Differences in Signaling Patterns on PI3K Inhibition Reveal Context Specificity in KRAS-Mutant Cancers

This article is featured in Highlights of This Issue, p. 1335

There are multiple examples of targeted anticancer drugs that are clinically effective in targeting different cancers driven by the same oncogene including the recent report of the activity of Tropomyosin Receptor Kinase (TRK) inhibitors across a wide range of tumors driven by TRK fusions (1). However, there is emerging evidence related to context specificity where differences in signaling in different tumor types driven by the same oncogene can result in disparate clinical outcomes; for example, the BRAF inhibitor, vemurafenib, causes clinical responses in patients with V600E-mutant BRAF-driven melanoma (2) but not colorectal cancer (3). This has been attributed to differences in EGFR signaling between the two tumor types (4).

Mutations in the oncogene KRAS are seen across a wide range of solid tumors, such as pancreatic cancer (97%), colon cancer (40%), and non–small cell lung cancer (NSCLC; 30%). Multiple treatment approaches to target KRAS have been proposed including post-translational modification (farnesyl transferase inhibitors), combinatorial inhibition of downstream pathways (5), or directly targeting the mutated KRAS protein (6).

There have been detailed studies of downstream signaling patterns of KRAS-mutated (KRAS^MT) NSCLC, pancreatic ductal adenocarcinoma (PDAC), and colorectal cancer to define patterns of signaling, feedback loops, and optimal combination therapy (7–10). In the light of emerging evidence of context specificity of drug response in cancer (3, 4) and the reports that patients with KRAS^MT NSCLC and colorectal cancer respond differentially when treated with combinations of MEK and PI3K pathway inhibitors (11, 12), we aimed to study differences in dynamic signaling patterns in three diseases with frequent KRAS mutations, i.e., NSCLC, PDAC, and colorectal cancer.

We chose to study differences in signaling patterns in 30 KRAS^MT cell lines (10 NSCLC, 10 colorectal cancers, and 10 PDAC) and 10 KRAS^MT cells isolated from patients with malignant effusions. It is known that differences in the type of KRAS mutations between these tumor types do occur, for example, G12C mutations occur more frequently in KRAS^MT NSCLC compared with colorectal cancer (5). The different KRAS mutations in our cell line panel are listed in Supplementary Table S1. An antibody-based platform was used to screen changes in 50 phosphoproteins which are relevant to KRAS signaling and are related to targets of the drugs used as probes (Fig. 1). We exposed the cell lines to clinically relevant concentrations of 7 targeted anticancer drugs: AZD5363 (ref. 13; AKT inhibitor), everolimus (m-TOR inhibitor), gefitinib (EGFR inhibitor), luminespib/NVP-AUY922 (ref. 14; HSP90 inhibitor), pictilisib/GDC-0941 (ref. 15; PI3K inhibitor), trametinib (MEK inhibitor), and vemurafenib (BRAF inhibitor). We chose these drugs as tools as they were known to inhibit signaling nodes related to KRAS signaling and/or where KRAS mutations were known to affect sensitivity to the drug. Further, all these drugs had been used in the clinic, and it was possible to use concentrations of the drugs that were clinically relevant. We used equimolar drug concentrations across cell lines and patient samples rather than individual GI₅₀ concentrations across all samples as we were focused on using clinically relevant concentrations, and it was not possible to determine GI₅₀ of drugs in cells isolated from ascites of patients as we did not establish cell lines from these samples. We planned to validate any interesting findings in more detailed experiments.

All cell lines were obtained from the ATCC (LGC Standards), Public Health England, Sigma-Aldrich, or The Francis Crick Institute's Cell Services. All drugs were sourced from Selleck Chemicals. Details of cell lines, culture media, and concentrations of drugs used are available in the Supplementary Data.

Data related to whole-exome sequencing baseline mRNA for 16,831 genes were extracted from the Cancer Cell Line Encyclopedia (CCLE) database. Data related to mutations and baseline mRNA expression in 26 and 28 of 30 cell lines, respectively (mutational data for COLO678, LIM-2099, LIM-1899 and PANC-1 cell lines, were not present in the CCLE database, and mRNA data were unavailable for LIM-1899 and LIM-2099). The data file “CCLE_Expression_Entrez_2012-09-29.gct” and annotation file “CCLE_Expression.Arrays.sif_2012-10-18.txt” were downloaded from the CCLE website. The data file was filtered in R Studio to only contain the cell lines used in this project, and then the values for each gene were median centered using the base R package. This data file was then loaded into the Broad Institute's GENE-E software (version 3.0.204) which was used to create Fig. 2 using default settings. Pearson's correlation between global mRNA expression is indicated by a blue–white–red color scale, normalized to the minimum correlation between cell lines seen (0.8367) and a hypothetical perfect correlation of 1.

Up to 1,000 mL of ascites or pleural fluid was collected from the patient and immunomagnetically separated using previously published methods (16). Pleural and ascitic fluid were used in the study after the investigators had obtained written, informed consent. The tissue collection protocols were approved by the Institutional Review Board and conducted in accordance with the Declaration of Helsinki.

MILLIPLEX MAP Akt/mTOR phosphoprotein kit, MILLIPLEX MAPK/SAPK signaling kit, and MILLIPLEX MAP RTK phosphoprotein kit (48-611MAG, 48-660MAG, and HPRTKMAG-01K, respectively; MerckMillipore) were combined with the following singleplex magnetic bead sets to produce three multiplex Luminex assays: phospho-NFkB, phospho-SRC, phospho-STAT3, phospho-STAT5 A/B, total HSP27, and GAPDH (46-702MAG, 46-710MAG, 46-623MAG, 46-641MAG, 46-608MAG, and 46-667MAG; MerckMillipore). Bio-Plex Pro phospho-PDGFRa, phospho-PDGFRb, and Akt (Thr308) (171-V50017M, 171-V50018M, and 171-V50002; Bio-Rad) were combined into a triplex assay. Ten microgram of protein was loaded per well, and manufacturers' protocols were followed throughout.

Additionally, an in-house multiplex Luminex assay was created utilizing a range of antibodies from Cell Signaling Technology, targeting proteins of interest. These were conjugated to Luminex MagPlex Microspheres (MC100XX-01, Luminex) via an xMAP Antibody Coupling Kit (40-50016, Luminex). A second set of antibodies, targeting phosphorylated versions of the antibodies from the first set, were biotinylated (Biotin Type A conjugation kit, ab102865; Abcam). This home-grown assay followed the Millipore protocol; however, during the optimization process, it was found that small amounts of unbound biotin in the secondary antibody mix were binding directly to protein, increasing background. To offset this as much as possible, the secondary antibody was added directly after primary incubation, and unbound protein was only washed off after secondary antibody incubation. Phosphoprotein levels were measured on the Luminex 200 system utilizing xPONENT v3.1 software.

We attempted to quality control our Luminex platform. Three test cell lines (A2780, HT29, and NCI-H520) were run in triplicate, and each repetition was run across five separate 96-well plates. All plates were run in a single sitting to avoid possible inter-daily fluctuations. Baseline phospho-protein levels were measured, and the coefficient of variation (CV) per analyte was calculated. There was considerable variation in the assays with the CV being >30% in at least one of the three cell lines in 12 of the 55 analytes tested. The CV of phosphoproteins of interest such as p-MEK and p-AKT was <10% and <20%, respectively, in all three cell lines tested during the validation (Supplementary Table S2).

Each Millipore Mapmate kit provided positive and negative lysate controls for each analyte (as listed in the manufacturer's manual). These were all run once at the start of the project. A protein concentration titration was also carried out to assess the appropriate amount of protein to use in each assay. Positive controls for the home-grown bead set were found via a literature search and were used in the optimization process.

The assay platform continues to be in development and would not meet the standards required for clinical decision-making. Samples of each experiment including untreated controls and samples treated with drugs were analyzed in the same run in order to reduce chances of variability affecting results. We ran the untreated control sample for each cell line in each individual experiment thrice and calculated a mean and SD for each phosphoprotein. Equal amounts of protein were loaded in individual wells. Following GAPDH normalization, only values for each phosphoprotein that were two SDs above or below the corresponding untreated control were considered to be significant. Normalizing the values of each phosphoprotein to its total protein would improve the quality and validity of the results. This was not done as it would require double the quantity of protein needed to analyze patient samples which we did not have. Other factors taken into consideration for not measuring total protein and normalizing all phosphoproteins to individual total proteins are that this would very significantly increase the cost and complexity of the screening assay.

All phosphoproteomic data were normalized to GAPDH. Importantly, for each cell line, three samples of control and one sample for each drug treatment were set up. An SD was calculated for each control, and if the drug-treated sample had a value more than 2 SDs above, it was classified as “increased,” and if 2 SDs below, it was classified as “decreased.” If it was within 2 SDs above or below the control, it was considered unchanged. We chose to use dichotomous, increased, decreased, or no change outputs as we had not validated the linearity of the absolute changes in phosphoprotein in our assays. We ran only controls and did not treat samples in triplicate because of the cost but repeated specific experiments in triplicate to validate any interesting changes seen.

For the purpose of analysis by logistic regression, only individual analytes which were considered to be significantly increased or decreased (2 SDs above or below the mean control) compared with control were used. These phosphorylation changes in cell lines from one tumor type were compared with the other two groups, e.g., NSLC versus colorectal cancer and PDAC, colorectal cancer versus NSCLC and PDAC, PDAC versus NSCLC and colorectal cancer using the generalized linear models function in R to compute logistic regression (RStudio, V1.1.383, RStudio, Inc.). FDR was corrected for via the Benjamini–Hochberg (BH) procedure (base R Function), and only changes that were significant after BH correction are mentioned in Fig. 3.

Cells were lysed, and phospho/Total AKT levels were measured using a phospho (ser473)/Total AKT whole cell lysate kit (Meso Scale Discovery; K15100D-1) following the manufacturer's guidelines.

Growth inhibition was assessed via standard 72-hour sulforhodamine B assays by previously described methods (17).

Data related to mutations and baseline mRNA signatures were available in 26 and 28 of the 30 cell lines, respectively. Unbiased hierarchical clustering did not reveal that KRAS-mutant cell lines of different tumor types of origin significantly clustered together based on either mutation (Fig. 2A) or in mRNA expression (Fig. 2B). The data have also been fitted to each tumor type/tissue type and represented in Supplementary Figs. S1 and S2.

Interestingly, p-MEK levels increased in more than 50% of cell lines in cells exposed to the PI3K inhibitor, pictilisib, AKT inhibitor, AZD5363, EGFR inhibitor, gefitinib, BRAF inhibitor, vemurafenib, and HSP90 inhibitor, luminespib. p-ERK levels were increased in more than 50% of cell lines upon exposure to the BRAF inhibitor, vemurafenib, AKT inhibitor, AZD5363, and EGFR inhibitor, gefitinib (Supplementary Table S3). These results suggest increased activation of MEK is a relatively common rewiring event across multiple different drugs in the cell line panel tested.

There were similarities between changes caused by drugs targeting a defined signaling pathway, for example, more than 50% of all the 30 cell lines showed reduction of p-S6 when exposed to pictilisib, AZD5363, and everolimus all targeting the PI3K pathway. There were differences in phosphorylation of proteins in drugs targeting the PI3K pathway, for example, p-AKT levels were increased in more than 50% of cell lines exposed to AZD5353 but not pictilisib or everolimus. The increase in phosphorylation of AKT despite inhibition of the target is known to be due to change in conformation of the drug target.

We then tested to see if changes in each analyte were different in the three different tumor types when exposed to an individual drug using logistic regression and correcting for multiple testing (Fig. 3). A total of 39 changes in phosphorylation were considered significantly different across different tumor types (Table 1).

Interestingly, a significantly lower number of colorectal cancer cell lines showed a reduction in phosphorylation of AKT^Ser473 when exposed to pictilisib at a clinically relevant, equimolar concentration compared with NSCLC and PDAC cell lines (P = 0.037). Conversely, p- AKT^Ser473 levels were reduced in NSCLC cell lines exposed to pictilisib compared with PDAC and colorectal cancer, although this was not statistically significant. Further, when exposed to pictilisib, significantly fewer NSCLC cell lines showed an increase in phosphorylation of MEK compared with colorectal cancer and PDAC cell lines (P = 0.036). We went on to further validate the findings of differential phosphoprotein changes between NSCLC, colorectal cancer, and PDAC cell lines when exposed to pictilisib.

We analyzed differences between phosphoprotein changes caused by all seven drugs by unbiased clustering, and there were no significant differences between types of KRAS mutation (Supplementary Note).

We chose 9 cell lines (3 each of colorectal cancer, PDAC, and NSCLC) from the original panel to validate our results, exposing them both for 1 hour. We conducted the experiments in triplicate allowing more detailed assessment of fold change compared with threshold-based binary representations used for logistic regression in the initial 1-hour screen. The patterns of nonreduction of p-AKT in colorectal cancer cell lines and no increase in p-MEK levels in the NSCLC cell lines were recapitulated at 1 hour of exposure to pictilisib and technically validated our findings in the larger screen (Fig. 4A). We also used a further assay (ELISA) to quantify p-AKT^Ser473 and total AKT at the 1-hour time-point, and the colorectal cancer cell lines exposed to pictilisib showed a lesser reduction of p-AKT^Ser473 normalized to total AKT compared with NSCLC cell lines (Supplementary Fig. S3).

We also determined the GI₅₀ of pictilisib and another pan-PI3K inhibitor, buparlisib, exposing the 9 cell lines to increasing concentrations of pictilisib and buparlisib for 1 hour to study changes in p-AKT and p-MEK. Changes in phosphorylation caused by pictilisib were concordant with that of buparlisib, suggesting that phosphoprotein changes were not due to the off-target effects of pictilisib (Fig. 4B). It was, however, noted that p-MEK levels increased in the NSCLC cell line H1944 in Fig. 4B while this was not the case in Fig. 3 and Fig. 4A. The concentration of pictilisib used in experiments in Figs. 3 and 4A was 96.3 nmol/L, whereas concentrations of pictilisib used in Fig. 4B were GI₅₀ and 5X GI₅₀, which was considerably higher (Supplementary Table S4), and this may account for this difference.

We studied p-MEK and p-AKT levels in cancer cells isolated from serous effusions (pleural effusions and ascites) in 10 patients with KRAS-mutant cancers (3 NSCLC and 7 colorectal cancers) exposed to pictilisib for 1 hour as we had done in the 30 cell line screen. We found the increase in the p-MEK levels was significantly raised in 3 of 7 samples of cells isolated from patients with colorectal cancer as opposed to 0 of 3 samples of cells isolated form NSCLC. No significant reduction of p-AKT was seen in either colorectal cancer or NSCLC samples (Fig. 4C). The number of samples was not sufficient for statistical analysis; however, the experiment provided proof of concept that it is possible to conduct ex vivo assays on cancer cells freshly isolated from cancer patients.

To our knowledge, this is the first report to study the differences in dynamic signaling patterns in KRAS^MT cell lines in an attempt to study context specificity of tumors of origin.

Antibody-based phosphoprotein platforms have been used to study signaling output in cancer cells exposed to drugs, with reverse-phase protein arrays being one of the more established platforms (18). We used the antibody-bead–based Luminex platform (19). The drugs used as probes were anticancer drugs that were either licensed (everolimus, gefitinib, trametinib, vemurafenib) or in phase II studies (AZD5363, luminespib, and pictilisib). We chose to use clinically relevant concentrations of these drugs at human Cmax and exposed the cell lines and patient samples to these concentrations for 1 hour. Most phosphorylation changes occur early and so we chose a 1-hour time-point to reflect this, as well as the fact that concentrations surrounding Cmax could be maintained for 1 hour. Our aim was to study differences in signal transduction at early time-points. We acknowledge that rewiring of signal transduction may change at later time-points which we did not study. Further, we attempted to standardize the treatment conditions by exposing cells to drugs at 80% confluence. The cells were exposed to drugs for 1 hour so the drugs are unlikely to cause major differences in cell density due to cell death or growth inhibition; however, it was not possible to rule out the effects of contact inhibition on signaling in individual cell lines. We did not establish cell lines from each patient sample, and cells derived from ascites or pleural effusions were EpCAM separated and exposed to drugs for 1 hour in suspension. It was not possible to ascertain the contribution of differences in tissue culture conditions, i.e., adherent versus suspension in the results while comparing cell lines and patient samples. The importance of PI3K signaling in KRAS^MT colorectal cancer, NSCLC, and PDAC has been described previously (20). Coexistence of activating PIK3CA mutations with KRAS mutations in the general population of NSCLC, colorectal cancer, and PDAC is rare (21, 22). However, the coexistence of PIK3CA mutations can significantly affect signaling output drug resistivity to signal transduction inhibitors, and we were careful while choosing our panel of cell lines (23) to exclude cell lines with known activating mutations in PIK3CA.

KRAS mutations are found in approximately 90% of PDAC, 40% of colorectal cancer, and 30% of adenocarcinoma subtype of NSCLC (5). Within NSCLC, a glycine to cysteine substitution is more common (5). Interestingly, within NSCLC there are differences in outcomes for patients depending on the type of KRAS mutation. While evaluating differences in KRAS mutations in NSCLC, the investigators found cell lines with KRAS^Gly12Cys and KRAS^Gly12Val had decreased growth factor–dependent AKT activation suggesting that there are differences in signaling patterns within different KRAS^MT cell lines (24). Further, analysis of known mutations (not exclusively KRAS) of the cell lines and baseline mRNA expression available in public databases did not reveal that the cell lines clustered significantly to tissue of origin, highlighting the importance of studying posttranscriptional differences such as rewiring of signal transduction.

Importantly, we found that in KRAS^MT colorectal cancer cell lines, exposure to the pan-PI3K inhibitor, pictilisib, had a compensatory increase in p-MEK levels as well as differentiating themselves by not having a reduction in p-AKT levels. In contrast, NSCLC cell lines did not have a compensatory increase in p-MEK levels. This differential phosphorylation of MEK by PI3K inhibitors in KRAS^MT cell NSCLC and colorectal cancer cell lines has not been described previously.

We went on to validate specific changes in cells seen in response to pictilisib by repeat experiments in triplicate. We also confirmed that these changes in p-MEK and p-AKT occurred when a second pan-PI3K inhibitor was used, which suggested that the phosphoprotein changes are not due to the off-target effects of pictilisib. Interestingly, in cancer cells isolated from patients with KRAS^MT, NSCLC did not show activation in p-MEK, whereas some of the samples isolated from patients with KRAS^MT colorectal cancer did. The number of samples studied did not allow robust statistical analysis, however shows proof of concept that such tissue can be used to study changes in phosphoproteins.

The context specificity of oncogenes to induce cancer has been described (25). In the setting of KRAS^MT cancers, there is increasing evidence that tissue of origin also dictates metabolism of branched chain amino acids (26). The signaling patterns within cancers cells are influenced by surrounding stromal cells as shown in KRAS-mutant cancer cells such PDAC (27). It is possible that stromal differences in diverse organs lead to context specificity of KRAS^MT cancers. The findings in the current study of signaling in cancer cells did not take stromal influences into consideration and could be considered as a weakness in the experimental design. However, our findings are of particular significance as they suggest inherent differences in signaling output in KRAS-mutant cancer cells sans stroma and may in part be responsible for context specificity of response to signaling inhibitors seen in clinical trials. Our study was confined to cell lines that had known KRAS mutations. However, it was not possible to answer the interesting question relating to context specificity of signaling extending to non–KRAS-mutant cell lines.

Context specificity is a critical facet that influences the precision medicine paradigm. There is already evidence that this is clinically important to drug response, as has been exemplified by BRAF inhibitors (3, 4). Our findings suggest there are inherent differences in dynamic signaling output within KRAS^MT cancer cells, which could influence drug response and should be taken into consideration while designing clinical trials. Finding new treatment paradigms for KRAS^MT cancers remains an urgent and unmet need.

A.R. Minchom is a consultant/advisory board member for Faron pharmaceuticals, Janssen pharmaceuticals, and Imugene, and has an expert testimony in LOXO. D. Cunningham reports receiving other commercial research support from Amgen REAL 3 Trial for the Royal Marsden NHS Foundation Trust, Sanofi Trial for the Royal Marsden NHS Foundation Trust, Janssen IMYC Trial for the Royal Marsden NHS Foundation Trust, Merck ICONIC/POLEM Trial for the Royal Marsden NHS Foundation Trust, Merrimack. PLATFORM Trial for the Royal Marsden NHS Foundation Trust, AstraZeneca FRGR Trial for the Royal Marsden NHS Foundation Trust, Celgene PROSPECT R Trial for the Royal Marsden NHS Foundation Trust, MedImmune PLATFORM Trial for the Royal Marsden NHS Foundation Trust, Bayer PROSPECT R Trial for the Royal Marsden NHS Foundation Trust, 4SC EMERGE Trial for the Royal Marsden NHS Foundation Trust, Clovis PLATFORM Trial for the Royal Marsden NHS Foundation Trust, and Eli Lilly PLATFORM Trial for the Royal Marsden NHS Foundation Trust. M.E.R. O'Brien reports receiving honoraria from the speakers' bureau of MSD, BMS Roche, Pierre Fabre, and Abbvie, and is consultant/advisory board member for MSD. J.S. de Bono reports receiving commercial research grant from AstraZeneca, GSK, Pfizer Oncology, Genentech/Roche, Astellas, MSD, Daiichi, and Merck Serono; reports receiving honoraria from the speakers' bureau of AZ, GSK, Genentech/Roche, Pfizer Oncology, Astellas, MSD, Daiichi, and Merck Serono; and is a consultant/advisory board member for AstraZeneca, Genentech, Pfizer Oncology, Astellas, MSD, Daiichi, GSK, and Merck Serono. B. Al-Lazikani reports receiving honoraria from the speakers' bureau of Astellas and is a consultant/advisory board member for GSK, Astex Pharmaceuticals, and ICR. U. Banerji is an employee of The Institute of Cancer Research, an academic institution involved in the development of PI3K, HSP90, HDAC, AKT, ROCK, RAF, and CHK1 inhibitors; reports receiving other commercial research support from AstraZeneca (for investigator-initiated phase I clinical trial), Chugai (for investigator-initiated phase I clinical trial), Onyx Pharmaceuticals (for investigator-initiated phase I clinical trial), BTG International (to complete investigator-initiated phase I clinical trial of ONX-0801), and Verastem (for investigator-initiated phase I clinical trial); and is a consultant/advisory board member for Lilly, Phoenix Solutions, Novartis, Astellas, and Astex Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E.A. Coker, J.S. de Bono, U. Banerji

Development of methodology: A. Stewart, E.A. Coker, J.S. de Bono, B. Al-Lazikani, U. Banerji

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Stewart, A. Georgiou, A.R. Minchom, S. Carreira, D. Cunningham, M.E.R. O'Brien, J.S. de Bono, U. Banerji

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Stewart, E.A. Coker, S. Pölsterl, A. Georgiou, A.R. Minchom, S. Carreira, D. Cunningham, F.I. Raynaud, J.S. de Bono, B. Al-Lazikani, U. Banerji

Writing, review, and/or revision of the manuscript: A. Stewart, E.A. Coker, A. Georgiou, M.E.R. O'Brien, F.I. Raynaud, J.S. de Bono, B. Al-Lazikani, U. Banerji

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): U. Banerji

Study supervision: J.S. de Bono, U. Banerji

The authors acknowledge institutional funding from Cancer Research UK as part of the Experimental Cancer Medicine Centre initiative (Ref: C12540/A25128); a Cancer Therapeutics Unit award (Ref: C2739/A22897), and a Cancer Therapeutics Centre award (Ref: C309/A25144). The authors also acknowledge the National Institute for Health Research (NIHR) Biomedical Centre initiative awarded to The Institute of Cancer Research and The Royal Marsden Hospital NHS Foundation Trust (Ref: IS-BRC-1215-20021). U. Banerji is a recipient of an NIHR Research Professorship award (Ref: RP-2016-07-028).

This work has been presented, in part, as an oral presentation at the TAT Congress, Paris, March 2018 [Abstract #430] and as an oral presentation at the AACR Annual Meeting, Washington DC, April 2017 [Abstract #996].

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.