor (35); however, these findings were consistent with the dynamic increases in ERBB family members that we observed in 502R cells (Fig. 2). To validate these hits, we treated parental and 502R with a larger panel of EGFR and ERBB family inhibitors in cell viability assays in two-dimensional (2D) adherent or 3D spheroid cultures in the presence of PF-04691502. 502R cells displayed a substantial increase in sensitivity to all EGFR and other ERBB family inhibitors tested (Table 1). The irreversible pan-ERBB inhibitor dacomitinib (PF-00299804) (36) produced the greatest increase in sensitivity.

Table 1 Confirmation of hits from the chemical sensitivity screen.

Hits selected from the sensitivity screen (fig. S3) were selected for follow-up validation. Shown are IC50 values for parental (Par) and 502R cells cultured under 2D or 3D conditions.

View this table:

Combined inhibition of PI3K/mTOR and ERBB signaling decreases CRC cell viability

Given our findings suggesting that ERBB signaling mediated resistance to PI3K/mTOR inhibition in CRC cells, we assessed the effects of combining PF-04691502 with the irreversible pan-ERBB inhibitor PF-00299804 (dacomitinib) (36). To this end, we assessed the effects of adding PF-00299804 at a fixed concentration of 40 nM, a dose in culture that is comparable to the achievable therapeutic exposure in patients (37), on the cytotoxicity of PF-04691502 in parental and 502R cells. Whereas the addition of PF-00299804 slightly increased sensitivity to PF-04691502 in parental cells (Fig. 4A), it reversed resistance in 502R cells, decreasing the IC50 to PF-04691502 by >10-fold to 108 nM (Fig. 4B), a level comparable to the parental (drug-naïve) cells (Fig. 4A). These results indicate that acquired resistance to PF-04691502 (PI3K/mTOR inhibition) leads to a far more robust combination effect with PF-00299804 (EGFR inhibition).

Fig. 4 ERBB family inhibition enhances the cytotoxicity of PI3K/mTOR inhibition.

(A and B) Dose-response curves in parental (A) and 502R cells (B) treated with PF-04691502 alone (red) or in combination with a fixed dose (40 nM) of the ERBB inhibitor PF-00299804 (black). Data are representative of three experiments. (C) Western blotting for markers of pathway activity as indicated and apoptosis induction (Bim) in parental (P) and 502R (R) cells under basal culture conditions (none) or in response to a 1 μM dose (Rx) for 24 hours of PF-04691502 or PF-00299804 alone (502 or 804, respectively) or in combination (502 + 804). Blots are representative of two experiments.

To elucidate the acute effects of the combination on signaling and to confirm whether apoptosis may have been elicited, we assessed MAPK and PI3K pathway activity markers and the proapoptotic marker Bim in parental and 502R cells treated with each drug individually or in combination. As shown above, 502R cells had increased basal abundance of pEGFR and total EGFR, as well as increased pERK, compared with parental cells (Fig. 4C, lanes 1 and 2). Relative to vehicle, the PI3K/mTOR inhibitor PF-04691502 markedly decreased the abundance of pAKT, pS6, and p4EBP1 in parental cells (Fig. 4C, lanes 1 and 3) and 502R cells (Fig. 4C, lanes 2 and 4). The ERBB inhibitor PF-00299804 expectedly elicited a marked decrease of pEGFR in 502R cells (Fig. 4C, lanes 2 and 6) and a slight decrease in parental cells, in which pEGFR abundance was already basally low (Fig. 4C, lanes 1 and 5). In addition, PF-00299804 reduced the abundance of pERK in both parental and 502R cells (Fig. 4C, lanes 1 and 2 compared with lanes 5 and 6, respectively). PF-00299804 treatment also partially decreased the abundance of pAKT and pS6, whereas that of p4EBP1 was unaffected, suggesting that the ERBB inhibitor had a partial inhibitory effect on the PI3K/mTOR pathway at the dose and time point observed. In response to PF-00299804, induction of the proapoptotic marker Bim was seen in parental and 502R cells (Fig. 4C, lanes 1 and 2 compared with lanes 5 and 6, respectively). In response to a combination of PI3K/mTOR (PF-04691502) and ERBB (PF-00299804) inhibition (Fig. 4C, lanes 7 and 8), the abundance of MAPK and PI3K signaling markers decreased to below detectable limits in both parental and 502R cells, whereas that of Bim increased, consistent with the combined cytotoxicity observed (Fig. 4, A and B). Together, these findings suggest that the rational combination of PF-04691502 and PF-00299804 is effective at extinguishing both the PI3K and MAPK pathways and may induce an apoptotic response in both parental and drug-resistant cells.

Combination of PF-04691502 and dacomitinib induces tumor regression in vivo

Given the combination efficacy in vitro, we tested the potential growth inhibitory effects in vivo using subcutaneous xenografts. Mice were injected subcutaneously with 502R cells from the in vitro step-up resistance procedure and treated daily with PF-04691502, PF-00299804, or the combination after tumors became established. Neither PF-04691502 nor PF-00299804 alone had any growth inhibitory effect; however, combination treatment produced ~48% tumor regression (Fig. 5A), consistent with the restoration of tumor sensitivity by the combination in vitro. Tumor lysates were extracted at the end of treatment and assessed by Western blot. Tumors from mice treated with PF-04691502 had increased total EGFR compared with tumors from vehicle-treated mice (Fig. 5B), which was consistent with our in vitro findings of increased EGFR in 502R cells compared to parental; an increase in pEGFR was not observed. A coordinate increase in the abundance of pERK was also observed in tumors from PF-04691502–treated mice, again suggesting the activation of ERBB/MAPK signaling after PI3K/mTOR inhibition and the possibility of its role as a mechanism of resistance in vivo. Conversely, in tumors from mice treated with the pan-EGFR inhibitor PF-00299804, the abundance of pERK was decreased compared to those from vehicle-treated mice, whereas the PI3K/AKT pathway marker pS6 was increased (Fig. 5B), suggesting potential reciprocal compensation from these pathways in vivo. However, in tumors from mice treated with both inhibitors, both PI3K/AKT and ERBB/MAPK pathway markers [pAKT (Thr308) and pS6, and pEGFR and pERK, respectively] were markedly reduced. Consistent with tumor regression (Fig. 5A), the proapoptotic marker cleaved PARP [poly(ADP-ribose) polymerase] was increased (Fig. 5B) in tumors from mice treated with the combination. Together, these data further support the potential utility of a combination strategy using PI3K/mTOR and ERBB family inhibitors in KRAS-mutant CRC.

Fig. 5 Combination therapy in vivo results in regression of 502R tumors.

(A) In vivo tumor growth inhibition analysis of GEMM 502R allografts in response to daily single or combination treatment. Data are means ± SE from six to eight mice each. (B) Western blots assessing pathway modulation in tumor lysates from mice receiving 14 days of daily single or combination therapy. Blots are representative of two experiments.

RNA sequencing reveals transcriptional changes associated with adaptive resistance to PF-04691502

To further characterize drug-resistant (502R) cells, distinguish fixed versus dynamic changes in transcriptional programs, and identify additional pathways altered in the resistant context, we performed transcriptome profiling using RNA sequencing in parental cells, 502R-“on” cells (which received continuous drug exposure), and 502R-“off” cells (which were released from the drug for 48 hours) (fig. S3). The expression of 465 genes was ≥1.5-fold greater in 502R-on cells than in parental cells and was similarly decreased by drug withdrawal in 502R-off cells compared to 502R-on cells (table S1). Gene set enrichment analysis identified canonical signaling pathways that had the highest enrichment of these “dynamic resistance genes” and included both ERBB signaling, which supports our findings regarding dynamic regulation of EGFR and ERBB family members (Fig. 2), and EMT signaling, which is a mechanism of resistance to targeted pathway inhibition in other tumor types (16) (table S2). In addition, top physiological systems represented in the dynamic expression signature included embryonic and organ/tissue development (table S3). Among the dynamic resistance genes with the greatest degree of differential regulation was Aldh1a1. Aldh1a1 (aldehyde dehydrogenase 1 family, member A1) is associated with stem cell–like characteristics in malignant and nonmalignant tissues (38), is a marker of tumor-initiating cells in various tumor types (including prostate, melanoma, pancreatic, and colon cancers) (3943), and is a predictive marker of drug resistance (44, 45). Aldh1a1 mRNA was increased ~12-fold in parental cells in response to 48 hours of exposure to PF-04691502 (Fig. 6A). In the absence of PF-04691502, 502R had >2000-fold greater Aldh1a1 mRNA abundance than parental cells, and this amount was further increased modestly in the presence of the drug (Fig. 6A). We hypothesized that parental cells display the capacity to up-regulate Aldh1a1 expression in response to acute drug pressure, but that the markedly increased basal Aldh1a1 expression in 502R reflects prolonged selective pressure for a subpopulation of cells with characteristics of stem-like progenitors. To test the hypothesis, we subjected parental cells to 5 days of treatment with PF-04691502 and used the ALDEFLUOR assay to assess Aldh1a1 activity, a readout of stem and progenitor cells in various lineages (46). Consistent with the increased transcript abundance of Aldh1a, PF-04691502 treatment increased ~5-fold the percentage of cells exhibiting Aldh1a1 activity (Fig. 6B). This finding may suggest that drug pressure either promotes expansion of a preexisting subpopulation of cells with high basal Aldh1a1 activity or induces the induction and activation of Aldh1a1 in a subpopulation of cells, both of which are feasible given the duration of treatment.

Fig. 6 Aldh1a1 and WNT pathway activity are increased in 502R cells.

(A) qRT-PCR comparing Aldh1a1 expression in GEMM parental or 502R cells cultured for 48 hours in the presence of DMSO or PF-04691502. (B) ALDEFLUOR-FACS assay detecting Aldh1a1 activity in GEMM parental cells cultured with DMSO or PF-04691502. (C) WNT pathway activity as determined by TOP-Flash assay in GEMM parental and 502R cells cultured for 48 hours in the presence of DMSO or PF-04691502. (D) Western blots for basal abundance of the indicated proteins using nuclear- or cytoplasmic-enriched fractions from parental and 502R cells. Data in (A) and (C) are means ± SE from three experiments. Blots and data in (B) and (D) are each representative of two experiments.

Resistant cells display increased WNT–β-catenin transcriptional activity

Given that aberrant WNT pathway regulation has been characterized in a broad spectrum of cancers, including tumor-initiating cells in colon cancer (47, 48), we chose to assess the status of this pathway in parental and 502R cells. To this end, we mined the RNA-Seq data for differential expression of canonical WNT target genes. We observed that several such genes were increased >1.5-fold in 502R compared to parental cells (table S1), suggesting that WNT pathway activity was increased in 502R cells. Consistent with this hypothesis, vehicle-treated 502R cells had greater TOP-Flash reporter activity (Fig. 6C) and greater abundance of total and active β-catenin (CTNBB1) (Fig. 6D) than did parental cells. PF-04691502 increased WNT reporter activity in both parental and 502R cells (Fig. 6C). In addition, 502R cells displayed a marked decrease in E-cadherin abundance (Fig. 6D), further supporting the possibility that these cells have undergone an EMT. Together, these data indicate that 502R cells have increased basal WNT pathway activity, which is further amplified under drug pressure, consistent with their increased stem-like characteristics.

Characteristics of PF-04691502 resistance were recapitulated in a human CRC cell line

To determine whether the hallmark characteristics of PF-04691502 resistance identified in our GEMM could be recapitulated in human models, we subjected the human KRAS-mutant CRC cell line HCT116 to PF-04691502 using the same step-up drug exposure regimen performed with GEMM cells (Fig. 1A). The resulting resistant line is referred to herein as HCT116-502R. PI3K/mTOR markers pS6 and pAKT (Ser473) were decreased in the presence of PF-04691502 (fig. S4A), indicating that the drug was still effectively blocking the PI3K/mTOR pathway in the resistant cultures. Consistent with our findings in the GEMM, in the presence of PF-04691502, the abundance of EGFR was increased and pFOXO3a was decreased in HCT116-502R cells, and both effects were reversed upon drug withdrawal (fig. S4A), indicating that EGFR abundance in HCT116-502R cells was also regulated by the drug-induced release of FOXO3a phosphorylation. However, EGFR abundance detected in the parental HCT116 cells was greater than that detected in the GEMM, and the magnitude of up-regulation was coordinately smaller. Therefore, to gain a better understanding of the spectrum of RTKs modulated by PF-04691502 pressure in HCT116-502R cells, we performed phospho-RTK arrays in parental and drug-resistant cultures. Consistent with the GEMM findings, we again observed a dynamic increase in pEGFR in response to drug exposure in HCT116-502R cells (fig. S4B). However, several additional RTKs were increased in the human model, perhaps owing to the increased heterogeneity of the established human cell line compared to the genetically simplified GEMM. Given the dynamic regulation of EGFR seen in the human model, we again assessed the combination of PF-04691502 with the pan-ERBB inhibitor PF-00299804. Compared to treatment with PF-00299804 alone, both HCT116 and HCT116-502R cells treated with the combination had increased sensitivity (fig. S4C). Whereas this finding was not entirely consistent with what was observed in the GEMM 502R cells, which displayed a greater sensitivity to the combination compared to parental, the finding that resistant cells displayed sensitivity to the combination was consistent. Together, these data indicate that the GEMM resistance model predicted one hallmark resistance marker (ERBB), which was also observed in the human model, and this was consistent with sensitivity to the rational combination of PF-04691502 and the pan-ERBB inhibitor PF-00299804.

Characteristics of PF-04691502 resistance in cells were recapitulated in patient-derived xenografts

As one approach to test the clinical relevance of the potentially novel resistance mechanisms to PF-04691502 described herein in the KRAS-mutant CRC model, we used patient-derived xenografts (PDXs) originating from KRAS-mutant CRC patients with advanced disease. PF-04691502 produced tumor growth inhibition compared to vehicle, but tumors still steadily grew (Fig. 7A), despite appreciably decreased abundance of pAKT and pS6 in the initial days of treatment (fig. S5). To determine whether the putative markers of dynamic resistance discovered in our PF-04691502 CRC resistance models herein were also present in these PDXs in response to continued drug pressure, we profiled PDX tumors from mice that received 28 days of treatment, as well as those from mice that received 28 days of treatment followed by 14 days of drug withdrawal. Consistent with our cell culture findings, the abundances of EGFR and ALDH1A1 were markedly increased in PDXs after 28 days of PF-04691502 treatment, and the abundance of each declined 14 days after drug withdrawal (Fig. 7B). Consistent with these findings, EGFR and ALDH1A1 stainings were increased in tumor sections taken from mice treated for 28 days with PF-04691502 compared to those treated with vehicle (Fig. 7C). Together, these results implicate a role for EGFR and stem cell–like characteristics in the resistance to PI3K/mTOR inhibition in human CRC.

Fig. 7 Dynamic EGFR and Aldh1a1 regulation is recapitulated in human PDX models.

(A) Tumor growth inhibition assessment in a KRAS-mutant PDX model in response to 28 days of treatment with vehicle (group a, 10 mice), PF-04691502 (group b, 16 mice), or PF-04691502 followed by 14 days of drug withdrawal (group c, 6 mice). Data are means ± SD. (B) Western blots of representative tumors (2 per group) to assess resistance markers after treatment as indicated [same groups as in (A)]. (C) Tumor section immunohistochemistry after mice were treated for 28 days with either vehicle or PF-04691502. Sections were stained with EGFR or Aldh1a1 (green), Ki-67 (red), or 4′,6-diamidino-2-phenylindole (DAPI, blue). Corresponding serial sections were also stained with hematoxylin and eosin (H&E). Images are representative of six tumors. Scale bars, 100 μm.


In the present study, we sought to identify mechanisms of resistance to targeted PI3K/mTOR inhibition in models of KRAS-mutant CRC because several such inhibitors are in clinical development aimed at treating this disease segment. To this end, we used our recently characterized Kras-mutant CRC GEMM as well as human KRAS-mutant CRC cell lines and PDX material. Although each model contains inherent limitations, we have gained appreciable insight into potential underlying mechanisms of resistance that could be anticipated in response to targeted inhibition of the PI3K-mTOR axis in this disease segment. Despite their limited heterogeneity and complexity relative to human tumors, GEMMs have previously been used by several groups as part of comprehensive drug discovery and resistance characterization efforts, which have yielded insights into resistance mechanisms that can be anticipated in response to targeted pathway inhibition (49, 50). We used our Kras-mutant GEMM given the defined genetic background, aggressive phenotype, recapitulation of several hallmarks of human KRAS-mutant CRC, baseline sensitivity to PF-04691502 (31), and rapid acquisition of resistance. This model acquired resistance in vitro after 3 months of increasing drug pressure. In parallel, we initiated the same experimental resistance using the KRAS-mutant CRC line HCT116. It should be noted that these cells required significantly more time to acquire resistance to PF-04691502, and so most of screening and follow-up validation was performed in the GEMM; nonetheless, the limited characterization work that we performed in HCT116 502R partially supported our findings in the GEMM. Chemical screens to identify compounds that were effective at restoring sensitivity to PF-04691502 identified inhibitors of the ERBB family of RTKs, and combination sensitivity was confirmed in viability assays. These results were intriguing, given that ERBB targeting agents have shown promise in a subset of KRAS wild-type advanced CRC (51, 52), yet previous studies have implicated that KRAS mutation status can be used as an exclusion criterion when therapeutically targeting ERBB (35, 53). The data presented herein suggest that there may be opportunities in targeting this family in KRAS-mutant tumors in the context of combination with PI3K/mTOR inhibition. The notion of a combination including an ERBB inhibitor in the context of mutant KRAS was also elucidated in a recent study (54) in which KRAS mutations were acquired in CRC cell lines in response to EGFR blockade, as well as plasma from patients in response to prolonged EGFR antibody treatment, and combined EGFR and MEK inhibition circumvented the resistance.

Acquired sensitivity to EGFR inhibition in our GEMM and human PI3K/mTOR resistance models was supported by the fact that these cells displayed the ability to dynamically regulate EGFR in response to the inhibitor, with EGFR displaying robust increases at the protein and mRNA levels in the presence of the drug and decrease when the drug was withdrawn. These findings are consistent with previous studies implicating compensatory activation of growth factor signaling through RTKs in response to kinase inhibitors (55), as well as induction of RTKs specifically in response to AKT inhibition, which relieves AKT-mediated feedback suppression of the transcriptional regulator FOXO and enables its translocation into the nucleus (32). Indeed, FOXO3a knockdown in our model resulted in a decrease of drug-induced EGFR activation. Further, AKT and PI3K inhibitors have recently been shown to activate FOXO3a, which in turn contributed to a metastatic phenotype in CRC models (33). Thus, our findings implicating a potential mechanistic role of FOXO3a in resistance to PF-04691502, along with the potential EMT phenotype observed in our resistant cells, are consistent with established roles for FOXO3a in response to inhibition of the PI3K-AKT axis. Also important to note is that our parental GEMM CRC model contains an Apc mutation, consistent with the abundant presence of mutations in the WNT pathway observed in human CRC. Given that 502R cells displayed heightened WNT pathway activity relative to parental cells, which already contain a mutation in Apc, this further highlights the potential crosstalk between the WNT pathway and FOXO in resistance to PI3K inhibitors.

The results from the genomic profiling of parental and resistant cells using RNA-Seq paired with gene set enrichment analysis identified ERBB signaling as a top pathway that was dynamically regulated in response to drug pressure and withdrawal. In addition, several other interesting signaling programs were noted, including EMT and an increase in markers of progenitor-like tumor-initiating cells. In regard to EMT, previous studies have identified an increase in migration, invasion, and EMT characteristics in cell lines generated to be resistant to cytotoxic agents such as oxaliplatin (18) or to targeted pathway inhibitors such as bevacizumab (17). In addition, previous studies have implicated the potential contribution of a subpopulation of stem-like progenitor cells in resistance to both chemotherapeutic and targeted therapies (20, 5658). Various groups have characterized the drug-resistant aspects of such stem-like subpopulations, including a quiescent state refractory to agents targeting rapidly dividing cells, enhanced DNA damage repair mechanisms, and decreased apoptotic machinery (56). Recent studies have implicated a potential mechanistic link between EGFR activation and the acquisition of stem-like properties including the increase in known stem cell markers and enhanced spheroid formation (59, 60); however, the role of EGFR in promoting stem cell properties in CRC has not been fully characterized.

In our model, drug pressure resulted in the expansion of a subpopulation of cells with stem-like characteristics, including enhanced Aldh1a1 activity, a well-characterized marker of stem cells and poor prognosis in CRC and other cancer types (3942, 44, 45). Previous studies suggest that such a cell population may be inherently resistant to PI3K/mTOR inhibition. When we examined a KRAS-mutant CRC PDX model, we observed increases in EGFR and ALDH1 upon continued treatment with PF-04691502; however, in a cohort of tumors where drug pressure was removed, levels were consistent with the control treatment group, indicating that perhaps this was a dynamic response selected for by drug pressure. Whether these dynamic resistance mechanisms are the result of a progenitor-like subpopulation of cells that fluctuate along with the drug pressure or a temporary rewiring of signaling in response to drug pressure remains to be fully elucidated.

Together, these findings implicate a potential role for EGFR induction as a mechanism of resistance to PI3K/mTOR inhibition in GEMM and human cell lines and in a clinically relevant model of KRAS-mutant CRC, and highlight the utility of the GEMM as a platform for rapidly identifying such mechanisms. The increase of the progenitor marker Aldh1a1 in the PF-04691502–treated parental cells, the in vitro PF-04691502–resistant cells, and the KRAS-mutant PDX samples supports the hypothesis that drug pressure may elicit the increase of a drug-resistant subpopulation, and serves as a reminder that successful rational combinations must successfully target these highly adaptive and drug-refractory cells. In addition, these findings highlight the potential utility of combined PI3K/mTOR and ERBB family inhibition in our GEMM KRAS-mutant CRC resistant cells, and warrant further characterization of this combination in additional models. Follow-up studies will aim to establish whether there is a mechanistic link between EGFR and ALDH1 induction and stem-like characteristics in response to targeted PI3K/mTOR inhibition.


GEMM and human cell lines

The generation and characterization of tumor-derived cell lines from KRAS-mutant primary GEMM CRC tumors were described previously (31). The human KRAS-mutant CRC cell line HCT116 was acquired from the American Type Culture Collection. GEMM and human cell lines were maintained in Dulbecco’s modified Eagle’s medium (DMEM) with 10% fetal bovine serum (FBS).

Generation of resistance in vitro

To generate drug-resistant models, cell lines were subjected to a step-up drug treatment regimen. Briefly, cell viability IC50s for PF-04691502 were determined in parental drug-naïve cell lines. Cells were then subjected to PF-04691502 at approximately their respective IC50s (~40 nM), and doses were gradually increased over the course of 3 months (GEMM) or 9+ months (human) until exposures were at 1 μM (GEMM), cells reached positive growth, and cell death was not observed. Resistant cells were then maintained in 1 μM PF-04691502 indefinitely before seeding for subsequent experiments.

Cell viability assays

Cell lines were plated at 1000 cells per well in 96-well culture plates in growth medium with 10% FBS. Cells were incubated overnight and treated with DMSO (0.1% final) or serially diluted compound for 4 days. Cell viability was assessed by incubating plates with CellTiter-Glo reagent (Promega) at room temperature for 30 min. IC50 values were calculated by plotting luminescence intensity against drug concentration in nonlinear curves using GraphPad Prism (GraphPad).

In vitro growth kinetics

Cell lines were plated at 1000 cells per well in 96-well culture plates in growth medium with 10% FBS. Cells were incubated overnight and treated with DMSO (0.1% final) or PF-04691502 (1 μM final) for 4 days, and viability was assessed as above.

Chemical library screening

Parental and resistant cells were plated at 1000 cells per well in 96-well culture plates in growth medium with 10% FBS. Cells were incubated overnight and treated for 4 days with a proprietary collection of compounds with known or putative activity to oncology targets. Resistant cells were maintained in 1 μM PF-04691502 for the experiment to identify compounds that would provide a combination benefit in these cells. After 4 days, 30 μl of CellTiter-Glo (Promega) was added to indirectly measure cell viability as a measure of proliferation using an Envision multireader (Perkin-Elmer).

Pathway modulation in response to dose titration

Parental and resistant cells were plated at 150,000 to 200,000 cells per well in six-well culture plates in growth medium with 10% FBS. Cells were incubated overnight and treated with DMSO (0.1% final) or a dose titration of PF-04691502 at the indicated final concentrations for 3 hours. Cells were washed with phosphate-buffered saline (PBS) and lysed in ice-cold radioimmunoprecipitation assay (RIPA) buffer with phosphatase inhibitor cocktail (Thermo Scientific), and proteins were quantitated by the BCA (bicinchoninic acid) assay (Pierce Biotechnology). Equal amounts of protein were resolved by polyacrylamide gels (Bio-Rad), Western-blotted, and probed with primary antibodies [pAKT (Ser473), total AKT, pS6RP (Ser235–236), p4EBP1, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH)] and then with secondary antibodies (all from Cell Signaling Technology). Bands were detected by chemiluminescence, and images were captured with an AlphaImager system (Protein Simple).

Combination treatment cell viability assays

Parental and resistant cells were plated at 1000 cells per well in 96-well culture plates in growth medium with 10% FBS in standard assay plates (2D) or at 2000 cells per well in U-bottomed ultralow attachment plates (3D) (Corning Inc.) and incubated overnight. The next day, cells were treated with a 10-point dilution of the ERBB family inhibitors including dacomitinib and erlotinib. Resistant cells were maintained in a background concentration of 1 μM PF-04691502 throughout the assay to identify combination benefit. After 4 days, 30 μl of CellTiter-Glo was added to indirectly measure cell viability as a measure of proliferation, as above.

Dynamic phospho-signaling

Parental (drug-naïve) cells were maintained in DMEM, and drug-resistant cells were maintained in 1 μM PF-04691502 until experimental plating. Cells were plated at 100,000 cells per well in six-well culture plates in growth medium with 10% FBS and incubated overnight. Cells were treated with PF-04691502 for 48 hours (1 μM final, 502R drug on) or maintained in 10% FBS + 0.1% DMSO for 48 hours (drug off). Cells were lysed, and Western blots were performed as above, with the following primary antibodies: pEGFR (Tyr1068), total EGFR, pERK (Thr202/Tyr204), pAKT (Ser473), pFOXO3a (Thr32), total FOXO3a, and GAPDH (all from Cell Signaling Technology). Alternatively, lysates were probed with phospho-RTK arrays (R&D Systems). For preparation of nuclear and cytoplasmic fractions, cells were harvested and lysed with the NE-PER nuclear fractionation kit (Pierce Biotechnology) according to the manufacturer’s protocol.

Quantitative real-time PCR

Parental and resistant cells were plated and treated +/− PF-04691502 as above. Cells were lysed, and RNA was extracted according to the manufacturer’s instructions (Qiagen) and reverse-transcribed using qScript master mix (Quanta Biosciences); complementary DNA was subjected to qRT-PCR with TaqMan Universal Master Mix and TaqMan primers (Life Technologies) for the indicated genes (Fig. 2C) on an ABI 7900 machine. Relative mRNA abundance was determined by normalizing to that of GAPDH and using the ΔΔCT method, and abundance was represented as fold of the parental control.

Combination viability assays

Cells were plated at 1000 cells per well in 96-well plates and allowed to adhere overnight. Cells were then treated with increasing doses of PF-04691502 either as a single agent or in combination with a fixed dose of 40 nM PF-00299804. After 4 days, 30 μl of CellTiter-Glo (Promega) was added to measure cell viability/proliferation using an Envision multireader (Perkin-Elmer). Results were shown as percent viability relative to control (DMSO) treatment, and the absolute 50% viability line was plotted.

In vivo tumor growth inhibition

Animals were maintained under standard clean room conditions in accordance with the Pfizer Institutional Animal Care and Use Committee. Female nu/nu mice (8 to 10 weeks old) were obtained from Charles River Laboratories. Parental and 502R cells for implantation were collected, resuspended in serum-free medium mixed with Matrigel (1:1; BD Biosciences), and implanted subcutaneously into the hind flank region at 1 × 106 cells per mouse at 100-μl total volume. Treatment was started when the average tumor size was 200 to 300 mm3. PF-04691502 (5 mg/kg), PF-00299804 (4 mg/kg), or the combination was and given orally once a day for 14 days (n = 6 mice per group). Animal body weights and tumor volumes were measured every 3 to 4 days. Tumors were measured with calipers, and tumor volumes and percent tumor growth inhibition were calculated as previously described (30). Data are presented as means ± SE.

Tumor lysate preparation

Representative tumors from each treatment arm above were excised at the end of treatment (14 days after treatment), macrodissected, and snap-frozen in liquid nitrogen. Tumor chunks were thawed and added directly to ice-cold RIPA buffer supplemented with a protease and phosphatase inhibitor cocktail (Thermo Scientific) in tubes containing microbeads for sonication (MP Biomedical). Samples were sonicated in a FastPrep24 sonicator (MP Biomedical), debris was cleared by centrifugation at high speed, total protein was quantitated, and samples were used for Western blotting.


Parental cells were treated with 0.5 μM PF-04691502 or DMSO control for 5 days, washed, trypsinized, counted, and assessed for Aldh1a1 activity using the ALDEFLUOR activity kit according to the manufacturer’s instructions (STEMCELL Technologies). Cells were subjected to FACS with a Guava easyCyte flow cytometer (Millipore). Cells were first gated using forward and side scatter and then assessed for ALDEFLUOR fluorescence intensity, which corresponded with Aldh1a1 activity.


Parental cells were treated with 0.5 μM PF-04691502 or DMSO for 7 days, washed, trypsinized, counted, stained with a FITC-labeled EGFR antibody (Abcam) for 2.5 hours, and then washed in 1× PBS. Cells were subjected to FACS, first gated by forward and side scatter, and then analyzed for FITC intensity using a Guava easyCyte flow cytometer (Millipore).

In vivo PDXs

All PDX animal experiments were carried out in accordance with the guidelines of the Canadian Council on Animal Care, the regulations under the Animals for Research Act, RSO 1980, and with Animal Use Protocols approved by both the Toronto Centre for Phenogenomics (TCP) and University Health Network (UHN) Animal Care Committees. To establish sufficient quantities of patient-derived tumors of uniform size for drug therapy experiments, we determined that in vivo implantation of viable chunks of tumor tissue was preferable to the injection of single-cell suspensions. PDXs were obtained from the Princess Margaret Cancer Centre. The original patient tumor used in this study had mutations in adenomatous polyposis coli (APC) (nonsynonymous mutation A573D; stop-gain mutation R1432X), KRAS (G12V), and TP53 (R175H). Tumors were harvested and cut into forty 1 × 1–mm chunks, each of which was implanted into CB17 SCID (severe combined immunodeficient) mouse recipients via a small skin incision and creation of a subcutaneous pocket [vehicle (group a), n = 10 mice; 28 days of PF-04691502 (group b), n = 16 mice; 28 days of PF-04691502 followed by 14 days of drug withdrawal (group c), n = 6 mice]. The incision was closed with nonabsorbable sutures, which were removed 7 to 10 days after implantation. Tumors were monitored and measured with calipers three times per week. Once tumors reached an average volume of 200 to 300 mm3, the mice were randomized into treatment groups, and daily oral gavage was initiated. For acute treatment, mice were treated with vehicle or PF-04691502 (7.5 mg/kg per day) for 4 days and then sacrificed, and tumors were harvested, lysed, and subjected to Western blot analysis for pathway modulation. For extended treatment, mice were treated with vehicle or PF-04691502 (7.5 mg/kg per day) for 28 days. At day 28, a subset of PF-04691502 mice were sacrificed, and a subset were taken off drug for an additional 14 days and then sacrificed. Tumor tissue was collected and split into two portions: one snap-frozen and the other formalin-fixed and paraffin-embedded for immunohistochemistry. Snap-frozen tissue was thawed and prepared for Western blots by sonication as described above.

PDX phospho-signaling

Lysates from PDX samples treated as described above were assessed by Western blot to measure the abundances of pAKT (Ser473 or Thr308), total ERK, pERK (Thr202/Tyr204), pS6 (Ser235–236 or Ser240–244), Aldh1a1, and GAPDH (all from Cell Signaling Technology).

Resistant subclones

Pooled 502R cells were seeded at a sparse density into 24-well plates at 500 cells per well in growth medium with 10% FBS and incubated overnight, and PF-04691502 was readministered for 72 hours; sterile cloning disks were used to isolate and pick up individual colonies of cells and transfer them to fresh 24-well plates. Individual wells were then expanded into separate sublines in the presence of PF-04691502 (1 μM final) and used for downstream assays.

Ectopic EGFR expression

Parental cells were transduced with a pMSCV-driven retroviral construct encoding full-length human EGFR at multiplicities of infection of 5 and 50 (denoted as low and high, respectively) and subjected to 48 hours of selection with puromycin. Stably transduced cells were then passaged twice, and cell lysates were collected for Western blot analysis as described above.

β-Catenin transcriptional reporter assay

GEMM parental (AKP T90) and PF-04691502–resistant cells (502R) were transfected with the Top-Flash reporter system (Promega) using Lipofectamine 2000 (Life Technologies), plated onto 96-well assay plates, incubated for 24 hours at 37°C, and treated +/− 1 μM PF-04691502 for 48 hours. Cells were lysed with Bright Glo luciferase buffer and analyzed on a Wallac Trilux Microbeta luminescence counter (PerkinElmer). The raw TOP FFluc was divided by raw Rluc (cotransfect with TOP), the raw FOP FFluc was divided by raw Rluc (cotransfect with FOP), and then normalized TOP/FOP ratio was obtained.

RNA-Seq data analysis

mRNA samples were prepared in triplicate for the three sample groups: parental, 502R (drug on), and 502R (drug off, drug removed for 48 hours), and sequenced using the Illumina HiSeq 2000 to generate paired-end, 75-bp reads (~44.1 million read pairs per sample). Sequence reads were quality controlled and mapped to the reference mouse genome sequence (NCBI37/mm9 assembly), using TopHat (61). The resulting alignments were assembled into gene transcripts based on a junction library file containing the mapped coordinates of mouse RefSeq genes from UCSC (University of California Santa Cruz), and their expression levels were estimated in fragments per kilobase per million (FPKM) values using Cufflinks (62). Differentially expressed genes in each treated group were identified with fold change and P value by Cuffdiff (62), using the parental sample group as reference. The raw data (RNA-Seq reads) were submitted to the European Nucleotide Archive (accession no. E-MTAB-2510).

Dynamic resistance signature and pathway enrichment

Dynamic resistance genes were defined as genes that were up-regulated in the RNA-Seq data set ≥1.5-fold in 502R (drug on) versus parental, and decreased ≥1.5-fold in 502R (drug off) versus (drug on). Significance thresholds were also set at P < 0.001 for each comparison. A total of 465 genes fit these criteria (see table S1) and were analyzed using Ingenuity Pathway Analysis software (Ingenuity) to identify enrichment of canonical pathways and biological functions represented in the signature.


Fig. S1. siFOXO3a knockdown in parental and 502R cells.

Fig. S2. Chemical screening to identify combinations that yield sensitivity to PF-04691502 in resistant cells.

Fig. S3. Schematic of RNA-Seq profiling.

Fig. S4. Characterization of resistance to PF-04691502 in a human CRC cell line.

Fig. S5. Acute pathway modulation in PDX tumors.

Table S1. List of dynamic resistance genes identified by RNA-Seq.

Table S2. Top canonical pathways enriched among dynamic resistance genes.

Table S3. Top physiological processes enriched among dynamic resistance genes.


Acknowledgments: We would like to acknowledge the Pfizer postdoctoral program for their support. Funding: This work was supported by Pfizer Global Research and Development. Author contributions: E.S.M. and K.E.H. conceived the idea for the mouse models; E.S.M., K.E.H., J.R., M.J.S., and P.J.B. generated the mouse models; P.J.B., P.J., T.D.M., T.X., J.I., N.V.L., J.L.C.K., O.G., D.S., P.O., T.V.A., and S.L.W. characterized the mouse and human models; T.D.M., N.E.B., B.G.W., C.A.O., D.S., P.O., T.V.A., S.L.W., P.R., J.G.C., V.R.F., and E.S.M. generated and characterized the PDX models; T.X. and P.R. performed statistical and computational analyses; P.J.B. and E.S.M. wrote the manuscript; P.J.B., O.G., B.G.W., C.A.O., D.S., P.O., T.V.A., S.L.W., P.R., J.G.C., V.R.F., K.H., and E.S.M. reviewed the manuscript. Competing interests: P.J.B., P.J., T.X., J.I., N.V.L., J.L.C.K., O.G., D.S., P.O., T.V.A., S.L.W., P.R., J.G.C., V.R.F., K.H., and E.S.M. are either former or current employees of Pfizer. Data and materials availability: RNA-Seq data are available at the European Nucleotide Archive ( under accession no. E-MTAB-2510.
View Abstract

Stay Connected to Science Signaling

Navigate This Article