Formalin-fixed paraffin-embedded (FFPE) patient samples (n=64,354) were submitted to a commercial CLIA-certified laboratory (Caris Life Sciences, Phoenix, AZ). The present study was conducted in accordance with guidelines of the Declaration of Helsinki, Belmont Report, and U.S. Common Rule. With compliance to policy 45 CFR 46.101(b) (4), this study was conducted using retrospective, de-identified clinical data. Therefore, the present study was considered IRB exempt, and patient consent was not required.

Gene fusion detection was performed on mRNA isolated from a FFPE tumor sample using the Illumina NovaSeq platform (Illumina, Inc., San Diego, CA) and Agilent SureSelect Human All Exon V7 bait panel (Agilent Technologies, Santa Clara, CA). FFPE specimens underwent pathology review to diagnose percent tumor content and tumor size; a minimum of 10% of tumor content in the area for microdissection was required to enable enrichment and extraction of tumor-specific RNA. Qiagen RNA FFPE tissue extraction kit was used for extraction, and the RNA quality and quantity was determined using the Agilent TapeStation. Biotinylated RNA baits were hybridized to the synthesized and purified cDNA targets and the bait-target complexes were amplified in a post capture PCR reaction. The resultant libraries were quantified, normalized and the pooled libraries are denatured, diluted and sequenced; the reference genome used was GRCh37/hg19 and analytical validation of this test demonstrated ≥97% Positive Percent Agreement (PPA), ≥99% Negative Percent Agreement (NPA) and ≥99% Overall Percent Agreement (OPA) with a validated comparator method.

Fusion topologies were assessed by mapping gene fusion sequences to the reference genome GRCh37/hg19 to identify chromosome number, sense/antisense strand, and genomic breakpoint coordinates for gene-pairs of each fusion event detected. For each tumor, detection of multiple isoforms of a unique gene-pair with the same 5’-3’ gene order, presumed to result from alternative splicing, and reciprocal fusion isoforms were scored as a single event in the topology distribution.

NGS was performed on genomic DNA isolated from FFPE tumor samples using the NextSeq platform (Illumina, Inc., San Diego, CA). Matched normal tissue was not sequenced. A custom-designed SureSelect XT assay was used to enrich 592 whole-gene targets (Agilent Technologies, Santa Clara, CA). All variants were detected with >99% confidence based on allele frequency and amplicon coverage, with an average sequencing depth of coverage of >500X and an analytic sensitivity of 5%. Prior to molecular testing, tumor enrichment was achieved by harvesting targeted tissue using manual microdissection techniques. Genetic variants identified were interpreted by board-certified molecular geneticists and categorized as ‘pathogenic,’ ‘presumed pathogenic,’ ‘variant of unknown significance,’ ‘presumed benign,’ or ‘benign,’ according to the American College of Medical Genetics and Genomics (ACMG) standards. When assessing mutation frequencies of individual genes, ‘pathogenic,’ and ‘presumed pathogenic’ were counted as mutations while ‘benign’, ‘presumed benign’ variants and ‘variants of unknown significance’ were excluded.

The datasets presented in this article from Caris Life Sciences are not publicly available because the raw data is protected proprietary information – these datasets are available for qualified researchers upon reasonable request and with permission of Caris Life Sciences.

We queried the TCGA PanCancer Analyses, MSK Impact and AACR GENIE data bases through cBioPortal (accession dates 6/18/20, 8/8/20, 8/8/20 respectively) for EGFR, ERBB2 and ERBB4 fusions (22–25).

H3122 and Ba/F3 cell lines were acquired through Dr. John Minna (The University of Texas Southwestern Medical Center, Dallas, TX) and Dr. Dan Theodorescu (University of Colorado Comprehensive Cancer Center, Aurora, CO), respectively. Cells were grown in RPMI-1640 with 10% FBS. The HEK293T cell line was purchased from ATCC. Alectinib was purchased from Selleck Chemicals. AKT pS437 (4058), AKT (2920), ERK pT202/Y204 (9101), ERK (9107), EGFR pY845 (6963) and HER2 pY877 (2241) were purchased from Cell Signaling Technology. EGFR (610017) and HER2 (610161) antibodies were purchased from BD Biosciences and anti-GAPDH was from Millipore.

Lentiviral constructs for EGFR and ERBB2 fusions were manufactured by Vector Builders Inc (Chicago, IL) in their pLV lentiviral vector under CMV promoters. Transduction into H3122 was performed as previously described (26) with the following additions: after puromycin selection, H3122 cells were cultured in the presence of 100nM alectinib to inhibit ALK signaling. Cell viability assays and signaling analysis via immunoblotting were performed in the presence of 100nM alectinib.

Cell proliferation assays were performed using the CellTiter 96 Aqueous One Solution from Promega according to the manufacturer’s instructions after 72 hours of incubation with indicated inhibitors.

Cells were lysed in T-PER (Thermo Scientific) with Halt Protease and Phosphatase Inhibitor (Thermo Scientific). Proteins were separated with SDS-PAGE, transferred to nitrocellulose membranes and stained with indicated antibodies. Secondary IR-Dye anti-mouse or anti rabbit IgG (LI-COR) antibodies were then added. Imaging was performed with the Odyssey Imager and Odyssey Image Studio software (LI-COR).

We identified a total of 1,251 ERBB family fusions across the four data sets including a total of 182,590 samples. EGFR fusions (n=811) were the most common overall, followed by ERBB2 (n=287) and ERBB4 fusions (n=153). We estimated that the frequency of EGFR fusions across data sets was 0.4%, ERBB2 was 0.2% and ERBB4 was 0.1% ( Figure 1A ). We did not assess the frequency of ERBB3 fusions in this study, as the HER3 protein is unable to form functional homodimers (12). For the Caris samples, profiled by whole transcriptome sequencing, we further analyzed the topology of ERBB fusion transcripts to determine if the ERBB fusions resulted from deletion, inversion, duplication or translocation events. The majority of EGFR and ERBB2 fusions were the result of inversions, while ERBB4 fusions were most often translocations ( Figure 1B ). We also found that ~80% of EGFR fusions were both in-frame and retained a kinase domain. Approximately 50% of ERBB4 and 40% of ERBB2 fusions had these two features ( Figure 1C ).

Next, we analyzed the identity of the ERBB gene fusion partners. EGFR-SEPT14 was the most common EGFR fusion, along with the reciprocal fusion SEPT14-EGFR that also results from the associated inversion event. However, based on the common breakpoint at EGFR exons 24-25, which retains the extracellular and kinase domains in the EGFR-SEPT14 fusion, the SEPT14-EGFR reciprocal fusion would harbor only a small C-terminal portion of the EGFR protein, and thus, potentially have different or no oncogenic consequences compared to EGFR-SEPT14 fusions. ERBB2-IKZF3 and ERBB2-PPP1R1B were the two most common ERBB2 fusions. IKZF2-ERBB4 was the most common ERBB4 fusion ( Table 1 ).

We assessed the most common cancer types harboring ERBB fusions. The majority of EGFR fusions were found in glioblastomas or gliomas, followed by NSCLC. ERBB2 fusions were most frequently observed in breast cancer followed by gastroesophageal cancers. Depending on the dataset, ERBB4 fusions were most often found in ovarian cancers, breast cancers or NSCLC ( Figure 2 ).

We categorized the fusions that we identified based on their genomic architecture and proposed mechanism of activation. The first class, classical 3’ fusions, have a 5’ gene fusion partner that introduces a new 5’ promoter that likely enhances expression and typically has protein features, such as coiled-coil domains, that may promote dimerization of the fusion kinase. Class II fusions have a 3’ gene fusion partner that potentially allows constitutive dimerization. Class III fusions are tandemly linked to another kinase domain that may promote activation, presumably through enhanced dimerization. Lastly, Class IV fusions are tethered to an adapter protein that we propose allows direct activation of downstream pathways ( Figure 3A ). We assigned each fusion we identified to a class and found that overall Class II fusions were the most common among ERBB fusions, followed by Class I. Interestingly, we noticed that ERBB2 fusions had an increased rate of Class III and class IV fusions, compared to EGFR or ERBB4 fusions ( Figure 3B ).

We then analyzed which mutations co-occur with ERBB fusions. EGFR fusions frequently co-occurred with TERT mutations (>60%) and EGFR mutations (~30%). We also observed that ERBB2 and ERBB4 fusions co-occurred with TP53 mutations (~60-80%) at a rate that was higher than the mutation rate found in the data sets queried (36-42%) ( Figure 4 ).

Class I and Class II fusions have been extensively characterized and evidence supports oncogenic activity (13, 27, 28). Given the lack of functional characterization of Class III and IV fusions, we selected several of the EGFR and ERBB2 fusions from the novel Class III and IV groups to analyze in vitro, including EGFR-ERBB4 (Class III), EGFR-SHC1 (Class IV), ERBB2-SHC1 (Class IV) and ERBB2-GRB7 (Class IV) ( Figure 5A ). First, we attempted to express these fusions in the Ba/F3 cell line, which is a murine B-cell cell line that is dependent on the addition of exogenous IL3 for proliferation but can become IL3-independent following oncogenic transformation. Ultimately, we were unable to transform the Ba/F3 cells with introduction of ERBB fusions. We hypothesized that because Ba/F3 cells may be lacking expression of signaling proteins necessary for these ERBB family fusions to successfully transform cells and therefore attempted to introduce ERBB fusions into a lung adenocarcinoma cell line. Previously we have demonstrated that H3122 can serve as model cells for novel fusions such as NRG1 fusions (29). Parental H3122 cells without ERBB fusions are insensitive to EGFR or HER2 inhibition (29). Other groups have also used similar techniques to model novel oncogenes in lung adenocarcinoma cell lines (30). We transduced the novel ERBB fusions into the EML4-ALK fusion expressing H3122 cell line using lentivirus. We then performed short term selection with the ALK inhibitor alectinib to eliminate non-transduced cells and to select for cells containing ERBB fusions, which should inhibit proliferation of untransformed cells that remain dependent on ALK signaling, but not those cells expressing the ERBB fusions ( Figure 5B ). Once the cells were able to proliferate in 100nM alectinib we assessed their sensitivity to several EGFR/HER2 inhibitors. We found that both the ERBB2-GRB7+ and ERBB2-SHC1+ cells were sensitive to the HER2 inhibitors afatinib, lapatinib and tarloxotinib suggesting both that the cells were now reliant on the ERBB2 fusion for proliferation and that they are sensitive to HER2 inhibition. The transformed cells were less sensitive to the EGFR inhibitor gefitinib ( Figure 5C ).

Next, we characterized the signaling changes in our ERBB+ cells in response to the EGFR inhibitors gefitinib, afatinib, osimertinib and tarloxotinib. Cells expressing either EGFR-ERBB4 or EGFR-SHC1 fusions showed a decrease in pEGFR following treatment, as well as decreased activation of the downstream signaling proteins pERK and pAKT. Additionally, we saw inhibition of the 3’ fusion partners pHER4 and pSHC1. As expected, the mutant-selective EGFR inhibitor osimertinib was less effective at inhibiting EGFR and downstream signaling, given the wild-type sequence of the EGFR kinase domain in these constructs ( Figure 5D ). We observed a similar pattern in our ERBB2 fusion cell lines where treatment with HER2 inhibitors (afatinib and tarloxotinib) resulted in decreased HER2 phosphorylation and decreased pERK and pAKT. Gefitinib and osimertinib, which have less activity against HER2, resulted in less inhibition of HER2 and downstream signaling ( Figure 5E ).