Background
Molecular analyses of primary breast cancer for both research and patient management are now commonplace. Measurements may be made on diagnostic core-cut biopsies or surgical excisions that frequently comprise a very small fraction of the tumour. In so-called window-of-opportunity studies patients are exposed to medical therapy between diagnosis and surgery [
1] and comparisons are made between samples taken at both time points. Valid interpretation of these studies depends on knowledge of any variability or systematic changes in the respective biomarkers that occur in the absence of treatment. Variability/heterogeneity may lead to false rejection of a true effect while systematic differences between diagnostic and surgical specimens may lead to artifactual changes being falsely ascribed to an intervention. For example, we have previously described the highly significant impact of specimen type (core-cut vs excision) on pAKT and pERK1/2 staining [
2]. Pre-treatment/post-treatment comparison of biomarkers might also be affected by the taking of the diagnostic biopsy and changes due to cold ischaemia between resection and tissue stabilisation/fixation.
The effect of cold ischaemia time has been studied in small cohorts of breast cancer with up to 24 hours elapsed time before fixation, snap freezing or placement in RNA later [
3‐
5]. No studies have directly examined the impact of the short time delay (20–60 minutes) resulting from sending specimens for X-ray, a frequent practice during breast cancer surgery to ensure the removal of the lesion (e.g. non-palpable mass, calcifications) and/or to check for adequate surgical margins, even in clinically palpable tumours. A small number of studies have evaluated gene expression changes over a longer period of time between biopsies [
6‐
8]. For example, Jeselsohn identified 14 genes, including nine immune-related that differed between core-cuts and excision taken from 21 patients 6–65 days apart (mean 30 days).
Our primary objectives were to use genome-wide expression profiling to determine more comprehensively the variability and systematic changes in the expression of genes or pre-specified gene sets or subtype classifications (i) between two core biopsies taken (A) immediately after excision and (B) after sample X-ray and (ii) between diagnostic core biopsies (D) and surgical core biopsies (S) two weeks later in the absence of any intervention.
Discussion
Multiple issues relating to intra-tumoural heterogeneity are at the forefront of contemporary molecular pathology. One concerns the degree to which a single core-cut biopsy can represent a biomarker’s expression across the tumour. We assessed this using a genome-wide approach. We also determined whether two common clinical practices around the time of surgery significantly affected the expression of particular genes or activation of certain pathways. Systematic changes resulting from either process would be relevant to any studies of excised breast cancer, since virtually all excisions occur after diagnostic core-cut and many will involve X-ray of the tumour before its fixation/stabilisation. Data from other studies may differ due to differences between the analytical platforms used.
The variability in whole genome expression data between tissue samples taken peri-surgically has been studied in only small tumour sets (greatest number 13, discussed below) [
4‐
7]. Pure study of intra-tumoural heterogeneity is best conducted by taking multiple samples from a tumour at the same time. However, the systematic changes occurring in our studies were very modest and will have had little to no perceptible impact on the overall correlations observed. The range of correlations was similar across both studies and overall provided data on 79 tumours. The poorest of the correlations was 0.86 with the large majority being above 0.95 and several being >0.99. Thus gene expression overall shows only modest variability across tumours.
Most investigators are more interested in the variation in expression across the tumour for their gene or genes of interest. Our on-line data (
http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE73237) will allow them to evaluate that. For illustration we chose 18 genes frequently studied in breast cancer. In general the correlation of the individual genes between the samples was higher for those genes with wide ranges, e.g.
TFF1 (6-log2 range) and
ERBB2 (5-log2 range) than those with narrow ranges, e.g.
SNAI2 (1.5-log2 range) and
MKI67 (<1.0-log2 range). The correlations between individual genes were all worse than those for the genome-wide analyses where there was an approximately 8-log2 range of expression.
We have previously reported that the 60-minute delay in fixation in study I had no significant impact on immunohistochemical expression of ER, PgR, Ki67, HER2, pAKT or pERK1/2 [
2]. Similarly, no genes were found to differ at an FDR <0.05. However, several genes related to stress (e.g.
DUSP1) and/or known as early-response genes (e.g.
RGS1,
RGS2, and
FOSB) were among those most highly ranked according to change. In study II, where the larger number of samples provided greater statistical power, the same genes (e.g.
RGS1, FOSB and
DUSP1) or similar genes (e.g.
FOS) ranked in the top ten genes with changed expression. This suggests that the changes in these early-response and stress pathways were true findings in both studies. It is important to note for study II that no record was made in POETIC of whether excised tumours were subject to X-ray before taking of RNAlater-stored core-cuts. At the Royal Marsden all impalpable tumours and most tumours resected via wide excision (totalling about 50 % of operations) are X-rayed. We have informally determined that similar approaches are in place across the UK. Some of the similarities in the genes changing between the studies may therefore have been due to a proportion of the tumours in study II being subjected to X-ray before stabilisation. It should be noted, however, that while the similarities in the gene changes between the two studies are consistent with delays due to X-ray being responsible in study II, there are multiple other factors that occur around surgery that could also contribute. These include the time taken for a sample to reach histopathology, where some centres may have taken cores for the POETIC study, and delays due to sentinel node biopsy, which may have occurred prior to the core being taken. Nonetheless the changes observed in study II are likely to represent those that occur between diagnostic and surgical samples in common practice and will affect the measurement/study of early-response genes in excised tumours.
Two smaller studies have assessed the impact of delay to fixation on global gene expression [
4,
5]. In the Borgan study, changes in
FOSB and
JUND, while perceptible after 60 minutes, were much greater after 3 hours. The correlation of these changes with time since tumour removal make it likely that they are due to stress of tissue cutting and/or its exposure changed oxygen tension as opposed to the impact of other procedures around surgery such as anaesthesia. The pathway and network analyses undertaken with study I revealed changes in oxidative phosphorylation and mitochondrial dysfunction. This is also consistent with the exposure of the core-cuts to changed oxygen tension or ischaemia. The correlation of mitochondrial dysfunction also correlated quantitatively with time between core-cut taking and fixation supports this change being causatively associated.
Despite the lack of change in the pre-specified immune signatures
IL6 expression increased in both studies and was among the genes identified by Jeselsohn in a similar but smaller study. The change in
IL6 levels in study II was sufficiently heterogeneous between tumours to nullify the highly significant correlation between the A and B samples in study I, suggesting that the
IL6 changes were more related to the effects of the initial biopsy than to the short delays around surgery.
IL6 is a pleiotropic cytokine secreted by T cells and macrophages in both systemic and localised immune activation. Its role in breast cancer has been reviewed by Dethlefsen and colleagues [
12]. Changes in
IGFBP2 and particularly
MYC in study II also confirmed those seen in the Jeselsohn study, but there was little support for the other ten genes identified as significant in that study. Like
IL6 these two genes are widely studied in breast cancer. Interpretation of data on them must take account of the effects of diagnostic biopsies.
Some smaller genome-wide analyses between paired biopsies either side of surgery have been reported. Riis et al. [
7] studied 13 patients with the time between diagnostic and surgical samples ranging between 2 and 8 weeks. As in the current study genes related to early response, including
FOSB and to oxidative stress including
DUSP1 were differentially expressed between the two samples. Similar increases in early-response genes including
FOS were also reported in 16 patients in whom fine-needle aspirates were taken presurgically and immediately after tumour excision but the time between samples was not stated [
8]. Neither of these small studies, identified
IL6, IGFB2 or
MYC as a changing gene but may have been due to their low statistical power.
There were no systematic differences in categorisation of the tumours into the intrinsic subgroups in either study but discordance was noted between the luminal A versus B subtypes, even after quality control of the RNA and removing technical platform bias with normalisation and standardisation of expression profiles. In study II, 15–20 % of tumours considered luminal A on one core-cut were typed as luminal B or normal-like on the other. Allocation of subtypes is made according to the highest correlation coefficient with the archetypical centroid for each subtype irrespective of the proximity of the correlations to the subtypes, although an early report [
13] described 43/115 (37 %) of tumours as having a low correlation to any of the subtypes. Not surprisingly, we found that subtype discordances were largely associated with small differences between correlations with luminal A and luminal B centroids. The level of discordance in subtyping is important to appreciate given the prominence of intrinsic subtyping in clinical studies of breast cancer and its use for determining whether to allocate chemotherapy [
14].
Acknowledgements
This study was funded in part by Mary-Jean Mitchell Green Foundation, Breast Cancer Now Research Centre. We acknowledge National Health Service (NHS) funding to the National Institute for Health Research (NIHR) Biomedical Research Centre at the Royal Marsden Hospital. The POETIC trial (C1491/A8671/CRUK/07/015, C1491/A15955, C406/A8962), from which samples were obtained for this study, was supported by Cancer Research UK as is the Institute of Cancer Research Clinical Trials and Statistics Unit (ICR-CTSU) through its core programme grant.
The study sponsors had no involvement in the design of this study, the literature review, data interpretation, writing of the manuscript or the decision to submit it for publication.
Competing interests
MCU Cheang and J Parker are listed as co-inventor for the PAM50 gene expression classifier patent.
Other authors declare that they have no competing interests.
Authors’ contributions
ELK extracted RNA from study II, analysed the data and drafted the manuscript. QG analysed the data and drafted the manuscript. MC contributed to the statistical design and interpretation of data, analysis of the intrinsic subtypes and drafting the manuscript. JP did the intrinsic subtype classifier and drafted the manuscript. LAM contributed to the interpretation of data, review and revision of the manuscript. IP assembled samples and extracted RNA from study I. MH, LZ, SD and MA sectioned and reviewed the histopathology of the samples. AD was immunohistochemistry coordinator and allowed data acquisition by reviewing the histopathology. JM provided data and composed Additional file
2: Table S1. FM contributed to study design and obtained the samples for study I. AS, CH and JR contributed to study conception and provided patient recruitment. IS, JB and MD were involved in conception and design, and drafting of the manuscript. All authors revised and approved the final manuscript.