Combined proliferation and apoptosis index provides better risk stratification in breast cancer

Breast cancer (BC) risk stratification is critical for predicting behaviour and guiding management decision‐making. Despite the well‐established prognostic value of cellular proliferation in BC, the interplay between proliferation and apoptosis remains to be defined. In this study, we hypothesised that the combined proliferation and apoptosis indices can provide a more accurate in‐vivo growth rate measure and a precise prognostic predictor.


Combined proliferation and apoptosis index provides better risk stratification in breast cancer
Aims: Breast cancer (BC) risk stratification is critical for predicting behaviour and guiding management decision-making. Despite the well-established prognostic value of cellular proliferation in BC, the interplay between proliferation and apoptosis remains to be defined. In this study, we hypothesised that the combined proliferation and apoptosis indices can provide a more accurate in-vivo growth rate measure and a precise prognostic predictor. Methods and results: Apoptotic and mitotic figures were counted in whole slide images (WSI) generated from haematoxylin and eosin-stained sections of 1545 BC cases derived from two well-defined BC cohorts. Counts were carried out visually within defined areas. There was a significant correlation between mitosis and apoptosis scores. High apoptotic counts were associated with features of aggressive behaviour, including high grade, high pleomorphism score and hormonal receptor negativity. Although the mitotic index (MI) and apoptotic index (AI) were independent prognostic indicators, the prognostic value was synergistically higher when combined. BC patients with a high combined AI and MI had the shortest survival. Replacing the mitosis score with the mitosis-apoptosis index in the Nottingham grading system revealed that the modified grade with the new score had a higher significant association with BC-specific survival with a higher hazard ratio. Conclusion: Apoptotic figures count provides additional prognostic value in BC when combined with MI; such a combination can be implemented to assess

Introduction
Breast cancer (BC) is the most frequent cancer and the second leading cause of death in women globally. 1,2 Tumour intrinsic features, such as proliferation, have been shown to have a substantial prognostic value, and their evaluation aids in the prognostic stratification of BC patients for therapeutic approaches. [3][4][5][6][7] Proliferation in BC is commonly assessed in routine practice using mitotic counts or the Ki67 labelling index. 8 Despite the importance of proliferation, the net BC growth rate relies upon the combination of cell division and cell loss. 9,10 Apoptosis can be used as a surrogate to assess cell loss, as it exhibits morphological features that can be used in its identification and measurementopposite to necrosis, which leads to loss of cellular details and is usually affected by other factors such as inflammation and hypoxia/angiogenesis.
Apoptosis is a strictly controlled programmed cell death with different biochemical and genetic mechanisms that play an important role in tissue homeostasis, and maintains the balance between cell survival and cell loss. 11 The impact of apoptosis in cancer has been the focus of research over several years [12][13][14] ; however, there are conflicting results regarding its prognostic value in BC. [15][16][17][18][19][20][21][22] The assessment of apoptosis as a prognostic marker in BC faces several limitations, including the diversity of its morphological appearances, the assessment subjectivity and the controversy regarding its prognostic value. However, the advent of digital pathology and artificial intelligence to provide an objective assessment of obvious and subtle morphological features utilising digitalised whole slide images (WSI) stimulated our interest in the assessment of such morphological features in potential prognostic applications.
The aim of the current study is to demonstrate the prognostic significance of apoptosis in BC and its molecular intrinsic subtypes. We also hypothesised that combining the apoptosis index (AI) with the mitosis index (MI) would provide a more precise evaluation of tumour biology and behaviour. Moreover, differential gene expression analyses of the cases based on apoptotic and mitotic indices were performed to decipher the potential underlying biological changes.

S T U D Y C O H O R T S
This study was conducted on two BC cohorts. (1) The Nottingham cohort (n = 715) included primary earlystage BC patients who presented to Nottingham City Hospital, Nottingham, UK between 1990 and 1998. 23 This is a well-characterised cohort with available clinicopathological data, including patient age at diagnosis, tumour grade, size, histological tumour type, the Nottingham prognostic index (NPI), molecular subtypes and treatment and outcome data. 24 The outcome events included BC-specific survival (BCSS), defined as the time (in months) from the date of the primary surgery to the time of death from BC, distant metastasis-free survival (DMFS), defined as the time (in months) from the primary surgery until the first event of distant metastasis and recurrencefree survival (RFS) defined as the time from date of curative surgery to the time of recurrence event. 3 The median follow-up time for this cohort was 178 months [interquartile range (IQR) = 90-232 months]. Haematoxylin and eosin (H&E)-stained slides were processed and scanned into high-resolution (0.19 lm/pixel) WSIs using a high-throughput scanner (Pannoramic 250 Flash III; 3DHistech, Budapest, Hungary), and viewed with CASE-VIEWER (version 2.2.0.85; 3D-Histech) on a full-screen panel (size = 27 inches; resolution = 1366 9 768). (2) For the Cancer Genomic Atlas Breast Cancer (TCGA-BRCA) cohort (n = 830), 25 the H&E images of this cohort were digitised and obtained from BC excision specimens from patients who have a median clinical follow-up of 31 months (IQR = 17-60 months). Images and clinical data of the TCGA database are available from the website of Cancer Genomics Browser of the University of California Santa Cruz (https://genome-cancer.ucsc.edu/). All the images were directly down-loaded from the cBioPortal website and viewed on the Aperio ImageScope. 26 The details of follow-up data of both cohorts are summarised in the Supporting information, Table S1.

A P O P T O S I S A N D M I T O S I S A S S E S S M E N T
In the Nottingham cohort four slides were evaluated for each case; the one with the highest proliferation index was selected for the study, while the available case images were used in the TCGA cohort. Apoptotic bodies and mitotic figures were scored within the same (3 mm 2 ) area using digitalised WSIs. Counting was performed within the mitotic hot-spot, defined as an area with the highest number of mitotic figures observed at low power and distant from in-situ carcinoma, necrotic and sclerotic areas. 27 Apoptosis was identified morphologically by the presence of a constellation of morphological features including cellular shrinkage, densely eosinophilic cytoplasm, loss of cellto-cell contact leaving a clear halo surrounding cells, as well as nuclear features such as nuclear pyknosis (small and condensed nuclei), karyorrhexis (nuclear fragmentation) and karyolysis (dissolved nuclear material) and formation of apoptotic bodies. Other morphological features, such as an intact nuclear membrane, absent hairy projection of nuclear material, blebbing and cytoplasmic buds, might not be a pathognomonic characteristic of apoptosis on their own but, rather, need to occur simultaneously with the other features 11,28 ( Figure 1A). Mitotic figures were identified as previously described. 8 MI was defined as the number of mitotic figures per 3 mm 2 ; similarly, the AI was defined as the number of apoptotic bodies within the same area assessed. To check the reliability and reproducibility of the scoring technique, apoptotic bodies were scored by two pathologists (A.I. and N.M.A.), using the same previous scoring method. To further assess the reliability of the morphologybased scoring criteria, a subset of the cases (n = 70) was stained with antibodies directed against cleaved caspase-3 (anti-Asp175 #9661; Cell Signaling, Danvers, NA, USA), which is an apoptosis-specific marker. Tonsil and kidney sections served as positive controls while negative control was achieved by omitting the application of the primary antibody. Full-face formalin-fixed paraffin-embedded tumour blocks were used for immunohistochemistry (IHC). A cleaved caspase-3 antibody was applied at 1:400 concentration and detected using the Novocastra Novolink polymer detection system (vode: RE7280-K; Leica, Newcastle upon Tyne, UK). Quantification of cleaved caspase-3-positively stained cells was performed within the same 3 mm 2 area used for AI and MI ( Figure 1B). AI detected morphologically was then compared to AI stained with cleaved caspase 3. In order to identify the key regulatory genes controlling both proliferation and apoptosis in BC, we attempted to identify differentially expressed genes (DEGs) between cases with high MI and AI (HM/HA) and those with low MI and AI (LM/LA). BC-related RNA-seq data from the TCGA database was processed using R software (version 3.4.3; https://cran.r-project. org/). DEGs were determined between the apoptotic and mitotic indices, defined as dysregulated genes with |log2FC| > 1 and false discovery rate (FDR) < 0.05 between groups. Validated DEGs were visualised through Venn graphs to minimise the systematic error from group classification. The overlap of genes with the same trends in both groups was considered as DEGs of interest. HA), over-representation analysis (ORA), gene ontology and molecular function analysis were conducted on WebGestalt.org. 29 A PPI network was constructed using the STRING (Search Tool for Retrieval of Interacting Genes/Proteins) website, https://stringdb.org/, 30 which was then visualised using Cytoscape (version 3.6.1). 30 The maximal clique centrality (MCC) algorithm was used to discover hub nodes, and was calculated by CytoHubba, a plugin in Cytoscape. In this study, the genes with the top 10 MCC values were considered hub genes. To confirm the reliability of the hub genes, the prognostic significance of hub genes in BC was validated using the Kaplan-Meier (KM) plotter (https://kmplot.com/ analysis/). 31

S T A T I S T I C A L A N A L Y S I S
The optimal cut-offs for apoptosis and mitosis counts were determined against BCSS in the Nottingham cohort using X-tile bioinformatics software version 3.6.1 (School of Medicine, Yale University, New Haven, CT, USA). 32 The apoptosis and mitosis counts were categorised into two groups (high and low). A high apoptosis score was defined as ≥ 30 apoptotic bodies/3 mm 2 , while the total number of mitotic figures was dichotomised into two scores, where a high score has ≥11 mitoses/3 mm 2 . The same cut-offs were applied to the TCGA cohort. BC cases were then categorised into four distinct groups (mitosis/apoptosis index), defined as score 1: low mitosis/low apoptosis (LM/LA), score 2: low mitosis/high apoptosis (LM/HA), score 3: high mitosis/low apoptosis (HM/LA) and score 4: high mitosis/high apoptosis (HM/HA). These scores were incorporated with the other two components of the Nottingham grading system (tubule formation and pleomorphism) to develop a modified grade based on the mitosis-apoptosis index. This results in an overall grade score ranging from 3 to 10. The cut-off of total scores was categorised into grade 1, 3-5 grade 2 6-8 and grade 3 9-10 based on X-tile software.
IBM-SPSS statistical software version 24.0 (SPSS, Chicago, IL, USA) was used for statistical analysis. The degree of interobserver agreement was assessed by the intraclass correlation coefficient (ICC) and Fleiss's kappa statistic. Pearson's correlation coefficient was utilised to measure the association between mitosis and apoptosis count. Outcome analysis was assessed using KM curves and the log-rank test. The Cox regression model was used for the univariate and multivariate analysis. For all tests, P < 0.05 (twotailed) was statistically significant.

Results
A high apoptosis score was observed in 38% of cases in the Nottingham cohort and 30% in the TCGA cohort, while a high mitosis score was seen in 37 and 33% of the cases in the Nottingham and TCGA cohorts, respectively. Data on the range, mean and median of mitosis and apoptosis counts in both cohorts are summarised in (Supporting information, Table S2). There was a significant positive linear correlation between AI and MI and in both Nottingham (r = 0.54, P < 0.001) and TCGA cohorts (r = 0.49, P < 0.001).
An excellent agreement was observed in apoptosis counting between both observers (ICC = 0.7, P < 0.001). Moreover, there was a strong positive correlation between apoptosis score and cleaved caspase-3 score (r = 0.8, P < 0.001), confirming the reliability of using the morphological scoring method ( Figure 1C).

A S S O C I A T I O N O F A P O P T O T I C S C O R E S W I T H C L I N I C O P A T H O L O G I C A L P A R A M E T E R S
In both cohorts there was a statistically significant correlation between high apoptotic scores and high tumour grade, with high mitotic scores, less tubular formation, higher nuclear pleomorphism, high NPI scores, ER receptor negativity and lymphovascular invasion (Table 1).

A S S O C I A T I O N O F A P O P T O T I C S C O R E S P A T I E N T O U T C O M E
Univariate survival analysis of apoptotic scores in the Nottingham cohort showed that patients with high apoptotic scores had a significantly shorter BCSS [hazard ratio (HR) = 3.79, 95% confidence interval (CI) = 2.78-5.17; P < 0.001], worse DMFS (HR = 3.12, 95% CI = 2.33-4.16; P < 0.001) and shorter RFS (HR = 2.28, 95% CI = 1.79-2.90; P < 0.001). When the cases were stratified according to the molecular subtypes, AI was predictive of BCSS in luminal BC (HR = 3.70, 95% CI = 2.58-5.31; P < 0.001) and triple-negative BC (TNBC) (HR = 4.62, 95% CI = 2.08-10.26; P < 0.001).
In the TCGA cohort, high apoptotic scores were also predictive of a higher incidence of death from BC (HR = 2.89, 95% CI = 1.63-5.13; P < 0.001) and a higher risk of recurrence (HR = 1.65, 95% CI = 1.00-2.71; P = 0.046). Additionally, when the cohort was stratified according to the molecular subtypes, the high apoptotic score was predictive of BCSS A multivariate Cox regression model adjusted for the standard prognostic covariates, including tumour size, grade and nodal stage, revealed that apoptotic score was an independent predictor of shorter BCSS in the Nottingham (HR = 3.50, 95% CI = 2.56-4.78; P < 0.001) and TCGA (HR = 1.70, 95% CI = 1.03-2.82; P = 0.039) cohorts.
Incorporating apoptosis score with other grade components in a multivariate Cox regression model showed that apoptosis score was an independent predictor of survival in the Nottingham and TCGA cohorts, respectively (HR = 3.49, 95% CI = 2.56-4.77; P < 0.001 and HR = 3.31, 95% CI = 1.37-7.95; P = 0.007), and even showed a higher significant association with BCSS than mitosis scores in both cohorts (Table 2).

A S S O C I A T I O N O F M I T O S I S -A P O P T O S I S I N D E X W I T H C L I N I C O P A T H O L O G I C A L P A R A M E T E R S A N D O U T C O M E
In both cohorts, high MI and high AI (HM/HA) class was significantly associated with aggressive features including higher tumour grade, higher NPI scores (P < 0.001), higher nuclear pleomorphism and fewer tubules (Table 3). Survival analysis in the Nottingham cohort revealed that the mitosis-apoptosis index was predictive of death from BC (HR = 2.00, 95% CI = 1.76-2.26; P < 0.001), DMFS (HR = 1.83, 95% CI = 1.63-2.05; P < 0.001) and RFS (HR = 1.49, 95% CI = 1.36-1.64; P < 0.001). When cases were classified according to the intrinsic molecular subtypes, HM/HA was predictive of shorter BCSS in the luminal subtype (HR = 8.11, 95% CI = 5.09-12.92; P < 0.001). HM/LA was also a predictor of death from BC in TNBC (HR = 14.25, 95% CI = 3.40-59.7; P < 0.001).
When mitosis was replaced with the mitosis-apoptosis index in a multivariate Cox regression model with other grade components, this index was an independent predictor of survival in the Nottingham (HR = 1.93, 95% CI = 1.70-2.18; P < 0.001) and TCGA (HR = 1.59, 95% CI = 1.19-2.13; P = 0.002) cohorts, respectively. Importantly, the mitosis-apoptosis index showed a more significant association with BCSS than mitosis score in the Nottingham cohort (HR = 1.39, 95% CI = 1.14-1.71; P = 0.002) and mitosis score was not significantly associated with outcome in the TCGA cohort (P = 0.225) ( Table 4).

A P O P T O S I S A N D N O T T I N G H A M G R A D I N G S Y S T E M
The modified Nottingham grade with the added mitosis-apoptosis index to the tubule formation, and pleomorphism scores showed a significant association with BCSS with a higher hazard ratio (HR = 3.63, 95% CI = 2.87-4.58, P < 0.001), compared to the original grade (HR = 2.11, 95% CI = 1.66-2.7; P < 0.001) ( Figure 4) and was independent of other variables (Table 5). Moreover, the modified grade was significantly associated with outcome in the TNBC while conventional grade was not.

D E G S A N D F U N C T I O N A L P A T H W A Y E N R I C H M E N T A N D P P I N E T W O R K R E S U L T S
In the search for pathways and genes that could be accountable for the worse prognosis in the HM/HA category, we aimed to identify overlapping genes in this category; the overlap of DEGs is shown in Figure 5. Up-regulated genes (n = 707) and 880 downregulated genes were identified in both groups (mitosis and apoptosis). All DEGs enriched in DNA-binding transcription and RNA polymerase II-specific activity were used to predict the PPI network on the STRING website. Fifty-four up-regulated genes were determined and analysed. These were visualised by Cytoscape (version 3.7.2), and 10 of these genes were identified as hub genes (TFDP1, E2F1, MYBL2, E2F3, ISL1, PAX6, E2F5, E2F2, NEUROD1, SOX9) ( Figure 6).

P R O G N O S T I C V A L U E O F H U B G E N E S U S I N G T H E K M P L O T T E R
Of the 10 hub genes, the KM analyses suggested that the higher expression level of TFDP-1, E2F1, E2F3 and MYBL2 were significantly associated with worse OS and RFS in BC patients (P < 0.05) (Supporting information, Figures S3 and S4). According to all the above-mentioned observations, confirmed overexpression of TFDP-1, E2F1, E2F3 and MYBL2 are associated with worse prognosis and lower overall survival in BC patients.

Discussion
Tumour growth rate is determined by the net outcome of cellular proliferation and loss, and combining both proliferation and apoptosis scores could provide a more precise estimate of tumour growth and hence behaviour. 33 The influence of apoptosis on tumour growth and patient prognosis in BC has been a focus of research, although the results have frequently been  conflicting; while some authors, including Lipponen et al., 34 Heatley et al. 35 and Ito et al., 36 confirmed their prognostic value, others found that their prognostic impact is relatively limited. 37,38 In this work, we hypothesised that the combined assessment of apoptosis and proliferation would provide a more comprehensive assessment of tumour aggressiveness; for this purpose, we evaluated the value of apoptosis alone and its added prognostic value to mitosis and constructed an apoptosis-mitosis model, in which the key controllers of tumour growth are incorporated.
We were able to score and compare two large cohorts of BC patients using digitalised WSIs and we strictly applied the above-mentioned morphological criteria. We found that a high apoptosis score showed significant associations with aggressive clinicopathological features and was an independent predictor of poor outcomes in both cohorts. Our results are in line with studies by De Jong et al., 18 Gonz alez et al. 19 and Zhang et al., 17 who proved that AI is an independent prognostic factor in BC and associated with worse outcomes, while others failed to confirm such independent significance. 15,16,20 Additionally, our findings demonstrated a significant correlation between the number of apoptotic cells and mitotic figures, in accordance with previous studies 16,17,39,40 suggesting that the regulators that control proliferation and apoptosis in normal tissues persist, even during malignancy. 41 We counted apoptotic bodies within the tumour with excellent interobserver reproducibility, which is consistent with the findings of van de Schepop et al., 42 and to further confirm the reliability of using this morphological scoring method we stained apoptotic cells in a proportion of cases with cleaved caspase 3, a biomarker specific to apoptotic cells. 43 The positive correlation between apoptosis count assessed by the morphological criteria and cleaved caspase-3 scores reflect the integrity of using the morphological scoring technique, with no extra staining procedures or costs.
The morphological assessment approach adopted in this study agrees with previous trials on a variety of tumour types, 20,[34][35][36]44,45 and this could be improved further through the introduction of automated or artificially assisted algorithms, as the implementation of such algorithms for automated quantification of both apoptosis and mitosis will enable rapid, accurate and efficient processing as well as minimising the interpathologists' variability. 46   (14) 25 (8) 136 (41.2) 65 (42) 49 (32) 18 (12) 23 (15) *Tubule formation    The mitosis-apoptosis index generated in this study demonstrated a significant correlation between the high mitosis-high apoptosis group of BC and aggressive behaviour and worse outcome. We believe that the high number of apoptotic cells in these cases reflects the inclination of some highly proliferative aggressive tumours to undergo apoptosis, and that higher rates of proliferation are associated with increased apoptosis in malignant cells. 47 However, in TNBC, HM/LA class was predictive of worse BCSS, where there would be a mutational disruption of these regulators, and this could further explain the  very aggressive nature of this category due to the unopposed cell division with high mitosis and low apoptosis activity. The combined assessment of apoptosis and mitosis in this study was in line with the results of the work performed by De Jong et al. 18 However, these authors combined the HM/LA and LM/HA groups as a single entity, despite the significant clinical and prognostic differences between them. In addition to the limited number of cases, their study did not consider the impact of incorporation of the combined scores on overall grade or their added value. Engels et al. 48 evaluated the same concepts but used IHC markers, and showed that either a single or combined assessment of proliferation and apoptosis markers can be predictive of clinical prognosis. However, our study is considered the first to develop an index incorporating both cell proliferation and cell death in a single parameter in order to stratify BC patients' risk of death and recurrence, using two different and large cohorts of patients. Moreover, when this index is incorporated into the Nottingham grade its prognostic value significantly increased, and it became a significant predictor of outcome in TNBC.
Interestingly, this study demonstrated that both mitosis and apoptosis were positively linearly correlated, and both showed significant association with poor outcomes in BC. These findings provide evidence that while assessing proliferation in BC, morphological differentiation between mitosis and apoptosis might be unnecessary. In fact, counting mitotic and apoptotic figures as one score, within a specified BC tumour area, and incorporating this score with tubule formation and pleomorphism score as a modified Nottingham grading system, would be a more practical approach to assess proliferation in BC with additional prognostic value. A subsequent study is undergoing to test this hypothesis and confirm these observations.
Using TCGA data, we were able to identify a set of up-regulated genes that were enriched in DNAbinding transcription activity and encode proteins that control the transcriptional activity of several genes involved in cell cycle progression from the G1 to S phase. In addition, they are involved in apoptosis regulation; these genes could be responsible for this unfavourable prognosis, and it is likely that targeting them would improve BC patient survival and recurrence. Transcription factor Dp-1 (TFDP1) was at the top of the list; it has been found that TFDP1 is a transcription factor that is involved in the cell cycle and apoptosis, as it has been reported to be associated with hepatocellular carcinoma 49 and colorectal cancer. 50 The E2F/DP-1 complex regulates the expression of various cellular promoters involved in the cell cycle, apoptosis and oncogenic transformation. [51][52][53] DP-1 has shown transforming activity in cooperation with activated Ha-ras 54 or ras plus E2F1. 55 MYB proto-oncogene-like 2 (MYBL2) is one of the members of the family of MYB transcription factors and Figure 6. Visualisation of the protein-protein interaction (PPI) network and the candidate hub genes. A, PPI network of the genes between DEG lists. The blue nodes represent the genes. Edges indicate interaction associations between nodes. B, Identification of the hub genes from the PPI network using the maximal clique centrality (MCC) algorithm. Edges represent the protein-protein associations. The red nodes represent genes with high MCC sores, while the yellow nodes represent genes with a low MCC sore. interacts with cell cycle protein cyclin A, promoting cell cycle progression. 56,57 Many studies have shown that MYBL2 is highly expressed in several human tumours and plays an important role in the progression of these tumours . 58,59 In addition, a previous study has shown that, in many tumours, MYBL2 promotes cell proliferation and metastasis. 60 In conclusion, we demonstrate that adding apoptosis to mitosis is associated with worse outcome; this newly designed apoptotic-proliferative tumour model is of critical importance when compared to the interpretation of proliferation only, and could potentially improve the performance of grading. By providing such detailed insight into the tumour apoptosis and proliferation in BC, as the major elements responsible for tumour progression and, thus, in determining patient prognosis, it can allow a better understanding of the biological behaviours of tumours and will lead to a targeted selection of BC patients who would truly benefit from an aggressive therapeutic regimen.

Supporting Information
Additional Supporting Information may be found in the online version of this article: Figure S1. Kaplan-Meier survival plot showing the association between mitosis score and. breast cancer-specific survival (BCSS) in A. All the cases, B.
Luminal BC, C. HER2+, and D. TNBC in the Nottingham cohort.And the association between mitosis score and BCSS in E. All the cases, F. Luminal BC, G. HER2+, and H. TNBC in the TCGA cohort.  The patients were stratified into a high-level group (red) and a low-level group (black) according to the best cut-off expression of the gene. Figure S4. Recurrence-free survival analysis (RFS) analysis of 10 hub genes in BC patients from the Kaplan plotter. (A) TFDP1, (B) E2F1 (C) E2F3 (D) MYBL2 (E) ISL1 (F) PAX6 (G) NEUROD1 (H) PDX1 (I) FOXA2 (J) OLIG2. The patients were stratified into a high-level group (red) and a low-level group (black) according to the best cut-off expression of the gene. Log-rank P < 0.05 was a statistically significant Table S1. The details of follow-up data in both Nottingham and TCGA cohorts. Table S2. Information on the range, mean and median of mitosis, apoptosis and combined mitosisapoptosis counts in both Nottingham ad TCGA cohorts. Table S3. Univariate analyses for apoptotic and mitosis index analysis according to (A) BCSS, DMFS, and RFS in the Nottingham cohort, and (B) BCSS and OS in the TCGA cohort.