Oncogenic KRAS engages an RSK1/NF1 pathway to inhibit wild-type RAS signaling in pancreatic cancer

To identify interactors of mutant KRAS in PDAC and better understand mutant KRAS biology, we performed BioID screening in mouse PDAC cells. Notably, our studies focused on using BioID with KRAS^G12D, the most common mutation in PDAC. To label proteins proximal to KRAS in PDAC cells, we designed and expressed various BioID fusion proteins with BirA and KRAS4B, the ubiquitously expressed Kras splice isoform (Fig. 1A) (28). We fused both mutant (Kras^G12D) and WT murine Kras (Kras^WT) to the C terminus of myc-tagged BirA to create BirA-Kras^G12D (B-G12D) and BirA-Kras^WT (B-WT). Since the polybasic domain and CAAX motif at the C-terminal are both required for the membrane localization of KRAS, we designed a membrane-localized BirA control (B-CAAX) by fusing the last 20 amino acids of murine KRAS to BirA (48, 49). As the signaling and biological function of KRAS depend on its membrane localization (50, 51), we sought to identify membrane-specific KRAS^G12D interactors by creating a double mutant BirA-KRAS fusion protein, which harbors an additional mutation in the CAAX motif (B-G12D/C185S), as a nontransforming cytosolic comparator. To make the screen relevant to PDAC, we performed the experiment in three cell lines (mT42, mT93, and mT95) derived from FRT-LSL-Kras^G12V-FRT; LSL-Trp53^R172H; Pdx1-Cre;R26-FlpOERT2 (FPC) mouse PDAC tumors (SI Appendix, Fig. S1A) (52). The endogenous expression of KRAS^G12V in this mouse model and isolated cell lines is distinguishable from our exogenously expressed BirA-KRAS^G12D by mass spectrometry (MS) and can be excised upon 4-hydroxy-tamoxifen (4-OHT) treatment. This minimizes competition from the endogenous mutant KRAS that may confound the identification of BirA-KRAS interactors.

We first validated the localization of our panel of B-WT, B-G12D, B-CAAX, and B-G12D/C185S proteins in mT42 FPC cells by isolating membrane and cytoplasmic fractions using sequential fractionation (53). As expected, B-WT, B-G12D, and B-CAAX proteins localized to the membrane, whereas the CAAX mutant B-G12D/C185S localized to the cytoplasm (Fig. 1B). We validated these findings by myc-tag immunofluorescence stains of B-WT, B-G12D, and B-CAAX proteins, which displayed plasma membrane localization, and B-G12D/C185S protein, which was predominantly cytosolic in PDAC cells (Fig. 1C). Together, these experiments demonstrate the appropriate localization of the various fusion constructs.

To ensure that these BirA fusion proteins were functional, we performed Western blotting analyses of RAS effector pathways in 293T and 3T3 cells expressing these constructs. Expectedly, only B-G12D activated RAS effector pathways, ERK and AKT, in HEK293T and 3T3 cells, respectively (SI Appendix, Fig. S1 B and C). Additionally, among the BirA-KRAS fusion proteins, B-G12D was the most potent stimulator of cell proliferation and exclusively conferred anchorage-independent growth in 3T3 cells (SI Appendix, Fig. S1 D and E). To determine the functionality of the BirA ligase, we infected FPC cells to express the B-WT, B-G12D, B-CAAX, and B-G12D/C185S fusion constructs and treated the cells with biotin to determine whether the known RAS interactor RAF1 would be preferentially biotinylated by any of these fusion proteins. Indeed, B-G12D preferentially biotinylated more RAF1 than B-WT and B-CAAX but biotinylated comparable levels of RAF1 as B-G12D/C185S (Fig. 1D). This finding suggests that the interaction of RAF1 with KRAS depends on the GTP-bound state of KRAS rather than its localization and does not necessarily cause RAS pathway activation as has been previously reported (54). We also assessed the functional properties of these BirA constructs in FPC cells with and without the excision of the endogenous mutant KRAS. We found that steady-state activation of ERK and AKT pathways were not markedly altered in mT42 cells stably expressing these constructs (SI Appendix, Fig. S1F), suggesting the establishment of feedback loops. We also found that the expression levels of BirA-KRAS fusion proteins were comparable to or less than endogenous KRAS levels. Nonetheless, B-G12D increased proliferation and restored colony formation after Kras^G12V excision in mT93 FPC cells (SI Appendix, Fig. S1 G and H). These results provide additional evidence that the BirA-KRAS^G12D fusion retains oncogenic properties and that proximity labeling using these constructs could identify functionally relevant interactors.

Having characterized the functional aspects of our BirA-KRAS fusion proteins, we sought to compare our KRAS proximity labeling approach to the standard immunoprecipitation approach for identifying protein interactors of KRAS. To this end, we harvested protein lysate from biotin-treated B-WT– and B-G12D–expressing FPC cells and simultaneously performed a streptavidin pulldown and immunoprecipitation followed by Western blot analysis. Given the same input amount and conditions, both approaches were able to detect similar amounts of RAF1, whereas streptavidin pulldown was superior in detecting the p110-α subunit of PI3-Kinase (Fig. 1E). Stoichiometry may limit the immunoprecipitation approach such that only a fraction of enriched B-G12D molecules coprecipitates p110-α; however, in the proximity labeling approach, a single molecule of B-G12D may biotinylate several molecules of p110-α, increasing its detection by streptavidin pulldown. Because BioID was more sensitive in detecting the interaction of p110-α with KRAS than immunoprecipitation, we set out to employ this approach to identify interactors by MS.

To detect interactors of oncogenic KRAS, we performed BioID-MS analysis on three biological replicates of murine FPC tumor cells (mT42, mT93, and mT95). For each parental cell line, we generated cell lines stably expressing each of our BioID constructs (B-WT, B-G12D, B-CAAX, and B-G12D/C185S). The comparison of B-G12D to B-WT and/or B-G12D to B-G12D/C185S would enable the detection of mutant-specific and/or membrane-specific interactors, respectively (Fig. 2A). To determine whether endogenous mutant KRAS^G12V could affect the interactors identified, these cells were treated with 4-OHT to ablate the expression of Kras^G12V or dimethyl sulfoxide (DMSO) as a vehicle control for 3 d prior to biotin treatment, followed by lysate harvest, streptavidin pulldown, and MS analysis. To account for differences in BirA expression in each cell line and the variability in the number of peptides in each run for MS, we normalized the quantitative peptide abundance to the BirA counts for each experiment.

Several observed peptides validated this approach for both the vehicle-treated (Fig. 2B) and 4-OHT–treated cells (SI Appendix, Fig. S2A). MS detected shared KRAS peptides and also identified the G12-containing peptide across the three full-length BirA-KRAS variants. Interestingly, G12V peptides were not found in any of the samples run, suggesting that either the BirA tag prevents KRAS^G12D:KRAS^G12V dimer formation or that such heterodimers may be too scarce to detect. Additionally, samples with an intact CAAX box enriched for membrane-localized proteins, such as FAM129B (also known as niban-like protein 1 or MINERVA), compared to samples from cytosolic B-G12D/C185S. B-G12D and B-G12D/C185S had similar biotinylation levels of RAF1, whereas B-WT and B-CAAX labeled less RAF1, consistent with our Western blot results (Fig. 1D).

To identify KRAS^G12D effectors, we compared the proteins enriched by B-G12D and B-WT from our BioID-MS experiments and nominated 91 protein matches that met the criteria of being enriched in at least two out of three biological replicates in the setting of the endogenous mutant Kras^G12V allele. To further prioritize this list, we also queried for membrane-specific candidates identified by comparing B-G12D to B-G12D/C185S and found 255 proteins in at least two out of three biological replicates. Among them, 66 candidate proteins were both mutant specific and membrane specific (SI Appendix, Table S1 and Fig. S2B). Interestingly, when sorted by membrane enrichment, ARAF was preferentially biotinylated at the membrane whereas RAF1 and BRAF were both comparably labeled by B-G12D and B-G12D/C185S (SI Appendix, Table S2 and Fig. S2C). Although the direct interaction of KRAS and ARAF is well established, these data suggest that our approach could map spatially distinct proximal interactors and uncover dynamic KRAS^G12D interactors on the membrane.

To further refine our list and prevent confounding effects of endogenous mutant KRAS due to competition for substrates, we also performed BioID-MS analysis on the same BirA-KRAS–expressing FPC cells treated with 4-OHT to excise the endogenous mutant Kras^G12V allele (SI Appendix, Fig. S2D). By prioritizing candidates that were nominated in the mutant-specific and membrane-specific comparison for both the vehicle and 4-OHT conditions, we further narrowed our list to 32 proteins (Fig. 2C). Of these proteins, 11 matches were known interactors or KRAS effectors, including RAF family members (ARAF, BRAF, and RAF1), SPRED1, SPRED2, NF1, MLLT4, RIN1, and mTORC2 complex members (MAPKAP1, RICTOR, and DEPTOR) (Fig. 2D and SI Appendix, Table S3). Of the remaining 21 candidates, PLEKHA2, MYO1E, and RPS6KA1 (RSK1) were also enriched in the B-G12D to B-WT comparison as previously reported in BioID screens in HEK293 cells (47, 55). Several known RAS interactors, including PI3K, KSR, RALBP, SOS1, and PIP5KA1, were not detected in our MS data set, suggesting that additional approaches may be needed to increase the sensitivity of BioID assay in PDAC cells. Nonetheless, ARAF and RSK1 were the most consistently enriched candidates in the B-G12D to B-WT comparison in the 4-OHT setting.

Of the interactors that were nominated to be both KRAS^G12D specific and plasma membrane localized, RICTOR, ARAF, and RSK1 were among those validated by Western blot analysis of the BioID samples (Fig. 2E). To validate the RSK1–KRAS interaction and determine whether it is stable and therefore amenable to immunoprecipitation, we performed myc-tag coimmunoprecipitation in parallel with streptavidin pulldown of lysates from FPC cells expressing either B-WT or B-G12D. Unlike ARAF, RSK1 was only detected by streptavidin pulldown of B-G12D samples and failed to appreciably enrich by immunoprecipitation (Fig. 2F and SI Appendix, Fig. S2E), suggesting that the RSK1–KRAS^G12D interaction is transient, weak, or indirect.

Unlike RAFs and the mTORC2 complex, RSK1 lacks a known RAS-binding domain (RBD). Therefore, we reasoned that the proximity of RSK1 to oncogenic KRAS may depend upon another protein in murine PDAC cells, similar to the dependency of RICTOR on MAPKAP1 for proximity to KRAS (47). To test this hypothesis, we employed both BioID and CRISPR/Cas9 knockout (KO) approaches. BioID enables the detection of transient or weak KRAS^G12D interactors, which are difficult to identify by immunoprecipitation, whereas the selective ablation of proteins in the KRAS^G12D biotinylated interactome enables the identification of proteins that promote the proximity of RSK1 to KRAS^G12D on the membrane. Accordingly, we investigated whether ablating membrane-specific proximal interactors of B-G12D, such as MAPKAP1, RICTOR, and NF1, would abolish the proximity biotinylation of RSK1 by B-G12D in mT42 and mT93 cells. As expected, MAPKAP1 KO decreased the KRAS^G12D-mediated biotinylation of RICTOR; however, the depletion of neither mTORC2 component decreased RSK1 biotinylation (Fig. 3A). Interestingly, only NF1 depletion abrogated RSK1 biotinylation, suggesting that NF1 selectively mediates RSK1 interaction with oncogenic KRAS (SI Appendix, Fig. S3 A and B). Since NF1 loss promotes WT RAS activation, competition for RSK1 between WT RAS and mutant KRAS could provide an alternative explanation for the decreased RSK1–KRAS^G12D interaction upon NF1 KO. To exclude this possibility, we excised the endogenous mutant KRAS in FPC cells expressing either B-G12D or B-WT and then ablated Nf1 with CRISPR/Cas9. We hypothesized that if WT KRAS could compete for RSK1 binding, then the complete ablation of oncogenic KRAS in B-WT–expressing cells would lead to elevated RSK1 biotinylation. On the contrary, we found that RSK1 was biotinylated in B-G12D but not B-WT–expressing cells (SI Appendix, Fig. S3B), suggesting that the decrease in RSK1–KRAS^G12D interaction upon NF1 loss cannot be explained by WT KRAS activation. Since RSK1 requires NF1 to interact with KRAS^G12D on the membrane, and SPRED proteins are essential for the membrane recruitment of NF1 to the RAS interactome (56, 57), we tested whether KO of Spred1 and Spred2 would decrease the RSK1–KRAS^G12D interaction, measured by RSK1 biotinylation. Although we could not confirm SPRED2 depletion at the protein level due to the lack of suitable antibodies, we validated successful gene editing at the target sites of both guide RNAs against Spred2 (SI Appendix, Fig. S3C). Whereas guides directed against Spred1 slightly decreased NF1 biotinylation, Spred2 but not Spred1 depletion abolished RSK1 biotinylation (Fig. 3C and SI Appendix, Fig. S3D), suggesting that a SPRED2/NF1 membrane complex bridges the interaction between RSK1 and oncogenic KRAS. In line with a mutant KRAS interactome comprising RSK1, SPRED2, and NF1, these proteins were nominated as the top 13th, 14th, and 20th interactors specific to KRAS^Q61H in the human cell map dataset (55). Taken together, our results demonstrate that the proximity of RSK1 to KRAS^G12D depends on NF1 and SPRED2 and suggest a function for these negative regulators of RAS signaling.

RSK1 to 4 are a family of serine/threonine kinases known to be activated by ERK signaling and translocated to the plasma membrane transiently and in an ERK-dependent manner upon EGF stimulation (58, 59). To test whether MEK/ERK pathway activation regulates the interaction between RSK1 and mutant KRAS, we treated B-G12D–expressing FPC cells with 500 nM of the MEK inhibitor trametinib or vehicle control during the biotin incubation of the BioID assay and performed streptavidin pulldown and Western blot for RSK1. MEK inhibition markedly decreased phospho-ERK levels and abrogated RSK1 biotinylation by B-G12D (Fig. 3D), suggesting that the RSK1–KRAS^G12D interaction depends on MEK/ERK pathway activation. To determine whether the abrogation of KRAS^G12D-mediated RSK1 biotinylation is a secondary effect of suppressing cellular proliferation rather than a specific effect of inhibiting the MEK/ERK pathway, we evaluated the proliferation of mT42 G12D-expressing cells upon treatment with 20 nM gemcitabine, a cytotoxic drug, or 5, 50, 200, and 500 nM trametinib. As expected, gemcitabine suppressed the proliferation of these cells. Similarly, trametinib inhibited their proliferation in a concentration-dependent manner (Fig. 3E). Having confirmed the effects of these drugs on cellular proliferation, we repeated the BioID experiment with increasing doses of trametinib, vehicle control, or 20 nM gemcitabine. Western blot analyses confirmed the loss of RSK1 biotinylation upon the inhibition of ERK activation, whereas RSK1 biotinylation was mostly unchanged in gemcitabine-treated cells (SI Appendix, Fig. S3E), thereby confirming that the loss of RSK1 biotinylation upon MEK inhibition is a direct consequence of MEK/ERK pathway inhibition. Since ERK-mediated phosphorylation of RSK1 has been previously linked to its membrane translocation (59), we sought to determine whether a decrease in RSK phosphorylation by ERK could explain the loss of RSK1 biotinylation upon MEK inhibition. To this end, we treated mT42 cells with 50 nM trametinib and performed Western blot analysis for phospho-RSK (S380). Phospho-RSK decreased upon MEK inhibition, supporting a model in which MEK/ERK pathway inhibition reduces RSK phosphorylation and regulates its ability to interact with NF1 (SI Appendix, Fig. S3F). Since ERK activation induces the expression of negative regulators of the RAS pathway, including SPRED, SPROUTY, and DUSP family proteins, we asked whether ERK deactivation via MEK inhibition decreases SPRED1/2 expression, thereby preventing the membrane localization and recruitment of the NF1/RSK1 complex to the mutant KRAS interactome. Indeed, we found that DUSP6, SPRED1, and SPRED2 were all potently down-regulated following MEK inhibition in mT42 FPC cells in a time- and concentration-dependent manner (SI Appendix, Fig. S3G), providing an additional explanation for the loss of RSK1 biotinylation by oncogenic KRAS in the context of MEK inhibition.

Having established an active KRAS interactome including NF1 and RSK1, we sought to determine a functional role for membrane-localized RSK1 in PDAC cells. RSK1 is a known ERK pathway effector, and membrane-targeted RSK1, considered to be constitutively active, can confer MEK independence and mediate pro-survival signals in IL-3–dependent cells (60). However, in other cellular contexts, membrane-localized RSK1 has also been shown to inhibit RAS signaling (61). As KRAS localizes to the plasma membrane, we used a myristoylation (myr) tag to target RSK1 to the plasma membrane (62) (SI Appendix, Fig. S4A). To identify whether membrane-targeted RSK1 regulates the RAS pathway or contributes to RAS-mediated transformation in PDAC cells, we retrovirally introduced RSK1 or myr-tagged RSK1 into parental mT42 cells and derivative cell lines stably expressing our tagged KRAS variants, B-WT, B-G12D, and B-G12D/C185S. For comparison, we also included overexpression of ARAF and myr-tagged ARAF as well as an empty vector control. These cells were plated sparsely and treated with 4-OHT for 3 d to excise endogenous Kras^G12V before crystal violet staining to measure foci formation. Expectedly, mT42 cells expressing B-G12D formed colonies normally despite the removal of endogenous Kras^G12V, whereas the parental cells and cells with overexpression of B-WT and B-G12D/C185S were unable to rescue this loss and therefore exhibited less colonies (SI Appendix, Fig. S4 B–D). Myr-tagged ARAF, but not untagged ARAF, was able to partially rescue the loss of KRAS^G12V, consistent with the observation that membrane-targeted ARAF does not require active KRAS to activate downstream signaling (63). In contrast, myr-RSK1 further inhibited colony formation of mT42 parental cells and cells with either B-WT or B-G12D/C185S following the excision of Kras^G12V. Similarly, compared to empty vector, RSK1, or kinase-dead RSK1 expression, myr-RSK1 expression in mT42 FPC cells markedly attenuated proliferation following Kras^G12V excision (Fig. 4A, Top) but had no effect on cellular fitness when endogenous Kras^G12V was present (Fig. 4A, Bottom). We further validated this finding in mT93 cells (SI Appendix, Fig. S4E) and then sought to dissect the molecular underpinnings of this finding. Since membrane-targeted RSK produces a growth suppression phenotype only in cells lacking oncogenic KRAS, we hypothesized that this phenotype depends on the inhibition of WT RAS signaling.

To determine whether membrane-targeted RSK1 suppresses cellular proliferation by inhibiting the RAS pathway, we introduced empty vector, RSK1, kinase-dead RSK1, or myr-RSK1 to mT42 cells and treated the cells with either 4-OHT or DMSO control. We confirmed that these constructs were correctly expressed and functional by immunoblotting (Fig. 4B). As expected, cells expressing RSK1 or myr-RSK1 showed increased activation of RSK1 substrates such as S6 and EPHA2. There were no notable differences in the phosphorylation levels of ERK and AKT in the presence of endogenous mutant KRAS. However, upon mutant KRAS abrogation, phospho-ERK levels were markedly reduced in myr-RSK1 compared to kinase-dead RSK1 and empty vector–expressing cells. Importantly, mT42 empty vector–expressing cells treated with 20 nM gemcitabine exhibited no attenuation of ERK activation, indicating that the reduction of RAS signaling induced by myr-RSK1 is not a consequence of reduced proliferation. We also performed these experiments in mT93 cells and obtained similar results (SI Appendix, Fig. S4F). Collectively, these findings suggest that membrane-targeted RSK1 selectively exerts negative feedback on WT RAS signaling.

To extend these findings to human PDAC, RSK1 and myr-RSK1 were ectopically expressed in MIA PaCa-2 cells, which harbor a KRAS^G12C mutation that can be targeted by the commercially available KRAS^G12C-specific inhibitor AMG 510 (64). To test whether membrane-localized RSK1 might affect cell proliferation as a function of the observed changes in RAS signaling pathways, we cultured RSK1-, myr-RSK1-, or empty vector–expressing MIA PaCa-2 cells with DMSO or 100 nM AMG 510. Whereas overexpression of RSK1 and myr-RSK1 had no effect on the proliferation of mT42 cells treated with DMSO, cells expressing these constructs, particularly myr-RSK1, exhibited increased sensitivity to 100 nM AMG 510 (Fig. 4C). To further validate these results using various concentrations of AMG 510, we performed dose curves in MIA PaCa-2 cells expressing empty vector, RSK1, and myr-RSK1 and similarly found that the blockade of oncogenic KRAS with AMG 510 sensitized these cells to RSK1 and myr-RSK–induced growth suppression (SI Appendix, Fig. S4G). This effect could not be induced by 20 nM gemcitabine (SI Appendix, Fig. S4H, Top), suggesting that AMG 510 elicits RSK1-mediated growth suppression through the blockade of oncogenic RAS pathway, rather than through its secondary effects on cellular fitness. To determine whether the attenuation of effector pathways downstream of oncogenic KRAS modulates the sensitivity to RSK1 expression, we treated MIA PaCa-2 cells expressing control and RSK constructs with 20 nM trametinib and performed proliferation assays. Interestingly, myr-RSK1 but not untagged RSK1 reduced proliferation following trametinib treatment (SI Appendix, Fig. S4H, Bottom). As the growth-suppressive effect of RSK1 occurs exclusively at the membrane, and membrane-targeted but not untagged RSK1 can bypass cellular control of its membrane localization, this result suggests that direct MEK inhibition has a greater effect on preventing RSK1 membrane localization compared to upstream oncogenic KRAS inhibition. Next, we sought to confirm whether membrane RSK1 inhibits RAS effector pathways following oncogenic KRAS blockade in MIA PaCa-2 cells in a similar manner to our prior findings in FPC cells. To this end, we treated MIA PaCa-2 cells expressing control and RSK constructs with 100 nM AMG 510 and harvested protein lysates for Western blot analyses at 1 h, 1 d, and 3 d after drug treatment. Remarkably, RSK1 and myr-RSK1 attenuated ERK activation especially following 1 d of AMG 510 treatment (SI Appendix, Fig. S4I). These findings nominate RSK1 as an interactor of mutant KRAS that bridges the downstream activation of ERK signaling to the upstream negative regulator of WT RAS, NF1. This suggests that RSK1/NF1 participates as a dual negative feedback inhibitor of WT RAS, consistent with previously described RSK1-mediated negative feedback that involves NF1 (65) and SOS1 (61). The resistance of PDAC cells to membrane-targeted RSK1 may reflect the indifference of oncogenic KRAS to NF1 GAP activity and less dependence on SOS1-mediated GTP loading compared to WT RAS.

Given the specificity of membrane RSK1 to WT RAS and because RSK1 has been previously shown to feedback inhibit RAS/ERK signaling through the inhibition of SOS1 or through its activation of NF1 (61), we asked whether WT RAS inhibition through selective pharmacological inhibition of SOS1 could phenocopy the effects of membrane-targeted RSK. Accordingly, we treated MIA PaCa-2 cells with a combination AMG 510, to abrogate oncogenic KRAS signaling, and a selective SOS1 inhibitor, BI-3406, to block WT RAS signaling. Although MIA PaCa-2 cells have undergone loss of heterozygosity and do not possess WT KRAS, this cell line expresses other RAS isoforms, HRAS and NRAS (SI Appendix, Fig. S4J), which are amenable to mutant KRAS-induced negative feedback inhibition and have been recently shown to exhibit increased activation following KRAS^G12C inhibition (66, 67). Indeed, we found that BI-3406, a selective SOS1 inhibitor, enhanced the AMG 510-induced growth suppression in MIA PaCa-2 cells (Fig. 4D and SI Appendix, Fig. S5A). We also obtained similar results using NCI-H358, a KRAS^G12C non–small cell lung cancer model (SI Appendix, Fig. S5B, Left). Expectedly, SUIT2 PDAC cells with KRAS^G12D were completely resistant to the drug combination (SI Appendix, Fig. S5B, Right). SOS1 inhibition may impair GDP-to-GTP KRAS^G12C cycling, thereby increasing the efficacy of AMG 510 through increased availability of its substrate rather than direct inhibition of WT RAS (68). To validate the effect of SOS1 inhibition on WT RAS specifically, we employed our FPC model, which precludes the contribution of mutant KRAS GTP loading through genetic deletion of KRAS^G12V. Kras^G12V deletion combined with SOS1 inhibition abrogated the proliferation of FPC cells (Fig. 4E and SI Appendix, Fig. S5C), thereby supporting the notion that the blockade or loss of oncogenic KRAS eliminates negative feedback signaling on WT RAS. Notably, excision of mutant RAS modestly enhanced the effect of trametinib in these cells (SI Appendix, Fig. S5D). The efficacy of SOS1 inhibition upon the loss of endogenous mutant KRAS implicates WT RAS in mediating the adaptive response to oncogenic KRAS-directed therapies and suggests combination strategies involving dual targeting of oncogenic KRAS and WT RAS in KRAS-driven cancers. Finally, we sought to determine whether, similarly to membrane-targeted RSK, pharmacological blockade of WT RAS signaling would attenuate downstream RAS effector pathways upon ablation of oncogenic KRAS. Indeed, treatment with 5 µM BI-3406 following the excision of endogenous mutant KRAS abolished both ERK and AKT activation, consistent with the inhibition of WT RAS signaling (Fig. 4F). Our findings collectively provide support for a model in which oncogenic KRAS elicits negative feedback on WT RAS through NF1/RSK1 (Fig. 5A). The loss of this feedback signaling through oncogenic KRAS deletion or blockade disinhibits WT RAS and provides a mechanism for adaptive resistance to oncogenic KRAS-directed therapies (Fig. 5B).

The BirA R118G biotin-protein ligase gene including the multiple cloning site was cloned from the pcDNA3.1 mycBioID plasmid deposited by the Roux laboratory. Murine Kras was inserted between the multiple cloning site of BirA R118G and SalI with GGCGGAAGCGGA, encoding for a short glycine linker (GGSG). The C185S mutation was made with the Q5 Site-Directed Mutagenesis Kit. BirA-CAAX control was created by fusing the last 20 amino acids of murine KRAS4B to the C terminus of BirA. The constructs were moved to pBABE-neo between BamHI and SalI for retroviral production. For RSK1 overexpression, human RSK1 complementary DNA (cDNA) was obtained from John Blenis via Addgene. The sequence (ggg​agt​agc​aag​agc​aag​cct​aag​gac​ccc​agc​cag​cgc) was added after the start codon of the RSK1 cDNA to fuse a myr-tag to the N terminus. Both RSK1 and myr-RSK1 were cloned into pBABE-neo between EcoRI and SalI. For gene editing, lentiCas9-Blast (addgene no. 52962, deposited by the Feng Zhang laboratory) was used (78). Guides were designed using the Benchling CRISPR tool against the first coding exon of each target gene with the best off-target scores inserted into the ipUSEPR lentiviral vector (SI Appendix, Table S4). Kinase-dead RSK plasmid was constructed by PCR-based cloning. Using p-BABE-RSK plasmid as a template, the primer containing the mutations (CAT​CTT​GGT​CCG​GAC​GCG​GTC​ACG​TAC​TTT​CAG​CGT​TGC​CTT​CTT​CAG​CAC​GCG​CAT​AGC​ATA​CAG​GTG​CCC​AC) and 5′ pBABE-RSK1 primer were used to amplify the PCR products. The PCR products digested with EcoRI/BPEBI and MfeI/BSM were subsequently subcloned into p-BABE-hRSK vector digested with EcoRI/BPEBI and MfeI/BSMI, respectively. The mutations were confirmed by Sanger sequencing.

Phoenix, 293T, 3T3, mT42, mT93, mT95, and SUIT2 cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) with 10% fetal bovine serum (FBS) at 37 °C in 5% CO₂ incubator. MIA PaCa-2 cells were cultured in Roswell Park Memorial Institute Medium (RPMI 1640) with 10% FBS. Cells for BioID experiments had their media changed to serum-free DMEM with 50 μM biotin for 24 h prior to cell harvest.

Cells were washed five times with ice-cold phosphate-buffered saline (PBS) prior to lysis with mild buffer (50 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), 150 mM NaCl, 0.7% Nonidet P-40, 10% Glycerol, 1mM ethylenediaminetetraacetic acid (EDTA) with Roche cOmplete Mini protease inhibitor tablet and Roche PhoSTOP phosphatase inhibitor. Lysates were spun at 16,900 × g for 10 min. The detergent-compatible (DC) Protein Assay (Bio-Rad) was used to quantify the protein concentration. For experiments using FPC cells, 2 μM of 4-hydroxytamoxifen (4-OHT) with DMSO as the vehicle was added to the media for 3 d prior to harvest. AMG 510 (Selleckchem, no. S8830) dissolved in DMSO was added to RPMI for 24 h or as indicated in MIA PaCa-2 experiments. Buffers and protocol for sequential cell fractionation experiments were followed as laid out in Baghirova et al. (53)

Cell lysates were harvested in mild Nonidet P-40 buffer with protease and phosphatase inhibitor as described above (50mM HEPES, 150 mM NaCl, 0.7% Nonidet P-40, 10% Glycerol, 1mM EDTA) with Roche cOmplete Mini protease inhibitor tablet and Roche PhoSTOP phosphatase inhibitor, at 0 °C. A total of 5 mg lysate in 1 mL total buffer was incubated with 30 μl MyOne Streptavidin C1 Dynabeads or 5 mg Myc-Tag (9B11) Mouse mAb (Cell Signaling, no. 2276) antibody for 3 h with subsequent 1-h incubation with 30 μl Protein G Dynabeads for streptavidin pulldown and myc-tag immunoprecipitation (IP), respectively. The beads were washed five times in lysis buffer before elution in NuPAGE 3-(N-morpholino)propanesulfonic acid (MOPS) sodium dodecyl sulfate (SDS) running buffer at 95 °C.

On-bead tryptic digestion of Dynabeads, liquid chromatography with tandem mass spectrometry (LC-MS/MS), and database searching is detailed in SI Appendix. Proteins quantified by only one peptide were omitted from further analysis. All quantification of peptides was normalized to that of BirA prior to producing relative enrichment by ratiometric analysis. The proteins were ranked by relative enrichment for each biological replicate. The sum of the three assigned ranks was used to indicate a general enrichment score for all three biological replicates to account for protein expression differences across cell lines. The candidate interactors were then sorted by their cumulative rank scores to identify consistently enriched interactors. Candidate interactors met their respective criteria if they were enriched in two of three biological replicates.

Mouse cell lines (3T3 cells and FPC cells) were plated at 1,000 cells per well while human cell lines (MIA PaCa-2, SUIT-2, and NCI-H358) were plated at 2,000 cells per well. Five wells were used for technical replicates. Pharmacological agents were added by HP D300 drug dispenser, and cell viability was assayed daily using the CellTiter-Glo Assay and read on a Spectramax i3 plate reader. For crystal violet staining, cells were plated in 6-well plates at a density of 5,000 cells per well and allowed to grow for 2 d or 2,000 cells per well for 3 d. The cells were washed with ice-cold PBS twice before methanol fixation for 20 min. These cells were then stained with a crystal violet solution (0.5% crystal violet, 95% ethanol in water).

Microsoft Excel and GraphPad Prism were used for graphical representation of data. Statistical analysis was performed using Student’s t test.