FGFR4 regulates tumor subtype differentiation in luminal breast cancer and metastatic disease

Abstract

Mechanisms driving tumor progression from less aggressive subtypes to more aggressive states represent key targets for therapy. We identified a subset of luminal A primary breast tumors that give rise to HER2-enriched (HER2E) subtype metastases, but remain clinically HER2 negative (cHER2-). By testing the unique genetic and transcriptomic features of these cases, we developed the hypothesis that FGFR4 likely participates in this subtype switching. To evaluate this, we developed 2 FGFR4 genomic signatures using a patient-derived xenograft (PDX) model treated with an FGFR4 inhibitor, which inhibited PDX growth in vivo. Bulk tumor gene expression analysis and single-cell RNA sequencing demonstrated that the inhibition of FGFR4 signaling caused molecular switching. In the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) breast cancer cohort, FGFR4-induced and FGFR4-repressed signatures each predicted overall survival. Additionally, the FGFR4-induced signature was an independent prognostic factor beyond subtype and stage. Supervised analysis of 77 primary tumors with paired metastases revealed that the FGFR4-induced signature was significantly higher in luminal/ER+ tumor metastases compared with their primaries. Finally, multivariate analysis demonstrated that the FGFR4-induced signature also predicted site-specific metastasis for lung, liver, and brain, but not for bone or lymph nodes. These data identify a link between FGFR4-regulated genes and metastasis, suggesting treatment options for FGFR4-positive patients, whose high expression is not caused by mutation or amplification.

Keywords: Breast cancer; Genetics; Oncology.

Conflict of interest statement

Conflict of interest: CMP is an equity stock holder, consultant, and Board of Directors member of BioClassifier LLC and GeneCentric Therapeutics. CMP and JSP are also listed as inventors on patent applications for the Breast Cancer PAM50 Subtyping assay. AP is a consultant/advisory board member for NanoString Technologies. AL is on the Advisory Board and consults for Novartis, Pfizer, Roche/Genentech, Eisai, and Celgene, and as a clinician is involved in clinical research managed by the Instituto de Investigación Sanitaria (INCLIVA) of Amgen, Astra Zeneca, Boehringer-Ingelheim, GSK, Novartis, Pfizer, Roche/Genentech, Eisai, Celgene, and Pierre Fabre.

Figures

Figure 1. Comparative genetic and transcriptomic analysis of FGFR4 family members in TCGA.

(A) Box-and-whisker plots of FGFR family gene expression levels by mRNAseq according to molecular subtype and HER2 status by IHC. Tumors without clinical data for HER2 status and normal-like samples have been removed from the analysis, resulting in 1028 patients. Box-and-whisker plots display the median value on each bar, showing the lower and upper quartile range of the data and data outliers. The whiskers represent the interquartile range. Comparison between more than 2 groups was performed by ANOVA. Statistically significant values are highlighted in red. Each mark represents the value of a single sample. (B) Oncoprint diagram depicting high mRNA gene expression, DNA copy-number alterations, and mutations of FGFR family genes. *FGFR mRNA, high expression: high expression of genes was considered where levels exceeded the third quartile of positive values of gene expression (normalized, log2-transformed, and median-centered RNAseq data). Putative copy-number calls on 1077 cases were determined using GISTIC 2.0 (84). Deep deletion: –2 (homozygous deletion); amplification: 2 (high-level amplification). Mutation types are defined only as missense mutations (single base pair) or truncating mutations (multiple base pairs). HER2 clinical status was defined as previously described (85). LumA, luminal A; LumB, luminal B.

Figure 2. Testing the impact of FGFR4 on tumor growth and tumor differentiation in a HER2E/cHER2– PDX (WHIM11).

(A) Tumor growth and (B) tumor weight of WHIM11 tumors treated with BLU9931 (0.6 g/kg/day) or lapatinib (1.833 g/kg/day) for approximately 18 days (5 animals per treatment arm). Data represent mean tumor volume ± SEM. (C) Correlation to the PAM50 centroids (basal, HER2E, luminal A, luminal B) of mice treated with BLU9931 and untreated mice. (D) BLU9931-treated, lapatinib-treated, or untreated tumors evaluated for a luminal tumor score (LTS) tested along with TGCA cohort grouped according to intrinsic subtype. (E and F) Supervised hierarchical cluster across 1100 breast tumors and 98 normal samples from TCGA data set using FGFR4-related signatures. Significantly upregulated genes defined as an FGFR4-repressed signature (E) and significantly downregulated genes defined as a FGFR4-induced signature (F). Each gene set or subcluster has been selected based on a node correlation greater than 0.5 and named according to the important biological process governed by it as identified by Gene Ontology. Comparison between more than 2 groups was performed by ANOVA with post hoc Tukey’s test: *P < 0.05; **P < 0.01; ***P < 0.001. Statistically significant values are highlighted in red. Box-and-whisker plots display the median value on each bar, showing the lower and upper quartile range of the data. The whiskers represent the interquartile range.

Figure 3. Testing FGFR4-associated signatures in genetically engineered or chemically inhibited breast cancer cell lines.

(A) Box-and-whisker plots depicting the Euclidean distance of each group of cell lines to the UNC337 tumor-intrinsic subtypes. Lower Euclidean distance suggests higher similarity to the subtype call. (B) Analysis of luminal tumor score (LTS) in each cell line tested along with TCGA grouped according to intrinsic subtype. (C) Average expression (signature score) of FGFR4-induced signature and FGFR4-repressed signature of each cell line described in A and B. Comparisons between 2 groups were performed by 2-tailed t test. Comparison between more than 2 groups was performed by ANOVA. Box-and-whisker plots display the median value on each bar, showing the lower and upper quartile range of the data and data outliers. The whiskers represent the interquartile range. Each mark represents the value of a single sample. Statistically significant values are highlighted in red.

Figure 4. Single-cell RNA sequencing of WHIM11 tumor treated with BLU9931 and drug released.

(A) Left panel: Uniform manifold approximation and projection (UMAP) plot of all combined cells (30,058 cells in total) that passed quality checks in untreated WHIM11 (n = 2; 9298 cells), treated with BLU9931 (n = 2; 9777 cells) (0.6 g/kg/day) for 14 days, and treated but released of drug for 18 days (n = 2; 10,983 cells). Cells and clusters are color coded by each cell population found. Right panel: Heatmap of significantly differentially expressed genes and main Gene Ontology annotations for each cluster in WHIM11 based on MSigDB. Significant genes were identified by Wilcoxon’s signed-rank t test. (B) UMAP plot of all the cells that passed quality checks in WHIM11 and divided by each experimental condition as untreated, treated with BLU9931 (0.6 g/kg/day) for 14 days, and treated but released of drug for 18 days. Cells and clusters are color coded by each cell population found. (D, F, and H) UMAP plots showing the expression of FGFR4, ERBB2, and ESR1 genes across all WHIM11 clusters in WHIM11 untreated, treated with BLU9931 (0.6 g/kg/day) for 14 days, and treated but released of drug for 18 days. (C, E, G, and I) UMAP plots showing the gene signature score (average values of genes present in each score respectively for each cell) of FGFR4-induced signature, FGFR4-repressed signature, proliferation signature, and luminal tumor score (LTS) across all WHIM11 clusters in WHIM11 untreated, treated with BLU9931 (0.6 g/kg/day) for 14 days, and treated but released of drug for 18 days.

Figure 5. Prognostic value of FGFR4-derived signatures.

(A) Average expression of FGFR4-related signatures in TCGA tumor molecular subtypes. Normal-like patients and true-normal tissues have been removed from the analysis. Statistical differences were calculated by ANOVA test. (B) Scatterplot showing the correlation between FGFR4-related signatures and luminal tumor score (LTS) (as calculated in TCGA data set using only HER2E, luminal A, and luminal B tumors). Correlation was measured using the Pearson correlation coefficient. (C) Average expression of FGFR4-related signatures depending on histological tumor grade in METABRIC data (Grade 1: Low grade or well differentiated; Grade 2: Intermediate grade or moderately differentiated; Grade 3: High grade or poorly differentiated). (D and E) Kaplan-Meier plots to test the prognostic ability of FGFR4 signatures in METABRIC (D) and MDACC (E) data sets (normal-like samples were removed from the analysis in both cohorts). Survival curve differences were calculated by the log-rank test and the estimates of survival probabilities and cumulative hazard with a univariate Cox proportional hazards model. (F) Multivariable Cox proportional hazards analyses using METABRIC data (normal-like samples were removed from the analysis). Hazard ratio (HR) = 1: no effect. HR < 1: reduction in hazard. HR > 1: increase in hazard. Signatures were evaluated as continuous variables and rank ordered according to the gene FGFR4 signature scores (induced and repressed) in 3 different levels: low, medium, and high (assigned by distribution in a given upper, middle, or lower tertile). Comparison between more than 2 groups was performed by ANOVA. Statistically significant values are highlighted in red. Box-and-whisker plots display the median value on each bar, showing the lower and upper quartile range of the data and data outliers. The whiskers represent the interquartile range. Each mark represents the value of a single sample. LumA, luminal A; LumB, luminal B.

Figure 6. Expression profile and FGFR4-associated signatures in matched breast primary tumors and metastases.

(A) Luminal tumor score (LTS) calculated values in 77 primary and 77 paired metastatic tumors in the luminal set. (B) Expression levels of FGFR4-induced (left) and FGFR4-repressed signatures (right) in 77 primary and 77 paired metastatic luminal tumors. (C) Proliferation scores for FGFR4-induced and -repressed signatures (see PAM50 subtype classification in Methods) in 77 primary and 77 paired metastatic luminal tumors. (D) Supervised hierarchical cluster analysis derived from the significantly different modules scores, in luminal and (E) basal sets. Significance of the differences between modules was calculated using 2-class SAM at an FDR of 0%. Significantly up- and downregulated modules are clustered together for each set (basal and luminal). Clusters (4 from luminal and 5 from basal set) were selected based on node correlation greater than 0.5. Comparisons between 2 paired groups were performed by paired, 2-tailed t test. Statistically significant values are highlighted in red. Box-and-whisker plots display the median value on each bar, showing the lower and upper quartile range of the data and data outliers. The whiskers represent the interquartile range. Each mark represents the value of a single sample. LumA, luminal A; LumB, luminal B; PRIM, primary tumor; MET, metastatic tumor.

Figure 7. Univariate analysis in 855 primary tumors with known first site of relapse.

Kaplan-Meier and Cox proportional hazards model analyses of metastasis-free survival (MFS) analyzed in their specific sites of relapse: brain (A), lymph node (B), lung (C), bone (D), liver (E), or any site of relapse (F). Survival curve differences were calculated by the log-rank test and the estimates of survival probabilities and cumulative hazards with a univariate Cox proportional hazards model. FGFR4-derived signatures were evaluated as continuous variables and rank ordered according to the gene FGFR4 signature scores (induced and repressed) in 3 different levels: low, medium, and high (assigned by distribution in a given upper, middle, or lower tertile). Hazard ratio (HR) = 1: no effect. HR < 1: reduction in the hazard. HR > 1: increase in hazard. Statistically significant values are highlighted in red.