Purpose:

AZD5363/capivasertib is a pan-AKT catalytic inhibitor with promising activity in combination with paclitaxel in triple-negative metastatic breast cancer harboring PI3K/AKT-pathway alterations and in estrogen receptor–positive breast cancer in combination with fulvestrant. Here, we aimed to identify response biomarkers and uncover mechanisms of resistance to AZD5363 and its combination with paclitaxel.

Experimental Design:

Genetic and proteomic markers were analyzed in 28 HER2-negative patient-derived xenografts (PDXs) and in patient samples, and correlated to AZD5363 sensitivity as single agent and in combination with paclitaxel.

Results:

Four PDX were derived from patients receiving AZD5363 in the clinic which exhibited concordant treatment response. Mutations in PIK3CA/AKT1 and absence of mTOR complex 1 (mTORC1)-activating alterations, for example, in MTOR or TSC1, were associated with sensitivity to AZD5363 monotherapy. Interestingly, excluding PTEN from the composite biomarker increased its accuracy from 64% to 89%. Moreover, resistant PDXs exhibited low baseline pAKT S473 and residual pS6 S235 upon treatment, suggesting that parallel pathways bypass AKT/S6K1 signaling in these models. We identified two mechanisms of acquired resistance to AZD5363: cyclin D1 overexpression and loss of AKT1 p.E17K.

Conclusions:

This study provides insight into putative predictive biomarkers of response and acquired resistance to AZD5363 in HER2-negative metastatic breast cancer.

Translational Relevance

AKT inhibitors such as AZD5363 (capivasertib) have emerged as a promising strategy for triple-negative breast cancers (TNBCs), the most aggressive subtype of the disease in need of new therapies tailored to specific tumor characteristics. Data from randomized phase II trials suggest that first-line treatment with AKT inhibitors combined with paclitaxel improve outcomes in metastatic TNBC, especially in tumors that harbor mutations in PTEN, PIK3CA, or AKT1. As confirmatory phase III trials are ongoing, the refinement of biomarkers of response and the identification of mechanisms of resistance to AKT inhibitors remain key to better tailor therapy for metastatic TNBC. In this work, we propose a composite response biomarker to AZD5363 in this setting, which is relevant for the clinical development of these drugs.

The PI3K/AKT pathway is implicated in numerous processes of carcinogenesis comprising tumor development, growth, and survival. PI3K is composed of a catalytic and a regulatory subunit encoded by the PIK3CA/B/D/G and PIK3R1/2 genes, respectively, being PIK3CA and PIK3R1 the most predominantly mutated in cancer (1, 2). Following receptor tyrosine kinases activation, PI3K phosphorylates phosphatidylinositol (4, 5)-diphosphate into phosphatidylinositol (3, 4, 5)-triphosphate (PIP3). This reaction is reversed by the phosphatase and tensin homologue in chromosome 10 (PTEN, encoded by the PTEN gene; ref. 3). Accumulation of PIP3 promotes the activation of AKT and downstream proteins such as the proline-riche AKT substrate of 40 kDa (PRAS40), the tuberous sclerosis complex 2 (TSC2), or the Forkhead Box O3A transcription factor (FOXO3A), among others (4). In addition, AKT relieves the negative regulation of glycogen synthase kinase 3 beta (GSK3β) on cyclin D1 (encoded by CCND1) transcription and it also promotes shuttling of the cyclin-dependent kinase inhibitors p21 and p27 (CDKN1A and CDKN1B gene products) in the cytoplasm, thereby repressing their nuclear cell cycle inhibitory function and promoting cell cycle progression.

On the basis of different studies, 32% to 57% of metastatic HER2-negative breast cancers (BCs) harbor mutations in PIK3CA, AKT1, or PTEN, leading to activation of the PI3K/AKT/mTOR signaling cascade (5, 6), and providing the rationale to implement inhibitors of the pathway (7). AZD5363/capivasertib, an oral bioavailable ATP-competitive AKT1/2/3 inhibitor, has shown preclinical activity in HER2-negative breast cancer models and the combination of AZD5363 with paclitaxel or fulvestrant has been explored in clinical trials (8–10).

Response to AZD5363 has been observed in patients with tumors harboring activating mutations in PIK3CA and AKT1 (11, 12). However, not all patients benefit equally from treatment with AKT inhibitors. The randomized phase II PAKT trial (NCT02423603) has shown that capivasertib in combination with paclitaxel may be a promising therapeutic strategy for triple-negative metastatic breast cancers (TNBC), particularly in patients with tumors harboring PIK3CA/AKT1/PTEN alterations (10). In contrast, clinical benefit for the same combination was not observed in estrogen receptor–positive (ER+) breast cancers irrespective of the PIK3CA/AKT1 status of the tumor (BEECH trial, NCT01625286; ref. 9). In the FAKTION trial (NCT01992952), the addition of capivasertib to fulvestrant led to a significant improvement in progression-free survival (PFS) when compared with fulvestrant alone in ER+/HER2 metastatic BC, albeit these biomarkers did not help stratify responders versus nonresponders (13). Hence, the identification of accurate predictive biomarkers of response and resistance may improve the clinical benefit of capivasertib-containing regimens.

In this work, we sought to identify potential biomarkers of response to AZD5363 in HER2-negative breast cancer, and to decipher the mechanism of action and acquired resistance to AZD5363-containing therapies.

Study design

This study was designed to identify predictive biomarkers of response to AZD5363 that can be effectively used for patient stratification. We assessed AZD5363 sensitivity in a cohort of 28 murine xenografts from patients with metastatic breast cancer. Tumors were harvested and formaldehyde-fixed or flash-frozen for posterior proteomic and genomic analyses. Genetic engineered cell lines were used for validation of results.

Collection of tumor samples and establishment of patient-derived xenografts

Fresh tumor samples from patients with breast cancer were collected following an Institutional Research Board approved protocol and the associated written informed consent. For RNA sequencing, the tumor samples were snap frozen. The study was compliant with the declaration of Helsinki.

Experiments were conducted following the European Union's animal care directive (2010/63/EU) and were approved by the Ethical Committee of Animal Experimentation of the Vall d'Hebron Research Institute (Barcelona, Spain), the Catalan Government or by the National Research Ethics Service, Cambridgeshire (14) and https://caldaslab.cruk.cam.ac.uk/bcape/. Experiments were ended when the tumor volume surpassed 1,500 mm3 or a decline in mouse welfare was observed, including mouse weight loss >20%.

Six-week-old female athymic nude mice (HsdCpb:NMRI-Foxn1nu, Harlan Laboratories) were housed in air-filtered laminar flow cabinets with a 12-hour light cycle and food and water ad libitum. Surgical or biopsy specimens from primary tumors or metastatic lesions were immediately implanted in mice as fragments of 50 mm3. Animals were supplemented with 1 μmol/L 17β-estradiol (Sigma) in their drinking water (15). Upon xenograft growth, tumor tissue was implanted on the lower flank of new recipient mice, and equally distributed when tumor volume was 150 to 300 mm3. Mice were treated with AZD5363 twice daily, on days 1 to 4 every week with 100 mg/kg of AZD5363 in 10% DMSO, 2% HCl 1mol/L in 25% kleptose, or with 7.5 mg/kg intraperitoneal paclitaxel in 0.9% NaCl solution, once weekly. In combination treatments, paclitaxel was injected one hour before the first AZD5363 dose in every cycle.

Tumor xenografts were measured with calipers and tumor volumes were determined using the ellipsoid formula: (length × width2) × (π/6). All the in vivo experiments contained an untreated control arm. The antitumor activity was determined by calculating the percentage change in tumor volume on day 25 (unless the maximum tumor volume was surpassed before, at which time point the data were taken): % change in tumor volume = (Vday 25Vday 1)/Vday 1 ×100. To classify the response of the subcutaneous implants, we used modified RECIST criteria, to be based on tumor volume (16, 17). For AZD5363-sensitive patient-derived xenografts (PDX), the best response was defined as the minimum value of the mean % tumor volume change that was sustained for at least 10 days. Complete response (CR) was set as best response ≤−95%, partial response (PR) as −95%<best response≤−30%, stable disease (SD) as −30%<best response≤+20%, and progressive disease (PD) as best response>+20%. The models that displayed an AZD5363 response categorized as SD, PR, and CR were considered AZD5363-sensitive. PDX versus patient responses were compared as follows: we considered concordant any response that derived in clinical/preclinical benefit (or nonbenefit), both in patient and PDX. Clinical/preclinical benefit included CR, any PR, or stable disease that lasted at least 24 weeks in the patient or 6 weeks PDX (arbitrary cut-off taken from internal laboratory data, as the standard time needed to observe sustained treatment responses in ER+ PDXs). Nonbenefit included PD.

To induce AZD5363 resistance, AZD5363-sensitive tumors were chronically treated with AZD5363 until individual tumors grew. At this time point, tumors were harvested and implanted into new recipient mice. Dosing schedule was reinitiated 3–4 days postsurgery and lasted until tumors reached 200 mm3.

Statistical analysis

GraphPad Prism 6.0 and R version 3.6 statistical software packages were used for statistical analyses. A bootstrap resampling procedure (n = 2,000) was used to calculate the SE in the percentage of change in tumor volume relative to untreated. D'Agostino–Pearson omnibus test was performed to check the normality assumption in all comparative studies. If the null hypothesis of normality was not rejected, we assumed Gaussian distribution of the samples, but if the sample size was too small or the hypothesis was rejected, we did not assume it. For the comparative experiments of biomarkers between AZD5363-sensitive and AZD5363-resistant groups, we used unpaired t test with Welch correction or Mann–Whitney test for normally and not normally distributed data. For biomarker comparison, the data were not summarized in PDX mean but all measurements were used for data analysis. A linear mixed model with PDX as a random factor was performed to test the change between baseline and on-AZD5363 levels and to compare baseline and on-AZD5363 treatment levels between sensitive and resistant groups. In addition, an interaction test was fitted to study if the magnitude of change was statistically different between sensitive and resistant models. Adjustment for multiple testing was performed in each biomarker by controlling the FDR according to the Benjamini and Hochberg method. The statistical comparison of therapy efficacy in vivo was performed with one-way ANOVA with Bonferroni posttest. The optimum cut-off points established in this study were selected by the Youden index, which maximizes the sum of the sensitivity and specificity in each biomarker analyzing the ROC curve. See further methods in Supplementary Material.

Genetic alterations in the PI3K/AKT pathway associated with AZD5363 sensitivity

PIK3CA/AKT1/PTEN alterations have been proposed as selection biomarkers in clinical trials testing the activity of AKT1/2/3 inhibitors (AKTi; refs. 10, 18, 19). We aimed to extend our understanding on the predictive value of genetic/proteomic alterations of these biomarkers as well as other putative alterations within the PI3K/AKT pathway. To this aim, we established 28 murine PDX models derived from HER2-negative breast tumors and tested the antitumor activity of AZD5363 using an intermittent schedule that mimics exposure in patients (12). Of note, four of these PDX models were derived from patient's biopsies taken before, during or after treatment with an AZD5363-containing therapy from two clinical trials: BEECH (NCT01625286; evaluating safety, tolerability, and efficacy of AZD5363 in combination with paclitaxel in breast cancer, including a subgroup of patients harboring PIK3CA mutations) and D3610C0001 part E (NCT01226316, evaluating AZD5363 in combination with fulvestrant in patients with advanced/metastatic ER+ breast cancers harboring AKT1 mutations). AZD5363 monotherapy blocked tumor growth in nine models, causing either disease stabilization or tumor regression. Consistently, AZD5363 induced a higher change in the final tumor volume compared with untreated tumors in sensitive models than in resistant PDXs (Fig. 1A and Supplementary Table S1; P < 0.0001, unpaired t test with Welch correction). Out of the nine AZD5363-responsive PDXs, five were TNBC and four were ER+. Among the nonresponders, 12 were TNBC and 7 ER+. Treatment response was neither associated with the molecular classification according to IHC nor with the PAM50 classification of the PDXs (Fig. 1A; refs. 20, 21).

Figure 1.

Therapeutic efficacy of AZD5363 in PDX correlates with patient's response and AKT phosphorylation by IHC. A, On the top, waterfall plot representing the tumor growth of 28 PDX treated with 100 mg/kg AZD5363 twice daily on days 1-4 every week. The percentage change from initial tumor volume is shown at the time point of best response. +20%, −30%, and −95% are marked by dotted lines to indicate the range of PD, SD, PR, and CR. Below, the percentage change in tumor growth inhibition of AZD5363-treated versus the control arm is represented. Striped bars indicate PDX models derived from patients who were treated with an AZD5363-containing regimen. Error bars indicate SD from at least two tumors. The box underneath summarizes the subtype classification, based on IHC or PAM50, as well as the relevant genomic alterations in the PI3K/AKT/MTOR pathway. PTEN status by IHC is also provided, based on a H-score ≤ 10 cut-offf. Regarding copy-number alterations, partially colored box indicates heterozygosity. B, Representation of PDXs versus patient response for the four models derived from BEECH and D3610C0001 part E. The size of the circles represents the number of subjects in each response category. C, IHC scoring of pAKT S473 in formalin-fixed, paraffin-embedded (FFPE) samples from untreated PDXs in relationship to AZD5363-treatment response. Mean of the PDXs in each group and SEM is indicated. PTEN+, H-score > 10; PTEN−, H-score ≤ 10; P value, unpaired t test with Welch correction. The optimum cut-off point was established by the Youden index, which maximizes the sum of the sensitivity and specificity of the biomarker analyzing the ROC curve.

Figure 1.

Therapeutic efficacy of AZD5363 in PDX correlates with patient's response and AKT phosphorylation by IHC. A, On the top, waterfall plot representing the tumor growth of 28 PDX treated with 100 mg/kg AZD5363 twice daily on days 1-4 every week. The percentage change from initial tumor volume is shown at the time point of best response. +20%, −30%, and −95% are marked by dotted lines to indicate the range of PD, SD, PR, and CR. Below, the percentage change in tumor growth inhibition of AZD5363-treated versus the control arm is represented. Striped bars indicate PDX models derived from patients who were treated with an AZD5363-containing regimen. Error bars indicate SD from at least two tumors. The box underneath summarizes the subtype classification, based on IHC or PAM50, as well as the relevant genomic alterations in the PI3K/AKT/MTOR pathway. PTEN status by IHC is also provided, based on a H-score ≤ 10 cut-offf. Regarding copy-number alterations, partially colored box indicates heterozygosity. B, Representation of PDXs versus patient response for the four models derived from BEECH and D3610C0001 part E. The size of the circles represents the number of subjects in each response category. C, IHC scoring of pAKT S473 in formalin-fixed, paraffin-embedded (FFPE) samples from untreated PDXs in relationship to AZD5363-treatment response. Mean of the PDXs in each group and SEM is indicated. PTEN+, H-score > 10; PTEN−, H-score ≤ 10; P value, unpaired t test with Welch correction. The optimum cut-off point was established by the Youden index, which maximizes the sum of the sensitivity and specificity of the biomarker analyzing the ROC curve.

Close modal

With combination treatment, the four PDXs derived from BEECH or D3610C0001 part E trials recapitulated the patient's response (Fig. 1B). The first model (PDX292.1) was derived from a biopsy collected prior to treatment with AZD5363 plus paclitaxel. This model mirrored the patient's confirmed PR (cPR, −40%; Supplementary Fig. S1A and B) lasting for 373 days, showing a tumor growth reduction of −23% in the PDX. PDX279.3, derived from a progression biopsy after 411 days of treatment (best response was cPR, −47%; Supplementary Fig. S1C and D), exhibited a short SD for 37 days, until final progression to the same treatment combination (PD, +107%), mimicking the therapy resistance observed in the patient. Third, PDX406.1 derived from a biopsy prior to AZD5363 plus fulvestrant exposure (best response in the patient being cPR for 461 days, −56%; Supplementary Fig. S1E and S1F), exhibited a PR (−80%) to the same combination. Finally, PDX433.2 was derived from an on-treatment biopsy after one cycle of AZD5363 plus fulvestrant in a patient with an unconfirmed, short PR (83 days, −68%; Supplementary Fig. S1G and S1H) and exhibited a PR (−53%) to the same treatment.

To identify genetic alterations associated with response that could serve as predictive biomarkers, targeted tumor-sequencing (MSK-IMPACT) or whole exome sequencing was performed in the PDX models (Supplementary Table S2). We first analyzed the composite selection biomarker used in the phase III clinical trials testing AKT inhibitors plus paclitaxel, namely, mutations in PIK3CA/AKT1/PTEN, and obtained an accuracy of 64% (Table 1). Unexpectedly, the exclusion of PTEN alterations increased the accuracy to predict AZD5363 monotherapy response to 86%. Of note, PTEN protein levels were fully concordant with PTEN alterations in this dataset (Fig. 1A). Addition of PIK3R1 in the composite biomarker did not improve accuracy (82%), although PIK3R1 alterations can result in PI3K-pathway activation (2). Nonetheless, we observed antitumor activity of AZD5363 in a head and neck squamous cell carcinoma xenograft model harboring PIK3R1 loss (H31; Supplementary Fig. S2A). We also tested if MTOR mutations modified the accuracy of the composite biomarker, because a positive relationship was found between treatment response and concomitant AKT1/mTORC1 pathway alterations, in gynecological cancers (11). In our panel, however, the absence of additional activating alterations downstream AKT (i.e., MTOR or TSC1/2 mutations) increased the accuracy of the composite biomarker to 89%. This is exemplified by the different response observed in AZD5363-sensitive PDX225- versus AZD5363-resistant PDX313, both with AKT1 p.E17K-mutations, but the latter harboring an in silico predicted activating mutation in MTOR p.W1456G [PolyPhen score (22): 0.967; Fig. 1A; Supplementary Table S2]. Consistently, the MTOR-mutant PDX313 displayed elevated mTORC1 activity (phospho S6 ribosomal protein, pS6 and phospho eukaryotic translation initiation factor 4E-binding protein 1, p4EBP1) compared with PDX225 (Supplementary Fig. S2B); showed lower AKT phosphorylation, likely due to negative feedback regulation (23), and displayed limited reduction of S6 and 4EBP1 phosphorylation upon treatment with AZD5363 (Supplementary Fig. S2B). Altogether, these data indicate that a composite biomarker that includes PIK3CA and AKT1 mutations, regardless of PTEN status and in the absence of mTORC1-activating alterations (e.g., MTOR mutations or TSC2 loss), is associated with preclinical sensitivity to AZD5363 monotherapy.

Table 1.

Genomic signature predictive of AZD5363 sensitivity.

Genomic signature predictive of AZD5363 sensitivity.
Genomic signature predictive of AZD5363 sensitivity.

Low AKT phosphorylation identifies AZD5363-resistant tumors

We surmised that the magnitude of AKT activity may also help to select the population of patients that will benefit from AZD5363 in the clinic. To test this hypothesis, we measured the levels of AKT phosphorylation at S473 and T308 sites in PDX models. By IHC, we found that low S473 phosphorylation on AKT significantly discriminated AZD5363-resistant PDXs from sensitive ones (Fig. 1A and C), an effect that was also observed by reverse-phase protein array (RPPA) or Western blot analysis (Supplementary Fig. S3A–S3C). Phosphorylation of T308 phosphosite was less accurate in discriminating AZD5363 resistance by all three techniques (Supplementary Fig. S3D–S3G). Of note, the majority of the PTEN-mutated PDXs had high levels of AKT phosphorylation at S473 (OR, 6.43).

Low S6 S235 phosphorylation on-treatment is associated with AZD5363 response

Absence of mTORC1 activating mutations increased the accuracy of the composite biomarker to predict AZD5363 response. In line with this observation, all tumors with low AKT phosphorylation (e.g., resulting from lack of PI3K activation or high mTORC1 activity) were resistant to AZD5363 (Fig. 1C). Also, previous studies on resistance to PI3K inhibitors suggested that active mTORC1 or overexpression of ribosomal S6 kinase 3 and 4 (RSK3/4) were associated with lack of response to PI3K inhibitors (24, 25), and this is concordant with the observations from PDX313 versus PDX225 (Supplementary Fig. S2B and S2C). Therefore, we hypothesized that AZD5363-resistant models may exhibit higher mTORC1 activity (or limited mTORC1 pathway inhibition upon pharmacological pressure), compared with sensitive PDXs. We then quantified the baseline phosphorylation of AKT and mTORC1 targets, including S6 and 4EBP1, and their modulation by AZD5363 in short-term AZD5363-treated tumors by Western blot analysis. As mentioned, baseline levels of pAKT S473 and pAKT T308 were significantly higher in sensitive than in resistant models (Figs. 1C and 2A). Overall, in this PDX cohort treatment with AZD5363 resulted in increased pAKT T308/S473, decreased pPRAS40 T246, pFOXO1/3A T24/T32, pS6 S235, pS6 S240, and p4EBP1 T37/46, and unchanged p4EBP1 T70 (Fig. 2). The mild modulation of p4EBP1 versus marked decrease in pS6 suggested that secondary pharmacologic activity on ribosomal protein S6 kinase (S6K1) might be relevant in vivo (8). In addition, the on-treatment values for pS6 S235 were significantly lower in sensitive models compared with the resistant ones (P <0.001) and they significantly predicted AZD5363 response (Supplementary Fig. S3H; cut-off: 0.26, sensitivity: 100%, and specificity: 92.3%). We noted that baseline pS6K1 T389 was similar in AZD5363-sensitive and AZD5363-resistant models, measured by RPPA (Supplementary Fig. S4A) and there was no evidence to attribute the residual pS6 S236 in AZD5363-resistant tumors, to higher baseline MEK/ERK/RSK signaling, nor to increased signaling after treatment with AZD5363 (Supplementary Fig. S4B and S4C; ref. 26). Instead, a trend toward higher protein kinase C-α (PKCα) and pPKCδ S664 in AZD5363-resistant models was observed (Supplementary Fig. S4D; ref. 27).

Figure 2.

PI3K/AKT pathway inhibition is independent of AZD5363 therapeutic efficacy. A, Analysis of pAKT T308 and S473 levels assessed by Western blot analysis in AZD5363-resistant (PD) versus sensitive (SD/PR/CR) PDX in the absence (dots) and presence (squares) of AZD5363 treatment. For illustration purposes, only the mean value of each PDX was plotted; however, for the statistical analysis all available data were used. A linear mixed model was used to compare baseline levels and on-AZD5363 treatment levels between sensitive and resistant groups using Benjamini and Hochberg to adjust for multiple testing (see more details in Statistical analysis). n.s., not significant. B, Analysis of AKT/mTORC1 pathway modulation by Western blot analysis as in A.

Figure 2.

PI3K/AKT pathway inhibition is independent of AZD5363 therapeutic efficacy. A, Analysis of pAKT T308 and S473 levels assessed by Western blot analysis in AZD5363-resistant (PD) versus sensitive (SD/PR/CR) PDX in the absence (dots) and presence (squares) of AZD5363 treatment. For illustration purposes, only the mean value of each PDX was plotted; however, for the statistical analysis all available data were used. A linear mixed model was used to compare baseline levels and on-AZD5363 treatment levels between sensitive and resistant groups using Benjamini and Hochberg to adjust for multiple testing (see more details in Statistical analysis). n.s., not significant. B, Analysis of AKT/mTORC1 pathway modulation by Western blot analysis as in A.

Close modal

We further analyzed whether the serum/glucocorticoid kinase/n-myc downstream regulated 1 (SGK/NDRG1) axis may bypass AKT signaling and maintain mTORC1 signaling and cell proliferation (28, 29). We quantified SGK/NDRG1 activity by measuring SGK1 levels and the downstream phosphorylation on NDRG1. We found no association between these biomarkers and resistance to AZD5363 (P = 0.91/P = 0.50, Mann–Whitney test; Supplementary Fig. S4E). Overall, we observed a superior inhibition of the S6 kinase axis in AZD5363-sensitive than in resistant PDXs, likely due to lack of parallel signaling (e.g., PKCα and/or PKCδ) in sensitive tumors.

Treatment with AZD5363 induces cell-cycle arrest via downregulation of cyclin D1 in AZD5363-sensitive BC PDXs

Preclinical studies of AZD5363 using cell line models revealed the ability of the AKT inhibitor to reduce cell proliferation and induce apoptosis (8). To further unravel the mechanism of action of AZD5363 in vivo, we measured Ki67 and cleaved caspase 3 in short-term AZD5363-treated PDXs. The percentage of Ki67-positive cells in untreated tumors was lower in AZD5363-sensitive models compared with the resistant ones (P = 0.005; Fig. 3A). In addition, treatment with AZD5363 resulted in a greater reduction of Ki67 in sensitive PDXs compared with the resistant tumors (P < 0.001). We also noted that AZD5363 did not induce cleaved caspase 3 across AZD5363-sensitive tumors, consistent with previous findings for this dosing schedule (Supplementary Fig. S4F; ref. 8).

Figure 3.

Proliferation arrest is induced by AZD5363 in sensitive tumors. A, Analysis of Ki67-positive cells by IHC in FFPE samples from untreated (dots) and AZD5363-treated (squares) PDXs in relationship to AZD5363-treatment response. For illustration purposes, only the mean value of each PDX was plotted; however, for the statistical analysis all available data were used. A linear mixed model was used to compare baseline levels and on-AZD5363 treatment levels between sensitive and resistant groups using Benjamini and Hochberg to adjust for multiple testing (see more details in Statistical analysis). B, Analysis of cyclin D1 expression by IHC in the same samples as in A. A linear mixed model was used to calculate the statistics comparing untreated and treated samples in each group (Benjamini and Hochberg adjustment method for multiple testing, see more details in Statistical analysis). The Youden index, which maximizes the sum of the test sensitivity and specificity, was used to select the optimum cut-off point.

Figure 3.

Proliferation arrest is induced by AZD5363 in sensitive tumors. A, Analysis of Ki67-positive cells by IHC in FFPE samples from untreated (dots) and AZD5363-treated (squares) PDXs in relationship to AZD5363-treatment response. For illustration purposes, only the mean value of each PDX was plotted; however, for the statistical analysis all available data were used. A linear mixed model was used to compare baseline levels and on-AZD5363 treatment levels between sensitive and resistant groups using Benjamini and Hochberg to adjust for multiple testing (see more details in Statistical analysis). B, Analysis of cyclin D1 expression by IHC in the same samples as in A. A linear mixed model was used to calculate the statistics comparing untreated and treated samples in each group (Benjamini and Hochberg adjustment method for multiple testing, see more details in Statistical analysis). The Youden index, which maximizes the sum of the test sensitivity and specificity, was used to select the optimum cut-off point.

Close modal

Given that S6K mediates efficient cap-dependent translation of cyclin D1 and that AKT/GSK3β axis regulates cyclin D1 stability, (4, 30), we posited that treatment with AZD5363 impaired cell-cycle progression through the CDK4/6–cyclin D1 restriction point in sensitive models. Therefore, we quantified cyclin D1 by IHC in PDXs before and after treatment with AZD5363. These experiments revealed that most of AZD5363-resistant PDXs expressed low levels of cyclin D1, compared with AZD5363-sensitive PDXs (Fig. 3B; S4D, 12 of 16, 75% vs. 2 of 8, 25%, ROC curve cut-off H-score ≤ 13.3, ROC P = 0.066), suggesting that, in AZD5363-resistant tumors, cell-cycle progression was not dependent on cyclin D1. We also observed that, although AZD5363 downregulated cyclin D1 in all the models expressing cyclin D1 (Fig. 3B), the reduction in cyclin D1 was more relevant in AZD5363-sensitive models (P < 0.001). Altogether, these results suggest that cyclin D1 downregulation and cell-cycle blockade is an important mechanism of action of AZD5363 in vivo.

CCND1 amplification and loss of AKT1 p.E17K mutation are associated with AZD5363 resistance

To uncover potential mechanisms of resistance to AZD5363, we treated PDX225 with AZD5363 overtime until two tumors became refractory to the drug after having exhibited a disease stabilization that lasted >100 days (Fig. 4A, left panel). The acquisition of resistance was confirmed after serial transplantation of the AZD5363-progressing tumors in new recipient mice, which were further treated with either vehicle or AZD5363, and AZD5363-resistant tumors were insensitive to AZD5363 treatment (Fig. 4A, right panel). In these tumors, we identified an amplification in 11q13, encompassing CCND1, encoding for cyclin D1, which was confirmed at the protein level by IHC (Fig. 4B and C). In addition, PDX225-resistant tumors exhibited an attenuated reduction of cyclin D1 and maintained Ki67 positivity upon treatment with AZD5363 (Fig 4C), suggesting that overexpression of CCND1 limited the antiproliferative activity of AZD5363 previously seen in this model. We further conducted experiments to investigate whether exogenous cyclin D1 overexpression conferred resistance to AZD5363 in 3D cultures of two cell lines harboring a PIK3CAmutation. Specifically, we observed that overexpression of cyclin D1 resulted in a 3.5-fold increase of the IC50 for AZD5363 in T47D cells but not in MCF7 cells (Supplementary Fig. S5A). This was not surprising as MCF7 cells express higher baseline levels of cyclin D1 compared with T47D and were intrinsically less sensitive to AZD5363. These data are consistent with an inverse correlation between cyclin D levels and sensitivity to AZD5363. Biochemically, cyclin D1 overexpression in T47D cells resulted in increased phosphorylation of pRb and higher expression of the E2F target gene cyclin E2 compared with the empty vector control (Supplementary Fig. S5B).

Figure 4.

Acquired resistance to AZD5363 is associated with CCND1 amplification in PDX225 or by loss of AKT1 p.E17K in patient 433. A, Tumor growth inhibition of the AZD5363-sensitive PDX225, acquisition of drug resistance and confirmation of lack of response to AZD5363 after serial transplantation. B, Analysis of the copy-number alterations (CNA) detected in AZD5363 sensitive (X-axis) and resistant (Y-axis) PDX225. Acquired CNA are highlighted in red. C, IHC staining of cyclin D1 and Ki67, in a representative FFPE tumor from AZD5363 sensitive (top) and resistant (bottom) PDX225 tumor. Scale bars, 100 μm. D, Analysis of AKT1 p.E17K allele frequency and levels of tumor markers in serum before treatment initiation, during and at progression in the same patient as the PDX data in E. E, Analysis of the variant allele frequency (VAF) in the AZD5363-sensitive PDX433.2 (derived from Pt433, on-treatment) versus PDX433.3 (derived from Pt433, post-AZD5363). Acquired changes in VAFs are highlighted in red.

Figure 4.

Acquired resistance to AZD5363 is associated with CCND1 amplification in PDX225 or by loss of AKT1 p.E17K in patient 433. A, Tumor growth inhibition of the AZD5363-sensitive PDX225, acquisition of drug resistance and confirmation of lack of response to AZD5363 after serial transplantation. B, Analysis of the copy-number alterations (CNA) detected in AZD5363 sensitive (X-axis) and resistant (Y-axis) PDX225. Acquired CNA are highlighted in red. C, IHC staining of cyclin D1 and Ki67, in a representative FFPE tumor from AZD5363 sensitive (top) and resistant (bottom) PDX225 tumor. Scale bars, 100 μm. D, Analysis of AKT1 p.E17K allele frequency and levels of tumor markers in serum before treatment initiation, during and at progression in the same patient as the PDX data in E. E, Analysis of the variant allele frequency (VAF) in the AZD5363-sensitive PDX433.2 (derived from Pt433, on-treatment) versus PDX433.3 (derived from Pt433, post-AZD5363). Acquired changes in VAFs are highlighted in red.

Close modal

We also obtained pretreatment, on-treatment, and posttreatment biopsies from patients participating in two phase II clinical trials testing the activity of AZD5363 in combination with either paclitaxel or fulvestrant. For patient 433 (best response PR, −68%; Supplementary Fig. S1H), amplicon sequencing comparing pretreatment, on-treatment and posttreatment liver metastases unveiled a decrease and subsequent loss of the AKT1 p.E17K mutation in the on-treatment and posttreatment biopsies, respectively (Fig. 4D). In agreement with this observation, the PDX model obtained at disease progression (PDX433.3) did not harbor the AKT1 p.E17K mutation, compared with the model derived from the on-treatment biopsy (Fig. 4E). The fact that 68% of the identified variants were conserved between PDX433.2 and PDX433.3, suggests that an AKT1 p.E17K-devoid subclone expanded upon treatment with AZD5363, rather than the mutation in AKT1 being reverted.

AZD5363 plus paclitaxel results in preclinical benefit in TP53 wild-type tumors

As AZD5363 is being combined with paclitaxel in the clinic, we also treated 15 PDX models with paclitaxel both as single agent and in combination with AZD5363 (Fig. 5A; Supplementary Table S1). Paclitaxel was administered at 7.5 mg/kg, a dose that mimicked its concentration in human plasma (Supplementary Fig. S5C). The best response to single-agent paclitaxel in this cohort of PDX was PD (Fig. 5A, top panel). In addition to PDX044 and PDX406.1, which were intrinsically sensitive to AZD5363 monotherapy, two models that lacked response to either paclitaxel or AZD5363 significantly benefited from the combined schedule, resulting in SD (PDX292.1 and PDX093; Fig. 5A, middle panel). IHC analysis of PDX292.1 and PDX093 revealed that the combination therapy displayed a trend toward lower percentage of Ki67-positive cells than the control arm or single-agent arm treatments in both PDX models (Fig. 5B), indicative of decreased proliferation. We noted that the combination-responders PDX292.1 and PDX093 were TP53 wt (wild-type). We also observed that treatment with AZD5363 and/or paclitaxel resulted in activation of p53 by phosphorylation on S15 in PDX292.1 (Fig. 5C). This occurred alongside with a marked reduction of cyclin B1 in the combination-treated tumor while p21 remained unchanged. This result suggests a p53-mediated inhibition of cyclin B1 transcript and p53 preventing G2–M transition (31). Accordingly, cyclin A2 and cyclin E2 were upregulated in the combination-treated tumor, indicating that tumor cells may have arrested in S-phase. In contrast, there was no evidence that knockout/knockdown of TP53 in MCF10A or in MCF7 cells impacted in AZD5363 plus paclitaxel response in two-dimensional (2D) cell cultures (Supplementary Fig. S5D and S5E).

Figure 5.

Therapeutic efficacy of AZD5363 in combination with paclitaxel is observed in TP53 wt PDX. A, Waterfall plot representing the tumor growth of 15 PDXs treated with paclitaxel (top) or with paclitaxel in combination with AZD5363 (bottom). Percentage change from initial tumor volume at the time point of best response is shown. +20%, −30%, and −95% are marked by dotted lines to indicate the range of PD, SD, PR, and CR. Striped bars indicate PDX models derived from patients who were treated with an AZD5363-containing regimen. Error bars indicate SD from at least two tumors. The box underneath summarizes the molecular subtype, TP53 status and whether the patient had received taxanes before PDX establishment. B, Quantification of Ki67 in two combination-sensitive models (PDX093 and PDX292). One-way ANOVA with Tukey multiple comparison test was used to calculate the statistics. C, Western blot data showing that treatment with AZD5363 and/or paclitaxel resulted in activation of p53 by phosphorylation on S15 in PDX292.1. D, RNA expression of Coutant et al. mutant TP53-related gene signature in paired samples from pre-, on- and post-AZD5363 biopsies (labeled as 1, 2, and 3, respectively) of patients 279, 292, and 320 from the BEECH trial. n.s., not significant.

Figure 5.

Therapeutic efficacy of AZD5363 in combination with paclitaxel is observed in TP53 wt PDX. A, Waterfall plot representing the tumor growth of 15 PDXs treated with paclitaxel (top) or with paclitaxel in combination with AZD5363 (bottom). Percentage change from initial tumor volume at the time point of best response is shown. +20%, −30%, and −95% are marked by dotted lines to indicate the range of PD, SD, PR, and CR. Striped bars indicate PDX models derived from patients who were treated with an AZD5363-containing regimen. Error bars indicate SD from at least two tumors. The box underneath summarizes the molecular subtype, TP53 status and whether the patient had received taxanes before PDX establishment. B, Quantification of Ki67 in two combination-sensitive models (PDX093 and PDX292). One-way ANOVA with Tukey multiple comparison test was used to calculate the statistics. C, Western blot data showing that treatment with AZD5363 and/or paclitaxel resulted in activation of p53 by phosphorylation on S15 in PDX292.1. D, RNA expression of Coutant et al. mutant TP53-related gene signature in paired samples from pre-, on- and post-AZD5363 biopsies (labeled as 1, 2, and 3, respectively) of patients 279, 292, and 320 from the BEECH trial. n.s., not significant.

Close modal

To further test whether this finding was applicable in patients, we asked whether tumors that progressed upon this combination in the clinic harbored a TP53-mutation. However, we could not assess the impact of TP53 mutation in primary resistance in the BEECH trial, as this alteration was detected in only 8 of 98 patients [using circulating tumor DNA (ctDNA)]. We then performed DNA and RNA sequencing in matched tumor samples (pretreatment, on-treatment and posttreatment) from 3 patients who responded to AZD5363 plus paclitaxel. Of note, these patients had a confirmed PR while on study (patient 279, patient 292, and patient 320 with −47%, −40%, and −49% tumor reduction, respectively) and progressed after 411, 373, and 402 days of treatment, respectively. Even though no TP53 mutation was detected in any of the biopsies, one of the patients (Pt 279) acquired an increased expression of genes involved in TP53-mutant signaling in the progression biopsy (Fig. 5D; Supplementary Fig. S5B; refs. 32, 33). Interestingly, the other two patients (Pt 292 and Pt 320), overexpressed AURKA and KIF5C in the progression biopsies, respectively (Fig. 5D; Supplementary Fig. S5B). AURKA encodes for Aurora kinase A and is related to increased AKT/β-catenin signaling as well as attenuation of p53 function (34, 35), both of which could result in AZD5363 plus paclitaxel resistance (36). Conversely, KIF5C codifies for a kinesin implicated in mitotic chromatids segregation and has been previously associated with docetaxel resistance in vitro (37). These results suggest that a TP53-mutant-like scenario either by deleterious mutation on TP53, expression of TP53-mutant gene signature, or by alterations in p53-related genes may limit the efficacy of AZD5363 plus paclitaxel therapy.

This study aimed to refine the selection of enrichment biomarkers of response to AZD5363 in the preclinical setting that could potentially inform clinical development of this agent in HER2-negative breast cancer. Using PDX models, we have identified a composite biomarker comprising PIK3CA/AKT1 mutations, in the absence of mTORC1-activating alterations, which enriches for AZD5363 monotherapy response.

Results of two randomized, double-blind, placebo-controlled, phase II trials—PAKT and LOTUS (NCT02162719)—showed that adding capivasertib (AZD5363) or ipatasertib (GDC-0068, another pan-AKT catalytic inhibitor), respectively, to paclitaxel increased PFS in patients with TNBC, especially in those harboring PIK3CA/AKT1/PTEN alterations (10, 18). In line with these results, our data suggest that PIK3CA and AKT1 mutations enrich for AZD5363 benefit in a series of HER2-negative PDX models, some of which were derived from patients that received capivasertib in the clinic. While this is true for PIK3CA and AKT1 mutations, PTEN alterations do not seem to associate with response to AZD5363 as single agent in our cohort of PDXs. However, it must be noted that we have not comprehensively tested these biomarkers in PDX models treated with the combination of AZD5363 and paclitaxel as in the clinic, and hence the predictive value of PTEN in the response to AKT inhibitors combined with paclitaxel cannot be completely ruled out.

When examining other alterations of the PI3K/AKT pathway, we found that mutations activating mTORC1 genes might contribute to resistance to AZD5363. In the phase I study of AZD5363 as single agent in AKT1-mutant tumors, the presence of concurrent PIK3CA or MTOR alterations was associated with improved PFS compared with patients without these alterations (11). Of note, concurrent AKT1 and MTOR mutations were found in two patients with endometrial cancer. This suggests that the impact of AKT1-downstream mutations in the response to AZD5363 may be tumor-specific and underscores that MTOR mutation functional analysis is needed to interpret the impact of these variants. The planned analysis of plasma samples collected at baseline and disease progression in the phase 1b/III studies of capivasertib in breast cancer will enable a more thorough assessment of genetic mechanisms of resistance in larger cohorts of patients, and will establish whether MTOR and/or TSC1 mutations are bona fide mechanisms of resistance in clinical samples.

The genetic composite biomarker based on mutations in PIK3CA/AKT1 and absence of mTORC1-activating alterations suggests that AKT-pathway “oncogene addiction” is multifactorial. In an attempt to identify a single-plex response biomarker, a proteomic analysis of PI3K/AKT “pathway dependency” was undertaken, showing that low baseline phosphorylation of AKT S473 efficiently discriminates AZD5363 resistance. This is in agreement with another study describing an association between high pAKT S473 and sensitivity of established cell lines to the AKT inhibitor ipatasertib (GDC-0068; ref. 38). Therefore, assessment of pAKT S473 levels or a composite AKT activation signature, could be used as biomarker of AZD5363-resistance and warrants further corroboration in the clinical setting.

Differences in AZD5363 responses were consistent with the reduction of Ki67-positive cells, efficiently distinguishing AZD5363 monotherapy resistant and sensitive models and suggesting antiproliferative activity of single AZD5363 treatment as its main mechanism of action. Ki67 reduction after neoadjuvant chemotherapy is an independent prognostic factor in different subtypes of BC, including TNBC and Luminal B tumors (39). Our data suggest that a reduction in Ki67 >25.4% is associated with the likelihood of response to AZD5363 as single agent (Fig. 3A). Mechanistically, we have shown profound S6/cyclin D1-axis downmodulation and consequent inhibition of proliferation (Ki67) associated with AZD5363 response in PDX models. Changes in FOXO1/3a phosphorylation did not distinguish sensitive from resistant PDXs. These findings contrast with the data reported in the STAKT trial, a window-of-opportunity study testing capivasertib as single agent in patients with newly diagnosed ER+ invasive breast cancer. In this trial, an increase in nuclear FOXO3a was observed in the tumors with uppermost Ki67 response (40). The reason for this discrepancy could be based on the endpoint analysis of the biomarker and/or of the clinical/preclinical response.

We describe two different mechanisms of acquired resistance to AZD5363 that are consistent with the proposed mode of action. First, we have shown that deregulation of cyclin D1 downstream of AKT is associated with AZD5363 resistance. Cyclin D1 plays a key role in the initiation of cell cycle upon mitogenic signal, and both AKT and S6K posttranscriptionally regulate cyclin D1 (4, 30). We observed a higher frequency of cyclin D1-expressing (but not overexpressing) models in AZD5363-sensitive tumors than in resistant tumors, likely evidencing an active PI3K/AKT/cyclin D1 signaling pathway in sensitive tumors. However, overexpression of cyclin D1 could overcome AKT/S6K regulation and promote progression through the G1-restriction point, counteracting AZD5363's antitumoral activity. This is relevant because CCND1 amplification is found in 19%–35% of primary and metastatic BC (6, 41, 42). Similarly, Liu and colleagues have recently shown that sensitivity to AKTi is compromised by CDK6 overexpression, a partner of cyclin D1 at the G1-restriction point (43). Hence, taken together, both observations suggest that deregulation of the CDK4/6–cyclin D1 complex results in AZD5363 resistance. This warrants the validation of the putative clinical benefit of AKT and CDK4/6 inhibitors combinations in the clinical setting. Second, we have shown that loss of the AKT1 p.E17K subclone results in resistance to AZD5363. This is similar to the loss of HER2 or ER as a mechanism of resistance to trastuzumab or endocrine therapy, respectively (44, 45). This observation is consistent with one out of seventeen cases reported by Hyman and colleagues that progressed to AZD5363 without recovering high frequency of the AKT1 p.E17K mutation in ctDNA (11). Importantly, progression to AZD5363 based on intratumor heterogeneity of the AKT1 p.E17K mutation and subclonal selection due to treatment pressure will not be trackable by ctDNA analysis, as there will be no recovery of AKT1 p.E17K in the blood test.

Finally, our experiments on the combination of AZD5363 and paclitaxel revealed that p53 may play a role in the therapeutic efficacy of this combination, presumably due to the ability of p53 to halt cell cycle upon drug-induced stress. However, these preclinical results need to be interpreted with caution as cell line experiments in 2D culture could not evidence a direct link between the TP53 status and treatment response and they also lack clinical validation, as the proportion of TP53 mutant patients in the BEECH trial was too small.

This study shows that both ER+ and TNBC models benefit from AZD5363 monotherapy if tumors are positive for the refined genetic composite biomarker (comprising PIK3CA/AKT1-mutant and MTOR/TSC1/2-wt) and that low pAKT S473 identifies AZD5363-resistant tumors. Interestingly, in the recently reported FAKTION trial (13), the addition of capivasertib to fulvestrant for patients with endocrine resistant advanced breast cancer resulted in significantly longer PFS compared with fulvestrant alone (HR, 0.57; 95% CI, 0.39–0.84; one-sided P = 0.0017; two-sided 0.0035; median PFS, 10.3 months for capivasertib compared with 4.8 months for placebo). These results suggest that the lack of ER blockade may have contributed to the inability of capivasertib to improve outcomes over paclitaxel alone in the ER+ breast cancer population in the BEECH trial, highlighting the complex crosstalk of ER and PI3K/AKT signaling (46–48).

In conclusion, this translational study provides insight into putative predictive biomarkers of response and resistance to AZD5363 in HER2-negative metastatic breast cancer, either as single agent or in combination with paclitaxel. These findings are relevant to inform the clinical development of the combination.

G. Villacampa reports receiving speakers bureau honoraria from MSD. A. Barnicle is an employee/paid consultant for AstraZeneca. E.C. de Bruin, A. Reddy, G. Schiavon, and B.R. Davies are employees/paid consultants for and hold ownership interest (including patents) in AstraZeneca. J. Arribas is an employee/paid consultant for Menarini Biotech. G.B. Mills is an employee/paid consultant for AstraZeneca, Chrysalis, Lilly, PDX Pharma, Signalchem Lifesciences, Symphogen, and Tarveda; reports receiving other commercial research support from Myriad Genetics and DSP with Nanostring; holds ownership interest (including patents) in Catena, Spindletop Ventures, ImmunoMET, Signal Chem, and Tarveda; and reports receiving other remuneration from Ionis and ImmunoMET. C. Caldas is an employee/paid consultant for AstraZeneca and Illumina; and reports receiving commercial research grants from AstraZeneca, Servier, Genentech, and Roche. R. Dienstamnn reports receiving speakers bureau honoraria from Roche, Servier, Ipsen, Amgen, Sanofi, and Merck Sharp Dohme. A. Prat is an advisory board member/unpaid consultant for Nanostring Technologies, Roche, Novartis, Bristol-Myers Squibb, Pfizer, AztraZeneca, Oncolytics Biotech, and Amgen. P. Razavi is an employee/paid consultant for Novartis, AstraZeneca, and Foundation Medicine; and reports receiving commercial research grants from Grail Inc. M. Scaltriti reports receiving commercial research grants from Menarini Ricerche, AstraZeneca, Daiichi-Sankio, Immunomedics, Puma Biotechnology, and Targimmune; reports receiving speakers bureau honoraria from AstraZeneca, Daiichi-Sankio, and Menarini Ricerche; and holds ownership interest (including patents) in medendi.org. N.C. Turner is an employee/paid consultant for AstraZeneca, Bristol-Myers Squibb, Lilly, Merck Sharpe and Dohme, Novartis, Pfizer, Roche/Genentech, Bicycle Therapeutics, Taiho, Zeno Pharmaceuticals, and Repare Therapeutics. C. Saura reports receiving speakers bureau honoraria from AstraZeneca, Bristol-Myers Squibb, Celgene, Daiichi Sankyo, Eisai, Roche, Genomic Health, MSD, Novartis, Pfizer, Philips Healthwork, Pierre Fabre, Puma, Synthon, and Sanofi. No potential conflicts of interest were disclosed by other authors.

Conception and design: A. Gris-Oliver, C. Saura, B.R. Davies, M. Oliveira, V. Serra

Development of methodology: A. Gris-Oliver, M. Sánchez-Guixé, Y.H. Ibrahim, R. Dienstmann, V. Serra

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Gris-Oliver, M. Palafox, L. Monserrat, A.Òdena, M. Sánchez-Guixé, Y.H. Ibrahim, J. Grueso, M. Parés, M. Guzman, O. Rodriguez, A. Bruna, C.S. Hirst, E.C. de Bruin, J. Arribas, G.B. Mills, C. Caldas, A. Prat, P. Nuciforo, P. Razavi, C. Saura, M. Oliveira, V. Serra

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Gris-Oliver, M. Palafox, F. Brasó-Maristany, M. Sánchez-Guixé, Y.H. Ibrahim, G. Villacampa, C.S. Hirst, A. Barnicle, E.C. de Bruin, A. Reddy, G. Schiavon, G.B. Mills, C. Caldas, R. Dienstmann, A. Prat, P. Nuciforo, P. Razavi, M. Scaltriti, N.C. Turner, C. Saura, M. Oliveira, V. Serra

Writing, review, and/or revision of the manuscript: A. Gris-Oliver, M. Palafox, L. Monserrat, F. Brasó-Maristany, A.Òdena, E.C. de Bruin, A. Reddy, G. Schiavon, G.B. Mills, C. Caldas, R. Dienstmann, A. Prat, P. Razavi, M. Scaltriti, N.C. Turner, C. Saura, B.R. Davies, M. Oliveira, V. Serra

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Gris-Oliver, M. Sánchez-Guixé, J. Grueso, M. Parés, M. Guzman, O. Rodriguez, C. Caldas, V. Serra

Study supervision: A. Gris-Oliver, M. Oliveira, V. Serra

We acknowledge AstraZeneca, the GHD-Pink program, the FERO Foundation, and the Orozco Family for supporting this study (to V. Serra). This study has also been supported by the Catalan Agency AGAUR (2017 SGR 540, to V. Serra). V. Serra is supported by the Miguel Servet Program (ISCIII, CP14/00228 and CPII19/00033). A. Gris-Oliver was awarded with a fellowship from the Agència de Gestió d'Ajuts Universitaris i de Recerca (AGAUR, 2015 FI_B 01075), M. Sánchez-Guixé, with a Marie Slodowska-Curie Innovative Training Networks PhD fellowship (H2020-MSCA-ITN-2015_675392), M. Palafox with a Juan de la Cierva Grant from the Ministerio de Economía y Competitividad (FJCI-2015-25412), L. Monserrat with a grant from FI-AGAUR (2019 FI_B 01199) and F. Brasó-Maristany with a grant from the Fundación Científica Asociación Española Contra el Cáncer (AECC_Postdoctoral17-1062). The xenograft program in the C. Caldas laboratory is supported by Cancer Research UK and also received funding from an EU H2020 Network of Excellence (EuroCAN). The RPPA facility is funded by NCI #CA16672 (to G.B. Mills). This work has been supported by NIH grants P30 CA008748 and RO1CA190642-01, the CDMRP grant BC171535P1, and the Breast Cancer Research Foundation (to M. Scaltriti).

The authors are grateful to all the patients who kindly consented to the use of their tumors to develop this study and to all the personnel involved in sample collection from the Breast Surgical Unit, Breast Cancer Center, Department of Radiology Vall d'Hebron University Hospital and Molecular Oncology Group at Vall d'Hebron Institute of Oncology (VHIO). Capivasertib (AZD5363) was discovered by AstraZeneca after a collaboration with Astex Therapeutics (and its collaboration with The Institute of Cancer Research and Cancer Research Technology Limited). AZD5363 was kindly provided by Barry Davies, Natascha Bezdenejnih-Snyder and Christopher Morrow (AstraZeneca). The authors thank the Cellex Foundation for providing research facilities and equipment, the Cancer Genomics Group at VHIO (Ana Vivancos) and Joanne Soong and the Center for Molecular Oncology at the MSKCC for providing technical and analytical support with patient and PDX sequencing. We thank also Chris Vellano and the Functional Proteomics RPPA Core facility at MD Anderson Cancer Center for RPPA data of the panel of PDXs used in this article. The authors are grateful to Myria Nikolaou, Jaime Rodríguez Canales and Martine Roudier from AstraZeneca for the analysis of pAKT immunohistochemistry staining. We also acknowledge Claire Rooney and Andrew Foxley (AstraZeneca) for helpful discussion of this manuscript.

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.

1.
Wong
KK
,
Engelman
JA
,
Cantley
LC
. 
Targeting the PI3K signaling pathway in cancer
.
Curr Opin Genet Dev
2010
;
20
:
87
90
.
2.
Cheung
LW
,
Hennessy
BT
,
Li
J
,
Yu
S
,
Myers
AP
,
Djordjevic
B
, et al
High frequency of PIK3R1 and PIK3R2 mutations in endometrial cancer elucidates a novel mechanism for regulation of PTEN protein stability
.
Cancer Discov
2011
;
1
:
170
85
.
3.
Cantley
LC
. 
The phosphoinositide 3-kinase pathway
.
Science
2002
;
296
:
1655
7
.
4.
Manning
BD
,
Cantley
LC
. 
AKT/PKB signaling: navigating downstream
.
Cell
2007
;
129
:
1261
74
.
5.
Bareche
Y
,
Venet
D
,
Ignatiadis
M
,
Aftimos
P
,
Piccart
M
,
Rothe
F
, et al
Unravelling triple-negative breast cancer molecular heterogeneity using an integrative multiomic analysis
.
Ann Oncol
2018
;
29
:
895
902
.
6.
Razavi
P
,
Chang
MT
,
Xu
G
,
Bandlamudi
C
,
Ross
DS
,
Vasan
N
, et al
The genomic landscape of endocrine-resistant advanced breast cancers
.
Cancer Cell
2018
;
34
:
427
38
e6.
7.
Rodon
J
,
Dienstmann
R
,
Serra
V
,
Tabernero
J
. 
Development of PI3K inhibitors: lessons learned from early clinical trials
.
Nat Rev Clin Oncol
2013
;
10
:
143
53
.
8.
Davies
BR
,
Greenwood
H
,
Dudley
P
,
Crafter
C
,
Yu
DH
,
Zhang
J
, et al
Preclinical pharmacology of AZD5363, an inhibitor of AKT: pharmacodynamics, antitumor activity, and correlation of monotherapy activity with genetic background
.
Mol Cancer Ther
2012
;
11
:
873
87
.
9.
Turner
NC
,
Alarcon
E
,
Armstrong
AC
,
Philco
M
,
Lopez Chuken
YA
,
Sablin
MP
, et al
BEECH: a dose-finding run-in followed by a randomised phase II study assessing the efficacy of AKT inhibitor capivasertib (AZD5363) combined with paclitaxel in patients with estrogen receptor-positive advanced or metastatic breast cancer, and in a PIK3CA mutant sub-population
.
Ann Oncol
2019
;
30
:
774
80
.
10.
Schmid
P
,
Abraham
J
,
Chan
S
,
Wheatley
D
,
Brunt
AM
,
Nemsadze
G
, et al
Capivasertib plus paclitaxel versus placebo plus paclitaxel as first-line therapy for metastatic triple-negative breast cancer: the PAKT trial
.
J Clin Oncol
. 
2020
;
38
:
423
33
.
11.
Hyman
DM
,
Smyth
LM
,
Donoghue
MTA
,
Westin
SN
,
Bedard
PL
,
Dean
EJ
, et al
AKT inhibition in solid tumors with AKT1 mutations
.
J Clin Oncol
. 
2017
;
35
:
2251
9
.
12.
Banerji
U
,
Dean
EJ
,
Perez-Fidalgo
JA
,
Batist
G
,
Bedard
PL
,
You
B
, et al
A phase I open-label study to identify a dosing regimen of the pan-AKT inhibitor AZD5363 for evaluation in solid tumors and in PIK3CA-mutated breast and gynecologic cancers
.
Clin Cancer Res
2018
;
24
:
2050
9
.
13.
Jones
RH
,
Casbard
A
,
Carucci
M
,
Cox
C
,
Butler
R
,
Alchami
F
, et al
Fulvestrant plus capivasertib versus placebo after relapse or progression on an aromatase inhibitor in metastatic, oestrogen receptor-positive breast cancer (FAKTION): a multicentre, randomised, controlled, phase 2 trial
.
Lancet Oncol
2020
;
21
::
345
57
.
14.
Bruna
A
,
Rueda
OM
,
Greenwood
W
,
Batra
AS
,
Callari
M
,
Batra
RN
, et al
A biobank of breast cancer explants with preserved intra-tumor heterogeneity to screen anticancer compounds
.
Cell
2016
;
167
:
260
74.e22
.
15.
Tentler
JJ
,
Tan
AC
,
Weekes
CD
,
Jimeno
A
,
Leong
S
,
Pitts
TM
, et al
Patient-derived tumour xenografts as models for oncology drug development
.
Nat Rev Clin Oncol
2012
;
9
:
338
50
.
16.
Gao
H
,
Korn
JM
,
Ferretti
S
,
Monahan
JE
,
Wang
Y
,
Singh
M
, et al
High-throughput screening using patient-derived tumor xenografts to predict clinical trial drug response
.
Nat Med
2015
;
21
:
1318
25
.
17.
Therasse
P
,
Arbuck
SG
,
Eisenhauer
EA
,
Wanders
J
,
Kaplan
RS
,
Rubinstein
L
, et al
New guidelines to evaluate the response to treatment in solid tumors. European organization for research and treatment of cancer, national cancer institute of the united states, national cancer institute of canada
.
J Natl Cancer Inst
2000
;
92
:
205
16
.
18.
Kim
SB
,
Dent
R
,
Im
SA
,
Espie
M
,
Blau
S
,
Tan
AR
, et al
Ipatasertib plus paclitaxel versus placebo plus paclitaxel as first-line therapy for metastatic triple-negative breast cancer (LOTUS): a multicentre, randomised, double-blind, placebo-controlled, phase 2 trial
.
Lancet Oncol
2017
;
18
:
1360
72
.
19.
Oliveira
M
,
Saura
C
,
Nuciforo
P
,
Calvo
I
,
Andersen
J
,
Passos-Coelho
JL
, et al
FAIRLANE, a double-blind placebo-controlled randomized phase II trial of neoadjuvant ipatasertib plus paclitaxel for early triple-negative breast cancer
.
Ann Oncol
2019
;
30
:
1289
97
.
20.
Prat
A
,
Parker
JS
,
Fan
C
,
Perou
CM
. 
PAM50 assay and the three-gene model for identifying the major and clinically relevant molecular subtypes of breast cancer
.
Breast Cancer Res Treat
2012
;
135
:
301
6
.
21.
Parker
JS
,
Mullins
M
,
Cheang
MC
,
Leung
S
,
Voduc
D
,
Vickery
T
, et al
Supervised risk predictor of breast cancer based on intrinsic subtypes
.
J Clin Oncol
2009
;
27
:
1160
7
.
22.
Adzhubei
IA
,
Schmidt
S
,
Peshkin
L
,
Ramensky
VE
,
Gerasimova
A
,
Bork
P
, et al
A method and server for predicting damaging missense mutations
.
Nat Methods
2010
;
7
:
248
9
.
23.
O'Reilly
KE
,
Rojo
F
,
She
QB
,
Solit
D
,
Mills
GB
,
Smith
D
, et al
mTOR inhibition induces upstream receptor tyrosine kinase signaling and activates Akt
.
Cancer Res
2006
;
66
:
1500
8
.
24.
Elkabets
M
,
Vora
S
,
Juric
D
,
Morse
N
,
Mino-Kenudson
M
,
Muranen
T
, et al
mTORC1 inhibition is required for sensitivity to PI3K p110alpha inhibitors in PIK3CA-mutant breast cancer
.
Sci Transl Med
2013
;
5
:
196ra99
.
25.
Serra
V
,
Eichhorn
PJ
,
Garcia-Garcia
C
,
Ibrahim
YH
,
Prudkin
L
,
Sanchez
G
, et al
RSK3/4 mediate resistance to PI3K pathway inhibitors in breast cancer
.
J Clin Invest
2013
;
123
:
2551
63
.
26.
Roux
PP
,
Shahbazian
D
,
Vu
H
,
Holz
MK
,
Cohen
MS
,
Taunton
J
, et al
RAS/ERK signaling promotes site-specific ribosomal protein S6 phosphorylation via RSK and stimulates cap-dependent translation
.
J Biol Chem
2007
;
282
:
14056
64
.
27.
Biever
A
,
Valjent
E
,
Puighermanal
E
. 
Ribosomal protein S6 phosphorylation in the nervous system: from regulation to function
.
Front Mol Neurosci
2015
;
8
:
75
.
28.
Castel
P
,
Ellis
H
,
Bago
R
,
Toska
E
,
Razavi
P
,
Carmona
FJ
, et al
PDK1-SGK1 signaling sustains AKT-independent mTORC1 activation and confers resistance to PI3Kalpha inhibition
.
Cancer Cell
2016
;
30
:
229
42
.
29.
Sommer
EM
,
Dry
H
,
Cross
D
,
Guichard
S
,
Davies
BR
,
Alessi
DR
. 
Elevated SGK1 predicts resistance of breast cancer cells to Akt inhibitors
.
Biochem J
2013
;
452
:
499
508
.
30.
Ma
XM
,
Blenis
J
. 
Molecular mechanisms of mTOR-mediated translational control
.
Nat Rev Mol Cell Biol
2009
;
10
:
307
18
.
31.
Innocente
SA
,
Abrahamson
JL
,
Cogswell
JP
,
Lee
JM
. 
p53 regulates a G2 checkpoint through cyclin B1
.
Proc Natl Acad Sci U S A
1999
;
96
:
2147
52
.
32.
Troester
MA
,
Herschkowitz
JI
,
Oh
DS
,
He
X
,
Hoadley
KA
,
Barbier
CS
, et al
Gene expression patterns associated with p53 status in breast cancer
.
BMC Cancer
2006
;
6
:
276
.
33.
Coutant
C
,
Rouzier
R
,
Qi
Y
,
Lehmann-Che
J
,
Bianchini
G
,
Iwamoto
T
, et al
Distinct p53 gene signatures are needed to predict prognosis and response to chemotherapy in ER-positive and ER-negative breast cancers
.
Clin Cancer Res
2011
;
17
:
2591
601
.
34.
Tenbaum
SP
,
Ordonez-Moran
P
,
Puig
I
,
Chicote
I
,
Arques
O
,
Landolfi
S
, et al
beta-catenin confers resistance to PI3K and AKT inhibitors and subverts FOXO3a to promote metastasis in colon cancer
.
Nat Med
2012
;
18
:
892
901
.
35.
Katsha
A
,
Belkhiri
A
,
Goff
L
,
El-Rifai
W
. 
Aurora kinase A in gastrointestinal cancers: time to target
.
Mol Cancer
2015
;
14
:
106
.
36.
Sasai
K
,
Treekitkarnmongkol
W
,
Kai
K
,
Katayama
H
,
Sen
S
. 
Functional significance of aurora kinases-p53 protein family interactions in cancer
.
Front Oncol
2016
;
6
:
247
.
37.
Tan
MH
,
De
S
,
Bebek
G
,
Orloff
MS
,
Wesolowski
R
,
Downs-Kelly
E
, et al
Specific kinesin expression profiles associated with taxane resistance in basal-like breast cancer
.
Breast Cancer Res Treat
2012
;
131
:
849
58
.
38.
Lin
J
,
Sampath
D
,
Nannini
MA
,
Lee
BB
,
Degtyarev
M
,
Oeh
J
, et al
Targeting activated Akt with GDC-0068, a novel selective Akt inhibitor that is efficacious in multiple tumor models
.
Clin Cancer Res
2013
;
19
:
1760
72
.
39.
von Minckwitz
G
,
Schmitt
WD
,
Loibl
S
,
Muller
BM
,
Blohmer
JU
,
Sinn
BV
, et al
Ki67 measured after neoadjuvant chemotherapy for primary breast cancer
.
Clin Cancer Res
2013
;
19
:
4521
31
.
40.
Robertson
JFR
,
Coleman
RE
,
Cheung
KL
,
Evans
A
,
Holcombe
C
,
Skene
A
, et al
Proliferation and AKT activity biomarker analyses after capivasertib (AZD5363) treatment of patients with ER(+) invasive breast cancer (STAKT)
.
Clin Cancer Res
2019
;
26
:
1574
85
.
41.
Gao
J
,
Aksoy
BA
,
Dogrusoz
U
,
Dresdner
G
,
Gross
B
,
Sumer
SO
, et al
Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal
.
Sci Signal
2013
;
6
:
pl1
.
42.
Cerami
E
,
Gao
J
,
Dogrusoz
U
,
Gross
BE
,
Sumer
SO
,
Aksoy
BA
, et al
The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data
.
Cancer Discov
2012
;
2
:
401
4
.
43.
Liu
J
,
Duan
Z
,
Guo
W
,
Zeng
L
,
Wu
Y
,
Chen
Y
, et al
Targeting the BRD4/FOXO3a/CDK6 axis sensitizes AKT inhibition in luminal breast cancer
.
Nat Commun
2018
;
9
:
5200
.
44.
Mittendorf
EA
,
Wu
Y
,
Scaltriti
M
,
Meric-Bernstam
F
,
Hunt
KK
,
Dawood
S
, et al
Loss of HER2 amplification following trastuzumab-based neoadjuvant systemic therapy and survival outcomes
.
Clin Cancer Res
2009
;
15
:
7381
8
.
45.
Osborne
CK
,
Schiff
R
. 
Mechanisms of endocrine resistance in breast cancer
.
Annu Rev Med
2011
;
62
:
233
47
.
46.
Bosch
A
,
Li
Z
,
Bergamaschi
A
,
Ellis
H
,
Toska
E
,
Prat
A
, et al
PI3K inhibition results in enhanced estrogen receptor function and dependence in hormone receptor-positive breast cancer
.
Sci Transl Med
2015
;
7
:
283ra51
.
47.
Toska
E
,
Osmanbeyoglu
HU
,
Castel
P
,
Chan
C
,
Hendrickson
RC
,
Elkabets
M
, et al
PI3K pathway regulates ER-dependent transcription in breast cancer through the epigenetic regulator KMT2D
.
Science
2017
;
355
:
1324
30
.
48.
Ribas
R
,
Pancholi
S
,
Guest
SK
,
Marangoni
E
,
Gao
Q
,
Thuleau
A
, et al
AKT antagonist AZD5363 influences estrogen receptor function in endocrine-resistant breast cancer and synergizes with fulvestrant (ICI182780) in vivo
.
Mol Cancer Ther
2015
;
14
:
2035
48
.