Combination of PI3K and MEK inhibitors yields durable remission in PDX models of PIK3CA-mutated metaplastic breast cancers

Samples from 323 unilateral invasive triple-negative primary breast tumors excised from women managed at Institut Curie (Paris and Saint-Cloud, France) between 1980 and 2015 were analyzed (Additional file 1: Table S1). Most patients (67%) were diagnosed and treated after 2000. All patients admitted to our institution before 2007 were informed that their tumor samples might be used for scientific purposes and were given the opportunity to refuse such use. Since 2007, patients admitted to our institution also give express consent for the use of their samples for research purposes, by signing an informed consent form. Patients (mean age, 56 years; range, 28–91 years) met the following criteria: primary unilateral non-metastatic TNBC, with full clinical, histological, and biological data and full follow-up at Institut Curie. Median follow-up was 7.8 years (range 8 months to 36 years). Seventy-eight patients had developed metastases within 10 years.

PDXs were established from the engraftment of primary breast tumors with a procedure described elsewhere [16‐18]. Female Swiss nude mice were purchased from Charles River Laboratories and maintained under specific pathogen-free conditions. The experimental protocol and animal housing were in accordance with institutional guidelines and with the recommendations of the French Ethics Committee (Agreement B75-05-18, France). Three metaplastic TNBC PDXs with genomic alterations were chosen for experimental analysis: HBCx-60, HBCx-165, and HBCx-178. BYL-719 (PI3K inhibitor) and selumetinib (MEK inhibitor) were purchased from Medchem Express. BYL-719 was administered five times per week, at doses of 35 mg/kg, by oral gavage. Selumetinib (MEK inhibitor) was administered five times per week, at doses of 100 mg/kg (50 mg/kg, bid), by oral gavage. Adriamycin (DOX, doxorubicin, Teva Pharmaceuticals) and cyclophosphamide (Endoxan, Baxter) were administered by the intraperitoneal (i.p.) route, at doses of 2 and 100 mg/kg, respectively, every 3 weeks. We included 10 mice per groups. Tumor growth was evaluated by measuring two perpendicular tumor diameters with calipers twice weekly. Individual tumor volumes were calculated as follows: V = (a × b)²/2, where “a” is the largest diameter, and “b” is the smallest diameter. For each tumor, volumes are expressed relative to the initial volume, as a relative tumor volume (RTV). Tumor growth inhibition (TGI) was assessed by dividing mean RTV (relative tumor volume) in the treated group by mean RTV in the control group at the same time. The statistical significance of TGI was assessed by comparing tumor volumes between the treated and control groups in paired Student’s t tests. Stable disease was defined as the percentage change in volume, between 0 and − 50.

We used gene expression arrays for the transcriptomic profiling of 64 PDX TNBCs. The concentration and integrity/purity of each RNA sample were assessed with the RNA 6000 LabChip kit (Agilent) and an Agilent 2100 bioanalyzer. GeneChip Human 1.1 ST arrays were hybridized according to Affymetrix recommendations, with the WT Expression Kit protocol (Life Technologies) and Affymetrix labeling and hybridization kits. Arrays were normalized according to the RMA normalization procedure, with the oligo package [19]. No additional human-mouse cross-hybridization filtering was applied, as our xenograft samples contained less than 5% mouse cells (percentage determined by RT-PCR quantifying transcripts of the ubiquitously expressed TBP gene with specific mouse and human primers pairs), a percentage too low to affect the expression profiles obtained with HuGene1.0 arrays [20]. The TNBC molecular subtypes of the PDXs were determined from gene expression data, with the TNBCtype software developed by Chen et al [21].

We analyzed 64 PDXs by the targeted NGS of 95 genes, chosen from the genes most frequently mutated in breast cancer (> 1%) and including potential therapeutic targets, as previously described [22]. Briefly, NGS was performed on an Illumina HiSeq 2500 sequencer, and the genomic variants were annotated with the COSMIC and 1000 genome databases. Variants with a low allelic frequency (< 5%) or low coverage (< 100x) were excluded from the analysis. Deleterious genomic alterations were defined as follows: (i) for oncogenes, only gain-of-function mutations were considered (i.e., hotspot missense mutations, in-frame insertions/deletions/splicing reported to be oncogenic), and (ii) for tumor suppressor genes (TSG), only loss-of-function mutations were considered (i.e., biallelic truncating alterations (nonsense mutations, frameshift insertions/deletions/splicing) or monoallelic truncating alterations associated with heterozygous deletion detected by copy number analysis). Genomic variants were biologically validated by comparison with the COSMIC, TumorPortal, and cBioportal databases [7, 15].

PDXs were profiled with Affymetrix genomics arrays: 24 with SNP 6.0 and 37 with the Cytoscan HD array. Genome-wide copy number analysis was performed with Affymetrix SNP arrays, as previously described [23, 24]. SNP 6.0 or Cytoscan HD arrays were used with 500 ng and 250 ng of gDNA, respectively, as the input material, as recommended by the manufacturer. Raw data were normalized with Genotyping console (SNP6.0 arrays) or Chromosome Analysis Suite (Cytoscan HD arrays). The focal amplification of oncogenes was defined as a log ratio > 1.58 (6 copies per diploid genome) and a maximum size of < 10 Mb. The biallelic inactivation of TSGs was defined as homozygous deletion or truncating mutations associated with heterozygous deletion. Copy number alterations were compared with cBioPortal data for TCGA breast cancer [25, 26]. All PDX copy numbers are represented by the Circular Binary Segmentation algorithm [27], as implemented in the DNAcopy package for R, with a minimum width of 3, an alpha risk of 1%, and up to 10,000 permutations. Downstream analysis of the sample population was performed with GISTIC2.0 [28], with default settings. CGH explorer and CGHcall were used for visual representation of the results and figures [29]. The BRCAness signature consisted of large-scale state transitions (LST), defined as chromosomal breaks between adjacent regions of at least 10 Mb initially described by Popova et al. with Gap methodology [30].

Total RNA extraction and RT-qPCR have been described elsewhere [31]. The TBP gene (GenBank accession no. NM_003194) encoding the TATA box-binding protein (a component of the DNA-binding protein complex TFIID) was quantified as an endogenous RNA control, and each sample was normalized on the basis of its TBP content [28]. N-fold differences in target gene expression relative to the TBP gene (“Ntarget”), were determined as Ntarget = 2^ΔCtsample, where the ΔCt value of the sample was obtained by subtracting the mean Ct value of the target gene from that of the TBP gene [32].

For the gene expression study in PDXs, mRNA levels were normalized to obtain a “basal mRNA level” (smallest amount of mRNA quantifiable (Ct = 35)) equal to 1. We analyzed the expression of genes involved in epithelial-mesenchymal transition (EMT) (SNAI2, VIM, ACTA2, SPARC, TCF7L2, CAV1) in all TNBC PDXs and included 49 with well-known transcriptomic classifications in the final analysis.

RPPA was performed as previously described [33] for 48 of the 64 TNBC PDXs (16 recent PDXs were not included in the RPPA analysis). We assessed pathway activation, by calculating a PI3K-AKT-mTOR and RTK-MAPK pathway score with normalized data. This score was obtained by calculating the sum of the protein levels for positive components and subtracting that for the negative components of the pathway (especially for PI3K-AKT-mTOR pathway): (i) PI3K p110 subunit β, P-AKT1 (Ser473), P-AKT1 (Thr308), P-4E-BP1, P-p70 S6 kinase, P-S6 ribosomal protein minus PTEN (Cell Signaling Technology®), and (ii) P-RSK2 (Cell Signaling Technology®).

Tumors from patients and xenografts were fixed in 10% neutral-buffered formalin, embedded in paraffin, and stained with hematoxylin and eosin. The same histologist (ML) compared morphology between patient and PDX tumors. Four representative MBCs with chondroid and squamous characteristics were compared (HBCx-130, HBCx-162, HBCx-165, HBCx-178).

Proteins were extracted from tumors using a Laemmli buffer (50 mM Tris HCL pH 8, 2 mM DTT, 2% SDS, 5% glycerol), supplemented with protease and phosphatase inhibitors. Lysates were resolved on 10% agarose gels, transferred onto nitrocellulose membranes (Bio-Rad, Hercules, CA, USA), and immunoblotted with rabbit antibodies against AKT, p-AKT (ser 473), S6, p-S6 (Ser235/236), ERK, p-ERK (Thr202/Tyr204), MEK, p-MEK (Ser217/221), and GAPDH (Cell Signaling®). After washes, membranes were incubated with the appropriate horseradish peroxidase-conjugated affinity-purified goat anti-rabbit secondary antibodies (Jackson ImmunoResearch Laboratories, Inc., Interchim).

Metastasis-free survival (MFS) was assessed by determining the interval between diagnosis and the detection of the first distant metastasis. Overall survival (OS) was determined as the interval between diagnosis and death. Survival was estimated by the Kaplan-Meier method.

We compared survival between triple-negative MBC and other histological subtypes in a clinical cohort of 323 TNBC patients treated at our institute, with a long follow-up. This cohort included 13 MBCs (4%), 43 apocrine BCs (13.3%), 36 medullary BCs (11.1%), 198 breast cancers of no special type (NST) (61.4%), and other rare forms of TNBC (10.2%) (oncocytic, acinic, lobular, adenoid cystic papillary, micropapillary, or mucinous). MBC patients had a worse prognosis than patients with other subtypes, in terms of metastasis-free survival (MFS) and overall survival (OS) (p = 0.02 and p = 0.01, respectively) (Fig. 1). The medullary TNBC subgroup had a particularly good prognosis.

We investigated the frequency of tumors with PIK3CA mutations with Sanger sequencing of PIK3CA. In the whole cohort of tumors, PIK3CA mutations were present in 15% of metaplastic tumors and 12.2% of the tumors of other subtypes (p = 0.66) (NST, 13%; apocrine, 14%; rare, 12%; medullary, 0%). PIK3CA mutations were of no prognostic values in the metaplastic subtype (data not shown).

In our cohort of 64TNBC PDX models (the description of the first 61 PDX were recently published [22]), nine displayed metaplastic differentiation (14%) (Table 1). Three displayed fusiform differentiation, three had squamous characteristics, and three had mixed components (chondroid–fusiform or squamous–fusiform). Figure 2a shows the morphological features of four MBCs (patients and PDXs): HBCx-162, HBCx-130, HBCx-165, and HBCx-178. HBCx-162 was characterized by an abundant chondromyxoid matrix, and HBCx-130, HBCx-165, and HBCx-178 were squamous cell MBCs. The specific morphological characteristics of the initial tumors were also present in the corresponding PDXs. Squamous cell metaplastic carcinomas have a particular set of characteristics: tumor masses bordering cystic cavities, high degrees of nuclear pleomorphism, large polygonal cells with abundant eosinophilic cytoplasm, keratinization beads sometimes present, and abundant inflammatory infiltrate. These highly specific aspects are found in both PDXs and patient tumors. A similar phenomenon is observed for metaplastic carcinomas with mesenchymal chondroid differentiation: each pair of tumors displays an abundant cartilaginous and myxoid matrix enclosing the carcinomatous cells.

Nine MBC PDXs were characterized at the transcriptomic level: four were classified as M (mesenchymal) or MSL (mesenchymal stem-like), 1 as BL2 (basal-like 2), and 3 were unstable. For further characterization of the phenotype of MBC PDXs, we performed a RT-PCR analysis of various genes involved in the epithelial-mesenchymal transition (EMT), a cancer cell phenotype previously associated with MBC [34], in 48 TNBC PDXs. We generated an expression heatmap for the SNAI2, ACTA2, VIM, SPARC, and TCF7L2 genes in the various TNBC transcriptomic subtypes (14 unstable and 5 NA tumors were excluded from the analysis) (Fig. 2b). The expression of the SNAI2, ACTA2, VIM, SPARC, and TCF7L2 genes was significantly stronger in the MSL/M subtypes than in the other subtypes (unpaired t test).

The genomic alterations (mutations and copy number alterations) found in MBC PDXs concerned the PI3K and RTK-MAPK pathways.

Figure 3a shows the major genomic alterations in metaplastic PDX. Eight genetic alterations affecting the PI3K-AKT-mTOR signaling pathway were identified in six of the nine metaplastic PDX, including four PIK3CA mutations (44.4%), three PTEN genetic alterations (33.3%), and one AKT1 amplification (11.1%). PIK3CA (activating mutations) was significantly more frequently altered in metaplastic TNCB than in other subtypes (Fisher’s test, p = 0.01) (Fig. 3b). Five of the nine metaplastic PDX (55%) harbored one alteration to the TP53 gene (50% for other subtypes). The additional genetic alterations of theranostic interest were two alterations to the BRAF gene (one mutation and one focal amplification), two focal amplifications of FGFR4, and one focal amplification of FGFR1, EGFR, and MET. Four of the nine (44%) characterized MBCs displayed genomic changes to both the PI3K-AKT-mTOR and RTK-MAPK pathways (versus 11/55 for other subtypes, 20%).

We confirmed the activation of the PI3K-AKT-mTOR and RTK-MAPK signaling pathways at the protein level, by analyzing the expression of the major effectors of the PI3K-AKT-mTOR pathway (PI3-kinase, p-AKT, p-4E-BP1, p70-S6-kinase, p-S6RP, PTEN) and the RTK-MAPK pathway (p-MEK1, p-RSK2) by RPPA analysis. This analysis confirmed the activation of the PI3K pathway in metaplastic BCs. RSK2 is an effector of the MAPK pathway. Most metaplastic BCs had high RSK2 scores. However, by contrast to the PI3K pathway, the results of genomic and protein analysis were not well correlated (Additional file 2: Figure S1).

Based on the high frequency of concomitant alterations of the PI3K-AKT-mTOR and RTK-MAPK pathways, we hypothesized that the combination of PI3K and MEK inhibitors might constitute an efficient treatment strategy for MBC PDX. The PI3K inhibitor BYL-719 (alpelisib) and the MEK inhibitor selumetinib were tested in monotherapy and in combination in the HBCx-60 PDX (FGFR4 amplification, PIK3CA mutation, and AKT1 amplification), HBCx-165 PDX (FGFR4 amplification, PIK3CA mutation) and HBCx-178(BRAF mutation and PIK3CA mutation) models. Two of these models are chemoresistant to anthracycline (HBCx-165 and HBCx-178).

In the HBCx-60 model, treatment with BYL-719 and selumetinib in monotherapy did not significantly decrease tumor growth (Fig. 4a). By contrast, the combination inhibited tumor growth, with a TGI of 94% relative to the control (p = 0.0018, Mann-Whitney test). In this chemosensitive model, the combination of targeted therapies induced a response similar to that obtained with chemotherapy. In the HBCx-165 model (Fig. 4b), treatment with BYL-719 and selumetinib in monotherapy reduced tumor growth, with a TGI of 74% (p = 0.028) and 66% (p = 0.16), respectively, but xenografts showed no tumor growth arrest or regression. Conversely, in the group of xenografts treated with the combination of BYL-719 and selumetinib (TGI of 91%), six mice had stable disease, and two xenografts displayed tumor regression (data not shown). The combination was superior to BYL-719 alone (p = 0.0012). Finally, in the HBCx-178 PDX model, treatment with BYL-719 resulted in a TGI of 70% with no tumor regression, whereas the combination of BYL-719 and selumetinib resulted in a complete response in six of ten mice (Fig. 4c).

Monotherapies induced no significant change, but the combination was highly effective, with significant tumor regression observed in all models and a complete response frequency of 60% for HBCx-178.

To assess inhibition of PIK3 and MAPK signaling pathways, we analyzed the phosphorylation status of AKT, S6, ERK, and MEK in treated tumors harvested at the end of treatment by western blot (Fig. 4d). P-AKT was strongly inhibited in xenografts treated by BYL-719 alone or associated with selumetinib in HBCx-60 and HBCx-165 PDX and, to a lesser extent, in HBCx-178. P-S6 was strongly inhibited in xenografts of the combination arm. In the 3 PDX models, P-ERK was inhibited in selumetinib-treated xenografts and strongly inhibited in xenografts treated with the combination BYL-719 + selumetinib. In the two models (HBCx-60 and HBCx-165), P-MEK was inhibited in BYL-719 treated tumors. Overall, these results show strong inhibition of both PIK3 and MAPK signaling pathways in tumors treated by BYL-719 associated with selumetinib.