MEK-Dependent Negative Feedback Underlies BCR–ABL-Mediated Oncogene Addiction

Translational studies with small-molecule tyrosine kinase inhibitors (TKI) in patients with chronic myelogenous leukemia (CML) have convincingly demonstrated that the clinical activity of these agents is achieved through inhibition of the intended target BCR–ABL (1). To date, the high rate of success associated with BCR–ABL TKIs provides the most compelling clinical evidence for the phenomenon of “oncogene addiction,” the exquisite reliance of cancer cells upon a pathologically activated oncogene. Oncogene addiction enables targeted therapies to effect clinical responses and simultaneously cause little toxicity. Even in the most advanced phases of CML, a substantial proportion of patients can achieve deep responses with BCR–ABL TKIs (2).

Our understanding of the molecular basis of oncogene addiction is poor. Sharma and colleagues proposed the term “oncogenic shock” to describe the effect that activated oncogenes have on the balance between prosurvival and proapoptotic signals. The oncogenic shock model proposes that upon acute inhibition of oncogene activity, a more rapid decay in prosurvival signals leads to a state that favors apoptosis (3). Although this model provides a useful conceptual framework in which to begin to understand the phenomenon of oncogene addiction, evidence to support the oncogenic shock model is largely circumstantial; molecular mechanisms that lead to the proposed heightened levels of prosurvival and proapoptotic signals have not been well characterized.

Over the past decade, numerous oncogenic kinases in a variety of cancers have been identified and clinically targeted, including EGF receptor (EGFR) and ALK in non–small cell lung cancer, BRAF^V600E in melanoma and colorectal cancer, and FLT3-ITD in acute myeloid leukemia (AML). However, efforts to extrapolate the clinical success of BCR–ABL TKI therapy in CML to other malignancies have failed; the majority of patients treated with other clinically active TKIs do not achieve responses of similar magnitude. For example, the selective BRAF inhibitor vemurafenib, which is approved for BRAF^V600E-positive metastatic melanoma, rarely effects deep reductions in tumor volume (4). Interestingly, MAP–ERK kinase (MEK) inhibitors are active in BRAF^V600E-expressing cells in vitro but, unexpectedly, not in cells with activated receptor tyrosine kinases (RTK) that activate the RAS–MEK–ERK pathway (5). Previous studies demonstrated that BRAF^V600E establishes a high level of extracellular signal–regulated kinase (ERK)–directed transcriptional output and MEK-dependent negative feedback of growth factor receptor (GF-R) signaling, whereas activated oncogenic RTKs do not. In addition, in contrast to RTKs, BRAF^V600E escapes MEK-dependent negative feedback (6).

It has been postulated that efficient bypass of BRAF kinase inhibition through GF-R–mediated reactivation of the RAS–MAPK signaling pathway may allow melanoma cells to survive in the tumor microenvironment. Recent experimental data have demonstrated that melanoma, colorectal, and thyroid cancer cells harboring BRAF^V600E mutations are inherently primed to circumvent BRAF inhibition by vemurafenib through rapid relief of negative feedback of GF-R signaling (7–11).

Here, we sought to characterize the molecular mechanisms that underlie BCR–ABL-mediated oncogene addiction in an effort to understand what makes this kinase the best-validated target in human cancer. We applied an unbiased kinetic quantitative phosphoproteomic analysis to CML cells transiently exposed to the BCR–ABL TKI dasatinib to identify candidate mediators of BCR–ABL-dependent cell survival. To test the importance of the observed signaling changes, we established a tissue- and species-relevant isogenic model system to molecularly characterize BCR–ABL-mediated oncogene addiction and validated our findings in patient-derived cell lines.

Previous work demonstrated that transient exposure (20 minutes) of CML cell lines to clinically relevant concentrations of dasatinib elicits apoptosis with kinetics similar to continuous TKI exposure, despite evidence that BCR–ABL kinase activity is largely restored within 4 hours of drug washout (12–14). We hypothesized that the phosphorylation status of a subset of proteins must be durably altered, and critical mediators of BCR–ABL-mediated cell survival would be included among this group. We therefore undertook an unbiased kinetic, quantitative assessment of phosphotyrosine-containing proteins in the patient-derived CML cell line K562 transiently exposed to a high-dose pulse (HDP) of 100 nmol/L dasatinib using stable isotope labeling by amino acids in culture (SILAC). We successfully identified 184 phosphotyrosine residues in 126 different proteins, representing the most comprehensive kinetic analysis of TKI-treated patient–derived CML cells to date (Supplementary Table S1). We compared the quantified phosphotyrosine profile before TKI treatment, after 20 minutes of TKI exposure, and at 3 and 6 hours after TKI washout (Fig. 1A).

We grouped phosphotyrosine peptides based on the pattern of tyrosine modification following HDP dasatinib treatment. Twenty-four tyrosine residues were transiently dephosphorylated, 31 were gradually dephosphorylated, 46 were not appreciably altered, and seven were hyperphosphorylated. Notably, 55 tyrosine residues were persistently dephosphorylated following TKI washout, and functional enrichment of these peptides revealed an overrepresentation of proteins involved in GF-R signaling pathways (Supplementary Table S2). Among these were tyrosine residues from STAT5A/B, ERK1/2, GAB1, and SHC1. Phosphotyrosine peptides associated with phosphoinositide 3-kinase (PI3K)–AKT pathway activation were either transiently dephosphorylated or not altered (Fig. 1B). Several of the signaling changes identified in the phosphoproteomic analysis were confirmed by Western immunoblot analysis in K562 cells and the independent patient-derived CML cell line KU812. Although tyrosine residues within the PI3K–AKT pathway were not durably altered in the phosphoproteomic analysis, serine phosphorylation of S6 (S235/S236) was durably altered in a time-dependent manner following TKI treatment (Fig. 1C). Similar phosphorylation changes were observed in K562 and KU812 cells treated with an HDP of the BCR–ABL TKI imatinib (Supplementary Fig. S1), arguing that the observed signaling changes are likely a consequence of BCR–ABL inhibition and not the result of unintended off-target kinase inhibition.

Although phosphorylation at the BCR–ABL activation loop site Y393, which is essential for kinase activity, was only transiently absent following HDP dasatinib treatment, phosphoproteomic and Western immunoblot analyses revealed substantial variation in phosphorylation changes of BCR–ABL tyrosine residues (Fig. 1D). Collectively, our phosphoproteomic and immunoblot analyses, coupled with previous studies, suggest a requirement for the maintenance of at least two of the three canonical BCR–ABL-activated pathways (JAK–STAT, RAS–MEK–ERK, and PI3K) for CML cell survival in vitro (15). Furthermore, these observations suggest that persistent activation of GF-R signaling pathways is critical for BCR–ABL-mediated oncogene addiction.

To investigate whether BCR–ABL appropriates GF-R signaling pathways to establish oncogene addiction, we sought to model oncogene addiction in vitro by establishing an isogenic cell line model system. Human-derived erythroleukemia TF1 cells require human granulocyte macrophage colony-stimulating factor (hGM-CSF) or human interleukin-3 (hIL-3) for proliferation and survival in vitro, but can be transformed to growth factor independence by BCR–ABL (16). We established pools of TF1/puro and TF1/BCR–ABL cells through retroviral transduction. Western immunoblot analysis confirmed the expression of phosphorylated BCR–ABL protein in TF1/BCR–ABL cells and activation of downstream targets CRKL, STAT5A/B, ERK1/2, and S6 (Supplementary Fig. S2A and S2B).

To determine whether TF1/BCR–ABL cells have acquired reliance upon BCR–ABL activity for survival and thereby truly represent a model of oncogene addiction, we measured the amount of apoptosis induced in response to simultaneous growth factor deprivation and dasatinib treatment. In TF1/puro cells, hGM-CSF deprivation alone led to decreased viability, and dasatinib had no additional impact on the extent of apoptosis induced. In sharp contrast, TF1/BCR–ABL cells underwent a statistically significant increase in apoptosis when treated with dasatinib (Fig. 2A). Moreover, continuous cotreatment with hGM-CSF and dasatinib fully rescued TF1/puro cells from apoptosis, whereas TF1/BCR–ABL cells were only partially rescued from dasatinib-mediated apoptosis by hGM-CSF. These results demonstrate that BCR–ABL establishes a state of oncogene addiction in TF1 cells, which is associated with functionally altered GM-CSF receptor (GM-CSFR) signaling. More broadly, these findings implicate rewiring of GF-R signaling pathways in the establishment of the BCR–ABL-addicted state.

To further characterize the GM-CSFR signaling axis and to investigate the molecular mechanisms through which BCR–ABL may subvert GF-R signaling, we cultured TF1/puro and TF1/BCR–ABL cells under serum- and growth factor–deprived conditions and then stimulated the cells with hGM-CSF, which activates the JAK2–STAT5, RAS–MAPK, and PI3K–AKT signaling pathways (17). Although robust activation of JAK2 was observed in both cell lines, GTP loading of RAS was observed only in hGM-CSF–stimulated TF1/puro cells (Fig. 2B, lane 3 vs. lane 7, and Supplementary Fig. S2C and S2D). We also assessed the effect of a 1-hour dasatinib pretreatment before stimulation with hGM-CSF. Again, increased RAS–GTP loading was observed only in TF1/puro cells (Fig. 2B, lane 4 vs. lane 8, and Supplementary Fig. S2C and S2D) and hGM-CSF–mediated STAT5, ERK, and AKT activation was substantially attenuated in TF1/BCR–ABL cells relative to TF1/puro cells. Collectively, these data demonstrate that BCR–ABL kinase activity negatively regulates GM-CSFR signaling, and this regulation persists for more than 1 hour following BCR–ABL kinase inhibition.

To determine whether GF-R signaling in BCR–ABL-expressing cells can be fully restored within 2 to 4 hours following kinase inhibition, as has been observed with vemurafenib-treated BRAF^V600E melanoma cells stimulated with hepatocyte growth factor (HGF), EGF, or neuregulin-1 (NRG1) (7, 9, 10), TF1/puro and TF1/BCR–ABL cells were treated with BCR–ABL TKI for up to 8 hours before hGM-CSF stimulation. Although a gradual increase in the hGM-CSF–mediated activation of STAT5, ERK, and AKT was observed with more prolonged BCR–ABL inhibition (Fig. 2C, lanes 9, 10, 11, and 12), the magnitude of pathway reactivation in TF1/BCR–ABL cells after 8 hours of BCR–ABL kinase inhibition was substantially less than that observed in TF1/puro cells (Fig. 2C, lanes 3, 4, 5, and 6). In addition, despite moderate hGM-CSF–mediated rephosphorylation of ERK in TF1/BCR–ABL cells after 8 hours of BCR–ABL inhibition, RAS activation was minimal. BCR–ABL inhibition was maintained throughout this period, as evidenced by dephosphorylation of the activation loop tyrosine Y393 (Fig. 2C, lanes 9, 10, 11, and 12).

We next assessed whether more prolonged BCR–ABL inhibition is required to enable a near complete restoration of GM-CSFR signaling. Indeed, treatment of TF1/puro and TF1/BCR–ABL cells with dasatinib for 24 hours before hGM-CSF stimulation resulted in more complete RAS activation in TF1/BCR–ABL cells (Fig. 2D). Similarly, activation of STAT5, ERK, and AKT was restored to levels comparable with those observed in TF1/puro cells (Fig. 2E). However, as demonstrated earlier, despite this delayed restoration of growth factor signaling, hGM-CSF failed to fully rescue TF1/BCR–ABL cells from TKI-mediated apoptosis (Fig. 2A). These results demonstrate an important biologic consequence of GF-R signaling rewiring by BCR–ABL: a substantial proportion of cells commit to apoptosis despite the eventual reestablishment of prosurvival growth factor signaling. Similar results were obtained when TF1/puro and TF1/BCR–ABL cells were treated with imatinib, indicating that these effects are due to inhibition of BCR–ABL and not other targets of the TKI (Supplementary Fig. S3A–S3D).

To extend our findings to a patient–derived CML cell line, we evaluated whether BCR–ABL rewires GF-R signaling in K562 cells, which express a functional erythropoietin receptor (EPOR; ref. 18). Relative to HEL erythroleukemia cells harboring the activating JAK2^V617F allele, K562 cells exhibit nearly undetectable levels of JAK2 Y1007 phosphorylation in the absence of human erythropoietin (hEPO; Supplementary Fig. S4A). Stimulation of K562 cells with hEPO led to a modest increase in JAK2 activation; 1 hour pretreatment with a BCR–ABL TKI before hEPO stimulation did not appreciably affect the degree of JAK2 activation (Fig. 3A).

We next assessed the effect of prolonged BCR–ABL inhibition (2–24 hours) on hEPO-mediated activation of JAK2 and downstream pathways in K562 cells. A time-dependent increase in hEPO-mediated STAT5, ERK, and AKT activation occurred with either dasatinib or imatinib pretreatment (Fig. 3B and Supplementary Fig. S4B). A statistically significant increase in the degree of hEPO-mediated activation of JAK2 (Fig. 3C and Supplementary Fig. S4C) and a substantial increase in RAS activity (Fig. 3D) were observed following 24 hours of BCR–ABL TKI pretreatment.

To determine whether restoration of EPO-R signaling following prolonged BCR–ABL inhibition is mediated through the canonical EPO-R–JAK2 axis, we used the selective JAK2 inhibitor TG101348. Combined 24-hour pretreatment of K562 cells with TG101348 and a BCR–ABL TKI completely prevented hEPO-mediated JAK2 activation and rephosphorylation of STAT5, ERK, and AKT (Fig. 3E and F, lanes 8, 9, 11, and 12). Similar to the partial rescue from apoptosis hGM-CSF provides TKI-treated TF1/BCR–ABL cells (Fig. 2A), hEPO partially rescues K562 cells from imatinib-mediated apoptosis (ref. 19 and Supplementary Table S3). This partial rescue from imatinib (or dasatinib)-mediated apoptosis was completely reversed by coadministration of TG101348. However, in the setting of active BCR–ABL, TG101348 had no proapoptotic effect on K562 cells (Supplementary Table S3). These observations suggest that JAK2 kinase is minimally active in the presence of BCR–ABL kinase activity, but can become critical for growth factor–mediated CML cell survival upon prolonged BCR–ABL inhibition.

To better understand the mechanism(s) responsible for the inability of GF-Rs to effectively transduce signal in BCR–ABL-expressing cells, time-dependent changes in global gene expression were assessed in K562 cells that were treated with dasatinib. In total, microarray analysis revealed 1,903 genes that were significantly differentially expressed at either 4, 8, or 24 hours of dasatinib treatment (Supplementary Table S4). As negative regulators of RAS–MAPK signaling were previously shown to be downregulated early after mitogen-activated protein kinase (MAPK) inhibition in BRAF^V600E cells (6), we reasoned that key negative regulators of EPO-R signaling in K562 cells would be enriched among genes significantly downregulated after 4 hours of dasatinib treatment. In total, 162 genes followed this pattern [false discovery rate (FDR) < 1%], including well-characterized members of the JAK–STAT and RAS–MAPK negative feedback machinery: SOCS1, SOCS2, CISH, SPRY2, SPRY4, SPRED1, SPRED2, and PIM1, but notably excluded were others, such as DUSP4 and DUSP6 (Fig. 4A). Other genes among these 162 potently downregulated genes after 4 hours of treatment were downstream effectors of MEK/ERK signaling, such as cyclin D1 (CCND1), as well as genes that traditionally comprise ERK-mediated transcriptional output including transcription factors associated with transformation (ETV5 and MYC) among other ERK targets (IER3 and EGR1).

We explored the 1,048 unique probes whose expression increased following dasatinib treatment to identify genes that become derepressed following prolonged BCR–ABL inhibition, as might be expected of positive effectors of GF-R signaling, such as JAK2 or the SRC family kinases. Several noncanonical dual-specificity phosphatases (DUSP) not traditionally involved in negative feedback regulation of the RAS pathway were among this group of genes (i.e., DUSP1, DUSP13, DUSP21, and DUSP28). Interestingly, EPOR expression increased significantly in a time-dependent manner in dasatinib-treated cells. The observed changes in a subset of these genes were validated by quantitative PCR (qPCR) in dasatinib- and imatinib-treated K562 cells (Fig. 4B). Similar changes in gene expression were observed in K562 cells treated with the MEK inhibitor PD0325901, suggesting that the RAS–MAPK pathway is primarily responsible for establishing the negative feedback network and transcriptional output associated with attenuation of myeloid GF-R signaling.

To assess the functional importance of MEK-dependent negative feedback toward GF-R signaling attenuation in CML cells, we treated K562 cells with PD0325901 for 24 hours. MEK inhibition alone (in the absence of hEPO stimulation) produced robust GTP loading of RAS (Fig. 4C, lane 3), whereas cotreatment with dasatinib and PD0325901 prevented RAS–GTP loading, demonstrating that RAS activation upon MEK inhibition is mediated by BCR–ABL and is not the result of exogenous or autocrine growth factors (Fig. 4C, lanes 3 and 4), as has been recently documented to occur in AML, where HGF secretion by leukemic cells can foster survival in the setting of kinase inhibitor treatment (20). hEPO stimulation of PD0325901-pretreated cells resulted in further RAS activation, demonstrating that efficient EPO-R signaling in K562 cells can be restored by MEK inhibition alone (Fig. 4C, lanes 3 and 7, and Supplementary Fig. S5A).

To investigate whether MEK-dependent negative feedback dampens EPO-R signaling at the level of JAK2, we assessed JAK2 phosphorylation in response to hEPO after a 24-hour pretreatment with PD0325901. hEPO-mediated JAK2 activation was comparable with that observed with a 24-hour dasatinib pretreatment (Fig. 4D and Supplementary Fig. S5B), demonstrating that MEK-dependent negative feedback acts at the level of EPO-R–JAK2 in K562 cells. No further increase in hEPO-mediated JAK2 activation was detected when cells were pretreated with the combination of dasatinib and MEK inhibitor for 24 hours before hEPO stimulation (Fig. 4D, lanes 6 and 7 vs. lane 8). In the absence of hEPO stimulation, MEK inhibition did not result in a detectable increase in JAK2 phosphorylation (Fig. 4D, lane 3).

To determine whether BCR–ABL is itself subjected to MEK-dependent negative feedback, we evaluated the effect of MEK inhibition upon the phosphorylation status of multiple BCR–ABL tyrosine residues, including the activation loop tyrosine (Y393). Across all BCR–ABL phosphotyrosine residues interrogated, little to no change in phosphorylation was observed (Supplementary Fig. S5C), suggesting that, similar to BRAF^V600E and in sharp contrast to RTKs (6), BCR–ABL both establishes and evades a high level of MEK-dependent negative feedback regulation.

The long delay observed in the restoration of EPO-mediated pathway activation in K562 cells sharply contrasts with the rapid (∼2–4 hours) and full reactivation of the MEK–ERK pathway in EGF-stimulated BRAF^V600E cells following pretreatment with vemurafenib (10). This finding, coupled with the inability of hEPO to fully rescue K562 cells from dasatinib-mediated apoptosis, suggests that TKI-treated CML cells face competing fates: growth factor–mediated survival versus commitment to apoptosis before complete restoration of GF-R signaling. To formally test this, we initially assessed the kinetics of apoptosis induction in TKI-treated K562 cells by Western immunoblot analysis. Although substantial caspase-3 cleavage was detected after 24 hours of dasatinib treatment, we observed very little evidence of apoptosis following 12 hours of treatment (Fig. 5A). To detect caspase-3 activity with a higher degree of sensitivity, we used a plasmonic nanosensor technique capable of detecting single molecule caspase-3 cleavage events through a change in light scattering intensity of a gold nanoshell pair (Fig. 5B and C; ref. 21). This assay reliably detected caspase-3 activity following only 8 hours of dasatinib treatment (Fig. 5D), demonstrating that a commitment to apoptosis can precede full restoration of GF-R signaling in patient-derived, BCR–ABL-addicted CML cells.

Although both the quality and quantity of negative feedback following inhibition of BRAF^V600E in melanoma and thyroid cancer cell lines seem to differ from inhibition of BCR–ABL in CML cells (10, 11), we sought to directly compare the molecular mediators and duration of negative feedback induced by BCR–ABL, BRAF^V600E, and an oncogenic RTK in the same cellular context. We therefore introduced BRAF^V600E, FLT3-ITD, FLT3^D835V, and BCR–ABL into TF1 cells. Although hGM-CSF–independent TF1 populations expressing BCR–ABL, FLT3-ITD, and FLT3^D835V were readily established, multiple attempts to obtain a factor-independent pool of BRAF^V600E-expressing TF1 cells failed. We carefully studied the immediate fate of TF1 cells infected with MIG–BRAF^V600E, MIG–BCR–ABL, MIG–FLT3-ITD, and MIG–FLT3D835V. Following transduction, infected cells were sorted and their growth in hGM-CSF was monitored. TF1/MIG–BCR–ABL, TF1/MIG–FLT3-ITD, and TF1/MIG–FLT3^D835V cells grew somewhat more slowly than control TF1/MIG cells, suggesting that both BCR–ABL- and oncogenic FLT3-dependent negative feedback partially dampen hGM-CSF responsiveness (Fig. 6A). Interestingly, the growth rate of TF1/BCR–ABL was significantly slower than that of TF1/FLT3-ITD and TF1/FLT3^D835V cells, suggesting that negative feedback elicited by BCR–ABL differs in magnitude from that elicited by oncogenic FLT3. In contrast, the growth of TF1/BRAF^V600E cells was significantly attenuated relative to all other cell lines despite the presence of hGM-CSF, suggesting that expression of BRAF^V600E profoundly diminishes hGM-CSF responsiveness. Indeed, high levels of ERK1/2 activation were observed in TF1/MIG–BRAF^V600E cells after transduction, and although we eventually observed hGM-CSF–supported growth of these cells, ERK1/2 activation was substantially diminished when this occurred (Supplementary Fig. S6). This observation suggests that compensation for BRAF^V600E signaling through attenuation of BRAF^V600E-directed negative feedback may be necessary for growth factor–mediated proliferation of TF1/BRAF^V600E cells.

To test whether the molecular mediators of negative feedback elicited by an oncogenic RTK differ from BCR–ABL, we first assessed the state of oncogene addiction mediated by FLT3-ITD expression in TF1 cells. Similar to TF1/BCR–ABL cells treated with BCR–ABL TKI, growth factor–deprived TF1/FLT3-ITD cells were significantly more sensitive to treatment with the potent selective FLT3 inhibitor AC220 (quizartinib) when compared with growth factor–deprived TF1/puro cells (Fig. 6B). However, in contrast to what we previously observed with TF1/BCR–ABL cells, hGM-CSF more fully rescued TF1/FLT3-ITD cells from AC220-mediated apoptosis. To assess whether the duration of negative feedback in TF1/FLT3-ITD cells following kinase inhibition differs from what we observed in TF1/BCR–ABL cells, we treated TF1/FLT3-ITD cells with AC220 for varying lengths of time before hGM-CSF stimulation. In contrast to TF1/BCR–ABL cells, hGM-CSF–mediated activation of STAT5A/B and ERK1/2 was fully restored following only 2 hours of kinase inhibition (Fig. 6C), suggesting that, similar to what is observed following BRAF^V600E inhibitor treatment in melanoma and thyroid cancer cell lines, FLT3-ITD–dependent negative feedback rapidly diminishes following FLT3 inhibition. Interestingly, the kinetics of GM-CSFR–mediated AKT phosphorylation following kinase inhibitor treatment differ between TF1/BCR–ABL and TF1/FLT3-ITD cells. In TF1/BCR–ABL cells, partial AKT rephosphorylation was observed after 2 to 4 hours of dasatinib treatment (Fig. 2C), whereas prolonged FLT3-ITD inhibition was necessary for GM-CSFR–mediated AKT phosphorylation in TF1/FLT3-ITD cells. These results highlight the substantial differences in negative feedback elicited by BCR–ABL and an oncogenic RTK.

The efficacy of BCR–ABL TKI therapy for CML surpasses that of all other U.S. Food and Drug Administration (FDA)–approved targeted kinase inhibitors, and thereby provides the most compelling clinical example of oncogene addiction. Our poor understanding of the molecular basis of this phenomenon has been due, in part, to a lack of effective models. Here, we transiently exposed CML cells to dasatinib and identified durable changes in effectors of GF-R signaling, which likely explain the cytotoxic effects of BCR–ABL inhibition. We then established and validated an isogenic system of BCR–ABL-mediated oncogene addiction and investigated its molecular features. We found that BCR–ABL-mediated oncogene addiction is the result of physiologic negative feedback inhibition that functionally and persistently dampens GF-R signaling at multiple nodes, including, in the case of EPO-R, at the level of GF-R transcription. As with BRAF^V600E in melanoma cells, BCR–ABL establishes a high level of MEK-dependent negative feedback. Relief from MEK-dependent negative feedback is required to enable full restoration of GF-R signaling and occurs only after prolonged BCR–ABL kinase inhibition. Importantly, negative feedback dampens GF-R signaling substantially longer in CML cells than what has been reported in BRAF^V600E-expressing melanoma cells. As a consequence, GF-R signaling is insufficient to fully rescue CML cells from the competing fate of apoptosis, which, for the first time, we demonstrate to be initiated as early as 8 hours following TKI treatment. As a result of this interplay between a commitment to apoptosis and the decay of negative feedback that facilitates restoration of prosurvival growth factor signaling, CML cells demonstrate a profound reliance upon BCR–ABL kinase activity for survival (Fig. 7A and B).

Both experimental and clinical evidence suggest that the effectiveness of BRAF^V600E inhibition in melanoma contrasts sharply with BCR–ABL inhibition in CML. In vitro, BRAF^V600E melanoma cell lines largely fail to undergo apoptosis when treated with BRAF inhibitors such as vemurafenib (22–25), and although the majority of melanoma patients treated with vemurafinib respond to some degree, most fail to achieve deep remissions (4). A number of BRAF^V600E-expressing melanoma, colorectal, and thyroid cancer cell lines can be rescued from BRAF inhibitor treatment by RTK ligands in vitro, leading to subsequent reactivation of MAPK and AKT signaling as soon as 1 hour after BRAF inhibitor treatment (7–11). In contrast, the corresponding time frame we observed in BCR–ABL-expressing cells is markedly different: only partial restoration of GF-R signaling occurs after 8 hours of TKI treatment, at which time a competing commitment to apoptosis has already been initiated in a substantial proportion of cells.

Despite the clinical differences observed with inhibitors of BRAF^V600E and BCR–ABL, these two oncoproteins share some biologic properties. Both are pathologically activated cytosolic kinases that establish and evade a high degree of MEK-dependent negative feedback (6). Also, a substantial core component of the ERK transcriptional output is similarly diminished when these oncoproteins are inhibited, including ERK pathway effectors, negative feedback regulators, and downstream transcription factors. For instance, both kinases seem to induce the expression of members of the Sprouty gene family, which typically target the ERK pathway for negative feedback inhibition at the level of RAF, RAS, and upstream RTKs.

There are, however, notable differences in the negative feedback networks generated by BCR–ABL and BRAF^V600E, particularly with respect to where feedback is operable. At the GF-R level, BCR–ABL kinase activity downregulates EPO-R expression in K562 cells, largely through MEK/ERK signaling, which may further contribute to the delay in restoring prosurvival GF-R–mediated signaling following BCR–ABL inhibition. In contrast, experimental evidence suggests that there is sufficient RTK expression in BRAF^V600E melanoma and colorectal cancer cells to rapidly restore prosurvival signaling (7–10). Additional differences exist in downstream components of negative feedback, particularly at the level of ERK phosphatases, which target either nuclear or cytosolic ERK for regulation downstream of the oncogenic insult. In sharp contrast to the effect of BRAF^V600E inhibition in melanoma cells where downregulation of DUSP6 is apparent at multiple levels (6, 10), we failed to detect DUSP6 at the protein level in CML cells (data not shown) and, in agreement with this observation, BCR–ABL inhibition failed to downregulate DUSP6 expression. These differences suggest that MEK-dependent negative feedback is oncogene-specific. Moreover, in BCR–ABL-expressing CML cells, feedback effectors of the parallel JAK–STAT pathway are activated (e.g., SOCS1 and SOCS2), whereas in BRAF^V600E melanoma cells, these feedback effectors are absent. These results reaffirm that the networks mediating negative feedback inhibition of oncogenic signaling are multifaceted, pleiotropic, and context-dependent. Indeed, expression of oncogenic alleles of FLT3 in TF1 cells demonstrates that activation of a clinically relevant myeloid RTK elicits a negative feedback response that is clearly distinct from that elicited by BCR–ABL in the same cellular context. Further study is required to dissect the importance of the observed qualitative and quantitative differences in negative feedback elicited by these kinases.

Our work provides insight into the role of JAK2 in the molecular pathogenesis of CML, which has been a subject of considerable controversy. It is known that CML stem cells persist for at least several years in most, if not all, patients with CML. It has been speculated that successful inhibition of BCR–ABL kinase activity in CML stem cells may be insufficient to elicit apoptosis in these cells due to their residence in a cytokine-rich bone marrow microenvironment that maintains their viability despite BCR–ABL TKI treatment (26). One group provided evidence that combined BCR–ABL and JAK2 inhibition achieves more complete eradication of CML cells in conditioned media in vitro (27), whereas another group recently demonstrated that JAK2 is completely dispensable for nearly all aspects of BCR–ABL-mediated myeloid disease initiation and maintenance (28). Although these studies offer conflicting views on the importance of JAK2 activity for CML pathogenesis, our results and proposed model of BCR–ABL-dependent attenuation of GF-R signaling resolves this apparent paradox. When BCR–ABL is active, JAK2 plays little or no role in signal transduction due to MEK-dependent negative feedback-mediated attenuation of GF-R signaling. After prolonged BCR–ABL inhibition, negative feedback is relieved and JAK2 may become critically important as a mediator of STAT5 phosphorylation in the setting of external growth factors. Our data provide mechanistic support for clinical efforts to combine BCR–ABL TKIs with potent JAK2 TKIs, which are predicted to have no activity in the absence of BCR–ABL inhibition, but might demonstrate a synthetic lethal interaction with BCR–ABL TKIs in the setting of exogenous growth factors, specifically as a consequence of the decay in negative feedback we describe here. If JAK2 serves a redundant function in facilitating prosurvival GF-R–mediated signaling in the bone marrow microenvironment, alternative approaches will be required.

The oncogenic shock model of oncogene addiction (3) was proposed with limited supporting data, and the molecular mediators of oncogene addiction have remained elusive. Our data strongly implicate persistent negative feedback in the promotion of apoptosis following kinase inactivation, and thus provide valuable molecular mechanistic insights into the basis of both oncogene addiction and the oncogenic shock model. The recent appreciation of how rapidly negative feedback decays in BRAF^V600E-expressing systems and thereby facilitates prosurvival signal transduction by GF-Rs has led to treatment approaches that combine vemurafenib with multikinase GF-R inhibitors. However, given the growing number of signaling molecules that have been found to become activated following inhibition of BRAF^V600E, safe and effective inhibition of all relevant GF-Rs may prove difficult. Our studies of BCR–ABL-mediated oncogene addiction suggest an alternative approach aimed at potentiating negative feedback in the setting of inhibition of pathologically activated kinases, which is predicted to enhance oncogene addiction and increase sensitivity to targeted therapy. In addition to possibly facilitating deeper clinical responses, this strategy may be applicable to a broad range of malignancies associated with activating mutations in actionable signaling molecules. Efforts to further dissect and modulate molecular mechanisms of negative feedback in relevant tissue contexts are required to formally test this prediction.

Cell lines were propagated in RPMI-1640 medium supplemented with 10% FBS (Omega Scientific), l-glutamine, and penicillin/streptomycin (Invitrogen). K562 and KU812 cells were purchased from American Type Culture Collection. TF1 cells were a kind gift from Michael Tomasson (Washington University, St. Louis, MO). All cell lines were authenticated by Promega short-tandem repeat (STR) analysis. To generate isogenic cell lines, TF1 cells were engineered to express the ecotropic-receptor (Eco-R) through retroviral transduction (pMOWS-EcoR plasmid) as previously described (29). TF1–Eco-R cells underwent a second round of retroviral transduction with pMSCV-puro or pMSCV-BCR–ABL, followed by puromycin selection. TF1/BCR–ABL cells were selected for growth factor independence and TF1/puro cells were supplemented with 2 ng/mL hGM-CSF (Peprotech) under normal growth conditions.

Imatinib and dasatinib were purified at the University of California, San Francisco (UCSF; San Francisco, CA). TG101348, PD0325901, and AC220 were purchased from Selleckchem. All drug exposures were performed at a density of 5 × 10⁶ cells/mL. For cytokine stimulation, cells were starved in 0.1% FBS-RPMI as indicated and subsequently stimulated with 1 U/mL hEPO (R&D Systems) or 10 ng/mL of hGM-CSF. Dasatinib HDP drug exposures were performed as described previously (12).

All statistical analyses were performed in Prism (GraphPad) using a two-way ANOVA with Bonferroni posttests.

K562 cells were grown for 6 days in customized RPMI-1640 supplemented with heavy (Cambridge Isotope Laboratories; CLM-2265 and CLM-2247) or light (Sigma-Aldrich) 30 mg/L arginine, 40 mg/L lysine, and 10% dialyzed FBS (Invitrogen) before drug treatment. The PY100 Phosphoscan Kit (Cell Signaling Technology) was used to enrich phosphotyrosine-containing tryptic peptides. To reduce sample complexity before tandem mass spectrometry (MS-MS) analysis, phosphotyrosine peptides were fractionated stepwise under alkaline conditions (20 mmol/L ammonium formiate, pH 10.3) with increasing amounts of acetonitrile (5% ACN, 10% ACN, 15% ACN, 20% ACN, and 90% ACN). Peptides were separated and analyzed using a nano-LC column coupled to a LTQ OrbiTrap XL (Thermo Scientific) at the UCSF Mass Spectrometry Core Facility. An in-house software analysis program was used to extract quantitative information for all identified peptides from each nano-LC analysis for all time points. Heatmap representation of the phosphoproteomic data was generated using GENE-E (http://www.broadinstitute.org/cancer/software/GENE-E/). We functionally characterized phosphosites durably altered by HDP dasatinib treatment relative to all identified phosphoproteins (background) using the DAVID tool with default parameters (30, 31)

Cells were lysed in JAK2 immunoprecipitation buffer (50 mmol/L Tris pH 7.6, 100 mmol/L NaCl, 1 mmol/L EDTA, 1 mmol/L EGTA, 0.5% NP40, and 0.1% Triton) or RAS immunoprecipitation buffer (50 mmol/L Tris pH 7.5, 125 mmol/L NaCl, 6.5 mmol/L MgCl₂, 5% glycerol, and 0.2% NP40) supplemented with 1% protease and 1% phosphatase inhibitors (Calbiochem). For JAK2 immunoprecipitations, normalized lysates were tumbled overnight at 4°C with anti-total JAK2 antibody (1:100; Cell Signaling Technology; cat. #3230). Immunoprecipitated JAK2 was collected with protein A Dynabeads (Invitrogen). For RAS–GTP pull-down assays, normalized lysates were tumbled for 1 hour at 4°C with 20 μL of RAS Assay Reagent (Millipore; #14-278).

Cell lysates were resolved by SDS-PAGE and transferred to nitrocellulose as described previously (12). Phospho and total antibodies for STAT5A/B (phospho-Y694/Y699; cat. 9351 and 9363), ERK1/2 (phospho-T202/Y204; cat. 4370 and 9107), AKT (phospho-S473; cat. 4060 and 9272), S6 (phospho-S235/S236; cat. 2211 and 2317), CRKL (phospho-Y207; cat. 3181), JAK2 (phospho-Y1007; cat. 3776 and 3230), BCR (phospho-Y177; cat. 3901), cABL (phospho-Y245 and phospho-Y204; cat. 2861 and 3009), and cleaved CASP3 (cat. 9664) were purchased from Cell Signaling Technology. cABL (cat. OP20) and total RAS (cat. 05-516) antibodies were purchased from Millipore. cABL (phospho-Y134; cat. AP3011a) and cABL (phospho-Y251; cat. AP3014a) antibodies were purchased from Abgent. Total glyceraldehyde-3-phosphate dehydrogenase (GAPDH; cat. sc-25778) was purchased from Santa Cruz Biotechnology. Odyssey IR imaging technology and software (LI-COR Biosciences) was used for Western immunoblot analysis visualization and quantitation was measured by pixel-integrated intensity. Cleaved caspase-3 was measured by flow cytometry using fluorescein isothiocyanate (FITC)–conjugated anti-active caspase-3 antibody (BD; cat. #550480). Analysis of CRKL phosphorylation by flow cytometry was performed as previously reported (12).

Total cellular RNA was extracted using the Qiagen RNeasy Kit. For Illumina gene expression analysis, RNA integrity was assessed on a Bioanalyzer using the Agilent RNA 6000 Nano Kit (5067-1511), and cRNA was generated by Ambion Illumina TotalPrep RNA Amplification kit (AMIL1791). SuperScript II (Invitrogen) was used to generate cDNA from 2 μg of extracted total cellular RNA. Samples were hybridized to the Illumina HT-12 platform by the UCSF Genomics Core, and both raw control and sample probe intensities were converted to expression estimates after adjusting for array background, variance stabilization, and normalization, all with the lumi pipeline (PMID: 18467348). Differential gene expression was assessed as a time course with a two-step regression strategy (combining least-squares and stepwise regression) to identify genes with significant temporal expression changes between experimental groups (PMID: 16481333). Genes were considered significantly differentially expressed if they arose at an FDR less than 1% (q-value < 0.01). Confirmatory qPCR analysis was performed using TaqMan probes and TaqMan Universal Master Mix II (Invitrogen): GAPDH (Hs02758991_g1), SOCS1 (Hs00705164_s1), GRB10 (Hs01065498_m1), SPRY4 (Hs01935412_s1), EPOR (Hs00959427_m1). Expression array data generated in this study were deposited in the NIH National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO), accession number: GSE51083 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE51083).

Techniques were adapted from ref. 21 and modified by C. Tajon, Y.-W. Jun, and C.S. Craik (unpublished data). Briefly, a pair of magnetoplasmonic Zn_0.4Fe_2.6O₄@SiO₂@ gold (Au) nanoshells (50 nm) linked by a PEGylated-peptide bearing a selective caspase-3 cleavage site were synthesized. Lysates were prepared from K562 cells pretreated for 8 hours with dimethyl sulfoxide (DMSO) or dasatinib, and introduced to nanosensors immobilized on a glass flow chamber. As a control, 100 μmol/L zDEVD-cmk was added for 3 hours. Imaging was performed on an inverted microscope (Nikon Ti-E) outfitted with a darkfield dry condenser and recorded at a temporal resolution of 10 Hz using an EMCCD detector (Andor iXon; 512 × 512 pixel chip). Intensity trajectories of each nanoparticle pair (n = 75) were analyzed by ImageJ. Total cleavage events observed 90 minutes following lysate introduction were counted, normalized against control, and plotted.

TF-1 EcoR-expressing cells were retrovirally transduced to express MIG–BCR–ABL, FLT3-ITD, FLT3^D835V, or BRAF^V600E as described above. Forty-eight hours after infection, GFP⁺ cells were sorted on a BD FACSARIA II (BD) and equivalent cell numbers were plated in 10% RPMI supplemented with 2 ng/mL hGM-CSF. Cells were counted as indicated in triplicate by Trypan blue exclusion. The medium and hGM-CSF were replaced as necessary and cell populations were maintained in an exponential growth phase. GFP positivity and phospho-ERK were monitored daily by flow cytometry. Growth curves were generated by multiplying the cell count by percentage GFP positivity.

N.P. Shah has received commercial research grants from Bristol-Myers Squibb and Ariad Pharmaceuticals and is a consultant/advisory board member of Bristol-Myers Squibb and Ariad Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Asmussen, E.A. Lasater, J. Oses-Prieto, N.P. Shah

Development of methodology: J. Asmussen, C. Tajon, J. Oses-Prieto, A. Burlingame, C.S. Craik

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.A. Lasater, J. Oses-Prieto, A. Burlingame

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Asmussen, C. Tajon, J. Oses-Prieto, B.S. Taylor, A. Burlingame, N.P. Shah

Writing, review, and/or revision of the manuscript: J. Asmussen, E.A. Lasater, C. Tajon, J. Oses-Prieto, B.S. Taylor, N.P. Shah

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Asmussen

Study supervision: C.S. Craik, N.P. Shah

Provision of single molecule imaging technique: Y.-W. Jun

The authors thank Kevin Shannon and Benjamin Braun for insightful discussions. N.P. Shah is a Leukemia & Lymphoma Scholar in Clinical Research, and wishes to acknowledge the generous support of Arthur Kern, Bob Bonstein, Mark Maymar, and the Edward S. Ageno family.

J. Asmussen received support from the UC Cancer Research Coordinating Committee (CRCC) Fellowship and from the NIH Training Grant T32 GM007175. Mass spectrometry analysis was provided by the Bio-Organic Biomedical Mass Spectrometry Resource at UCSF (A. Burlingame, Director) supported by funding from the Biomedical Technology Research Centers program of the NIH National Institute of General Medical Sciences, 8P41GM103481. E.A. Lasater was supported by NIH National Cancer Institute T32CA108462-07. Y.-W. Jun was supported by the National Institute of Biomedical Imaging and Bioengineering and NIH (NIH 1R21EB015088). C.S. Craik received support from the NIH R01CA128765. C. Tajon was supported by the UCSF Quantitative Biosciences Consortium (QBC) Fellowship and NIGMS-IMSD award (R25-GM56847).