Atypical KRAS^G12R Mutant Is Impaired in PI3K Signaling and Macropinocytosis in Pancreatic Cancer

Mutational activation of the KRAS proto-oncogene is the predominant oncogenic alteration in the top three major cancers (lung, colorectal, and pancreatic), ranked by mortality, in the United States (1). Thus, effective anti-KRAS therapeutic strategies are expected to make a significant impact on cancer treatment. To date, there are no effective pan–KRAS-mutant therapies (2). However, with inhibitors specifically targeting the glycine-to-cysteine (G12C) missense mutation now entering clinical evaluation (3), an emerging premise is that KRAS mutation–selective therapies may be feasible (4). Cancer-associated mutations in KRAS cluster in one of three hotspots, with a majority (84%) of mutations causing single amino-acid substitutions at G12 (1). Of the six possible single-base missense mutations that can occur at G12, G12D is the most predominant (42%), followed by G12V, G12C, G12A, G12S, and G12R, the last of which occurs infrequently in most cancers.

However, although the KRAS^G12R mutation is rare in lung and colorectal cancers (∼1%), it is the third most common KRAS mutation in pancreatic ductal adenocarcinoma (PDAC; ref. 5). Recent The Cancer Genome Atlas (TCGA) analyses of PDAC suggested that KRAS^G12R-mutant cancers can harbor additional KRAS mutations, implying that KRAS^G12R may be functionally distinct from other KRAS driver mutations (6). Unexpectedly, we found that KRAS^G12R fails to bind the PI3K catalytic subunit p110α, an essential effector for KRAS-driven cancer initiation and maintenance (7), resulting in KRAS-independent macropinocytosis, a nutrient uptake process that has been shown to be necessary for PDAC tumor growth (8). Using X-ray crystallography, we found that KRAS^G12R, unlike KRAS^G12D, disrupts the switch II (SII) region critical for effector interaction. Instead, KRAS-independent p110γ PI3K activity supports macropinocytosis in KRAS^G12R-mutant PDAC. Finally, we addressed the possibility that KRAS^G12R- and KRAS^G12D-mutant PDAC may exhibit distinct therapeutic vulnerabilities. KRAS^G12R-mutant PDAC showed limited preferential sensitivity to MEK/ERK inhibitor monotherapy, with only a modest difference in comparison with PDAC models harboring the more common KRAS^G12D/V mutants. However, using drug sensitivity and resistance testing (DSRT) to probe 525 different inhibitors, we observed that KRAS^G12R-mutant PDAC was also sensitive to autophagy inhibitors. We found that the recently reported combination of ERK/MAPK inhibition and the autophagy inhibitor chloroquine can be a rapidly translatable, effective therapeutic strategy for patients harboring these mutations. Thus, we have demonstrated structurally and functionally distinct properties for KRAS^G12R that lead to alternative activation of macropinocytosis through PI3Kγ and enhanced sensitivity to combined ERK/MAPK pathway and autophagy inhibition.

Despite its status as the third most prevalent KRAS mutation in PDAC, after G12D and G12V (Supplementary Fig. S1A), KRAS^G12R has been poorly studied. Given the role of oncogenic KRAS in altering metabolic activities to support cancer growth (9), we first sought to determine whether KRAS^G12R plays a role similar to the more common KRAS mutants in driving metabolic perturbations. Accordingly, we examined macropinocytosis, a nutrient-scavenging process shown to sustain PDAC tumor growth (10). Upon quantifying macropinocytosis activity by uptake of FITC-tagged dextran, we detected variable levels of macropinocytosis in our panel of 10 KRAS-mutant PDAC cell lines (Fig. 1A). In agreement with Commisso and colleagues (10), transient siRNA suppression of KRAS reduced macropinocytosis in KRAS^G12D-, KRAS^G12V-, and KRAS^G12C-mutant cell lines (Fig. 1B–D). Surprisingly, suppression of KRAS did not reduce macropinocytosis in the KRAS^G12R-mutant PDAC lines (Fig. 1B–D), indicating that they displayed KRAS-independent macropinocytosis. This unexpected finding raised the possibility that the growth of KRAS^G12R-mutant PDAC may also be KRAS-independent. However, as we have described previously for KRAS^G12D/V/C-mutant PDAC cell lines (11), we observed that both anchorage-dependent and anchorage-independent in vitro growth of the KRAS^G12R-mutant cell lines were also KRAS-dependent (Fig. 1E and F). Furthermore, loss of each KRAS mutant was associated with increased apoptosis and impaired cell-cycle progression (Supplementary Fig. S1B and S1C).

We applied two strategies to determine if KRAS^G12R could stimulate macropinocytosis in other cellular contexts. First, we ectopically expressed different KRAS mutants in three cell models sensitive to mutant KRAS-driven growth transformation (12, 13) and quantified macropinocytosis by FITC-dextran uptake. Stable expression of G12D or G12V, but not G12R, increased macropinocytosis in RIE-1 rat intestinal epithelial cells (Fig. 2A–C) in a time-dependent manner (Supplementary Fig. S2A and S2B). Similar results were observed in NIH/3T3 mouse fibroblasts (Supplementary Fig. S2C–S2E) and hTERT-immortalized HPNE human pancreatic duct–derived epithelial cells (Supplementary Fig. S2F–S2H).

We also assessed macropinocytosis using BODIPY-conjugated BSA (DQ-BSA), which has a fluorescent signal that is quenched until it is released upon proteolytic degradation following macropinosome fusion with lysosomes (10). Concordant with the results obtained with FITC-dextran, stable expression of KRAS^G12D/V stimulated macropinocytosis relative to empty vector and KRAS^G12R in RIE-1 cells (Fig. 2D). Thus, although KRAS^G12R could cause the same morphologic transformation of RIE-1 cells as KRAS^G12D/V (Supplementary Fig. S2I), KRAS^G12R uniquely failed to stimulate macropinocytosis.

Our second strategy was to determine whether exogenous KRAS^G12R could rescue macropinocytosis in siRNA-suppressed KRAS^G12D/V-mutant PDAC cell lines. However, exogenous expression of mutant KRAS in KRAS-mutant PDAC cells is growth inhibitory, and stable suppression of endogenous mutant KRAS can lead to confounding compensatory activities (14). To overcome these technical issues, we used a doxycycline-inducible expression vector to transiently express exogenous mutant KRAS^G12R concurrently with KRAS siRNA oligo treatment to transiently silence endogenous KRAS^G12D in AsPC-1 cells (Fig. 2E–G) and KRAS^G12V in Pa04C cells (Fig. 2H–J). Ectopic expression that restored the same mutant as endogenous KRAS was able to rescue macropinocytosis, whereas ectopic expression of KRAS^G12R failed to restore KRAS^G12D/V-dependent PDAC macropinocytosis.

Previously, KRAS^G12R has been reported to have decreased intrinsic GTPase activity and be insensitive to GAP-mediated hydrolysis, similar to other KRAS^G12 mutants (15, 16). However, sensitivity to guanine exchange factors (GEF) had been unexplored. As SOS1 is considered the main RASGEF, we measured the k_cat of exchange with KRAS wild-type (KRAS^WT) and G12 mutants using the catalytic domain of SOS1 (SOS^cat). Similar to what has been described for HRAS^G12V (17), we found that SOS^cat activity toward KRAS^G12V was reduced 20% (0.01653 μM⁻¹s⁻¹) and 60% toward KRAS^G12D (0.007982 μM⁻¹s⁻¹) compared with KRAS^WT (0.01986 μM⁻¹s⁻¹). In contrast, SOS^cat was inactive toward KRAS^G12R (Fig. 3A). To determine if this impaired activity was due to loss of allosteric interaction or catalysis, we evaluated the activity of the CDC25 catalytic fragment of a different RASGEF, RASGRP1. Although its CDC25 domain has a similar catalytic mechanism as the SOS1 CDC25 domain, the RASGRP1 REM-CDC25 catalytic fragment (RASGRP1^cat) lacks the requirement for allosteric-mediated RAS activation (18). KRAS^G12R was activated equivalently to KRAS^WT by RASGRP1^cat (Fig. 3B), indicating that KRAS^G12R fails to interact with the allosteric binding site of SOS1.

As the loss of binding to the allosteric domain of SOS^cat suggested a large structural change in KRAS^G12R, we determined the structure of human KRAS4B^G12R (residues 1–169) bound to the GTP analogue GMPPNP at a resolution of 1.5 Å (Fig. 3C; Table S1). Both switch I (SI; residues 30–40) and SII (residues 60–76) undergo conformational changes during RAS GDP–GTP cycling and are critical for interaction with effectors and regulators (19). In the structure of GMPPNP-bound KRAS^G12R, SI is in the same conformation as in GMPPNP-bound RAS^WT. However, in KRAS^G12R, helix α2 in SII is partially unfolded and contains only one helical turn instead of four. The G12R mutation alters SII by displacing Q61 and forming a direct interaction with E62 and T35, a residue that interacts with the Mg²⁺ ion and γ-phosphate in the GTP-bound state (Fig. 3D). An overlay of the crystal structures of GMPPNP-bound KRAS^G12R and KRAS^G12D (20) shows the distinct structural alterations in SII induced by the G12R mutation (Fig. 3E and F).

The distribution of specific G12 mutations differs across RAS isoforms (COSMIC v89). For example, G12V represents 50% of all HRAS^G12 mutations, whereas G12D is much less common in HRAS than in KRAS (18% vs. 42%), and G12R represents just 3% of all HRAS^G12 mutations in cancer. Thus, we wondered whether specific mutations may have isoform-specific consequences. Comparison with a structure of GMPPNP-bound HRAS^G12V (Supplementary Fig. S3A and S3B; ref. 21) shows that HRAS^G12V and KRAS^G12D are similar, where the valine and aspartic acid side chains point away from the nucleotide-binding pocket. This is in contrast to KRAS^G12R, where the arginine side chain points directly into the pocket. Thus, HRAS^G12V and KRAS^G12D but not KRAS^G12R retain an ordered SII region (Fig. 3G and H). Comparison of our GMPPNP-bound KRAS^G12R structure with a previously determined GMPPNP-bound HRAS^G12R structure (Supplementary Fig. S3C; ref. 16) shows similar side chain conformations and interactions formed by R12, resulting in equivalent disruptions in SII and Q61/E62, and no significant isoform-specific structural differences were detected. Further, no differences in thermal stability between the G12 mutants and KRAS^WT were observed, as measured by monitoring nucleotide binding as a function of temperature (Supplementary Fig. S3D).

As we observed a decrease in SOS^cat binding to KRAS^G12R, we hypothesized that KRAS^G12R would exhibit altered effector specificity, potentially accounting for the inability of KRAS^G12R to stimulate macropinocytosis. First, we utilized well-characterized RAS mutants (T35S, E37G, and Y40C) that are differentially impaired in interaction with the major RAS effectors (22–24) to determine which pathways were critical for RAS-driven macropinocytosis. However, KRAS^G12D and KRAS^G12V carrying comutations at any of the three effector-binding domain residues failed to stimulate macropinocytosis in RIE-1 cells (Supplementary Fig. S4A–S4C), suggesting that multiple effector pathways are involved in driving KRAS-dependent macropinocytosis.

Second, we utilized reverse-phase protein array (RPPA) analyses (25) to profile RIE-1 cells overexpressing KRAS^G12 mutants, providing an isogenic background to evaluate KRAS signaling. Surprisingly, the only statistically significant effector signaling defect observed in the G12R-expressing cells was in the PI3K–AKT pathway (Fig. 4A; Supplementary Fig. S4D). Phosphorylation of AKT and AKT substrates (PRAS40, MDM2, FKHR, and S6K) was decreased in cells expressing KRAS^G12R relative to G12D or G12V (Fig. 4B; Supplementary Fig. S4E). We verified the RPPA results by immunoblot analysis and found that both basal and serum-stimulated activation of AKT were elevated (30%–59%) in RIE-1 cells stably expressing KRAS^G12D or KRAS^G12V compared with control cells, whereas it was suppressed in KRAS^G12R-expressing cells (Fig. 4C and D). In contrast, MEK–ERK activation was elevated similarly in cells expressing each of the three mutants (Fig. 4E).

The decreased phosphorylation of AKT and its downstream substrates in KRAS^G12R-expressing RIE-1 cells suggested that KRAS^G12R may be impaired in binding to and/or activating the catalytic subunit of PI3Kα (p110α), an upstream activator of AKT and validated driver of macropinocytosis (26). To directly address this possibility, we utilized a well-characterized fluorescence-based solution-phase assay (27) to measure the dissociation constants of KRAS^WT, KRAS^G12D, and KRAS^G12R with recombinant full-length p110α as well as the isolated RAS binding/association (RBD/RA) domains of RAF and other effectors. In agreement with previous studies (15), the dissociation constant of CRAF–RBD with KRAS^WT was 61.0 ± 0.53 nM, which was decreased approximately 8-fold in KRAS^G12D and KRAS^G12R. The dissociation constant of p110α with KRAS^WT and KRAS^G12D was approximately 2 μM; however, we did not detect any interaction between p110α and KRAS^G12R (Fig. 4F and G). Dissociation constants of the isolated RA domains from RGL2 and PLCϵ were comparable among all KRAS mutants tested (Supplementary Fig. S4F).

To determine whether the inability of KRAS^G12R to stimulate macropinocytosis is due solely to its impaired interaction with p110α, we investigated whether coexpression of a membrane-targeted and constitutively activated p110α variant (p110α-CAAX; ref. 28) would rescue macropinocytosis. Although expression of p110α-CAAX alone was not sufficient to stimulate macropinocytosis (Fig. 4H–J), coexpression of p110α-CAAX with KRAS^G12R in RIE-1 cells stimulated macropinocytosis to levels similar to those observed with KRAS^G12D/V alone. As expected, coexpression of p110α-CAAX did not further enhance macropinocytosis in KRAS^G12D/V-expressing cells. Similarly, activation of endogenous PI3K signaling using insulin was only sufficient to elevate macropinocytosis in RIE-1 cells expressing KRAS^G12R (Fig. 4H–J). Conversely, treatment with AZD8186, a PI3Kα/β inhibitor, decreased macropinocytosis in RIE-1 cells expressing KRAS^G12D/V (Fig. 4K–M). Thus, we conclude that the inability of KRAS^G12R to stimulate macropinocytosis in RIE-1 cells can be attributed solely to its impaired interaction with p110α.

As KRAS^G12R fails to bind p110α in vitro, we speculated that KRAS^G12R-mutant PDAC may also exhibit decreased PI3K signaling in vivo. Therefore, we applied RPPA to KRAS^G12R PDAC cell lines to analyze the expression/activity of more than 160 different signaling proteins (Fig. 5A). Although we observed that the RPPA profiles of KRAS^G12R-mutant cell lines clustered separately from KRAS^G12D/V-mutant cell lines, no clear mutation-specific signature was readily apparent (Supplementary Fig. S5A). Upon limiting our analysis to the 60 most differentially phosphorylated/expressed protein features (30 increased, 30 decreased), two prominent distinctions were observed in KRAS^G12R-mutant cell lines: upregulated levels of MYC and downregulation of signaling through PI3K–AKT–mTORC1–S6K–S6 (Fig. 5A).

Given that MYC is a critical KRAS effector in glucose metabolism (29), we sought to determine what role, if any, MYC plays in macropinocytosis and survival of KRAS^G12R-mutant PDAC cell lines. As we recently determined for KRAS^G12D/V-mutant PDAC (30), the growth of KRAS^G12R-mutant PDAC was also MYC-dependent (Supplementary Fig. S5B). However, transient knockdown of MYC reduced macropinocytosis in KRAS^G12D/V- but not KRAS^G12R-mutant cell lines (Fig. 5B and C; Supplementary Fig. S5C). Furthermore, as we recently described (30), MYC protein stability is dependent on KRAS activity. Consistent with this premise, transient knockdown of KRAS decreased MYC protein levels in all PDAC cell lines, independent of the specific KRAS mutation (Fig. 5D).

We then sought potential connections among the top 30 proteins that were the most decreased in phosphorylation or expression according to RPPA pathway mapping (Fig. 5A; Supplementary Fig. S5A). Utilizing the Reactome pathway database and bioinformatics tools to identify signaling relationships among these proteins in an unbiased manner (https://reactome.org/; ref. 31), we again found that the most decreased proteins were related to PI3K–AKT–mTOR signaling (Fig. 5E). Going back to the RPPA analyses, we found that the absolute RPPA intensities of these proteins were significantly decreased in the KRAS^G12R PDAC cell lines compared with the KRAS^G12D lines (Fig. 5F).

Despite the inability of KRAS^G12R alone to drive macropinocytosis in a variety of cell models, KRAS^G12R-mutant PDAC cell lines nevertheless exhibited robust macropinocytosis (Fig. 1A). As we had determined that the inability of KRAS^G12R to drive macropinocytosis in cell models was due solely to loss of p110α activation, we then addressed the possibility that another PI3K isoform activated independently of KRAS may compensate for this deficiency. Whereas expression of the p110α-related PI3K catalytic subunit isoforms p110δ and p110γ is generally restricted to hematopoietic cells (32), both isoforms are upregulated in PDAC (Fig. 6A; ref. 33). To address their potential role in macropinocytosis, we utilized PI3K isoform–selective inhibitors. As expected, the pan-PI3K inhibitor pictilisib (p110α/β/δ/γ) significantly reduced macropinocytosis in all KRAS-mutant PDAC cell lines tested (Fig. 6B), supporting the requirement for PI3K activity in this process. The p110α-selective inhibitor alpelisib and the p110α/β inhibitor AZD8186 reduced macropinocytosis only in KRAS^G12D/V-mutant lines but had no effect in any of the seven KRAS^G12R-mutant PDAC lines (Fig. 6B; Supplementary Fig. S6A and S6B, respectively), consistent with the inability of KRAS^G12R to activate p110α. Finally, the p110γ-specific inhibitor IPI-549 not only decreased macropinocytosis in all KRAS^G12R-mutant lines, but unexpectedly also did so in the KRAS^G12D/V-mutant lines (Fig. 6B). As expected, treatment with the pan-PI3K inhibitor reduced pAKT levels in all cell lines (Fig. 6C). However, treatment with the selective p110α or p110γ inhibitors did not result in a consistent reduction in pAKT levels that correlated with a reduction in macropinocytosis. These results indicate that PI3K but not AKT activity is required to support macropinocytosis.

Verifying the p110γ-specific activity of IPI-549, siRNA suppression of PIK3CG (encoding p110γ) reduced macropinocytosis in all cell lines evaluated, regardless of the specific KRAS mutation (Fig. 6D). Coupled with our evidence that macropinocytosis in KRAS^G12R-mutant PDAC lines is KRAS-independent (Fig. 1B–D), this result indicates that p110γ contributes to macropinocytosis regardless of whether that process is KRAS-dependent or KRAS-independent. Furthermore, PI3K can activate the RAC1 small GTPase, a known regulator of RAS-dependent macropinocytosis, through activation of RACGEFs (e.g., PREX1/2; ref. 34). In agreement with previously published results (35, 36), transient siRNA suppression of RAC1 reduced macropinocytosis in all lines tested, irrespective of KRAS mutation status (Fig. 6E).

Given that KRAS^G12R displayed impaired p110α binding and PI3K signaling, and that the PI3K–AKT axis is a known driver of resistance to inhibitors of the ERK/MAPK cascade (37), we speculated that KRAS^G12R-mutant PDAC cells may be preferentially sensitive to inhibitors of the MEK–ERK pathway compared with PDAC cells expressing KRAS mutants that retain the ability to activate p110α. To address this possibility, we analyzed a panel of 52 PDAC cell lines treated with the MEK1/2-selective inhibitor selumetinib. Mean activity area (AA) was obtained from four-parameter drug–response curves (Fig. 6F) and integrates IC₅₀ (Fig. 6G) with maximum response (A_max) at the highest drug concentration [AA = Σi[0-min(0,Ai/100)]. There was >80% agreement between AA and IC₅₀ values in classifying cell lines as sensitive (AA ≥ 2.5; IC₅₀ < 1 μmol/L) or resistant (AA < 1.8; IC₅₀ ≥ 10 μmol/L; Supplementary Fig. S6C and S6D). On average, PDAC cell lines harboring KRAS^G12R mutations were more sensitive to MEK inhibition than those harboring other KRAS mutations, yielding higher AA values and lower IC₅₀ values.

To exclude the possibility that other genetic mutations associated with KRAS mutational status might drive the selumetinib drug–response phenotype, we determined the mutation status of the 250 most common cancer genes in all 52 cell lines. Like the large genotyping efforts of TCGA and the International Cancer Genome Consortium on clinical specimens, we found that TP53 (34/52), CDKN2A (8/52), and SMAD4 (8/52) genes followed KRAS as the most commonly mutated genes in PDAC (Supplementary Fig. S6E). We did not find a significant co-occurrence of mutations specific to the KRAS^G12R cell lines among the 250 genes analyzed, including single-nucleotide variants or small insertions and deletions.

Utilizing additional model systems, we found that growth of KRAS^G12R-mutant PDAC organoids was more sensitive than KRAS^G12D-mutant organoids to MEK1/2 inhibition (selumetinib; EC₅₀ of 12.3 nmol/L vs. 575 nmol/L; Fig. 6H and I). Finally, we transplanted PDAC patient-derived xenograft (PDX) tumors into immunocompromised mice and treated them with 35 mg/kg selumetinib twice daily for 28 days. Prior to treatment (F3 or F4 generation; F0, surgical resection specimens; tumor volume 200–250 mm³), these PDX tumors retained most cytoarchitectural features of the original patient tumor, including a significant human stromal component (Fig. 6J and K). Selumetinib suppressed growth of the KRAS^G12R-mutant PDX tumors to a significantly greater extent than the KRAS^G12D-mutant tumors (Fig. 6L and M). To confirm that growth suppression was related to the mechanism of action of selumetinib, we examined phospho-ERK, cyclin D1 (T286), and p27^KIP1 (T187) in untreated and treated tumors. Inhibition of MEK induces G₁–S cell-cycle arrest via a reduction of cyclin D1 and the induction and stabilization of p27^KIP1. Although phosphorylated cyclin D1 levels (T286) were slightly elevated in both tumor sets, quantitative RPPA profiling of pooled lysates from tumors (n = 6 per group) showed effective inhibition of ERK phosphorylation in both PDX models with no difference in the induction of p27^KIP1 levels between the two models (Fig. 6N and O). In summary, selumetinib achieved MEK inhibition in the xenotransplanted mice.

Finally, because we previously found that KRAS-mutant PDAC cell lines displayed inconsistent sensitivities to MEK versus ERK inhibition (11), we also utilized the ERK1/2-selective inhibitor SCH772984 (ERKi) in five KRAS^G12D- and five KRAS^G12R-mutant PDAC organoids. ERK inhibition was moderately more effective at inhibiting the proliferation of KRAS^G12R versus KRAS^G12D PDAC; however, the difference was not statistically significant (Supplementary Fig. S7A–S7C). In summary, a subset of KRAS^G12R-mutant PDAC models exhibited increased sensitivity to MEK or ERK inhibition compared with KRAS^G12D-mutant PDAC. Overall, the effect was modest, and we could not conclude that KRAS^G12R mutation status alone provides a definitive genetic marker to merit single-agent treatment of patients with PDAC with MEK or ERK inhibitors.

To assess whether KRAS^G12R- and KRAS^G12D-mutant PDAC exhibits differential sensitivity to other oncology drugs, we performed a 525-inhibitor DSRT analysis on seven KRAS^G12R PDAC lines and compared the results with 16 KRAS^G12D/V-mutant PDAC cell lines (Supplementary Fig. S8A). In this assay, the KRAS^G12R-mutant lines showed a trend toward increased sensitivity to MEK/ERK inhibition compared with the KRAS^G12D/V-mutant lines, and surprisingly they were significantly more sensitive to PI3K inhibition, likely due to the lack of KRAS signaling through this pathway (Fig. 7A). Given the DSRT results and unique dependence on PI3Kγ in KRAS^G12R cell lines, we asked whether KRAS^G12R-mutant organoids would be more susceptible to concurrent ERK/MAPK and PI3Kγ inhibition. In agreement with the DSRT and macropinocytosis data, KRAS^G12R-mutant organoids showed increased sensitivity to PI3Kγ inhibition (hT2 EC₅₀ = 2.6 μmol/L vs. hM1A, resistant), and combined ERK/MAPK and PI3Kγ inhibition was synergistic in the KRAS^G12R-mutant organoid line (Supplementary Fig. S8B and S8C).

We recently determined that concurrent inhibition of ERK1/2 and autophagy synergistically suppressed the growth of KRAS-mutant PDAC (38), leading to our initiation of clinical trials to evaluate the combination of MEK or ERK inhibitor together with the autophagy inhibitor hydroxychloroquine. In the present study, we noted from our DSRT analyses that KRAS^G12R PDAC showed increased single-agent sensitivity to both chloroquine and mepacrine compared with KRAS^G12D/V lines (Fig. 7A). We speculated that this may be due to the impaired ability of KRAS^G12R to activate PI3K–AKT–mTORC1 signaling, which suppresses autophagy (9). We further speculated that KRAS^G12R PDAC lines may be hypersensitive to this combination. To directly determine if KRAS^G12R and KRAS^G12D differ in their regulation of autophagy, we applied the dual-fluorescence autophagic flux assay to RIE-1 cells stably expressing the different KRAS mutants. As we observed previously (38), autophagy was suppressed in KRAS^G12D/V-expressing cells relative to empty-vector control cells. In contrast, autophagy was not suppressed in KRAS^G12R-expressing cells (Supplementary Fig. S8D–S8F). Thus, KRAS^G12R is impaired in regulation of two distinct lysosome-associated metabolic activities: macropinocytosis and autophagy.

We next determined whether KRAS^G12R-mutant PDAC lines are more sensitive to combined ERK and autophagy inhibition. As expected, chloroquine alone was sufficient to decrease cell viability, but had no effect on cytotoxicity as a single agent (Fig. 7B). However, the decrease in viability upon dual treatment with ERK inhibitor plus chloroquine was only additive, with markedly less synergy than observed in non–KRAS^G12R-mutant PDAC cell lines (Fig. 7B), yet the combination increased cytotoxicity in most of the KRAS^G12R cell lines. Finally, in addition to chloroquine, a nonspecific inhibitor of autophagy, we also tested two ULK-specific inhibitors, which block the initiating kinases of the autophagic cascade and have been shown to reduce autophagy in PDAC (38). Both ULK inhibitors, MRT68921 (ULK1/2i; Fig. 7C) and SBI0206965 (ULK1i; Fig. 7D), were more effective than chloroquine at reducing cell viability when combined with ERKi, and ULK1/2i was synergistic with ERK inhibition in four out of five KRAS^G12R-mutant PDAC cell lines. Thus, we speculate that, although KRAS^G12R activation of p110α is impaired, the compensatory activation of p110γ that allows KRAS^G12R to potently drive PDAC growth also minimizes its differential drug sensitivity, as revealed in the DSRT analysis (Supplementary Fig. 8A), compared with KRAS^G12D PDAC.

The KRAS^G12R mutation is rare in cancer overall, yet found at high prevalence in PDAC. Furthermore, KRAS^G12R mutations in PDAC are associated with additional co-occurring KRAS mutations (6), and KRAS^G12R status is correlated with better overall survival (13 months) compared with KRAS^G12D (8 months) or KRAS^G12V (10 months; ref. 39), supporting our premise that the KRAS^G12R mutation may have driver functions that are distinct from the more common G12 mutants. We therefore evaluated the cellular and biochemical properties of KRAS^G12R. Surprisingly, KRAS^G12R fails to interact with and activate a key effector, p110α PI3K, essential for KRAS-driven cancer development (40). In addition, KRAS^G12R is impaired in stimulating macropinocytosis, a metabolic activity critical for PDAC tumor growth (10). However, we found that overexpression of the related p110γ PI3K compensates for the impaired promotion of macropinocytosis by KRAS^G12R, providing a possible explanation for the unusually common occurrence of this KRAS mutation in PDAC.

Although there are six possible single-base change mutations at codon 12, their overall frequencies are widely divergent. Until recently, it was assumed that the basis for this was different susceptibilities to mutational insults. The KRAS^G12C mutation, which is found at a high frequency in smoking-associated lung cancers but not in other cancers (1), provides strong evidence for this premise. However, complicating this issue are tissue-specific differences, for example that KRAS^G12R is rare in cancer overall yet common in pancreatic cancer. Thus, an alternative hypothesis states that the biological potency of each mutant KRAS protein drives cancer development and thereby dictates how common or rare a given mutation will be. Our finding that KRAS^G12R, unlike the more prevalent G12D or G12V mutant proteins, failed to bind and activate a key effector, PI3Kα, supports a biological basis for the low prevalence of this mutation overall in cancer. As we discuss below, we also identified a possible basis for how KRAS^G12R, despite this defect, can still effectively drive PDAC development and growth. In summary, our findings, together with other recent studies (41, 42), support the importance of oncogenic potency in driving the differential frequency of specific RAS mutations seen in cancers that arise from different tissues.

The RAS–PI3Kα interaction has been shown to be necessary for tumor growth: Ablation of RAS interaction with p110α reduced tumor initiation and maintenance in RAS-driven lung and skin cancer (7, 43). Therefore, although impaired engagement with PI3Kα may provide a biological basis for the rarity of KRAS^G12R mutations in other KRAS mutant–associated cancers, such as lung and colorectal cancers, conversely, why KRAS^G12R constitutes the third most prevalent KRAS mutation in PDAC has been puzzling. Our studies identified two possible KRAS-independent activities that enable KRAS^G12R mutations to be more common in PDAC compared with other cancer types.

We showed that ectopic expression of activated PI3Kα rescued the macropinocytosis defect seen when KRAS^G12R was evaluated in cell models, arguing that this mutant was defective solely in activation of PI3Kα. In PDAC, a possible mechanism to restore PI3K signaling may be exogenous insulin from pancreatic endocrine tissue acting on adjacent tumor cells. However, this would not explain the growth of metastatic KRAS^G12R-mutant PDAC in non-pancreas tissue. Thus, other mechanisms that activate PI3K may also restore the potent cancer-driving functions of KRAS^G12R.

Another mechanism we identified in KRAS^G12R-mutant PDAC involved overexpression of the related isoform p110γ. Consistent with this compensatory mechanism, whereby p110γ provides the functions of both p110γ and p110α, we found that treatment with a p110γ- but not p110α-selective inhibitor blocked macropinocytosis. However, in KRAS^G12D-mutant PDAC cells, a p110α- or p110γ-selective inhibitor (or a pan-PI3K inhibitor) were all capable of blocking macropinocytosis, indicating that both α and γ isoforms were active and required in this context. Studies in Dictyostelium discoideum have shown that distinct isoforms of PI3K serve distinct roles in supporting macropinocytosis, with one isoform driving cup formation and the other cup closure (26). Similarly, macropinocytosis in KRAS^G12D-mutant PDAC cells may require distinct functions of different PI3K isoforms.

We observed that MYC protein stability is regulated by both KRAS^G12D and KRAS^G12R, and that both KRAS^G12D- and KRAS^G12R-mutant PDAC cell lines retain MYC-dependent growth. However, macropinocytosis was dependent on MYC in KRAS^G12D- but not KRAS^G12R-mutant PDAC cell lines. This connection between MYC and macropinocytosis in KRAS^G12D-mutant PDAC was unexpected, as no association between them has been described previously.

Another surprising finding from our studies was that the lack of elevated macropinocytosis when KRAS^G12R was ectopically expressed in simple cell models was not reflected in bona fide cancer cells endogenously harboring KRAS^G12R mutations. Indeed, macropinocytosis was elevated in both KRAS^G12R- and KRAS^G12D-mutant PDAC cell lines, consistent with the critical role of macropinocytosis in pancreatic cancer growth (10). However, in the KRAS^G12R-mutant cell lines, this elevation was KRAS-independent. An important lesson learned is that the model systems are useful in demonstrating what a given KRAS mutant can or cannot do alone, but the phenotypes observed do not necessarily reflect what will occur in tumor cells expressing that same mutant in a context that is genetically and epigenetically more complex.

Upon identifying defective PI3Kα signaling by KRAS^G12R, an activity associated with driving resistance to ERK/MAPK inhibition, we anticipated that we might be able to identify a therapeutic approach that would be preferentially effective in KRAS^G12R-mutant PDAC. Taking both candidate and unbiased approaches, we found that KRAS^G12R-mutant PDAC showed increased sensitivity to inhibition of ERK/MAPK or autophagy compared with KRAS^G12D-mutant PDAC. We speculated that perhaps KRAS^G12R-mutant PDAC may be hypersensitive to combined inhibition of ERK/MAPK and autophagy, which we recently identified as an effective therapeutic strategy for KRAS^G12D/V-mutant PDAC (38). We were surprised that, whereas this combination synergistically suppressed the growth of KRAS^G12D-mutant PDAC, it was only additive in KRAS^G12R-mutant PDAC. Although all KRAS-mutant PDACs are responsive to this combination, and as trials have begun and are planned with combinations that inhibit MEK or ERK and autophagy, other genetic markers will be needed beyond specific KRAS mutational status, to stratify patients for treatment with these combinations.

PATC43 and PATC50 cell lines were provided by J. Fleming (MD Anderson Cancer Center, Houston, TX); RWP1, FA6, and Paca44 were from N. Lemoine (Barts Cancer Institute, London, United Kingdom); IMIM-PC-1, SK-PC-1, and SK-PC-3 were from F. Real (Center Nacional de Investigaciones Oncologicas, Spain); MDAPanc-48 and MDAPanc-81 were from E. Frazier (MD Anderson Cancer Center, Houston, TX); L3.3 and FG were from I. Fidler (MD Anderson Cancer Center, Houston, TX); A818 was from H. Kalthoff (University of Kiel, Germany); PSN.1 was from the European Collection of Authenticated Cell Cultures; SNU-213, SNU-324, and SNU-410 were from the Korean Cell Line Bank; HuP-T3, HuP-T4, YAPC, DAN-G, and PaTu-8988T were from the DSMZ German Collection of Microorganisms and Cell Cultures; PACADD-135, PACADD-159, PACADD-165, and PACADD-183 were from F. Rueckert (University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany); KP-2, KP-3, and TCC-Pan2 were from the Japanese Collection of Research Bioresources Cell Bank; PK-1, PK-8, PK-45P, PK-59, KP-4, NOR-P1, and T3M-4 were from Riken Cell Bank; and SB.rb and SB.bm were primary patient-derived pancreas cancer cell lines from our institution (U. Rudloff, NCI, Bethesda, MD). The remaining PDAC cell lines were obtained from the ATCC and were maintained in either DMEM or RPMI-1640 supplemented with 10% FBS. The model cell lines (HPNE-DT, HEK293T, and RIE-1) were all maintained in DMEM supplemented with 10% FBS. All cell lines were maintained in a humidified chamber with 5% CO₂ at 37°C. Cell lines used in experiments were passaged for 1 month or 10 passages before a new aliquot was thawed. Cell line authenticity was verified by short tandem repeat profiling (April 17, 2017), and all lines were monitored monthly for Mycoplasma contamination using the Lonza MycoAlert Mycoplasma Detection Kit.

Macropinocytosis assays were performed as described previously (10). Cells were incubated for 30 minutes with 50 or 100 μg/mL DQ-BSA, followed by a 90-minute chase in serum-free DMEM/RPMI before fixation. Approximately 50 to 100 cells in 9 to 12 fields of view per condition were imaged on a Zeiss 700 confocal microscope (63×, 1.4 N.A. objective). For overall macropinocytosis levels in PDAC cells, power and gain levels were determined using the AsPC-1 cell line. For RIE-1 cells, power and gain levels were set using the KRAS^G12D condition in each experiment. For conditions where siRNA, inhibitors, or other treatments were applied, power and gain levels for each cell line were set using the corresponding control condition. The cell outline was mapped using a differential interference contrast image. Macropinocytotic index was quantified using ImageJ by taking the macropinosome area divided by the total cell area and multiplying by 1,000.

Human KRAS4B mutant proteins and membrane-targeted, constitutively activated chimeric p110α terminating in the HRAS C-terminal “CAAX” membrane targeting sequence (28) were ectopically expressed from the retroviral expression vector pBabe in RIE-1, NIH/3T3, and HPNE-DT cells. The doxycycline-inducible lentiviral expression vector pInducer was used to ectopically express siRNA-resistant KRAS4B mutants in AsPC-1 and Pa04C PDAC cells. Viral particles were generated by transient transfection of each expression vector into HEK293T cells using Fugene6 (Promega) with the pCL-10A1 packaging system for retrovirus or the psPAX2 and pMD2.G packaging system for lentivirus, according to the manufacturer's recommended protocol. Infection of cell lines was performed in growth medium supplemented with 8 mg/mL polybrene.

siRNA Silencer Select oligonucleotides against scrambled (Negative Control No. 1), KRAS (s7940 and s7939), MYC (s9129 and s9130), p110γ (s10533 and s10534), and RAC1 (s11711 and s11712) sequences were obtained from Invitrogen and transfected into cells by using Lipofectamine RNAiMAX following the manufacturer's recommended protocol. After 16 hours, the medium was exchanged, representing time 0 for siRNA-mediated knockdown.

To study the effect of KRAS silencing on growth, cells were treated for 48 hours with a KRAS-specific siRNA, trypsin-digested, counted, and plated. To measure growth on plastic, cells were plated in duplicate in 6-well dishes at a density of 2 × 10³ (MIA PaCa-2 and AsPC-1), 3 × 10³ (A818-4, PSN-1, and PK-8), or 5 × 10³ (Pa04C, PATC43, PATC50, TCC-PAN2, and HuP-T3) cells per well. Plates were developed after 7 days by removing the medium and fixing cells with 4% paraformaldehyde and crystal violet. To monitor 3-D proliferation, 50 μL 0.6% Bacto agar per well was placed into clear-bottom 96-well plates. Cells were mixed into 1% SeaPrep agarose and plated at a density of 5 × 10³ (MIA PaCa-2 and AsPC-1), 7.5 × 10³ (A818-4 and PK-8), or 1 × 10⁴ (Pa04C, PATC43, PATC50, PSN-1, TCC-PAN2, and HuP-T3) cells per well. To quantitate cell number, cell viability was determined by staining with AlamarBlue after 7 days, according to the manufacturer's protocol. Each biological replicate experiment included four technical repeats for each cell line. A corresponding 6-well plate was generated for Western blot analysis to verify KRAS knockdown.

Organoids (hM1A and hT2) were kindly provided by the Tuveson laboratory (Cold Spring Harbor Laboratory). Organoids were cultured at 37°C in 5% CO₂. Cells were seeded in growth factor–reduced Matrigel (Corning; catalog no. 356231) domes and fed with human complete feeding medium: advanced DMEM/F12-based WRN-conditioned medium (L-WRN; ATCC CRL-3276), 1× B27 supplement, 10 mmol/L HEPES, 0.01 μmol/L GlutaMAX, 10 mmol/L nicotinamide, 1.25 mmol/L N-acetylcysteine, 50 ng/mL hEGF, 100 ng/mL hFGF10, 0.01 μmol/L hGastrin I, 500 nmol/L A83-01, 1 μmol/L PGE2, and 10.5 μmol/L Y27632 (44).

Organoids were generated under Institutional Review Board (IRB)–approved protocols at Dana-Farber Cancer Institute (DFCI #14-408, 17-000) or Cold Spring Harbor Laboratory using previously published protocols (44, 45). All patients provided written informed consent, and the studies were conducted in accordance with recognized ethical guidelines. Organoids were cultured at 37°C in 5% CO₂. Organoids were dissociated, and 3,000 to 5,000 single cells per well were seeded in 150 μL of 10% growth factor–reduced Matrigel (Corning) and 90% human organoid feeding medium + 10.5 μmol/L Y27632 (Selleckchem) into poly(2-hydroxyethyl methacrylate)-coated clear flat-bottom 96-well plates (Corning; Ref. 3903). Two days after seeding, organoids were treated with ERKi SCH772984 (0 nmol/L to 10 μmol/L) or MEKi AZD6244 (selumetinib; 0 to 2.5 μmol/L), randomized on a Tecan D300e drug dispenser. On day 7, the perimeter wells were refilled with 1× PBS. Ten days after inhibitor treatment, organoids were imaged with a Molecular Devices SpectraMax i3x MiniMax 300 imaging cytometer. After image acquisition, organoid viability was assessed using the CellTiter-Glo 3D Cell Viability Assay (Promega; catalog no. G9683) on a SpectraMax i3x (UNC) or Clariostar Plate Reader (LC-3200; DFCI) plate reader according to the manufacturer's protocol. Depending on the assay, duplicate or triplicate measurements were recorded for each inhibitor concentration.

Cells were washed twice with ice-cold PBS, lysed in 1% NP-40 buffer (25 mmol/L Tris HCl, pH 7.4, 100 mmol/L NaCl, 10 mmol/L MgCl₂, 1% NP-40) supplemented with phosphatase (Sigma) and protease (Roche) inhibitors, scraped, and collected in chilled Eppendorf tubes. Lysates were cleared by centrifugation at 15,000 × g for 15 minutes at 4°C, and the protein concentration was determined using Bradford assay (Bio-Rad). Standard immunoblotting procedures were followed. Membranes were blocked in 5% milk diluted in TBS with 0.05% Tween 20 (TBST). To determine the levels of activated proteins, blot analyses utilized phospho-specific antibodies to AKT (T308 and S473), p70S6K (T389), MEK1/2 (S217/S221), ERK1/2 (T202/Y204), and RSK (T359/S363; Cell Signaling Technology), with corresponding antibodies recognizing total proteins, as well as p110 (all isoforms), RAC1, and MYC (Cell Signaling Technology) to measure total protein levels. Immunoblotting for total KRAS (Sigma) protein was done to verify siRNA suppression and for GAPDH (Cell Signaling Technology) and vinculin (Sigma) to verify equivalent loading of total cellular proteins. Anti-HA antibody (Covance) was used to probe for overexpressed KRAS.

Apoptosis analyses were performed using the TACS Annexin V-FITC Kit (Trevigen, Inc.) following the manufacturer's protocol. Briefly, spent culture medium containing detached cells was collected, mixed with trypsinized cells, and centrifuged at 300 × g for 5 minutes. After being washed once in ice-cold 1X PBS, cells were incubated in Annexin V Incubation Reagent (1% Annexin V–FITC, 1× propidium iodide solution, in 1× calcium-containing binding buffer) at room temperature for 15 minutes in the dark before diluting with 1× binding buffer. Cells were analyzed on a BD LSRFortessa flow cytometer, and data were collected and exported using FACSDiva v8.0.1 and analyzed with Cytobank. Using an SSC-A (y) versus FSC-A (x) dot plot, a “cells” gate was used to exclude small and large debris. Cells were analyzed for apoptosis in a PI-A (y) versus FITC-A (x) dot plot, and a quadrant gate was established around the healthy population of DMSO control cells (PI-A negative, FITC-A negative). For cell-cycle analysis, cells were harvested, washed once in PBS, and suspended in fresh PBS. Ten volumes of ice-cold 70% ethanol were added to each tube dropwise while gently vortexing. Cells were incubated overnight at 4°C. The fixed cells were washed once in PBS and suspended in PBS containing 40 μg/mL propidium iodide (Life Technologies, Inc.) and 100 μg/mL RNase A (Life Technologies, Inc.). Cells were incubated for 2 to 3 hours at 37°C and analyzed using FACSDiva v8.0.1 on a BD LSRFortessa flow cytometer. After separating cells from debris in an SSC-A (y) versus FSC-A (x) dot plot, doublets were excluded with a FSC-H (y) versus FSC-A (x) dot plot. Singlet cells were analyzed for propidium iodide staining to determine DNA content at 2N (G₀–G₁), 4N (G₂–M), or in between (S). Data were analyzed with the Multicycle DNA algorithm in FCS Express.

Samples for RPPA analyses were prepared and performed as described previously (46). In brief, cell lysates were prepared and arrayed on an Aushon 2470 automated system (Aushon BioSystems). Selected arrays were stained with Sypro Ruby Protein Blot Stain (Molecular Probes) following the manufacturer's instructions to quantify the total protein. Remaining arrays were pretreated with Reblot Antibody Stripping solution (Chemicon) for 15 minutes at room temperature, followed by two washes with PBS, and incubated for 5 hours in I-block (Tropix) prior to antibody staining (47). Using an automated system (DakoCytomation), arrays were incubated with 3% hydrogen peroxide, biotin-blocking system, and an additional serum-free protein block to reduce nonspecific binding. Each array was probed for 30 minutes with one antibody targeting the protein of interest. Arrays were probed with a total of 157 antibodies previously tested for their specificity and ability to capture the linear dynamic range of the protein of interest (48, 49). Biotinylated anti-rabbit (Vector Laboratories, Inc.) or anti-mouse secondary antibody (DakoCytomation) coupled with a commercially available tyramide-based avidin/biotin amplification system (Catalyzed Signal Amplification System; DakoCytomation) were used for signal amplification. Fluorescence detection was achieved using the IRDye 680RD Streptavidin (LI-COR Biosciences) system. Sypro Ruby and antibody-stained slides were scanned on a Tecan laser scanner (TECAN) using the 580 nm and 620 nm channels. Images were analyzed using commercially available software (MicroVigene Version 5.1.0.0, Vigenetech) as described previously (50).

Supervised hierarchical clustering was performed using R (version 3.4.1). The heat map, containing the 60 most differentially expressed protein features, was generated using the ComplexHeatmap package. The RPPA standardized intensity data were log₂ transformed, and the median of three independent biological replicates was determined for each feature. Data from cells harboring KRAS^G12D or KRAS^G12R mutations were separated into two groups. A Wilcoxon rank sum test was applied to identify the top 30 and bottom 30 features that were most differentially expressed between the two groups, out of the 162 total protein features evaluated. Heat maps were row-normalized using a Z-score. Rows and columns were clustered via hierarchical clustering (Euclidean distance).

Cells (1,000–2,500 cells per well, depending on cell line) were seeded in 96-well plates and incubated for 24 hours before addition of drug. Increasing concentrations of drug, with DMSO (vehicle) as a negative control, were added to the wells in three replicates using a digital dispenser liquid handling device (TECAN D300e). For the 525-inhibitor DSRT screen, plates were incubated at 37°C for 5 days after addition of drug and analyzed using the Promega CellTiter-Glo assay reagent (Promega). Plates were read on a GloMax 96 Microplate Luminometer (Promega). Percent cell viability was calculated by normalizing raw luminescence values to vehicle-control (DMSO-treated) samples. For cotreatment with the ERK inhibitor and chloroquine or the ULK inhibitors, cells were stained with Calcein AM (Invitrogen) according to the manufacturer's recommended protocol, and the wells were counted using a SpectraMax MiniMax 300 imaging cytometer. For survival curves, the average of 10 wells from a day 0 control plate was used to determine baseline viability, whereas the average of the vehicle-control wells at day 5 was used to determine maximum viability. All data were analyzed using SoftMax version 5 and GraphPad Prism version 7.03 using a 4-parameter drug–response curve. Generation of the dose–response curves as well as scoring and clustering the data were performed as previously described (51), using the cell lines indicated in the figure.

Patients provided written informed consent on IRB-approved protocol NCI 09-C-0179, and the studies were conducted in accordance with recognized ethical guidelines. All animal studies were conducted at the NCI in Bethesda, MD, according to protocols and policies of the Institutional Animal Care and Use Committees. Immediately after surgical resection, 2 × 2 mm² fragments from viable tumor areas were implanted subcutaneously into the right flanks of 6- to 8-week-old NOD/SCID mice (F0 generation). Tumor fragments were retransplanted 2 to 3 times when tumors reached 2 × 2 cm² (F3 and F4 generation). When tumor volumes reached 200 mm³, at least 10 mice per arm were treated by oral gavage with vehicle or selumetinib (35 mg) twice daily for 28 days. Tumor volume was measured twice weekly via bidirectional caliper measurements. Tumor lysates were obtained from harvested tumors after 28 days and subjected to RPPA analysis. In addition to hematoxylin and eosin staining, tumors were stained for collagen type I (ab34710, Abcam), trichrome, smooth muscle actin (SMA; #A2547, Sigma-Aldrich), and CD31 (ab28364; Abcam). Brightfield images (immunostaining) were acquired using an Aperio ScanScope XT (Aperio) for whole slide scanning at 20× magnification and analyzed using ImageScope Microvessel Analysis.

The cDNA sequence encoding truncated human KRAS4B (residues 1–169) was cloned into the pQlinkH bacterial expression vector (gift from Konrad Buessow, Max Planck Institute for Molecular Genetics, Berlin, Germany; Addgene plasmid # 13667) containing an N-terminal 6x-His purification tag followed by a Tobacco Etch Virus (TEV) protease cleavage site (52). RASGRP1 (residues 50–468, pET28a) and SOS^cat (residues 564–1,049, pPROexHTb) contained a similar vector architecture. Bacterial expression vectors encoding KRAS-GTP effector interacting domains included CRAF-RBD (residues 54–131, pQlinkH), RGL2-RA (residues 647–736, pGEX3T-2; ref. 53), and PLCϵ-RA (residues 2,113–2,221, pTriEx4; ref. 54). All 6x-His–tagged proteins were expressed in BL21 (DE3) Rosetta2 cells and purified following the Qiagen Nickel NTA purification protocol, and the 6X-His tags were removed using TEV protease. For pGEX vectors, proteins were purified following the Glutathione Sepharose 4B purification protocol (Amersham Pharmacia Biotech). The GST-tag was cleaved overnight using thrombin protease while dialyzing in wash buffer. If necessary, the proteins were further purified by size exclusion chromatography (Superdex-75 10/300 GL column; GE Life Sciences) and judged greater than 95% pure by SDS-PAGE analysis. Insect cell-expressed full-length human PI3K (p110α/p85) protein was provided by H. Li and J. Wu (Genentech) and expressed and purified as described previously (55).

We performed nucleotide exchange by replacing GDP with nucleotide analogues. For loading GMPPNP and 2′-/3′-O-(N'-methylanthraniloyl)β, γ-methyleneguanosine 5′-triphosphate (Jena Biosciences), KRAS^G12R was incubated with alkaline phosphatase beads and GMPPNP (20-fold excess) for 3 hours with gentle rotation, followed by removal of alkaline phosphatase and purification using a desalting column. Purified protein was checked for nucleotide exchange by high-performance liquid chromatography, which showed close to 95% replacement of the bound nucleotide (56). Exchange for 2′-/3′-O-(N'-methylanthraniloyl)GDP (mGDP; Jena Biosciences) was performed following previously published methods (57).

Nucleotide exchange assays were performed using a Cary Eclipse Fluorescence Spectrophotometer (Agilent), as described previously (57). The minimal catalytic fragment of the RASGEF SOS^cat or GRP1^cat was used to stimulate nucleotide dissociation. All experiments were performed in triplicate.

The fluorescence GTP-binding assay was adapted from a previous protocol (58). In brief, KRAS loaded with mGppCp (1.5 μmol/L) was incubated with increasing concentrations of CRAF-RBD, PLCϵ-RA, or RGL2-RA in Buffer D (20 mmol/L HEPES, 50 mmol/L NaCl, and 5 mmol/L MgCl₂ at pH 7.4) or PI3Kα in assay Buffer D + 5% glycerol. Nucleotide dissociation was initiated by addition of 1,000-fold excess of unlabeled nucleotide at 25°C. The rate of dissociation was monitored by the change in fluorescence at an excitation wavelength of 355 nm and emission at 448 nm, using a Spectramax M5 plate reader with a 384-well Greiner plate. Fluorescence nucleotide dissociation curves were fit to a one-phase exponential decay equation using GraphPad Prism version 7.03. The dissociation rate was fit against the ligand concentration using previously published methods (58). All experiments were performed in triplicate.

KRAS proteins were preloaded with mGDP as described. Using a Cary Eclipse Fluorescence Spectrophotometer with an attached circulating water bath (PCB-1500, Agilent), the temperature was ramped at 1°C/minute from 25°C to 80°C while monitoring mGDP fluorescence every 30 seconds (excitation 365 nm, emission 435 nm). A cuvette containing only mGDP was used as a control, where the intrinsic fluorescence loss over temperature was fit to a linear curve and added to the KRAS curves to correct for fluorescence quenching due to temperature change. The resulting curve was transformed by a first derivative, and the T_m was calculated from the inflection point.

Concentrated KRAS^G12R bound to GMPPNP was used for crystallization screening. The final crystallization buffer consisted of 2.1 mol/L sodium malonate, pH 7.5, and 40 mmol/L dimethyl octylphosphine oxide. For data collection, crystals were flash-frozen (100 K) in the mother liquor supplemented with 30% (v/v) sodium malonate, pH 7.0, as a cryoprotectant. The diffraction data were collected using synchrotron radiation at beamlines ID-24-E at the Advanced Photon Source, Argonne National Laboratory. All measured diffraction spots were indexed, integrated, and scaled using the XDS package (59).

The structure of GMPPNP-bound KRAS^G12R was solved by molecular replacement using the program PHASER, based on the structure of GMPPNP-bound wild-type KRAS4B (60). The initial map showed interpretable electron density of the G-domain except for the switch regions. Further model improvement was performed with alternate rounds of refinement using Phenix.refine (61) and model building via COOT (62). Water molecules were included near the end of refinement. The majority of the model has a clear and well-interpretable electron density map with the exception of a few solvent-exposed side chains, which were omitted in the final model. The final data collection and refinement statistics are given in Supplementary Table S1. Figures were prepared using the program PyMOL.

Data were analyzed by GraphPad Prism built-in test (One-Way ANOVA, Dunnett multiple comparisons test). Data are presented relative to their respective control, or where noted in the figure legend, comparing the G12R condition with other conditions. For all graphs, error bars indicate mean ± SEM for n ≥ 3 independent experiments (except where noted) and P values on graphs are denoted by ****, P < 0.0001; ***, P < 0.0002; **, P < 0.0021; and *, P < 0.032, as determined in GraphPad Prism. Number of samples analyzed per experiment and whether data are representative or an average is indicated in the figure legend.

The atomic coordinates and structure factors for the crystal structure of GMPPNP-bound KRAS^G12R have been deposited in the Protein Data Bank under accession code 6CU6. The datasets generated for the current study are available from the corresponding authors upon request.

M. Pierobon has ownership interest (including patents) in Theranostics Health. E.F. Petricoin III is chief science officer at Perthera, Inc., a consultant at Avant Diagnostics, Inc., and a consultant at Ceres Nanosciences, Inc.; has ownership interest (including patents) in Perthera, Inc., Avant Diagnostics, Inc., and Ceres Nanosciences, Inc.; and has an unpaid consultant/advisory board relationship with Serpin Pharma, Inc. A.J. Aguirre is a consultant at Merck, Oncorus, and Arrakis, and reports receiving a commercial research grant from Mirati Therapeutics. B.M. Wolpin is a consultant at Celgene, GRAIL, G1 Therapeutics, and BioLineRx, and reports receiving commercial research grants from Celgene and Eli Lilly. A.D. Cox is a consultant at Eli Lilly and has an unpaid consultant/advisory board relationship with Mirati Therapeutics. C.J. Der is SAB at Mirati Therapeutics and Warp Drive Bio; a consultant at Deciphera Pharmaceuticals, Eli Lilly, Astex Pharmaceuticals, Kymera, Ribometrix, and Jazz Pharmaceuticals; and reports receiving commercial research grants from Mirati Therapeutics and Deciphera Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G.A. Hobbs, N.M. Baker, A.M. Miermont, U. Rudloff, C.J. Der

Development of methodology: G.A. Hobbs, J. Wang, P. Gautam, D. Esposito, E.F. Petricoin III, A.J. Aguirre, K. Wennerberg, U. Rudloff

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.A. Hobbs, N.M. Baker, A.M. Miermont, R.D. Thurman, M. Pierobon, T.H. Tran, A.O. Anderson, A.M. Waters, J.N. Diehl, B. Papke, R.G. Hodge, J.M. DeLiberty, J. Wang, P. Gautam, K.L. Bryant, S.L. Campbell, E.F. Petricoin III, D.K. Simanshu, A.J. Aguirre, B.M. Wolpin, K. Wennerberg, U. Rudloff

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G.A. Hobbs, A.M. Miermont, R.D. Thurman, M. Pierobon, J.N. Diehl, R.G. Hodge, J.E. Klomp, C.M. Goodwin, J.M. DeLiberty, R.W.S. Ng, P. Gautam, K.L. Bryant, S.L. Campbell, E.F. Petricoin III, D.K. Simanshu, K. Wennerberg, U. Rudloff, A.D. Cox, C.J. Der

Writing, review, and/or revision of the manuscript: G.A. Hobbs, A.M. Miermont, R.D. Thurman, M. Pierobon, P. Gautam, S.L. Campbell, E.F. Petricoin III, A.J. Aguirre, B.M. Wolpin, K. Wennerberg, U. Rudloff, A.D. Cox, C.J. Der

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.M. Goodwin, A.D. Cox, C.J. Der

Study supervision: D.K. Simanshu, U. Rudloff, A.D. Cox, C.J. Der

C.J. Der and A.D. Cox were supported by grants from the NCI (R01CA42978, U01CA199235, P50CA196510, R35CA232113, and P01CA203657), the Department of Defense (W81XWH-15-1-0611), and the Lustgarten Pancreatic Cancer Foundation (388222). For C.J. Der and K. Wennerberg, research was supported by the 2015 Pancreatic Cancer Action Network-AACR Research Acceleration Network Grant, Grant Number 15-90-25-DER. K. Wennerberg was supported by NCI P01CA203657. S.L. Campbell and E.F. Petricoin III were supported by NCI CA203657. A.J. Aguirre was supported by the Lustgarten Foundation, the Dana-Farber Cancer Institute Hale Center for Pancreatic Cancer Research, the Doris Duke Charitable Foundation (2017066), the Pancreatic Cancer Action Network (18-35-AGUI), and NCI K08CA218420-02, P50CA127003, and U01CA224146. B.M. Wolpin was supported by the Lustgarten Foundation and DFCI Hale Family Center for Pancreatic Cancer Research. G.A. Hobbs was supported by NCI F32CA200313 and T32CA009156; N.M. Baker by NCI F31CA180628; R.D. Thurman, J.E. Klomp, and C.M. Goodwin by NCI T32CA009156. C.M. Goodwin was supported by NCI F32CA221005. R.G. Hodge was supported by the 2018 Debbie's Dream Foundation-AACR Gastric Cancer Research Fellowship, in memory of Sally Mandel, Grant Number 18-40-41-HODG. A.M. Waters was supported by a fellowship from the American Cancer Society (PF-18-061); J.N. Diehl by NCI T32CA071341 and a fellowship from the Slomo and Cindy Silvian Foundation. A.M. Miermont and U. Rudloff were supported by the Intramural Research Program of the NIH, NCI (ZIA BC 011267). B. Papke was supported by the Deutsche Forschungsgemeinschaft (DFG PA 3051/1-1). K.L. Bryant was supported by T32CA009156 and by a Pancreatic Cancer Action Network/AACR Pathway to Leadership grant. Support to T.H. Tran, D. Esposito, and D.K. Simanshu was provided by NCI under contract HHSN261200800001E. Structural work used Northeastern Collaborative Access Team beamline (GM103403) at the Advanced Photon Source (DE-AC02-06CH11357). The UNC Microscopy Services Laboratory and Flow Cytometry Core Facility were supported in part by P30CA016086. Research reported in this article was supported by the Center for AIDS Research award number 5P30AI050410. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.