In our current study, we demonstrated that multiplexed PCR-based methods can accurately predict ER, PR, and HER2 status in breast cancer patients when coupled with a RF-based approach that takes into consideration multiple genes that are associated with the unique biology of breast cancer. Most notably, for predicting PR status, incorporation of PR target or co-regulated genes such as
GATA3,
ESR1, and
FOXA1 adds predictive benefit when compared to
PGR alone. In the USO 01062 training set that incorporated central IHC determination, the accuracy of assigning breast cancer subtypes was much superior compared to local testing. In our training set of 24 HER2+ samples locally assessed from the USO 01062 trial, 16 were positive by central staining. This large discordance, even utilizing similar methodologies, may perhaps explain the lack of concordance between HER2 IHC local determination and our HER2 RFP algorithm in the analysis of the larger test set. Similarly, a high degree of discordance between HER2 status between local and central/reference sites has been shown in independent studies, ranging from 13 to 26.6 % [
13,
14]. Of note, we found significant survival differences in the HER2 positive group between those patients classified by IHC or RFP, an observation that was not observed in the HR+ or TNBC groups. This poorer disease-free survival observed with the HER2 RFP is more consistent with the literature suggesting a negative prognostic impact of HER2 status prior to the approval of trastuzumab-containing regimens, with only approximately 30 % of the HER2 positive patients receiving trastuzumab-based therapies following completion of the experimental adjuvant therapy [
6]. A similar gene expression analysis was carried out by Prat et al. in HER2 IHC positive breast cancer patients treated with a trastuzumab-containing regimen, in which the HER2-enriched group, as defined by PAM50, derived a significant survival benefit compared to the group defined as non-HER2-enriched [
26].
Molecular subtyping using PAM50 is not an effective surrogate for IHC status since it identifies distinct subtypes (luminal A, luminal B, basal-like, and HER2-enriched) with different prognostic implications [
9]. In a similar study assessing
ESR1,
PGR, and
ERBB2 gene expression using quantitative RT-PCR, Du, and colleagues determined that single gene expression had a concordance rate of 87 % within the luminal subtype, 75 % within the HER2+ subtype, and 48 % within the triple negative subtype [
17]. Of the 52 % discordant cases in the triple negative subtype, 12 out of the 17 (71 %) cases were determined to be luminal by RNA expression. The authors demonstrate the prognostic differences between the three IHC subtypes, with the luminal subgroup being the more favorable group followed by the triple negative and then HER2+ group. Interestingly, by RT-PCR subtyping, the HER2+ and triple negative subgroups had reversed prognostic trend, although, this may have been driven by the small HER2+ patient population in this sample set, comprising of only 9 cases. In a second study, Haibe-Kains et al., suggested that the simplest three gene classification model, SCMGENE, which comprised
ESR1,
PGR, and
AURKA was largely concordant with other more complex gene expression models in terms of subtype prediction [
27]. In addition, the authors show that their SCMGENE model demonstrated similar prognostic implications when compared to more established models such as PAM50, MAMMAPRINT, and ONCOTYPE gene expression signatures, and may be adequate for clinical management of patients. Comparing RT-PCR methods to IHC, Bastien et al. demonstrated an AUC, sensitivity, and specificity for ER to be 0.90, 0.96, and 0.74, respectively; PR to be 0.90, 0.84, and 0.85, respectively; and HER2 to be 0.95, 0.94, and 0.85, respectively [
11]. When the authors compared the prognostic differences between the different methodologies, the ER+/
ESR1+ and ER−/
ESR1+ subgroups had similar clinical outcomes, and performed better than patients who were ER−/
ESR1− or ER+/
ESR1−, suggesting that RT-PCR assignment of
ESR1 status was more prognostic and accurate than IHC for ER [
11]. This observation may be driven by the ability of RT-PCR technologies to accurately detect low percentage ER+ cells within the tumor population. As such, tumors that have ER IHC expression 1–9 % have a worse prognosis compared to tumors that have an ER IHC expression of 10 % or greater [
28], which may, in part, explain the results of Bastien and colleagues. Interestingly, we observed notable differences in the RFP for PR predicted genes based on the Allred cutoff. Using an Allred cutoff of 3,
ESR1 was the gene that best predicted PR IHC status. However, with the cutoff extended to 4,
PGR became the most predictive gene. These results suggest that extending the Allred score to 4 for a positive PR result may more accurately predict PR status, an observation that warrants further clinical investigation.
In conclusion, central confirmation of breast cancer receptor status is critical for the interpretation of clinical data from experimental trials. Our study suggests that local and central testing show substantial overall concordance, but that a PCR-based classifier offers a reasonable strategy to identify the set of samples that may have been erroneously classified and subject them to further testing. Such an approach has added valued in that it conserves 3–5 slides that would be used for repeat IHC testing, and also yields expression data on up to 90 breast cancer-related genes. As tissue can be limited in patients who have gone through multiple lines of therapy in the metastatic setting, multiplexed assays provide an attractive method of obtaining valuable biomarker data for association with clinical outcomes. We proposed two RF-based approaches, RF and RF-KNN. While both approaches performed similarly, the former classifier requires the usage of the same 96 genes for test samples, and the latter, through implementing a preliminary feature selection step, relies on a smaller set of genes for prediction, providing flexibility in panel development. In the current report, we demonstrate the utility of PCR-based multiplexed assays to accurately confirm ER, PR, and HER2 status and obviate the need for central IHC confirmation of subtype, while obtaining additional biomarker data on a diverse set of breast cancer-related genes.