H3T assay design and performance characteristics
The H3T assay was developed to quantify the total expression of HER3 protein in FFPE specimens. The H3T assay is based on proximity binding of two derivatized antibodies and is similar in design to the VeraTag HER2 assay (HERmark), as previously described [25, 29] (Fig. 1a).
The performance characteristics of the VeraTag H3T assay are illustrated in Fig. 1b–f. H3T assay accuracy was cross-validated using a panel of cell lines expressing different levels of HER3 protein, prepared either as FFPE blocks, cell lysates or fixed cells, by comparing H3T measurements obtained using the VeraTag assay to measurements derived using several standard, well established technologies (Fig. 1b). ELISA and VeraTag measurements display a strong, linear correlation (R
2 = 0.97, p = 0.0002) across more than 2 orders of magnitude of HER3 expression. Based on flow cytometry data, HER3 receptor levels varied from several hundred to thirty-thousand receptors per cell in the selected cell lines. As expected, a strong correlation (R
2 = 0.89, p = 0.005) was found between VeraTag H3T measurements and the number of receptors per cell. HER3 IHC was also performed on FFPE sections prepared from cell line pellets. As expected, MB453 cells stained strongly, consistent with high HER3 expression. Staining for the other cell lines was much weaker, with the IHC limit of detection near the HER3 expression levels in MB468 cells. In contrast to IHC where only slight differences in staining were observed between the MB468, MB231, and SKOV3 cell lines, the VeraTag assay revealed a ~20-fold difference in H3T expression levels across these cell lines, consistent with FACS and ELISA results.
The precision (intra-assay variation) of the H3T assay was demonstrated by generating replicate measurements of H3T expression in cell pellets prepared by FFPE and tested in the same assay batch (Fig. 1c). Assay reproducibility (inter-assay variation) was demonstrated by performing duplicate measurements of H3T in 57 breast tumors prepared as FFPE specimens and tested in different assay batches (Fig. 1d). The excellent correlation between duplicate measurements (Spearman r = 0.94, p < 0.0001) showed a mean CV of 18 %.
The specificity of the VeraTag assay for HER3 detection was evaluated by substituting non-specific isotype matching control (ITC) antibodies for each of the HER3 specific antibodies (Ab6, B9A11) in the assay. This experiment was performed in both cell lines and breast tumors prepared as FFPE specimens (Fig. 1e). The overlap in assay signal between tumors at the low end of the H3T dynamic range and tumors at the high end of the ITC range is small and thus was used to establish a less than minimum (LTM) signal threshold. Samples that generate LTM H3T values were considered negative for HER3 within the sensitivity limits of the assay.
The potential contribution to the H3T assay signal by non-specific Ab6 and B9A11 binding to fat and non-tumor stroma was evaluated by comparing H3T measurements obtained using multiple paired tumor sections that were or were not macro-dissected to remove non-tumor tissue prior to testing. H3T measurements were not affected by the presence of fat or non-tumor stroma in the tumor section (Fig. 1f). The specificity of the two HER3 antibodies used in the assay was further confirmed by the absence of cross-reactivity in SDS-PAGE/western blot and VeraTag assays using cell lines expressing variable levels of HER3, HER2, and HER1 (data not shown).
Impact of HER3 expression on clinical outcomes in trastuzumab-treated MBC
H3T results were generated for 89 specimens with available tissue in a cohort of patients with MBC that were treated with trastuzumab and had well-documented clinical outcomes (PFS and OS). The clinical characteristics of the patients are described in Supplementary Table S1, and correlation of clinical characteristics is reported in Table S2. HER2 tumor status and trastuzumab treatment were based largely on HER2 IHC results with the remainder by HER2 FISH. However, retrospective testing by HERmark, a quantitative HER2 protein assay, and central laboratory performed HER2 FISH revealed better outcomes in patients with HERmark HER2-positive tumors and, to a lesser extent, those with HER2 FISH-positive tumors . In the current study, we sought to determine whether those patients with the most favorable outcomes as predicted by HERmark or HER2 FISH assays could be subdivided into groups of different outcomes by applying a quantitative HER3 cutoff. Sixty-one of the 89 patients that had tissue available for H3T analysis were assessed as HER2-high or H2T-high (H2T > 13.8) by the HERmark H2T assay, as previously reported . We were unable to derive precise H3T values for four H2T-high cases. Based on the correlation between H3T expression levels and PFS for the remaining 57 H2T-high cases, we used positional scanning analyses of the hazard ratio and selection by the lowest p value to establish a H3T assay cutoff (H3T = 3.5) that best discriminated patient subgroups with significantly different outcomes. Using this cutoff, Kaplan–Meier analyses were performed to compare outcomes (PFS and OS) for the two H2T-high groups, i.e., HER3-low (H3T ≤ 3.5) and HER3-high (H3T > 3.5), against the H2T-low (HER2-normal) group (Fig. 2a, b; Table 1). Within the H2T-high group (H2T > 13.8), patients with HER3-low tumors experienced longer PFS than patients with HER3-high tumors (PFS: HR = 2.7, p = 0.0021), although no significant difference in OS was observed. Similar results were found when focusing on the HER2 FISH-positive subgroup (Fig. 2c, d; Table 1). The clinical outcomes of the H2T-high (H2T > 13.8)/HER3-high (H3T > 3.5) group were comparable to the H2T-low group (H2T ≤ 13.8; Table 1). Within the H2T-low group, the four patients with H3T > 3.5 had PFS (p = 0.6) and OS (p = 0.14) that were indistinguishable from the poor outcomes of the other H2T-low patients.
Impact of HER3 expression on clinical outcomes independent of p95 and very high H2T levels
Next, a multivariate analysis was undertaken to determine the role of HER3 in the context of both H2T and p95. A regression tree analysis was used to ensure that previously established univariate cutoffs for H2T and p95 were similar when considering H3T, H2T, and p95 simultaneously. Using this same cohort of trastuzumab-treated MBC patients, we previously reported that within the confirmed HER2-positive group (H2T > 13.8 or FISH/CEP17 ≥ 2.0), patients with tumors that expressed elevated levels of p95 (p95-high) experienced less favorable outcomes compared to the subgroup of patients with tumors that expressed low p95 levels (p95-low) . Also in the confirmed HER2-positive group, patients with tumors that expressed very high H2T levels of H2T > 68.5 in the current cohort  and H2T > 125 in an early breast cancer cohort  experienced less favorable outcomes compared to the rest of the HER2-positives. Having applied the VeraTag HER3 assay to measure H3T expression, we next considered HER3 expression in the context of p95 and quantitative HER2 expression to explore whether these additional measurements could further explain the varied clinical outcomes of patients within the confirmed HER2-positive group. No significant correlations were observed between either the levels of H3T and H2T (Spearman p = 0.7), or H3T and p95 (Spearman p = 0.6).
We used the regression tree method as a multivariate analysis to scan the continuum of H2T, H3T, and p95 measurements for significant association with PFS. Subgroups of patients defined by these biomarkers were identified that experienced different clinical outcomes in response to trastuzumab-based therapies. The optimal regression tree and the Kaplan–Meier curves derived from the subgroups defined by the tree are presented in Figs. 3, 4, and 5. PFS and OS for each group are given in Table 2. The first split of the tree is based on a H2T cutoff value of 16.1, which separates patients with low HER2 expression (H2T < 16.1) on the right from those with high HER2 expression (H2T ≥ 16.1) on the left. This cutoff value is similar to the H2T ≤ 13.8 clinical cutoff previously reported , and used in the univariate analysis of HER3 expression described above. These patients are segregated next by intermediate HER2 expression (16.1 ≤ H2T ≤ 68.3) versus very high HER2 expression (H2T > 68.3). It is important to note that this HER2-intermediate group identified by regression tree analysis is essentially the HER2-positives (i.e., not HER2-low), excluding only those expressing extremely high levels of HER2, greater than approximately fivefold above the cutoff for HER2-positivity. The patients within the HER2-intermediate subgroup were further separated based on H3T and p95 expression levels (Table 2; Fig. 3). Subgroup A, characterized as H2T-intermediate, H3T-low, and p95-low, experienced the best clinical outcomes (Fig. 5; Table 2). In this cohort of HER2-positive MBC patients treated with trastuzumab, multiple factors were found to be statistically significant independent correlates of shorter PFS and OS including elevated p95 expression (p95 > 3.75), elevated HER3 expression (H3T > 3.89), very high levels of HER2 expression (H2T > 68.3), and low (normal) levels of HER2 expression (H2T ≤ 16.1).