The Pan-Tumor Landscape of Gene Amplifications and Copy Number Amplification Ratio for Established and Emerging Clinical Targets

Abstract

Purpose:

Gene copy number (CN) amplifications and protein overexpression are common drug targets, and detection relies on various methodologies, including next-generation sequencing–based CN, immunohistochemistry (IHC), and in situ hybridization (ISH). We investigated the pan-tumor landscape of amplifications and developed AmpRatio, a novel method of CN quantitation.

Experimental Design:

Pan-tumor tissue (N = 486,340) and liquid (N = 85,635) samples underwent hybrid capture–based comprehensive genomic profiling. A genome-wide CN model for each sample was generated to estimate the purity, ploidy, and segment-level CN. AmpRatio was calculated by dividing gene CN/sample ploidy. A US-based deidentified clinicogenomic database was utilized to assess the relationship between ERBB2 AmpRatio and HER2 IHC/FISH and outcomes on anti-HER2 therapies.

Results:

Amplifications with varying degrees of gain were reported in 38.6% of pan-tumor tissue samples, most frequently MYC (5.6%), 11q13 (5.2%), ERBB2 (5.2%), and CCNE1 (3.2%). ERBB2 AmpRatio was associated with HER2 positivity by IHC/FISH in gastroesophageal [overall percent agreement (OPA) 90%] and breast (OPA 95%) cancers. Among patients treated with anti-HER2 therapies, ERBB2 AmpRatio significantly stratified outcomes within the ERBB2-amplified and IHC-defined HER2+ and HER2-low/ultralow populations. High concordance (sensitivity 88%) of amplification detection in liquid biopsy versus tissue was associated with higher AmpRatio and ctDNA tumor fraction ≥20%.

Conclusions:

CN amplifications are prevalent and diverse biomarkers, and AmpRatio is variable across genes and tumor types. ERBB2 AmpRatio is associated with outcomes to HER2-directed therapies and may have utility alongside IHC for clinical decision-making. With the increasing number of therapies targeting amplifications/overexpression, it will be important to define harmonized methods for CN quantification for optimal patient selection.

Translational Relevance.

Copy-number (CN) amplifications and protein overexpression are a prevalent class of oncogenic drivers associated with sensitivity to a growing number of approved and investigational therapies. Various methodologies are used for patient selection, including next-generation sequencing–based CN, in situ hybridization (ISH), and immunohistochemistry (IHC), but the relationship between these metrics is not well understood. Our study assesses the landscape of CN amplifications in genes associated with approved and investigational agents and quantified the degree of amplification using AmpRatio, a novel normalized CN metric that accounts for the broader genomic context. We assessed the association between ERBB2 AmpRatio and HER2 positivity by IHC/ISH and its utility in stratifying outcomes on anti-HER2 regimens independently and within IHC/ISH-defined HER2 subsets. These data suggest that AmpRatio may have added utility along with amplification and IHC/ISH-defined HER2 status in refining the definition of clinically relevant gains and serve as an additional variable for consideration in trials and clinical practice.

Introduction

Gene copy number (CN) amplifications and protein overexpression represent established and emerging therapeutic targets across tumor types. The monoclonal antibody (mAb) trastuzumab was approved in 1998 for the treatment of metastatic breast cancers (mBC) that overexpress HER2 protein as detected by immunohistochemistry (IHC), and recent evidence from the DESTINY trials has led to the approval of the antibody–drug conjugate (ADC) trastuzumab deruxtecan (T-DXd) for all previously treated, unresectable, or metastatic HER2+ (IHC 3+) solid tumors (1, 2). There is significant interest in using MET tyrosine kinase inhibitors (TKI), mAbs, and ADCs to target MET amplification as a primary driver and as an acquired resistance alteration in non–small cell lung cancer (NSCLC), including the recent approval of telisotuzumab vedotin, a MET-directed ADC for patients with NSCLC with high MET protein overexpression (IHC 3+; refs. 3–10). There are also several ongoing trials assessing the efficacy of therapeutics targeting CCNE1, EGFR, FGFR2, and MYC amplifications across tumor types (11–17). As therapies are developed for more varied targets, multiplexed testing using next-generation sequencing (NGS) may become increasingly cost-effective and necessary because of tissue scarcity in certain tumor types.

Several studies have shown a correlation between the degree of amplification or overexpression and treatment response. In NSCLC, higher response rates to MET TKIs are associated with higher CN focal MET amplifications (3, 4, 18). The magnitude of ERBB2 amplifications has also been shown to correlate with response to trastuzumab in breast, colorectal (CRC) and gastroesophageal (GE) cancers (19–23). In addition to the pan-tumor approval of T-DXd for all HER2+ (IHC 3+) solid tumor patients, the DESTINY-Breast04 and DESTINY-Breast06 trials led to the approval of T-DXd for patients with HER2-low [IHC 1+ or IHC 2+/in situ hybridization (ISH−] or HER2-ultralow (IHC 0; 0%–10% weak or incomplete membrane staining) advanced breast cancer, demonstrating the potential clinical utility of detecting low-level gains in addition to high-level amplifications (24). However, emerging evidence suggests that there may still be a role for NGS to stratify responders within HER2 IHC subsets of breast cancer (25).

Multiple methodologies can be utilized to detect CN amplification or protein overexpression including IHC, ISH, and NGS. Guidelines for HER2 testing in breast and GE cancers currently recommend using IHC to determine HER2 positivity with reflex to ISH to resolve equivocal IHC 2+ cases, and there is an approved NGS-based companion diagnostic (CDx) for ERBB2 amplifications in breast cancer (26–29). Although there is some evidence demonstrating an association between ERBB2 CN and response to trastuzumab, the difference between quantitative CN metrics reported by different NGS assays, and the relationship between these values and IHC/ISH, remain largely unresolved (30, 31). We sought to address the variability in CN amplification quantitation from NGS by developing a quantitative, normalized CN metric that accounts for the broader genomic context resulting in a more robust, reproducible estimate of gene CN, a metric we term AmpRatio. In this study, we characterized the pan-cancer landscape of CN amplifications as detected by an FDA-approved tissue NGS assay and quantified the degree of amplification with AmpRatio. We also examined the association between ERBB2 AmpRatio and IHC/ISH-based HER2 status, and its association with outcomes on HER2-directed therapies. Finally, due to the known challenges of detecting CNAs in liquid biopsy (32), we assessed the sensitivity of amplification detection by liquid biopsy in paired tissue and liquid samples from 2,440 patients.

Materials and Methods

Study design and patient selection

Comprehensive genomic profiling (CGP) results from 486,340 tissue and 85,635 liquid pan-tumor samples in the Foundation Medicine (FMI) genomic database were assessed for gene CN amplifications. This study also used the US-based deidentified Flatiron Health–FMI breast, GE, and CRC Clinico-Genomic Database (FH-FMI CGDB), in which clinical data from the FH Research Database (33) were linked to genomic data derived from FMI’s CGP tests by deidentified, deterministic matching (34). This cohort represents a subset of the FMI genomic database and included 3,389 patients with advanced GE (aGE) cancer, 6,388 patients with metastatic breast cancer (mBC), and 10,154 patients with metastatic CRC (mCRC) who underwent FMI CGP testing (FoundationOne® or FoundationOne®CDx) and received care in the FH network between 01/2011 and 09/2024. Patients who received their FMI CGP report >60 days after their last FH visit date were excluded to ensure that patients remained in the FH network at the time of CGP.

CGP

For tissue biopsy samples, DNA was extracted from formalin-fixed, paraffin-embedded sections, and CGP (FoundationOne® or FoundationOne®CDx) was performed on hybridization-captured, adapter ligation-based libraries for a mean coverage depth of >550× for 315 or 324 cancer-related genes, respectively, including selected introns from 28 or 36 genes frequently rearranged in cancer (35, 36). For liquid biopsy samples, cell-free DNA was extracted from blood plasma to create adapted sequencing libraries before hybrid capture and sample-multiplexed sequencing to a median unique exon coverage of >6000× for 324 genes (FoundationOne®Liquid CDx; ref. 37). Testing was performed in a Clinical Laboratory Improvement Amendments–certified/College of American Pathologists-accredited laboratory (Foundation Medicine, Inc.). Results were analyzed for short variants (SV; base substitutions, short insertions, and deletions), CN alterations (CNAs; amplifications and homozygous losses), and rearrangements (35, 36). Prevalence of reported amplifications and co-occurring genomic alterations were restricted to alterations designed as known or likely to be pathogenic, and alterations characterized as variants of unknown significance were excluded. Pathogenic/likely pathogenic alterations were called using a multi-step method leveraging annotations which include reporting in COSMIC (RRID: SCR_002260), functional knowledge of the gene affected, internal insights, and clinical/functional characterization in the literature, as described previously (35, 38). Tumor mutational burden (TMB) was calculated across a 0.8 to 1.2 megabase region by counting the number of somatic non-driver synonymous/non-synonymous mutations, a method previously validated against whole-exome sequencing (39). The presence of homologous recombination deficiency (HRD) was assessed using a scar-based HRD signature (HRDsig) algorithm which leverages a broad set of CN features to examine genome-wide patterns associated with HRD and is offered as a laboratory professional service (40). Genomic ancestry was predicted by a classifier utilizing single-nucleotide polymorphisms (SNP) targeted by the CGP assay and trained on data from the 1000 Genomes Project (RRID: SCR_006828) to classify patients as belonging to one of the following subpopulations: African, East Asian, European, South Asian, and admixed American (AMR; ref. 41). For liquid samples, circulating tumor DNA (ctDNA) tumor fraction (TF), offered as a laboratory professional service, was calculated using a composite algorithm that incorporates aneuploidy, variant allele frequency, fragment length information, clonal hematopoiesis predictions, and known tumor-associated alterations detected on FoundationOne®Liquid CDx (42, 43). Clinical features (e.g., age, sex, and cancer diagnosis) were extracted from test requisition forms and pathology reports.

Gene CN determination

To determine gene CN, a genome-wide CN model for each sample was generated by normalizing the sequence coverage obtained at all targeted regions (exons, introns, and SNPs) by a set of reference samples (panel of normals). A genome-wide log₂-ratio coverage profile was generated and combined with the allele frequencies of targeted heterozygous SNPs to produce an overall segmented profile and estimated CN model. This model consists of the sample purity, ploidy, and a CN for each genomic segment. The purity and ploidy were defined as the fraction of tumor cells in the biopsy and the average CN, respectively. AmpRatio, currently offered as a laboratory professional service, is the ratio of the gene CN to the sample ploidy and is calculated: modeled gene CN/sample ploidy. This metric maintains CN-model invariance across sample purities and allows for a more robust and reproducible estimation of gene CN (Supplementary Figs. S1 and S2). AmpRatio values expectedly trend with modeled gene CN, in which most amplifications with modeled CN ≥ 11 have AmpRatio ≥3 (Supplementary Fig. S3).

For ERBB2, amplifications were defined as CNAs with a modeled CN ≥ median CN + 3 in tissue and liquid samples for all tumor types, in alignment with the approved CDx label in breast cancer (29). For all other genes, amplifications were defined as CNAs with AmpRatio ≥3 for tissue samples and modeled CN ≥ median CN + 4 for liquid samples, in alignment with current clinical reporting.

IHC/FISH annotation

HER2 IHC and ERBB2 FISH results were abstracted from the electronic health records and annotated in the mBC and aGE CGDBs. HER2/ERBB2 status was abstracted from unstructured text leveraging technology-enabled human abstraction and extraction using natural language processing with or without machine learning (44–46). Quantitative FISH ratios were not abstracted, and FISH results were reported as amplification-positive/negative. HER2 IHC status was abstracted as IHC 0, 1+, 2+, or 3+. Comparisons between AmpRatio and abstracted HER2 IHC and FISH were restricted to tests performed on biopsies from the same day.

Statistical analysis

All analyses made use of R (version 4.4.0; RRID: SCR_001905). The Fisher's exact test was used for the comparison of discrete variables, and the Kruskal–Wallis test was used for the comparison of continuous variables. FDR correction for multiple hypothesis testing was performed with the Benjamini–Hochberg procedure. The association between gene AmpRatio and coalteration prevalence was evaluated with successive logistic regressions assessing the relationship between coalteration status for each gene/alteration type and AmpRatio as a continuous variable. All statistical tests were two-sided and performed using P < 0.05 as the threshold for statistical significance.

Differences in time-to-event outcomes were assessed with the log-rank test and Cox proportional hazard models. To evaluate the association between the degree of amplification and outcomes, patients were stratified into AmpRatio tertiles. Real-world overall survival (rwOS) was defined as the time from treatment start to date of death. Patients with no record of mortality were right-censored at their last date of confirmed activity. A patient’s entry date was defined as the later of the date of the patient’s second FH network visit or their first eligible FMI CGP report, and overall survival risk intervals were left truncated to the entry date to account for immortal time. Real-world progression-free survival (rwPFS) was calculated from the treatment start date to the time of disease progression or death, and patients with no record of progression or mortality were right-censored at their last clinic note date. For survival analysis and visualization, the survival (RRID: SCR_021137) and survminer (RRID: SCR_021094) R packages were used.

Institutional Review Board approval

For the FMI genomic analysis, approval for this retrospective study, including a waiver of informed consent and a Health Insurance Portability and Accountability Act (HIPAA) waiver of authorization, was obtained from the WCG Institutional Review Board (IRB; Protocol No. 20152817). The CGDB is a deidentified database, in which data are linked by an independent third party. For the CGDB analysis, Institutional Review Board approval of the study protocol was obtained prior to study conduct and included a waiver of informed consent based on the observational, noninterventional nature of the study (WCG IRB, Protocol No. 1342690). Studies were conducted in accordance with the Declaration of Helsinki.

Results

Pan-tumor landscape of gene CN amplifications

Of 486,340 pan-tumor tissue samples, CN amplifications were reported in 187,514 (38.6%) samples (Table 1; Supplementary Fig. S4). Amplifications were most prevalent in GE, breast, and gallbladder cancers, detected in 68.1%, 59.1%, and 54.0% of tissue samples, respectively (Fig. 1A). Across all tumor types, the most frequently amplified genes were MYC (5.6%), the 11q13 amplicon (CCND1/FGF19/FGF3/FGF4; 5.2%), ERBB2 (5.2%), CCNE1 (3.2%), FGFR1 (3.2%), and EGFR (2.8%). MDM2 and MET amplifications were detected in 2.7% and 1.0% of pan-tumor tissue samples, respectively. CN amplifications were the predominant gain-of-function (GOF) alteration type for most genes, except for EGFR, KRAS, and PIK3CA, in which SVs were more common (Fig. 1B). Although amplicons such as MYC, 11q13, and ERBB2 were prevalent across tumor types, others had notable disease enrichments, including EGFR (24.8%), CDK4 (10%), and PDGFRA (6.7%) in glioma, ZNF703, FGFR1, and WHSC1L1 in breast (9.8%–13.8%), KRAS in GE (12.0%), and AR in prostate (10.8%; Fig. 1C). MDM2 amplifications were also detected across cancer types, occurring without an inactivating TP53 alteration [TP53-wildtype (wt)] in 75% of cases. TP53-wt MDM2 amplifications were more highly amplified than amplifications co-occurring with a TP53 alteration (median AmpRatio 9.1 vs. 4.8, P < 0.0001), and disease-specific co-alteration enrichments were also observed (Supplementary Figs. S5 and S6).

Table 1.

Demographics and clinical characteristics of pan-tumor patients harboring tissue CN amplifications in the FMI genomic database.

Characteristic	All amp+ cases	CCNE1	EGFR	ERBB2	MET
N	187,514	15,616	13,438	25,159	5,039
Sex, n (%)	​	​	​	​	​
Female	106,863 (57)	10,756 (68.9)	6,243 (46.5)	14,875 (59.1)	2,036 (40.4)
Male	80,650 (43)	4,860 (31.1)	7,195 (53.5)	10,283 (40.9)	3,002 (59.6)
Median age (range)	65 (9, 88)	66 (6, 88)	64 (4, 88)	64 (0, 88)	65 (4, 88)
Ancestry, n (%)	​	​	​	​	​
AFR	21,486 (11.5)	2,211 (14.2)	1,189 (8.8)	3,011 (12)	517 (10.3)
AMR	15,507 (8.3)	1,334 (8.5)	1,086 (8.1)	2,227 (8.9)	403 (8)
EAS	6,766 (3.6)	518 (3.3)	672 (5)	1,038 (4.1)	233 (4.6)
EUR	141,799 (75.6)	11,361 (72.8)	10,316 (76.8)	18,612 (74)	3,839 (76.2)
SAS	1,956 (1)	192 (1.2)	175 (1.3)	271 (1.1)	47 (0.9)
Median TMB (range)	3.8 (0, 1,110.1)	3.6 (0, 603.8)	3.6 (0, 486.3)	3.8 (0, 1,110.1)	5 (0, 284.5)
Median AmpRatio (range)	4.8 (1.3, 468.1)	5.3 (3.8, 8.9)	13.4 (5.7, 30.5)	3.8 (2.1, 14.2)	5.9 (4, 10.3)
Median sample ploidy (range)	3 (1, 10)	3 (1, 8)	3 (1, 8)	3 (1, 9)	3 (2, 8)

Demographic and clinical characteristics for all amplification-positive (amp+) tissue samples and samples positive for amplifications associated with approved and/or investigational agents highlighted in this study.

Abbreviations: AFR, African; AMR, Admixed American; EAS, East Asian; EUR, European; SAS, Southeast Asian.

Figure 1.

Pan-tumor amplification landscape. A, Prevalence of reported tissue gene amplifications (CN ≥ median CN + 3 for ERBB2 and AmpRatio ≥3 for all other genes) across tumor types. Diseases with ≥2000 tissue samples were included. B, Distribution of GOF variant types (top), prevalence (middle), and AmpRatio (bottom) for the 25 most prevalent amplicons across pan-tumor tissue samples. Genes grouped together in an amplicon were restricted to amplicons for which each gene co-occurred with all others in ≥90% of cases. Prevalence and AmpRatio were reported for the first gene listed in each amplicon. The dashed yellow line denotes AmpRatio = 3, the threshold for amplification for all genes except ERBB2. C, Prevalence of reported tissue amplifications by disease for tumor types with ≥2000 cases. Genes included were the subset with approved or investigational therapies targeting amplification or overexpression highlighted in this study, or those with ≥5% prevalence in a focus tumor type. Alts, alterations; CCA, cholangiocarcinoma; CNS, central nervous system; CRC, colorectal cancer.

The median pan-tumor sample ploidy was 2.9 (IQR 2.0–3.5), and bladder, prostate, NSCLC, colorectal cancer, and breast cancers were frequently triploid, whereas gastrointestinal stromal tumor, thyroid, and central nervous system tumors were largely modeled as diploid (Supplementary Fig. S7). The median AmpRatio of all amplifications pan-tumor was 4.8 and was relatively consistent across genes, with notably higher degrees of amplification in EGFR (13.4), CDK4 (8), MDM2 (7.5), and KRAS (6.8; Fig. 1B). Shorter CN segments were also associated with higher AmpRatios (P < 0.0001), and CDK4, ERBB2, and EGFR amplifications tended to be part of smaller, more focal amplicons (Supplementary Figs. S8 and S9).

ERBB2 amplifications are recurrent with varying degrees of gain across cancer types

ERBB2 amplifications were detected in 5.2% of pan-tumor tissue samples and were most prevalent in GE (16.7%), gallbladder (16.0%), salivary gland (12.1%), bladder (11.6%), uterine (10.6%), and breast cancers (9.7%). In most disease types, the majority of GOF ERBB2 alterations were CN amplifications. However, ERBB2 amplifications in small intestine and cervical cancer comprised <50% of pathogenic ERBB2 alterations and ERBB2 activation in these diseases was primarily driven by SVs. The median AmpRatio of ERBB2 amplifications pan-tumor was 3.8. Salivary gland tumors had the highest median ERBB2 AmpRatio (12.8), followed by GE cancer (7.1), breast cancer (7.0), and colorectal cancer (6.6), whereas neuroendocrine tumors (1.8), kidney cancer (2.2), and NSCLC (2.3) tended to have lower levels of ERBB2 gain (Fig. 2A; Supplementary Fig. S10).

Figure 2.

ERBB2 amplification landscape. A, Distribution of GOF alteration types (top), prevalence (middle), and AmpRatio (bottom) for ERBB2 amplifications (CN ≥ median CN + 3) across tumor types. The dashed yellow line denotes AmpRatio = 3, the threshold for amplification for all genes except ERBB2. B, rwOS for 167 patients with ERBB2-amplified mBC treated with 1L trastuzumab-based regimens stratified by ERBB2 AmpRatio tertiles. C, rwOS for 167 patients with ERBB2-amplified aGE cancer treated with 1L trastuzumab–based regimens stratified by ERBB2 AmpRatio tertiles. D, rwOS for 55 ERBB2-amplified patients with mCRC treated with trastuzumab-based regimens stratified by ERBB2 AmpRatio tertiles. Alts, alterations; CCA, cholangiocarcinoma; CNS, central nervous system; CRC, colorectal cancer.

Evaluating the co-alteration landscape relative to ERBB2 AmpRatio across all ERBB2 CN segments in breast, CRC, and GE cancers, ERBB2 and CDK12 rearrangements and TP53 SVs were associated with increasing ERBB2 AmpRatio (all P < 0.0001) in all three diseases (Supplementary Figs. S11–S13). In breast cancer, ERBB2 AmpRatio was inversely associated with the presence of HRDsig and AKT1, CDH1, ESR1, PTEN, and BRCA1 SVs (all P < 0.0001). In GE cancer, increasing ERBB2 AmpRatio was inversely associated with KRAS and PIK3CA SVs and EGFR, KRAS, MDM2, PTEN, and MET CNAs (all P < 0.0001). In CRC, increasing ERBB2 AmpRatio was associated with ERBB2 SVs (P < 0.0001), whereas MAPK and PI3K pathway alterations (KRAS, NRAS, BRAF, NF1, and PIK3CA) and TMB ≥10 were associated with lower ERBB2 AmpRatio (all P < 0.01).

Higher ERBB2 AmpRatio was associated with response to HER2-directed therapies

The impact of the degree of ERBB2 amplification on outcomes to anti-HER2 regimens was assessed for 812 patients with mBC, 256 patients with aGE cancer, and 70 patients with mCRC in the CGDB (Supplementary Fig. S14–S16). We first assessed the ability of AmpRatio to stratify response in patients with ERBB2amp+ cancers. Among 167 patients with ERBB2amp+ mBC who received a first-line (1L) trastuzumab-based regimen (Supplementary Table S1), we observed a positive association between ERBB2 AmpRatio and rwOS (30.1, 57.3, and 59.8 months) and rwPFS (8.8, 15.9, and 11.6 months) from first to third AmpRatio tertiles. Patients with amplifications in the second and third tertiles had significantly longer rwOS compared with the first tertile (both P < 0.01), but the largest increase in rwOS was observed between the first and second tertiles, with a nonsignificant difference between the second and third tertiles (P = 0.93; Fig. 2B; Supplementary Fig. S17A). When evaluating a range of exploratory ERBB2 AmpRatio cutoffs, the minimum hazard ratio (HR) for rwOS was observed at AmpRatio = 5 and began to increase at thresholds ≥ AmpRatio 10, although the 95% confidence intervals (CI) across the tested thresholds were overlapping with no clear optimal cutoff that best dichotomized outcomes (Supplementary Fig. S18A and S18B).

A positive association between AmpRatio and rwOS was also observed for 135 patients with ERBB2amp+ mBC treated with T-DXd (Supplementary Table S2), and similarly in this cohort, the largest increase in both rwOS (17.1 vs. 45.0 months, P = 0.049) and rwPFS (8.3 vs. 15.9 months, P = 0.07) was observed between the first and second tertiles, with nonsignificant differences in rwOS (45 vs. 41.2 months, P = 0.42) and rwPFS (15.9 vs. 14.9 months, P = 0.74) between the second and third tertiles (Supplementary Fig. 19). In evaluating exploratory AmpRatio cutoffs, across AmpRatios 2 to 5, the HRs for rwOS and rwPFS remained fairly consistent and began to increase above AmpRatio 5, although the CIs were similarly overlapping (Supplementary Fig. 20).

Among 167 patients with ERBB2amp+ aGE cancer who received a 1L trastuzumab–based regimen (Supplementary Table S3), we also observed a positive association between ERBB2 AmpRatio and rwOS (10.9, 13.1, and 22.4 months) and rwPFS (6.7, 8.3, and 12.2 months) across AmpRatio tertiles. Patients with ERBB2 amplifications in the third AmpRatio tertiles had significantly longer rwOS and rwPFS (both P < 0.0001) than patients in the first tertile (Fig. 2C; Supplementary Fig. S17B). Evaluating a range of exploratory AmpRatio cutoffs, a statistically significant difference in outcome was observed for most cutoffs evaluated, with consistently decreasing HRs, indicating a greater survival benefit, with increasingly higher-level amplifications (Supplementary Fig. S18C and S18D).

Among 55 patients with ERBB2amp+ mCRC treated with trastuzumab-based regimens in any line (Supplementary Table S4), we also observed longer rwOS (12.3, 23, and 28.5 months) and rwPFS (3.2, 5.5, and 7.6 months) from first to third ERBB2 AmpRatio tertiles with significantly longer rwOS and rwPFS (both P = 0.03) for patients in the third versus first tertile (Fig. 2D; Supplementary Fig. 17C), and exploratory analysis suggested similar predictive value for AmpRatio cutoffs between 3 and 30, though this was limited by small cohort sizes (Supplementary Fig. S18E and S18F).

ERBB2 AmpRatio positively correlates with HER2 IHC/ISH classification and patient outcomes

The association between ERBB2 AmpRatio and HER2 status as defined by IHC/FISH was assessed for 3,650 patients with mBC with ERBB2 NGS and abstracted HER2 IHC/FISH results. Increasing ERBB2 AmpRatio was associated with HER2 positivity (IHC 2+/FISH+ or IHC 3+), with strong performance for HER2+ prediction (AUC = 94%; Supplementary Fig. S21). 6.4% of mBC cases with ERBB2 AmpRatio [1, 2) were HER2+, which increased to 44% and 97% in the ERBB2 AmpRatio [2, 3) and ≥3 subsets. 79% (295/372) of ERBB2-amplified cases were HER2+, and conversely, 74% (295/398) of HER2+ cases were ERBB2amp+, with an overall percent agreement (OPA) of 95% (3,470/3,650; Fig. 3A; Supplementary Fig. S22A; Supplementary Table S5).

Figure 3.

Association between ERBB2 AmpRatio, IHC/FISH-defined HER2 status, and outcomes on HER2-directed therapy. A, Association between ERBB2 AmpRatio and HER2 status as defined by IHC and FISH for 3,650 patients with mBC with ERBB2 AmpRatio as determined by NGS and abstracted IHC/FISH testing. Where applicable, the IHC distribution was separated by ERBB2amp status within a given AmpRatio bin. B, rwOS for 128 patients with HER2+ (15 IHC 2+/FISH-positive and 113 IHC 3+) mBC treated with 1L trastuzumab-based regimens stratified by ERBB2 AmpRatio tertiles. C, rwOS for 238 patients with HER2-low or ultra-low (55 IHC 0, 110 IHC 1+, and 73 IHC 2+/FISH-negative) mBC treated with 2L+ T-DXd stratified by ERBB2 AmpRatio tertiles. D, Association between ERBB2 AmpRatio and HER2 status as defined by IHC and FISH for 1,168 patients with aGE cancer with ERBB2 AmpRatio as determined by NGS and abstracted IHC/FISH testing. Where applicable, the IHC distribution was separated by ERBB2amp status within a given AmpRatio bin. E, rwOS for 109 patients with HER2+ (26 IHC 2+/FISH-positive, 83 IHC 3+) aGE cancer treated with 1L trastuzumab–based regimens stratified by ERBB2 AmpRatio tertiles. 2L+, second-line or greater; FISH, fluorescence in situ hybridization.

We next assessed the ability of AmpRatio to stratify outcomes in patients with mBC with HER2+ disease by IHC, regardless of amplification status by NGS. In 128 patients with HER2+ (15 IHC 2+/FISH-positive, 113 IHC 3+) mBC treated with 1L trastuzumab-based regimens (Supplementary Table S6), ERBB2 AmpRatio correlated with rwOS (20.2, 57.8, and 62.9 months) and rwPFS (7.1, 19.8, and 11.6 months) across AmpRatio tertiles, though the longest rwPFS was observed in patients with an ERBB2 AmpRatio in the second tertile (P = 0.04). rwOS was significantly longer in patients with ERBB2 AmpRatio in both the second and third tertiles (P = 0.006 and P = 0.003) compared with the first tertile but did not significantly differ between the second and third tertiles (P = 0.79; Fig. 3B; Supplementary Fig. S23A). In this cohort, the HRs for rwOS and rwPFS remained largely consistent with overlapping CIs across AmpRatio thresholds 2 to 10 before paradoxically increasing with AmpRatio thresholds >10 (Supplementary Fig. S24A and S24B).

Among patients HER2+ mBC treated with T-DXd and T-DM1 (Supplementary Tables S7 and S8), rwOS trended with ERBB2 AmpRatio, with significantly longer rwOS in patients with an ERBB2 AmpRatio in the third tertile compared with the first tertile for T-DXd (P = 0.009) and T-DM1 (P = 0.003). Again, patients treated with T-DXd with ERBB2 AmpRatio in the second and third tertiles had numerically similar rwOS (45 and 46.2; P = 0.25) with the largest increase observed again between the first and second tertiles (Supplementary Figs. S25 and S26). There was similarly no clear optimal AmpRatio cutoff; however, there was a notable increase in the rwPFS HR point estimate at AmpRatio thresholds ≥10 for patients treated with T-DM1 (Supplementary Figs. S27 and S28).

Sensitivity analyses were performed within the HER2+ cohort to assess the impact of the independent predictive value of IHC 2+/FISH-positive versus IHC 3+ status. The association between outcomes and ERBB2 AmpRatio were largely consistent in the IHC 3+ population, but the analysis in the IHC 2+/FISH-positive cohort was limited by small sample sizes. Although we observed trends suggesting potential discriminatory value in this population, this analysis was underpowered to draw any independent conclusions (Supplementary Figs. S29–S31).

Among 238 patients with HER2-low or ultralow (55 IHC 0, 110 IHC 1+, and 73 IHC 2+/FISH-negative) mBC treated with 2L+ T-DXd (Supplementary Table S9), patients with ERBB2 AmpRatio in the second and third tertiles had numerically similar rwOS (13.9 and 15.1 months) and rwPFS (6.3 and 5.2 months), all with significantly longer responses (P < 0.05) compared with the first tertile (Fig. 3C; Supplementary Fig. S23B). In this cohort, the minimum rwOS and rwPFS HR was observed at AmpRatio 0.75, but this did not dichotomize outcomes significantly better than other thresholds (Supplementary Figs. S24C and S24D).

ERBB2 AmpRatio was also associated with HER2 positivity for 1,168 patients with aGE cancer in the CGDB with ERBB2 NGS and abstracted HER2 IHC/FISH results, with similarly strong performance for HER2+ prediction (AUC = 90%; Supplementary Fig. S21). The prevalence of HER2 positivity increased from 5.9% in samples with ERBB2 AmpRatio <1 to 11.6%, 52.3%, and 94.5% in ERBB2 AmpRatio [1, 2), [2, 3), and ≥3 subsets, respectively. Overall, 85.4% (216/253) of ERBB2amp+ cases were also HER2+, and conversely, 72.5% (216/298) of HER2+ cases harbored an ERBB2amp, with 90% OPA (Fig. 3D; Supplementary Table S10; Supplementary Fig. S22B).

Finally, we assessed the ability of binary ERBB2amp status and ERBB2 AmpRatio to stratify outcomes in patients with GE with IHC HER2+ disease. In 109 patients with HER2+ aGE cancer treated with 1L trastuzumab-based regimens (Supplementary Table S11), the presence of an ERBB2amp by NGS was associated with longer rwOS (15.2 vs. 8.5 months, P = 0.06) and rwPFS (8.3 vs. 5.4 months, P = 0.001; Supplementary Fig. S32). ERBB2 AmpRatio further stratified response to 1L trastuzumab in this cohort as ERBB2 AmpRatio was associated with rwOS (9.8, 12.4, and 17.1 months) and rwPFS (5.7, 7.4, and 12.0 months; Fig. 3E; Supplementary Fig. S23C). Evaluating a range of AmpRatio cutoffs, the HRs for rwOS and rwPFS consistently decreased with increasing AmpRatio (Supplementary Fig. S24E and S24F). Similarly, the sensitivity analysis assessing the association of AmpRatio within the separate IHC 3+ and IHC 2+/FISH-positive population revealed consistent trends in the IHC 3+ patients and was limited by small sample sizes in the IHC 2+/FISH-positive cohort (Supplementary Fig. S33).

MET amplifications are rare pan-tumor alterations and negatively correlate with actionable fusions

MET amplifications (defined as AmpRatio ≥3 in tissue) were present in 1.0% of pan-tumor tissue samples and most frequently detected in GE cancer (4.0%), gallbladder cancer (2.6%), and NSCLC (2.4%). Amplifications comprised the majority of MET activating alterations in most diseases; however, MET SVs including MET exon 14 skipping alterations were more prevalent in NSCLC, kidney cancer, and melanoma. The median AmpRatio for MET amplifications pan-tumor and in NSCLC were 5.9 and 5.2, respectively, with higher-level amplifications observed in liver cancer (13.3), GE cancer (8.8), and CRC (8.1; Fig. 4A; Supplementary Fig. S10).

Figure 4.

Emerging amplification target landscape distribution of GOF alteration types (top), prevalence (middle), and AmpRatio (bottom) for (A) MET, (B) EGFR, and (C) CCNE1 amplifications (AmpRatio ≥3) in tissue samples across tumor types. Alts, alterations; CCA, cholangiocarcinoma; CRC, colorectal cancer; Gi-Neuro, gastrointestinal neuroendocrine tumor.

Evaluating the co-driver landscape versus MET AmpRatio for all NSCLC MET CN segments, KRAS SVs and amplifications (both P < 0.0001), ALK rearrangements (P = 0.04), and CD74 fusions (primarily fused with NRG1 or ROS1; P = 0.04) were negatively correlated with MET AmpRatio, whereas TP53, MET, and EGFR SVs, BRAF, HGF, EGFR, and ERBB2 amplifications, and TMB ≥10 were associated with increasing MET AmpRatio (all P < 0.01; Supplementary Fig. S34).

EGFR amplifications are focal high-level events across tumor types

EGFR amplifications were present in 2.8% of pan-tumor tissue samples and were significantly more amplified compared with other genes (median AmpRatio 13.4 vs. 4.7, P < 0.0001). Amplifications were highly prevalent in glioma (24.8%) and were also highly amplified (median AmpRatio 31.1). EGFR amplifications in GE, small intestine, and breast cancers also tended to be highly amplified (median AmpRatio 15.2–21.3), whereas the median AmpRatio of EGFR amplifications in NSCLC was 7.4 (Fig. 4B; Supplementary Fig. S10).

Evaluating the NSCLC co-alteration landscape across all EGFR CN segments, EGFR SVs and rearrangements (P < 0.0001), MET amplifications (P = 0.004), and RB1 losses (P < 0.0001) were associated with increasing EGFR AmpRatio. Conversely, KRAS, STK11, ERBB2, KEAP1, and BRAF SVs, PTEN SVs and CN losses (all P < 0.0001), ALK, RET, ROS1, and CD74 (most frequently fused to NRG1) rearrangements (all P < 0.05), and TMB ≥10 (P < 0.0001) were all associated with lower EGFR AmpRatio (Supplementary Fig. S35).

CCNE1 amplifications are highly prevalent in gynecologic malignancies

CCNE1 amplifications were present in 3.2% of pan-tumor tissue samples and were most frequently observed in ovarian (12.0%), uterine/endometrial (9.5%), gallbladder (9.4%), and GE (7.6%) cancers. CCNE1 amplifications had a median pan-tumor AmpRatio of 5.3 and was highly amplified in gallbladder (11.4) and GE (7.9) cancers compared with other tumor types, including ovarian (5.1) and uterine/endometrial tumors (5; Fig. 4C; Supplementary Fig. S10).

In ovarian tumors, CCNE1 amplifications were enriched in serous carcinomas (14.3%) and carcinosarcomas (13.4%) but were less common in clear cell (2.2%) and endometrioid carcinomas (2.1%), although the AmpRatio distribution between subtypes did not significantly differ (4.3–5.2; P = 0.12; Supplementary Fig. S36). In serous carcinomas, TP53 SVs and AKT2, KRAS, BCL2L1, ERBB2, and ARAF amplifications were associated with increasing CCNE1 AmpRatio (all P < 0.0001). Conversely, HRDsig positivity, CDKN2A/B/MTAP loss, and BRCA1/2, KRAS, NF1, NRAS, and BRAF SVs were negatively associated with CCNE1 AmpRatio (all P < 0.0001; Supplementary Fig. S37).

Amplification detection in liquid biopsy is associated with ctDNA TF and AmpRatio

Amplifications were detected in 17.4% (15.2% with AmpRatio ≥3) of pan-tumor liquid biopsy samples (Supplementary Fig. S4) compared with 38.6% of tissue samples. The median TF of liquid biopsy samples harboring an amplification was 23% (IQR 11%–46%), and the prevalence increased to 54.2% (44.2% with AmpRatio ≥3) among high-shed liquid samples with TF ≥ 20%, which was a consistent trend across genes and tumor types (Fig. 5A; Supplementary Figs. S38 and S39). However, 44% of amplifications detected in liquid biopsies were in samples with TF < 20%. The median AmpRatio for all liquid CN amplifications was 4.1, and samples harboring amplifications with an AmpRatio above the median (e.g., EGFR, MDM2, MET, KRAS, and ERBB2) had lower TF compared with genes that tend to be less highly amplified (e.g., BCL2L1, GNAS, and CCND2; median TF 22% vs. 31%, P < 0.0001; Fig. 5B).

Figure 5.

Amplification detection in liquid biopsy. A, Prevalence of reported liquid biopsy amplifications with CN ≥ sample ploidy +3 for ERBB2 and CN ≥ sample ploidy +4 for all other genes. Prevalence in all liquid biopsy samples/prevalence in liquid biopsy samples with ctDNA TF ≥ 20% is plotted in the top graph and annotated in each cell. B, Distribution of ctDNA TF among pan-tumor liquid biopsy samples harboring an amplification in each gene (left) and the pan-tumor AmpRatio distribution for each gene (right). C, Pan-tumor PPA of amplification detection in liquid biopsy compared with tissue between paired patient tissue and liquid samples collected ≤30 days apart. In alignment with current clinical reporting, concordance was assessed between tissue amplifications with CN ≥ median CN + 3 for ERBB2 and AmpRatio ≥3 for all other genes versus liquid amplifications with CN ≥ median CN + 3 for ERBB2 and CN ≥ median CN + 4 for all other genes. CCA, cholangiocarcinoma; CRC, colorectal cancer; NPV, negative predictive value.

We assessed the sensitivity of amplification detection in liquid biopsy compared with tissue in 2,440 patient paired tissue and liquid samples collected ≤30 days apart. Across all genes, positive percent agreement (PPA) of amplification detection in liquid was 38.5%, which increased to 88.1% among samples with TF ≥ 20% (Supplementary Fig. S40). Similarly, for selected biomarkers, the PPA ranged from 81.6% to 100% in the TF ≥ 20% subset (Fig. 5C). Among the 323 concordantly detected amplifications, the tissue and liquid-based AmpRatios were highly concordant (Pearson R = 0.87; P < 0.001) and more correlated than the tissue- and liquid-based modeled CN values (R = 0.69; Supplementary Fig. S41). The median TF of concordant amplification-positive liquid samples was 21%, versus 0.57% in amplification-negative liquid samples (P < 0.0001), and the median AmpRatio of tissue amplifications was 7.6 for amplifications that were detected in the paired liquid versus 4.7 for amplifications that were not detected in liquid (P < 0.0001; Supplementary Fig. S42).

Discussion

CN amplifications are prevalent and diverse biomarkers distributed across cancer types, many of which are targetable with approved and emerging therapies. Across our dataset of nearly 500,000 solid tumor samples, almost 40% harbored ≥1 gene CN amplification and clinically relevant amplifications in ERBB2, MET, EGFR, and CCNE1 were detected across tumor types with varying degrees of CN gain. To quantify the degree of CN gain, we introduced AmpRatio, a CN metric normalized by the sample ploidy to produce a more analytically robust and reproducible CN estimate across varying sample purities and ploidies. AmpRatio had distinct distributions across genes and tumor types, and amplifications in diseases with higher degrees of aneuploidy such as bladder, prostate, and NSCLC, all of which were frequently triploid with high prevalences of CN amplifications, may benefit the most from having a quantitative CN value normalized by ploidy.

Although the therapeutic implications of CN amplifications are well established in several solid tumors, the metrics of detection and quantification remain imperfect. The current standard of care to determine HER2 status in breast and gastrointestinal cancers includes IHC and reflexive FISH for IHC 2+ cases. Although they typically yield concordant results, discrepancies have been reported, potentially due to a combination of lack of HER2 standardization in non-breast cancers, tumor heterogeneity, and variability between laboratories, with evidence suggesting that other testing modalities including DNA NGS and mRNA expression may yield additional insights (47–50). Underlying algorithmic differences may also result in differences in reported quantitative NGS-based CN values between assays. The NSCLC National Comprehensive Cancer Network (NCCN) guidelines reference NGS as a means of detecting high-level MET amplification, defined as ≥10 copies (10), but it is worth noting that this cutoff seems to be determined from FISH-based profiling in the GEOMETRY mono-1 trial, and there is little evidence relating this to other NGS-based quantitative CN metrics (51, 52).

Our study demonstrates that ERBB2 AmpRatio is associated with response to HER2 mAb and ADC regimens in ERBB2-amplified mBC, aGE cancer, and mCRC and additionally stratifies outcomes within patients with IHC HER2+ mBC and aGE cancer. 20.7% and 14.6% of ERBB2-amplified mBC and aGE cancer cases were not HER2+, suggesting that IHC/FISH-based HER2 status alone may not identify all likely responders. When evaluating the association of a range of exploratory AmpRatio thresholds with rwOS and rwPFS to anti-HER2 regimens, there was no clear optimal cutoff that dichotomized response, but disease-specific associations were observed. We observed a monotonic decrease in the HRs for patients with aGE cancer treated with trastuzumab-based regimens across a range of increasing AmpRatio thresholds, suggesting that higher AmpRatios are consistently associated with increased survival benefit. The strength and consistency of this effect was not seen in the mCRC and mBC cohorts. Interestingly, among patients with mBC treated with HER2 mAb or ADCs, we frequently observed the largest increase in rwOS between the first and second tertiles with nonsignificant differences between the second and third tertiles. Caveated by overlapping CIs, the lowest HRs in the ERBB2amp+/HER2+ subsets were observed at an AmpRatio threshold of 5 to 10, with increasing HR point estimates at thresholds greater than AmpRatio 10. These results suggest that unlike in aGE cancer, there may be underlying biological or clinical differences in mCRC and mBC with high-level amplifications. There is growing evidence that lower-level CN gains may be relevant to newer therapy modalities such as ADCs, for which even modest overexpression of the target can predict response (9, 53), and this study demonstrates that ERBB2 AmpRatio also stratifies outcomes to 2L+ T-DXd in patients with HER2-low and ultra-low mBC.

IHC is a semi-quantitative metric, and these associations between ERBB2 AmpRatio and outcomes to HER2-directed therapies within IHC-defined HER2 subgroups provide preliminary evidence for its utility as a quantitative biomarker to further stratify populations that are currently managed as a single clinical entity. For example, we saw significantly better outcomes in patients with aGE HER2 IHC+ with AmpRatio in the third tertile compared with the first and second, suggesting that IHC may not have sufficient quantitative range to potentially delineate those patients who may be exceptional responders. The therapy mechanism of action such as whether the amplification is targeted directly or through synthetic lethality (e.g., PKMYT1 inhibitors targeting CCNE1), may be another consideration. These results suggest that the appropriate CN cutoff may vary by disease and treatment scenario and that incorporation of quantitative CN biomarkers such as AmpRatio into clinical trials may allow for more granular and personalized patient stratification to best inform clinical decision making.

We additionally investigated the genomic co-alteration landscape across all CN segments, inclusive of CN loss, neutral, and gain for emerging biomarkers and observed several potentially clinically relevant associations. MET and EGFR AmpRatio in NSCLC tended to be inversely related to the presence of canonical drivers, with notable exceptions for cases in which the MET or EGFR amplification may be associated with acquired resistance or histologic transformation, and CCNE1 AmpRatio in ovarian was inversely related to HRDsig positivity.

The detection of biomarkers, particularly CNAs, using liquid biopsy is a known challenge because of low ctDNA shed, which was reflected by the low sensitivity of amplification detection in liquid samples with low TF. In liquid samples with TF ≥ 20%, we observed high concordance of amplification detection between liquid and tissue biopsies, highlighting the utility of TF to guide confidence in negative liquid CNA results, particularly given that TF ≥ 20% accounts for 17.9% of pan-tumor liquid samples, although this also varies by disease. Detection of amplifications in liquid biopsy was a function of the degree of amplification and TF, as evidenced by the high proportion of amplification-positive samples with TF < 20%, and both should be important considerations when leveraging liquid biopsy for amplification detection. When TF is sufficient for liquid-based amplification detection, we observed a strong correlation between the tissue and liquid-based AmpRatio values, supporting the use of liquid biopsy as a viable alternative when tissue may be unavailable for NGS or IHC, such as at therapy progression.

There were several limitations to this study. First, our cohort was restricted to patients with CGP performed during routine clinical care, and the CGDB was further restricted to patients treated in the FH network, which may not be representative of the general population, which may affect reported amplification prevalences and negative predictive value of liquid-based detection. The frequency of ERBB2 amplifications in this study of patients with primarily advanced disease is lower than that classically associated with breast cancers, which likely reflects the success of anti-HER2 therapies used in earlier treatment settings, which result in a lower percentage of advanced HER2+ breast cancer cases being submitted for CGP. Furthermore, due to the lack of clinical annotation in the FMI genomic database used for prevalence and co-alteration analysis, this cohort likely included pre- and post-treatment specimens which may also affect the reported amplification prevalences, particularly for genes associated with acquired resistance. Finally, the outcomes analyses were limited by small sample size and real-world treatment patterns and response assessments, and results may differ if evaluated in a larger, prospective cohort.

In this study, we summarized the landscape of CN amplifications across tumor types and demonstrated the validity of AmpRatio as a quantitative CN metric associated with outcomes to HER2-directed therapies. AmpRatios were highly variable, and the threshold for clinically relevant amplifications may vary by disease, gene, and the therapy mechanism of action. With the rapidly growing number of novel agents targeting gene amplification or protein overexpression under investigation, it will be important to define harmonized methods for determining biomarker definitions and selecting patients that will derive the most benefit from these therapies.

Supplementary Material

Supplemental Tables S1-S11

Supplemental Table 1: Demographics and clinical characteristics of the ERBB2-amplified mBC cohort treated with first-line trastuzumab-based regimens Supplemental Table 2: Demographics and clinical characteristics of the ERBB2-amplified mBC cohort treated with T-DXd Supplemental Table 3: Demographics and clinical characteristics of the ERBB2-amplified aGE cohort treated with first-line trastuzumab-based regimens Supplemental Table 4: Demographics and clinical characteristics of the ERBB2-amplified CRC cohort treated with trastuzumab Supplemental Table 5: ERBB2 NGS and HER2 IHC/FISH concordance for 3,650 mBC patients Supplemental Table 6: Demographics and clinical characteristics of the HER2+ mBC cohort treated with 1L trastuzumab-based regimens Supplemental Table 7: Demographics and clinical characteristics of the HER2+ mBC cohort treated with T-DXd Supplemental Table 8: Demographics and clinical characteristics of the HER2+ mBC cohort treated with T-DM1 Supplemental Table 9: Demographics and clinical characteristics of the HER2-low/ultra-low mBC cohort treated with 2L + T-DXd Supplemental Table 10: ERBB2 NGS and HER2 IHC/FISH concordance for 1,168 aGE patients Supplemental Table 11: Demographics and clinical characteristics of the HER2+ aGE cohort treated with 1L trastuzumab-based regimens.

Supplemental Figures S1-S42

Supplemental Figure 1: Modeled CN vs. AmpRatio Supplemental Figure 2: Amplification ratio purity invariance Supplemental Figure 3: Modeled gene CN vs. AmpRatio Supplemental Figure 4: CONSORT diagram Supplemental Figure 5: Pan-tumor MDM2 amplification landscape Supplemental Figure 6: MDM2 co-alteration landscape Supplemental Figure 7: Tissue ploidy distribution by tumor type Supplemental Figure 8: CN segment length vs. AmpRatio Supplemental Figure 9: Amplicon focality Supplemental Figure 10: Pan-tumor AmpRatio distribution of emerging targets Supplemental Figure 11: Breast ERBB2 co-alteration landscape Supplemental Figure 12: GE ERBB2 co-alteration landscape Supplemental Figure 13: CRC ERBB2 co-alteration landscape Supplemental Figure 14: Breast CGDB CONSORT diagram Supplemental Figure 15: GE CGDB CONSORT diagram Supplemental Figure 16: CRC CGDB CONSORT diagram Supplemental Figure 17: rwPFS for ERBB2-amplified mBC, aGE, and mCRC patients treated with trastuzumab-based regimens Supplemental Figure 18: Exploratory AmpRatio cutoffs for ERBB2-amplified mBC, aGE, and mCRC patients treated with trastuzumab-based regimens Supplemental Figure 19: Outcomes for ERBB2-amplified mBC patients treated with T-DXd Supplemental Figure 20: Exploratory AmpRatio cutoffs for ERBB2-amplified mBC patients treated with T-DXd Supplemental Figure 21: Performance of ERBB2 AmpRatio for predicting HER2-positivity Supplemental Figure 22: Association between ERBB2 AmpRatio and HER2 IHC/FISH Supplemental Figure 23: rwPFS on anti-HER2 regimens within HER2 IHC-defined subgroups Supplemental Figure 24: Exploratory AmpRatio cutoffs for anti-HER2 regimens within HER2 IHC-defined subgroups Supplemental Figure 25: Outcomes for HER2+ mBC patients treated with T-DXd Supplemental Figure 26: Exploratory AmpRatio cutoffs for HER2+ mBC patients treated with T-DXd Supplemental Figure 27: Outcomes for HER2+ mBC patients treated with T-DM1 Supplemental Figure 28: Exploratory AmpRatio cutoffs for HER2+ mBC patients treated with T-DM1 Supplemental Figure 29: Sensitivity analysis for outcomes of HER2+ mBC treated with 1L trastuzumab-based regimens Supplemental Figure 30: Sensitivity analysis for outcomes of HER2+ mBC treated with T-DXd Supplemental Figure 31: Sensitivity analysis for outcomes of HER2+ mBC treated with T-DM1 Supplemental Figure 32: Outcomes for HER2+ aGE patients treated with 1L trastuzumab-based regimens Supplemental Figure 33: Sensitivity analysis for outcomes of HER2+ aGE treated with 1L trastuzumab-based regimens Supplemental Figure 34: NSCLC MET co-alteration landscape Supplemental Figure 35: NSCLC EGFR co-alteration landscape Supplemental Figure 36: Ovarian CCNE1 amplification landscape Supplemental Figure 37: Serous Ovarian CCNE1 co-alteration landscape Supplemental Figure 38: ctDNA TF distribution by alteration type Supplemental Figure 39: ctDNA TF distribution by tumor type Supplemental Figure 40: All-gene liquid amplification concordance Supplemental Figure 41: Tissue and Liquid AmpRatio Correlation Supplemental Figure 42: Distribution of ctDNA TF and AmpRatio for discordant vs. concordant liquid samples.

Data Availability

The authors declare that all relevant aggregate data supporting the findings of this study are available within the article and its supplementary information files. The data that supports the findings of this study originated from FH and FMI. In accordance with the HIPAA, FMI does not have IRB approval or patient consent to share individualized patient genomic data, which contains potentially identifying or sensitive patient information and cannot be reported in a public data repository. More information and requests for data sharing by license or by permission for the specific purpose of replicating results in this article can be made to the corresponding author or submitted to data.governance.council@foundationmedicine.com, PublicationsDataAccess@flatiron.com, and cgdb-fmi@flatiron.com.

Authors’ Disclosures

J.K. Lee reports personal fees from FMI and Roche during the conduct of the study. J.C.F. Quintanilha reports other support from FMI and Roche during the conduct of the study, as well as other support from FMI and Roche outside the submitted work. B. Fendler reports financial incentives through Roche stock. C.F.B. Tambaoan reports personal fees from FMI and Roche outside the submitted work. R. Graf reports personal fees from FMI during the conduct of the study, as well as personal fees from Roche outside the submitted work. L. Pusztai reports grants and personal fees from Merck, AstraZeneca, and Natera, grants from Pfizer and Exact Sciences, personal fees from Radionetics, Daiichi Sankyo, and BeOne Medicines, and other support from Ataraxis outside the submitted work. H. Singh reports grants from AstraZeneca, personal fees from Zola Therapeutics, Dewpoint Therapeutics, UpToDate, Merck, Sharp & Dohme, and Blueprint Medicines, and other support from DAVA Oncology outside the submitted work. M. Strickland reports personal fees from Astellas Pharma, Bayer, and Bristol Myers Squibb outside the submitted work. T.A. O’Meara reports personal fees from Third Rock Ventures outside the submitted work. S.M. Tolaney reports grants and personal fees from Novartis, Pfizer/Seagen, Merck, Eli Lilly and Company, AstraZeneca, Genentech/Roche, Bristol Myers Squibb/SystImmune, Daiichi Sankyo, Gilead Sciences, Menarini/Stemline Therapeutics, Jazz Pharmaceuticals, and Olema Pharmaceuticals; personal fees from Eisai, Blueprint Medicines, Reveal Genomics, Sumitovant Biopharma, Artios Pharma, AADI Bioscience, Bayer, Natera, Tango Therapeutics, eFFECTOR Therapeutics, Hengrui USA, Cullinan Oncology, Circle Pharma, Arvinas, BioNTech, Launch Therapeutics, Zuellig Pharma, Johnson & Johnson/Ambrx, Bicycle Therapeutics, BeiGene Therapeutics, Mersana Therapeutics, Summit Therapeutics, Avenzo Therapeutics, Aktis Oncology, Celcuity, Boehringer Ingelheim, Samsung Bioepis, Tempus, Boundless Bio, Denali Therapeutics, and Roche; and grants from Exelixis, NanoString Technologies, and OncoPep outside the submitted work. T.A. Yap reports other support from the University of Texas MD Anderson Cancer Center [Vice President, Head Clinical Development in the Therapeutics Discovery Division, with commercial interest in DNA damage response and other inhibitors (IACS30380/ART0380 licensed to Artios)]; grants from Acrivon, Artios, AstraZeneca, Bayer, BeiGene, BioNTech, Blueprint Medicines, Bristol Myers Squibb, Boundless Bio, Clovis Oncology, Constellation Pharmaceuticals, Cyteir Therapeutics, Eli Lilly and Company, EMD Serono, Forbius, F-star, GlaxoSmithKline, Genentech, Haihe Biopharma, IDEAYA Biosciences, ImmuneSensor Therapeutics, Insilico Medicine, Ionis Pharmaceuticals, Ipsen, Jounce Therapeutics, Karyopharm Therapeutics, KSQ Therapeutics, Kyowa Kirin, Merck, Mirati Therapeutics, Novartis, Pfizer, Ribon Therapeutics, Regeneron Pharmaceuticals, Repare Therapeutics, Rubius Therapeutics, Sanofi, Scholar Rock, Seattle Genetics, Tango Therapeutics, Tesaro Therapeutics, Vivace Therapeutics, Zenith Pharmaceuticals, NCI Cancer Center, Department of Defense, V Foundation, and NIH; and personal fees from AbbVie, Acrivon Therapeutics, Adagene, Almac, Aduro, Amphista Therapeutics, Artios Pharma, Astex Pharmaceuticals, AstraZeneca, Athena, Atrin Pharmaceuticals, Avenzo Therapeutics, Avoro Capital, Axiom Pharma, Baptist Health System, Bayer, BeiGene, BioCity Biopharma, Blueprint Medicines, Boxer Capital, Bristol Myers Squibb, C4 Therapeutics, Calithera Biosciences, Cancer Research UK, Carrick Therapeutics, Circle Pharma, Clovis Oncology, Cybrexa Therapeutics, Daiichi Sankyo, Dark Blue Therapeutics, Diffusion Pharma, Duke Street Bio, 858 Therapeutics, EcoR1 Capital, Ellipses Pharma, EMD Serono, Entos Pharmaceuticals, F-star, Genesis Therapeutics, Genmab, Glenmark Pharmaceuticals, GLG Pharma, Globe Life Sciences, GlaxoSmithKline, Guidepoint, IDEAYA Biosciences, Idience, Ignyta, I-Mab, ImmuneSensor Therapeutics, Impact Therapeutics, Institut Gustave Roussy, Intellisphere, Janssen Pharmaceuticals, Kyn Therapeutics, MEI Pharma, Mereo BioPharma, Merck, Merit Pharmaceutical, Monte Rosa Therapeutics, Natera, Nested Therapeutics, Nexus Pharmaceuticals, Nimbus Therapeutics, Novocure, Odyssey Therapeutics, OHSU, OncoSec, Ono Pharmaceutical, Onxeo, PanAngium Therapeutics, PEGASCY-GROUP, Par Pharmaceutical, Pfizer, Piper Sandler, Pliant Therapeutics, Prolynx, Radiopharm Theranostics, Repare Therapeutics, resTORbio, Roche, Ryvu Therapeutics, Swiss Group for Clinical Cancer Research (SAKK), Sanofi, Schrödinger, Servier, Synnovation Therapeutics, Synthis Therapeutics, Tango Therapeutics, TCG Crossover, TD2 Oncology, Terremoto Biosciences, Tessellate Bio, Theragnostics, Terns Pharmaceuticals, TOLREMO therapeutics, Tome Biosciences, Thryv Therapeutics, Trevarx Biomedical, Varian Pharmed, Veeva, Versant Ventures, Vibliome Therapeutics, Voronoi Inc., XinThera, Zai Lab, and Ziel Biosciences outside the submitted work. J. Ross reports employment with FMI, as well as equity ownership in Roche Holdings and Tango Therapeutics. A. Gasco Hernandez reports employment with FMI. B. Decker reports other support from FMI and Roche during the conduct of the study, as well as a patent for Liquid Biopsy Copy Number Detection Enhancements Using Framentomic Features pending. R.S.P. Huang reports personal fees from FMI, a wholly owned subsidiary of Roche, during the conduct of the study. S.J. Klempner reports personal fees from Gilead Sciences, Amgen, Boehringer Ingelheim, BeiGene, I-Mab, EsoBiotec, AstraZeneca, Astellas Pharma, Merck, Natera, Taiho Pharmaceutical, Elevation Oncology, and Jazz Therapeutics outside the submitted work, and reports being a member (uncompensated) of the NCCN Guidelines Committee for Gastric and Esophageal Cancers, the NCI Task Force for Esophagogastric Cancers, Debbie’s Dream Foundation Medical Advisory Board, and the Hope for Stomach Cancer Advisory Board. E.S. Sokol reports other support from FMI and Roche during the conduct of the study. A.B. Schrock reports other support from FMI and Roche during the conduct of the study. No disclosures were reported by the other authors.

Disclaimer

The Editor-in-Chief of Clinical Cancer Research at the time this article was being considered at the journal, is an author of this article. In keeping with AACR editorial policy, a senior member of the Clinical Cancer Research external editorial team managed the consideration process for this submission and independently rendered the final decision about acceptability.

Authors’ Contributions

J.K. Lee: Formal analysis, visualization, methodology, writing–original draft, writing–review and editing. J.C.F. Quintanilha: Formal analysis, visualization, methodology, writing–review and editing. K.-T. Chen: Software, formal analysis, visualization, methodology, writing–review and editing. B. Fendler: Software, formal analysis, visualization, methodology, writing–review and editing. C.F.B. Tambaoan: Software, formal analysis, methodology, writing–review and editing. R. Graf: Formal analysis, supervision, methodology, writing–review and editing. N. Odzer: Supervision, methodology, writing–review and editing. L. Pusztai: Supervision, writing–review and editing. M. Lustberg: Supervision, writing–review and editing. H. Singh: Supervision, writing–review and editing. M. Strickland: Supervision, writing–review and editing. T.A. O’Meara: Supervision, writing–review and editing. S.M. Tolaney: Supervision, writing–review and editing. T.A. Yap: Supervision, writing–review and editing. J. Ross: Supervision, writing–review and editing. A. Gasco Hernandez: Supervision, writing–review and editing. B. Decker: Supervision, writing–review and editing. R.S.P. Huang: Supervision, methodology, writing–review and editing. S.J. Klempner: Supervision, methodology, writing–review and editing. E.S. Sokol: Conceptualization, supervision, methodology, writing–review and editing. A.B. Schrock: Conceptualization, supervision, methodology, writing–original draft, writing–review and editing.