Dissecting mechanisms of resistance to targeted drug combination therapy in human colorectal cancer

Genomic alterations in cancer cells result in vulnerabilities that clinicians can exploit using molecularly targeted drugs, guided by knowledge of the tumour genotype. However, the selective activity of these drugs exerts an evolutionary pressure on cancers that can result in the outgrowth of resistant clones. Use of rational drug combinations can overcome resistance to targeted drugs, but resistance may eventually develop to combinatorial therapies. We selected MAPK- and PI3K-pathway inhibition in colorectal cancer as a model system to dissect out mechanisms of resistance. We focused on these signalling pathways because they are frequently activated in colorectal tumours, have well-characterised mutations and are clinically relevant. By treating a panel of 47 human colorectal cancer cell lines with a combination of MEK- and PI3K-inhibitors, we observe a synergistic inhibition of growth in almost all cell lines. Cells with KRAS mutations are less sensitive to PI3K inhibition, but are particularly sensitive to the combined treatment. Colorectal cancer cell lines with inherent or acquired resistance to monotherapy do not show a synergistic response to the combination treatment. Cells that acquire resistance to an MEK–PI3K inhibitor combination treatment still respond to an ERK–PI3K inhibitor regimen, but subsequently also acquire resistance to this combination treatment. Importantly, the mechanisms of resistance to MEK and PI3K inhibitors observed, MEK1/2 mutation or loss of PTEN, are similar to those detected in the clinic. ERK inhibitors may have clinical utility in overcoming resistance to MEK inhibitor regimes; however, we find a recurrent active site mutation of ERK2 that drives resistance to ERK inhibitors in mono- or combined regimens, suggesting that resistance will remain a hurdle. Importantly, we find that the addition of low concentrations of the BCL2-family inhibitor navitoclax to the MEK–PI3K inhibitor regimen improves the synergistic interaction and blocks the acquisition of resistance.


Introduction
Over recent years an improved understanding of the molecular basis of cancer has led to the concept of precision medicine, where treatment with targeted drugs is guided by knowledge of the patient's tumour genotype [1]. Although this approach can be successful, the selective pressure that targeted agents exert can result in the outgrowth of resistant clones [2][3][4][5]. The molecular responses of tumours to therapeutics targeting the MAPK pathway are paradigms for the development of resistance [3,4,6]. Early studies of BRAFand MEK-inhibitors targeting this pathway established that they were particularly active in BRAF mutant melanoma [6][7][8]. However, even in this patient cohort, the response is generally temporary with most patients relapsing within a year [6]. To overcome resistance, an RAF-MEK inhibitor combination has been approved for the treatment of mutant BRAF melanoma and, more recently, has been shown to be effective in BRAF mutant colorectal cancer [9][10][11][12]. However, despite improved responses to these combination therapies tumours still recur [12,13].
Colorectal cancers frequently have genetic aberrations in the MAPK or phosphatidylinositol 3-kinase (PI3K) pathways [14] and targeting these pathways can inhibit tumour growth [7,[15][16][17]. However, a significant proportion of colorectal tumours have genetic abnormalities that activate both pathways [14], resulting in a reduced response to monotherapy. The interconnectedness of these two pathways suggests that combination of MEK-and PI3Kinhibitors is a rational therapeutic approach. Several studies have explored combinations of MEK-and PI3K-inhibitors in vivo and in vitro and have reported both positive and negative effects of the combination treatment versus monotherapy [16,[18][19][20][21].
It is important to understand the mechanisms responsible for resistance to targeted cancer drugs to inform their future clinical use. In this study we focus on colorectal cancer and find that a combination of MEK-and PI3K-inhibitors is synergistic for growth inhibition across a panel of human colorectal cell lines, particularly in KRAS mutant cells with a reduced sensitivity to PI3K inhibition. We also find that prolonged exposure to a PI3K inhibitor plus an MEK-or ERK-inhibitor leads to the emergence of resistance; importantly, however, this resistance can be overcome by cotreatment with a BCL2-family inhibitor.

MEK-and PI3K-inhibitors act synergistically in human colorectal cancer cells
Initially, we determined the potency of PI3K pathway inhibitors in a panel of 29 human colorectal cancer cell lines, in order to aid selection of exemplars for further analysis (Supplementary Table 1-2; Supplementary Fig. 1).
To determine the relationship between inhibitor selectivity profile and its ability to block PI3K signalling, we also measured AKT phosphorylation in HCT116 human colorectal cancer cells (Fig. 1a), including additional PI3K Fig. 1 Synergistic inhibition of HCT116 human colorectal cancer cell growth by MEK and PI3K inhibitors. a Comparison of the EC 50 for inhibition of AKT SER473 and AKT THR308 phosphorylation with the GI 50 of PI3K inhibitors (EC 50 values, n = 2) using an electrochemiluminescent ELISA. b, c Effect on HCT116 cell growth of 96 h exposure to pictilisib (pan-PI3K inhibitor) and cobimetinib (MEK inhibitor). d Effect of CCT245731 (an inactive analogue of pictilisib) and cobimetinib (Supplementary Tables S1-2) or single/repeated treatment with cobimetinib. e Combination indices for combination treatments featuring PI3K pathway inhibitors with different selectivity profiles. Column colours indicate PI3K pathway inhibitor selectivity: orange = pan-class I, cyan = class I isoform-selective, red = dual mTOR-pan-class I, blue = mTOR. The pink region indicates CI values associated with an additive effect, determined by combining two doses of cobimetinib (from (d)). Mean values (s.e.m.), n ≥ 3 pathway inhibitors to widen the analysis. For the majority of the inhibitors the EC 50 values for inhibition of AKT SER473 and AKT THR308 phosphorylation are similar and correlate with GI 50 values for cell growth inhibition (Fig. 1a). The mTOR-selective inhibitors and one of the dual mTOR-pan class I PI3K inhibitors, dactolisib, are an exception, with disparate EC 50 values for inhibition of AKT SER473 and AKT THR308 phosphorylation (Fig. 1a).
We next used HCT116 and SW480 colorectal cancer cell lines, initially with the cobimetinib (MEK inhibitor; Supplementary Table 3) and pictilisib (class I PI3K inhibitor) to establish the combination methodology. We observe a synergistic interaction with CI values of 0.515 ± 0.030 and 0.305 ± 0.026 respectively at the combination EC 50 (Fig. 1b,  c; Supplementary Fig. 2a). As controls, substituting pictilisib with CCT245731 (an inactive analogue [15,22]; Supplementary Table 1) eliminates the synergistic response ( Fig. 1d), while a combination of cobimetinib with a second treatment of cobimetinib, predicted to give an additive value of 1, results in a measured CI of 1.241 ± 0.146 (Fig. 1d).
Having established that we could detect a synergistic interaction, we tested a selection of MEK inhibitors with different PI3K-pathway inhibitors in HCT116 and SW480 cells (Supplementary Tables 1 and 3). Combinations of all three MEK inhibitors, PD0325901, cobimetinib or selumetanib with pictilisib gave synergistic inhibition of cell growth (Fig. 1e). Similarly, all three pan-class I PI3K inhibitors, pictilisib, ZSTK474 and taselisib are equally synergistic with the MEK inhibitor cobimetinib. Alpelisib, which is highly selective for the alpha-isoform of PI3K, is the most effective of all of the isoform-specific PI3K inhibitors tested in combination with cobimetinib, producing effects equivalent to the pan-class I PI3K inhibitors. The dual mTOR-pan-class I PI3K inhibitors apitolisib, PI-103 and dactolisib are less effective than their pan-class I PI3Kselective counterparts when used in combination with cobimetinib. As previously described [21], we found that combining any of the mTOR-specific inhibitors, AZD8055, torkinib or WYE125132, with cobimetinib results in an additive, rather than synergistic, response in both lines ( Fig.  1e; Supplementary Fig. 2b).
Based on their cell panel GI 50 potency, activity against AKT phosphorylation and combination efficacy ( Fig. 1; Supplementary Fig. 1-2), we selected the pan-class I PI3K inhibitor pictilisib and the MEK inhibitor cobimetinib for use in subsequent experiments to explore the basis of resistance to combined MEK−PI3K inhibitor treatment.

Combined MEK-and PI3K-inhibition induces apoptosis
Monotherapy with either cobimetinib or pictilisib does not induce apoptosis biomarkers, whereas monotherapy with the pictilisib alone increases the level of LC3B, potentially indicative of autophagy (Fig. 2a, b). In contrast, the combination treatment induces apoptosis-associated caspase activation in both HCT116 and SW480 cells with no detectable autophagy-associated LC3B signal (Fig. 2a, b; Supplementary Fig. 2c).
We used an siRNA minipanel to identify regulators of the apoptotic response induced by the combination regimen. At least 2/3 of the siRNAs targeting the expression of antiapoptotic BCLXL and MCL1 induced apoptotic caspase activity. In contrast, siRNA silencing of proapoptotic genes BIM, BBC3, BID and BAX significantly reduces combination-induced apoptosis (Fig. 2c). We confirmed that BAX and BBC3 are required for the induction of apoptosis by the combination treatment using isogenic knockout HCT116 BAX−/− and HCT116 BBC3−/− cell lines (Fig. 2d) [23,24]. Interestingly, the cobimetinib-pictilisib combination treatment retains its synergistic activity in both knockout lines (HCT116 BAX−/− CI = 0.518 ± 0.035, HCT116 BBC3−/− CI = 0.604 ± 0.035), indicating that apoptosis defined by caspase activation is not exclusively required for these inhibitors to act in synergy.
Combined MEK-and PI3K-inhibition acts synergistically on RPS6 phosphorylation and Forkhead-regulated gene expression Next, we determined the molecular response to combined MEK-and PI3K-inhibition and observe a decrease in the levels of phosphorylated ERK and AKT ( Fig. 3a; Supplementary Fig. S3a). However, the combination treatment is no more effective at inhibiting ERK or AKT phosphorylation than the single treatments. Similarly, the combination treatment is no more effective at inhibiting 4E-BP1 phosphorylation ( Fig. 3a; Supplementary Fig. 3a) or global capdependent protein synthesis than monotherapy ( Fig. 3b; Supplementary Fig. 3b). In contrast, we observe a significant synergistic reduction in the level of phosphorylated RPS6 following combination treatment, compared to levels after monotherapy (HCT116 CI = 0.268 ± 0.023; SW480 CI = 0.395 ± 0.066; P < 0.05; Fig. 3a; Supplementary  Fig. 3a).

MEK-and PI3K-inhibitor combination treatment has a synergistic effect in multiple colorectal cell lines
To determine CI values in our panel of 47 human colorectal cancer cell lines we first established GI 50 values for cobimetinib, pictilisib or the dual mTOR-pan-class I PI3K inhibitor apitolisib ( Supplementary Fig. 5). We observe that KRAS-mutant PIK3CA-wild-type cells are significantly less sensitive to the PI3K inhibitor pictilisib than the wild type (P = 0.019, q = 0.0498) or KRAS-wild-type PIK3CAmutant cells (P = 0.0076, q = 0.0384). Cells with KRAS codon 12 mutations (KRAS CDN12 /PIK3CA WT ) are the least sensitive to PI3K inhibition overall (P = 0.0051, q = 0.0108; Supplementary Fig. 5). Most BRAF mutant cell lines are very sensitive to the MEK inhibitor cobimetinib compared to wild-type cells (BRAF MUT P = 0.0487, q = 0.0804; BRAF MUT /PIK3CA MUT P = 0.0085, q = 0.0281).
We find that the interaction between cobimetinib and pictilisib is synergistic in all but one colorectal cell line, with CI values ranging from 0.296 to 0.884 ( Fig. 4a; Supplementary Table 4). There is a strong synergistic response (CI < 0.5) to the combination treatment in many KRAS mutant lines, but this is not statistically significant across the whole panel ( Supplementary Fig. 6a). As observed with the HCT116 and SW480 cell lines, the synergistic effect of the combination treatment is reduced when the dual mTOR-PI3K inhibitor apitolisib is used in place of pictilisib ( Supplementary Fig. 6b).
The growth of COLO320 cells is insensitive to MEK inhibition (cobimetinib GI 50 = 1455 nM; Supplementary inhibition. Levels of a LC3 staining, indicative of autophagy, and b caspase activity, indicative of apoptosis, following 24 h treatment with cobimetinib and/or pictilisib. c Caspase 3/7 activity following transfection of cells with siRNA against proapoptotic genes for 48 h and subsequent treatment with 5× GI 50 cobimetinib and pictilisib, or the vehicle control, for 24 h. Caspase activity is expressed relative to that in cells treated with Allstars negative control siRNA (*P ≤ 0.001). d Caspase activity in the HCT116 parental cells and HCT116 BBC3−/− or HCT116 BAX−/− cell lines after a combined treatment with cobimetinib and pictilisib. All plots show mean values (s.e.m.), n ≥ 3 Table 4) and shows no evidence of synergy with the combination. We also find that SW620 cells (lymph node metastasis)-which are derived from the same patient as SW480 cells (primary tumour) (Fig. 4b)-are less sensitive to pictilisib ( Supplementary Fig. 6c) and also to the cobimetinib-pictilisib combination (SW480 CI = 0.310 ± 0.018, SW620 CI = 0.596 ± 0.017, P < 0.001). Consistent with this, the combined MEK-PI3K inhibitor treatment of SW620 cells does not induce apoptosis and has no additional effect on RPS6 phosphorylation or Forkheadregulated gene expression (Fig. 4c,  Overall, across the large panel of colorectal cell lines with a range of genetic backgrounds, combined MEK-and PI3K-inhibition is highly synergistic, particularly in KRAS mutant cancers. In addition, we find that inherent resistance or reduced sensitivity to the MEK−PI3K inhibitor combination is rare, but can result from reduced or loss of sensitivity to either arm of the combination therapy.
Cells with acquired insensitivity to MEK-or PI3Kinhibitor monotherapy are also resistant to combination therapy We found that some cells that survived 7 days exposure to the combination treatment were able to proliferate when fresh drug-free medium was added (Fig. 5a). To dissect out potential resistance mechanisms, we first examined cells Fig. 3 Combined MEK-PI3K inhibition has a synergistic effect on RPS6 phosphorylation and FOXO-regulated transcription in HCT116 human colorectal cancer cells. a Expression of phospho-or total proteins following 6 h exposure to cobimetinib and/or pictilisib. b Level of cap-dependent protein synthesis initiation (firefly luciferase) after 24 h exposure to cobimetinib and/or pictilisib. Data normalised to cap-independent protein synthesis initiation (EMCV IRES-driven renilla luciferase expressed from the same bicistonic mRNA) and expressed relative to vehicle control. c Transcript abundance determined by RT-PCR after 14 h exposure to cobimetinib and/or pictilisib (*P ≤ 0.001). All data are mean values (s.e.m) shown relative to that for cells treated with the vehicle control (n ≥ 3) with acquired resistance to the individual elements of the combined regimen. Exposure of HCT116 cells to increasing concentrations of MEK or PI3K inhibitors over a period of 8-10 weeks generated an MEK inhibitor resistant line (HCT116 MEKRES ) with a GI 50 for cobimetinib (401 ± 55 nM) 13-fold greater than that of the parental line and also a PI3K inhibitor resistant line (HCT116 PI3KRES ) with a GI 50 for pictilisib (3119 ± 266 nM; Fig. 5b) 5-fold greater than that of the parent line. HCT116 MEKRES cells fail to show a synergistic response to the combination treatment, even when the concentration of cobimetinib used is increased (Fig. 5b). Treating HCT116 PI3KRES cells with a concentration of cobimetinib and pictilisib that is effective on the parental cells does not result in a synergistic inhibition of cell growth but, in contrast to the HCT116 MEKRES cells, there is a synergistic response to the combination treatment when a fivefold higher concentration of pictilisib is used (CI = 0.589 ± 0.026; Fig. 5b).
Whole-exome sequencing showed that >92% of the single nucleotide variants (SNV) and insertions or deletions (INDELS) were conserved between the parent and resistant cells, confirming that the two resistant lines were derived from parent HCT116 cells (Fig. 5c). Among the variants unique to the HCT116 PI3KRES cells are two frameshift mutations in PTEN (PTEN K14fs* and PTEN N323fs* ). Sequencing of PTEN cDNA clones isolated from HCT116 PI3KRES cells detected only mutant cDNAs (Supplementary Fig. 7), consistent with the loss of PTEN protein and elevated phosphorylation of AKT SER473 (Fig. 5d) that still remains sensitive to pictilisib treatment at high concentrations ( Fig. 5f), explaining the synergistic response to the combination treatment at high pictilisib concentrations.
Exome sequencing of HCT116 MEKRES identified a MEK1 L215P mutation previously detected in other MEK inhibitor resistant cell lines or patients that prevents MEK inhibitor binding [26]. These HCT116 MEKRES cells are cross-resistant to the MEK inhibitors selumetanib (GI 50 > 2000 nM) and trametinib (Supplementary Fig. 8a and Supplementary Table 3) but remain sensitive to the ERK inhibitor VTX11e [27] and to VTX11e-pictilisib combination therapy (CI = 0.551 ± 0.028), indicating that the MEK-ERK pathway is still required by HCT116 MEKRES cells (Fig. 5e). ERK1/2 phosphorylation is less responsive to cobimetinib treatment in the HCT116 MEKRES cell line, consistent with the inability of cobimetinib to bind mutated MEK (Fig. 5f).

Sustained combination treatment results in the acquisition of resistance
We next cultured HCT116 colorectal cancer cells in increasing concentrations of cobimetinib and pictilisib and find they acquired resistance to the combination treatment over a period of 8-10 weeks. This resistance is due to reduced sensitivity to MEK inhibition (Fig. 6a, b). As observed for HCT116 MEKRES , we find that the combination- are predicted to alter protein sequence. c Levels of caspase 3/7 activity, indicative of apoptosis, in SW620 cells exposed to pictilisib or cobimetinib for 24 h. d Transcript abundance in SW620 cells exposed to pictilisib and/or cobimetinib for 6 h. Mean levels (s.e.m.) are shown relative to those in cells treated with a vehicle control (n ≥ 3) resistant HCT116 DUALRES cells are cross-resistant to selumetanib (GI 50 > 2000 nM) and trametinib ( Supplementary  Fig. 8a). Replacing the MEK inhibitor cobimetinib arm of the combination treatment with the ERK inhibitor VTX11e (Fig. 6c) restores synergy (CI = 0.693 ± 0.034 nM), indicating that a dependence on signalling downstream of MEK still remains. Importantly, exposure of the HCT116 DUALRES line to increasing concentrations of VTX11e and pictilisib subsequently resulted in the acquisition of resistance to this combination treatment as a result of loss of sensitivity to VTX11e at the level of ERK1/2 (Fig. 6c, d). This cell line (HCT116 DUALERKRES ) retains resistance to the original cobimetinib-pictilisib combination (Supplementary Fig. 8b). Subsequently, we generated HCT116 cells with acquired resistance to ERK inhibitor monotherapy (HCT116 ERKRES ;   Fig. 6f), which unlike the HCT116 DUALERKRES line, is sensitive to MEK inhibition-consistent with derivation from the parent HCT116 cells (Supplementary Fig. 8c).
Exome sequencing confirmed that the combination-and VTX11e-resistant lines (HCT116 ERKRES , HCT116 DUALRES and HCT116 DUALERKRES ) are also derived from parental HCT116 cells (Fig. 6f). HCT116 DUALRES had acquired an MEK2 V215E mutation that prevents inhibitor binding, analogous to that detected in HCT116 MEKRES cells. We found that HCT116 DUALERKRES also shared 644 variants with HCT116 DUALRES cells, including the MEK2 V215E mutation, consistent with its derivation from HCT116 DUALRES . In addition, HCT116 DUALERKRES cells acquired 114 unique variants including an ERK2 Y36H mutation. The ERK2 Y36H mutation is also detected in the HCT116 ERKRES cells.  Consistent with their derivation from the parent cells, the HCT116 ERKRES cells lack the MEK2 V215E mutation and the additional sequence variants found in HCT116 DUALRES and HCT116 DUALERKRES cells.

The Y36H mutant found in ERK inhibitor resistant cells is insensitive to ERK inhibitors
Next, we determined the functional impact of the ERK2 Y36H mutation. The values of K m and V max found for ATP corresponding to ERK2 WT and ERK2 Y36H enzymes are similar ( Supplementary Fig. 9a). ERK2 Y36H exhibited 19-fold resistance to VTX11e, 5-fold resistance to the ERK inhibitor ravoxertinib [28] and 2.7-fold resistance to SCH772984, a chemically distinct ERK1/2 inhibitor (Fig.  7a, Supplementary Fig. 9a-b; Supplementary Table 5) [29]. The differential resistance of the ERK2 Y36H protein to these inhibitors reflects the activity of the inhibitors in HCT116 DUALERKRES and HCT116 ERKRES cells, with a hierarchy of fold-resistance in the order of VTX11e > ravoxertinib > SCH772984 (Supplementary Fig. 10). HCT116 DUALERKRES and HCT116 ERKRES cells are also resistant to ulixertinib, a VXT11e derivative undergoing clinical studies, but not to the mechanistically distinct DEL-22379, which targets ERK dimerisation ( Supplementary  Fig. 10) [30][31][32].
Modelling of the ERK Y36H mutation using published Xray structures suggests that repulsion between the electrondense imidazole ring of H36 and the partial negative charge of the aryl chlorine in VTX11e, also found in ulixertinib and ravoxertinib, predicted to make binding to ERK2 Y36H unfavourable ( Supplementary Fig. 11a-c) [27,28,33]. The aryl chlorine is essential for the selectivity of VTX11e for ERK2 over other kinases (Supplementary Table 2). In contrast, this mutation is predicted to have little effect on the binding of SCH772984 to ERK2, as both Y36 or H36 are predicted to fold under the active site P-loop Fig. 6 Characterisation of HCT116 human colorectal cancer cells with acquired resistance to MEK or ERK inhibitor-based combination therapy. a Surviving fraction of HCT116 DUALRES cells treated with pictilisib and/or cobimetinib for 96 h. b Level of pERK1 THR202/TYR204 / pERK2 THR185/TYR187 in HCT116 DUALRES cells exposed to pictilisib or cobimetinib for 6 h. Data are shown relative to levels in cells treated with a vehicle control. c Surviving fraction of HCT116 DUALRES  ( Supplementary Fig. 11d-e), explaining why ERK2 Y36H cells are only slightly less sensitive to SCH772984 than wild type.

BCL2-family inhibitor blocks the acquisition of resistance to MEK-PI3K combination treatment
BCL2-family inhibitors can act synergistically with MEKor PI3K-inhibitors [35][36][37], and also block the acquisition of resistance to MEK inhibition [38]. We find that BCLXL siRNA induced apoptosis when combined with pictilisib and cobimetinib (Fig. 2c). These observations prompted us to expose HCT116 cells to a cobimetinib−pictilisib combination treatment plus the BCL2-family inhibitor navitoclax. Cotreatment with navitoclax (GI 50 = 7362 ± 247 nM) improves the CI for pictilisib plus cobimetinib in HCT116 cells (CI = 0.406 ± 0.037), potentially by acting in synergy both with pictilisib (CI = 0.552 ± 0.080) and cobimetinib (CI = 0.636 ± 0.042; Fig. 8a). This effect was less pronounced when pictilisib was replaced with the dual mTOR/ pan-class I PI3K inhibitor apitolisib (Supplementary Fig.  12a). We also found that addition of a fixed concentration of navitoclax increases sensitivity to the cobimetinib-pictilisib combination in HCT116 cells (2500 nM navitolcax CI = 0.290 ± 0.021; 750 nM navitoclax CI = 0.387 ± 0.043; Fig.   Fig. 7 The ERK2 Y36H mutation does not affect MAPK-pathway signalling but induces resistance to small molecule ERK inhibitors. a Effect of VTX11e on ERK2 kinase activity. b Abundance of ERK2, pERK1/2 and total ERK1/2 in the parental HCT116 line, drug combination-resistant HCT116 DUALERKRES line and recombinant lines overexpressing ERK2 WT or ERK2 Y36H . c Fraction of HCT116 and ERK2-overexpressing cells surviving 96 h treatment with cobimetinib or VTX11e. d Level of ERK2 and ERK1/2 in cells transfected with siRNA targeting ERK1/2 expression (C = mock control, N = Allstars negative control nontargeting siRNA; ERK1 siRNA nos. 6,7,8,11 and ERK2 siRNA nos. 10, 11, 12, 13; Qiagen, Germany). e Fraction of HCT116 or HCT116 DUALERKRES cells surviving 96 h exposure to pictilisib after siRNA knockdown of ERK1/2 expression for 48 h prior to drug treatment. For all data n = 3 and, where applicable, error bars indicate s.e.m. GADPH was used as a loading control in immunoblots 8b). HCT116 DUALRES cells with previously acquired resistance to the combined MEK−PI3K inhibitor regimen through loss of sensitivity to MEK inhibition retained the synergistic interaction between pictilisib and navitoclax, but that was not further improved by the addition of cobimetinib ( Supplementary Fig. 12b). In the HCT116 DUALRES cells, replacement of cobimetinib with the ERK inhibitor VTX11e rescued the potent synergistic activity of the triple combination, but again synergy was lost in HCT116 DUALERKRES cells with previously acquired resistance to the ERK inhibitor combination regimen ( Supplementary Fig. 12c-e). Importantly, in longer-term clonogenic assays, we find that cotreatment with 750 nM navitoclax reduced the number of HCT116 and SW480 cells surviving 7 days exposure to the cobimetinib-pictilisib combination, such that we were unable to generate stable clones with acquired resistance in longer-term cultures (Fig. 8c, d).

Discussion
Combination treatment regimens have several advantages over monotherapy: they may increase antitumour effects within acceptable toxicity limits, are more likely to be effective against a heterogeneous tumour population and may delay or block the development of drug resistance [39]. Here, we used a panel of human colorectal cancer cell lines to explore response to mono-and combination-therapies targeting MEK and PI3K. Initial sensitivity profiling work was followed up with a focused strategy using a single colorectal cancer cell line model to identify mechanisms of resistance. While the latter approach may not define the full spectrum of resistance in colorectal cancer, we did find resistance mechanisms, namely loss of PTEN and mutation of MEK1/2, that were previously reported in clinical resistance to MEK and PI3K inhibitors [26,40], thus validating the approach.
Our results indicate that MEK-PI3K inhibitor combination treatment affects multiple nodes at the intersection of the MAPK and PI3K signalling pathways; these include reduced RPS6 phosphorylation and induced Forkheadregulated gene expression [25]. RPS6 is phosphorylated by p70 and p90 S6 kinases, which are themselves regulated by MAPK and PI3K signalling, while the localisation and stability of the Forkhead transcription factors is regulated by AKT and ERK-mediated phosphorylation respectively Fig. 8 Navitoclax administration prevents cells acquiring resistance to a pictilisb-cobimetinib combination treatment. a Three-way combination treatment of HCT116 cells with cobimetinib, pictilisib and navitoclax at 1:1:1 ratios of their respective GI 50 s. b Combination treatment of HCT116 and SW480 cells with cobimetinib and pictilisib at a 1:1 ratio of their respective GI 50 s ± a low fixed concentration of navitoclax. For all plots (a, b) data are mean (s.e.m.) values, n ≥ 3. c, d Photographs of c HCT116 and d SW480 cells treated with 5× GI 50 pictilisib-cobimetinib (combination treatment) or the combination treatment ± 750 nM navitoclax. Cells were seeded in duplicate six-well plates for 2 d before 7 d drug treatment or 7 d drug treatment plus a further 5 d in drug-free media [41][42][43]. Whether or not the overall synergistic effect of combined MEK-and PI3K-inhibition requires activity at multiple nodes, or just at one critical node, remains to be determined. We also find that BIM, BBC3 and BAX are required for the combination treatment to induce apoptosis, in line with previous studies [35,44]; however, apoptosis measured by apoptotic protease activation is itself not essential for synergy since we observe that isogenic BBC3 or BAX knockout cells retained the synergistic growth inhibition response to MEK−PI3K inhibitors in the absence of caspase activation.
We find the selectivity profile of the PI3K inhibitor impacts the effectiveness of the combination regimen, as mTOR, dual mTOR-PI3K inhibitors and PI3K-inhibitors lacking activity against PIK3CA or poor potency against AKT THR308 phosphorylation have reduced synergy in combination with MEK inhibition. The consistency of these observations across multiple PI3K inhibitors, the majority of which have been clinically tested, suggests that other PI3K pathway inhibitors that can inhibit the class I alpha-isoform but lack MTOR inhibitory activity, such as copanlisib recently licensed for the treatment of follicular lymphoma [45], will have a similar activity profile in combination regimes. Almost all of the 47 human colorectal cancer cell lines tested exhibit a synergistic response to the optimal MEK−PI3K inhibitor combination treatment. However, colorectal cancer cell lines with inherent or acquired resistance to either of the treatment arms exhibit a reduced synergistic response to the combination regimen. Importantly, by exposing sensitive parent HCT116 cells to combination treatments for a prolonged period we generated cell lines that had lost the synergistic response that was initially exhibited to the MEK-PI3K combination, and subsequently also to the ERK-PI3K inhibitor combination that was at first able to overcome the resistance to the original MEK-PI3K combination. In both cases, loss of synergy is associated with the appearance of MEK and ERK mutations that were predicted or shown to disrupt inhibitor binding [26,46] without overtly affecting MEK/ERK activity.
In contrast, resistance to PI3K inhibitor monotherapy was found to involve pathway deregulation through loss of PTEN, which we found can be overcome with higher concentrations of PI3K inhibitor. The absence of mutations preventing inhibitor binding to the PIK3CA lipid kinase domain may reflect the fact that these types of mutation are not well tolerated, because they tend to also inhibit enzyme activity [47]. This suggests that combination resistance-inducing mutations preferentially occur in MEK1/2 or ERK2 because mutations that disrupt inhibitor binding are better tolerated in these targets and do not affect enzymatic activity. Alternatively, or in addition, the parent HCT116 cell line used in this study is highly dependent on MAPK signalling, and disrupting this pathway has a greater impact on the efficacy of the combination regimen.
ERK inhibitors are in early clinical studies, including trials to determine if these agents can rescue resistance to BRAF-and MEK-inhibitors [32,48,49]. In addition to our own observations, studies expressing exogenously mutated forms of ERK, or exposing cell lines to ERK inhibitors, have identified mutations associated with resistance to ERK inhibitors [34,50,51]. Interestingly, a recent computational approach to identify resistance mutations likely to arise following drug treatment predicts mutation of ERK2 Y36 as one of a number of mutation hotspots generating resistance to ERK inhibition [52]. Together, those studies and our own data suggest that resistance to ERK inhibitors may be inevitable in the clinic and that switching the inhibitor may delay, but will not avoid, the emergence of resistance.
Overall, we clearly show that a combination treatment targeting the MEK-and PI3K-pathways is insufficient to block the acquisition of resistance in human colorectal cancer cell line models, but the addition of a third agent, a BCL2 inhibitor, is able to do this. Several studies have shown that addition of a third agent can overcome acquired resistance or enhance the combinatorial effect of single agent MAPK-and PI3K-inhibitor regimens [53][54][55]. These observations suggest that the addition of a third agent, in our case a BCL2-family inhibitor, to create triple combinations may generate a fitness threshold that is too high to surpass by resistance mechanisms involving gene mutation or amplification. Importantly, the synergistic effect of the three-way combination was lost in cells with prior acquisition of resistance to one arm of the combination, suggesting that the triple therapy strategy is better suited to preventing the acquisition of resistance rather than overcoming existing intrinsic or acquired resistance from prior treatment. Preclinical studies have established a strong rationale for combined inhibition of PI3K-and MAPKpathways and numerous early clinical studies exploring this approach are either in progress or completed [56]. Results presented as meeting abstracts generally report only modest activity of the combinations, often limited by toxicity [56]. Hence the addition of a third agent, while potentially beneficial, would require careful consideration of tolerability. It is feasible that short exposures to higher concentrations may have similar or greater antitumour activity while being better tolerated than continuous exposures to lower doses. For example, short intermittent exposures to a MEK-PI3K inhibitor combination treatment are sufficient to drive an in vivo response in preclinical models [16,53]. Thus it is possible that the addition of a third agent may be tolerated clinically in such a dosing regimen and provide the opportunity to overcome or ameliorate the major clinical of clinical problem of resistance to targeted therapies.
Cell culture models, assays and treatment All cell lines were obtained from accredited cell banks, validated by DNA profiling and confirmed free of Mycoplasma spp. by PCR. HCT116 BAX−/− , HCT116 BBC3−/− and parent isogenic cell lines were obtained from Horizon Discovery (UK). GI 50 values (concentration that inhibits cell proliferation by 50%) were determined by sulphurhodamine blue staining [57]. Combination indices (CI) were determined using median effect analysis [58]. Cells were seeded in 96-well plates and at 2 days test compounds were added at a ratio of 1:1 of their respective GI 50 values. After a further 4 days, cells were sulphurhodamine blue stained and the CI calculated using Calcusyn (Biosoft, UK). A CI value of 1 indicates an additive effect, <1 synergy and >1 antagonism. Longer term colony growth assay experiments were performed in six-well plates. At the end of the experiment cells were fixed and stained using Crystal Violet. Apoptosis was quantified by assaying cleavage of a fluorescent caspase 3/7 substrate (Promega, USA) and autophagic cells were quantified using an LC3B immunofluorescence assay (Thermo Fisher Scientific, USA) [59].
Stable resistant lines were generated by incremental exposure to inhibitors. Cells were passaged when they reached confluence and compound exposure increased until the concentration reached 64 × GI 50 or the solubility limit.
Reporter cells were generated by transfection with a plasmid encoding a cap-dependent firefly luciferase and an EMCV IRES-driven Renilla luciferase (Promega, USA) downstream of an EF1α promoter [60]. ERK2 WT or ERK2 Y36H open reading frames were cloned into the expression vector pEFIRES-P [60]. Cells were transfected using lipofectamine (Thermo Fisher Scientific, USA) and cells with stable vector incorporation selected with puromycin.
Cells were reverse transfected with targeting siRNAs or positive or negative control siRNAs (Qiagen, Germany) using hiperfect cationic lipid (Qiagen, Germany). Cells were transfected in 96-well plates for viability assays or six-well plates to confirm the effects on target protein expression.
ERK2 kinase activity was determined using an EZ Reader II (PerkinElmer, USA). Conversion to product, and inhibition, was calculated relative to no enzyme and all reagent conditions. IC 50 values (concentration that reduced substrate conversion by half) were calculated using a fourparameter logistics fit (Dotmatics, UK).

Statistical analysis
Normal data distribution was determined using a D'Agostino and Pearson normality test. Significance for normally distributed data was determined with a one-way ANOVA and t test analyses and for non-normally distributed data using a Kruskal−Wallis test. A false discovery rate (q value) for multiple testing was calculated using the procedure of Benjamini, Krieger and Yekutieli (Graphpad, USA).

Compliance with ethical standards
Conflict of interest The authors are employees of The Institute of Cancer Research, which has a commercial interest in the discovery and development of PI3K inhibitors, including pictilisib, and operates a rewards-to-inventors scheme. PW and PAC have been involved in a commercial collaboration with Yamanouchi (now Astellas Pharma) and with Piramed Pharma and intellectual property arising from the program was licensed to Genentech. PAC is also involved in commercial collaborations with Merck KGaA and Astex Pharmaceuticals. PW was previously a scientific founder, consultant, Scientific Advisory Board member and stock holder of Piramed Pharma (acquired by Roche). He was a scientific founder, consultant, Scientific Advisory Board member, Board member and is a stockholder of Chroma Therapeutics. PW is also a consultant/advisor to Astex, CV6 Therapeutics and NextechInvest, and is a Non-Executive Director and stockholder of STORM Therapeutics. ICR receives research funding and income from numerous commercial partners.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons. org/licenses/by/4.0/.