KRAS Codon-Specific Mutations Differentially Toggle PI3K Pathway Signaling and Alter Sensitivity to Inavolisib (GDC-0077)

Abstract

PIK3CA and KRAS are among the most frequently mutated oncogenes and often co-mutated in colorectal cancers. Understanding the impact of KRAS codon–specific mutations on cross-talks between the PI3K and MAPK pathways and response to targeted therapies, such as the p110α-specific inhibitor inavolisib (GDC-0077), is critical for advancing precision oncology. Focusing on colorectal PIK3CA + KRAS co-mutated models, we found that KRAS^G12D-mutated cells were more sensitive to inavolisib than models with KRAS^G13D, or other MAPK pathway mutations, even though the PI3K and MAPK pathways were active in both genotypes. In most co-mutated models, regardless of the type of KRAS alteration, the combination of inavolisib with MAPK pathway inhibitors showed synergy in vitro and in vivo. Our work highlights how specific codon substitutions in KRAS differentially toggle pathway activity and alter sensitivity to inavolisib, which could inform whether patients would benefit more from single-agent inavolisib or combination with MAPK pathway inhibitors.

Introduction

The PIK3CA gene, which encodes the p110α catalytic kinase subunit of the PI3K complex, is one of the most frequently mutated oncogenes across multiple cancer types. Mutations in PIK3CA occur in approximately 35% to 40% of hormone receptor–positive (HR⁺) breast cancers, 30% of gastrointestinal cancers, and 24% of ovarian cancers (1). Within The Cancer Genome Atlas and METABRIC datasets, about two thirds of PIK3CA mutations are considered hotspot mutations occurring in the kinase (predominantly at codon H1047) or helical domains (codons E542/E545) of p110α (1, 2). Oncogenic mutations in PIK3CA increase the kinase activity of p110α and downstream components of the PI3K pathway such as AKT and PRAS40, which ultimately leads to unchecked cell survival and proliferation (3). The MAPK signaling pathway comprises RAS isoforms (KRAS, NRAS, and HRAS), RAF isoforms (ARAF, BRAF, and CRAF), MAPK extracellular signal–regulated kinase (MEK), and extracellular signal–regulated kinase (ERK1/2). RAS proteins are GTPases which cycle between an active (GTP) and inactive (GDP) form to promote RAF protein dimerization and activation (4). KRAS point mutations occur in 30% to 50% of colorectal cancer, most often at the G12 and G13 codons (1, 5), and differentially influence downstream signaling (6–9). Many mechanisms and modes of cross-talk between the MAPK and PI3K signaling pathways have been unveiled (10), including cross-inhibition (11–13), cross-activation (14, 15), feedback (16) and pathway convergence (17, 18), and direct interactions between KRAS and p110α (19), which can affect response to pathway inhibitors (20–22).

Significant advances have been made in recent years toward targeting hyperactive PI3K and MAPK pathways in cancers (4, 18, 23). In HR⁺ HER2-negative (HER2⁻) breast cancer, the recent approval of the p110α inhibitors alpelisib and inavolisib and the AKT inhibitor capivasertib demonstrates the potential of inhibiting the PI3K pathway for treating PIK3CA-mutant HR⁺/HER2⁻ tumors (24–26). Although initial response of HR⁺/HER2⁻ tumors to targeted therapy is effective, the inhibition of one intracellular pathway is usually defeated by resistance mechanisms such as feedback signaling through parallel pathways (27–29). The development of p110α inhibitors in other diseases has been more challenging (30). For instance, KRAS-mutated lung cancers depend on both the PI3K and MAPK pathways (31, 32) and inhibition of MEK in lung cancer cells results in activation of the PI3K pathway (33). Hence, concomitant inhibition of both pathways may be required to achieve durable treatment efficacy.

As drug development efforts to target both pathways progress, it is critical to understand the mechanism of the cross-talk between the PI3K and MAPK pathways in mutated contexts. To this end, we used inavolisib (GDC-0077), an approved p110α-specific inhibitor, that also induces receptor tyrosine kinase–dependent degradation of mutated p110α (25, 34, 35). Focusing on the effects of mutations in MAPK pathway effectors on a PIK3CA-mutated background, we observed differences in the activity of the PI3K pathway and inavolisib response depending on whether cancer cells harbored a KRAS^G12D or KRAS^G13D mutation. By combining inavolisib with different MAPK pathway inhibitors, we measured strong combination benefits. Our results could support new clinical development opportunities for inavolisib, particularly for patients with colorectal cancer, a disease with significant co-occurrence of PIK3CA and KRAS mutations.

Materials and Methods

Cell lines

Cell lines were obtained from our in-house tissue culture cell bank. All cell lines inventoried in the cell bank underwent authentication by short tandem repeat profiling, SNP fingerprinting, and Mycoplasma testing. Cell lines were cultured in RPMI supplemented with 10% FBS, 100 units/mL penicillin, 2 mmol/L L-glutamine, and 1× HEPES at 37% under 5% CO₂. Details about vendors, gender, and the subculturing conditions for each cell line are provided in the Supplementary Table S1. Upon receipt from the in-house tissue culture cell bank, the cells were expanded two to three times for the experiments in this study. We followed the subculturing procedures outlined in Supplementary Table S1, using 0.5% trypsin for this expansion.

Some cell lines are related to each other. LS-180 is the parent of LS-174T and HM7, GP2d and GP5d are from the same patient, HCT 116 is the parent of ATRFLOX, and DLD-1 and HCT-15 are from the same patient. In experiments in which related cell lines are used, statistical analyses are done with all lines, as well as after averaging related lines.

Cell line authentication by short tandem repeat profiling

Short tandem repeat (STR) profiles were determined for each line using the Promega PowerPlex 16 System. This was performed once and compared with external STR profiles of cell lines (when available) to determine cell line ancestry. For loci analyzed, detection of 16 loci (15 STR loci and amelogenin for gender identification), including D3S1358, TH01, D21S11, D18S51, penta E, D5S818, D13S317, D7S820, D16S539, CSF1PO, penta D, AMEL, vWA, D8S1179, and TPOX, was performed.

Cell line quality control and SNP fingerprinting

SNP profiles were performed each time new stocks were expanded for cryopreservation. Cell line identity was verified by high-throughput SNP profiling using Fluidigm multiplexed assays. SNPs were selected based on minor allele frequency and presence on commercial genotyping platforms. SNP profiles were compared with SNP calls from available internal and external data (when available) to determine or confirm ancestry. In cases in which data were unavailable or cell line ancestry was questionable, DNA or cell lines were repurchased to perform profiling to confirm cell line ancestry. SNPs analyzed were rs11746396, rs16928965, rs2172614, rs10050093, rs10828176, rs16888998, rs16999576, rs1912640, rs2355988, rs3125842, rs10018359, rs10410468, rs10834627, rs11083145, rs11100847, rs11638893, rs12537, rs1956898, rs2069492, rs10740186, rs12486048, rs13032222, rs1635191, rs17174920, rs2590442, rs2714679, rs2928432, rs2999156, rs10461909, rs11180435, rs1784232, rs3783412, rs10885378, rs1726254, rs2391691, rs3739422, rs10108245, rs1425916, rs1325922, rs1709795, rs1934395, rs2280916, rs2563263, rs10755578, rs1529192, rs2927899, rs2848745, and rs10977980. All stocks were tested for Mycoplasma prior to and after the cells were cryopreserved. The method used was the Lonza MycoAlert Mycoplasma Detection Kit (LT07-318).

Cell growth rates and morphology were also monitored for any batch-to-batch changes. Cells were used for experiments within 3 weeks of receipt from the cell bank.

KRAS short hairpin RNA and gene rescue stable cell line generation

Short hairpin RNAs (shRNA) targeting KRAS 3′ untranslated region (UTR; GAT​CCG​TCT​GAT​ACA​ATA​TAC​GTC​TGC​TAG​TAG​TGA​AAT​ATA​TAT​TAA​ACT​AGC​AGA​CGT​ATA​TTG​TAT​CTT​ACG​GTA​C) were synthesized and cloned into the doxycycline (dox)-inducible PiggyBac transposase expression vector with constitutive mCherry expression.

In six-well culture plates, we transfected cells (∼2 × 10⁵) with 750 ng of transposon vector (KRAS 3′UTR shRNA1) and 250 ng of PiggyBac transposase. After 48 hours, we detached and re-plated cells in T175 flasks. Once confluent, we selected for mCherry expression by flow cytometry, and clones were expanded to create stable lines. The targeting efficiency of dox-induced knockdown (KD) was accessed by KRAS qRT-PCR using KRAS primers and probes (Applied Biosystems).

To rescue the KD phenotype, KRAS wild-type (WT), KRAS^G12D, and KRAS^G13D cDNA were synthesized and cloned into the dox-inducible PiggyBac transposase expression vector with constitutive blue fluorescent protein expression. These plasmids were transfected into the KRAS-KD cells using the same transfection method described above. The cells were selected for mCherry and blue fluorescent protein expression by flow cytometry and clones were expanded to create stable lines. The expression of KRAS was validated by qRT-PCR using KRAS primers and probes (Applied Biosystems).

Kinase inhibitors

Inavolisib (35), cobimetinib (36), GDC-0994 (37), and mutant KRAS^G12D inhibitor MRTX1133 (38) were synthesized at Genentech, Inc. In Fig. 1, the concentration is 1 μmol/L, which was used to improve the dynamic range of the responses across all cell lines. In Fig. 2 [phosphorylated PRAS40 (pPRAS40) and phosphorylated ERK (pERK) inhibition], we used higher concentration to ensure that all cell lines including less responsive ones were affected by inavolisib treatment. In Fig. 3A (GP2d cells, sensitive to inavolisib), we used a low (0.1 μmol/L) starting concentration. In Fig. 3B (HCC 116 cells, resistant to inavolisib), we used 10 μmol/L as the starting concentration. In the remaining experiments focused on cellular efficacy, we used 0.3 μmol/L to illustrate the combination benefits with the same concentrations across different cell lines.

Figure 1.

MAPK pathway mutations reduce inavolisib efficacy in PIK3CA hotspot-mutant cell lines. A, GR value for cell lines treated with 1 μmol/L of inavolisib split by PIK3CA mutational status. P values from one-sided rank sum tests. See Supplementary Table S2. B, GR value for PIK3CA hotspot-mutant cell lines treated with 1 μmol/L of inavolisib against MAPK pathway activity score. Color refers to MAPK pathway mutational status and shape to PIK3CA mutation. Spearman correlation and P value are shown. See Supplementary Table S2. C, GR value for PIK3CA hotspot-mutant cell lines treated with 1 μmol/L of inavolisib split by MAPK pathway mutations. P values from one-sided rank-sum tests. D, Prevalence and co-occurrence of PIK3CA and KRAS mutations in 373,069 tumor samples as detected in tissue biopsies in the Foundation Medicine’s FoundationCORE database. Only diseases with at least 1,000 samples and at least 5% prevalence of both PIK3CA and KRAS mutations are shown. ORs and P values are calculated by the Fisher exact test. E, Prevalence and co-occurrence of PIK3CA and KRAS mutations in 44,600 colorectal tumor samples classified by codon substitution. CRC, colorectal cancer; mut., mutant.

Figure 2.

Mutant KRAS regulates both PI3K and MAPK pathways. A, Baseline levels of pAKT (left, P value from a one-sided rank-sum test, n = 13) and pERK (right, P value from a rank sum test) normalized to total AKT and ERK levels measured by the Meso Scale Discovery Assay Kit in lysates of PIK3CA + KRAS co-mutated cells. Data split by KRAS mutational status (see Supplementary Table S3). B, Inhibition of pPRAS40 (left) and pERK (right) levels measured by the Meso Scale Discovery Assay Kit in lysates from cell lines treated with 5 μmol/L of inavolisib for 1 hour. Data split by KRAS mutational status (see Supplementary Table S3). C, Levels of PI3K and MAPK pathway markers measured by Western blotting in lysates from PIK3CA + KRAS co-mutated (heterozygous for KRAS) colorectal cancer cells transfected with a control siRNA or a combination of KRAS, HRAS, and NRAS siRNAs. Cells were harvested after 72 hours. Representative results of n = 2 experiments. D, p110α level after ubiquitinated (Ub) protein was pulled down and levels of PI3K and MAPK pathway markers in subcellular fractionation analyzed by Western blotting in HCC1954 PIK3CA-mutant isogenic cells transfected with gene plasmids expressing either WT KRAS, KRAS^G12D, or KRAS^G13D. Cells were harvested 48 hours after transfection. a.u., arbitrary unit.

Figure 3.

Exogenous expression of KRAS^G12D or KRAS^G13D alters response to inavolisib. A and B, Confluence relative to DMSO for (A) GP2d and (B) HCT 116 cells transfected with dox-inducible shRNA against endogenous KRAS alleles and treated with different concentrations of inavolisib, either with or without dox. Left, Growth curves from a representative experiment for each engineered model of GP2d (A) and HCT 116 (B) with dox-inducible shKRAS. Plain lines show the no dox condition and dashed line the plus dox condition. Colors show different concentrations of inavolisib. Vertical black lines show the cutoff used for calculation of relative confluency. Right, Results from n = 3 experiments aggregated. C and D, Confluence relative to DMSO for (left) GP2d and (right) HCT 116 cells transfected with dox-inducible shRNA against endogenous KRAS alleles and exogenous expression of either (C) KRAS^G12D or (D) KRAS^G13D treated with different concentrations of inavolisib and either with or without dox (n = 2 for GP2d; n = 3 for HCT 116). See Supplementary Fig. S3E and S3F for representative growth curves. KI, knock-in.

Antibody reagents

Antibodies to p110a (4249), phospho-AKT-S473 (4060), pS6 S235/236 (2211), total AKT (2920), phospho-ERK-Thr202/Tyr204 (4370), total ERK (4695), phospho-MEK1/2 (2338), total MEK1/2 (8727), phospho-PRAS40-Thr246 (2997), cleaved PARP (5625), and GAPDH (5174) were obtained from Cell Signaling Technology. The antibody to β-actin (A5441) was purchased from Sigma-Aldrich. The antibodies to RAS (ab52939), p85a (ab133595), and p85b (ab28356) were obtained from Abcam. The ubiquitin reagent TUBE1 (UM101) was obtained from LifeSensors.

Immunoblotting

One million cells were plated in each of a four-well plate the day before the treatment. After incubating cells with the indicated concentration of inavolisib or cobimetinib for the time indicated, the cells were harvested with 1× Cell Extraction Buffer (Invitrogen) supplemented with protease inhibitors and phosphatase inhibitors from Roche. The Bicinchoninic Acid Protein Assay Kit (Pierce) was used to measure protein concentration. From each treatment, 20 μg of protein was separated by electrophoresis through NuPAGE Bis-Tris 4% to 12% gradient gels (Invitrogen); proteins were transferred onto nitrocellulose membranes using the iBlot system (Invitrogen).

pERK, phosphorylated AKT, and pPRAS40 ELISA assay

Cells were plated in 96-well tissue culture–treated assay plates. The following day, test compounds were serially diluted in DMSO and added to cells (final DMSO concentration of 0.5%). The cells were then incubated with drugs for 1 hour and lysed; pERK, phosphorylated AKT (pAKT), or pPRAS40 levels were quantified using the Meso Scale Discovery Assay Kit according to the product manual.

siRNA transfection

Transfection of siRNA was carried out using Lipofectamine RNAiMAX reagent (Thermo Fisher Scientific) 72 hours in advance of drug treatment.

KRAS plasmid transient transfection

To assess the effect of mutant KRAS on the PI3K pathway, the HCC1954 (PIK3CA^H1047R, HER2+, breast) line was transiently transfected with pRK5 flag KRAS WT, KRAS^G12D, and KRAS^G13D plasmids using Lipofectamine 3000 (Thermo Fisher Scientific). Transfection was performed using methods described by the vendor. At 48 hours after transfection, the cells were harvested for subcellular fractionation and ubiquitin pull-down assay described in the methods.

Subcellular fractionation

Cells were washed once with PBS before scraping into 0.8 mL/dish containing hypotonic lysis buffer (25 mmol/L Tris–HCl pH7.5, 10 mmol/L NaCl, 1 mM EDTA, and protease and phosphatase inhibitors). The cells were lysed by 30 strokes in a Dounce homogenizer and subjected to centrifugation at 1,500 g (3,000 RPM) for 5 minutes to pellet nuclei and unbroken cells, followed by centrifugation of the supernatant at 100,000 g (44,000 RPM) in TLA55 rotor for 40 minutes. The supernatant (800 μL) was collected (S100 fraction) and the pellet was resuspended in 200 μL hypotonic lysis buffer plus 1% NP40 (P100 fraction). The resuspended pellet was centrifuged for 5 minutes at high speed in a microfuge, and the supernatant was collected.

Ubiquitin pull-down assay

Cells were lysed in 20 mmol/L Tris–HCl pH 7.5, 137 mmol/L NaCl, 1 mmol/L EDTA, 1% NP40, and 10% glycerol plus protease and phosphatase inhibitors. For the ubiquitinated protein pull-down experiment, the cells were lysed in a lysis buffer containing 200 μg/mL TUBE1 (LifeSensors UM101). Lysates were isolated and 50 μL of glutathione agarose beads was added to them (Sigma, G4705). The samples were incubated overnight and the captured ubiquitinated protein was eluted in SDS-reducing sample buffer.

Foundation Medicine cancer genomic data

Comprehensive genomic profiling using hybrid capture-based next-generation sequencing was performed on tumor tissue specimens (N = 373,069) submitted to Foundation Medicine during routine clinical care. Assays were designed to identify all classes of alterations in at least 324 genes (39). PIK3CA and KRAS likely and known alterations were considered for our analyses.

Cell viability assays

In 384-well plates, 1,000 cells/well were seeded in a volume of 45 μL the day before the treatment and incubated at 37°C under 5% CO₂. Compounds were added with 5 μL of media to each well such that the desired concentration was achieved. Cell viability was assessed using CellTiter-Glo (Promega) after 5 days of incubation. Cell viability data were normalized using the growth rate inhibition (GR) method based on a reference division time as described by Hafner and colleagues (40). Combination benefit was assessed by excess over single agent and synergy by excess over Bliss independence (41). The caspase-3 level was determined using Caspase-3 assay (Promega) after 2 days of incubation. Total luminescence was measured on an EnVision Plate Reader (PerkinElmer).

IncuCyte (quantitative measurement of cell proliferation)

Cells were plated in RPMI medium supplemented with 10% FBS, 100 units/mL penicillin, 2 mmol/L L-glutamine, and 1× HEPES in the presence or absence of 0.2 μg/mL dox. After 72 hours, the cells were re-plated (2,000–4,000 cells/well) in 96-well plates and incubated overnight. The next day, either DMSO or inavolisib was added at the indicated concentrations. Proliferation rates based on cell confluence were determined by live cell imaging using IncuCyte. For each condition (clone with or without dox), the time at which the DMSO-treated samples reach 80% confluency was used as reference (or the last time point if 80% confluency was not reached). At this time point, the confluency of the inavolisib-treated conditions was normalized against the DMSO-treated condition.

Colony formation assays

Twelve-well plates were seeded with 10,000 to 20,000 cells/well in a volume of 1 mL depending on the cells’ doubling time and incubated at 37°C under 5% CO₂ for 24 hours. Compounds were added the next day at indicated concentration for 7 days. The medium was replaced with fresh medium with drug for another week of incubation if needed. Then the medium was aspirated and 100 μL of 0.5% crystal violet was added for 15 minutes. The dye was discarded and the cells were washed with PBS for 5 to 10 minutes five times. PBS was then aspirated and the cells were imaged using GelCount (Scintica).

In vivo efficacy studies

Female NCR nude and BALB/c nude mice were obtained from Taconic Biosciences and Charles River Laboratories, respectively. Mice were housed at Genentech according to standards established by the Institutional Animal Care and Use Committee. All in vivo studies were approved by Genentech and Institutional Animal Care and Use Committee and adhered to the NIH Guidelines for the Care and Use of Laboratory Animals. HCT 116 and LS-174T colon cell lines were cultured in vitro, harvested in log-phase growth, and resuspended in Hank’s Balanced Salt Solution containing Matrigel (BD Biosciences) at a 1:1 ratio by volume for in vivo inoculation. Female NCR nude mice were subcutaneously inoculated with 5 × 10⁶ HCT 116 cells and BALB/c nude mice with 2 × 10⁶ LS-174T cells in the right flank in a volume of 100 μL. Tumors were allowed to grow to a volume in an initial range before mice were randomized to treatment groups at the start of dosing to create closely matched baseline average tumor sizes across regimens.

Mice were given cobimetinib (5 mg/kg, 3× per week) and inavolisib (25 mg/kg, daily) in 0.5% methylcellulose and 0.2% Tween 80 solution by oral gavage. Tumor sizes and mouse body weights were recorded twice weekly over the course of the study. Tumor volumes were measured in two perpendicular dimensions (length and width) using Ultra-Cal IV calipers (model 54 − 10 − 111; Fred V. Fowler Co.) and calculated according to the formula tumor size (mm³) = (longer measurement × shorter measurement²) × 0.5. Animal body weights were measured using an Adventurer Pro AV812 scale (Ohaus Corporation). Percent animal weight change was calculated using the formula body weight change (%) = [(current body weight/initial body weight)−1 × 100].

Analyses and comparisons of tumor growth were performed using a package of customized functions in R (version 3.6.2; R Foundation for Statistical Computing), which integrates software from open-source packages as described in Forrest and colleagues (42). Growth contrast (GC) represents the difference in AUC-based growth rates (endpoint gain integrated in time) between the treatment and reference groups. The more negative the GC value, the greater the antitumor effect. The 95% confidence intervals are based on the fitted model and variability measures of the data.

Single-cell analysis

GP2d, LS-174T, and HCT 116 cells were seeded to six-well plates. All samples were treated with indicated compound combinations starting at different days and harvested together. Upon sample collection, the cells were trypsinized from the plate and hashed by using the cholesterol-modified oligo (CMO) following the MULTI-seq protocol (43). Briefly, the cells from each sample were washed and resuspended in 180 μL of PBS + 1% BSA and transferred to a round-bottom 96-well plate. Then 20 μL of each anchor–barcode mix was added to the suspension for a final concentration of 2 μmol/L. After the incubation for 15 minutes at room temperature, 20 μL of co-anchor was added to each sample (final concentration of 2 μmol/L) and incubated for another 15 minutes at room temperature. The cells were then washed by 500 μL of cold PBS + 1% BSA twice before pooling together. In total, we used 32 different barcodes for 64 samples (32 different conditions, each with two replicates). The same replicate from different samples (32 in total) was pooled into one large cell pool (two pools in total) and loaded on the FACS to sort for the live cells.

After sorting, the cells were loaded on the 10x Genomics Chromium X machine with the Chromium Single Cell 3′ V3.1 Kits. In total, six lanes (three lanes per cell pool) were loaded with the target cell recovery of 20,000 cells/lane. The downstream library construction was performed following the manufacturer’s manual with a few modifications to accommodate the MULTI-seq protocol (41): (i) during the cDNA amplification, 1 μL of the 5′-CTTGGCACCCGAGAATTCC-3′ primer (10 μmol/L) was added to the master mix; (ii) the supernatant from the 0.6× solid phase reversible immobilization beads cleanup for cDNA was kept and further cleaned up by using 2.0× solid phase reversible immobilization beads; and (iii) the CMO libraries were made by performing eight cycles of the indexing PCR using the supernatant. The transcriptome and CMO libraries were then sequenced by using either the Illumina NovaSeq or NextSeq machines.

Demultiplexing and read alignment were performed using CellRanger and processed with standard parameters. For each cell line, the cells with at least 7,000 reads were selected and processed using the gpsa pipeline. MAPK signature score was based on 10 genes and the E2F target signature score on the Hallmark signature from MSigDB v7.0. Plots were performed in R.

Results

MAPK pathway mutations alter the phenotypic response to inavolisib

We first compared the response of a large panel of cancer cell lines to inavolisib using the GR method (44) to avoid biases due to differences in nominal growth rates among cell lines. Across a panel of 379 cell lines from multiple diseases, inavolisib effectively reduced the growth rate of cell lines harboring a PIK3CA hotspot mutation (defined here as a substitution mutation in codons E542, E545, or H1047) as reflected by a median GR value of 0.34 at 1 μmol/L. GR trended to be weaker in cell lines expressing non-hotspot PIK3CA mutations with a median GR value of 0.55 (P = 0.069, rank-sum test), whereas PIK3CA WT cell lines were the least sensitive (median GR value of 0.78, Fig. 1A; Supplementary Table S2). We also observed the same differences in sensitivity to inavolisib based on PIK3CA status in a panel of more than 700 cell lines analyzed with the PRISM assay (Supplementary Fig. S1A; ref. 45). Furthermore, we observed significant differences in dependency on the PIK3CA gene between cell lines with PIK3CA hotspot mutations, non-hotspot mutations, and WT PIK3CA as quantified by the Chronos score of the Achilles dataset (Supplementary Fig. S1B; ref. 46). These results confirmed that most cell lines with PIK3CA hotspot mutations are highly dependent on PI3K signaling, whereas those with non-hotspot PIK3CA mutations are less so, and most lines with WT PIK3CA are not dependent on PI3K signaling.

Focusing on PIK3CA hotspot-mutant cell lines, we observed a wide range of GR values from negative (cytotoxic response) to one (no effect) and found a significant correlation with those lines’ baseline MAPK pathway activity score (Spearman ρ = 0.38, P = 0.019; Fig. 1B; ref. 47). Splitting PIK3CA hotspot-mutant cell lines by MAPK pathway mutational status revealed that cell lines with mutations in the MAPK pathway were significantly more resistant to inavolisib (median difference in GR value of 0.55, P = 3.9e–5, rank-sum test; see Supplementary Table S2) compared with those without MAPK pathway gene mutations. Strikingly, among PIK3CA + KRAS co-mutated cell lines, KRAS^G12D-mutant lines were more responsive to inavolisib than those with any other MAPK pathway mutations (P = 0.015, rank-sum test; P = 0.088 when averaging lines established from the same patient; Fig. 1C). This difference in sensitivity to inavolisib was further confirmed by the PRISM data and the Chronos scores for PIK3CA dependency (P = 0.036 and P = 0.033, respectively, one-sided rank-sum test; Supplementary Fig. S1C and S1D).

In analyzing Foundation Medicine’s FoundationCORE database of real-world patients with cancer whose tumor samples underwent comprehensive genomic profiling, we found that, in contrast to all solid tumors, PIK3CA and KRAS mutations were most likely to co-occur in colorectal (OR = 2.07, P = 7.0e–190, Fisher exact test) and gastric cancers (OR = 1.47, P = 1.4e–4; Fig. 1D). Among colorectal cancer samples with PIK3CA hotspot mutations, 22% had a KRAS^G12D mutation (Fig. 1E). Given that no clinical data with inavolisib in patients with KRAS-mutant tumors were available, we interrogated available data from the first-in-human phase Ia trial of the p110α inhibitor alpelisib (48). Among the patients with tumors harboring both a PIK3CA mutation and MAPK pathway gene mutation, alpelisib led to tumor regression in two patients harboring KRAS^G12D co-mutations (Supplementary Fig. S1E). Although this result will have to be validated in larger studies, it suggested that PI3KCA-mutant tumors with KRAS^G12D mutation might tend to be more sensitive to p110α inhibition than those harboring BRAF or other KRAS mutations.

Mutant KRAS differentially regulates both the PI3K and MAPK pathways

To study the cross-talk between the PI3K and MAPK pathways when both pathways harbor mutations, we selected 13 cell lines, a majority of which are colorectal, with PIK3CA + KRAS co-mutations (either G12D or G13D) and no other RAF/RAS mutation (Supplementary Table S3). First, we observed that baseline pAKT expression normalized to total AKT expression was significantly higher in lines with a KRAS^G12D mutation compared with those with a KRAS^G13D mutation [Fig. 2A (left); P = 0.011, rank-sum test; P = 0.1 when averaging related cell lines]. In contrast, no significant difference was observed in pERK levels normalized to total ERK levels between KRAS^G12D- or KRAS^G13D-mutated lines [Fig. 2A (right); P = 0.73]. Next, we found that high doses of inavolisib treatment completely suppressed PI3K signaling (as measured by phosphorylation of PRAS40) in all PIK3CA-mutant cell lines regardless of KRAS mutational status [P = 0.69, Kruskal–Wallis test, Fig. 2B (left)], but MAPK signaling (as measured by ERK phosphorylation) was unchanged in PIK3CA-mutant cell lines harboring a KRAS mutation [P = 0.037, Kruskal–Wallis test, Fig. 2B (right)]. This contrasted with our results in PIK3CA-mutant cell lines with WT MAPK pathway from other diseases, in which inavolisib reduced MAPK signaling in addition to fully suppressing PI3K signaling (Fig. 2B; Supplementary Fig. S2A).

We next asked whether mutant KRAS regulated the PI3K pathway in PIK3CA + KRAS co-mutated cells by knocking down all RAS isoforms. From the pan-RAS KD, we found that both pERK and pAKT are reduced in all cell lines (Fig. 2C), although the magnitude of downregulation could be affected by the efficiency of the RAS KD. This contrasted with PIK3CA-mutant RAS-WT cell lines from other diseases in which pan-RAS KD reduced pERK levels but had no effect on pAKT levels independently of the type of PIK3CA mutation (Supplementary Fig. S2B). To confirm that mutant KRAS affects the activity of mutant p110α, we measured p110α ubiquitination, which occurs at the membrane and is a consequence of p110α activation (34). To differentiate the effect of WT versus mutant KRAS, we transfected p110α-mutant HCC1954 cells with either KRAS WT, G12D-mutant, or G13D-mutant plasmids. We found that KRAS^G12D and KRAS^G13D, but not WT KRAS, induced ubiquitination of membrane-associated mutant p110α (Fig. 2D), indicating that mutant KRAS can recruit p110α and activate it. Overall, these data suggested that mutant KRAS regulates both the PI3K and MAPK signaling pathways in PIK3CA-mutant cell lines, whereas WT KRAS does not.

Exogenous mutant KRAS expression confirms signaling differences between KRAS^G12D and KRAS^G13D mutations

To validate the differential effect of KRAS codon-specific mutations on pathway signaling and inavolisib response, we knocked down the endogenous KRAS alleles in two PIK3CA + KRAS co-mutated cell lines with a dox-inducible shRNA targeting the 3′UTR (Supplementary Fig. S3A–S3D). Strikingly, KRAS KD only minimally affected inavolisib response in KRAS^G12D-mutant GP2d cells as measured by confluence during a time-course experiment (Fig. 3A; Supplementary Fig. S3E). In contrast, KRAS KD in KRAS^G13D-mutant HCT 116 cells enhanced sensitivity to inavolisib (Fig. 3B; Supplementary Fig. S3F). In addition, the introduction of exogenous KRAS^G12D after knockdown of the endogenous KRAS alleles had limited effect on inavolisib response in GP2d and marginally sensitized HCT 116 cells (Fig. 3C; Supplementary Fig. S3E and S3F) similar to KRAS KD alone. The expression of exogenous KRAS^G12D was also associated with increased pAKT levels (Supplementary Fig. S3C and S3D). In contrast, exogenous expression of KRAS^G13D after knockdown of the endogenous KRAS alleles led to weaker response to inavolisib in GP2d and HCT 116 cells (Fig. 3D; Supplementary Fig. S3E and S3F). These results suggested an association between KRAS codon-specific mutations, pAKT levels, and inavolisib response.

In summary, our results highlighted a hierarchy of pathway cross-talk and inavolisib response among PIK3CA hotspot-mutant lines: Lines with no mutation in the MAPK pathway exhibit the greatest sensitivity to inavolisib because of limited cross-talk from KRAS proteins to the PI3K pathway and weaker MAPK pathway activity upon inavolisib treatment. In KRAS-mutant lines, KRAS can affect PI3K signaling and inavolisib is not inhibiting MAPK signaling. In terms of proliferation, KRAS^G12D-mutant lines exhibited intermediate sensitivity to inavolisib, and cell lines with other MAPK pathway mutations such as KRAS^G13D were the least responsive.

MAPK pathway inhibitors synergize with inavolisib in PIK3CA + KRAS co-mutated cell lines

To further characterize the differential impact of inhibition of the PI3K and MAPK pathways on KRAS^G12D- and KRAS^G13D-mutant cells, we performed single-cell RNA sequencing of PIK3CA/KRAS co-mutated cells treated with either inavolisib, cobimetinib (GDC-0973, a MEK inhibitor), or their combination for 24 hours. We chose 0.3 μmol/L for each drug to ensure a robust signal across all tested lines. The Uniform Manifold Approximation and Projection showed regions of high proliferative cells (dark color reflecting high E2F score) for each cell line. The overlaid contour plots show the distribution of the cells for individual conditions. Changes in the location of the contour revealed that inavolisib has a strong effect on the KRAS^G12D-mutant GP2d and LS-174T cells but a minimal effect on the KRAS^G13D-mutant HCT 116 cells (Fig. 4A). Cobimetinib reduced the MAPK pathway activity score across the three cell lines (Fig. 4B) but strikingly, had a limited effect on the transcriptional profile of KRAS^G12D-mutant cells as shown by the overlap of the contour of cobimetinib-treated cells with the one of DMSO-treated cells in the Uniform Manifold Approximation and Projection. In contrast, cobimetinib induced a major shift in the transcriptional profile of KRAS^G13D-mutant HCT 116 cells (Fig. 4A). Distribution of the E2F target signature scores—a proxy for proliferation—confirmed the sensitivity difference between KRAS mutations: inavolisib had the strongest effect on E2F scores in KRAS^G12D-mutant cells whereas cobimetinib was most efficacious in KRAS^G13D-mutant HCT 116 cells (Fig. 4C). These results further showed that the cell transcriptional state of KRAS^G12D-mutant cells is more dependent on PI3K signaling than MAPK signaling, whereas KRAS^G13D-mutant cells show the opposite.

Figure 4.

Combination treatment of inavolisib with cobimetinib is synergistic. A, Distribution of cells in the UMAP from single-cell RNA sequencing data for (left) GP2d, (middle) LS-174T, and (right) HCT 116 cells treated with DMSO, inavolisib, cobimetinib, or their combination for 24 hours. Underlying heatmap shows the averaged E2F target score for all cells independently of treatment. UMAP, Uniform Manifold Approximation and Projection. B, Distribution of MAPK pathway activity scores for single-cell RNA sequencing data split by treatments and cell lines. C, Distribution of E2F target scores for single-cell RNA sequencing data split by treatments and cell lines. D, Excess over Bliss score for PIK3CA + KRAS co-mutated cell lines treated with inavolisib and cobimetinib plotted against MAPK pathway activity score. Color refers to MAPK pathway mutational status and shape to PIK3CA mutation. Spearman correlation and P value are shown. CRC, colorectal cancer; mut., mutant. E, Representative images of crystal violet assay for PIK3CA-mutant cell lines with indicated KRAS mutation treated with either DMSO, 0.3 μmol/L inavolisib, 0.3 μmol/L cobimetinib, or a combination of the two drugs for 2 to 3 weeks. F, Fitted tumor volumes for LS-174T xenograft tumors treated with either vehicle, 25 mg/kg of inavolisib daily, 5 mg/kg of cobimetinib thrice a week, or a combination of the two drugs for 3 weeks (n = 8 per group). G, Fitted tumor volumes for HCT 116 xenograft tumors treated with either vehicle, 25 mg/kg of inavolisib daily, 5 mg/kg of cobimetinib thrice a week, or a combination of the two drugs for 3 weeks (n = 8 per group).

Single-cell data revealed that E2F target scores were heterogeneous upon single-agent treatment with either inavolisib or cobimetinib alone but the combined treatment induced a reduction of proliferation across the whole cell population (Fig. 4C), suggesting a combination benefit. As validation, we measured the growth inhibition of 11 colorectal cancer lines with PIK3CA + KRAS mutations and observed synergy (excess over Bliss above 0.15) in all tested lines but one (Fig. 4D). For example, at a single-agent dose of 0.33 μmol/L of cobimetinib or 0.37 μmol/L of inavolisib, all cell lines had GR values above 0.2 (partial growth inhibition). When combined, six of 11 cell lines had GR values below 0.1, reflecting stasis or cytotoxic response (Supplementary Fig. S4A). The strongest synergies (measured by the average excess over Bliss) were observed in lines with the highest MAPK pathway activity scores (Spearman ρ = 0.71, P = 0.019; Fig. 4D), most of which were resistant to inavolisib. The strong proliferation inhibition at submicromolar doses of inavolisib and cobimetinib was confirmed using a crystal violet clonogenic assay (Fig. 4E) and the cytotoxic response with a caspase-3/7 activity assay, which showed increased activity levels upon combination in four of five colorectal cancer cell lines (Supplementary Fig. S4B).

To systematically identify drugs that showed increased efficacy over single agent when combined with inavolisib in PIK3CA + KRAS^G13D–mutant cell lines, we screened a library of 720 small-molecule tool compounds as single agents and in combination with 1 μmol/L of inavolisib. We selected three colorectal PIK3CA + KRAS^G13D–mutant cell lines resistant to inavolisib (HCT-15, HCT 116, and T84) and used the PIK3CA-mutant, KRAS WT breast cancer cell line MFM-223 as a control. MAPK pathway inhibitors including cobimetinib showed the strongest combination effect across all KRAS^G13D-mutant lines (FDR <0.002; median increase of GR area over the curve around 0.3; see Supplementary Fig. S4C for HCT 116 and Supplementary Dataset S1 for other lines) but had limited combination effects for the MFM-223 (KRAS WT) cell line (Supplementary Dataset S1). We validated the combination effect in PIK3CA + KRAS co-mutated lines using the ERK inhibitor GDC-0994 (37) and the KRAS^G12D-specific inhibitor MRTX1133 (38). GDC-0994 showed excess over single agent above 0.15 in the majority of lines (six of seven tested; Supplementary Fig. S4D), independently of the type of KRAS mutation, whereas MRTX1133 showed combination benefit with inavolisib only in the KRAS^G12D-mutant lines as expected (Supplementary Fig. S4E). We thus consistently observed a strong combination benefit across colorectal cancer PIK3CA + KRAS co-mutated cell lines when combining inavolisib with a MAPK pathway inhibitor.

Combination of inavolisib and cobimetinib shows increased efficacy in colorectal cancer tumor xenograft models

To validate our findings in vivo, we investigated the antitumor efficacy of inavolisib alone (25 mg/kg, daily) and in combination with cobimetinib (5 mg/kg, thrice a week) in two xenograft models: LS-174T (PIK3CA and KRAS^G12D mutated; Fig. 4F; Supplementary Fig. S5A) and HCT 116 (PIK3CA and KRAS^G13D mutated; Fig. 4G; Supplementary Fig. S5B). Group-level efficacy estimates were obtained using the GC metric achieved through comparing AUC-based growth rates of drug-treated tumors with vehicle-treated tumors (42). In the LS-174T model, single-agent cobimetinib effects on tumor growth were weak (−0.047 GC) but inavolisib significantly reduced tumor growth (−0.110 GC). When combined, tumor growth was further reduced (−0.148 GC; Supplementary Fig. S5A and S5C). On the other hand, the HCT 116 model showed a modest reduction in tumor growth when treated with cobimetinib (−0.009 GC) or inavolisib (−0.032 GC) compared with vehicle control. The combination of inavolisib and cobimetinib resulted in a significant antitumor effect which exceeded the effect observed with either agent alone (−0.088 GC; Supplementary Fig. S5B and S5D). For these studies, no sample was collected for pharmacodynamic biomarkers analysis. Treatment was well tolerated in all groups; the observed body weight loss in most cases was less than 5% for both models (Supplementary Fig. S5E and S5F). Taken together, these results confirmed that inavolisib more strongly inhibits growth in PIK3CA + KRAS^G12D–mutated than PIK3CA + KRAS^G13D–mutated tumors. Additionally, treatment with inavolisib with a concomitant inhibition of the MAPK pathway further reduced growth in PIK3CA + KRAS^G12D–mutated tumors and addressed weak response to inavolisib in PIK3CA + KRAS^G13D–mutated tumors.

Discussion

In summary, we used inavolisib, a p110α-selective inhibitor, to identify the codon-specific effects of PIK3CA and KRAS mutations on the activity and dependency of the PI3K and MAPK pathways. Although WT RAS regulates the MAPK pathway and not the PI3K pathway, mutant KRAS regulates both pathways in a nuanced manner: the KRAS^G12D mutation is associated with higher pAKT levels compared with the KRAS^G13D mutation. Although most mutations in MAPK pathway genes reduce sensitivity to inavolisib in PIK3CA hotspot-mutant cell lines, we found that the KRAS^G12D mutation is an outlier as PIK3CA + KRAS^G12D co-mutated cell lines were more sensitive to inavolisib compared with other variants and this inavolisib sensitivity can be further enhanced with the KRAS^G12D-specific inhibitor MRTX1133. In contrast, inavolisib had limited effect as a single agent on PIK3CA + KRAS^G13D co-mutated cell lines, but when combined with the MEK inhibitor cobimetinib or the ERK inhibitor GDC-0994, the effect was synergistic and overcame weak response in vitro and in vivo. Although our experiments with exogenous expression of KRAS-mutant proteins in GP2d and HCC 116 support a causal relationship between KRAS codon–specific mutation and inavolisib response, analyses of the response to combination treatments were correlative and would require further mechanistic validations, especially in vivo studies, to enable further translation of these findings.

Taken together, our data uncovered that, in the context of PIK3CA + KRAS co-mutations, MAPK pathway activity is independent of mutant p110α activity and inavolisib treatment. PIK3CA + KRAS^G12D co-mutated cells show higher AKT phosphorylation than PIK3CA + KRAS^G13D–mutated cells and are still dependent on the PI3K pathway (Fig. 5A), whereas the occurrence of a KRAS^G13D mutation can lead to resistance to inavolisib (Fig. 5B). At the phenotypic level, targeting of a single pathway is not enough to inhibit E2F activity and fully suppress proliferation across PIK3CA + KRAS co-mutated cells likely because of inherent heterogeneity of the response. We found that only the concomitant inhibition of both pathways leads to the strongest effect on tumor proliferation.

Figure 5.

Model of cross-talk between PI3K and MAPK pathways in PIK3CA + KRAS co-mutated tumors based on KRAS codon substitution. A, G12D-mutant KRAS affects both PI3K and MAPK pathways. PI3KCA + KRAS^G12D–mutated cells depend on both PI3K and MAPK pathways. B, G13D-mutant KRAS affects both PI3K and MAPK pathways. PI3KCA + KRAS^G13D co-mutated cells depend primarily on the MAPK pathway.

Our in vitro and in vivo findings have several translational implications. As expected, inavolisib monotherapy resulted in significant efficacy in PIK3CA-mutated tumors with WT MAPK pathway and, less expected, also in PIK3CA + KRAS^G12D co-mutated tumors (Fig. 1C), which represent 22% of PIK3CA + KRAS co-mutated colorectal cancer tumors (Fig. 1E). In addition, we showed that PIK3CA + KRAS^G13D co-mutated tumors (12% of PIK3CA + KRAS co-mutated colorectal cancer tumors) were relatively insensitive to inavolisib alone but the response could be enhanced when inavolisib was combined with inhibitors of the MAPK pathway. These observations, along with the trend found in the alpelisib trial data, provide useful insights toward identifying patients who could benefit from inavolisib monotherapy or a combination with a MAPK pathway inhibitor based on codon-specific KRAS mutations. The clinical strategy for combined inhibition of p110α and MEK requires further investigation as tolerability remains a challenge, but this could potentially be addressed by intermittent treatment with the MEK inhibitor (49) or the use of mutant-specific KRAS inhibitors (38, 50), which are currently at various stages of clinical development.

In conclusion, we found that not only mutations of oncogenes can substantially change pathway activity and cross-talk between pathways but that the specific codon substitutions matter in both signaling and dependencies. Although this work was focused on the role of KRAS mutations, further investigation into the impact of the different PIK3CA mutations may also provide more granularity to our understanding of the interactions between the PI3K and MAPK pathways. Given the recent advances in the development of mutant-specific KRAS inhibitors, our work provides new avenues for the clinical development of inavolisib and may influence the design of future clinical trials in diseases with a high co-occurrence of PIK3CA and KRAS mutations such as colorectal cancers, thus contributing to the advancement of precision medicine in oncology.

Supplementary Material

Supplemental Figure 1

Supplemental Figure 1: MAPK pathway mutations reduce inavolisib efficacy in PIK3CA hotspot mutant cell lines

Supplemental Figure 2

Supplemental Figure 2: Effect of inavolisib and RAS knock-down on AKT and MAPK pathway activity in RAS WT cell lines

Supplemental Figure 3

Supplemental Figure 3: GP2d and HCT 116 models expressing exogenous mutant KRAS have different sensitivity to inavolisib

Supplemental Figure 4

Supplemental Figure 4: Combination treatment of inavolisib with cobimetinib is synergistic

Supplemental Figure 5

Supplemental Figure 5: Combination treatment of inavolisib with cobimetinib in vivo

Supplemental Table 1

Supplemental Table 1: References for reagents and material

Supplemental Table 2

Supplemental Table 2: Cell line properties and inavolisib viability data

Supplemental Table 3

Supplemental Table 3: Cell line properties and levels of PI3K and MAPK signaling proteins

Supplemental Dataset 1

Supplemental Dataset 1 contains the difference in GR AOC values in the chemicogenomic screen for CRC cell lines HCT 116 (Supplemental Figure 4C), HCT-15, and T84 as well as Breast Cancer cell line MFM-223.

Data Availability

Raw data and count matrix are accessible through the Gene Expression Omnibus portal (series GSE248412; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE248412). Difference in GR area over the curve values in the chemicogenomic screen for colorectal cancer cell lines HCT 116, HCT-15, T84, and MFM-223 is available in Supplementary Dataset S1.

Authors’ Disclosures

K.W. Song, C.C. Ong, E. Lin, K.E. Hutchinson, S. Xie, J. Tan, Y. Liang, Z. Modrusan, D. Maddalo, M. Hafner, and A. Dey report employment with Genentech and ownership of Roche stock. N.M. Sodir reports other support from Revolution Medicines outside the submitted work. D.X. Jin reports personal fees from Foundation Medicine and other support from Roche Holding AG outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

K.W. Song: Conceptualization, validation, investigation, writing–original draft, writing–review and editing. C.C. Ong: Investigation, writing–original draft. E. Lin: Investigation, writing–original draft. J. Lau: Investigation, writing–original draft. N.M. Sodir: Investigation, writing–original draft. D.X. Jin: Formal analysis. K.E. Hutchinson: Writing–original draft, writing–review and editing. S. Xie: Resources, writing–original draft. J. Tan: Investigation, writing–original draft. Y. Liang: Investigation, writing–original draft. Z. Modrusan: Resources, writing–original draft. S.E. Martin: Resources, writing–original draft. D. Maddalo: Resources, writing–original draft. M. Hafner: Conceptualization, data curation, formal analysis, visualization, writing–original draft, writing–review and editing. A. Dey: Conceptualization, resources, writing–original draft, writing–review and editing.