Tumor and stromal cell targeting with Nintedanib and Alpelisib overcomes intrinsic bladder cancer resistance

Abstract

Bladder cancer is a highly prevalent tumor, requiring the urgent development of novel therapies, especially for locally advanced and metastatic disease. Nintedanib is a potent anti-fibrotic angio-kinase inhibitor, which has shown clinical efficacy in combination with chemotherapy in locally advanced muscle invasive BC patients. Nintedanib inhibits Fibroblast Growth Factor receptors (FGFRs), validated targets in patients with BC harboring FGFR3/2 genetic alterations. Here, we aimed at studying its mechanisms of action to understand therapy resistance, identify markers predictive of response, and improve the design of future clinical trials. We have used a panel of genetically well-characterized human BC cells in order to identify the molecular and transcriptomic changes induced upon treatment with Nintedanib, in vitro and in vivo, at the tumor and stroma cell levels. We showed that BC cells display an intrinsic resistance to Nintedanib treatment in vitro, independently of their FGFR3 status. However, Nintedanib has higher antitumor activity on mouse xenografts. We have identified PI3K activation as a resistance mechanism against Nintedanib in BC and evidenced that the combination of Nintedanib with the PI3K inhibitor Alpelisib has synergistic antitumor activity. Treatment with this combination is associated with cell cycle inhibition at the tumoral and stromal levels and potent non-tumor cell autonomous effects on α-smooth muscle (SMA)-positive tumor infiltrating cells and tumor vasculature. The combination of Nintedanib with PI3K inhibitors not only reversed BC resistance to Nintedanib but also enhanced its anti-angiogenic effects.

INTRODUCTION

Bladder cancer (BC) is the 10th most frequent tumor worldwide, its incidence being 3–7 times higher in men than in women in the Western world. The vast majority of bladder tumors are urothelial carcinomas (UCs) and 80% of newly diagnosed UCs are non muscle-invasive (NMI) neoplasms. The remaining 20% are muscle-invasive (MI) tumors. While NMI-BC has a good prognosis and is rarely fatal, MI-BC is an aggressive disease and the 5-year survival of patients ranges from 60%, for those with localized disease, to 15% for those with metastatic disease (1).

Standard first-line treatment for patients with MI-BC consists of cisplatin-based neoadjuvant chemotherapy followed by radical cystectomy. This approach is associated with a significant deterioration of the quality of life. Therefore, there is a growing trend to use bladder-preserving strategies. Within two years of surgery, up to 50% of patients will develop distant recurrences (2). For patients with metastatic disease there is no curative therapy and the standard treatment consists of cisplatin combination chemotherapy. However, almost half of the patients are ineligible to receive this drug. Immune checkpoint inhibitors (ICI) and carboplatin-based chemotherapy are the main treatment options for this group of patients but response rates are in the range of 30–40% (3). Second-line ICI have demonstrated anti-tumor activity in approximately 30% of patients with metastatic BC (4). The poor prognosis associated with MI-BC highlights the urgent need of developing novel treatment strategies for this disease.

Until recently, targeted therapies had not become part of the arsenal of drugs to treat BC but the recent development of more selective targeted drugs has opened new therapeutic opportunities. Fibroblast growth factor receptor 3 (FGFR3) is the most commonly mutated oncogene in BC; hotspot mutation incidence is higher in NMI (60–70%) than in MI-BC (10–15%). FGFR3 gene fusions have also been reported, although at much lower frequency (5,6). The value of FGFR3 as a BC therapeutic target has been recognized with the approval of the pan-FGFR inhibitor Erdafitinib by the FDA as a second line treatment for patients with altered FGFR3/2 having failed a platinum-containing regimen (7). Recent results using the anti-Vascular endothelial growth factor 2 (VEGFR2) antibody Ramucirumab point to angiogenesis as one of the most promising BC targets (8).

Given the demonstrated value of the FGFR pathway and tumor angiogenesis as targets in BC, the angiokinase inhibitors should be considered promising drugs. Nintedanib is a small-molecule receptor tyrosine kinase (RTK) inhibitor targeting FGFR1–3, VEGFR1–3, and Platelet-derived growth factor receptor (PDGFR) α and β that has been approved for the treatment of lung cancer and idiopathic pulmonary fibrosis (9,10). Neoadjuvant Nintedanib, in combination with standard chemotherapy, has been recently reported to improve survival of patients with locally advanced MI-BC (11). The mechanisms involved are not well understood. In order to rationally design future clinical trials combining Nintedanib with other therapies it is mandatory to elucidate its mechanism of action. The growth promoting effects of RTKs - such as FGFR3, VEGFR- and PDGFR - on angiogenesis and other stromal components may contribute to the antitumor effects of these drugs. This knowledge should help to stratify patients for therapy. To answer some of these questions, we have performed preclinical in vitro and in vivo studies to evaluate the effects of Nintedanib using a panel of genetically well-characterized human BC cell lines (12). We show that the anti-tumor effects of Nintedanib are enhanced when used in combination with Alpelisib, an isoform-specific PI3K inhibitor that is approved for the treatment of advanced or metastatic breast cancers harboring PIK3CA mutations (13).

MATERIAL AND METHODS

Cell lines and cell culture

The source of the human BC cell lines used (MGH-U3, 5637, UM-UC-7, VM-CUB-1, UM-UC-6, J82, RT4, SW1710, UM-UC-3, SCaBER, 639V, 97–7, UM-UC-17, T24, SW-780, UM-UC-9, RT112) has been reported by our group (12). Cell lines were authenticated using short tandem repeat (STR) DNA profiling. Cells were maintained in high glucose Dulbeccós Modified Eaglés Medium (Sigma-Aldrich, D5671) supplemented with 10% Fetal Bovine Serum (GIBCO, 11520646) and 1 mM Sodium Pyruvate (Lonza, BE13–115E). Cells were grown at 37 ºC in 5% CO₂. All cells used were Mycoplasma-free.

Reagents

MEK (PD325901) and GDC-0941 (Pictilisib) were purchased from MedChemExpress. Nintedanib and Alpelisib were obtained from Boehringer-Ingelheim and Novartis, respectively.

Cell viability assays

For drug sensitivity and in vitro drug combination assays, BC cells were seeded at 7 ×10³ cells/well in triplicate in 96-well plates (Corning, Falcon). Viability was quantified 72h later using the CellTiter Glo Luminescent Cell Viability Assay (Promega; G7573). The combination ratios for Nintedanib and Alpelisib are specified in Supplementary Table S1. Area under the curve (AUC) and GI50 for single drugs were calculated using GraphPad Prism 8.0 software (RRID:SCR_002798). Combination drug GI50 and combination index (CI) were determined using Compusyn software (http://www.combosyn.com/). Cells were classified according to experimentally obtained AUC values into Nintedanib-sensitive (AUC≤0.5), Nintedanib-poor-responders (0.5>GI50 ≤0.71), and Nintedanib resistant (GI50 ≥0.71). For apoptosis assays, cells were seeded at 2 ×10⁵ cells/well in 6-well plates in triplicate and treated with vehicle (DMSO), 2 μM or 10 μM Nintedanib for 24h, 72h, or 6d. After this, the culture medium was recovered, cells were trypsinized and both fractions were pooled and centrifuged at 1500 rpm for 5 minutes. The pellets were resuspended with light vortexing in ice-cold 90% EtOH/PBS solution and stored at 4 ºC. Cell cycle distribution and sub-G0/G1 fraction were determined by DNA staining using propidium iodide solution (50 μg/mL RNase, 50 μg/mL propidium iodide in PBS) and flow cytometry (FACSCalibur, BD Biosciences). Data were analyzed using FlowJo v10.2 software (BD Biosciences, RRID:SCR_008520).

Western blotting and protein-based assays

Proteins were extracted from cultured cells and lysed in radioimmunoprecipitation buffer (20mM Tris-HCl, pH 8.0, 137 mM NaCl, 1 mM MgCl₂, 1 mM CaCl₂, 10% glycerol, 1% NP40, 0.5% sodium deoxycholate, 0.1% SDS) containing protease (Complete-EDTA free protease inhibitor cocktail, Roche) and phosphatase inhibitors (Phosphatase inhibitor cocktail 3, Sigma-Aldrich), 1 mM sodium orthovanadate, and 1 mM phenylmethylsulfonyl fluoride (PMSF). Bio-Rad Protein Assay Dye Reagent (BIO-RAD) was used to measure protein concentration. Proteins (30 μg) were resolved by SDS-PAGE and transferred to nitrocellulose membranes. Antibodies against the following antigens were used: pAKT (1:600, Cat#9271S, RRID:AB_329825), AKT (1:1000, Cat#9272S, RRID:AB_329827), pERK1/2 (1:1000, Cat# 9101S, RRID:AB_331646), ERK1/2 (1:1000, Cat# 9102, RRID:AB_330744), all from Cell Signaling Technology. Vinculin expression was assessed as loading control using anti-Vinculin (1:5000, Sigma-Aldrich Cat#V9131, RRID:AB_477629). HPRP-conjugated secondary antibodies were from DAKO-Agilent Technologies. Western blots were developed using Luminata Classico HRP substrate (Merck-Millipore, 10680720). Phospho-specific and total protein antibodies were applied either independently to different membranes, and processed in parallel, or sequentially using the same membrane, after stripping with Restore Western Blot Stripping Buffer (Thermo Scientific, 21059). For dot blot-based phospho-array experiments, protein lysates of frozen tumor biopsies from 14 day-treated vehicle-, Nintedanib-responder, Nintedanib-non responder, and combination-treated VM-CUB-1 tumors were prepared (n=3 each) and protein concentration measured as indicated above. Equivalent amount of proteins from the biological replicates were then pooled and subjected to the phospho-RTK array (R&D Systems, ARY001B), following manufacturer’s guidelines. Image J software was used for dot-blot signal quantification. For this experiment, protein lysates from tumor sample pools were pooled.

In vivo experiments

For xenograft transplantation, cells (1.5×10⁶ in 100 μL of 80% matrigel/PBS) were injected subcutaneously into 8–10 week-old Hsd:Athymic Nude-Foxn1nu female mice (Jackson Laboratories). Once the tumor volume reached 0.1 cm³, mice were randomized to receive: vehicle (0.5% methylcellulose, 0.1% Tween80), Alpelisib (25mg/kg), Nintedanib (40mg/kg), or the combination [Alpelisib (25mg/kg); Nintedanib (40mg/kg)]. Mice received compounds by oral gavage 6 days/week. The general aspect of the animals was verified daily by means of visual inspection, signs of pain, or abnormal motility. Mice were weighed at least once weekly. Animal procedures were approved by the institutional Ethics and Animal Welfare Committee (Instituto de Salud Carlos III, Madrid, Spain) and the General Guidance of the Environment of Madrid Community in accordance with the guidelines for Ethical Conduct in the Care and Use of Animals as stated in The International Guiding Principles for Biomedical Research involving Animals, developed by the Council for International Organizations of Medical Sciences (CIOMS).

Immunohistochemical (IHC) analyses

IHC was performed using sections of formalin-fixed paraffin-embedded tissues. After deparaffinization and rehydration, antigens were retrieved by boiling in citrate buffer pH 6 for 10 min. Endogenous peroxidase was blocked with 3% H₂O₂ in PBS for 30 min. Then, sections were incubated with 3% BSA/0.1% Triton in PBS for 1h at room temperature and with primary antibody overnight at 4 °C. After washing with 0.1% Triton/PBS, sections were incubated with the appropriate secondary antibodies (Envision kit for mouse or rabbit Ig, Agilent Cat# K4001, RRID:AB_2827819 and Cat #4003, RRID:AB_2630375 ) for 45 min, washed, and reactions were developed with DAB (Dako, K3468). Sections were lightly counterstained with haematoxylin, dehydrated, and mounted. Primary antibodies used for IHC were raised against: SMA (1:2000, Sigma-Aldrich Cat# A2547, RRID:AB_476701), CD31 (1:100, Abcam Cat# AB28364, RRID:AB_726362), pHistone H3 (pH3) (1:200, Millipore Cat#06–570, RRID:AB_310177), and cleaved Caspase-3 (Asp 175) (1:200, Cell Signaling Cat# 9661, RRID:AB_2341188). IHC protocols have been optimized by the CNIO Histopathology Unit using appropriate controls: no primary antibody, isotype control, and positive control tissues. After digitalization of the sections, 8 regions of interest (2000× 2000 pixels) were taken from each tumor and exported to TIFF with ZEN Zeiss software. IHC staining was quantified using the Color Deconvolution plugin of Image J software (RRID:SCR_003070) (14). Alternatively, IHC staining of whole tumor sections was quantified with QuPath software (RRID:SCR_018257, https://qupath.github.io). In both cases, DAB signal was normalized to the hematoxylin-stained area. CD31 image analysis was used to assess vascular density (vasculature area/total area) and vascular length density (skeletonized vasculature/total area) using Image J Vessel analysis plugin (15).

RNA extraction and RNA-sequencing (RNA-seq)

Total RNA was isolated from frozen tissues and cultured cells using ReliaPrep RNA Miniprep Systems (Promega, Z6012) or TRItidyG (ITW Reagents, A4051). Reverse transcription was performed using TaqMan Reverse Transcription Reagents (ThermoFisher Scientific, N8080234). Libraries for whole-transcriptome sequencing were constructed using the TruSeq RNA Sample Preparation Kit v2 (VM-CUB-1 tumor samples) or the QuantSeq 3’ mRNA-Seq Library Prep Kit for Illumina protocol (ScaBER tumor samples) and sequenced using the 75-bp paired-end protocol on a NextSeq500 (RRID:SCR_014983) (Illumina, San Diego, CA, USA).

RNA-seq data analysis

The Nextpresso version 1.9.1 analysis pipeline (16) was used to process the data, which were aligned to the version GRCh37/hg19 of the human or mm10 of the mouse reference genomes. In this pipeline, HTseq-count (RRID:SCR_011867) and DESeq2 (RRID:SCR_000154) packages were used to generate count matrices and to run differential expression gene (DEG) analysis, respectively. FDR adjusted p-value (q<0.05) were considered significant. Gene set enrichment analyses (GSEA) against ranked gene lists generated by DEseq [in which genes are ranked based on log2-fold change (treatment vs vehicle)] were computed using the GSEAPreranked tool of GenePattern v3.9 (RRID:SCR_003199, RRID:SCR_003201) (17,18), using the KEGG, Biocarta, PID and REACTOME databases. FWER p-val < 0.05 was used to establish significance. Cytoscape (v3.9.1) (RRID:SCR 003032) (19) was used to cluster gene pathways in functionally grouped networks. Annotation of the category was added manually. Cell type enrichment analysis was performed using xCell (http://xCell.ucsf.edu/) (20).

Plot Generation

Heatmaps and Venn diagrams were generated using tidyverse, pheatmap, and eulerr R packages (RRID:SCR_019186, RRID:SCR_016418, RRID:SCR_022753). GraphPad Prism 8.0 (RRID:SCR_002798) software was used to build all other plots and graphs. Cytoscape (v3.9.1) EnrichmentMap plugin (RRID:SCR_003032, RRID:SCR_016052) (19) was used for the visualization of significant enriched pathway clusterization into similarity-based networks.

Statistical analysis

One-way analysis of variance ANOVA was used when more than two groups were compared. Mann-Whitney U-Test was used to compare differences between two independent groups when the data did not follow a normal distribution. Statistical analyses were performed with GraphPad Prism version 8.0.1 software. Xcell data statistics were performed by R software: ANOVA test was used to compare differences between group means, and Tukey test was applied as post hoc test. Two-sided P values <0.05 were considered significant.

Data Availability

Raw and processed sequencing data have been deposited in the Gene Expression Omnibus repository (GSE182833 and GSE218575).

RESULTS

Nintedanib has modest in vitro anti-proliferative effects on bladder cancer cells

We tested the sensitivity of a panel of 17 well-characterized human BC cell lines (12) to Nintedanib by calculating GI50 and AUC values. Only 3 out of 17 cell lines (RT4, SW1710, and UM-UC-6) were considered sensitive, displaying GI50s of 0.8μM, 1.9μM, and 2.4μM, respectively. Eleven additional cell lines (UM-UC-3, 639V, SCaBER, RT112 UM-UC-9, UM-UC-17, 97–7, T24, SW-780, MGH-U3, and 5637) responded poorly. J82, VM-CUB-1, and UM-UC-7 cells were resistant (Fig. 1A, Supplementary Table S2). To evaluate Nintedanib effects at longer time points, we selected one sensitive, one mid-responder, and two resistant cell lines and measured apoptosis by flow cytometry after treating cells for 1, 3, and 6 days with low (2μM) or high (10μM) drug concentrations. Nintedanib displayed low cytotoxic activity, even on SW1710-sensitive cells (Fig. 1B). Finally, we evaluated whether Nintedanib sensitivity was associated with FGFR3 genetic status: cells harboring FGFR3 mutations or gene fusions did not show higher sensitivity than the wild type ones (Fig. 1C).

Figure 1. Effects of Nintedanib on cell proliferation and apoptosis in vitro are independent from FGFR3 genetic alterations.

(A) Percentage of viable cells after 72h of exposure to Nintedanib (0.1–10μM), relative to DMSO treatment (%± SEM; n=3). (B) Percentage of apoptotic cells (sub G0/G1) upon treatment with Nintedanib (n=3). The median is represented by the horizontal line within the density plot. * P<0.05, two-tailed Mann Whitney U-test. (C) GI50 (left) and normalized AUC of the Nintedanib dose-response curve (right) of bladder cancer cells classified by FGFR3 genetic status.

PI3K activation does not correlate with Nintedanib sensitivity but it mediates intrinsic resistance of BC cells

All tested BC cell lines displayed, at least, a partial resistance to Nintedanib. RTK crosstalk and downstream feedback mechanisms involving multiple signaling cascades have been shown to reduce the effects of kinase inhibitors. To determine whether upstream RTKs are reactivated upon Nintedanib treatment, we used dot blot-based phospho-RTK arrays with protein extracts from Nintedanib-sensitive SW1710 cells treated with the drug for 0, 1, 4, or 7 h at a concentration in the GI50 range (2μM). All the RTKs tested were transiently dephosphorylated upon treatment but there was a subset of RTKs that were reactivated at later time points, therefore being candidates to participate in downstream pathway activation (Fig. 2A). Among the reactivated RTKs were several EGFR family members (e.g. EGFR, ERBB2, and ERBB3).

Figure 2. MAPK and PI3K pathway signaling is not persistently suppressed upon in vitro treatment with Nintedanib.

(A) Heatmap showing relative ImageJ software quantification of the signals of all RTKs included in the array. Phospho-receptors showing reactivation at later time points are highlighted in red. (B) WB analysis of the indicated proteins using lysates of cells treated for indicated times with DMSO, 2 or 10μM Nintedanib. (C,D) Cell viability (mean ± SEM) after 72h of treatment with Nintedanib (GI50), PD325901 (0.5μM) (C), or GDC-0941 (1μM) (D), either alone or in combination (n=3–4). * P<0.05, two-tailed Mann-Whitney U-test.

The modest anti-tumor effects observed, and the alternative RTK activation in SW1710 cells, prompted the analysis of the effects of Nintedanib (2 and 10 μM) on PI3K and MAPK, two or the main downstream signaling mediators, in sensitive (SW1710) and resistant (J82, VM-CUB-1 and UM-UC-7) cells (Fig. 2B). Upon drug treatment, MAPK pathway activation, determined by phospho-ERK1/2 (pERK1/2) levels, was induced in SW1710 and VM-CUB-1 cells. However, pERK1/2 levels were transiently down-regulated in UM-UC-7 and J82 cells. The PI3K pathway, assessed by phospho-AKT (pAKT) levels, was transiently down-regulated in sensitive (SW170) and resistant (UM-UC-7) cells. pAKT levels were not reduced at any time upon Nintedanib addition to resistant J82 and VM-CUB-1 cells, even at higher drug concentrations. Importantly, J82 and VM-CUB-1 harbor PIK3CA mutations (Supplementary Table S3). These data demonstrate that Nintedanib does not lead to a sustained inhibition of the MAPK and PI3K pathways, neither in sensitive nor in resistant cells. However, these findings do not rule out their contribution to Nintedanib resistance.

To address this possibility, we compared the effects of inhibiting MEK (PD325901) and PI3Kα/δ (GDC-0941) - alone or in combination with Nintedanib - in a subpanel of representative cell lines (Fig. 2C and 2D). The Nintedanib/GDC-0941 combination, unlike the Nintedanib/PD325901 combination, had significantly higher anti-proliferative effects on all tested cells, except for MGH-U3, independently from their sensitivity to Nintedanib.

PI3K inhibition synergizes with Nintedanib to impair BC cell growth in vitro and in vivo

Based on these results, we analyzed the effects of the combination of Nintedanib with Alpelisib (13), a FDA-approved PI3Kα-specific inhibitor (Fig. 3A). This drug combination had synergistic effects and impaired the in vitro growth of 6 out of 11 BC lines (J82, RT4, SW1710, SCaBER, VM-CUB-1, UM-UC-7). Synergy was defined by an experimentally-calculated combination index (CI) <1 by the Chou-Talalay method (21) and a GI50 <10.5 μM (calculated as the sum of GI50s of each drug, Supplementary Table S3). We then tested the effects of this drug combination in vivo using SW1710, SCaBER, and VM-CUB-1 cells xenotransplanted into athymic nude mice. The Nintedanib/Alpelisib combination impaired tumor growth in the three tumor models tested; in SW1710 and VM-CUB-1 xenografts, it showed a greater antitumor effect than single compound therapy. There was no major toxicity or weight loss (Supplementary Figure S1A). Surprisingly, Nintedanib alone significantly reduced the growth of SCaBER and VM-CUB-1 tumors, but not of SW1710 xenografts, suggesting the involvement of non-tumor cell-autonomous effects (Fig. 3B). Nintedanib/Alpelisib-treated tumors showed lower tumor cell mitotic index (pH3) at 7 day in VM-CUB-1 cells and at 14 day in ScaBER cells. SW1710 tumors also exhibited increased apoptosis (cleaved-caspase 3) upon Nintedanib and Nintedanib/Alpelisib treatment (Fig. 3C and 3D, Supplementary Fig. S1B and S1C).

Figure 3. Synergistic effects of Nintedanib and Alpelisib in vitro and in vivo.

(A) Percentage of viable cells relative to DMSO- treated cells (mean ± SEM; n=3) after 72h of exposure to Nintedanib, Alpelisib, or their combination. (B) Relative tumor volume ± SEM of SW1710, SCaBER, and VM-CUB-1 xenografts (n=8–10 per group) treated with vehicle, Alpelisib (25mg/kg), Nintedanib (40mg/kg), or their combination (same doses). * P<0.05; **, P<0.001; ***, P<0.0005; two-tailed Mann-Whitney U-test. (C) Normalized DAB signal of staining of the indicated markers in different treatment groups from SW1710, SCaBER and VM-CUB-1 xenografts. Days of treatment is also indicated. The median is represented by the horizontal line within the density plot; *, P<0.05, **, P<0.01 two-tailed Mann-Whitney U-test. (D) Representative images of IHC analysis of the indicated antigens in VM-CUB-1 xenograft tumors treated for 13 days with vehicle, Nintedanib, Alpelisib, or their combination. Scale bar, 50 μm. (E) Heatmap showing the two-way hierarchical clustering of the indicated phospho-proteins level from pooled tumor protein lysates. Each row represents the ImageJ software signal quantification of the phospho-array signal for a specific antigen and each column represents the data corresponding to pooled protein lysates from vehicle-, Nintedanib-responder, Nintedanib-non responder, and combination-treated VM-CUB-1 tumors (n=3 each).

Among Nintedanib-treated VM-CUB-1 tumors, we observed a dichotomous behavior: some tumors showed major responses whereas others did not (Supplementary Fig. S1D). To explore the molecular mechanisms underlying the in vivo Nintedanib resistance, VM-CUB-1 tumor lysates from control and drug-treated mice were analyzed by dot blot-based phospho-arrays. Unsupervised clustering of the samples showed that Nintedanib non-responder tumors displayed a signaling activation pattern similar to that of vehicle-treated tumors. In contrast, Nintedanib-responder tumors clustered with combination-treated tumors and showed reduced activation of almost all the kinases analyzed (Fig. 3E). These data demonstrate that PI3K activation mediates in vivo resistance to Nintedanib.

Transcriptome analysis reveals that Nintedanib and Alpelisib differentially target the tumor and the stroma

To gain insight into the mechanisms underlying Nintedanib and Alpelisib synergism in vivo, we performed RNA sequencing of ScaBER and VM-CUB-1 xenografts treated for 14 and 7 days respectively, with either vehicle (n=4 each), Nintedanib [responders (VM-CUB-1 (n=2) and ScaBER (n=4) and VM-CUB-1 non-responders (n=3)], Alpelisib (n=4), or their combination (n=4) (Supplementary Fig. S2A). Reads were mapped separately against the human and mouse genomes to differentiate tumor vs. stromal expression changes. DESeq was used to identify differentially expressed transcripts between vehicle-treated and drug-treated tumors (FDR ≤ 0.05). The transcriptomic effects observed upon treatment with single drugs or the combination on the stroma of ScaBER and VM-CUB-1 xenografts differed markedly (Supplementary Fig. S2B). In order to extract the shared molecular changes driven by the Nintedanib/Alpelisib combination in tumor cells and in the tumor microenvironment, and to minimize the impact of sample size on the differentially expressed genes (DEGs), we performed Gene Set enrichment analysis (GSEA) using pre-ranked mouse and human gene lists, in which genes are ranked by log2-fold change of drug-treated (Nintedanib, Alpelisib or Combination) vs vehicle-treated samples. For these analyses VM-CUB-1 Nintedanib responders and non-responders were also analyzed separately.

The Nintedanib/Alpelisib combination induces changes in the expression of genes/pathways involved in cell cycle, DNA repair, and ECM remodeling

To identify consistent effects on the human transcriptome across treatments, we represented the normalized enrichment scores (NES) for all the gene sets from the interrogated databases for the following comparisons: Nintedanib-responder vs vehicle, Alpelisib vs vehicle, and Combination vs vehicle. In order to infer the biological processes impacted by the different treatments, gene sets were classified in functional categories (Fig. 4A). The percentage of gene sets undergoing changes in activity with the same directionality in VM-CUB-1 and ScaBER cells by Nintedanib, Alpelisib, and their combination were 53.2%, and 55.5 %, respectively, reflecting heterogeneity in the response of the tumor xenografts. Notably, 2 gene sets associated with epithelial differentiation (keratinization) were the most up-regulated pathways in all treatment conditions when compared with vehicle-treated tumors.

Figure 4. The Nintedanib/Alpelisib combination down-regulates cell cycle and DNA repair pathways in tumor cells.

(A) Scatterplots show the normalized enrichment scores (NES) of each gene pathway emerging from the GSEA analysis of the ScaBER and VM-CUB-1 human transcriptomes for the indicated comparisons. Dot colors indicate the functional category of the pathway. Dots circumscribed in dotted line rectangles indicate a significant NES (FDR<0.05) either in VM-CUB-1 (black), ScaBER (blue), or both (red) (B) Plots display the NES for the ECM, cell cycle, and DNA repair pathways showing the same directionality in ScaBER and VM-CUB-1 human transcriptomes. Pathways colored in red correspond to a significant NES (FDR<0.05). (C) Enrichment map of the pathways included in cell cycle, DNA repair, and ECM categories for the Nintedanib/Alpelsib vs vehicle GSEA for VM-CUB-1 human transcriptome data. Blue nodes represent pathways with positive NES (FDR<0.05) and red nodes represent pathways with negative NES (FDR<0.05). Node size is proportional to the number of genes in the gene-set and edge thickness is proportional to the overlap among genesets.

Rho GTPase-related pathways were significantly down-regulated by Alpelisib in VM-CUB-1 tumors and the same trend was observed in ScaBER cells (Fig. 4A). Nintedanib-responder and combination-treated tumors showed significantly down-regulated DNA repair and cell cycle pathways, while there was an up-regulation of protein synthesis, RNA metabolism, and extracellular matrix (ECM) related pathways (Fig. 4A).

To evaluate broadly the transcriptomic changes in each specific treatment condition, we represented the NES of the consistently deregulated pathways in VM-CUB-1 and ScaBER xenografts for each group, including Nintedanib non-responders (Fig. 4B, Supplementary Table S4). GSEA showed an up-regulation of ECM and a down-regulation of cell cycle and DNA repair pathways in Nintedanib-responders and Nintedanib/Alpelisib-treated tumors; the effects were lesser in Alpelisib-treated and Nintedanib non-responder samples. These findings suggest that these processes are associated with the tumor cell response to Nintedanib/Alpelisib treatment in vivo and are partially recapitulated by Nintedanib monotherapy. Interestingly, despite the fact that the transcriptomic effects of the combination were greater that those induced by Nintedanib alone, ScaBER xenografts showed a similar growth rate upon treatment with Nintedanib or Nintedanib/Alpelisib.

To visualize the interconnection between the individual gene sets within each pathway category we used Cytoscape (Fig. 4C). ECM-related pathways are enriched in syndecan proteoglycans, integrin signaling, collagen metabolism, and ECM degradation. Down-regulated DNA repair gene sets are enriched in the homologous recombination pathway.

The Nintedanib/Alpelisib combination induces down-regulation of cell cycle and angiogenesis-related genes/pathways in the stroma

A similar approach was used to analyze mouse gene expression changes in stroma cells. Similar to the human transcriptome analysis, the percentage of mouse gene sets displaying concordant regulation in VM-CUB-1 and ScaBER by Nintedanib, Alpelisib and their combination were 63%, 54%, and 53%, respectively. Down-regulated gene sets in Nintedanib-responder and Nintedanib/Alpelisib-treated tumors were related to cell cycle, VEGFR, NOTCH, and Rho GTPase pathways. Immune and metabolic pathways were up-regulated (Fig. 5A). Alpelisib also induced the down-regulation of selected Rho GTPase related gene sets and the up-regulation of DNA repair gene sets.

Figure 5. The Nintedanib/Alpelisib combination down-regulates cell cycle gene expression and up-regulates immune and carbohydrate metabolism related genes in stromal cells.

(A) Scatterplots show NES for each pathway emerging from the GSEA analysis of the indicated comparisons in ScaBER and VM-CUB-1 mouse transcriptomes. Dot colors indicate the functional category of the pathway. Dots circumscribed in dotted line rectangles indicate a significant NES (FDR<0.05) either in VM-CUB-1 (black), ScaBER (blue), or both tumor models (red). (B) Plots display the NES for the metabolism, immune, cell cycle, and VEGFR and NOTCH pathways showing the same directionality in ScaBER and VM-CUB-1 mouse transcriptomes. Pathways colored in red correspond to a significant NES (FDR<0.05). (C) Enrichment map of the pathways included in the metabolism, immune, and cell cycle categories for the mouse Nintedanib/Alpelsib vs vehicle GSEA in VM-CUB-1 mouse transcriptomes. Blue nodes represent pathways with positive NES (FDR<0.05) and red nodes represent pathways with negative NES (FDR<0.05). Node size is proportional to the number of genes in the gene-set and edge thickness is proportional to the overlap among genesets.

We explored the enrichment of the above mentioned pathways in individual treatment groups (Fig. 5B, Supplementary Table S5 unlike in the human transcriptomic analyses, VM-CUB-1 and ScaBER differed in the significance of pathway enrichment. This suggests that Nintedanib, Alpelisib, and their combination have different impact on the stroma in different models, likely reflecting variable stromal cell composition. We also observed that the stromal transcriptomic changes induced by Nintedanib/Alpelisib were partially recapitulated in Nintedanib-treated tumors.

Stromal cell cycle genes were down-regulated by all treatments in both models, except for Alpelisib-treated ScaBER tumors which did not respond to this compound. Nevertheless, these pathways were only significantly down-regulated in Nintedanib- and Nintedanib/Alpelisib-treated VM-CUB-1 tumors. Finally, we observed a significant down-regulation of VEGFR and NOTCH pathways in Nintedanib- and Nintedanib/Alpelisib-treated VM-CUB-1 tumors (Fig. 5B, Supplementary Fig. S2C) suggesting an important role of angiogenesis and vasculogenesis (22,23).

All treatments led to an up-regulation of metabolic and immune pathways but the significance of the corresponding enrichment was only associated to response in ScaBER xenografts. Up-regulated metabolic gene sets related mainly to carbohydrate metabolism. Up-regulated immune pathways included those related to Interleukin and complement activation (Fig. 5C).

Endothelial cells and CAFs are targeted by the Nintedanib/Alpelisib combination

To assess the effects of Nintedanib/Alpelisib on the stroma, we used xCell, a software for gene signature-based digital cell type enrichment analysis across 64 immune and stromal cell types (20). Only 5 cell types were differentially enriched among the different treatment groups (one-way ANOVA test P<0.05) in VM-CUB-1 xenografts (Fig. 6A). Post-hoc pairwise comparisons showed that endothelial cell signatures were significantly down-regulated in samples from Nintedanib/Alpelisib- and Nintedanib-treated (both, responders and non-responders) mice, when compared with either vehicle- or Alpelisib-treated mice. Fibroblast gene signatures were enriched in samples from Nintedanib- and Nintedanib/Alpelisib- samples although the differences were not statistically significant (P=0.069). M2 macrophage signatures were significantly enriched in Nintedanib non-responders when compared with all the other groups.

Figure 6. The Nintedanib/Alpelisib combination inhibits tumor and stromal cell proliferation and targets CAFs and tumor angiogenesis.

(A) Heatmap illustrates xCell enrichment scores of signatures showing significantly different enrichment scores between treatment groups in VM-CUB-1 xenografts. (B) Normalized DAB signal of IHC staining of the indicated markers in different treatment groups from VM-CUB-1, SW1710, and SCaBER xenografts. Days of treatment is also indicated. The median is represented by the horizontal line within the density plot; *, P<0.05, **, P<0.01 two-tailed Mann-Whitney U-test. (C) Representative images of IHC analysis of the indicated antigens in VM-CUB-1 tumors treated for 13 days with vehicle, Nintedanib, Alpelisib, or their combination. Scale bar, 50 μm.

To validate these findings, we applied IHC using antibodies detecting SMA and CD31 to identify activated fibroblasts and endothelial cells, respectively, in biopsies from VM-CUB-1, SW1710, and SCaBER xenografts. The stromal compartment of these tumors is quite diverse: while VM-CUB1 and SCaBER xenografts display high SMA+ cell infiltration, these cells are almost absent in the SW1710 counterpart (Fig. 6B, C and Supplementary Fig. S2D). In VM-CUB-1 tumors, there were no significant differences in SMA-positive areas at day 7, but there was a significant reduction in samples from all treatment groups at day 13. In SCaBER tumors, the number of SMA-expressing cells was reduced only by the Nintedanib/Alpelisib combination. Regarding CD31, VM-CUB-1 xenografts are the only ones exhibiting high vessel density. In them, all treatments significantly diminished vascular density but only Nintedanib- and Nintedanib/Alpelisib-treated mice showed a significantly reduced vascular length density. The effects observed were greater when both drugs were combined (Fig. 6B, C).

Altogether, these findings suggest the Nintedanib/Alpelisib combination synergistically impacts on BC growth by affecting both tumor and stromal cells.

DISCUSSION

Here, we show that Nintedanib and the PI3K inhibitor Alpelisib cooperate to impair BC cell growth in vitro and in vivo. We also provide evidence on the molecular changes underlying these effects: PI3K inhibition sensitizes BC cells to Nintedanib and boosts Nintedanib-driven antitumor effects. The pleiotropic effects of Nintedanib on the different tumor microenvironments might account for the heterogeneous responses observed in our in vivo models when using Nintedanib as single agent therapy. Thus, the in vivo effects of Nintedanib are greater in highly vascularized, SMA+ infiltrated xenografts of intrinsically resistant VM-CUB-1 cells than in less vascularized and activated fibroblast-poor SW1710 tumors.

BC cells are largely intrinsically resistant to Nintedanib, regardless of their FGFR3 status. This is unlike other tumors harboring PDGFRA or FGFR2 genetic alterations - such as gastric, endometrial, breast and non-small cell lung cancer - where a stronger association with response is observed (24). Previous clinical trials using multi-RTK inhibitors similar to Nintedanib have demonstrated that their therapeutic effects might be independent from the FGFR status (25,26). However, more recent FGFR-selective inhibitors have demonstrated higher efficacy in tumors harboring FGFR genetic alterations (27,28).

The anti-proliferative and apoptotic effects of Nintedanib were accompanied by a transient down-regulation of MAPK and AKT signaling pathways (24,29). RTK crosstalk and PI3K/AKT reactivation mediate resistance to multiple RTKs, including FGFR inhibitors. Experimental and in silico analyses have revealed that EGFR family members (EGFR, ERBB2, and ERBB3) are central circuitry RTK network controllers (30) involved in the resistance to FGFR inhibitors in several tumors, including BC (31). These findings are confirmed here by in Nintedanib-treated SW1710 cells, possibly contributing to PI3K/AKT reactivation and resistance generation. Our data indicate that the Nintedanib/Alpelisib combination contributes to block pathway crosstalk, leading to an efficient inhibition of tumor cell proliferation.

VM-CUB-1 xenografts exhibited variable sensitivity to Nintedanib. Differences in drug delivery to the tumor, among others, may account for such variation although we did not find differences in tumor vascularity between Nintedanib-responders and non-responders (Fig 6B). The antitumor effect may be also influenced by the pre-existence, or the emergence, of cell subpopulations exhibiting drug tolerance, a phenomenon identified in other instances using single cell sequencing (32). Nintedanib non-responders displayed a MAPK/PI3K signaling activation pattern similar to that of vehicle-treated tumors. Larger sample size and single cell analyses might help to determine the mechanisms of tumor response to Nintedanib as a single-agent.

To identify mechanisms involved in the response to the drug combination, we analyzed the transcriptome of tumors. Nintedanib/Alpelisib induced a down-regulation of cell cycle genes and an up-regulation of ECM and immune-related pathways in the tumor cells. Importantly, these gene expression changes were partially recapitulated by Nintedanib monotherapy both in VM-CUB-1 and ScaBER xenografts. Further experiments might clarify the impact of these transcriptomic changes on tumor growth inhibition. In the stroma, Nintedanib and Nintedanib/Alpelisib induced an up-regulation of multiple genes involved in glucose metabolism, including pyruvate dehydrogenase kinase (PDK) 1 and 3. This kinase phosphorylates and inactivate PDH, an enzyme that links glycolysis to the activity of the tricarboxylic acid cycle. In addition, an up-regulation of lactate dehydrogenase A (LDHA) was observed. A metabolic shift towards enhanced glycolysis has been associated with resistance to Nintedanib and other tyrosine-kinase inhibitors in mouse breast cancer models (33), although these transcriptomic changes occurred at the tumor cell levels. Despite the fact that used athymic mice, Nintedanib and Alpelisib up-regulated the expression of several genes coding for pro-inflammatory cytokines (IL1, IL6 and IL12), complement, and related signaling molecules (JAK/STAT) which might contribute to the antitumor effects at the stromal cell level.

In addition, the transcriptomic analysis revealed - and IHC confirmed - effects on endothelial cells in VM-CUB-1 tumors, but not in SW1710 and ScaBER xenografts that exhibit reduced angiogenesis. Tumor vascularization can be induced through several mechanisms implying the formation of new vessels from pre-existing ones (angiogenesis) or de novo formation (vasculogenesis) (22). In VM-CUB-1 xenografts, there was reduced vascular density and vascular length density upon treatment with Nintedanib and with the Nintedanib/Alpelisib combination, indicating an effect not only on angiogenesis but also vasculogenesis. In agreement with our results, previous studies have shown an anti-angiogenic effect of Nintedanib on tumor xenografts (24,29,34) and its inhibition of retinal and corneal pathological neovascularization (35,36). PI3Kα is also a well-known regulator of endothelial and lymphatic cell function and PI3K inhibitors display anti-angiogenic activity (37,38). Therefore, the drug combination used here may have added value to broadly target this pathway. The transcriptomic data suggest that Nintedanib reduces tumor vascularization through the down-regulation of Notch, Dll4, and VEGFR family genes, which are central effectors of angiogenesis and vasculogenesis (22,35).

The therapeutic effects of Nintedanib in cancer and fibrosis have been associated with its ability to suppress the proliferation and activation of α-SMA-expressing hepatic stellate cells and α fibroblasts in lung adenocarcinoma, intrahepatic cholangiocarcinoma and melanoma (39–42). We now extend these observations to BC, a tumor in which the stroma has not been studied in-depth (43). Preclinical studies in melanoma models suggest that the inhibitory effect of Nintedanib on cancer fibroblasts attenuates the immunosuppressive effects of the tumor microenvironment, thus enhancing the effect of ICIs (42). A recent publication also shows that Nintedanib, in combination with a MEK inhibitor, reprograms the immunosuppressed landscape of pancreatic tumors in mice (44). A limitation of our study is the use of immunocompromised mice, where we could not evaluate the impact of Nintedanib/Alpelisib combination on the immune compartment. Moreover, the stroma of tumors implanted subcutaneously does not fully recapitulate the features of primary bladder tumors or metastases. The development of adequate mutant Fgfr3-driven mouse models of MI-BC should contribute to better investigate the effects of Nintedanib/Alpelisib on the tumor microenvironment.

Despite the evidence of genetic alterations in the RTK/MAPK pathway in 45% of MI-BC (6), specific or multi-target RTK inhibitors have shown little or no therapeutic efficacy in patients with metastatic BC (45–47). Our work provides rationale for combining two FDA-approved compounds, Nintedanib with Alpelisib in stroma-rich tumors. It is both intriguing and exciting that anti-tumor activity has recently been reported for Nintedanib in combination with chemotherapy in the neo-adjuvant setting (11) and for the Pazopanib/Everolimus combination in metastatic BC (48). These findings strongly point to an in-depth analysis of the tumor microenvironment as the disease progresses. This information will be crucial to optimally exploit the potential of chemotherapy, targeted therapies, and immune oncology drugs in BC.