SRC inhibition enables formation of a growth suppressive MAGI1-PP2A complex in isocitrate dehydrogenase-mutant cholangiocarcinoma

Abstract

Intrahepatic cholangiocarcinoma (ICC) is an aggressive bile duct malignancy that frequently exhibits isocitrate dehydrogenase (IDH1/IDH2) mutations. Mutant IDH (IDHm) ICC is dependent on SRC kinase for growth and survival and is hypersensitive to inhibition by dasatinib, but the molecular mechanism underlying this sensitivity is unclear. We found that dasatinib reduced p70 S6 kinase (S6K) and ribosomal protein S6 (S6), leading to substantial reductions in cell size and de novo protein synthesis. Using an unbiased phosphoproteomic screen, we identified membrane-associated guanylate kinase, WW, and PDZ domain containing 1 (MAGI1) as an SRC substrate in IDHm ICC. Biochemical and functional assays further showed that SRC inhibits a latent tumor-suppressing function of the MAGI1–protein phosphatase 2A (PP2A) complex to activate S6K/S6 signaling in IDHm ICC. Inhibiting SRC led to activation and increased access of PP2A to dephosphorylate S6K, resulting in cell death. Evidence from patient tissue and cell line models revealed that both intrinsic and extrinsic resistance to dasatinib is due to increased phospho-S6 (pS6). To block pS6, we paired dasatinib with the S6K/AKT inhibitor M2698, which led to a marked reduction in pS6 in IDHm ICC cell lines and patient-derived organoids in vitro and substantial growth inhibition in ICC patient-derived xenografts in vivo. Together, these results elucidated the mechanism of action of dasatinib in IDHm ICC, revealed a signaling complex regulating S6K phosphorylation independent of mTOR, suggested markers for dasatinib sensitivity, and described a combination therapy for IDHm ICC that may be actionable in the clinic.

INTRODUCTION

Intrahepatic cholangiocarcinoma (ICC) is an aggressive cancer arising from the biliary tree and has been increasing in incidence worldwide for the past 4 decades (1). Early stages of ICC are often asymptomatic, and most cases present without identifiable risk factors at a late stage of the disease. For unresectable ICC, combination chemotherapy with gemcitabine and cisplatin has been the first-line therapy, with folinic acid, fluorouracil, and oxaliplatin (FOLFOX) as a second line (2). More recently, the US Food and Drug Administration has approved durvalumab in combination with gemcitabine and cisplatin due to increased survival in the phase 3 TOPAZ-1 trial (3). Despite substantial advances in our understanding of the etiology, diagnosis, and treatment of cholangiocarcinoma over the last decade, the 5-year survival rate for ICC remains at 9%, with an overall median survival of 11.7 months and median progression-free survival of 8.0 months (2). There are no widely used treatment options available for patients with ICC that progress on the current standard of care.

Recent genetic screens have identified actionable genetic mutations in ICC, many of which are associated with epigenetic processes (4, 5). Of particular interest are isocitrate dehydrogenase 1/2 (IDH1/2) mutations, which are present in ~18 to 37% of ICC cases in North America and Europe (6). IDH is a metabolic enzyme present in the cytoplasm and mitochondria that reduces isocitrate to a-ketoglutarate (aKG). Hotspot mutations in the isocitrate binding domain of IDH are known to promote cancers in many cell lineages through gain-of-function enzymatic activity, which produces the oncometabolite R (−)–2-hydroxyglutarate (2-HG) (7). 2-HG inhibits aKG-dependent dioxygenases, which are often epigenetic modifiers, and contributes to large-scale changes in the genomic landscape, ultimately resulting in impaired cholangiocyte differentiation (7-9). IDH mutations are common in other cancers, including acute myeloid leukemia (AML) and glioma, and high-potency compounds targeting IDH mutations have been approved for treatment of patients with AML (5, 6, 9). Ivosidenib (AG-120), which blocks the function of mutant IDH1/2 at nanomolar concentrations, was used in a recent randomized phase 3 trial to prolong progression-free survival in patients with IDH-mutant (IDHm) ICC (10). Although encouraging, the objective response rate (with partial response) was only 2% (11), highlighting the need for additional treatment strategies. Given the large-scale biological changes that occur because of IDH mutations, a more detailed understanding of additional cellular dependencies caused by the presence of 2-HG could provide therapeutic alternatives.

A high-throughput drug screen of a large panel of cancer cell lines, including 17 biliary tract cancers, found that IDHm ICC cells demonstrate a marked response to the multikinase inhibitor dasatinib (12). Dasatinib, a multikinase, SRC, and BCR-ABL inhibitor, induced apoptosis in IDHm ICC cell lines but not in IDH WT cell lines. Further, dasatinib induced necrosis in an IDHm patient-derived xenograft (PDX) model and was also effective in reducing tumor burden in a genetically modified mouse model harboring an IDH2 mutation (12). The direct target of dasatinib in IDHm ICC was confirmed to be SRC (13). SRC is a nonreceptor tyrosine kinase that drives aggressiveness and poor prognosis in a number of cancers. SRC activity is generally modulated by phosphorylation events at Y416 and Y527, which are activating and inhibitory, respectively (chicken numbering is used for consistency; Y419 and Y530 in humans) (14-16). SRC is a well-described proto-oncogene and is involved in multiple signaling cascades controlling differentiation, angiogenesis, motility, and proliferation. Introduction of the SRC T341I “gatekeeper” mutation prevents dasatinib from binding to the SRC drug pocket and rescues dasatinib-induced cytotoxicity in IDHm ICC cells (12). Together, these data revealed a SRC dependency in IDHm ICC, but very little is known about the mechanism of this dependency and the survival signaling involved.

In this study, we identified a previously unknown SRC-regulated survival mechanism in IDHm ICC cells. Inhibition of SRC in IDHm ICC cells by dasatinib treatment inhibited p70 S6 kinase (S6K) and ribosomal protein S6 (S6), members of the mammalian target of rapamycin complex 1 (mTORC1) pathway, and led to reductions in cell size and de novo protein synthesis. There were no effects on other upstream (AKT and mTOR) components of the mTOR pathway. Using an unbiased phosphoproteomic screen, we identified a scaffolding molecule, membrane-associated guanylate kinase, WW, and PDZ domain containing 1 (MAGI1), as a SRC substrate in IDHm ICC. Upon SRC inhibition, MAGI1 formed a tumor-suppressive complex with protein phosphatase 2A (PP2A), facilitating S6K dephosphorylation and suppression of S6K/S6 signaling. We also found that inhibition of mutant IDH could partially rescue S6K/S6 signaling and dasatinib-induced cytotoxicity. Together, this study characterized the molecular mechanism underlying dasatinib hypersensitivity and revealed a suppressive MAGI1-PP2A signaling complex that functions to inactivate S6K/S6 in IDHm ICC.

RESULTS

Dasatinib inhibits pS6K and protein synthesis independent of AKT/mTORC1 in IDHm ICC

IDHm ICC cells are uniquely sensitive to dasatinib (Fig. 1, A and B). We examined the change in the activity of canonical SRC downstream survival pathways in response to dasatinib in three IDH wild-type (WT) (HuCCT1, CCLP1, and ICC2) and three IDHm (SNU-1079, RBE, and ICC5) cell lines. At 6 hours of dasatinib treatment, extracellular signal-regulated kinase (ERK) and signal transducer and activator of transcription 3 (STAT3) were unaffected; however, phosphorylation of S6K and its downstream target S6 was ablated in all three IDHm lines tested (Fig. 1C). In contrast, dasatinib did not affect any of the pathways evaluated in the three IDH WT ICC lines. Inhibition of pS6 was confirmed by immunofluorescence staining for pS6 and total S6 (fig. S1) Dasatinib treatment showed no effect on AKT or mTOR abundance (Fig. 1C) or on 90-kDa ribosomal S6 kinase and S6K2 phosphorylation (fig. S2).

Fig. 1. Dasatinib inhibits pS6K and de novo protein synthesis independent of AKT/mTORC1 in IDHm ICC.

(A) Proliferation and survival curves of human IDH WT (black) versus IDHm (red) ICC cell lines treated with increasing doses of dasatinib. (B) Apoptosis assay measuring percentage of IDH WT and IDHm cells stained positive for annexin V by flow cytometry at 48 hours after dasatinib treatments. Data are shown as mean ± SEM between triplicates and are representative of three independent experiments (Student’s two-tailed t test). (C) IDH WT or IDHm ICC cell lines were treated with dasatinib (50 to 500 nM) for 6 hours, and protein lysates were probed for the indicated proteins in the mTOR, ERK, and STAT3 survival pathways by Western blot (WB). (D) Percent reduction in cell size of IDH WT or IDHm ICC cell treated with 100 nM dasatinib for 24 hours, as measured by flow cytometry. Representative flow cytometry forward scatterplots indicative of cell size shift comparing DMSO (solid) and dasatinib-treated (dotted) in RBE (red, top left) and HuCCT1 (black, bottom left). Percentage of average cell size change in IDHm and WT lines (right, one-way ANOVA). (E) Representative IDH WT and IDHm ICC cells were treated as in (C). Thirty minutes before harvest, cells were exposed to 1 μM puromycin. Lysates were probed for the indicated proteins or puromycin-labeled proteins by Western blot. **P < 0.01; ***P < 0.001; ****P < 0.0001. ns, not significant. N.D., not defined.

Although the functions of phosphorylated S6K and S6 remain controversial and are likely context dependent, the best-characterized roles of S6K include the regulation of cell size and translation (17, 18). We therefore investigated whether dasatinib affected either cell size or translation in IDHm ICC cell lines. Dasatinib reduced cell size of all three IDHm ICC lines tested as measured by flow cytometry while having minimal impact on IDH WT cell lines (Fig. 1D). The reduction in cell size was observed within 24 hours of dasatinib treatment and before increased apoptosis was detected (fig. S3). Because cell size is intimately linked to protein synthesis, we measured de novo protein synthesis of IDHm ICC line upon treatment with dasatinib, using puromycin incorporation as a surrogate. Treating IDHm ICC lines with concentrations of dasatinib as low as 50 nM for just 6 hours resulted in a marked reduction in protein synthesis as shown by reduced puromycin uptake, whereas rates of translation remained robust in IDH WT ICC lines treated with doses as high as 500 nM (Fig. 1E). To confirm whether the mTOR pathway was intact in IDH WT cells, we challenged both IDH WT and mutant cells with dual mTORC1 and mTORC2 inhibitors Torin1 and AZD2014. Treatment with either mTOR inhibitor resulted in suppression of pS6K and pS6 at comparable doses (fig. S4, A and B). Together, these data showed that dasatinib-induced cell death is associated with the inhibition of S6K/S6 signaling, rapid suppression of protein synthesis, and apoptosis in IDHm ICC cell lines but not their WT counterparts.

Inhibition of SRC is both necessary and sufficient for killing IDHm ICC through inhibition of S6K/S6 signaling

We next examined whether SRC was necessary and sufficient for regulating signaling and protein synthesis. We have previously generated two endogenous SRC T341I gatekeeper mutant IDHm ICC cell lines, SNU-1079 and RBE (12). Both SNU-1079 and RBE SRC T341I mutants were highly resistant to growth inhibition and apoptosis by dasatinib compared with their parental counterparts at increasing dasatinib concentrations (Fig. 2, A and B). The resistant phenotype in the SRC gatekeeper lines in both SNU-1079 and RBE was associated with rescue of the pSRC Y416 mark as well as pS6K and pS6 abundance (Fig. 2C). None of the other survival signaling pathways, including pAKT, pSTAT3, pERK, and other mTOR downstream targets, such as phospho-eukaryotic translation initiation factor 4E-binding protein 1 (p4E-BP1) and phospho-Unc-51-like kinase 1 (pULK1), was affected (Fig. 2C). Furthermore, dasatinib failed to inhibit de novo protein synthesis as indicated by sustained puromycin uptake in either SNU-1079 or RBE SRC T341I lines, even at concentrations as high as 500 nM (Fig. 2D). These results suggested that SRC is a necessary target to bring about the cytotoxic and translation inhibitory effects of dasatinib.

Fig. 2. Inhibition of SRC is both necessary and sufficient in killing IDHm ICC through inhibition of S6K/S6 axis.

(A) Proliferation curves of parental IDHm ICC cell lines (red) or isogenic lines harboring a genomic SRC T341I gatekeeper mutation rendering endogenous SRC dasatinib-resistant (blue) treated with increasing doses of dasatinib. (B) Parental SRC WT and SRC gatekeeper lines were treated with dasatinib at the indicated doses for 48 hours and assessed for the induction of apoptosis by measuring annexin V positivity by flow cytometry. (C) Cells were treated with increasing concentrations of dasatinib (50 to 500 nM) for 6 hours, and lysates were probed for the indicated proteins in mTOR, ERK, and STAT3 survival pathways by Western blot. (D) Cells were treated as in (C). Thirty minutes before harvest, cells were exposed to 1 μM puromycin. Lysates were probed for the indicated proteins or puromycin-labeled proteins by Western blot. (E) IDH WT (black) or IDHm (red) ICC cell lines were transduced with either a control shRNA or two independent shRNAs against SRC, and lysates were probed with antibodies of the indicated proteins by Western blot.

To further investigate whether inhibiting SRC alone was sufficient to suppress S6K signaling, we knocked down SRC in three IDHm (ICC5, RBE, and SNU-1079) and three IDH WT ICC lines (ICC2, HuCCT1, and CCLP1) using two independent short hairpin RNAs (shRNAs). SRC knockdown (KD) resulted in reduced pS6K and pS6 in all three IDHm ICC lines while having no effect on any of the IDH WT lines (Fig. 2E). SRC KD did not change other phosphorylation marks, including pERK, pSTAT3, pAKT, p4E-BP1, and pULK1 (Fig. 2E). Together, these data demonstrate that SRC promotes cell survival and regulates both pS6K/pS6 signaling axis and protein translation in IDHm but not IDH WT ICC.

MAGI1 is a substrate of SRC and modulates downstream S6K signaling

We next sought to address how SRC regulates the S6K/S6 axis. A Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) analysis was performed that did not identify any known physical interaction between SRC and S6K (fig. S5). Because SRC is a tyrosine kinase and both S6K and S6 are activated by phosphorylation of serine/threonine residues, we hypothesized that one or more intermediate molecules may be involved. Furthermore, we saw no evidence that other upstream or downstream components of the mTOR pathway were affected by SRC inhibition (Fig. 2E). We therefore took both unbiased and hypothesis-driven approaches to identify other molecules that may be involved in SRC-mediated regulation of S6K.

Two IDHm ICC SRC WT and T341I isogenic cell line pairs (RBE and SNU-1079) were treated with dasatinib at 20 nM for 1 hour. Phosphopeptides were extracted from the tryptic digests of the protein lysates and were subjected to mass spectrometry–based multiplexed quantitative phosphoproteomics to characterize dynamic changes in the phosphoproteome (19, 20). Top “hits” would be represented by phosphotyrosine peptides that were reduced in abundance after dasatinib treatment in parental SRC WT lines but not affected by dasatinib treatment in SRC T341I isogenic derivatives. Phosphorylated MAGI1 (pMAGI1) Y373 represented the top “hit” in both SRC WT/T341I pairs tested (Fig. 3A). pMAGI1 Y373 was barely detected in either of the IDH WT ICC cell lines tested (Fig. 3B). MAGI1 is a large scaffolding protein with six PDZ domains, two WW domains, and a kinase dead guanylate kinase (GUK) domain that is thought to localize to tight junctions and function as a tumor suppressor (21, 22). One study identified a potential role for MAGI1 in the phosphatase and tensin homolog (PTEN)/PI 3-kinase (PI3K)/Akt pathway (23). pMAGI1 Y373 can be found in the second WW domain (Fig. 3C) and has been previously described to be a potential substrate for the protein-tyrosine phosphatase receptor type Z phosphatase (24). To explore whether MAGI1 may serve as a SRC substrate, we raised rabbit antisera against pMAGI1 Y373 and coexpressed SRC and either WT MAGI1 or a MAGI1 Y373F mutant (which cannot be phosphorylated) in 293T cells. Coexpression of SRC and WT MAGI1 but not the Y373F mutant resulted in strong phosphorylation of MAGI1 at Y373 (Fig. 3D). Coexpression of SRC with WT MAGI1 full-length and truncation mutants lacking GUK, WW, or GUK-WW domains further demonstrated that Y373 phosphorylation was abolished when both WW domains were absent but not GUK (Fig. 3E). Coimmunoprecipitation (co-IP) in 293T cells revealed that SRC strongly bound to MAGI1 full-length and GUK deletion truncation mutant (Fig. 3F). The WW and GUK-WW truncation mutants showed reduced binding with SRC, with the WW truncation mutant having the lowest affinity (Fig. 3F).

Fig. 3. MAGI1 is a substrate of SRC and modulates downstream S6K signaling.

(A) Identification of SRC substrates by phosphoproteomic screen. Mutant IDH1 ICC SRC WT and SRC T341I gatekeeper mutant pairs were treated with 20 nM dasatinib for 1 hour, and phosphopeptides were extracted from the tryptic digests of the protein lysates, followed by mass spectrometry–based multiplexed quantitative phosphoproteomics. Phospho-tyrosine peptide, pMAGI1 Y373 (red dot), represented the top candidate that was inhibited by dasatinib in both SRC WT lines SNU-1079 (left) and RBE (right) but not their corresponding SRC gatekeeper lines. (B) Relative intensity of pMAGI1 Y373 (left) and total MAGI1 (right) signals in mutant IDH1 (SNU-1079 and SNU-1079 SRC T341I) and WT IDH1 (HuCCT1 and CCLP1) ICC lines treated with DMSO or dasatinib from phosphoproteomic screen. (C) Schematic of MAGI1 showing domain structure and the pMAGI1 Y373 site. (D) 293T cells were transfected with vector control, SRC, GFP-tagged MAGI1 WT, or GFP-MAGI1 Y373F, and lysates were probed with rabbit antisera against pMAGI1 Y373 or the indicated antibodies by Western blot. (E) 293T cells were transfected with vector control, SRC, myc-tagged MAGI1 full length, or myc-tagged MAGI1 deletion of GUK, WW, or GUK-WW domains, and lysates were probed with pMAGI1 Y373 antisera. (F) 293T cells were transfected with vector control, SRC, flag-tagged MAGI1 full length, MAGI1 GUK-deleted, MAGI1 WW-deleted, or MAGI1 GUK-WW–deleted truncation mutants or cotransfected with SRC, flag-tagged MAGI1 full length, and truncation mutants. Lysates were then immunoprecipitated (IP) with flag antibody and analyzed by WB along with 1% input and probed for SRC and flag antibodies. (G) IDHm ICC cells were treated with control siRNA or siRNA against MAGI1 and then exposed to increasing doses of dasatinib (5 to 50 nM) for 6 hours. Lysates were then probed with antibodies against the indicated proteins by Western blot. (H) IDHm cells RBE expressing Cas9 plus either control single guide RNA (sgRNA) or sgRNA targeting MAGI1 (clone 2 and clone 4) were then treated as in (G) and analyzed by Western blot. (I) The same control and MAGI1 knockout clones 2 and 4 as in (H) were treated with dasatinib at indicated doses for 48 hours and subjected to annexin V apoptosis assay (two-way ANOVA; **P < 0.01; ***P < 0.001; ****P < 0.0001).

We next sought to determine whether MAGI1 played a functional role in the S6K/S6 pathway. We transfected three different IDHm ICC cell lines with either control small interfering RNA (siRNA) or pooled siRNA against MAGI1 and treated them with increasing doses of dasatinib. siRNA-mediated KD of MAGI1 partially rescued pS6K/pS6 in all three lines tested, whereas p4E-BP1 remained unchanged (Fig. 3G). To further corroborate these findings, we used CRISPR-Cas9 editing to generate two MAGI1 knockout (KO) RBE clones. Dasatinib treatment in both KO clones also resulted in partial rescue of pS6K/pS6 abundance (Fig. 3H) as well as reduction in annexin V–positive cells, suggesting resistance to dasatinib-induced apoptosis (Fig. 3I). KD and KO of MAGI1 both led to a modest increase in pS6 even in dimethyl sulfoxide (DMSO)–treated cells (Fig. 3, G and H). Thus, we hypothesized that MAGI1 acts as a negative regulator of the S6K/S6 pathway. To examine whether MAGI1 regulated other components of the mTOR pathway, we knocked down MAGI1 in six human ICC lines, finding that it did not affect mTOR signaling in either IDH WT or IDHm cells (fig. S6). Together, these data suggested that SRC-mediated phosphorylation of MAGI1 prevents MAGI1 from suppressing the S6K/S6 pathway.

Dasatinib inhibits S6K/S6 signaling through activation of PP2A

Because MAGI1 is a scaffolding molecule, without intrinsic kinase activity, it is unlikely that MAGI1 directly regulates phosphorylation of S6K. T389 of S6K is a direct substrate of mTORC1; however, we saw no evidence that either upstream or downstream components of the mTORC1 pathway were affected by dasatinib treatment. In addition, overexpressing S6K increased S6K/S6 signaling at baseline but failed to rescue dasatinib-induced suppression of pS6 (fig. S7, A and B). These findings suggested that SRC/MAGI1 may regulate S6K in an mTOR-independent manner. To our knowledge, mTOR is the only kinase capable of phosphorylating S6K T389; thus, we hypothesized that a phosphatase may instead be recruited to mediate SRC/MAGI1 regulation of S6K. Of the known human phosphatases, PP2A has been shown to directly bind to S6K and dephosphorylate T389 without affecting phosphorylation of AKT or 4E-BP (25).

The PP2A holoenzyme is composed of three subunits, A (structural), B (regulatory), and C (catalytic). The human genome encodes only two distinct PP2A-A subunits and two distinct PP2A-C subunits, but there are at least 12 different PP2A-B genes that allow for a variety of regulatory mechanisms and substrate specificities (26). To determine whether PP2A may dephosphorylate S6K T389 in dasatinib-treated IDHm ICC cells, we cotreated cells with okadaic acid, a potent and specific inhibitor of PP2A and protein phosphatase 1 [cell-free median inhibitory concentration (IC₅₀) = 0.1 and 10 nM, respectively] (27) and dasatinib. Okadaic acid counteracted dasatinib-mediated inhibition of pS6K and pS6 in a time-dependent manner in all three IDHm lines, with rescue seen as early as 15 min of treatment (Fig. 4A). To further verify that the inhibition of pS6K and pS6 came from PP2A phosphatase activity, we transfected pooled siRNA against PPP2CA in three IDHm lines. PPP2CA KD increased baseline S6K and S6 phosphorylation, as well as partially rescued pS6K and pS6 abundance upon dasatinib treatment. Neither SRC nor 4E-BP1 phosphorylation was affected, suggesting a specific phosphatase activity targeted against the S6K/S6 axis (Fig. 4B). KD was confirmed by quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR) showing that catalytic A subunit mRNA was lost with no effect on catalytic B subunit mRNA (Fig. 4C).

Fig. 4. Dasatinib suppresses S6K signaling through induction of protein phosphatase 2A.

(A) IDHm ICC RBE, SNU-1079, and ICC5 cells were treated with dasatinib 100 nM for 6 hours and/or the 1 μM PP2A inhibitor okadaic acid for the indicated time points before harvest. Lysates were then analyzed by Western blot and probed for phospho- and total SRC, S6K, S6, and 4E-BP1. Quantification of pS6K and pS6 is presented below each blot and normalized to the corresponding total protein. (B) IDHm ICC cells were treated with control siRNA or siRNA against PPP2CA and then exposed to increasing doses of dasatinib (5 to 50 nM) for 6 hours. Protein lysates were probed for the indicated proteins in the SRC and mTOR pathways by Western blot. (C) mRNA expression of PPP2CA and PPP2CB by qRT-PCR in individual ICC lines confirms specific depletion of PPP2CA by siRNA. (D and E) Change of PP2A catalytic subunit posttranslational modification at pY307 in two IDH1 WT and IDHm ICC pairs (D) and two SRC WT and SRC T341I gatekeeper pairs in IDHm lines (E) upon increasing doses of dasatinib (50 to 500 nM). Lysates were then analyzed by Western blot as above.

To confirm that pMAGI1 was necessary to maintain pS6K, we generated MAGI1 Y373E phosphomimetic and MAGI1 WT reexpressing lines in RBE MAGI1 KO cells. When challenged with dasatinib, the MAGI1 Y373E phosphomimetic mutant could rescue pS6K and pS6 abundance as compared with MAGI1 WT reexpressing and parental RBE lines, whereas p4E-BP1 remained unchanged (fig. S8). Together, our data reveal that SRC-mediated S6K/S6 survival signaling is facilitated through binding and direct phosphorylation of MAGI1 at Y373, which then blocks the inhibitory effect of PP2A on S6K/S6.

PP2A catalytic activity is thought to be modulated by two major posttranslational modifications, pY307 (generally repressive) and methyl-L309 (generally activating) (28). Phosphorylation of PP2A catalytic subunit (PP2AC) at Y307 decreases its activity by inhibiting the interaction between the catalytic and regulatory subunits of the holoenzyme, affecting proper trafficking to the target (29, 30). PP2AC pY307 has been shown to correlate with SRC activity in some contexts (29), and reactivation of PP2A has become a therapeutic avenue of interest for a number of different cancer types (31-33). To understand the role of this relationship in IDHm ICC and to see whether reactivation of PP2A may be occurring through inhibition of SRC, we treated our panel of cell lines with dasatinib and probed for pY307 by Western blot. IDHm ICC cell lines, but not IDH WT cell lines, demonstrated reduction in PP2A pY307 after treatment with dasatinib (Fig. 4D). Moreover, the two SRC gatekeeper IDHm lines demonstrated partial rescue of the dasatinib-induced decrease in PP2AC pY307 (Fig. 4E). These findings suggested that SRC phosphorylates MAGI1 Y373 and thereby inhibits PP2AC through posttranslational modifications. Given a recent report that Y127 and Y284 on PP2AC act as additional phospho-acceptor sites upon SRC activation and the lack of consensus on the specificity of PP2AC pY307 antibody (34), it remains unclear how the interplay of the phosphorylation at these three sites control overall activity of PP2A.

SRC inhibits the growth suppressive MAGI1-PP2A complex to activate S6K

Because MAGI1 is a scaffolding protein, we sought to determine whether it brought together SRC and PP2AC into a signaling complex that could regulate S6K. First, we examined whether the phosphorylation status of MAGI1 at Y373 influenced SRC binding to MAGI1. We coimmunoprecipitated SRC with MAGI1 WT, Y373E (phosphomimetic), or MAGI1 Y373F (nonphosphorylatable) mutants, with or without dasatinib treatment. The MAGI1 Y373F mutant displayed highest binding to SRC. Dasatinib treatment led to increased SRC binding for both WT MAGI1 and MAGI1 Y373E mutant but not for the MAGI Y373F mutant (Fig. 5A). Dasatinib belongs to a class of inhibitors that stabilizes the open conformation of SRC, which may account for the increased binding (35). However, SRC inhibition does not seem to increase SRC binding to a nonphosphorylatable MAGI1 Y373F (Fig. 5A). These data indicate that, although phosphorylation status of MAGI1 Y373 influences SRC binding, it is not correlated with SRC activity.

Fig. 5. SRC interacts with MAGI1-PP2A complex to regulate S6K.

(A) 293T cells were cotransfected with WT SRC and flag-tagged MAGI1 Y373 WT/Y307E/Y307F for 2 days and treated with DMSO or 50 nM dasatinib for 6 hours before lysis. Lysates were then immunoprecipitated with flag antibody and analyzed by Western blot along with 1% input and probed for SRC and flag antibodies. (B) 293T cells were transfected with vector control, HA-tag PP2A catalytic subunit (PP2AC), flag-tagged MAGI1 Y373 WT, flag-tagged MAGI1 Y373E, or flag-tagged MAGI1 Y373F or cotransfected with HA-PP2AC and flag-tagged MAGI1 Y373 WT/E/F mutants. Cells were treated with DMSO or dasatinib before harvesting for co-IP as in (A) and analyzed by Western blot with HA and flag antibodies. (C) 293T cells were transfected with HA-tagged PP2AC and flag-tagged MAGI1 full-length or without WW, GUK, or GUK-WW domains. Lysates were immunoprecipitated with flag antibody. (D) Flag-tagged MAGI1 full-length WT and HA-PP2AC Y307 WT/E/F mutants were cotransfected in 293T cells treated with DMSO or dasatinib for 6 hours and processed for immunoprecipitation as in (A). HA-tagged PP2AC and endogenous S6K pulled down were analyzed by WB. (E) 293T cells were cotransfected with HA-tagged S6K and flag-tagged MAGI1 Y373 WT/Y307E/Y307F and treated with DMSO or dasatinib before harvesting for co-IP as in (A).

Next, we examined the physical interaction between MAGI1 and PP2AC. PP2AC coimmunoprecipitated with MAGI1 in 293T cells, and the binding was significantly increased upon dasatinib treatment (Fig. 5B and fig. S9). This supports the hypothesis that, upon SRC inhibition, MAGI1 recruits a suppressive signaling complex that facilitates downstream dephosphorylation of S6K. We also performed co-IP of PP2AC with MAGI1 Y373E and Y373F mutants with or without dasatinib treatment and found that PP2AC appeared to bind to MAGI1 Y373F more strongly than Y373E, and this binding did not increase with SRC inhibition (Fig. 5B). This suggested that SRC-mediated phosphorylation of MAGI1 may inhibit PP2AC binding. To determine the MAGI1 domain that binds to PP2AC, we coexpressed WT MAGI1 full-length and truncation mutants lacking GUK, WW, or GUK-WW domains with PP2AC. We found that PP2AC bound less strongly to the truncation mutant lacking both the GUK and WW domains (Fig. 5C). Next, we performed co-IP of MAGI1 with WT, Y307E (phosphomimetic), or Y307F (nonphosphorylatable) PP2AC with or without dasatinib treatment. PP2AC Y307E showed weaker, whereas Y307F showed stronger, binding to MAGI1 at baseline (Fig. 5D). Although dasatinib treatment resulted in increased binding of both WT and Y307F PP2AC to MAGI1, Y307E phosphomimetic mutant did not show any change (Fig. 5D). These data suggested that phosphorylated or inactive PP2AC Y307 cannot bind MAGI1, thus preventing dephosphorylation of S6K/S6 signaling. Endogenous S6K appeared to bind to MAGI1 equally well with or without dasatinib when coexpressed with PP2AC WT and Y307 mutants (Fig. 5D). To further confirm S6K binding to MAGI1, we performed co-IP between MAGI1 WT or Y373E/F mutants (flag-tagged) with S6K [hemagglutinin (HA)–tagged] with or without dasatinib. S6K interacted equally with the WT and MAGI1 phospho-mutants, and the interaction did not change upon dasatinib treatment (Fig. 5E), suggesting that it is the PP2A-MAGI1 interaction that acts as a determinant of pS6K/pS6 signaling.

These results showed that SRC phosphorylation of MAGI1 Y373 and PP2A Y307 prevents the formation of a suppressive MAGI1-mediated signaling complex, limiting access of PP2AC to S6K and leading to S6K/S6 hyperactivation. Inhibition of SRC allows PP2A to bind MAGI1 and dephosphorylate MAGI1-bound S6K.

Elevated pS6 determines intrinsic and acquired resistance to dasatinib

Our data suggested that IDHm ICC is dependent on SRC for cell survival. Immunofluorescence revealed that IDHm cell lines had lower abundance of pS6 compared with IDH WT (Fig. 6, A and B). This finding was recapitulated by immunohistochemistry on samples obtained from patients with ICC (Fig. 6, C and D), suggesting that higher baseline abundance of pS6 may determine resistance. To test this hypothesis, we inhibited mutant IDH1 and therefore accumulation of 2-HG, with a potent and specific pharmacological inhibitor, ivosidenib (AG-120). Treatment with AG-120 in all three human IDHm ICC cell lines inhibited cellular 2-HG concentrations with an average IC₅₀ of 90 nM, without causing growth inhibition using doses as high as 10 μM in proliferation assay (Fig. 6, E and F). Pretreating IDHm ICC cells with AG-120 resulted in a substantial rescue from dasatinib-induced cytotoxicity, as shown by crystal violet staining, annexin V apoptosis assay, and caspase-3 and poly(ADP-ribose) polymerase (PARP) cleavage (Fig. 6, G to I). Furthermore, AG-120 pretreatment (4 days) in the three IDHm ICC lines showed a higher baseline abundance of pS6K/pS6 compared with IDH WT levels and partial rescue of dasatinib-mediated inhibition of pS6K/pS6 (Fig. 6J). Knock-in (KI) of the WT allele of IDH1 into IDHm ICC cells phenocopied treatment with AG-120 and also resulted in elevated pS6K/pS6 at baseline and a partial rescue from dasatinib-induced inhibition of pS6 (Fig. 6, K and L).

Fig. 6. 2-HG is required in dasatinib-induced cytotoxicity and inhibition of pS6K.

(A) Immunofluorescence staining of pS6 and total S6 in RBE, SNU1079, ICC5 (IDH1 mutant), and HuCCT1 (WT) cells. Scale bars, 50 μm. (B) Quantification of pS6 median staining intensity. Each data point represents one cell (one-way ANOVA test, ****P < 0.0001). (C) Immunohistochemistry staining of pS6 in samples from patients with IDH WT or IDH1 R132C ICC. Scale bars, 50 μm. (D) Quantification of pS6 IHC staining intensity by blinded histopathology scoring (one-way ANOVA test, *P < 0.05). (E) IC₅₀ curves of IDH1 mutant-specific inhibitor AG-120 in suppressing cellular 2-HG concentrations in three IDHm (SNU-1079, RBE, and ICC5) and IDH WT (RBE KI c5 and RBE KI c9) cell lines. (F) Proliferation curves of IDHm (red) and IDH WT (black) ICC cell lines with increasing doses of AG-120. (G to H) SNU-1079 cells were treated with regular media, DMSO, or 5 μM AG-120 for 1, 2, 3, or 4 days followed by dasatinib treatment for 72 hours at indicated doses and subjected to crystal violet staining (G), annexin V apoptosis assay (Student’s two-tailed t test, **P < 0.01; ****P < 0.0001) (H), and Western blot for apoptosis markers cleaved caspase-3 and PARP (I). (J) IDHm ICC cells were pretreated with either DMSO (black) or 5 μM AG-120 (red) for 4 days, followed by DMSO or increasing doses of dasatinib (5 to 50 nM) for 6 hours. Cells were then harvested, and lysates were probed for the indicated proteins by Western blot. (K) IDHm RBE cells and their isogenic WT knocked-in (KI) clone 9 were subjected to increasing doses of dasatinib (50 to 500 nM) treatment for 6 hours and analyzed for the indicated proteins by Western blot. (L) Baseline expression of pSRC, pS6K, and pS6 in a panel of human ICC cell lines (IDHm in red, WT in black, and AG-120 treated in gray). (M) Dasatinib-resistant clones generated through continuous dasatinib treatment were subjected to increasing doses of dasatinib treatment, and abundance of pS6 was analyzed by Western blot.

Next, we generated dasatinib-resistant clones over several months by culturing RBE cells in increasing concentrations of dasatinib. All three resistant clones showed higher IC₅₀ values than the parental lines in response to dasatinib and other SRC inhibitors (fig. S10). These clones with acquired dasatinib resistance demonstrated increased abundance of pS6 at baseline and maintained higher pS6 with increasing doses of dasatinib (Fig. 6M). We reasoned that the activation of PP2A through inhibition of SRC was not enough to suppress pS6 abundance, and further inhibition of S6 upstream kinase activity may be needed to synergize with dasatinib in IDHm cells to counter both intrinsic and acquired resistance through elevated pS6.

To test this, we treated cells with S6K inhibitors, including PF-4708671 (36), LY-2779964 (37), and M2698 (38). M2698, a potent, orally bioavailable, selective inhibitor against S6K, AKT1, and AKT3, had the most robust activity (fig. S11). Combination of clinically achievable doses of both M2698 and dasatinib reduced pS6 to undetectable levels in IDHm ICC cells, whereas there was no change in IDH WT cells (Fig. 7A). Treatment of patient-derived organoids (PDOs) with dasatinib in combination with M2698 improved the reduction of pS6 abundance in IDHm ICC PDOs when compared with dasatinib alone, whereas there was no change in IDH WT PDOs (Fig. 7, B to E). We then tested the efficacy of dasatinib and M2698 alone and in combination in IDHm and IDH WT ICC PDX models. Mice with established tumors were randomized into four arms—vehicle control, dasatinib only, M2698 only, and dasatinib + M2698 combination. PDX62 (IDH1 R132C) treated with a single treatment of either dasatinib or M2698 resulted in significant inhibitory effects relative to control, and dasatinib + M2698 inhibited tumor growth to a much greater extent compared with either single drug alone (Fig. 7F). In contrast, the same combination regimen did not cause an observable change in growth in IDH WT PDX (Fig. 7G). Dasatinib + M2698 resulted in tumor shrinkage in PDX62 (IDH1 R132C) at day 29, which was not seen in either single arm (Fig. 7H). Treatment with either agent resulted in prolonged survival, which was further improved in the combination arm (Fig. 7I). Immunohistochemistry (IHC) showed an increase of apoptosis and reduction of proliferation in the combination arm as measured by cleaved caspase-3 and Ki67 staining, respectively (Fig. 7, J and K). The dosages of dasatinib and M2698 were 30 and 10 mg/kg, respectively, which were both below the single-agent effective doses used in previous studies (12, 39). The combination treatment did not cause substantial changes in body weight (fig. S12). Together, the in vivo data from PDX models provide proof-of-principle evidence that dasatinib in combination with the S6K/AKT inhibitor M2698 could represent a therapeutic option for patients with IDHm ICC.

Fig. 7. Combination treatment with SRC and S6K inhibitors suppresses IDHm ICC growth in patient-derived models.

(A) IDH WT and IDHM ICC cells were treated with increasing doses of dasatinib (50 to 500 nM) with or without S6K1/AKT inhibitor M2698 at 10 nM for 6 hours. Cells were then harvested, and lysates were probed for the indicated proteins by Western blot. (B to E) Patient-derived organoids ICC195, FHICC19 (WT), and FHICC17 (IDH1m) were treated with increasing doses of dasatinib (100, 500, and 1000 nM) for 16 hours, and the levels of pSRC, pS6K, and pS6 were analyzed by Western blot (B). FHICC17 IDH1m organoids were also treated with the combination of dasatinib and S6K1/AKT inhibitor M2698 at 20 nM for 16 hours, and the levels of pSRC, pS6K, and pS6 were analyzed by Western blot (C). Quantifications of pS6 levels in dasatinib-treated organoid lines ICC195, FHICC19, and FHICC17 are presented in (D), and quantifications of pS6 in dasatinib plus M2698-treated FHICC17 IDH1m organoids are presented in (E). (F to K) NSG mice with subcutaneously implanted IDH1 WT PDX (ICC 195) and IDH1 R132C mutant PDXs (PDX62) were treated with vehicle control, dasatinib (30 mg/kg), M2698 (10 mg/kg), or dasatinib (30 mg/kg) + M2698 (10 mg/kg) daily for 28 days by oral gavage (PDX62, n = 9, 10, 9, and 12 respectively; ICC195, n = 5 each arm). Part of the IDH1 mutant PDX cohort was harvested at day 28 of treatment (n = 3, 4, 3, and 4 for vehicle, dasatinib, M2698, and dasatinib + M2698 respectively), and the remaining mice were monitored for survival up to 70 days after treatment. Tumor volume fold change of IDH1 mutant PDX (F) and IDH1 WT (G), tumor volume change at day 29 compared with day 1 for IDH1 mutant PDX (H), and survival plot for IDH1 mutant PDX (Kaplan-Meier analysis and log-rank P values are shown between groups) (I). (J) Histological analysis of tumors from PDX62 vehicle, dasatinib, M2698, and combo groups. Left column: Hematoxylin and eosin (H&E) staining; middle column: IHC staining for Ki67 (proliferation marker); and right column: IHC staining for cleaved caspase-3 (cell death marker). Representative images of each experimental group are shown. Scale bar, 250 μm. (K) Quantification of Ki67 (top) and cleaved caspase-3 (bottom) IHC staining. Each dot represents the percentage of cells with positive staining in a randomly selected area. Five areas per slide were quantified. Data are mean ± SEM (one-way ANOVA multiple comparisons; *P < 0.05; ****P < 0.0001).

DISCUSSION

This study delineated the molecular mechanisms behind the hypersensitivity of IDHm ICC to dasatinib. We demonstrated that dasatinib-induced SRC inhibition leads to specific reduction of S6K/S6 signaling through dephosphorylation of an adaptor protein, MAGI1, and activation of PP2A (fig. S13).

PP2A holoenzyme is a serine/threonine phosphatase that is composed of structural (A), regulatory (B), and catalytic (C) subunits and targets many substrates implicated in oncogenic pathways, including MYC (40), ERK, MEK (41), AKT (42), and S6K (43, 44). As a tumor suppressor, PP2A is frequently mutated or functionally inhibited in many common malignancies, including breast, prostate, lungs, colon, melanoma, etc. Ongoing studies are aiming to simultaneously inhibit oncogenic kinases and activate PP2A to enhance antitumor activity. The current study presents evidence for the involvement of PP2A in cholangiocarcinoma. We demonstrated that PP2A-mediated dephosphorylation of S6K is the key effector response upon SRC inhibition in IDHm ICC cells. This reveals a vulnerability of IDHm ICC cells to PP2A activation.

The specific mechanism by which MAGI1 modulates PP2A and suppresses S6K signaling remains to be fully characterized. Although SRC is reported to inhibit PP2A by phosphorylating its catalytic subunit (45), our findings suggest that this interaction is subject to additional regulation. Dasatinib inhibits SRC equally in IDH WT and mutant cells, but S6K and S6 phosphorylation are only reduced in the mutant counterparts. Although MAGI1 was similarly abundant in both IDH mutant and IDH WT cells, pMAGI1 at Y373 was detected at much lower abundances in WT than IDH mutant cells based on phosphoproteomic data (Fig. 3B). The difference in MAGI1 baseline phosphorylation may contribute to the differential activation of PP2A and subsequent sensitivity to dasatinib between IDH WT and mutant ICC cells. Increasing evidence suggests that MAGI1 functions as a tumor suppressor (21, 46). With multiple PDZ domains, MAGI1 can bring together various binding partners to facilitate signaling, an example of which is the membrane recruitment of PTEN (23). MAGI1 recruited activated PP2A and S6K to turn off survival signaling in an SRC-regulated manner. We hypothesize that SRC is active and phosphorylates MAGI1 at Y373 and PP2AC at Y307, inhibiting both PP2A phosphatase activity and the formation of the MAGI1-PP2A tumor-suppressive complex leading to downstream activation of S6K/S6 survival signaling in IDHm cells. When IDHm cells are treated with dasatinib, SRC-mediated phosphorylation is blocked, derepressing both MAGI1 and PP2A.The now-active PP2AC dephosphorylates S6K, leading to inactivation of S6K/S6 signaling and cell death.

mTOR controls protein synthesis, at least in part, through direct phosphorylation of the tumor suppressor eukaryotic translation initiation factor 4E-BP1 and S6K. Evidence suggests the presence of a mechanism to control the activity of S6K and 4E-BP1 independent of mTORC1 activity and independent of each other. Treatment with rapamycin, a specific inhibitor of mTORC1, has shown differential regulation of 4E-BP1 and S6K in a cell-specific manner. Rapamycin potently inhibits S6K activity, but 4E-BP1 recovered phosphorylation over 6 hours despite mTOR inhibition (47). Similarly, primary B lymphocytes isolated from 4-week-old Eμ-Myc mice show an unexpected and specific increase in mTORC1-dependent phosphorylation of 4E-BP1, whereas S6K was not altered (48). Our discovery of PP2A-mediated dephosphorylation of S6K in a MAGI1-dependent manner reveals a signaling complex that negatively regulates S6K/S6 activity. This discovery presents a potential explanation for the dissociation of 4E-BP1 and S6K phosphorylation.

We are also reminded of similarities involving interactions among viral and cellular proteins. Adenovirus type 9 (Ad9) E4-ORF1 and high-risk human papillomavirus (HPV) E6 proteins bind to the PDZ domains of MAGI1, resulting in MAGI1 being aberrantly sequestered in the cytoplasm by the Ad9 E4-ORF1 protein or being targeted for degradation by high-risk HPV E6 proteins. The authors of these studies have surmised that the tumorigenic potentials of these viral oncoproteins may depend, in part, on an ability to inhibit the function of MAGI1 in cells (49-52). Polyomavirus middle-T antigen (MT), SRC, and PP2A also form a similar signaling complex (53). Like MAGI1, MT contains no intrinsic kinase activity and relies on recruitment of SRC to activate downstream oncogenic cell signaling. In an ordered sequence of interactions, MT binds to the core dimer of PP2A (54, 55) and then to a member of the SRC family of tyrosine kinases, usually pp60^c-src (56) or pp62^c-yes (57). This activates the kinase activity of SRC, which phosphorylates tyrosines within MT. Three of these phosphotyrosines act as binding sites for the SH2 or PTB domains of PI3K (MT Y315) (58), ShcA (Y250) (59, 60), and phospholipase C-γ1 ([PLC-γ1] Y322) (61). As a consequence of their interaction with MT, each of these polypeptides is, in turn, tyrosine phosphorylated, which activates PI3K- and PLC-γ1–dependent signaling pathways and creates a binding site on ShcA for Grb2 (60). The guanine nucleotide exchange factor Sos1 and the adapter molecule Gab1 (62) are brought into the MT complex through their interactions with Grb2, thereby activating Ras and the ERK kinase cascade (63, 64). However, unlike MT complex, SRC-MAGI1-PP2A appears to be growth suppressive through inactivation of S6K/S6 signaling. Further assessment of the similarities and differences between MAGI1 and MT may provide new insights in exploring the biochemical regulation of this SRC-MAGI1-PP2A survival signaling complex in ICC and other cancers.

Inhibition of 2-HG by treatment with AG-120 and IDH1 WT KI could partially reverse dasatinib hypersensitivity and S6K/S6 signaling, suggesting that 2-HG promotes reliance on SRC in ICC cells. One possible mechanism is through acting on one or more of the ~70 aKG-dependent dioxygenase family enzymes that have diverse cellular functions, including epigenetic modifications, DNA damage repair, collagen synthesis, and hypoxia response. A recent report suggested that 2-HG can activate mTOR signaling through inhibition of histone demethylase lysine demethylase 4A, one of the αKG dioxygenase family enzymes (65). Experiments performed in endogenous IDH1 R132C fibrosarcoma line HT1080, in which the authors reported that inhibiting IDHm by earlier generation of mIDH1-specific inhibitor AGI-5198 resulted in a reduction of mTOR signaling, showed an opposite effect to what we have observed (62). This may indicate a difference in tissue specificity in terms of the implicated dioxygenase enzymes targeted by 2-HG. It remains to be elucidated whether any specific aKG dioxygenase family enzymes are targeted by 2-HG in ICC to drive disease pathogenesis. Regardless, the data we presented may have important clinical implications, because treating ICC with ivosidenib could make ICC cells more resistant to dasatinib; therefore, combination or sequential applications of dasatinib and ivosidenib in patients with ICC may have to be avoided.

We also presented evidence that pS6 abundance predicts intrinsic and extrinsic resistance to dasatinib in multiple ICC cell line models, clinical samples, and PDOs. Targeting pS6 is of critical importance in overcoming dasatinib resistance, and we demonstrated that combination treatment with dasatinib and M2698 could effectively suppress pS6 signaling. Inhibitors targeting mTOR and PI3K have shown limited efficacy because inhibition of a single node in the mTOR/PI3K pathway can lead to compensatory activation, usually of AKT, via release of a negative feedback loop (66, 67). M2698 has the potential to block the AKT compensatory feedback loop while avoiding the adverse effects of pan-AKT inhibition (ipatasertib, capivasertib, GSK690693, and MK-2206), including those associated with AKT2 inhibition (68). M2698 has been well tolerated in a phase 1 clinical trial (38). We suggest that M2698 and dasatinib could be a plausible combination for patients with IDHm ICC.

One of the major limitations in this study is the low number of human IDHm models available. There are only three established IDHm ICC cell lines, and these cannot generate tumors in mice. We studied in vivo efficacy using a PDX model; however, we found that the tissue grafting take rate was very low, which limited the power of the preclinical efficacy study. In addition, the current study did not address the immune influence of dasatinib and M2698 treatments in vivo because of the use of non-obese diabetic severe combined immunodeficiency gamma mouse (NSG) mice lacking an intact immune system.

In summary, we reported the molecular mechanism behind the hypersensitivity of IDHm ICC to dasatinib, involving SRC inhibition followed by MAGI1-PP2A activation that suppresses S6K/S6 signaling and leads to a reduction in global protein translation and survival. We identified pS6 as an indicator of dasatinib sensitivity and described a combination therapy of dasatinib and an S6K/AKT inhibitor that reduces pS6 and improves cell growth inhibition over dasatinib alone, specifically in IDHm ICC.

MATERIALS AND METHODS

Study design

The goals of this study were to dissect the underlying mechanism by which IDHm ICC cells are sensitive to SRC inhibition and to evaluate combination treatments to overcome resistance for potential clinical applications. These objectives were accomplished by (i) measuring signaling responses to drug treatments and gene silencing by analysis of survival pathways; (ii) assessing cellular changes by functional assays such as flow cytometry for apoptosis, cell size, and puromycin uptake for translation; (iii) identifying SRC substrates by unbiased phosphoproteomic screen and physical interaction of SRC, MAGI1, PP2A, and S6K signaling complex by co-IPs; (iv) evaluating marker for dasatinib resistance and the effect of combining dasatinib with S6K inhibitor M2698 in overcoming treatment resistance using cell line, PDO (in vitro), and PDX (in vivo) models. For in vitro experiments, we used the maximum number of available human IDHm cell lines and successfully derived PDO lines. For in vivo experiments, sample size and treatment regimen were determined on the basis of published literature and past experience. Mice were randomized into treatment arms. Investigators were blinded to the treatment effect, and a blinded pathologist performed all histological analyses for murine studies. All in vitro experiments were performed with a minimum of two replicates or as indicated.

Cell lines

Cell lines were obtained from the Riken Bioresource Center (RBE and HuCCT1) and the Korean Cell Line Bank (SNU-1079) or were derived as previously described (ICC2 and ICC5) (12). CC-LP-1 was a gift from P. J. Bosma (Academic Medical Center, Amsterdam, Netherlands). RBE IDH1 S132R WT KI pool cells were generated from parental RBE cells with IDH1 R132S by Synthego using guide RNA sequence TCATAGGTCGTCATGCTTAT. RBE KI monoclonal cell lines were then generated from pooled cells by limiting dilution. Each KI clone was sequenced to confirm the conversion of encoded amino acid from serine to arginine. Cell lines were grown at 37°C under 5% CO₂ in their required growth medium (Gibco) supplemented with 10% fetal bovine serum and 1% penicillin and streptomycin. Cells were passaged by trypsinization.

PDO culturing and drug treatments

Human organoids were derived and cultured according to the methods detailed previously (69). Human ICC tissue from surgical resected samples were obtained from R. Yeung [Department of Surgery, University of Washington; Institutional Review Board (IRB) no. 00001852]. Fresh tumor chunks were minced and digested with collagenase II (5 mg/ml; Gibco), deoxyribonuclease I (10 μg/ml; Sigma-Aldrich), and Y-27632 Rho kinase inhibitor (10.5 μM; Sigma-Aldrich) in human feeding media in a rotating incubator set at 37°C and 35 rpm rotation for three rounds of 15 min each. Cells isolated from human tissue were embedded in Matrigel (Corning) and cultured in a 24-well plate in human complete media. Media were changed every 2 to 4 days. For drug treatment, media were changed to human complete medium containing drug at the desired concentration. After 16 hours of drug treatment, organoids were harvested by resuspending domes in Cell Recovery Solution (Corning) and washed with cold phosphate-buffered saline before snap-freezing the pellets for subsequent protein isolation and Western blotting as described below.

Protein isolation and Western blot

Protein lysates were prepared by lysing cells directly in NP-40 buffer supplemented with a protease inhibitor cocktail (cOmplete EDTA-free, Roche Applied Science) and 5 μM phosphatase inhibitors (phosphatase inhibitor cocktail sets I and II, Calbiochem). Cells suspended in lysis buffer were sonicated for 10 s of active sonication, followed by 20 s of rest for three cycles at 20% amplitude. The lysate was then centrifuged at 14,000 rpm for 10 min at 4°C, and the supernatant was harvested. A BCA protein assay kit (Pierce) was used to measure and normalize protein concentration. Thirty micrograms of the cell lysate was run on a 4 to 20% gradient polyacrylamide gel with SDS (Bio-Rad) and electroblotted onto polyvinylidene difluoride membranes (Millipore). Membranes were blocked in tris-buffered saline with tween-20 (TBS-T) with 5% nonfat milk and 0.1% Tween and probed with primary antibodies (table S1). Horseradish peroxidase–conjugated secondary antibodies (Vector Biolaboratories) were used to detect membrane-bound proteins. Blots were developed using the Clarity Max Western ECL Blotting Substrate (Bio-Rad). Signaling experiments are representative of at least two independent experiments.

Phosphoproteomics

Protein digestion and TMT labeling

Cell pellets were resuspended in cell lysis buffer [75 mM NaCl, 50 mM Hepes (pH 8.5), 10 mM sodium pyrophosphate, 10 mM sodium fluoride, 10 mM β-glycerophosphate, 10 mM sodium orthovanadate, 1 mM phenylmethylsulfonyl fluoride, 3% SDS, and complete mammalian protease inhibitor tablet (Roche)] by passing the suspension through a 21-gauge needle 20 times. Dithiothreitol was used to reduce disulfide bonds, and free thiols were alkylated with iodoacetamide as described previously (70). Reduced and alkylated proteins were then precipitated following the methanol / chloroform method precipitation as described previously (70). Precipitated proteins were reconstituted in 300 μl of 1 M urea in 50 mM Hepes (pH 8.5). Vortexing and sonication were used to aid solubility. Proteins were then digested in a two-step process, first with 3 μg of endoproteinase Lys-C (Wako) for 17 hours at room temperature (RT) and then with 3 μg of sequencing-grade trypsin (Promega) for 6 hours at 37°C. The digest was acidified with trifluoroacetic acid (TFA). Peptides were desalted over Sep-Pak C18 solid-phase extraction cartridges (Waters). The peptide concentration was determined using a BCA assay (Thermo Fisher Scientific), and a maximum of 50 μg of peptides was aliquoted, then dried under vacuum, and stored at −80°C before labeling with tandem mass tag (TMT) reagents. Peptides were labeled with 10-plex TMT reagents (Thermo Fisher Scientific). TMT reagents were suspended in dry acetonitrile (ACN) at a concentration of 20 μg/μl. Dried peptides were resuspended in 30% dry ACN in 200 mM Hepes (pH 8.5), and 5 μl of the appropriate TMT reagent was added to the sample, which was incubated at RT for 1 hour. The reaction was then quenched by adding 6 μl of 5% (w/v) hydroxylamine in 200 mM Hepes (pH 8.5) and incubated for 15 min at RT. The solutions were acidified by adding 50 μl of 1% TFA, combined into one sample, and desalted. If the number of samples exceeded 10, then samples were split across two TMT sets, and a bridge sample generated by pooling a part of all samples was added to each TMT set (71).

Basic pH reversed-phase liquid chromatography sample fractionation

Basic pH reversed-phase liquid chromatography (bRPLC) was used to perform sample fractionation with concatenated fraction combining. Briefly, samples were resuspended in 5% formic acid (FA)/5% ACN and separated over a 4.6 mm by 250 mm ZORBAX Extend C18 column (5 μm, 80 Å, Agilent Technologies) on an Agilent 1260 high-performance liquid chromatography system outfitted with a fraction collector, degasser, and variable wavelength detector. A two-buffer system (buffer A: 5% ACN and 10 mM ammonium bicarbonate; buffer B: 90% ACN and 10 mM ammonium bicarbonate) was used for separation, with a 20 to 35% gradient of buffer B over 60 min at a flow rate of 0.5 ml/min. A total of 96 fractions were collected, which were combined in a total of 24 fractions. The combined fractions were dried under vacuum and reconstituted with 8 μl of 5% FA/5% ACN, 3 μl of which was analyzed by liquid chromatography–mass spectrometry 2/ mass spectrometry 3 (LC-MS2/MS3).

Phosphopeptide enrichment

Peptides were subjected to enrichment for phosphopeptide enrichment using a 4:1 ratio of titanium dioxide beads to peptide (w/w) (19, 20). Peptides were resuspended in 2 M lactic acid in 50% ACN and added to 1.8 mg of titanium dioxide beads. The mixture was shaken gently for 1 hour. Beads were collected by centrifugation and washed three times with 2 M lactic acid in 50% ACN and three times with 50% ACN/0.1% TFA. Phosphopeptides were eluted with 2 × 200 μl of 50 mM KH₂PO₄ (pH 10) and acidified with 1% TFA. Eluted phosphopeptides were desalted, lyophilized, and labeled with 2 μl of 10-plex TMT reagents 127n to 130c as described above. The combined sample was enriched for phosphotyrosine-containing peptides using phosphotyrosine antibody–conjugated beads (Cell Signaling Technology) following the protocol provided by the manufacturer. Unbound peptides (phosphoserine and phosphothreonine peptides) were desalted, lyophilized, and fractionated by bRPLC using a gradient of 5 to 28% buffer B. A total of 96 fractions were collected, and fractions were combined into 12 fractions. Bound peptides (phosphotyrosine peptides) were eluted and desalted. All 13 fractions were resuspended in 5% ACN/5% FA and analyzed on an Orbitrap Fusion mass spectrometer using LC-MS2/MS3 for identification and quantification of the phosphopeptides.

Mass spectrometry data acquisition and analysis

Combined sample fractions were dried, resuspended in 5% ACN/5% FA, and analyzed in 3-hour runs via LC-M2/MS3 on an Orbitrap Fusion mass spectrometer using the Simultaneous Precursor Selection (SPS)–supported MS3 method (19, 70, 72, 73). Two MS2 spectra were acquired per peptide upon higher energy collisional dissociation fragmentation and collision-induced dissociation fragmentation followed by an SPS-MS3 spectrum on the CID fragment ions (19, 20). MS2 spectra were assigned using a SEQUEST-based in house built proteomic analysis platform (74) allowing phosphorylation of serine, threonine, and tyrosine residues as a variable modification. The Ascore algorithm was used to evaluate the correct assignment of phosphorylation within the peptide sequence (75). On the basis of the target-decoy database search strategy (76) and using linear discriminant analysis and posterior error histogram sorting, peptide and protein assignments were filtered to a false discovery rate of <1% (74). Peptides with sequences that were contained in more than one protein sequence from the UniProt database were assigned to the protein with the most matching peptides (74). TMT reporter ion intensities were extracted as that of the most intense ion within a 0.03 window around the predicted reporter ion intensities in the collected MS3 spectra. Only MS3 with an average signal-to-noise value of larger than 40 per reporter ions as well as with an isolation specificity of larger than 0.75 was considered for quantification (70). A two-step normalization of the protein TMT intensities was performed by first normalizing the protein intensities over all acquired TMT channels for each protein on the basis of the median average protein intensity calculated for all proteins. To correct for slight mixing errors of the peptide mixture from each sample, a median of the normalized intensities was calculated from all protein intensities in each TMT channel, and the protein intensities were normalized to the median value of these median intensities.

shRNA transfection

Viral particles containing human short hairpin (sh) SRC #1 (TRCN0000195339) target sequence: 5′-CATCCTCAGGAACCAAC AATT-3′ and shSRC #2 (TRCN0000199186) target sequence: 5′-CTG ACTGAGCTCACCACAAAG-3′ were synthesized using retroviral (pCL-ECO) packaging plasmids with pCMV-VSV-G (Addgene). pLKO.1 shRNA with target sequence 5′-GCAAGCTGACCCTG-AAGTTCAT-3′ was used as a negative control. Cells were incubated with virus and polybrene (8 μg/ml; Millipore, #TR-1003-G) for 24 hours and subsequently selected in puromycin (2.5 μg/ml) for at least 2 days. Western blots were performed as previously described, and results are representative of two independent experiments.

siRNA transfection

Cells were transfected with pooled siRNA targeting MAGI1c (Dharmacon), PPP2CA/B (Dharmacon), or SRC (Dharmacon) at a final concentration of 40 nM using Lipofectamine RNAimax. Cells were harvested 48 to 72 hours after transfection and processed as stated previously. Results are representative of two independent experiments.

Plasmids

pCDNA3.1 murine full-length myc-MAGI1, myc-MAGI1 deletion mutants, and individual myc-MAGI1 domain constructs were gifts from M. Baccarini (23). Human MAGI1c-Y373F–green fluorescent protein (GFP) mutants were generated from pCMV6-AC-GFP-MAGI1c WT (Origene, #RG212712) by site directed mutagenesis using the following primers: CATAGTAGATACCAAAGCAGGG TCTTCAATCTTTTCCCAAC and GTTGGGAAAAGATTGAA GACCCTGTCTTTGGTATCTACTATG. pCMV SRC WT and pRK7-HA-S6K1-WT (Addgene, #8984) were gifts from J. Cooper and J. Blenis.

CRISPR guide RNA sequences

Guide RNA sequence used to generate human MAGI1 KO in Cas9 expressing stable RBE cells: G*A*A*GGGUUUCGUGUAAAAAA, A*U*C*AAGAGCUUGGUCCUAGA, and U*C*G*UGGCUUU GGCUUCACGG (MAGI1); G*C*A*CUACCAGAGCUAACUCA (nontargeting control). * indicates 2′-O-methyl analogs and 3′-phos-phorothioate internucleotide linkages. All sequences had an added Synthego modified EZ scaffold at 3′ (Synthego).

Immunohistochemistry and image analysis

Samples from patients with ICC

Surgical resected tumor samples from patients with ICC were obtained from R. Yeung following University of Washington Medicine IRB protocol #00001852. Tissue samples were fixed for 7 days in 10% buffered formalin phosphate, embedded in paraffin, and sectioned (5-μm thickness) by the Fred Hutchinson Cancer Center Experimental Histopathology Core. Immunohistochemistry was performed as previously described (77). Briefly, sections were hydrated followed by antigen retrieval using sodium citrate buffer. Sections were stained using anti-phospho-S6 ribosomal protein (Ser^235/236) primary antibody (Cell Signaling Technology, #4858) at 1:50 dilution. Biotinylated secondary antibody was used at 1:200. Sections were then stained with hematoxylin (Thermo Fisher Scientific, #6765007), dehydrated, and mounted. Stained slides were visualized using the Zeiss Observer. Z1 microscope at ×20 magnification. Representative images were captured using uniform brightness and contrast between samples.

PDX tissue samples

Sample preparation and immunohistochemistry experiments were performed as previously described (72). Briefly, tumors were fixed immediately after excision in a 4% buffered formalin solution for a maximum of 24 hours at RT before being dehydrated and embedded in paraffin. Fixed tissue samples embedded in paraffin were sectioned to a 3-μm thickness, and slides were heated in the instrument at 75°C for 8 min and deparaffinized with EZ prep solution (Ventana Medical System, catalog no. 950-102 2L). Antigen retrieval was performed at 95°C for 64 min using the cell conditioning 1 buffer (Ventana Medical System, catalog no. 950-124 2L). Subsequent incubation of 8 min with CM inhibitor (ChromoMap DAP kit) was used for peroxidase blockade. For primary antibodies anti-Ki67 (1:250 dilution; Roche, #05278384001; RRID:AB_2631262) and anti-cleaved caspase-3 (Asp¹⁷⁵; 1:100 dilution; Cell Signaling Technology, #9661; RRID:AB_2341188), slides were first incubated at 37°C for 24 or 60 min, respectively, and for a further 8 min with UltraMap anti-rabbit antibody (horseradish peroxidase; Roche, #05269717001; PRID:-AB_2924783). As a detection system, a CM ChromoMap DAB kit (Roche, #760-159) was used according to the manufacturer’s instructions, followed by counterstaining with hematoxylin II (Ventana Medical System, #760-2021) for 8 to 12 min and bluing reagent (Ventana Medical System, #760-2037) for 4 min, dehydration, and mounting processes. Slides were scanned in the NanoZoomer 2.0-HT slide scanner (Hamamatsu Photonics) and visualized in the NDP.view2 software (Hamamatsu Photonics) or QuPath.

PDX studies

All mouse procedures were conducted in accordance with the Animal Research Reporting of In Vivo Experiments (ARRIVE) guidelines.

PDX ICC195 (IDH WT)

All experiments at Fred Hutchinson Cancer Center were conducted under protocol PROTO202000037 and approved by the Institutional Animal Care and Use Committee. Human PDX ICC195 was developed by implanting a resection of a IDH WT, primary ICC tumor from a patient (IRB-approved protocol #00001852). Tumor fragments with the size of ~1- to 2-mm³ were rinsed in RPMI and implanted subcutaneously into the right flanks of 6- to 8-week-old female NSG (NOD scid gamma) mice. When tumors reached ~100 to 200 mm³, mice were randomized into four groups for treatment with vehicle control, dasatinib (30 mg/kg), M2698 (10 mg/kg), or dasatinib (30 mg/kg) + M2698 (10 mg/kg) combo daily by oral gavage for 28 days. All drugs were dissolved in 100 mM citrate buffer (pH 3). Tumor growth and body weight were monitored two times a week. Tumor volumes were calculated using the formula: V = (length × width²)/2. PDX tumors were harvested at the end of treatment.

PDX62 (IDH1mut, R132C)

The animal procedures conducted at the Vall d’Hebron Institute of Oncology were approved by the Ethical Committee for the Use of Experimental Animals in accordance with the regulations of the Government of Catalonia. PDX62 was generated by subcutaneous implantation of a metastatic liver biopsy from a patient with ICC tumor (78). PDX62 tumor pieces (~3 to 4 mm) were subcutaneously implanted into the right flanks of 6- to 8-week-old female NOD. CB-17-Prkdc scid/Rj mice (Janvier Labs, RRID:MGI:3760616). Animals were housed in air-filtered flow cages with a 12:12 light/dark cycle, and food and water were provided ad libitum. Upon xenograft growth (150 to 200 mm³), PDX62-bearing mice were randomized into four groups and treated as described above. Mice were euthanized by CO₂ inhalation when tumors reached 1 to 1.5 cm³ or severe weight loss occurred, according to institutional guidelines.

Statistical analyses

Statistical significance was determined by specific tests and is presented as means ± SEM as indicated in the figure legends. Statistical analyses were performed using GraphPad Prism. Student’s two-tailed t test was used when comparing data from two groups, and one-way or two-way analysis of variance (ANOVA) was used when comparing more than two groups to determine significance, which was set at a P value of <0.05.