Resistance limits the effectiveness of receptor tyrosine kinase (RTK)-targeted therapies. Combination therapies targeting resistance mechanisms can considerably improve response, but will require an improved understanding of when particular combinations will be effective. One common form of resistance is bypass signaling, wherein RTKs not targeted by an inhibitor can direct reactivation of pathways essential for survival. Although this mechanism of resistance is well appreciated, it is unclear which downstream signaling events are responsible. Here, we apply a combined experimental- and statistical modeling–based approach to identify a set of pathway reactivation essential for RTK-mediated bypass resistance. Differences in the downstream pathway activation provided by particular RTKs lead to qualitative differences in the capacity of each receptor to drive therapeutic resistance. We identify and validate that the JNK pathway is activated during and strongly modulates bypass resistance. These results identify effective therapeutic combinations that block bypass-mediated resistance and provide a basic understanding of this network-level change in kinase dependence that will inform the design of prognostic assays for identifying effective therapeutic combinations in individual patients. Cancer Res; 76(18); 5219–28. ©2016 AACR.
Therapies targeting aberrant receptor tyrosine kinase (RTK) signaling are effective in treating a subset of multiple malignancies, including breast carcinoma and lung adenocarcinoma (1, 2). Despite transient effectiveness, the resulting survival benefit of these therapies is limited by resistance mechanisms that allow tumor cells to escape the effect of therapy. Resistance etiology varies widely, including mutation of the drug target to block the effect of therapy, amplification of the drug target to overcome inhibition, pharmacokinetic barriers that block trafficking of drug to tumor cells, and "bypass" switching to alternative pathways not targeted by therapy (3, 4). In the case of RTK-targeted therapies, many nontargeted RTKs may become activated to provide bypass resistance (5). Two well-studied combinations are the ability of HER3 to provide resistance to HER2-targeted therapy in breast carcinoma and the ability of Met to provide resistance to EGFR-targeted therapies in lung carcinoma (6–10). In each case the resistance-conferring receptors may contribute to innate or adaptive resistance and can become activated by multiple means, including ligand-mediated autocrine or paracrine induction, amplification, or mutation (8, 11, 12).
Combination therapy can effectively combat resistance but will require accurate identification of relevant combinations for individual tumors. However, many receptors exist that can cause resistance and are activated through both tumor cell-intrinsic and -extrinsic means. This dictates that a more fundamental understanding of network-level bypass, and methods to identify effective combinations for individual tumors, will be required (6). An improved understanding of exactly which pathways must become reactivated to provide resistance may also identify widely effective therapeutic combinations not dependent upon the particular RTK providing bypass signaling.
Here, we have undertaken a combined experimental and computational approach to understand bypass resistance. We examine four cell lines in which activation of a noninhibited RTK can provide bypass resistance to develop a multipathway understanding of the process (6). Through modeling and validation experiments, we identify a core set of pathway activation that determines whether cells are ultimately viable. Individual RTKs activate these pathways to varying extents, and thus have varying ability to drive bypass resistance. This observation in turn explains why RTK expression alone poorly predicts resistance capacity. We subsequently apply this model to identify therapeutic combinations that can block bypass resistance from multiple driving receptors simultaneously. This information will be valuable for a basic understanding of bypass resistance, development of prognostic tools to identify resistance mechanism, and design of effective therapeutic combinations.
Materials and Methods
BT474, SKBR3, and HCC827 were obtained from ATCC in 2015, confirmed by STR profiling by the source, and have been passaged for fewer than 6 months. PC9 was obtained from Sigma-Aldrich in 2015, confirmed by STR profiling by the source, and has been passaged for fewer than 6 months. HOP-62, HOP-92, H322M, and H522 confirmed by STR profiling were obtained from Merrimack Pharmaceuticals in 2016. BT474 and PC9 identities were additionally confirmed in 2016 through RNA sequencing experiments. BT474 and SKBR3 were grown in DMEM, whereas PC9, HOP-62, HOP-92, H322M, H522, and HCC827 were grown in RPMI1640, both supplemented with 10% FBS and 1% penicillin/streptomycin, at 37°C and 5% CO2.
Viability measurements were performed using CellTiter Glo reagent according to the manufacturer's protocol (Promega). Cells were seeded in 96-well plates at a density of 200 per well, and treated the next day with the indicated GFs (Peprotech) and inhibitors (LC Labs). After 72 hours, CellTiter Glo reagent was mixed with each well and luminescence was quantified.
Cell signaling and protein abundance quantitation
Cells were treated for 4 hours with the indicated GFs and inhibitors in serum-free medium without prior stimulation, then lysed in 10 mmol/L Tris-HCl pH 8.0, 1 mmol/L EDTA, 1% Triton-X 100, 0.1% Na deoxycholate, 0.1% SDS, and 140 mmol/L NaCl, with protease and phosphatase inhibitor supplemented before use (Bio-Rad). Protein concentration was normalized by bicinchoninic acid assay.
A bead-based multiplexed ELISA-based assay was used for signaling measurement (Bio-Rad). In all cases, pErk is Erk1/2 (pT185/pY187, pT202/pY204), pMek is Mek pS217/221, pGSK3 is GSK3α/β pS21/9, pAkt is Akt pS473, pP38 is P38 pT180/pY182, pcJun is c-Jun pS63, pSTAT3 is STAT3 pY705, and pJNK is JNK pT183/pY185. RTK abundance was quantified by multiplexed ELISA (Millipore). Absolute quantification of receptor abundance was performed by comparison to recombinant standards of each receptor (R&D Systems). Lysates were incubated with beads overnight, and then the beads were washed with 0.1% (v/v) Tween-20 in TBS. Detection antibody and streptavidin–phycoerythrin were incubated for 60 and 10 minutes, respectively. Signal was quantified from each bead set using a MagPix Luminex reader (Bio-Rad).
For quantification of total cJun, cells were treated and lysed identically to those used in ELISA measurement. Lysates were normalized according to protein concentration, and then separated by SDS-PAGE and transferred to a nitrocellulose membrane. Each membrane was incubated overnight with antibodies against cJun (Cell Signaling Technology, cat. no. 9165) or paxillin (BD Biosciences, cat. no. 610052).
Partial least squares regression and generalized linear modeling
Partial least squares regression was performed using MatLab (MathWorks). Phosphosignaling time courses for each condition were summarized either by calculating the AUC via trapezoidal integration, or the steady-state levels by averaging the 2- and 4-hour time points. Z-scored phosphosignaling measurements were regressed against z-scored viability measurements. The SD of the loadings were calculated by jackknife, leaving out individual GF/drug combination treatments (13).
Generalized linear regression (GLM) models were developed relating cell signaling as measured at 4 hours to viability measurements at 72 hours using R. Before regression, each phosphorylation and viability measurement was z-score normalized. GLM models were fit using glm using the formula Viab ∼ pAkt + pErk + pGSK + pcJun + pJNK + pP38 + pAkt × pErk. For sampling the model parameters posterior distribution, MCMCglmm from the package of the same name was used without thinning, a burn-in of 1,000 iterations, and sampling over 10,000 iterations. Markov chain mixing was achieved according to the Geweke diagnostic. A default model prior of each parameter normally distributed around 0 with a variance of 1 was used.
Bliss synergy calculation
The interaction between Mek and JNK inhibition was calculated according to the Bliss independence model. Viability measurements were normalized to the value with no inhibitor treatment and then inverted to be in the form of percent inhibition:
where is the percent inhibition effect of a particular combination, is the viability measurement for a particular combination, and is the viability measured in the absence of inhibitor. Thus, for a condition where the measured viability was 70% of what was measured without inhibitor, this would become an effect of 30%. The results of combination inhibitor treatment were then calculated by Bliss independence, using the equation:
where , , and are the effects of the individual inhibitors and combination, respectively. For example, if one inhibitor decreased viability by 10% and the other by 20% when administered alone, the predicted combined effect would be , or 28%. The measured and predicted effects were then plotted on an inverted scale, so that positive effects (inhibition of viability) were negative, to aid comparison with the untransformed viability measurements. The effects and difference between that predicted and measured were plotted on the same color scale for ease of comparison.
AXL and a kinase-dead variant K562R were amplified from previously used vectors (14). Met and PDGFRb were amplified from pDONR223-PDGFRb and pDONR223-MET from William Hahn (Dana-Farber Cancer Institute, Boston, MA) and David Root (Broad Institute, Cambridge, MA; Addgene plasmids 23893 & 23889; ref. 15). Each receptor was then cloned into IRES-EGFP2 (Clontech). Cells were seeded densely on 10-cm plates, and then transfected the next day with 5 μg of each plasmid with a corresponding amount of 5-μL Lipofectamine 2000 in OptiMEM according to manufacturer's instructions. After 4 hours, the media were exchanged into full serum media lacking antibiotics.
For evaluation of resistance-mediated selection, the day after transfection, cells were split into six-well plates and, after adhering, were placed in serum-free media with inhibitor and GF as indicated. The next and following day, wells were trypsinized, spun down, and then resuspended in PBS. Fluorescence, forward scatter, and side scatter were then immediately quantified on an Accuri C6 (BD Biosciences).
For evaluating receptor overexpression, cells were trypsinized the day after transfection and fixed in 4% PFA in PBS for 1 hour, blocked in Odyssey Blocking Buffer (LI-COR) for 1 hour, then stained with antibodies against Met, AXL, or PDGFRb overnight. The next day, cells were repeatedly washed and stained with Alexa Fluor 594–conjugated antimouse antibodies for 1 hour. After additional washing, fluorescence, forward scatter, and side scatter were immediately quantified on an Accuri C6 (BD Biosciences).
RTK expression is essential but not sufficient for bypass resistance
To better understand the process of bypass resistance and its relationship to signaling network state we first selected two HER2-overexpressing breast carcinoma cell lines, both studied before for their sensitivity to the HER2-targeted inhibitor lapatinib and their capacity for bypass resistance in the presence of HRG (6). We treated each cell line with 0, 1, or 5 μmol/L of lapatinib either alone or in the presence of 50 ng/mL of different growth factors (GF; Fig. 1A). Cells exhibited a dose-dependent decrease in viability measured at 72 hours strongly counteracted by simultaneous addition of HRG and partially by other GFs (Fig. 1B). To expand our perspective beyond HER2-dependent signaling dysregulation, we additionally selected two EGFR-dependent lung adenocarcinoma cell lines, PC9 and HCC827, that are accordingly sensitive to the EGFR inhibitor erlotinib. We measured viability with 0 or 1 μmol/L erlotinib in the presence of 50 ng/mL of different GFs (Fig. 1C). Particular GFs counteracted the erlotinib-induced decrease in viability in a consistent pattern to that observed previously (6).
Because of manifold bypass possibilities among RTKs and their critical function in diverse biological processes, prospective diagnostic measurements are needed to identify which among many targeted therapies would be most effective to apply in precise combination to overcome bypass resistance. A first notion might be to measure and focus on treating those receptors that are most abundant or most abundantly activated within a tumor (16). Therefore, we next sought to directly determine the relationship between receptor abundance and capacity to provide bypass resistance. We measured the absolute abundance of seven RTKs across four cell lines using a recombinant standard. RTK abundances varied across several orders of magnitude, and for many cases, the RTK that provided primary drug sensitivity was not the most abundant receptor (Supplementary Fig. S1A). To test the relationship between RTK abundance and resistance-promoting capacity (viability with inhibitor and GF normalized to that with just inhibitor), we plotted viability in the presence of inhibitor against the abundance of different GF-activated RTKs (Fig. 1D and E). Each data point indicates RTK abundance matched to the viability promoting effect upon treatment with each cognate GF (Fig. 1A). The viability of cells in the absence of erlotinib or lapatinib was predominantly dependent on EGFR or HER2 signaling, respectively. Therefore, we plotted in green the abundance of the inhibitor-targeted RTK and cell viability in the absence of inhibitor. Although abundance significantly correlated with resistance capacity in the case of three cell lines, no significant correlation was observed for the HCC827 (Fig. 1D and E). Strikingly, in HER2-dependent cells, HER3-HRG was relatively potent in its resistance-conferring capacity (Fig. 1D), while it was less potent than predicted in EGFR-dependent cell lines (Fig. 1E).
We considered that drug treatment or sustained serum withdrawal may influence RTK expression and thus the correlations we observed. For example, if HER2/HER3 expression is lost by either manipulation in PC9 cells, the change would explain the relative resistance-promoting inefficiency of HRG. Thus, we examined the influence of each over 24 hours in PC9 cells (Fig. 1F). Only HER3 changed in abundance more than two-fold, increasing up to seven-fold at 24 hours. This change in fact exacerbates the discrepancy and removes the correlation between viability and abundance (Fig. 1E, tan line; r = 0.84, P < 0.05 to r = 0.62, P > 0.05). Although more modest in effect, EGFR and MET displayed a simultaneous decrease of roughly 2-fold in abundance by 24 hours with serum withdrawal, or increase of 50% in the presence of 1 μmol/L erlotinib (Supplementary Fig. S1B). Therefore, inhibitor-induced changes in RTK abundance only reduced the correlation between RTK abundance and resistance-promoting capacity in the case of PC9 cells.
Despite mixed correlation between RTK abundance and resistance capacity within cell lines, we did observe a consistent threshold effect across all four cell lines, wherein all receptors that promoted resistance more than 30% were more abundant than 2 × 106 per cell (Fig. 1G). We defined resistance promotion as the GF-induced increase in viability, scaled between 0.0 and 1.0, to compare across cell lines. To examine whether some threshold of expression might be necessary but not sufficient for resistance, we explored the results of two previous studies examining expression and ligand-induced resistance across a large number of cell lines (6, 17). Similarly, we observed that expression below a certain level correctly predicted that ligand-induced bypass resistance would not occur, but also that high receptor expression could not predict resistance-promoting capacity (Fig. 1H). Therefore, the abundance of a particular RTK is unlikely to accurately predict whether it is driving resistance.
RTK overexpression identifies qualitative differences in resistance-promoting capacity
The four cell lines of this study were selected on the basis of their sensitivity to an RTK-targeted therapy and ability to display resistance upon the addition of one or more GFs (6). These criteria may artificially select for an improved correlation between RTK abundance and resistance-conferring capacity because cell lines lacking a resistance-promoting bypass receptor would not have been included. Therefore, we took an orthogonal approach, overexpressing individual RTKs in PC9 cells and then determining the extent to which each receptor could promote resistance to erlotinib when highly abundant (Fig. 2).
Many cell lines exhibit clonal heterogeneity that is reflected in their response to targeted inhibition (18, 19). To account for this, we took a short-term overexpression approach, transiently overexpressing RTKs within an IRES-EGFP vector, to minimize clonal selection within the population of cells (Fig. 2A). If overexpression of a particular RTK improved the resistance capacity of the cells, we expected to observe selection of transiently transfected cells in the presence of inhibitor. We first selected overexpression of the RTK AXL as an orthogonal control, as the receptor's role in resistance to EGFR-targeted treatment, often without exogenous ligand, is well appreciated (11, 20). Selection of transfected cells was quantified by flow analysis (Fig. 2B). We observed selection for AXL-transfected cells in the presence of erlotinib, dependent upon kinase activity of the receptor, providing confidence in our approach (Fig. 2C).
Our previous observations indicate that Met is inefficient (promotes viability less than its abundance would suggest) but capable of EGFR-associated bypass resistance (Fig. 1C and E). In contrast, in previous work across a panel of cell lines, PDGFRb was never observed to promote resistance, although very few cell lines showed expression of the receptor (6). Overexpressing each of these receptors and then selecting among a mixed population with erlotinib, we observed selection of Met-expressing cells in the presence of erlotinib, but no selection of PDGFRb-positive cells even when overexpressed (Fig. 2D and E). We verified as a control that PDGFRb was indeed being overexpressed in GFP-positive cells (Supplementary Fig. S2). Thus, receptor expression alone is insufficient to predict mechanism of resistance, and receptors qualitatively differ in their resistance-promoting capacity. Furthermore, in this cell line, PDGFRb is incapable of promoting resistance at any expression level obtained by transient overexpression.
Conserved RTK-specific signaling exists despite differences in bypass capacity
To examine the role of downstream pathway activation, we measured a panel of phosphorylation sites for each of the four cell lines, 4 hours after treatment with inhibitor and GFs, to capture the signaling consequences of each treatment (Fig. 3A and B). Basic analysis of these signaling measurements captured known pathway associations. For example, pAkt and pGSK were clustered together in three of the four cell lines. For SKBR3, BT474, and HCC827, resistant conditions clustered with those in the absence of erlotinib (Fig. 3A and B). For PC9 cells, all of the conditions with erlotinib clustered separately from those without (Fig. 3B). Overall clustering profiles were still quite distinct between cell lines, likely a reflection of different receptor and intracellular component expression levels (Fig. 3A and B).
We applied GLM modeling, regressing our phosphorylation measurements against viability, to understand which phosphorylation sites could predict the particular GF conditions leading to resistance. To capture dependencies on coordinate pathway activation, we included an interaction term of Akt and Erk. We elected to not include other interaction terms, or more complex relationships, due to less prior evidence implicating the other pathways measured. In the EGFR-dependent cell lines PC9 and HCC827, pcJun (and pErk for HCC827) was identified as particularly important (Fig. 3C and Supplementary S3). A model combining measurements from both EGFR-dependent cell lines, with only significantly nonzero parameters shown, and 95% confidence intervals shown in parentheses, was: Viability = 0.17 (0.01 − 0.33) pErk + 0.75 (0.57 − 0.92) pcJun. Positive weighting of pcJun and pErk suggests that further phosphorylation contributes positively to resistance (Fig. 3C and Supplementary S3). In contrast, the regression models for the HER2-dependent cell lines SKBR3 and BT474 positively weighted pAkt but negatively weighted pcJun (Fig. 3C and Supplementary S3), indicating greater pcJun measurement corresponded to conditions with less viability. A model combining measurements from both HER2-dependent cell lines, with only significantly nonzero parameters shown, and 95% confidence intervals shown in parentheses, was Viability = 1.00 (0.69 − 1.32) Akt − 0.30 (0.51 − 0.09) cJun. The parameters identified as important by each model correlated with which phosphorylation sites were most inhibited by inhibitor treatment (Fig. 3D, Wilcoxon signed-rank test, P < 0.01). Model parameters were similar among but not between EGFR- or HER2-dependent cell lines, indicating that this influenced which pathway activation changes were essential for resistance (Supplementary Fig. S3). In total, this regression modeling identified that multiple pathway activation measurements are essential to predicting resistance conditions, and formed novel hypotheses as to the contribution of JNK pathway activation.
EGFR bypass resistance requires sustained JNK pathway signaling
Complex signaling dynamics can govern cellular phenotypic response to extracellular cues (21). Therefore, we wished to examine whether the inclusion of pcJun in our models of EGFR inhibitor bypass resistance might be due to incompletely capturing the complete dynamic response to ligand and inhibitor treatment. To determine this, we measured a dense time-course of pErk, pMek, pcJun, and pAkt response in PC9 cells without erlotinib treatment, with simultaneous inhibitor treatment, or with 4-hour pretreatment (Fig. 4A). These phosphosites were chosen as Akt and Mek/Erk reactivation are widely implicated in bypass resistance (6), and were the four phosphosites repeatedly selected in models based on a single time point (Fig. 3C). As expected, the responses at shorter timescales were often distinct from the sustained responses observed. Similar to the measurements at 4 hours, these dynamic responses showed that each pathway was inhibited by erlotinib treatment, usually quite rapidly. Although the responses to HGF stimulation changed little with addition of erlotinib, the bypass effects of other GFs were more complex. For example, addition of IGF entirely restored pAkt levels, but not pMEK or pcJun. FGF restored pcJun levels, but not pAkt or pMEK. Also, although HRG strongly activated all pathways in the absence of erlotinib, its response was dependent upon EGFR activity. These differences all highlight a potential source of variation in the resistance capacity of particular RTKs.
To examine the relationship between pcJun, Erk pathway activation, and bypass resistance over differing timescales, we plotted both pathways with viability measurements from each condition (Fig. 4B and C). Regardless of whether pMek or pErk measurements were used, conditions separated by their eventual viability only at later time points (Fig. 4B and C; 2–4 hours). In both cases pcJun was required for separation, again demonstrating that pcJun is required to predict viability. To further evaluate the contribution of signaling at different timescales, we summarized the response of each condition using the AUC of each phosphorylation time course, capturing early response, as well as the sustained (>2 hours) measurement. Using partial least squares regression, models could accurately predict the response of PC9 cells with the sustained (>2 hours) measurement (Fig. 4D and E). pcJun was required for predictive capacity, but all earlier time points (AUC) were dispensable (Fig. 4D). Sustained pMek/pErk was positively weighted with viability along principal component 1, whereas pcJun was weighted with viability along both principal components (Fig. 4E), consistent with our earlier modeling (Fig. 3C). In total, these results indicate that measurement of sustained pcJun and Erk pathway activation is necessary for accurate prediction of bypass resistance in PC9 cells.
Validation of the bypass resistance model identifies combination therapy approaches
With confidence that pcJun and Erk pathway activation were coordinately necessary for predicting bypass resistance development in PC9 cells, we sought to determine whether this interaction might suggest effective therapeutic combinations. We first measured the pathways influenced by treatment with a Mek inhibitor U0126 to ensure that pcJun was not simply responding to Erk activation. Although Erk1/2 phosphorylation was completely abrogated by Mek inhibition, cJun phosphorylation was only modestly reduced in PC9, and not at all in HCC827 (Fig. 5A and Supplementary S4A). pcJun was most reliably and strongly abrogated by JNK inhibition (Supplementary Fig. S4A); however, total cJun levels also varied, indicating this phosphorylation measurement is likely a measure of both protein abundance and phosphorylation changes (Supplementary Fig. S5). Thus, pcJun is most prominently modulated by JNK activity in these cells but is also influenced by Erk activity.
A first prediction of our model is that Erk and JNK inhibition would have additive effectiveness in decreasing PC9 viability (Fig. 3D). Indeed, we observed that combination treatment with a JNK (SP600125) and Mek inhibitor coordinately reduced PC9 viability in the absence (Fig. 5B) or presence (Fig. 5C) of erlotinib. We observed the same coordinate reduction in viability with JNK and Mek inhibition across a panel of HCC827 and four other lung carcinoma cell lines, as well as with another JNK inhibitor JNK-IN-7 (Supplementary Fig. S4B and S4C). We used the Bliss independence model to examine the nature of combination Mek and JNK inhibition on viability (Supplementary Fig. S6A). This model recapitulated the results of the combination inhibitor treatments, consistent with an additive interaction between both inhibitors (Supplementary Fig. S6B and S6C).
Next, we predicted that bypass resistance via Met is coordinated through activation of JNK and Erk. Previous studies suggested that activation of JNK by Met occurs via Src family kinases (22). Indeed, Met-mediated pcJun, but not pMek, was blocked by Src inhibition (dasatinib; Fig. 5D). Consistent with this and our model's prediction of the coordinate importance of Erk/JNK activation to bypass resistance, Src (Fig. 5E and Supplementary S4D) or JNK (Fig. 5F and Supplementary S4E) inhibition cooperated with Mek inhibition to block increases in Met-induced viability.
In contrast to the case with EGFR-dependent cells, our models of signaling–viability relationship for HER2-dependent cell lines indicated that JNK inhibition should in fact increase viability (Fig. 3D). To test this, we first examined the effect of JNK inhibition on viability in the presence or absence of lapatinib, and as predicted observed a dose-dependent increase (Fig. 5G). We wondered if this effect could modulate the ability of particular GFs to provide resistance in HER2-dependent cell lines. In particular, although lapatinib is thought to be an EGFR/HER2 dual-targeting inhibitor, we observed potent STAT3 phosphorylation suggestive of EGFR activation with EGF treatment in each HER2-dependent cell line, and an increase in STAT3 phosphorylation with lapatinib treatment (Fig. 5H). This indicated that some amount of EGFR signaling must be maintained in the presence of lapatinib. With respect to Akt phosphorylation, although more modest than HRG treatment, EGF treatment increased Akt phosphorylation roughly 4-fold in the presence of lapatinib (Fig. 5I). We predicted that this increase in Akt phosphorylation may have more pronounced effects in the presence of JNK inhibition. Although EGF treatment only leads to a modest increase in viability in the absence of JNK inhibition, with JNK inhibition the GF was able to make BT474 cells completely resistant to lapatinib (Fig. 1B and 5J). This indicates that JNK pathway activity strongly modulates the bypass signaling response of cells, and that our model accurately captures these effects through measurement of pcJun.
Although targeted therapies lead to survival benefits in patients, these benefits are usually short-lived due to resistance mechanisms. Many studies have focused on various mechanisms of resistance with the notion that identifying particular resistance mechanisms would lead to widely efficacious treatment combinations. These have left us with panoply mutations, expression changes, and tumor-extrinsic factors that can contribute to resistance, but little in the way of a unified understanding of these processes.
Here, we have focused on a family of resistance mechanisms that arise due to redundancy in the cellular signaling machinery. By a combined experimental and computational approach, we observe that bypass resistance can be predicted by coordinate measurement of multiple pathways. The pathways that must be reactivated for resistance in each case, however, depend on which RTK, and thus signals, have been therapeutically targeted. Because of the pathway dependencies, and particular RTKs being biased in their relative pathway activation, not all RTKs are similar in their capacity to provide resistance to a particular therapy (Figs. 1 and 6). Previous works have similarly recognized that pathway reactivation must occur for bypass resistance, but have not attempted to capture what those signaling changes might be beyond Erk and Akt activation (5, 6, 11). Contribution of other pathways such as JNK would explain why Erk/Akt activation is not always entirely predictive of resistance (6).
Our results indicate that the effects of JNK pathway activation, whether promoting or inhibiting resistance, depend upon the cellular context and/or the original RTK inhibited. The effects of the JNK pathway have similarly shown opposing effects on cell survival, apoptosis, and proliferation dependent upon cellular context, as well as activation duration and intensity, in many other cases (23). For example, surviving melanoma cells after Raf/Mek inhibitor treatment are enriched for higher JNK activation, and Raf/JNK dual inhibitor treatment is synergistic (24). However, positive feedback in JNK signaling response predicts neuroblastoma patient survival, and JNK inhibition drives tumor growth in a MYCN-driven spontaneous neuroblastoma model (25). In breast cancer, our observed negative relationship between cell survival/proliferation and JNK activity is supported by the oncogenic role of dominant-negative MAP2K4 mutants (26). However, dissecting the exact contexts in which JNK activity promotes or inhibits resistance will require more detailed study of the pathway's diverse effects.
The widespread nature of redundancy-mediated resistance means we need precise and prognostic ways of identifying combination therapies. Just among RTK-targeted therapies, inhibitors for cMET, IGF1R, PDGFRb, EGFR, HER3, AXL, and HER2 have entered clinical trials. How will patients be matched to effective combinations of these therapies, while avoiding toxicity? A most basic approach might be to quantify the abundance of various RTKs within tumor cells, and selectively target those with highest abundance or activation (16). Our results, however, indicate that receptors display qualitative differences in their resistance capacity (Fig. 1 and 2), suggesting this approach would be poorly predictive of effective therapeutic combinations. Rather, a more effective approach might be to identify receptors providing the particular pathway activation essential to conferring resistance. It was recently shown that selective measurement of Grb2–EGFR interaction could prognostically predict therapeutic response, more so than abundance or phosphorylation of EGFR (27). Similarly, our results indicate that if different receptors rely on the same adapter proteins for essential pathway reactivation, assays might be developed to evaluate which receptor is interacting with a particular adapter, identifying the particular RTK-mediating resistance.
In which cases does resistance in essence represent reestablishment of the same signaling state as opposed to fundamentally different requirements for cancer cell survival? Our results suggest that RTK bypass resistance often corresponds to reactivation of the particular pathways lost by inhibitor treatment, resembling intracellular signaling in the absence of inhibitor (e.g., HRG in BT474 and SKBR3; Fig. 3A). Particular RTKs can provide measurable signaling responses and yet fail to produce resistance due to mismatch between the requisite and provided signaling changes (Fig. 4A). An outstanding question is whether more global transcriptional mechanisms of resistance, such as epithelial–mesenchymal transition, operate through fundamental changes in the pathways relied on for survival, or if other factors such as RTK expression changes still lead to activation of the same survival pathways (28). The many pathways involved in the development of resistance will necessitate modeling approaches such as those undertaken here to address these mechanisms.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: S. Manole, A.S. Meyer
Development of methodology: S. Manole, A.S. Meyer
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Manole, E.J. Richards, A.S. Meyer
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Manole, E.J. Richards, A.S. Meyer
Writing, review, and/or revision of the manuscript: S. Manole, E.J. Richards, A.S. Meyer
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Manole, A.S. Meyer
Study supervision: A.S. Meyer
This work was supported by NIH 1-DP5-OD019815-01 (A.S. Meyer) and in part by the Koch Institute Support (core) grant P30-CA14051 from the NCI.
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.
The authors would like to thank Douglas Lauffenburger, Forest White, Allison Claas, Sarah Schrier, and Annelien Zweemer for helpful discussions.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received January 18, 2016.
- Revision received May 27, 2016.
- Accepted June 16, 2016.
- ©2016 American Association for Cancer Research.