The 70-gene signature analysis

A 3 mm punch biopsy of the tumor was obtained from the specimen within 15 min after surgery. Since a percentage of tumor cells >30% was required to consider the sample suitable for testing, the sample had to be obtained from a region free of necrosis or fibrosis. Tissue sample was submerged in an RNA-stabilizing solution before being collected in a dedicated delivering box. A single kit was used for each patient. Cryogenic preservation and H&E staining were performed by Agendia (Amsterdam). A customized microarray containing 70 genes was analyzed as previously described.¹⁸ Mammaprint results were expressed as high risk (poor signature) and low risk (good signature).

Immunohistochemistry

Standard immunohistochemistry analyses were performed by two of the authors (L.DT., B.F.) from our Department of Pathology, who were blinded for the 70-gene signature results. Sampling sections of different thickness were taken with microtome, 2 μm for IHC study/H&E staining and 3 μm for FISH procedure. Estrogen and progesterone receptor levels were assessed using monoclonal antibodies (Dako clone 1D5, 1:50 and PGR636, 1:50, respectively); HER2 status testing was performed using rabbit polyclonal antibodies (Dako, 1:1500); MIB-1 anti-ki-67 antibodies (Dako, 1:400) were used to determine cell proliferation rate. Fluorescence in situ hybridization test (FISH) using a Pathvysion kit (Vysis/Abbott) was performed when immunohistochemical results were equivocal (2+) as evaluated by the ASCO-CAP 2007 criteria. Tumor grade was evaluated by Elston–Ellis criteria and peritumoral lymphovascular invasion by Rosen-Oberman's.

Hormone receptor expression was defined as negative (<1%), intermediate positive (1–49%) and highly positive (≥50%).³ A cut-off of 20% was selected for Ki-67 LI according to literature data.¹⁹

Risk assessment tools and treatment recommendations

Risk assessment using criteria proposed by the St. Gallen Consensus in 2005 included clinicopathological features as age, tumor size, nodal involvement, tumor grade, peritumoral lymphovascular invasion, proliferation rate, ER, PgR and HER2 expression and defined 3 risk categories (low, intermediate and high).¹⁶

The algorithm for systemic adjuvant treatment of the NCCN guidelines 2009 (the most current at time of our study) and recommendations of the St. Gallen Consensus of 2005, 2007 and 2009 were chosen as reference for the indication of adjuvant treatment.^{9, 10, 16 and 17}

Our Institutional Multidisciplinary team (MDT) included breast surgeons, medical oncologists, radiologists, pathologists and radiation oncologists dedicated to breast cancer diagnosis and treatment. The MDT gave indications for adjuvant local and systemic treatments according to the above mentioned international guidelines. Mammaprint results were disclosed after the multidisciplinary meeting for case management discussion.

The protocol was approved by the Institutional Review Board and all patients gave written informed consent.

Statistical analysis

Data were described as numbers and percentages, or mean and standard deviation, where appropriate. Differences between groups were explored with chi square test, with fisher correction when necessary. Factors that could be prognostic of MP results (high vs. low risk) were analyzed with logistic regression. AP < 0.05 was considered significant. All analyses were made with Stata11 (www.stata.com).

Results

From July 2009 to August 2010, 124 consecutive patients attending the Breast Surgery Unit were enrolled in the study and a tumor sample was sent for the assay. Patient and tumor characteristics are summarized inTable 1. Since in the vast majority of tumors ER was expressed in ≥50% of cells, tumors were classified as ER and PgR negative (<1% of stained cells) or positive (≥1% of stained cells) for subsequent analyses.

Mammaprint results were available for 106 samples (Table 2). Eighteen cases (14%) were excluded from the analysis because the microarray was not performed due to <30% of neoplastic cells in the tumor sample. According to Mammaprint results, 53 tumors (50%) had a low risk and 53 (50%) had a high risk signature. Sixteen cases had ≥4 positive nodes, among these, 5 were classified as low risk-MP and 11 as high risk-MP (data not shown). Poorly differentiated, ER and PgR−ve, HER2+ve and highly proliferating tumors (Ki-67 ≥ 20%) were more likely to be classified as high risk-MP, while as expected, clinicopathological features as tumor size and nodal status were not associated with MP results.

We stratified MP risk categories into the following 4 immunohistochemistry (IHC)-defined tumor subtypes: ER+ve/PgR+ve/HER2−ve (N = 80), ER+ve/PgR+ve/HER2+ve (N = 11), ER−ve/PgR−ve/HER2+ve (N = 6) and ER−ve/PgR−ve/HER2−ve (N = 7). Two patients had discordant expression of hormone receptors. While the majority of cases in the latter 3 groups were high risk-MP, in the ER+ve/PgR+ve/HER2−ve group, 31 (39%) tumors were classified as high risk and 49 (61%) were classified as low risk-MP (data not shown).

We further focused the analysis in the ER+ve/PgR+ve/HER2−ve group (N = 80) with the aim of identifying clinicopathological features strongly correlated with MP results that could potentially be considered a reliable surrogate of MP. Since the predictive value of MP for the benefit of CT has been described in node-negative and in up to 3 positive node tumors,⁸ 12 tumors with ≥4 positive nodes were excluded from this analysis. The remaining 68 tumors that met these criteria were 24 high and 44 low risk-MP lesions. Table 3 shows the correlation between clinicopathological features and MP results for these 68 tumors. On univariate analysis tumor proliferation (Ki-67) had a statistically significant association with MP results while tumor size and grade showed a non-significant borderline correlation. After multivariate analysis, Ki-67 remained as the only significant predictor of MP results (OR 3.71, 95% CI 1.26–11.0, P = 0.018). Results were similar when Ki-67 was considered as a continuous variable (data not shown).

We performed a logistic regression analysis to determine the cut-off value of Ki-67 able to predict MP results. However, while none out of 44 tumors in the low risk-MP group had a Ki-67 ≥ 40%, 8 out of 24 high risk-MP tumors had a Ki-67 < 20%, precluding the definition of a clear cut-off.

When a cut-off of <15% was considered to define low and high Ki-67, the proliferation index was confirmed as the only predictive factor of the 70-gene signature both in the univariate and in multivariate analyses (data not shown). However, while among patients with high risk-MP only 2 pts had tumors with Ki-67 < 15% vs 22 with Ki-67 ≥ 15%, among pts with low risk signature this cut-off equally divided the patients in two identical groups (22 with Ki-67 < 15% vs 22 with Ki-67 ≥ 15%), failing to improve the predictivity as compared with the higher cut-off.

We also correlated MP results with St. Gallen risk categories and then focused on the intermediate risk group for which therapeutic decisions may be controversial (Table 4). Of 106 tumors, 80 were classified as intermediate St. Gallen risk; 35 were high and 45 low risk-MP. The 70-gene signature results showed a statistically significant correlation with ER and PgR status, HER2 overexpression and proliferation rate. Multivariate analysis showed that Ki-67 remained as the single predictor of MP risk also in this subset (OR 6.51, 95% CI 2.31–18.34, P < 0.001) (Table 4).

Finally, we correlated MP results with recommendations for adjuvant systemic treatment drawn from international guidelines as NCCN 2009 and St. Gallen Consensus 2007 and 2009.^{9, 10 and 17}

Overall, a higher number of pts were assigned CT in addition to endocrine therapy (ET) by the NCCN algorithm as compared to St. Gallen (35 vs 12), especially in the low risk-MP group (22 vs 5). Conversely, the number of candidates for ET with an undetermined indication for chemotherapy (ET ± CT group) was higher according to St. Gallen recommendations than to NCCN guidelines (25 vs 19 in high risk-MP and 43 vs 27 in low risk-MP, respectively).

With special focus on the ET ± CT group, we correlated indications for systemic treatment given by our institutional MDT with MP results (Table 5). Endocrine therapy alone was administered in 11/19 (58%) and in 11/25 (44%) pts (according to NCCN and St. Gallen, respectively) with high risk-MP. Conversely, ET plus CT was administered in 2/27 (7%) and 9/43 (21%) pts (by NCCN and St. Gallen, respectively) with low risk-MP. Overall, MP results would have led to different treatment decisions in 13/46 (28%) and in 20/68 (29%) pts according to NCCN guidelines and St. Gallen recommendations, respectively. Discordant cases represent patients with MP-low risk who were assigned to ET and CT and those with MP-high risk assigned to ET alone. When the indications of our MDT in the ET ± CT group were correlated with Ki-67 (defined by a cut-off of 15%), a discordance was observed in 20/46 (43%) and 23/68 (41%) in the NCCN and the St Gallen groups, respectively, attributable to a greater indication for ET alone in pts with high Ki-67 tumors, while the proportion of pts with low Ki-67 tumors who were indicated ET + CT was decreased as compared to MP categories (data not shown).

Discussion

The use of multigene assays has improved the prognostic definition of breast cancer as compared to that of standard clinicopathological features.⁴ Whether this leads to a better prediction of the benefit of adjuvant systemic treatments is currently undefined and represents the primary endpoint of large ongoing or recently completed randomized prospective trials.^20 and 21

Different from other gene signatures such as Oncotype DX recurrence score and Genomic Grade Index, the 70-gene prognosis signature development was not driven by a biological hypothesis and initially it was validated in both, ER+ve and ER−ve tumors.^4 and 5

Subsequent studies have aimed to refine the proper setting where the use of the 70-gene signature may add a value beyond that of standard clinicopathological features, focusing on specific patient and tumor characteristics (nodal status, tumor size, menopausal status).^{12, 13 and 14}

In the present study we attempted to determine whether any of the standard routinely available clinicopathological features was associated with the 70-gene signature risk categories and could be considered a reliable surrogate of MP results. We focused our analysis on tumor subsets that raise more challenges for treatment decisions.

First, we considered a subset of tumors that are generally assumed to be quite homogeneous for clinicopathologic characteristics as ER+ve/PgR+ve/HER2−ve tumors. A better definition of further subgroups within these tumors is particularly relevant because of the existing controversies on the benefit of adjuvant chemotherapy, even in pts with nodal involvement.^7 and 8 Our population of ER+ve/PgR+ve/HER2−ve tumors could grossly represent luminal A tumors given that the vast majority of them expressed very high levels (>50%) of both ER and PgR and a median proliferation rate of 17%. However 35% of these tumors had a high risk signature. As expected, clinicopathological features as nodal status and tumor size, which are routinely used in clinical practice to identify prognostic subgroups were not associated with the Mammaprint results. The only significant predictor of the 70-gene signature category was tumor proliferation measured by Ki-67. Ki-67 ≥ 20% was associated with a nearly 4-fold increased likelihood of having a high risk-MP result.

These results are biologically plausible and consistent with previous data indicating proliferation as the major biological pathway involved in the growth of ER+ve/HER2−ve tumors.²² A meta-analysis of microarray datasets performed by independent groups suggested that the assignment to different prognostic categories largely, if not exclusively, depends on the expression of proliferation related genes and that their prognostic power is limited to ER+ve tumors since the proportion of low risk ER−ve tumors ranges from 0 to 5%.²³

Unfortunately we failed to settle a definite cut-off for Ki-67 to discriminate high from low risk tumors, as one third of tumors in the high risk signature had a Ki-67 < 20%. The small sample size as well as biological heterogeneity may account for this finding.

Despite a number of methodological flaws which might adversely affect Ki-67 reproducibility, its measurement is acquiring a growing role as a prognostic marker.^{19, 24 and 25} Cut-off values of 11% or of 14% have been proposed to distinguish luminal A from luminal B tumors.^{24, 26 and 27} Other studies have used a value of 20% for dividing high from low proliferating tumors.^19 and 25 Differences in methods used across studies to determine Ki-67 preclude the definition of a consistent cut-off that could be used in clinical practice.²⁸ This inconsistency may also have a biological rationale since it has been proposed that the expression of proliferation related genes is a continuum and that the cut-off established to split luminal A from luminal B tumors may be arbitrary.^23 and 29 Tumor heterogeneity and different proliferation areas within the tumor may also account for inconsistencies between low proliferation and high risk signature.

Other studies have compared the 70-gene signature prognostic groups with the St. Gallen categories, focusing on the intermediate risk category.^{13, 14 and 15} This group includes, by definition, tumors with heterogeneous characteristics and raises controversies for treatment decisions. In our study we observed a fairly even distribution of these tumors between the 70-gene high and low risk categories with 35 tumors falling in the former and 45 in the latter categories. Proliferation rate was confirmed as the main predictor of the 70-gene signature results also in this group.

Literature reports show that the use of gene signatures may lead to a change in treatment indications given according to standard guidelines in 20–70% of cases depending on the guideline, the assay and the population examined.^30 and 31 Since the added prognostic value of multigene signatures (compared to clinicopathological features) is the identification of a larger number of low risk tumors, the use of these assays usually leads to a decrease in indication for CT. In the present study we selected a very challenging group, the one with undetermined benefit from CT according to NCCN and St. Gallen recommendations.^9 and 10 We confirmed that overall the results of the 70-gene signature would have changed the indications of the MDT in about 30% of cases. Our findings go along with the design of the MINDACT trial where a proportion of about 30% of tumors is expected to show discordancy between clinicopathological features and the 70-gene signature results.¹⁹ In our study, MP results would have indicatated CT in 58% of cases in whom, according to NCCN criteria, its benefit is undetermined and in whom our MDT prescribed ET alone. The indication for CT would have been 44% if St. Gallen criteria were applied to define the ET ± CT subgroup. This discordance was lower in the MP-low risk group, especially for the NCCN group. Ki-67 correlated more closely with MDT indications for ET than did low MP-risk, but it would have shifted an even greater proportion of pts to receive ET + CT than the high risk signature.

Conclusions

In conclusion, the use of the 70-gene signature was feasible in our cancer center with 86% of samples being adequate for the assay. A substantial proportion of ER+ve/PgR+ve/HER2–ve tumors with pathological features resembling luminal A subtype were assigned to the high risk group by the 70-gene signature. Proliferation index was the only strong predictor of MP results in this subgroup although a cut-off value was not identified.

In pts with undetermined benefit for CT according to NCCN and St. Gallen, MP results would have led to discordant recommendations for adjuvant therapy compared to those based on standard clinicopathological features in about 30% of pts increasing the indication for CT.

The results of MINDACT and other large randomized trials are warranted in order to understand whether we should rely on multigene assays rather than on standard clinicopathological features for treatment decisions.

Conflicts of interest statement

The author(s) indicated no potential conflict of interest. No member of Agendia participated in study design, analysis of results or manuscript writing.

References

Copyright © 2013 Elsevier Ltd. All rights reserved.