Breast cancer molecular classification: evaluation of the gene expression profile and comparison with immunohistochemical parameters

Background. Breast cancer is heterogeneous disease, with different patterns of gene expression leading to differences in behavior and prognosis (Perou CM, 2000; Sorlie T, 2001; Sorlie T, 2003). A number of factors such as tumour size, histologic grade, regional lymph node involvement, lymphovascular invasion, and expression of both the estrogen (ER) and progesterone (PGR) hormone receptors, her2 (human epidermal growth factor receptor 2) protein overexpression or gene amplification (or both) and Ki67 value are considered important prognostic and predict factors. The use of these common parameters and the methods used to evaluate them have shown poor prognostic value and a limited ability to predict the clinical evolution of the disease. The literature data show that only a part of the patients undergoing adjuvant chemotherapy, on the basis of these criteria, receives a real benefit from the treatment (EBCTCG, Lancet. 2005). In the recent years, breast cancer management is already changing in light of the new molecular analysis. In 2000, through the study of 8,102 genes, Perou showed that the phenotypic diversity of breast cancer corresponded to specific gene expression patterns (Perou CM, 2000). This study identified four groups of samples that might be related to different molecular features of mammary epithelial biology: ER+/luminal-like, basal-like, Erb-B2+ and normal breast. On the basis of Perou data, in 2009 Parker selected 50 genes that showed the capacity to predict disease recurrences, named PAM50 (Prediction Analysis of Microarray) (Parker JS 2009). Consequently, breast cancer was classified into four subtypes based on the differences in molecular patterns: Luminal A, Luminal B, HER-2-enriched and Basal-like (Goldhirsch A 2013). Subsequently, breast cancer molecular classification was adopted in St. Gallen Consensus 2011 that combined molecular classification with specific therapies. In particular, for luminal A breast cancer the therapy is endocrine therapy while for luminal B breast cancer endocrine plus cytotoxic therapy (Goldhirsch A, 2011). PAM50/PROSIGNA (Nanostring technology), analyzing a panel of 50 genes classified breast cancer in four intrinsic molecular subtypes (Luminal A, Luminal B, HER-2-enriched, and Basal-like ) and, correlating this result with the stage of the disease, it estimates the risk of recurrence to 10 years after surgery. Currently, although the Panel experts of recent guidelines believe strongly that genomic assays, such as PAM50, are valuable for determining whether or not to recommend adjuvant chemotherapy in women with breast cancer, however tests are not universally accessible largely owing to costs above routine pathology testing (Burstein H. J. 2019). Consequently, in clinical practice, immunohistochemistry (IHC) determination of ER level, PR level, HER2 status and Ki67 proliferation serves as a surrogate for broad classification of ER-positive tumors into more favorable 'luminal A-like' or less favorable 'luminal B-like' cancers (Goldhirsch A 2013; Senkus E 2015; Burstein H. J. 2019). Nevertheless, there are many pre-analytical and analytical factors that introduce variability in the IHC assay results. Moreover, quantitative evaluation of immunohistochemistry is not standardized among different laboratories, and there is a lack of criteria for the definition of IHC positivity (Dowsett M, 2011; Luporsi E, 2012; Hashmi A, 2019). In particular the choice of the Ki67 cut-off has a major impact in clinical practice in terms of predict and prognostic features (Nielsen OT, 2021). The definition of the positivity threshold of Ki67 is a topic widely debated in literature. St. Gallen consensus Expert Panel 2013 suggested 14% as cut off for Ki67, because his value correlated with the gene-expression definition of Luminal A breast cancer (Goldhirsch A, 2013). However in a foot note of this document, the majority of expert panel voted that a threshold more o equal to 20% was indicative of high Ki67 status (Goldhirsch A, 2013). Subsequently, the 15th St. Gallen panel proposed that Ki67 scores should be interpreted in light of local laboratory values, and recommended to use the median expression of each lab to define high and low Ki67 values (Coates AS, 2015). The current guidelines remain skeptical about the technical validity of Ki67 IHC assays (Burstein HJ, 2019) because for any tumor biomarker test, including Ki67, there are many factors that may affect the result, including collection, processing, and archiving of the specimen (preanalytical) to staining, analysis, and reporting (analytical), and finally to ensuring ongoing quality of the analytical assessment (Nielsen TO, 2021). Consequently, treatment allocation may be based on a suboptimal way of capturing the tumor's biological pathway, and thus its ability to respond to a certain treatment. On these basis the aims of this work are: AIM 1. Comparison between PROSIGNA/PAM50 molecular intrinsic subtype and IHC-based subtypes classification in patients with luminal early breast cancer. Moreover, the comparison between the IHC evaluation and molecular classification should allow to identify the subgroup of patients with breast cancer for which the evaluation of the gene expression profile is needful for a correct diagnostic-therapeutic framework. AIM 2. The reassessment of IHC data by a single pathologist with experience in breast cancer and comparison between this new revised IHC-based subtypes and PAM50 molecular classification, with the aim to reduce inter-observed variability and to assess its impact on the concordance between PAM50 and IHC breast cancer classification. AIM 3. The analysis of our cohort with PREDICT2.0 and comparison between risk of recurrence evaluated by PROSIGNA (ROR score) and treatment benefit predictions according PREDICT 2.0. Moreover, today a number of predictive models are now available to help estimate the survival for individual patients. Among these, Predict is an online prognostication and treatment benefit following surgery for women with invasive breast cancer (Gordon WC, 2010). In particular, Predict estimates the benefit of chemotherapy in women with breast cancer based on clinicopathological factors such as tumor size, tumour grade, lymph node status, ER expression, HER2 status and mode of detection. We evaluated the concordance between the risk class (ROR score) assigned by the PROSIGNA test and the benefit obtained from the adjuvant treatment according to PREDICT. Methods and materials PAM50/Prognostic Gene Signature Assay and IHC data were evaluated in 132 patients with early breast cancer between June 2014 and September 2020 at Campus Bio-Medico University of Rome - Molecular Diagnostic Lab. The inclusion criteria are: postmenopausal status, hormone receptor-positive (ER+/any PGR) (luminal breast cancer), Her2 negative (IHC or IHC/FISH), node-negative or node-positive early-stage (1-3 lymph-nodes) and tumor size greater than 0,4 centimeter. Immunoistochemical subtypes were defined according international guidelines (Goldhirsch A 2013; Coates AS, 2015). Before we defined Luminal A-like or Luminal B-like tumors according to the IHC surrogate definitions of breast cancer subtypes proposed in the Goldhirsch A, 2013 guideline (Goldhirsch A, 2013): Luminal A-like tumors were defined as HER2-negative, ER positive with a low Ki67 assessment (<14% and <20%) and Luminal B-like tumors were defined as HER2-, ER positive with a high Ki67 determination (=14% and =20%). Moreover, IHC surrogate definitions of breast cancer subtypes were defined according St Gallen consensus 2015: a median of Ki67 score in our laboratory was 20%, therefore a value of 30% or above could be considered clearly high and a value of 10% or less clearly low (Coates AS, 2015). For aim 2, in 86 cases the immunoistochemical slides were available for the re-evaluation by a pathologist expert in breast cancer. Subsequently, for aim 3 the benefit of chemotherapy was calculated for all patients using the Predict 2.0 online tool. Results Aim 1. Comparing PAM50 molecular classification with IHC surrogate subtypes classification applying 14% cut-off of Ki67, 45/132 (34,1%) cases were discordant: 27/132 (20,5%) breast cancer were classified as Luminal A-like breast cancer, but in this subset of patients 4/27 (14,8%) of these resulted as Luminal B at PAM50 test; 105/132 (79,5%) breast cancer were defined as Luminal B-like according IHC evaluation, but in this subset of patients 41/105 (39%) resulted Luminal A at PAM50 test (K value: 0.31, minimal agreement; 95%IC: 0.17-0.44 (Pr(>z) <0,0001). Using Ki67 cut-off of 20%: 52/132 (39,4%) patients were classified as Luminal-A like according IHC, while in this group PAM50 showed 17/52 (32,7%) Luminal B tumors; and 80/132 (60,6%) breast cancer were defined Luminal B like but PAM50 identified 29/80 (36,2%) Luminal A ones. Comparing the two subtyping methodologies, using 20% Ki67 cut-off, 46/132 (34,8%) of tumors had a different classification (K value: 0.29, minimal agreement; 95%IC: 0.13-0.45 (Pr(>z) 0,0002). According St. Gallen consensus 2015: 23/132 (17,4%) breast cancer should be classified as Luminal A like and 34/132 (25,8%) cases as Luminal B like, while 75/132 (56,8%) cancer should be classified as "Luminal intermediate" breast cancer. In cases with Ki67=10% PAM50 test identified 4/23 (17,4%) Luminal B breast cancer and in the group of cases whit Ki6=30% PAM50 showed 4/34 (11,8%) Luminal A breast cancer. Moreover, in the subset of patient with Ki67 intermediate (10%<30%), PAM50 assay identified 41/75 (54,7%) Luminal A and 34/75 (45,3%) Luminal B breast cancer. Considering case with Ki67=10% and Ki67=30%, the concordance between IHC and PAM50 evaluation is good (K value: 0.71, moderate agreement; 95%IC: 0.52-0.89 (Pr(>z) <0,0001), but it should be noted that in 75/132 (56,8%) ("Luminal indeterminate") the IHC approach failed to distinguish between Luminal A and Luminal B breast cancer. Aim 2. If we evaluate the concordance between PAM50 and IHC breast cancer classification after slides re-evaluation we see that when we using 14% cut-off of Ki67, 21/86 (24,4%) breast cancer were discordant between two subtyping methodologies (K value: 0.52, weak agreement). Moreover if we applied 20% cut-off of Ki67, 20/86 (23,2%) cases resulted discordant between IHC evaluation and PAM50 test (K value: 0.53, weak agreement; 95%IC: 0.35-0.71 (Pr(>z) <0.0001). According St Gallen 2015 recommendation, 32/86 (37,2%) breast cancer should be classified as Luminal A like and 17/86 (19,8%) cases as Luminal B like, while 37/86 (43%) cancer should be classified as "luminal intermediate" breast cancer. In the group of case with Ki67=10% PAM50 test identified 2/32 (6,2%) Luminal B breast cancer and in the group whit Ki67=30% PAM50 showed 1/17 (5,9%) Luminal A breast cancer. Moreover, in the subset of patient with Ki67 intermediate (10%<30%), PAM50 assay identified 19/37 (51,4%) Luminal A and 18/37 (48,6%) Luminal B breast cancer. Considering case with Ki67=10% and Ki67=30%, the concordance between IHC and PAM50 evaluation is almost perfect [K value: 0.86 (almost perfect agreement); 95%IC: 0.72-1 (Pr(>z) <0,0001)], but it should be noted that 37/86 (43%) patients are classified as Luminal indeterminate breast cancer. Aim 3. In all cases the "additional benefit" of chemotherapy in terms of OS (overall survival) were evaluated using Predict 2.0 and these results were compared with risk of recurrence (ROR score) assessed by PAM50/PROSIGNA assay. In cases with chemotherapy treatment benefit <3% and >5% the concordance between two tests were moderate (K value = 0,59). In 29/132 (21,9%) Predict tool does not give precise indications about the prescription of chemotherapy (chemotherapy estimated between 3% and 5%) and in this group of patients the PROSIGNA test solved the "uncertainty" of the PREDICT 2.0 results in 22/29 (75,9%) cases: 16/29 (55,2%) resulted as "high risk" and 6/29 (20,7%) as "low risk". Only 24,1% (7/29) of patients showed intermediate risk according to PROSIGNA Conclusions The data showed that the concordance rates between Prosigna/PAM50 subtype (i.e. Luminal A vs. Luminal B) and IHC subtype (Luminal A-like vs. Luminal B-like) using Ki67 cutoffs of 14% and 20% is very low. Instead, according St Gallen 2015 there was a good concordance between immunohistochemistry and PAM50 classification in cases of low and high Ki67 value (kappa value = 0,7 moderate agreement) , even if identified an intermediate group (10%<30%) which in our series includes 75 (56,8%) cases, for whom immunohistochemistry does not allow to identify patients who could benefit from cytotoxic therapies. The re-evaluation of Ki67 by a single pathologist with expertise in breast cancer, even if it proved that a more accurate evaluation of ki67 improves the agreement between IHC and PAM50 classification, nevertheless it confirmed a substantial discrepancy between the two methods when using only one ki67 cut-off (K value = 52 and 53 using 14% and 20% cut-off of Ki67, moderate agreement) and a strong agreement for very high values (>30%) and very low (<30%) of ki67 (K value = 0,87). Nevertheless, according St Gallen consensus 2015, there are an intermediate range of ki67 which included 43% breast cancer of our cohort, for which the evaluation of the gene expression profile is essential for a correct therapeutic approach. Moreover, the last objective of the study enabled to evaluate the correlation between the risk of recurrence class (ROR score) assigned by the PROSIGNA test and the benefit of the adjuvant treatment according to PREDICT2.0. The analysis of our data showed that PREDICT 2.0 correlates with PROSIGNA results (ROS score) when the advantages obtained from chemotherapy is >5%, however, when it is <5%, the tool is not reliable. In cases of advantages obtained from chemotherapy is comprised between 3% and 5% according Predict 2.0, the role of PROSIGNA/PAM50 evaluation is crucial: the molecular test allows a correct classification in the 75,9% of patients in this group (55,2% high risk, 20,7% low risk). In particular, in patients for whom the benefit of chemotherapy is doubtful (PREDICT2.0 result between 3% and 5%), PROSIGNA allows a more correct classification, reducing the number of "intermediate cases". It should be considered, in fact, that the data used by PREDICT2.0 to define the usefulness of an adjuvant treatment include immunohistochemical data which, as already underlined, are characterized by important limits and poor reproducibility of the results. In conclusion, although IHC evaluation, in particular Ki67 value, has repeatedly shown to be prognostic and predictive of chemotherapy response, the clinical value of Ki67 in identifying low risk outcome patients or Luminal A disease who might be safely spared chemotherapy remains uncertain. Immunohistochemical evaluation is not efficient for the biological classification of the disease in cases with ER+ and intermediate Ki67 value (between 10% and 30%). The data suggested that immunoistochemistry assay could be considered a good "screening test" to identify patients who should be tested with molecular test. In the subset of cases with ki67 intermediate (Ki67 grey zone) PAM50 test could facilitate the oncologist to establish the best therapeutic option for breast cancer patients.