A pan-cancer analysis of HER2 index revealed transcriptional pattern for precise selection of HER2-targeted therapy

Abstract

Background: The prevalence of HER2 alterations in pan-cancer indicates a broader range of application of HER2-targeted therapies; however, biomarkers for such therapies are still insufficient and limited to breast cancer and gastric cancer.

Methods: Using multi-omics data from The Cancer Genome Atlas (TCGA), the landscape of HER2 alterations was exhibited across 33 tumor types. A HER2 index was constructed using one-class logistic regression (OCLR). With the predictive value validated in GEO cohorts and pan-cancer cell lines, the index was then applied to evaluate the HER2-enriched expression pattern across TCGA pan-cancer types.

Findings: Increased HER2 somatic copy number alterations (SCNAs) could be divided into two patterns, focal- or arm-level. The expression-based HER2 index successfully distinguished the HER2-enriched subtype from the others and provided a stable and superior performance in predicting the response to HER2-targeted therapies both in breast tumor tissue and pan-cancer cell lines. With frequencies varying from 12.0% to 0.9%, tumors including head and neck squamous tumors, gastrointestinal tumors, bladder cancer, lung cancer and uterine tumors exhibited high HER2 indices together with HER2 amplification or overexpression, which may be more suitable for HER2-targeted therapies. The BLCA.3 and HNSC.Basal were the most distinguishable subtypes within bladder cancer and head and neck cancer respectively by HER2 index, implying their potential benefits from HER2-targeted therapies.

Interpretation: As a pan-cancer predictive biomarker of HER2-targeted therapies, the HER2 index could help identify potential candidates for such treatment in multiple tumor types by combining with HER2 multi-omics features. The discoveries of our study highlight the importance of incorporating transcriptional pattern into the assessment of HER2 status for better patient selection.

Funding: The National Key Research and Development Program of China; Clinical Research and Cultivation Project of Shanghai ShenKang Hospital Development Center.

Keywords: Biomarker; HER2-amplification; HER2-enriched subtype; Machine learning; Multi-omics analysis; Pan-cancer.

Fig. 1

Landscape of multi-omics HER2 status in pan-cancer. a. Multi-omics features of HER2 in 33 TCGA cancer types, including CNV, SNV, mRNA, protein and phosphor-protein of HER2. Each column represents a tumor sample, for CNV and SNV, colors indicate different HER2 status; for mRNA, protein and phospho-protein, heights of the bar stands for expression levels. Thresholds are marked with black lines through bars in which bars exceeding the line are regarded as HER2-overexpression. Cancer types are separated by dash lines and labeled at the top, in which HER2- targetable cancers are in bold. b. Frequencies of HER2 aberrations upon CNV, SNV, mRNA, protein and phosphor-protein levels in pan-cancer. c. Sankey plot of multi-omics HER2 status in pan-cancer. d. Multi-omics HER2 status of tumors with HER2 mutation in pan-cancer

Fig. 2

Genomic characteristics of tumors with increased HER2 SCNA in pan-cancer. a. Comparisons between PGAs of tumors with and without HER2-amplification in pan-cancer (left: PGAs of the whole genome; right: PGAs of Chr17). b. SCNA patterns of Chr17 in pan-cancer. X axis and y axis represent for gene location on the Chr17 and its amplification frequency, respectively. The upper line indicates for tumors with low-level HER2 amplification, and the bottom line indicates for tumors with high-level HER2 amplification. The color intensity of each dot on the upper line suggests for the association between low-level SCNA status of this specific gene with low-level HER2 SCNA status. While the color intensity of each dot on the bottom line suggests for the association between high-level SCNA status of this specific gene with high-level HER2 SCNA status. c. HER2 amplicon. X axis and y axis represent for gene location on the Chr17q and its amplification frequency in HER2-amplified tumors, respectively. Six genes including PGAP3, ERBB2(HER2), MIR4728, MIEN1, GRB7 and IKZF3 co-amplified in over 95% HER2-amplified tumors. d. Comparisons of 12 SCNA patterns of Chr17q22-23 in BRCA and STAD, indicating Chr17q22-23 is selectively amplified in gynecologic tumors.

Fig. 3

Transcriptomic heterogeneity of clinical HER2-positive BRCA. a. Fractions of four intrinsic subtypes of 75 clinical HER2-positive BRCA samples. Fraction of different HER2-CNV status in each subtype are displayed using horizontal bars. b. Comparisons of HER2 mRNA, protein and phosphor-protein among four intrinsic subtypes. c. Clustering heatmap using differential expressed pathways identified by GSVA. Subtype of each sample is annotated at the top. d. Consensus clustering of the pairwise correlation of HER2 pathway and 216 BIOCARTA pathways in HER2-enriched, Luminal A and Luminal B subtypes.

Fig. 4

Construction of HER2 index using machine learning. a. HER2 index of TCGA training cohort derived using HER2 signature. Each bar represents a single tumor sample in BRCA, with height standing for HER2 index and color indicating the corresponding subtype. The HER2 index was defined as Spearman correlation between mRNA expression matrix and weighted HER2 signature. The performance was evaluated via leave-one-out cross-validation, and the average AUC of TCGA training cohort was 0.979. b. Correlations between HER2 index and expressions of HER2 mRNA, protein and phosphor-protein (Spearman). c. fgsea. HER2 signature was compared with hallmark gene sets, C2 gene sets (KEGG, BIOCARTA and REACTOME pathways) and C5 gene sets (GO-BP level) from MSigDB.

Fig. 5

The predictive value of HER2 index over the response to HER2-targeted therapy. a. The HER2 index acted as the only significant predictor of pCR rate of trastuzumab-contained neoadjuvant treatment among 13 signatures tested via both univariate (OR= 97.862, p=0.017) and multivariate Cox regression analysis (OR = 92.768, p = 0.035) incorporated with ER status, age and tumor size (GSE50948, n = 63). b. Signatures including HER2 index (HR = 0.04,p = 4.9e-4 for univariate; HR = 0.029, p = 0.001 for multivariate), ERBB2(HR = 0.44, p = 3.5e-4 for univariate; HR = 0.45, p = 7.89e-4 for multivariate) and ESR1(HR = 3.29, p = 3.4e-4 for univariate; HR = 3.609, p = 3.46e-4 for multivariate), Rb.sig (HR = 0.57, p = 0.034 for univariate; HR = 0.475, p = 0.0165 for multivariate), Immune2(HR = 0.475, p = 0.0482 for multivariate), and T.cell (HR = 0.543,p = 0.0448 for multivariate) exhibited significant RFS hazard ratios in patients receiving trastuzumab-contained adjuvant regimen in univariate and/or multivariate logistic analysis combined with ER status, age, grade and T stage (GSE55348, n = 51). The PAM50 subtype was not included for assessment because no event of recurrence happened in the HER2-enriched subgroup. c. Pan-cancer cell lines scored high in HER2 index were more vulnerable to the knockout of HER2 gene (Spearman r = -0.37, p < 2.2e-16). d. The HER2 index exhibited a significantly negative correlation with IC50 of pan-HER inhibitors both in breast cancer cell lines and pan-cancer cell lines (Spearman). pCR, pathological complete response; RFS, recurrence-free survival; IC50, the half maximal inhibitory concentration.

Fig. 6

Evaluation of the similarity with HER2 expression pattern across pan-cancer tumor types and subtypes. a. Comparison of HER2 expression pattern assessed by HER2 index in 33 tumor types. HER2-CNV status is labeled in which red for “Amp”, green for “Del”, blue for “Neutral” and grey for missing data. The medians of the index are labeled as black dots. b. The multi-omics feature of HER2 in the samples with higher index in ten top scored HER2-aberrant tumor types. The cutoff for log10 (HER2 mRNA) (1.74), HER2 protein (0.274) and phosphor-protein (0.383) are labeled as dash lines in the three graphs respectively. c. ROC analysis. BLCA.3 and HNSC.Basal could be identified by classifiers using HER2 index.