HER2+ breast cancer cell lines SKBR3 (ATCC^® HTB-30™), BT474 (ATCC^® HTB-20™), and HCC-1954 (ATCC^® CRL-2338™) were obtained from the American Type Culture Collection (ATCC). KPL4 and JMT1 were obtained from the Laboratory of Roberto Bianco, Department of Clinical Medicine and Surgery, University of Naples “Federico II”, Italy. SKBR3, HCC-1954, KPL4 and JIMT1 cells were grown in RPMI-1640 medium (Gibco, Invitrogen, Milan, Italy) supplemented with 10% fetal bovine serum (FBS; Gibco, Invitrogen, Milan, Italy), 100 U/mL penicillin, and 100 μg/mL streptomycin. BT474 cells were grown in DMEM medium (Gibco, Invitrogen, Milan, Italy) supplemented with 10% fetal bovine serum (FBS; Gibco, Invitrogen, Milan, Italy), 100 U/mL penicillin, and 100 μg/mL streptomycin. All cells were maintained under 5% CO2 in a humidified incubator at 37°C. Cells were routinely tested for mycoplasma using MycoFluor (Mycoplasma Detection Kit MycoAlert, M7006, Invitrogen, Milan, Italy). Trastuzumab and pertuzumab were purchased from SelleckChemed (Munich, Germany). Alpelisib was provided by Novartis Pharma (Basel, Switzerland). All experiments were conducted within two months of thawing of the acquired cells. Alpelisib-Resistant (AR) and Alpelisib-Trastuzumab-Resistant (ATR) derivatives were generated by exposing HCC1954 and KPL4 cell lines to gradually increasing doses of alpelisib (from 0.125μM to 1μM) or concomitant alpelisib (from 0.125μM to 1μM) plus trastuzumab (from 1μg/ml to 10μg/ml).

Four to eight thousand cells per well were seeded in 24-well plates and treated with the reported concentrations of trastuzumab and pertuzumab for 7 days and analyzed with the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazoliumbromide (MTT) assay according to the manufacturer’s instructions (Sigma–Aldrich, Italy) as previously described (17, 18). Experiments were performed three times; values represent means ± SEM from quadruplicate samples for each treatment. The percentage of absorbance of treated samples versus untreated is reported as a percentage of viable cells/controls. IC50 values were generated through GraphPad Prism v8.01 using the log (inhibitor) versus response-variable slope (four parameters) model as previously described (19).

HER2+ breast cancer cells were transfected with Lipofectamine RNAiMAX^® (Invitrogen) and 20 nM siRNA [siControl - OriGene cat.#; siAKR1C1#1-2-3 OriGene_SR301163], and seeded in 24-well multiwells (for cell growth assays) or 60 mm plates (for immunoblot assays) and incubated at 37°C, 5% CO₂ for 24 hours. The next day, the transfection medium (without antibiotics/antifungals) was replaced with the complete culture medium in the presence or absence of the drug. The culture medium ± drugs was replaced every 72 hours. On the sixth day post-transfection, the cells were analyzed using the MTT method. For immunoblot assays, cells were harvested and lysed 72 hours post-transfection.

The cells were plated, allowed to grow for 24 h before treatments and then treated with the drugs for 48h. All cells were maintained under 5% CO₂ in a humidified incubator at 37°C. The cell pellet was obtained after treatment with trypsin and centrifugation for 5 minutes at 1300 rpm. Protein preparations from cells were obtained by lysing samples with RIPA lysis buffer (ChemCruz, Dallas, TX). The samples were incubated on ice for 30 minutes and then centrifuged at 13200 rpm for 20 minutes at 4°C. Protein concentration was measured by the Bio-Rad Protein Assay (Bio-Rad, Milan, Italy). Twenty-five–microgram aliquots were electrophoresed through 8–10% SDS polyacrylamide gels. Prestained molecular weight standards were from Bio-Rad. The membrane was stained with Ponceau S (Sigma–Aldrich, Italy) to evaluate the success of transfer. Free protein binding sites were blocked with nonfat dry milk and Tween-20/TBS solution (5%). The membranes were washed, stained with specific primary antibodies (Ab) and then with secondary antisera, conjugated with horseradish peroxidase (1:3.000; Santa Cruz Biotechnology, Inc., Santa Cruz, CA). The following primary antibodies were used: Ab anti-phosphorylated (p-) HER2 (CST#2243, 1:1000), Ab anti-PTEN (CST #9559, 1:1000), Ab anti-AKT (CST#9272, 1:1000), Ab anti-p-AKT S473 (CST#9271, 1:1000), Ab anti-S6K, Ab anti-p-S6K T389 (CST#9205, 1:1000), Ab anti-beta-actin (CST#4970, 1:5000). The luminescent signal was visualized with the ECL Western blotting detection reagent kit (Bio-Rad, Milan, Italy) and detected by light-sensitive films (FUJIFILM Super RX-N 100 NIF).

KPL4 and HCC1954 cells were plated in complete medium and after 24 h treated with vehicle (DMSO), 1 µM alpelisib or 1 µM alpelisib + 10 µg/ml trastuzumab for 48 hours. KPL4-AR and HCC1954-AR cells were plated in complete culture medium in the presence of 1 µM alpelisib. KPL4-ATR and HCC1954-ATR cells were plated in complete culture medium in the presence of 1 µM alpelisib + 10 µg/ml trastuzumab. Total RNA of the HCC1954 and KPL4 cell lines was extracted with RNeasy Mini Kit (Qiagen, Germany). The RNA-seq libraries were prepared using the TruSeq RNA Sample Preparation Kit (Illumina) and the Agilent Automation NGS system per manufacturers’ instructions. Samples were sequenced on an Illumina HiSeq platform with paired-end 150 bp reads. Sequence reads were trimmed to remove possible adapter sequences and nucleotides with poor quality using Trimmomatic v.0.36. The trimmed reads were mapped to the Homo sapiens GRCh38 reference genome available on ENSEMBL using the STAR aligner v.2.5.2b. The STAR aligner is a splice aligner that detects splice junctions and incorporates them to help align the entire read sequences. BAM files were generated as a result of this step. Unique gene hit counts were calculated by using featureCounts from the Subread package v.1.5.2. The hit counts were summarized and reported using the gene_id feature in the annotation file. Only unique reads that fell within exon regions were counted. If a strand-specific library preparation was performed, the reads were strand-specifically counted. After extraction of gene hit counts, the gene hit counts table was used for downstream differential expression analysis. Using DESeq2, a comparison of gene expression between the groups of samples was performed. The Wald test was used to generate p-values and log2 fold changes. Genes with an adjusted p-value <0.05 and absolute log2 fold change > 1 were called as differentially expressed genes for each comparison. A gene ontology analysis (20) was performed on the statistically significant set of genes by implementing the GeneSCF v.1.1-p2 software. The goa_human GO list was used to cluster the set of genes based on their biological processes and determine their statistical significance. The GSEA software (Broad Institute, Massachusetts, USA) and MSigDB v6.2 Hallmark gene sets collection were used to calculate enrichment scores.

Kaplan–Meier survival curves were generated using an online meta-analysis tool for public microarray datasets (21). This integrative data analysis tool was employed to evaluate the prognostic value of the following genes: ARL4A, SNRPG, ODC1, AKR1C1, TRIM16L, AKR1C2, GLB1L2, and BDH1. Kaplan–Meier plots were generated after averaging the probes. The selected clinical cases were HER2+ (defined by immunohistochemistry) breast tumors divided into two groups (high and low) by the median expression value of each gene. P values were determined by the log-rank test.

HCC 1954 cells were maintained as described above. Ovariectomized 5- to 6-week-old athymic mice (Charles River, Italy) received a subcutaneous injection into the right flank of 1×10⁶ HCC1954 cells suspended in Matrigel + RPMI-1640 (1:1). When the volume of the cell xenografts reached ~150–200 mm³, mice were randomly assigned (n=10 per group) to receive vehicle, 30 mg/kg trastuzumab (via intraperitoneal administration twice weekly), 30 mg/kg alpelisib (administered orally via gavage, daily) or the combination of 30 mg/kg trastuzumab plus 30 mg/kg alpelisib. Alpelisib was dissolved in PEG400/acidified water (0.1 N HCl, pH 3.5; 20/80, v/v). Tumor dimensions and body weights were recorded twice weekly starting on the first day of treatment. The tumor volume was calculated according to the formula π/6 × (larger diameter) × (smaller diameter)². The in vivo antitumor activity was measured using the rate-based T/C as previously described (22). All animal experiments were performed according to protocols approved by the Animal Care and Use Committee of Federico II University.

Statistical analyses were performed using GraphPad Prism v9.01. Quantitative data are shown as mean ± SEM. A p-value <0.05, calculated by two-tailed Student’s t-test, ANOVA with Bonferroni’s or Dunnett/Tukey’s post-hoc tests, or Pearson/Spearman correlation coefficient test was considered as statistically significant.

We selected a panel of PIK3CA mutant (KPL4, JMT1, and HCC1954) and PIK3CA WT (SKBR3 and BT474) HER2+ breast cancer cell lines. The estrogen receptor (ER), the progesterone receptor (PR), and the PIK3CA status of each cell line are shown in Supplementary Table 1. First, we investigated the efficacy of the anti-HER2 monoclonal antibodies trastuzumab and pertuzumab in our panel of HER2+ breast cancer cell lines (Figure 1). In line with the results of previous studies (3, 11), trastuzumab or the combination of trastuzumab plus pertuzumab inhibited the growth of BT474, SKBR3 and HCC1954 cell lines. On the contrary, the anti-HER2 monoclonal antibodies had no effect or only a modest inhibitory effect on JIMT1 and KPL4 cell lines, respectively. We next assessed the efficacy of alpelisib in our models and found that Alpelisib treatment resulted in a dose-dependent inhibition of cell growth of KPL4, HCC1954, SKBR3 and BT474 cell lines but not of JIMT1 cells that had a complete de novo resistance to the drug (Figure 2A). The calculated alpelisib IC50 values are reported in Figure 2B. The KPL4 cell line was the most alpelisib-sensitive, while the intrinsic alpelisib-resistant JIMT1 model was excluded from subsequent experiments. We next investigated the efficacy of anti-HER2 monoclonal antibodies (trastuzumab or trastuzumab plus pertuzumab), alpelisib, or their combinations in our panel of HER2+ breast cancer cell lines. Specifically, KPL4, HCC1954, SKBR3 and BT474 cell lines were subjected to vehicle (DMSO, Sigma–Aldrich, Italy), 10μg/ml trastuzumab, 10μg/ml trastuzumab + 10μg/ml pertuzumab, 1μM alpelisib, 1μM alpelisib + 10μg/ml trastuzumab, and the triplet combination of 1μM alpelisib + 10μg/ml trastuzumab + 10μg/ml pertuzumab (Figure 3). We observed that alpelisib monotherapy was more effective than trastuzumab or trastuzumab + pertuzumab in inhibiting cell growth in the PIK3CA mutant KLP4 and HCC1954 cell lines (growth inhibition >90% and >70%, respectively). Among PIK3CA WT models, no differences in cell growth inhibition were observed between trastuzumab or trastuzumab + pertuzumab vs. alpelisib treatments in BT474 cells, differently, alpelisib had a higher inhibitory effect than trastuzumab whereas trastuzumab + pertuzumab in the SKBR3 cells did not. Interestingly, the addition of alpelisib to the anti-HER2 therapy resulted in a more pronounced growth inhibition in 3 out of the 4 cell line models (BT474, SKBR3 and HCC1954) (Figure 3A). Western blot analysis of selected components of the PI3K pathway revealed a greater reduction of phosphorylated AKT (p-AKT) and ribosomal S6 kinase protein (p-S6K) levels upon treatment with alpelisib plus anti-HER2 therapy (trastuzumab or trastuzumab + pertuzumab) versus alpelisib or anti-HER2 therapy alone in SKBR3 and BT474 cell lines (Figure 3B). In PIK3CA mutant KPL4 and HCC1954 models, the addition of alpelisib to anti-HER2 therapy resulted in a higher inhibition of p-S6K but not p-AKT protein levels (Figure 3B). We also assessed the potential drug interaction between alpelisib and trastuzumab in PIK3CA mutant HER2+ breast cell lines using Chou-Talalay method (23), where a combination index (CI) < 1 indicates synergy between drugs. Viability assays using increasing concentrations of alpelisib and trastuzumab demonstrated a pharmacological synergism between these agents in both HCC1954 and KPL4 models (CI=0.24 and CI=0.36, respectively; Figure 4).

The inhibitory activity of alpelisib on tumor growth was evaluated in vivo using xenografts of HCC1954 cells. When tumors reached a volume of approximately 200 mm³, mice were randomized into the following treatment groups: vehicle (n=10), trastuzumab (n=10), alpelisib (n=10) and trastuzumab + alpelisib (n=10) (Figure 5). In the absence of treatment (vehicle), HCC1954 xenografts had a high growth rate, reaching a tumor volume of 1500 mm³ in about 15 days. In line with the results of our in vitro experiments, tumor growth was moderately slower in tumors treated with trastuzumab than in those treated with vehicle. In addition, tumor growth was slower in alpelisib-treated tumors than in those treated with vehicle. Alpelisib treatment substantially inhibited tumor growth unlike treatment with vehicle or trastuzumab. Finally, the combination of alpelisib plus trastuzumab reduced growth versus the individual drugs, although the difference with the alpelisib treatment arm was not statistically significant (Figure 5). Notably, total regression of the tumor was not found in any of the treatment arms. Overall, our data support the potential role of alpelisib for the treatment of HER2+/PI3KCA mutated breast cancers.

To investigate the molecular mechanisms promoting resistance to alpha-specific PI3K inhibition, we generated derivatives with acquired resistance to alpelisib and to alpelisib plus trastuzumab (ATR) from KPL4 and HCC1954 cells by exposing cells to increasing doses of alpelisib or alpelisib + trastuzumab. Next, we analyzed the transcriptomic profiles of the HCC1954 and KPL4 cell lines treated with short term (48 hours) alpelisib or alpelisib + trastuzumab together with their resistant derivatives in order to identify transcriptional alterations underlying alpelisib resistance. The differentially expressed genes (DEG) in alpelisib-resistant vs sensitive HER2+/PIK3CA mutant breast cancer cell lines are reported in Supplementary Table S1. Gene ontology analysis of the DEG between sensitive vs resistant cells showed a common enrichment for biological processes associated with oxidation reduction (“oxidation reduction process”), proliferation and cell division (“cell division”, “cell proliferation”), immune response (“inflammatory response”, “innate immune response”) and RNA synthesis (“positive regulation of transcription, DNA template”) in AR and ATR models (Supplementary Figure S1). We next focused on the significantly up-regulated genes in AR and ATR models. Globally 8 genes (ARL4A, SNRPG, ODC1, AKR1C1, TRIM16L, AKR1C2, GLB1L2, and BDH1) were found to be commonly overexpressed (log2 fold-change >1, adjusted p-value <0.05) in our resistant models (Figure 6).

To investigate the impact of these genes in determining alpelisib sensitivity, we used genomic and cellular data from the Broad Institute’s “Cancer Dependency Map Portal” (https://depmap.org/portal/achilles/). In particular, we analyzed the association between the expression levels of each gene and the in vitro growth inhibitory effect of alpelisib in 26 breast cancer cell lines. This analysis revealed that the expression of 6 genes (AKR1C1, AKR1C2, TRIM16L, ODC1, ARL4A and SNRPG) negatively correlates with alpelisib sensitivity (Figure 7). We also investigated the prognostic value of these genes by evaluating their expression in primary HER2+ breast tumors. To conduct this analysis, we used the publicly available clinical and gene expression data from the “Kaplan-Meier Plotter” database (https://kmplot.com/analysis/). Among the 8 genes analyzed, only AKR1C1 and SNRPG had a significant prognostic value. In detail, patients with HER2+ breast cancer and high expression of AKR1C1 had a higher risk of recurrence than did patients with low AKR1C1 expression (Hazard Ratio [HR]=1.53, 95% Confidence Interval [CI] 1.01-2.32, p-value = 0.043). Conversely, in the same patient population, elevated expression of SNRPG was associated with a better prognosis (HR=0.6, 95% CI 0.39-0.92, p-value = 0.018) (Figure 8). Lastly, we examined the effects of changes in AKR1C1 expression on cell growth of alpelisib sensitive and resistant lines (HCC1954, HCC1954 AR, KPL4 and KPL4 AR). AKR1C1 knock-down by siRNA resulted in a significant reduction of cell growth of parental cells (range 20-25%). However, in HCC1954 and KPL4 AR models, which had higher AKR1C1 protein levels versus parental cells, AKR1C1 knock-down resulted in a more pronounced cell growth inhibition (range 70-75%). We also found that knockdown of AKR1C1 enhanced the effect of alpelisib in inhibiting cell growth in both HCC1954 and KPL4 cell lines (Supplementary Figure S2). Overall, these data indicate that AKR1C1 may play a potential role in determining the sensitivity to alpelisib and sustaining cell growth of models with acquired resistance to alpelisib (Figure 9).