Integrated proximal proteomics reveals IRS2 as a determinant of cell survival in ALK-driven neuroblastoma

Abstract

Oncogenic anaplastic lymphoma kinase (ALK) is one of the few druggable targets in neuroblastoma, and therapy resistance to ALK-targeting tyrosine kinase inhibitors (TKIs) comprises an inevitable clinical challenge. Therefore, a better understanding of the oncogenic signaling network rewiring driven by ALK is necessary to improve and guide future therapies. Here, we performed quantitative mass spectrometry-based proteomics on neuroblastoma cells treated with one of three clinically relevant ALK TKIs (crizotinib, LDK378, or lorlatinib) or an experimentally used ALK TKI (TAE684) to unravel aberrant ALK signaling pathways. Our integrated proximal proteomics (IPP) strategy included multiple signaling layers, such as the ALK interactome, phosphotyrosine interactome, phosphoproteome, and proteome. We identified the signaling adaptor protein IRS2 (insulin receptor substrate 2) as a major ALK target and an ALK TKI-sensitive signaling node in neuroblastoma cells driven by oncogenic ALK. TKI treatment decreased the recruitment of IRS2 to ALK and reduced the tyrosine phosphorylation of IRS2. Furthermore, siRNA-mediated depletion of ALK or IRS2 decreased the phosphorylation of the survival-promoting kinase Akt and of a downstream target, the transcription factor FoxO3, and reduced the viability of three ALK-driven neuroblastoma cell lines. Collectively, our IPP analysis provides insight into the proximal architecture of oncogenic ALK signaling by revealing IRS2 as an adaptor protein that links ALK to neuroblastoma cell survival through the Akt-FoxO3 signaling axis.

Introduction

Neuroblastoma (NB) is the most common extracranial childhood tumor. Tumors arise from the sympathetic nervous system and account for about 15% of pediatric cancer mortality (1). NB is characterized by clinical and biological heterogeneity causing non-uniform responses to treatment. The likelihood of cure varies widely according to age at diagnosis, disease stage and tumor biology. Despite advances in treatment, high-risk NB continues to have a poor prognosis and a survival rate below 50% (1). Moreover, current therapeutic strategies in oncology, which rely on targeting of oncogenic drivers, cannot be applied to NB because there are few recurrent somatic mutations associated with high-risk NBs (2). An improved understanding of how key oncogenic drivers support the disease will allow for the identification of sensitive nodes that are targetable, resulting in improved patient outcome and survival.

The most malignant tumors harbor amplification of the MYCN oncogene (around 20%), which is used as a biomarker for NB risk stratification (3, 4). Transgenic mouse and zebrafish models show that MYCN overexpression serves as a tumor-initiating factor (5); however, cooperating genes accelerate NB tumor development and pathogenesis. These genes are clinically critical because MYCN remains a challenging target to directly inhibit (6). Among MYCN cooperating genes, mutated anaplastic lymphoma kinase (ALK) synergistically induces NB tumors in mouse and zebrafish models (7–9) and represents a target for precision therapy of high-risk NBs (2). ALK is a receptor tyrosine kinase (RTK) for which several ligands have been identified, including heparin and members of the FAM150 protein family (10, 11). ALK is the major predisposition gene for familial NB (12, 13), and oncogenic ALK signaling drives a substantial subset of sporadic NBs. Thus, up to 3% and 8-10% of these cases are supported either by ALK amplification (ALK^Amp) or gain-of-function point mutations, respectively (2, 12–16). These mechanisms render the ALK receptor constitutively active, and thus, tractable for therapeutic intervention in NB.

Several highly potent and selective ALK-targeted tyrosine kinase inhibitors (TKIs) - TAE684, crizotinib, LDK378 (ceritinib) and lorlatinib - effectively block ALK-driven cell growth in NB cell lines and tumor models (8, 13, 15, 17–20). While TAE684 did not advance beyond preclinical use, US Food and Drug Administration approved crizotinib as a first-in-line drug for ALK-positive non-small cell lung cancer (NSCLC) and designated LDK378 and lorlatinib as breakthrough therapies (21, 22). These compounds are currently undergoing clinical evaluation for ALK-positive malignancies including NB (ClinicalTrials.gov: NCT01121588, NCT01742286, NCT03107988) (23). Although crizotinib, LDK378 and lorlatinib may prove beneficial for high-risk NB, therapy resistance will mostly likely challenge any real clinical benefit as has been evident for crizotinib in NSCLC and NB (23–25). Thus, the aim of our study was to increase knowledge about the TKI-sensitive nodes, as this may reveal ways to improve treatment efficacy by identifying residual signaling proteins or unaffected nodes that could be co-targeted to ultimately postpone or overcome resistance.

Mass spectrometry (MS)-based quantitative phosphoproteomics is a powerful technology to unravel RTK signaling by providing an unbiased and comprehensive way of measuring signaling responses (26, 27). However, phospho-signaling networks are dynamic and highly complex due to the interplay between adaptor proteins, kinases and phosphatases. Thus, oncogenic network rewiring involving proximal and distal nodes can be challenging to decipher from properties of individual signaling layers such as the phosphoproteome alone. Detailed mechanistic insight can be derived from integration of various proteomics data sets and aid in moving towards a comprehensive understanding of network connectivity (28–30). We and others have specifically shown that integration of proximal signaling information through analysis of the RTK interactome and phosphoproteome can identify protein subnetworks driving cell fate decision-making (29, 30). Consequently, understanding the role of proximal signaling adaptors can be beneficial in the quest to identify immediate downstream amplifiers of oncogenic signaling drivers with pronounced impact on cell-fates and thus, previously unknown targetable nodes.

A few phosphoproteomics studies have successfully addressed constitutive ALK signaling through an inhibitor-based approach using crizotinib in NSCLC and NB cells (30–32). However, for NB specifically, the depth of analysis regarding the number of quantified phosphotyrosine sites has been insufficient to capture many of the important regulatory sites (31). Here, we measured the responses of NB cells with ALK^Amp to treatment with TAE684, crizotinib, LDK378 and lorlatinib by quantitative proteomics. Our integrated proximal proteomics (IPP) strategy included multiple signaling layers − the ALK interactome, phosphotyrosine interactome, phosphoproteome and proteome − to provide a quantitative map of ALK TKI-sensitive signaling nodes in NB cells. Our IPP resource offered insight into the architecture of ALK proximal signaling by revealing a critical role for insulin receptor substrate 2 (IRS2) as the adaptor molecule linking ALK to NB cell survival through the Akt-forkhead box protein O3 (FoxO3) signaling axis.

Results

Integrated proximal proteomics (IPP) refines aberrant ALK signaling networks in NB cells

To study the aberrant ALK signaling network in NB cells we used the NB1 cell line which harbors amplification of full-length ALK and thus has high endogenous expression and activation of ALK (33). To modulate constitutive ALK activity, we treated NB1 cells for 48 h with different concentrations of three ALK-targeting TKIs (TAE684, LDK378 and crizotinib), measured cell viability and generated dose-response curves (Figure 1A). Each of the three TKIs reduced NB1 cell viability in a dose-dependent manner, confirming ALK as an oncogenic driver of cell growth in this model. Thus, we considered the NB1 cell line to be a therapeutically relevant model for our subsequent proteomics analyses. We first determined the half maximal inhibitory concentration (IC₅₀) for each inhibitor and used concentrations within the 95% confidence interval (CI): 100 nM for TAE684, 250 nM for LDK378 and 500 nM for crizotinib (Figure 1A). Then, we ensured that ALK signaling was effectively inhibited at these concentrations by performing time-course analyses (Figure 1B and C, fig. S1A and B). Immunoprecipitation of ALK revealed that all inhibitors markedly reduced the levels of phosphorylated ALK (Tyr¹⁶⁰⁴) and abrogated the interaction with the ALK adaptor protein Shc (34), as soon as 15 min after treatment (Figure 1B, fig. S1A). Furthermore, two ALK downstream targets, extracellular signal-regulated kinase (ERK), and Akt (also known as protein kinase B), displayed a time-dependent reduction in phosphorylation as an indicator of their inhibition (Figure 1C, fig. S1B). Due to detection of residual phosphorylated ALK in LDK378- and crizotinib-treated cells up to 30 min after treatment (Figure 1C and fig. S1B), we chose the 30-minute time point for all subsequent analyses and used the inhibitors at their respective IC₅₀-concentrations to ensure robust inhibition of ALK signaling.

Fig. 1. Multilayered proteomics approach to study potentially druggable ALK signaling in neuroblastoma cells.

(A) Cell viability of NB1 neuroblastoma cells in response to treatment (48 h) with different concentrations of crizotinib, LDK-378 and TAE-684. Data are presented as mean ± SEM of N=3-6 independent experiments. (B, C) Lysates from NB1 cells treated with either DMSO or LDK-378 (250 nM) for different times and immunoprecipitated for ALK (B) or immunoblotted (B, C) as indicated (N=3 independent experiments). p, phospho. Arrows indicate protein variants as previously described (84–86). (D) Schematic representation of the proteomics strategy using crizotinib, LDK-378 and TAE-684 to inhibit constitutive ALK signaling in NB1 cells. Drug-induced changes in the ALK interactome and phosphoproteome were measured after 30 min of inhibitor treatment including mapping of the ALK phosphotyrosine interactome and proteome analysis of untreated NB1 cells. (E) Overview of results from ALK interactome (yellow; N=2 independent experiments for each inhibitor; Significance B test; P<0.05), phosphotyrosine interactome (green; N=4 pulldowns for each pY-peptide (bait) and non-p-peptide (control); t test for significance, P<0.05, S-score>1), phosphoproteome analysis (blue; N=2 independent experiments for each inhibitor) and proteome (pink; N=2 independent experiments). (F) Number of total regulated phosphorylation sites by amino acid distribution (determined as previously described (28)). (G) Overlap between the TKI-regulated phosphoproteome and the identified and quantified proteome (upper) and adaptors with decreased ALK association (interactome) upon TKI treatment (lower). See also Figures S1-S2 and Data Files S1-S4.

To investigate proximal ALK signaling in NB cells, we performed a large-scale MS-based quantitative proteomics analysis of four signaling layers: interactome, phosphotyrosine interactome, phosphoproteome, and proteome (Figure 1D). To enable quantitative comparisons, we used a combination of stable isotope labeling by amino acids in cell culture (SILAC) (35) (for the interactome and phosphoproteome) and label-free approaches (for the phosphotyrosine interactome and proteome) (fig. S1C) combined with high-resolution liquid chromatography–tandem mass spectrometry (LC-MS/MS). We reasoned that combining the quantitative proteomics analysis for three ALK-targeting TKIs (TAE684, LDK378 and crizotinib) and analyzing the overlapping effects on ALK signaling would help to control for off-targets and serve as a better readout for ALK signaling inhibition compared to the previously used single-treatment strategy (30, 31). Moreover, we performed three triple-SILAC experiments allowing us to measure the effect of each inhibitor in duplicate to obtain more robust insights into inhibitor-specific effects (fig. S1C). To focus on ALK proximal signaling, we performed MS analysis on ALK immunoprecipitates from lysates of inhibitor-treated NB1 cells to analyze the interactome. We identified and quantified 1,467 proteins and obtained good reproducibility between the effects of each inhibitor across SILAC experiments (fig. S2A-B and Data File S1). We identified 51 proteins whose association with ALK was significantly abrogated upon treatment with the three inhibitors (Figure 1E and Data File S1). In addition to known signaling interactors of full-length ALK such as Shc and FRS2 (34, 36), we identified several proteins not previously reported to associate with full-length ALK, including the tyrosine-protein phosphatase non-receptor type 11 (PTPN11) and IRS2 (Data File S1). To confirm the interaction between ALK and the identified interactors we additionally performed an ALK phosphotyrosine interactome analysis (fig. S1C). Because the phosphorylated tyrosine residues on ALK serve as docking sites for the immediate downstream mediators of ALK signaling, we generated phosphotyrosine-containing peptides derived from ALK and their corresponding non-phosphorylated peptides, incubated them with lysate from NB1 cells treated with DMSO or LDK378 and analyzed enriched proteins pulled down by these peptides by LC-MS/MS (fig. S1C). We reasoned that including lysates from ALK inhibitor-treated cells would allow us to identify direct binders of ALK phosphotyrosines, assuming that secondary indirect binders would depend on ALK-driven phosphorylation for interaction. From this peptide pulldown analysis, we identified 20 proteins with phosphotyrosine binding (PTB), phosphotyrosine interaction (PI) or Src-homology 2 (SH2) domains. The presence of these domains suggested that these proteins directly bound to phosphorylated ALK tyrosine-containing peptides (Figure 1E, Data File S2).

To analyze the phosphoproteome, pooled lysates from each triple-SILAC experiment were enriched for phosphorylated peptides by anti-phosphotyrosine immunoprecipitation followed by two sequential rounds of TiO₂ enrichment and analysis by LC-MS/MS (fig. S1C). We identified and quantified 16,617 phosphorylated sites, 13,327 of which were confidently localized to serine (84.0% of the total or 11,198 sites), threonine (10.9% or 1,451 sites), or tyrosine (5.1% or 678 sites) residues in the peptide sequence (class I) within 4,637 proteins (Figure 1E and 1F, and Data File S3). Ultimately, this study expanded the coverage of identified and quantified phosphotyrosine sites in ALK-driven NB cells compared to a previous study (fig. S2C and S2D) (31). Although Chen et al. analyzed three NB cell lines, the total number of phosphotyrosine sites did not exceed 397, and there was limited overlap between cell lines (fig. S2C). However, despite differences in experimental setups regarding ALK TKIs, concentrations and treatment time, there was a good agreement in the regulation of sites in ALK, IRS2 and ERK1/2 (MAPK1/3) between our analyses and that of Chen et al. (fig. S2D). We deemed phosphorylation sites to be regulated if their ratios were higher or lower than the 2.5% most up- or down-regulated non-phosphorylated peptides, respectively; thus, cut-offs were individually determined for each inhibitor as previously described (29). Among the regulated phosphorylated sites 1,733 showed an increased ratio (13%), whereas 1,424 sites had decreased ratios (10.7%) (Figure 1F). Phosphorylated tyrosine residues were almost two-fold enriched among the up- and down-regulated sites (from 5.1% to 9.1% and 11.5%, respectively) (Figure 1F), supporting a greater role for tyrosine phosphorylation in the signaling downstream of aberrant ALK. The total 3,157 regulated phosphorylation sites were derived from 1,849 proteins and compared to the measured NB1 cell proteome of 10,066 proteins, showing good reproducibility, the regulated phosphoproteome comprised 14% of the measured proteome (Figure 1G, fig. S2E). These findings underscore the extent to which inhibitor treatment perturbs the cellular machinery of signaling proteins. Moreover, 37% of regulated proximal signaling adaptors also displayed regulation at the phosphorylation level suggesting a tight control of proximal functions in signaling transmission and control (Figure 1G).

IPP reveals drug-sensitive ALK adaptors

To refine ALK proximal signaling across multiple signaling layers, we integrated the four proteomics data sets and focused on the thirty most regulated ALK adaptors from the interactome analysis (Figure 2A). We ranked the interactors according to highest fold-change in SILAC ratio upon inhibitor-treatment, revealing several well-established sub-complexes such as the Shc-, Gab1- and IRS-complex. Many of the identified interactors displayed regulation involving several phosphorylated residues, including ALK (5 Tyr residues, 1 Ser residue, 1 Thr residue) and IRS2 (8 Tyr residues, 4 Ser residues, 1 Thr residue) (Figure 2A). Because of the large number of regulated phosphotyrosine sites for both ALK and IRS2, we proposed that IRS2 was a central node for ALK signaling transmission. Furthermore, when the protein abundance as measured by iBAQ quantification in the NB1 proteome analysis (37) was taken into account (Figure 2A), it was evident that the more prominently regulated interactors, or those with the highest fold changes in SILAC ratio upon inhibitor treatment, were of lower abundance in the proteome compared to ALK, whereas proteins ranked with lower SILAC ratios in the interactome generally were more abundant. These findings suggest that functional protein-protein interactions are achieved through a high degree of specificity independently of protein abundance. For each interactor, we mapped the results of the ALK phosphotyrosine interactome to include only significant interactions (Figure 2A). Here, the direct binding of the majority (8 of 11) of PTB, PI and SH2 domain-containing interactors to one or more ALK phosphotyrosine residues was confirmed (for example, IRS2 binding to Tyr¹⁰⁹⁶, Tyr¹⁵⁰⁷, and Tyr¹⁵⁸⁴). Moreover, a robust association (Figure 2A; marked in red) for Shc1 and Shc3 binding to Tyr¹⁵⁰⁷ in the Shc PTB domain consensus sequence (NPTpY) (34, 36) was identified, confirming the validity of our approach. Other proteins with robust association included the regulatory subunits of phosphoinositide 3-kinase (PI3K), PIK3R1 and PIK3R2; the protein tyrosine phosphatase PTPN11, and the adaptor protein SH2B2 (Figure 2A, fig. S3A). The phosphotyrosine interactome revealed additional PTB, PI and SH2 domain-containing interactors such as the phospholipase PLCγ and the transcription factors STAT1 and STAT3 (fig. S3A), but these were not affected by TKI-treatment, and thus, we considered them to be TKI-insensitive interactors in our NB cell model. Interactors found in both the interactome and the phosphotyrosine interactome may directly associate with ALK (Figure 2B). PI3K subunits bind to many sites as shown in other cellular contexts (38). This finding is consistent with our hypothesis on the key role played by IRS2 in the control of ALK proximal signaling as IRS2 also contains multiple PI3K binding motifs (38). A functional network based on STRING of TKI-sensitive ALK interactors grouped these into two main clusters with functions related to RTK signaling and glycolysis (Figure 2C). These findings were also confirmed by Gene Ontology (GO) term enrichment analysis for biological process (Figure 2D). We validated the interaction of PI3K (p85 and p110), PTPN11, Grb2, IRS2 and SH2B1 with full-length ALK by immunoprecipitation analyses and showed that TKI-treatment substantially abrogated their association to ALK (Figure 2E). Moreover, we validated the decrease in phosphorylation of IRS2 tyrosine residues and in the ALK/IRS2 interaction by the ALK-TKIs by immunoprecipitating IRS2 (Figure 2F). Because of the association between IRS2 and PI3K, we confirmed that IRS2 interacts with PI3K in NB1 cells through two tyrosine residues of IRS2 (Tyr⁶⁷⁵ and Tyr⁹⁷⁸) that reside in a classical PI3K PTB-binding motif (39) (Figure 2G, fig. S3B and S3C).

Fig. 2. Proximal signaling proteomics reveals drug-sensitive ALK adaptors.

(A) Four integrated heatmaps representing results from the ALK interactome, phosphotyrosine interactome, phosphoproteome and proteome analysis. The list includes ALK and the 30 most TKI-sensitive (as assessed by decreased association) ALK adaptor proteins (shown by gene name) and displays the median log2 SILAC ratio (yellow) from the ALK interactome analysis, the number of phosphorylation sites (blue; phosphoproteome) with significantly decreased ratios in response to at least two inhibitors, the relative protein abundance by iBAQ value (pink; proteome) and the significantly associated with phosphotyrosine-specific binding (green; phosphotyrosine interactome; P<0.05 and in red S-score>1). (B) Overview of site-specific phosphotyrosine interactions. Data is a graphical summary of figure S3. TKI-sensitive as well as TKI-insensitive adaptors are highlighted. Only proteins containing SH2, PTB and PI domains are included. Each square indicates a median SILAC ratio for the indicated ALK tyrosine residue. (C) Functional association network based on STRING and visualized by Cytoscape. ALK is grey and IRS2 has a pink halo. (D) Significantly overrepresented GO terms for biological process among proteins listed in (A). (E-F) Lysates from NB1 cells treated with either DMSO, crizotinib, TAE-684 or LDK-378 for 30 min and immunoprecipitated for ALK (E) or IRS2 (F) and immunoblotted as indicated (N=3 independent experiments). (G) Volcano plot showing the phosphotyrosine specific interactors of the phosphorylated IRS2 Tyr⁶⁷⁵-containing peptide. SH2, PI, PTB and Cbl-like PTB domain-containing proteins are indicated by gene name and a star. Log2 ratios of fold change the median intensities of pull-downs (N=4 independent experiments) of phosphorylated peptide (bait) versus non-phosphorylated peptide (control) (x-axis) are plotted versus −log10 of the p-values derived from a t-test. Significant associations are represented above the S-curve. See also figure S3 and Data Files S1-S2.

The druggable ALK phosphoproteome identifies ALK-driven phosphorylation of FoxO3

To explore ALK signaling downstream of the constitutively active receptor, we searched for phosphorylated sites that had either up- or down-regulated ratios by at least two inhibitors and found 697 sites with down-regulated ratios and 634 sites with up-regulated ratios (Figure 3A). Small molecule inhibitors directed toward a common primary target can share potential off-targets. Thus, we examined the protein expression levels and the potential phosphorylation regulation of the top-5 off-targets reported by Klaeger et al. for LDK378 and crizotinib including those reported by the suppliers (fig. S4A) (40). 10 of 24 potential off-targets were not detected in our deep proteome analysis of NB1 cells. Moreover, among the 14 detected proteins, only two proteins - the RTK epidermal growth factor receptor (EGFR) and focal adhesion kinase 1 (PTK2) - were identified with a single site each being down-regulated by minimum two ALK inhibitors (Data File S3). Knowing that our inhibitor-based approach may represent the combination of on-target and off-target effects, we conclude that the main reported off-targets did not confound the interpretation of our data.

Fig. 3. Phosphoproteomics identifies ALK-driven phosphorylation of FoxO3.

(A) Overlap in number of identified and quantified phosphorylation sites in NB1 cells treated with TAE684, LDK378 and crizotinib. (B) Kinase substrate motif enrichment analysis (Fisher’s exact test) including phosphorylation sites common to at least two out of three inhibitors and displaying either decreased (697 sites) or increased (634 sites) ratio in response to inhibitors. (C) Sequence motif analysis by iceLogo of the ± 6 amino acid residues flanking the regulated phosphorylation site (left: tyrosine-specific, right: serine/threonine) compared to sites (tyrosine, serine and threonine) that are not regulated. (D) KEGG pathway enrichment analysis (Fisher’s exact test) for proteins with sites displaying a decreased and increased ratio in response to TKI treatment. (E) Overview of phosphorylation regulation of transcription factors ranked according to their five most prominently decreased phosphorylation sites in response to ALK TKIs. Each square corresponds to ALK-inhibitor-induced changes in phosphorylation sites as indicated. (F) Lysates from NB1 cells treated with either DMSO, crizotinib, TAE-684 or LDK-378 for 30 min and immunoblotted as indicated (N=3 independent experiments). See also Figure S4.

Kinase substrate motif enrichment analysis of these commonly regulated sites (by two out of three inhibitors) revealed overrepresentation of Src substrates among both the down- and upregulated pools whereas kinases such as phosphorylase kinase, MAPKAPK1 and Akt targeted the downregulated phosphorylation sites (Figure 3B). While proline-directed kinase substrates were overrepresented among up-regulated sites, a greater diversity of kinase substrates was evident among down-regulated sites. For instance, substrate motifs of PKA, Akt and PKC were significantly overrepresented among down-regulated sites. This finding was confirmed by sequence motif enrichment analysis, which showed overrepresentation of arginine (R) in the -3 and -5 position relative to the serine or threonine phosphorylation site (Figure 3C), which is the canonical motif for PKA, Akt, and PKC (39). Moreover, phosphotyrosine sites with ALK kinase substrate motifs were overrepresented in the down-regulated pool, supporting the validity of our TKI-approach to unravel ALK signaling (Figure 3B and fig. S4B). Among these, PTPN11, Shc and IRS2 had the greatest decrease in SILAC ratios, confirming a dual regulation of ALK proximal adaptors at the interactome and phosphorylation levels (fig. S4B). Finally, the sequence motif of down-regulated phosphotyrosine sites revealed the presence of SH2-domain binding motifs for PTPN11 (SHP2) (pY(I/V)X(I/V)) and PI3K (pYMXM and pYXXM), underscoring their involvement as effectors of ALK phosphotyrosine signaling (Figure 3C). Consistently, down-regulated phosphotyrosine sites of IRS2 mainly contained the PI3K-motifs (fig. S3C).

Analysis of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment confirmed that proteins with down-regulated phosphorylation sites were shared among other canonical RTK signaling pathways (Figure 3D) in agreement with a previous study (31). In contrast, proteins with up-regulated sites were significantly associated with pathways related to axon guidance, prolactin signaling, regulation of actin cytoskeleton and signaling related to the class O subfamily of Forkhead box (FoxO) transcription factors (Figure 3D). The transcriptional activity of FoxO proteins is controlled by post-translational modifications, including Akt-mediated phosphorylation at a critical inhibitory site (Ser²⁵³), which excludes FoxO from the nucleus resulting in repression of transcriptional activity in NB cells (41). In support of ALK TKI-dependent inhibition of Akt (Figure 1C, fig. S1B), FoxO3 ranked third among transcription factors with highly down-regulated phosphorylation sites including Ser²⁵³, which was also confirmed by Western blot analysis (Figure 3E and 3F). These findings indicate that aberrant ALK sustains downstream phosphorylation of FoxO3 at a critical inhibitory site.

Besides refining ALK signaling based on the shared targeted nodes by TAE684, crizotinib and LDK378, our setup also allowed the study of effects for each inhibitor individually. All inhibitors affected common pathways (such as MAPK signaling), but also unique KEGG pathways (fig. S4C). Unique pathways included axon guidance, phosphatidylinositol and chemokine signaling among the down-regulated pool and regulation of actin cytoskeleton and RNA transport for the up-regulated pool (fig. S4C). Comparison of sequence motifs for down-regulated sites showed a similar overrepresentation of the SH2-binding motifs of PKA/PKB/PKC and PTPN11 for each inhibitor (39). In contrast, the sequence motifs for up-regulated sites were less biased towards known motifs, although crizotinib and TAE684 tended to upregulate motifs with a glutamine (Q) in the +1 position relative to the serine or threonine phosphorylation site, the preferred motif for the DNA damage responsive kinases ATM and ATR, suggesting increased signaling due to DNA damage (fig. S4D) (39).

ALK-driven NB cells display differential responses to signaling inhibition compared to ALK^EML4-ALK NSCLC cells

In NB, the aberrant activation of MAPK/ERK and Akt signaling pathways associates with poor outcomes and refractory disease (42, 43). These canonical RTK pathways also act downstream of oncogenic ALK. In lung cancer cells driven by the ALK fusion protein ALK^EML4-ALK, initial treatment with a MAPK inhibitor in combination with ALK inhibition has superior efficacy but does not benefit ALK-driven NBs (44, 45). Therefore, we characterized the downstream pathway dependencies in different ALK-driven NB cell lines (NB1: ALK^Amp, SH-SY5Y: ALK^F1174L, CLBGA: ALK^R1275Q), including NBL-S cells with wildtype ALK (ALK^WT) and the ALK-driven NSCLC cell line (H3122: ALK^EML4-ALK). All cell lines differentially responded to treatment with LDK378 and the ALK-targeting inhibitor lorlatinib in terms of cell viability (fig. S5A and S5B), despite downstream MAPK/ERK and Akt signaling inhibition, which served as surrogate markers for ALK inhibition in SH-SY5Y and CLBGA cells with low levels of phosphorylated ALK (fig. S5C). The ALK^WT NBL-S cells were least responsive both in terms of cell viability (fig. S5A and S5B) and downstream signaling (fig. S5C) with minimal inhibition of MAPK signaling. To examine the effect of downstream signaling inhibition, cells were treated with the MEK inhibitor U0126 and the PI3K inhibitor LY294002. Despite an inhibitory effect on protein signaling at concentrations of 10 μM and 50 μM, the effect on cell viability was minimal at these concentrations (fig. S5A-C). Moreover, the combination of LDK378 with U0126 or LY294002 resulted in minor additional reductions (5-10%) in cell viability compared to LDK378 alone (fig. S5D-E). The viability of SH-SY5Y cells was significantly reduced upon combination treatment compared to LDK378 alone whereas NB1, NBL-S and H3122 cells only showed this effect for some concentrations of LDK378. Whereas the effects of LDK378 treatment were enhanced by additional inhibition of residual MAPK and PI3K/Akt signaling, the tested combination therapies only had modest effects. Therefore, we tested additional inhibitors in combination with LDK378 to target nodes of residual signaling to explain residual cell viability despite ALK inhibition. Given the presence of Src and proline-directed kinase substrate motifs among the phosphorylation sites with increased SILAC ratios in response to ALK inhibition (Figure 3B), we used dasatinib to target Src and the inhibitor KD025 to target Rho-associated kinase 2 (Rock2). The latter was targeted because of the identification of regulation of a proline-directed site (Data File S3) and because Rock is a therapeutic target in NB (46). Dose-response cell viability assays in NB1, SH-SY5Y, CLBGA, NBL-S and H3122 revealed sensitivity to both dasatinib and KD025 with the exception of minimal effect of KD025 in H3122 cells (fig. S6A and S6B). However, the observed limited sensitivity to dasatinib was ascribed to measuring viability after only 48 hours of treatment. The combination of LDK378 and dasatinib showed a combinatorial effect in SH-SY5Y, CLBGA and NBL-S cells, whereas combinatorial effects were minimal for NB1 and H3122 cells (fig. S6C). For the combination of LDK378 and KD025, SH-SY5Y cells showed an additive effect across a range of concentrations, whereas the other cell lines only showed such a response for 1-3 of the total of 6 tested concentrations (fig. S6D).

Constitutively active ALK regulates NB cell survival through the IRS2-PI3K-Akt-FoxO3 axis

Several lines of evidence from our IPP analysis pointed towards dual regulation of IRS2 and a key role for IRS2 as a proximal signaling adaptor that linked amplified full-length ALK to PI3K signaling in NB cells (Figure 2A and 2E-2G). Furthermore, the ALK TKI-treatment markedly reduced the phosphorylation of downstream Akt and FoxO3 (Figure 1C, 3F, fig. S1B). To further establish a role for IRS2 downstream of ALK, we depleted NB1 cells for ALK using siRNA and performed a quantitative phosphoproteomics analysis using a tandem mass tag (TMT)-11-plex approach allowing us to analyze the effects of ALK-depletion and two different concentrations of lorlatinib in NB1 cells (fig. S7A-C). We analyzed lorlatinib at both low and high dose because our IC₅₀ determinations showed higher values (fig. S5A) than previously reported (19, 20), which we ascribe to our relatively shorter treatment time of 48 hours compared to 5 days. We identified and quantified 24,891 phosphorylated sites (pTyr: 430, pSer: 21,157 and pThr: 3,304) (Data File S5) and samples clustered according to treatments (Figure 4A). Lorlatinib treatment or ALK-depletion inhibited MAPK and Akt signaling, including inhibition of FoxO3 and IRS2 phosphorylation (Figure 4B-C, fig. S7D). Whereas ALK depletion by siRNA significantly downregulated phosphorylation of serines and threonines on IRS2 (Figure 4D, Data File S5), lorlatinib treatment downregulated phosphorylation of IRS2 Tyr⁵⁷⁶ and Tyr⁸²³. These IRS2 tyrosine sites were also identified upon ALK depletion (Data File S5); however the lack of significant regulation (depletion compared to control) is most likely a consequence of incomplete ALK depletion or the inherent differences in the dynamics of ALK inhibition when comparing a 48-hour siRNA depletion experiment with a 30-minute TKI treatment-based experiment. Moreover, a minimal effect on the activation loop tyrosines on IGF-1R(Tyr¹¹⁶¹ and Tyr¹¹⁶⁵)/insulin receptor (INSR) (Tyr¹¹⁸⁵ and Tyr¹¹⁸⁹) phosphorylation seemed to rule out a potential compensatory crosstalk between ALK and IGF-1R as previously shown to exist for NSCLC upon chronic ALK inhibition (47). However, ALK inhibitor-treated NB1 cells respond to insulin-like growth factor-1 (IGF-1) in terms of increased ERK phosphorylation suggesting that IGF-1R retains its signaling capacity despite ALK inhibition and rules out its off-target inhibition (fig. S7E). In contrast, a response to insulin was not detected probably due to the relative lower expression levels of this receptor (approximately 20-fold lower compared to IGF-1R) in the NB1 cells (fig. S4A and S7F).

Fig. 4. ALK inhibition by lorlatinib and ALK depletion by siRNA in NB1 cells reduces IRS2, Akt, FoxO3 and ERK phosphorylation.

(A) Cluster dendrogram of the TMT-11-plex phosphoproteomics data showing the relation between analyzed samples; 10 nM lorlatinib-treated (lorlatinib low conc.) cells, 10 μM lorlatinib-treated (lorlatinib high conc.), and ALK-depleted cells to siRNA control and DMSO-treated cells (see also figure S7C). Hierarchical clustering was performed in Perseus using quantile-based normalized intensities for identified and quantified phosphorylated peptides. (B, C) Volcano plots of - log₁₀ transformed FDR-corrected p-values versus log₂(fold change) of median phosphorylation site intensities measured for NB1 cells upon low dose lorlatinib (10 nM) (B) and siRNA depletion of ALK (C) as measured by TMT-multiplexing analysis and mass spectrometry. Fold change in (B) represents lorlatinib low dose treated NB1 cells (N=2) compared to DMSO-treated control siRNA (siCTRL) cells (N=3). Fold change in (C) represents ALK depletion by two different siRNA ALK-targeting sequences as well as their mix (N=3) compared to siCTRL cells (N=3). Statistical analysis was performed for N=2-3 independent experiments by two-sided t-test and significance was determined based on a FDR<0.05 and hyperbolic curve threshold of s0=0.1 using Perseus. (D) Overlap between significantly downregulated phosphorylated sites for the conditions comparing 10 nM lorlatinib-treated (lorlatinib low conc.) cells to DMSO-treated cells, 10 μM lorlatinib-treated cells to DMSO-treated cells, and ALK-depleted cells to siRNA control cells. See also figure S7 and Data File S5.

A search of the Cancer Cell Line Encyclopedia (CCLE) revealed that NB cell lines were in the top three cancer cell lines for high expression of IRS2, ALK and FOXO3 (48) (fig. S8). These findings prompted us to further examine the contribution of IRS2 to control PI3K-Akt-FoxO3 signaling downstream of ALK in NB1 cells as well as SH-SY5Y (ALK^F1174L) and CLBGA (ALK^R1275Q). Although the point-mutated cell lines had low ALK abundance, they displayed sensitivity to ALK inhibition in terms of cell viability and downstream signaling including inhibition of FoxO3 phosphorylation (fig. S5A-C). Moreover, depletion of IRS2 in these three NB cell lines was efficient and specific and resulted in the concomitant reduction of phosphorylation of Akt and FoxO3, but minimal effect on phosphorylation of ERK (Figure 5A and 5B, fig. S9A). Furthermore, we confirmed these findings in ALK^WT NBL-S cells, suggesting that IRS2 served as an important link to PI3K-Akt-FoxO3 signaling in NB in general (fig. S9B). Whereas IRS2 has not previously been shown to associate with aberrant ALK in NB cells, the closely related family member IRS1, interacts with the NPM-ALK fusion in anaplastic large-cell lymphoma cells and upon overexpression, with full-length ALK in murine 32D murine myeloid cells (49, 50). The preference for IRS2 reported in our study may be cell-context dependent and explained by a differential expression pattern. Indeed, we found IRS2 to be 100-fold and 40-fold more abundant based on iBAQ values compared to IRS1 in NB1and SH-SY5Y cells, respectively, which likely explains the apparent preference of ALK for IRS2 (fig. S9C and S9D) (51). Because of the link of IRS2 to PI3K-Akt-FoxO3 and because Akt and FoxO3 have important roles in cell survival and apoptosis (52), we examined the effect of IRS2 depletion on cell viability and measured the activity of caspases, which serve as important mediators of apoptosis. For all three ALK-driven cell lines, IRS2 depletion significantly reduced cell viability over a 48 h-time period compared to control cells (Figure 5C). Furthermore, IRS2-depleted cells showed increased caspase-3/7 activity (Figure 5D) and increased cleavage of caspase-3 was confirmed in NB1 and SH-SY5Y cells (and not detected in CLBGA and NBL-S), suggesting that IRS2 protects ALK-driven NB cells from apoptosis (fig. S9E).

Fig. 5. ALK regulates NB cell survival through the IRS2-FoxO3 axis.

(A) Lysates from NB1, SH-SY5Y and CLBGA cells depleted for IRS2 using siRNA and immunoblotted as indicated (A) and quantified in (B). Blots are representative of N=3-4 independent experiments. (C, D) Cell viability (C) and caspase activity normalized to cell viability (D) upon siRNA-mediated depletion of IRS2. (E) Immunoblots of cleaved caspase 3 and quantification for NB1 and SH-SY5Y cells upon IRS2 depletion. Blots are representative of N=3-4 independent experiments. Data is represented relative to siRNA control (C or siCTRL) for each cell line and values are mean ± SD of N=3-4 (A, B, E) or N=4-5 (C, D) independent experiments. ^*P <0.05, ^**P <0.01, ^***P <0.001 compared with siRNA control (One sample t test on log-transformed fold changes relative to siRNA control). See also Figure S6.

Discussion

In this study, we applied an IPP approach to characterize aberrant ALK signaling in NB cells on a systems-wide scale. Combining a pharmacological approach using three clinically relevant ALK TKIs with quantitative proteomics refined our understanding of ALK proximal and downstream signaling, expanded our knowledge on phosphotyrosine signaling and led us to identify IRS2 as determinant of NB cell survival through the PI3K-Akt-FoxO3 axis (Figure 6). The data represents a resource for aberrant ALK signaling in NB. It provides insights into the druggable ALK signaling network by revealing TKI-sensitive nodes and expanding our knowledge on how ALK-driven NB growth is supported by a complex network of adaptors to the activated receptor, phosphorylation of downstream proteins and protein abundance.

Fig. 6. The constitutive ALK signaling network in NB1 cells.

Model summarizing main findings from our integrated proteomics approach to unravel oncogenic ALK signaling in NB1 cells linking IRS2 to survival signaling (highlighted in dark orange). ALK interactors belonging to the GO terms “transmembrane RTK signaling pathway” and “insulin receptor signaling pathway” (Figure 2D) are represented (light orange) together with proteins belonging to the KEGG pathways “ErbB signaling”, “neurotrophin signaling”, and “insulin signaling” from the downregulated phosphoproteome (Figure 3D). Transcription factors are represented based on their relation to the KEGG pathways or their phosphorylation regulation (Figure 3E; >2 regulated phosphorylation sites). ALK TKI-sensitive nodes are highlighted by red boxes. Arrows indicate activation, T-bars indicate inhibition, and dotted arrows indicate translocation.

We were intrigued to find that several components of the INSR signaling network were shared with ALK (Figure 2D and 3D). At first, these findings may not seem surprising because ALK belongs to the INSR superfamily and has an intracellular kinase domain homologous to that of INSR (53). However, because INSR and its closely related family member IGF-1R play crucial roles in breast, prostate and thyroid cancers (54), our understanding of INSR and IGF-1R signaling may provide further insight into ALK signaling. Molecular details about the ALK-IRS2 interaction and the contribution of SH2B proteins remain to be elucidated. SH2B proteins promote insulin signaling by both enhancing INSR catalytic activity and inhibiting tyrosine dephosphorylation of IRS proteins; thus, it is intriguing to speculate that aberrant ALK sustains constitutive signaling through a similar mechanism (55, 56). SH2B proteins interact specifically with the phosphorylated tyrosine sites in the activation loop of the INSR and link through IRS proteins to PI3K signaling (57). However, further studies are required to establish the molecular basis of these interactions in the context of aberrant ALK in NB cells.

IRS2 is an adaptor protein that has been well characterized in the context of the regulation of cellular glucose metabolism by INSR and IGF-IR signaling (58). In this study, we described a specific role for IRS2 in ALK-driven survival signaling and attributed this role to a particular cell context and the greater abundance of IRS2 compared to IRS1 in NB cells (fig. S9C and S9D). However, despite a high level of homology and shared functions between IRS1 and IRS2, knockout and RNAi studies show that IRS proteins also have distinct biological functions. For example, IRS1 knockout mice display a generic growth defect, whereas IRS2 knockout mice show defective growth in only a few tissues such as the brain and pancreas (59, 60). Moreover, IRS2 promotes proliferation of neuronal precursors (59). Increasing evidence underscore the need to understand how tumor cell metabolism is regulated because the Warburg effect (61) enables cancer cells to fuel proliferation, survival, and invasion, by undergoing metabolic reprogramming to specifically exploit aerobic glycolysis. In this context, non-redundant roles for IRS1 and IRS2 have emerged because although they can both activate PI3K, only IRS2 promotes aerobic glycolysis in mammary tumor cells (62). Moreover, IRS2 but not IRS1 can protect NB cells from apoptosis caused by high glucose levels (63). In our study, we identified in the ALK interactome a network of glycolytic enzymes including phosphoglycerate mutase 1 (PGAM1), fructose-bisphosphate aldolase A (ALDOA), glucose-6-phosphate isomerase (GPI) and triosephosphate isomerase (TPI1) (64). Moreover, we also identified an ALK TKI-regulated phosphotyrosine site on TBC1D4, a Rab GTPase-activating protein that controls GLUT4 trafficking and thus glucose uptake in multiple cell types (65). Together, these findings support a role for IRS2 in specifically regulating glucose metabolism and it is possible that IRS2 plays a more substantial role at the intersection of ALK-driven cancer progression and metabolism than previously anticipated.

While many of the ALK inhibitors used in this study are in clinical trials for ALK-driven NBs, the parallel preclinical studies demonstrate their efficacy (8, 19, 21). However, long-term clinical use of these compounds is anticipated to drive emergence of therapy resistance. Moreover, intrinsic resistance may challenge and limit clinical success given that only 1 of 11 crizotinib-treated NB patients with ALK mutations showed complete response (23). Our study addresses the immediate adaptive response to ALK inhibition after short-term treatment. Despite our attempts to use the findings of this study to discover new effective combinatorial treatments, the results were disappointing. Inhibiting ALK in combination with MAPK, PI3K, Src or Rock2 did not add major reductions to cell viability in the ALK-driven NB cell lines that we tested (fig. S5, D and E and S6, C and D). Whereas these attempts served to target additional downstream signaling, other RTKs may sustain bypass signaling to drive growth and proliferation. Given the modest effects on viability observed when targeting MAPK and PI3K signaling alone, we reason that simultaneous inhibition of both signaling arms are required as shown effective in RAS-driven lung cancer (66). Whereas the MAPK and PI3K pathways are canonical to many RTKs, a shift in dependency of from one RTK to another would allow a compensatory loop to re-activate the very same cascade and thus, underscore plasticity within the signaling network rewiring. Related to this issue, we showed that ALK-inhibited NB1 cells retained the ability to boost downstream MAPK signaling in response to IGF-1. Therefore, a strategy co-targeting other RTKs (for example, IGF-1R in combination with ALK inhibition) should be tested, in addition because combined IGF-1R- and ALK-targeted therapy has proven effective for NSCLC (47). Knowing whether co-targeting RTK bypass signaling can avoid or postpone development of therapy resistance may ultimately guide the development of strategies for NB patients with poor outcomes on ALK-targeted therapy. Nonetheless, given the resource-richness of the IPP data provided here, additional avenues for hypothesis-generation and exploring combinatorial treatments in more depth may emerge. Moreover, it is important to consider whether more advanced experimental models are required to better represent the underlying molecular circuits to identify novel and effective combination therapies to overcome resistance. For instance, therapy resistant NB cell lines developed to grow despite long-term ALK inhibition could serve this rationale.

In NB cells from high-grade tumors, the physiological response to FoxO3 activation (such as FoxO3-induced cell death) is impaired and modulated by wildtype p53 which interferes with FoxO3-promoter recognition (67). Along these lines, we observed limited caspase-mediated responses in NB1 cells upon IRS2-Akt-FoxO3 pathway inhibition compared to the point-mutated SH-SY5Y and CLBGA cells. Although all cell lines are reported to have wildtype p53, additional factors may influence and modulate the functional survival response in the NB1 cells (68, 69). For instance, p53 is a direct transcriptional target of MYCN in NB cells (70), which may ultimately interfere with the actions of activated FoxO3, and thus, cause partial resistance to FoxO3-induced cell death. Ultimately, a more complex transcription factor rewiring in NB1 cells may explain the observed functional heterogeneity in terms of survival signaling. Moreover, whether ALK point-mutated cells rely more on IRS2 for oncogenic survival signaling downstream of other RTKs warrants further investigation. However, data reported by Chen et al. support regulation by ALK given the reduced phosphorylation of IRS2 Tyr⁶⁷⁵ upon crizotinib treatment in the point-mutated SH-SY5Y (ALK^F1174L) and NB1643 (ALK^R1275Q) cell lines (31). Nevertheless, proximal signaling may be critically different between ALK amplified compared to ALK point-mutated NB cell lines. How and if this differential proximal dependency can be exploited therapeutically to impact clinical efficacy and resistance to ALK targeted therapy remain to be determined. In conclusion, the present study shows how a focused IPP approach can broaden our understanding of proximal mediators of oncogenic ALK signaling. Using this integrative proteomics approach, we identified a key role for IRS2 in PI3K-Akt-FoxO3 survival signaling in NB cells.

Materials and Methods

Reagents

TAE684, crizotinib (PF-02341066), LDK378 (ceritinib), lorlatinib, dasatinib and KD025 were purchased from Selleck Chemicals (Selleckchem). LY294002 and U0126 were purchased from Cell Signaling Technology. The stock solutions were prepared in DMSO and stored at −20°C. IGF-1 were purchased from Peprotech and insulin from Sigma-Aldrich. The following antibodies were used: rabbit anti-phospho-ALK (Tyr¹⁶⁰⁴), mouse anti-ALK (31F12), rabbit anti-ALK (C26G7), rabbit anti-phospho-Akt (Ser⁴⁷³) and anti-Akt, mouse anti-phospho-ERK1/2 (Thr²⁰²/Tyr²⁰⁴), rabbit anti-ERK1/2, rabbit anti-phospho-FoxO3a (Ser²⁵³) (D18H8) and anti-FoxO3a (D19A7), rabbit anti-PI3 Kinase p110α and anti-PI3 Kinase p85, rabbit anti-IRS1, rabbit anti-IRS2 (L1326), mouse anti-phosphotyrosine (P-Tyr-100), rabbit anti-cleaved caspase-3, rabbit anti-phospho-IGF-I Receptor β (Tyr¹¹³¹)/Insulin Receptor β (Tyr¹¹⁴⁶), rabbit anti-Insulin Receptor β (4B8), rabbit anti-IGF-I Receptor β (D23H3) (Cell Signaling Technology); rabbit anti-IRS2 for immunoprecipitation, mouse anti-GAPDH, anti-SHP2 [M163] and anti-Shc (Abcam); mouse anti-Grb2 (BD Biosciences, San Jose, CA); mouse anti-vinculin (Sigma-Aldrich); mouse anti-phosphotyrosine (4G10) (Millipore, Bedford, MA); goat-anti-rabbit HRP-conjugated secondary antibody, goat-anti-mouse HRP-conjugated secondary antibody (Jackson ImmunoResearch Laboratories). For phosphoproteomics: anti-phosphotyrosine immunoaffinity beads (P-Tyr-100 and P-Tyr-1000) (Cell Signaling Technology).

Cell culture and SILAC labeling

The human neuroblastoma cell lines NB1, SH-SY5Y, CLBGA and NBL-S and the human lung cancer cell line H3122 (kindly provided by Stefan Mueller, Evotec, Munich, Germany) were cultured in RPMI 1640 including 2 mM L-glutamine (Gibco), supplemented with 10% fetal bovine serum (Gibco) and penicillin-streptomycin (100 units/mL and 100 µg/mL; Gibco). Cell lines were maintained at 37°C in a humidified atmosphere at 5% CO₂. For SILAC-based quantitative mass spectrometry, NB1 cells were labeled in SILAC RPMI (PAA Laboratories) supplemented with 10% dialyzed fetal bovine serum (Sigma-Aldrich), 2 mM L-glutamine (Gibco), penicillin-streptomycin (100 units/mL and 100 µg/mL; Gibco) for at least 2 weeks to ensure complete incorporation of amino acids. Three cell populations were obtained: one labeled with natural variants of the amino acids (light label; Lys0, Arg0) (Sigma-Aldrich), the second labeled with medium variants of amino acids (L-[2H4]Lys (+4) and L-[13C6]Arg (+6)) (Lys4, Arg6) and the third labeled with heavy variants of the amino acids (L-[13C6,15N2]Lys (+8) and L-[13C6,15N4]Arg (+10)) (Lys8, Arg10). Medium and heavy variants of amino acids were purchased from Cambridge Isotope Labs.

Cell viability assay

Cells were seeded in 96-well microplates one day before the start of treatment. At the onset of the experiment, growth medium containing inhibitors was added to the cells with final concentrations as indicated in the figures. Control cells were treated with similar amount of vehicle as the treated cultures (0.1 % DMSO). Cell viability was measured after 48 hours of treatment using the ATPlite 1step Luminescence Assay System (PerkinElmer Life Sciences) or CellTiter-Glo Luminescent Cell Viability Assay (Promega) according to the manufacturer’s instructions. Luminescence was measured using an EnSpire 2300 Multilabel Reader (PerkinElmer Life Sciences) or a Tecan infinite M200 Pro multimode microplate reader. Three independent biological experiments, each prepared in triplicate or quadruplicate for each concentration, were performed with reproducible results. An IC₅₀ value for each inhibitor was determined corresponding to the concentration giving 50% reduction in cell viability. The IC₅₀ values were derived from a five-parameter non-linear curve fitting using GraphPad Prism software and the experimentally used IC₅₀ concentrations were within the 95% confidence interval. The combination effect of LDK378 and dasatinib or KD025 (figure S7C and D) was determined using the Bliss independence model to calculate interaction scores (ΔI) as previously described (71), and for positive ΔI’s, any two-drug combinations were considered synergistic.

Cell stimulation, lysis and Western blotting

NB1, SH-SY5Y, CLBGA, NBL-S and H3122 cells were cultured in complete medium and treated for the indicated time points with inhibitor (TAE684, LDK378, crizotinib, lorlatinib, LY294002 or U0126) as indicated with their respective the IC₅₀ values reported. Control cells were treated with similar amount of DMSO as the treated cultures (0.1 % DMSO). IGF-1 and insulin stimulations were performed using 10 ng/mL for 10 minutes after ALK inhibitor pre-treatment. Cell lysates, protein concentration determination and Western blotting was carried out as previously described (29).

Immunoprecipitation

Cell lysates for immunoprecipitation were prepared as described previously (29). Cell lysates (1-2 mg) were incubated overnight at 4°C with anti-ALK antibody (Cell Signaling Technology) or anti-IRS2 antibody (Abcam) with subsequent binding to Protein G-Sepharose for 1 h. After five washes with lysis buffer, the bound proteins were eluted by boiling in SDS sample buffer and resolved by SDS–PAGE and analyzed by Western blotting.

RNA interference and transfection

NB1, SH-SY5Y, CLBGA and NBL-S cells were transfected using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer’s instructions, and all the assays were performed 48 to 72 hours after transfection. Double-stranded siRNA oligonucleotides targeting human IRS2 (sequence #1: 5′-UCAUCCCACCCUGUUUCCCUGAAUU-3′; sequence #2: 5′-GGUCCAUCUUCAGAGAAGAAAUCAA-3′; sequence #3: 5′-CAUGGGAGAGGGUUAUGCCUUUGAA-3′, sequence #4: 5′-CCAGCAGAUUGAUAGCUGUACGUAU-3′) and human ALK (sequence #1: 5′-GACAAGAUCCUGCAGAAUA-3′; sequence #2: 5′-GGAAGAGUCUGGCAGUUGA-3′; sequence #3: 5′-GCACGUGGCUCGGGACAUU-3′, sequence #4: 5′-GGUCAUAGCUCCUUGGAAU-3′) were purchased from GE Dharmacon. Cells were transfected either individually (80 nM) or with a mixture of all four IRS2-targeting siRNA duplexes (20 nM each siRNA). As control, siRNA Control duplex (siGenome Non-Targeting siRNA Pool #2) (GE Dharmacon) was used at a final concentration of 80 nM. Silencing of gene expression was monitored by Western blotting of cell lysates with an antibody against IRS2.

Caspase 3/7 activity measurements

Transfected cells were seeded in 96-well microplates. Activities of caspase-3 and -7 together with cell viability were measured 24 and 72 hours post-transfection using the Caspase-Glo 3/7® assay kit (Promega) and ATPlite 1step Luminescence Assay System (PerkinElmer Life Sciences) according to the manufacturer’s instructions. Luminescence was measured using an EnSpire 2300 Multilabel Reader (PerkinElmer Life Sciences). 3-4 independent biological experiments, each prepared in triplicate or quadruplicate for each transfection condition, were performed with reproducible results.

Sample preparation for MS-based interactome analysis

For each experiment cells from three SILAC conditions (fig. S1C) were lysed in immunoprecipitation lysis buffer at 4°C. ALK was immunoprecipitated from 8 mg of lysate in parallel for each SILAC condition before immunoprecipitated eluates were combined prior to SDS-PAGE processed for by Coomassie staining and in-gel digestion essentially as described in (29).

Sample preparation for MS-based phosphotyrosine interactome analysis

Peptide pull-downs were performed on NB1 cell lysate (prepared using lysis buffer containing 50 mM Tris pH 8.5, 150 mM, NaCl, 10 mM potassium chloride, 0.1% Triton X-100, 0.5 mM DTT with the addition of 5 mM β-glycerolphosphate, 5 mM sodium fluoride, 1 mM sodium orthovanadate and 1 complete® inhibitor cocktail tablet per 10 mL (Roche). from DMSO- and LDK378-treated cells using biotinylated peptides; phosphotyrosine-containing peptides the corresponding non-phopshorylated version (fig. S1C). Peptides included the following phosphotyrosines (+/- 6 amino acids flanking the indicated phosphotyrosine residues) from ALK; Tyr¹⁰⁷⁸, Tyr¹⁰⁹², Tyr¹⁰⁹⁶, Tyr¹¹³¹, Tyr¹²⁷⁸, Tyr¹²⁸³, Tyr¹²⁸³, Tyr¹³⁵⁹, Tyr¹⁵⁰⁷, Tyr¹⁵⁸⁴, and Tyr¹⁶⁰⁴, and from IRS2; Tyr⁶⁷⁵, Tyr⁷⁴², and Tyr⁹⁷⁸ including the non-phosphorylated counter peptide synthesized by Intavis and coupled to sepharose-streptavidin beads (GE Healthcare) in a buffer containing 50 mM Tris pH 8.5, 150 mM, NaCl and 0.1% NP-40. Pull-downs (1 mg cell lysate per pull-down), elution and on-bead trypsin digestion were carried out in 96-well multiscreen filter plates essentially as described in Eberl et al. (72).

Sample preparation for MS-based phosphoproteome analysis

For each experiment cells from three SILAC conditions (fig. S1C) were lysed in immunoprecipitation lysis buffer at 4°C. Proteins were precipitated overnight at −20°C in 4-fold excess of ice cold acetone. The acetone-precipitated proteins were solubilized in denaturation buffer (10 mM HEPES, pH 8.0, 6 M urea, 2 M thiourea), and 5 mg of protein from each SILAC condition were mixed a 1:1:1 ratio (total 15 mg per SILAC mix), and reduced with 1 mM DTT followed by alkylation with 5.5 mM CAA. Proteins were subjected to Lys-C digestion for 3 hours (Wako) and then, after a four-fold dilution using 50 mM Tris, pH 8.0, digested with trypsin (Sigma-Aldrich) overnight at room temperature. The peptide mixtures were desalted and concentrated on a C18-SepPak cartridge (Waters) and eluted with 50% ACN. Eluted peptide mixtures were dried almost to completeness in a SpeedVac and dissolved in MOPS buffer (50 mM MOPS, pH7.2, 10 mM sodium phosphate, 50 mM NaCl) and subjected to anti-phosphotyrosine IP using a mixture of anti-phosphotyrosine antibodies (P-Tyr-100 and P-Tyr-1000, PTMScan Kit, Cell Signaling Technology). Immunoaffinity beads were washed with increasing salt concentration (50 mM NaCl, 150 mM NaCl) and phosphopeptides were eluted with 0.1% TFA before loading onto a C18 Stage tip. The unbound fraction from the anti-phosphotyrosine IP was acidified with TFA, desalted using a C18 SepPak (Waters) and eluted with 50% ACN, 0.1% TFA. Peptide mixtures were adjusted to 80% ACN, 6% TFA and phosphopeptides were further enriched by two sequential rounds of titansphere chromatography as previously described (73).

TMT labeling and phosphopeptide enrichment

Cells were either transfected with ALK-targeting siRNA, control siRNA or treated with two different concentrations of the inhibitor lorlatinib and DMSO (figure S7C). Cells were washed in PBS 48 hours post-transfection or 30 minutes post-inhibitor treatment and lysed 10 min at 99°C in 6 M guanidine-HCl, 100 mM Tris pH 8.5, 5 mM TCEP and 10 mM CAA and whole cell extracts were sonicated. Cell lysates were digested by Lys-C (Wako) in an enzyme/protein ratio of 1:100 (w/w) for 1 hour, followed by a dilution with 25 mM Tris buffer pH 8.5, to 2 M guanidine-HCl and further digested overnight with trypsin (Sigma-Aldrich) 1:100 (w/w). Protease activity was quenched by acidification with TFA and the resulting peptide mixture was concentrated on C18 Sep-Pak (Waters). Peptides were eluted with 40% ACN, followed by 60% ACN. The combined eluate was reduced by speedvac, and the final peptide concentration was estimated by measuring absorbance at A280 on a NanoDrop (Thermo Scientific). 300 μg peptide from each sample was labeled with one of eleven different TMT-reagents according to the manufacturer’s protocol (ThermoScientific). After labeling, the samples were mixed and adjusted to 80% ACN, 6% TFA and phosphopeptides were further enriched by two sequential rounds of titansphere chromatography as previously described (51). The eluted phosphopeptides were concentrated in a SpeedVac and fractionated with high-pH-reversed-phase fractionation as described (74).

Sample preparation for MS-based proteome analysis

Cells were washed in PBS and lysed 10 min at 99°C in 6 M guanidine-HCl, 100 mM Tris pH 8.5, 5 mM TCEP and 10 mM CAA and whole cell extracts were sonicated. Cell lysates were digested by Lys-C (Wako) and trypsin (Sigma-Aldrich) and resulting peptides were processed by high-pH fractionation (14 fractions per biological replicate; two replicates in total) as described in Batth et al. (75).

Liquid chromatography-tandem mass spectrometry (LC-MS/MS)

Peptides from all samples were eluted from C18-stagetips using 40% ACN, 0.5% acetic acid. Peptides were analyzed using online nanoflow LC-MS/MS on a Q Exactive Plus (interactome and phosphoproteome), a Q Exactive HF (phosphotyrosine interactome and proteome), or a Q Exactive HF-X (TMT-11-plex phosphoproteomics) mass spectrometer (Thermo Fisher Scientific), which was interfaced with an EASY-nLC system (Proxeon, Odense, Denmark) equipped with a nanoelectrospray ion source essentially as described (76).

MS data analysis

Raw MS files were analyzed by MaxQuant software version 1.5.3.33 or 1.6.0.17 (TMT-11-plex phosphoproteomics) using the Andromeda search engine (77, 78) by which the precursor MS signal intensities were determined and SILAC triplets were automatically quantified. Proteins were identified by searching the HCD-MS/MS peak lists against a target/decoy version of the human Uniprot protein database (release 2015_03 or release April 2017 for TMT-11-plex phosphoproteomics) using default settings. TMT correction factors were edited for the TMT labels and the reporter ion mass accuracy was set to 0.002 Da. Carbamidomethylation of cysteine was specified as fixed modification and protein N-terminal acetylation, oxidation of methionine, pyro-glutamate formation from glutamine and phosphorylation of serine, threonine and tyrosine residues were considered as variable modifications. The “maximum peptide mass” was set to 7500 Da, the “modified peptide minimum score” and “modified maximum peptide score” was set to 25. Everything else was set to default values.

Bioinformatic analysis

For interactome data, ratios of identified and quantified interactors were normalized to the ratio of ALK to account for uneven efficiency during individual IPs performed in parallel. Significantly changing druggable ALK interactors were determined by significance B testing (P < 0.05) using Perseus version 1.3.9.10 (Data File S1) (77).

For the phosphotyrosine interactome data, a minimum of 3 razor and unique peptides were required across eight conditions, and label-free quantitation (LFQ) intensities (79) were required for one of four phosphotyrosine-peptide conditions (DMSO; N=2 independent biological replicates, LDK378; N=2 independent biological replicates) with no restrictions for non-phopshopeptide (Data File S2). Quantitative interaction proteomics analysis was performed by t-test based comparison of protein intensities between each phosphotyrosine-containing peptide (bait) and the non-phosphorylated counterpart peptide (control) using the web tool, pulldown.jensenlab.org. The NB1 cell line proteome was used to correct for protein abundance in the pulldown analysis. Data was analyzed with ratio cutoff of 2.0 (log2), P-value cutoff of 3.0 (-log10) and infinity P-value cutoff of 3.125 (-log10). Significance was concluded whenever s-score > 1.

For the phosphoproteomics data, only peptides with a phosphorylation site localization probability of at least 0.75 (class I, Data File S3) were included in the bioinformatics analyses. To identify phosphorylation sites with significantly regulated ratios we compared to the ratio distributions of all quantified phosphopeptides with all non-phosphorylated peptides, which we expect not to change and therefore specify our technical variance. To determine the level of regulation cut-offs for up- and downregulation were set to allow for an estimated 5% false positive rate based on the distribution of ratios of identified and quantified non-phosphorylated peptides as described in (29). Regulated interactors and phosphorylation sites were considered common and representative of ALK signaling whenever deemed regulated by two out of three inhibitors. Proteome data was filtered for common contaminants and protein quantifications were reported as a median of two replicates by iBAQ intensities (table S4) (37). Analysis of GO term enrichment related to biological process (interactome) and KEGG pathway annotation enrichment (phosphoproteome) was performed using the DAVID bioinformatics resource (80). For the GO analysis, gene sets derived from the pool of downregulated by inhibitors (2 out of 3) was used. For the KEGG analysis, gene sets derived from each pool of regulated phosphorylation sites (up- and downregulated) for each inhibitor as well as the commonly regulated interactors (2 out of 3 inhibitors) was used. Significance was concluded when P < 0.05 by Fisher’s exact test. The protein association network based on ALK interactome data was obtained using the STRING database (version 10) (81). All active interaction sources were included in the network and a medium confidence score over 0.4 was required. To asses for sequence bias around the regulated phosphorylation sites, sequence motif logo plots (+/- 6 amino acids adjacent to the identified phosphorylated sites) were generated and visualized using the IceLogo software (82) with default parameters (P < 0.01). The analysis was performed independently for the group of phosphorylation sites with up- and downregulated SILAC ratios, and compared with non-regulated site sequences, which was used as a common background. The non-regulated phosphorylation sites were defined as sites with ratios within less than one standard deviation away from the mean of the distribution of identified non-phosphorylated peptides. Linear sequence motifs for kinase substrates were annotated using Perseus version 1.3.9.10 and analyzed for overrepresentation among the upregulated phosphopeptides compared to the unregulated phosphopeptides using a Fishers’s exact test. Motifs with P < 0.05 considered significant.

For the TMT-11-plex phosphoproteome all measured peptide intensities were normalized using the “normalizeQuantiles” function from the Bioconductor R package LIMMA (83). Subsequent data analysis was performed using Perseus version 1.5.1.12. The quantile normalized ratios were further normalized by median subtraction in the rows and the data were filtered for contaminants and reverse hits. Only peptides with a phosphorylation site localization probability of at least 0.75 (class 1, Data File S5) were included in the bioinformatic analyses. Volcano plots were generated by plotting the –log₁₀ transformed and FDR-adjusted p-values (q-value threshold of 0.05) derived from a two-sided t-test versus log₂ transformed fold changes. Significance was determined based on a hyperbolic curve threshold of s0=0.1using Perseus.

Statistical analysis

Statistical analysis of MS data is described in the Bioinformatic analysis section. For experiments with effects measured as fold changes relative to a control within each experiment, fold changes were log₂-transformed and significance determined based on a one-sample t-test asking if the mean (of minimum 3 independent experiments) was different from 0 (Figure 5B-E, figure S7B). Significance testing for data in figure S5C-D was performed by a two-sample t-test with a Sidak-Bonferroni correction for multiple testing. A statistically significant difference was concluded when P < 0.05.

Data and materials availability

The raw MS data and associated tables have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository with the dataset identifiers PXD006404 and PXD009477. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.