Breast Cancer Res Treat (2015) 149:363–371 DOI 10.1007/s10549-014-3241-y
PRECLINICAL STUDY
INPP4B and RAD50 have an interactive effect on survival after breast cancer Xiaofeng Dai • Rainer Fagerholm • Sofia Khan Carl Blomqvist • Heli Nevanlinna
•
Received: 19 September 2014 / Accepted: 10 December 2014 / Published online: 21 December 2014 Ó Springer Science+Business Media New York 2014
Abstract Genes sharing similar genomic landscape have the potential to interactively orchestrate certain clinicopathological features of a disease. Deletion of the RAD50 gene is a common event particularly in basal-like breast cancer, and often occurs together with deletions of BRCA1, RB1, TP53, PTEN, and INPP4B. In this study, we investigate whether these co-deleted genes have interactive effects on survival in breast cancer. Using publicly available TCGA data, we employed Cox’s proportional hazards models to test whether genomic deletions of these genes, or reduced protein or transcript levels associate with breast cancer patient survival in an interactive manner. Further validation was obtained at the transcriptional level by including 1,596 additional cases from 13 publicly available gene expression data sets from the KM-plotter database. Our results indicate that RAD50 and INPP4B associate interactively with breast
Xiaofeng Dai and Rainer Fagerholm have contributed equally to this work.
Electronic supplementary material The online version of this article (doi:10.1007/s10549-014-3241-y) contains supplementary material, which is available to authorized users. X. Dai R. Fagerholm S. Khan H. Nevanlinna (&) Department of Obstetrics and Gynecology, University of Helsinki and Helsinki University Central Hospital, PO Box 700, 00029 HUS Helsinki, Finland e-mail: [email protected] X. Dai National Engineering Laboratory for Cereal Fermentation Technology, School of Biotechnology, JiangNan University, Wuxi 214122, China C. Blomqvist Department of Oncology, University of Helsinki and Helsinki University Central Hospital, Helsinki, Finland
cancer survival at the transcriptional, translational, and genomic levels in the TCGA data set (p(interaction) \ 0.05). While neither of the genes was independently prognostic on its own, low INPP4B levels in combination with above median RAD50 abundance associated with increased hazard, both at the mRNA (HR 2.39, 95 % CI 1.20–4.76) and protein (HR 2.92, 95 % CI 1.42–6.00) levels, whereas concomitant deletion or low expression of both genes associated with unexpectedly improved survival. A similar pattern was observed in the KM-plotter data set (p(interaction) = 0.0067). We find that RAD50 and INPP4B expression levels have a synergistic influence on breast cancer survival, possibly through their effects on treatment response. Keywords Breast cancer RAD50 INPP4B Survival Interaction Abbreviations TCGA The Cancer Genome Atlas CNV Copy number variation HR Hazard ratio CI Confidence interval RPPA Reverse-phase protein microarray OS Overall survival RFS Relapse-free survival ER Estrogen receptor T Tumor size N Lymph node metastasis
Introduction Breast cancer is the most common type of cancer among women worldwide and one of the leading causes of death. It is a morphologically and molecularly heterogeneous disease
123
364
that can be categorized into a number of biologically distinct, intrinsic subtypes based on gene expression profiles [1, 2]. Tumors belonging to the same intrinsic subtype often share similar patterns of DNA copy number variation (CNV) [3–7]. Thus, genes in the frequently altered regions may have interactive effects on the clinical and pathological features of cancer. It could be further hypothesized that the expression levels of such combinations of genes may associate with cancer phenotype and prognosis independently of CNV, reflecting possible deregulation at the transcriptional or posttranscriptional levels. RAD50 is a highly conserved DNA double-strand break repair gene involved in DNA repair and frequently deleted in basal-like tumors [8]. It plays an essential role in maintaining genomic integrity and preventing tumorigenesis, and has been suggested to be a breast cancer susceptibility gene associated with genomic instability, although results vary between studies [9, 10]. The genomic locus containing RAD50 has been reported to be deleted in a large subset of breast carcinomas, particularly in the basal-like subtype, where a copy of RAD50 is lost in roughly 50 % of tumors. Loss of RAD50 commonly co-occurs with the deletion of one or more of the tumor suppressor genes BRCA1, INPP4B, PTEN, RB1, and TP53 [11]. BRCA1 is a major breast cancer susceptibility gene that is centrally involved in DNA homologous recombination repair, cooperatively with RAD50 [12, 13]. RB1 and TP53 are cell cycle regulators with crucial roles in breast cancer subtype differentiation and epithelial-to-mesenchymal transition [14–16]. The tumor suppressors INPP4B and PTEN are members of the PI3K/Pten/mTOR pathway, a complex network that controls proliferation and cell survival and is deregulated in over 70 % of breast carcinomas [17]. Concomitant loss of expression of several DNA damage response genes, particularly genes in separate pathways, may have complex prognostic and predictive implications in cancer, as defective DNA repair is not only a driver of tumorigenesis, but can also sensitize tumors to treatment [18]. We therefore hypothesized that concomitant loss of expression in two such genes may influence breast cancer survival in an interactive manner. To evaluate this hypothesis, we have retrieved and analyzed publicly available gene expression, protein expression, and CNV data to test for pairwise interactions between RAD50 and the genes commonly co-deleted with it in association with breast tumor characteristics and patient survival.
Materials and methods The TCGA data set The Cancer Genome Atlas (TCGA) is a publicly accessible database of molecular data in various types of cancer. The molecular marker data consist of (I) genome-wide SNP-based
123
Breast Cancer Res Treat (2015) 149:363–371
CNV data as determined by GISTIC 2.0 software [21]; (II) reverse-phase protein microarray (RPPA) data on 171 protein markers, with an emphasis on markers currently used for breast cancer classification due to their value in treatment decisions, markers implicated in breast cancer pathophysiology, and markers implicated in the pathophysiology of other cancer lineages [22, 23]; and (III) lowess-normalized microarray-based mRNA expression data from an Agilent 244K Custom Gene Expression G4502A-07-3 platform [23]. All TCGA data used in this study originate from primary solid breast tumors, and were retrieved using the TCGA cBioPortal interface and the associated R package ‘cgdsr’ [19, 20] which provides a programmable interface to the database (see http://www.cbioportal.org/public-portal/cgds_r.jsp for detailed documentation). We included all available tumor samples, defined by the TCGA case list brca_tcga_all (‘‘All Tumors’’) within the cancer study brca_tcga [‘‘Breast Invasive Carcinoma (TCGA, Provisional)’’]. The tumor marker data sets for all genes of interest (RAD50, BRCA1, INPP4B, PTEN, RB1, and TP53) were obtained from the following TCGA Genetic Profiles: brca_tcga_gistic (‘‘Putative copynumber alterations from GISTIC’’), brca_tcga_mrna [‘‘mRNA expression (microarray)’’], and brca_tcga_RPPA_protein_level [‘‘protein/phosphoprotein level (RPPA)’’]. Clinical data for the TCGA breast cancer cases were downloaded as a data matrix from https://tcga-data.nci.nih.gov/ tcga/dataAccessMatrix.htm. From this data set, we employed information on patient survival, as well as data on age at diagnosis, estrogen receptor (ER) status, HER2 immunohistochemistry, tumor size (T), and nodal status (N). The use of these parameters in the statistical analyses is explained in detail below. The experimental procedures, as well as data acquisition, normalization, and quality control for the TCGA data, have been previously described [23]. In total, 526–892 TCGA cases were available for the different stages of the study; see Table 1 for a more detailed summary.
Validation data set To validate our findings at the gene expression level, we downloaded a data set of 1,809 normalized breast cancer microarrays from the Kaplan–Meier Plotter database [24], consisting of the following GEO data sets: GSE11121, GSE12093, GSE12276, GSE1456, GSE16391, GSE2034, GSE2990, GSE3494, GSE4922, GSE5327, GSE6532, GSE7390, and GSE9195. These data sets have been obtained with Affymetrix HG-U133A, HG-U133 Plus 2.0, and HG-U133A 2.0 platforms, shown to be compatible for analysis with the Agilent-based TCGA gene expression data [25]. In total, the data set [with available relapse-free survival (RFS) information] consisted of 1,596 cases (Table 1). The gene expression and clinical information for
Breast Cancer Res Treat (2015) 149:363–371 Table 1 Descriptive statistics of the data sets used in this study
365
TCGA CNV
TCGA Protein
TCGA mRNA
Validation mRNA
Cases
892
729
526
1,596
Events
92
78
65
664
Follow-up endpoint
OS
OS
OS
RFS
Median
19.4
19.6
24.8
77.3
MADa
25.4
26.7
24.8
61.9
59
58
58
57
13.34
13.34
13.34
14.83
Negative Positive
191 (21 %) 660 (74 %)
167 (23 %) 520 (71 %)
117 (22 %) 402 (76 %)
259 (16 %) 964 (60 %)
Missing
41 (5 %)
42 (6 %)
7 (1.3 %)
373 (23 %)
T1
231 (26 %)
177 (24 %)
133 (25 %)
–
T2
521 (58 %)
433 (59 %)
311 (59 %)
–
T3
102 (11 %)
86 (12 %)
59 (11 %)
–
T4
35 (4 %)
32 (4 %)
20 (4 %)
–
Missing
3 (0.3 %)
1 (0.1 %)
3 (0.6 %)
1,596 (100 %)
N0
420 (47 %)
338 (46 %)
255 (48 %)
973 (61 %)
N1?
458 (51 %)
381 (52 %)
260 (49 %)
187 (12 %)
Missing
14 (1.6 %)
10 (1.4 %)
11 (2.1 %)
436 (27 %) –
Follow-up time (months)
Age at diagnosis Median MAD
a
ER status
Tumor size category (T)
Lymph node metastasis (N)
Metastasis at diagnosis (M)
a
Median absolute deviation
M0
788 (88 %)
651 (89 %)
497 (94 %)
M1
15 (2 %)
15 (2 %)
14 (3 %)
–
Missing
89 (10 %)
63 (9 %)
15 (3 %)
1,596 (100 %)
these data sets were downloaded from http://kmplot.com/ analysis/index.php?p=download. Survival interaction analysis Cox proportional hazards models were constructed for the purpose of evaluating the independent and interactive effects of INPP4B and RAD50 on breast cancer survival. The analysis was conducted at three levels, using gene expression, protein expression, and CNV data. The expression levels were first binarized at the median, and fitted into two Cox models: one with both genes represented individually, and one that included an interaction term between the two. A likelihood ratio test was then conducted to examine whether the interaction model is a better fit for the prognostic data. Any observed interaction was approximately visualized using Kaplan–Meier plots representing different RAD50:INPP4B combinations adjusted to the mean of any covariates included in the analyses (e.g., age at diagnosis, ER status). Given the information available in each data set, patient survival
refers to overall survival (OS) in TCGA, and RFS in the validation data set (in which there are only 357 cases with OS data, and 96 events, compared to 1,592 cases and 689 events with RFS data). OS analyses were adjusted for patient age at diagnosis. In the case of gene expressionrelated survival, combined meta-analysis of the two data sets was conducted using the R package ‘rmeta’. In addition to the main analysis, parallel survival analyses were conducted using ER status as a stratifier in the TCGA data. The validation analyses were stratified by study and, to avoid further splitting the data, ER status was used as a covariate instead of a stratifier. Initially, we tested for interaction using TCGA CNV data. Here, cases with genomic amplifications were removed, and hazard ratios were calculated for deletions (hemizygous and homozygous combined) versus normal copy number. Five pairs were analyzed in total: RAD50 together with BRCA1, INPP4B, PTEN, RB1, and TP53. If a statistically significant (p \ 0.05) interaction was detected for a pair of genes, the pair was further analyzed at the protein and gene expression level.
123
366
Gene expression and protein expression data were binarized at the median. For the purpose of comparison of the results with the CNV analysis, where hazard ratios were calculated for deletion versus normal copy number, ‘‘high’’ (above median) expression was set as the reference category for the survival analyses. In order to assess the independent effect of expression levels, genomic copy number was included as a covariate in these analyses. The Agilent (TCGA) mRNA expression data of INPP4B and RAD50 were obtained as the median signal of all probes targeting each gene. On the Affymetrix (validation) arrays, INPP4B was represented by a single probe set, whereas there are two RAD50 probe sets: 208393_s_at and 209349_at. Affymetrix 208393_s_at was chosen to represent RAD50, as the 209349_at probe set gives non-specific BLAST hits across the genome, and 47 % of its target sequence is flagged by RepeatMasker [26]. Other statistical analyses To evaluate the connection between tumor histopathology and the genes of interest, a histopathological association analysis was conducted. Samples were labeled with high and low expression of INPP4B and RAD50 as described above, and analyzed for association between the expression levels of INPP4B and RAD50 and histopathological markers including ER, PR, HER2, T, N, and TNM stage. The statistical significance was assessed by a Chi-square test. The spearman correlation test was used to detect correlation between CNV, protein, and gene expression levels.
Results RAD50:INPP4B and RAD50:BRCA1 deletions have an interactive effect on survival after breast cancer No interaction pertaining to OS was detected between RAD50 and RB1, PTEN, or TP53 deletions (Supplementary Table 1). Deletions in RAD50 and INPP4B did associate with OS in an interactive manner, however (p(interaction) = 0.0129; p = 0.0086 when stratified by ER status) (Table 2). While RAD50 and INPP4B deletions associated with increased hazard when the other gene had normal copy number (HR 2.11 and 1.81, respectively), concomitant deletion of both appeared to have a protective effect (HR 0.24, 95 % CI 0.08–0.73). The survival curves for different CNV combinations between RAD50 and INPP4B are displayed in Fig. 1a. A similar interactive effect was observed between RAD50 and BRCA1 deletions (p(interaction) = 0.0247): BRCA1 deletions associated with adverse prognosis when RAD50 copy number was normal (HR 2.49, 95 % CI
123
Breast Cancer Res Treat (2015) 149:363–371
1.36–4.57), but concomitant deletion of both genes reduced the hazard (HR 0.24, 95 % CI 0.08–0.78) (Supplementary Table 1). RAD50 and INPP4B expression levels have an interactive effect on survival, independently of copy number variation Next, we investigated whether the RAD50:INPP4B and RAD50:BRCA1 interactions seen at the CNV level could be further seen at the transcriptional and translational levels. Since CNV at these loci associates with gene expression [11], we included copy number as a covariate in the Cox models in order to assess the independent effect of gene expression. The tests for interaction between RAD50 and INPP4B were statistically significant in both the protein and transcript level analyses (p(interaction) = 0.0420 for protein expression; p = 0.0217 at the transcript level), and the pattern of hazard ratios was similar: concomitant low expression of both genes associated with reduced hazard, whereas low INPP4B expression in particular associated with increased hazard when RAD50 expression was high (Table 2; Fig. 1b, c). The transcript and protein levels of RAD50 and INPP4B were not prognostic on their own in models without interaction terms. At the transcript level, we could detect no interaction between RAD50 and BRCA1 in the TCGA data set (p(interaction) = 0.7894; Supplementary Table 1). Further analysis at the protein level was not possible due to lack of BRCA1 protein data at TCGA. To further evaluate our findings regarding RAD50 and INPP4B, we repeated the interaction analysis in the validation data set of 1,596 gene expression samples, and the results were roughly similar (p(interaction) = 0.0067; p = 0.0211 when stratified by ER status) (Table 2; Fig. 1d). The interactive effect of concomitant low expression of both genes was again protective, although here the hazard ratios were somewhat more modest than in the TCGA data set, and the INPP4B(low):RAD50(high) category did not as clearly associate with increased hazard. It must be noted that instead of overall survival, RFS was used as the end point in this analysis, due to data availability. INPP4B and RAD50 protein expression levels are associated with hormone receptor status and lymph node metastasis As was expected based on existing literature, we observed some correlation between INPP4B and RAD50 copy numbers (r = 0.148, p \ 0.0001). A somewhat stronger correlation was also seen at the transcript level (r = 0.352, p \ 0.0001) and at the protein level (r = 0.228, p \ 0.0001).
Breast Cancer Res Treat (2015) 149:363–371
367
Table 2 Survival analysis and interaction tests between combinations of RAD50 and INPP4B in the TCGA and validation data sets. Interaction terms between RAD50 and INPP4B are indicated by an asterisk (*) I. Genomic deletion (TCGA)a
Main analysis
Stratified by ER status
N = 653 (65 events)
N = 626 (60 events)
HRc
95% CI
p
HRc
95% CI
1.19
(0.66–2.14)
0.57
0.94
(0.49–1.83)
0.86
1.08
(0.6–1.92)
0.8
1.00
(0.54–1.84)
1
RAD50_deletion
2.11
(1.09–4.06)
0.026
1.75
(0.86–3.58)
0.12
INPP4B_deletion
1.81
(0.95–3.45)
0.07
1.80
(0.92–3.53)
0.087
RAD50_del*INPP4B_del
0.24
(0.08–0.73)
0.012
0.21
(1.5–15.16)
0.0082
No interaction RAD50_deletion INPP4B_deletion Interaction term included
II. Protein level (TCGA)a
p(interaction) = 0.0129
p(interaction) = 0.0086
Main analysis
Stratified by ER status
N = 715 (74 events)
N = 674 (68 events)
p
No interaction
HRd
95% CI
p
HRd
95% CI
p
RAD50_low
1.06
(0.64–1.77)
0.82
1.13
(0.7–1.83)
0.62
INPP4B_low
1.56
(0.89–2.75)
0.12
1.57
(0.97–2.53)
0.066
RAD50_low
2.04
(0.96–4.3)
0.062
1.94
(0.97–3.88)
0.06
INPP4B_low
2.39
(1.2–4.76)
0.014
2.35
(1.27–4.37)
0.0068
RAD50_low*INPP4B_low
0.32
(0.12–0.88)
0.027
0.38
(0.15–0.96)
0.041
Interaction term included
III. Transcript level (TCGA)a
p(interaction) = 0.0420
p(interaction) = 0.0276
Main analysis
Stratified by ER status
N = 511 (61 events)
N = 505 (59 events)
No interaction
HRd
95% CI
p
HRd
95% CI
p
RAD50_low
0.69
(0.38–1.24)
0.21
0.79
(0.45–1.38)
0.4
INPP4B_low
1.80
(0.99–3.29)
0.055
1.80
(1.04–3.11)
0.036
RAD50_low
1.49
(0.65–3.42)
0.34
1.49
(0.68–3.28)
0.32
INPP4B_low
2.92
(1.42–6)
0.0036
2.89
(1.44–5.79)
0.0028
RAD50_low*INPP4B_low
0.27
(0.09–0.81)
0.019
0.33
(0.12–0.93)
0.036
Interaction term included
IV. Transcript level (Validation)b
p(interaction) = 0.02167
p(interaction) = 0.0389
Main analysis
Stratified by ER status
N = 1,594 (663 events)
N = 1,221 (417 events)
No interaction
HRd
95% CI
p
HRd
95% CI
p
rad50_low
0.94
(0.79–1.11)
0.46
0.99
(0.79–1.24)
0.93
inpp4_low
1.10
(0.94–1.29)
0.24
1.00
(0.81–1.23)
0.97
rad50_low
1.18
(0.93–1.48)
0.17
1.27
(0.94–1.71)
0.12
inpp4b_low rad50_low*inpp4b_low
1.34 0.65
(1.08–1.66) (0.47–0.89)
0.0066 0.0065
1.25 0.61
(0.95–1.64) (0.41–0.91)
0.11 0.015
Interaction term included
p(interaction) = 0.0067 a
p(interaction) = 0.0211
Age at diagnosis was included as a covariate in all TCGA analyses but is omitted from the table for space and clarity
b
The validation analyses were stratified by data set identity (GEO)
c
Hazard ratio compared to normal copy number
d
Hazard ratio compared to above median expression
123
368
Breast Cancer Res Treat (2015) 149:363–371
Fig. 1 Kaplan–Meier plots illustrating the cumulative survival among different RAD50:INPP4B combinations of a genomic deletions in TCGA data, b median split protein expression in TCGA data, c median split gene expression in TCGA microarray data and, d median split gene expression in the validation (KM-plotter) data set.
The curves have been adjusted for ER status and, in the case of TCGA protein and gene expression, for copy number variation at the corresponding genomic loci. The x-axes indicate follow-up time in years, while cumulative survival is shown on the vertical axes
As seen in Table 3, the protein expression levels of INPP4B and, to a lesser degree, RAD50, were associated with ER status (p \ 0.0001). The combined RAD50(high):INPP4B(high) category was predominantly ER positive (96.9 %), and so was the RAD50(low):INPP4B(high) group (93.3 %). The expression levels of the two proteins were also somewhat correlated with lymph node metastasis (N; p = 0.0102), with the RAD50(low):INPP4B(low) group unexpectedly associating with the lowest frequency of positive lymph nodes.
confounder in these analyses. We therefore repeated all interaction analyses with ER status as a stratifier in the Cox models (Table 2). The statistics were nearly identical to the ER-independent analyses across all stages of the study, and the tests for interaction remained significant after ER stratification despite the loss of some cases due to missing data (especially in the validation data set). Additionally, to determine if the pattern of hazard ratios differs between ER-positive and ER-negative cases, we ran the gene expression interaction analysis separately in ERpositive and ER-negative subgroups (Supplementary Table 2). Low INPP4B transcript abundance combined with above median RAD50 level associated with particularly high hazard in the ER-positive subgroup (HR 3.58, 95 % CI 1.66–7.71), and INPP4B was in fact prognostic even in a model without an interaction term (HR 2.14,
The interaction between RAD50 and INPP4B is not dependent on ER status Due to the strong association of both RAD50 and INPP4B with ER status, it was necessary to rule out ER as a
123
Breast Cancer Res Treat (2015) 149:363–371 Table 3 Association analysis between different combinations of RAD50:INPP4B protein levels and key tumor characteristics in the TCGA data set
369
RAD50:INPP4B protein level All
High:high
Low:high
High:low
Low:low
p value
\0.0001
ER status Negative
166 (24.2 %)
6 (3.1 %)
10 (6.7 %)
48 (31.4 %)
102 (53.4 %)
Positive
519 (75.8 %)
185 (96.9 %)
140 (93.3 %)
105 (68.6 %)
89 (46.6 %)
Negative
240 (54.4 %)
67 (55.4 %)
43 (45.7 %)
60 (61.2 %)
70 (54.7 %)
Positive
201 (45.6 %)
54 (44.6 %)
51 (54.3 %)
38 (38.8 %)
58 (45.3 %)
HER2 status ns
Nodal metastasis (N) Negative
338 (47.1 %)
90 (45.0 %)
61 (38.4 %)
77 (47.8 %)
110 (55.8 %)
Positive
379 (52.9 %)
110 (55.0 %)
98 (61.6 %)
84 (52.2 %)
87 (44.2 %)
0.0102
Tumor size category (T) 1 2
177 (24.4 %) 431 (59.4 %)
61 (30.0 %) 112 (55.2 %)
39 (24.2 %) 100 (62.1 %)
36 (22.4 %) 97 (60.2 %)
41 (20.4 %) 122 (60.7 %)
3
86 (11.8 %)
20 (9.9 %)
15 (9.3 %)
21 (13.0 %)
30 (14.9 %)
4
32 (4.4 %)
10 (4.9 %)
7 (4.3 %)
7 (4.3 %)
8 (4.0 %)
95 % CI 1.16–3.97, p = 0.015). The test for interaction was statistically significant in the ER-positive subgroup (p(interaction) = 0.0363). The low power in the ER-negative subgroup (17 events) rendered the analysis statistically uninformative, but the protective effect of the interaction term could be observed in this subgroup as well.
Discussion In this study, RAD50 and INPP4B were found to associate with breast cancer survival in an interactive manner. We used TCGA CNV data as a starting point, and discovered that the interaction detected at the CNV level can also be seen at the transcript and protein levels, independently of copy number, possibly reflecting further deregulation at the transcriptional or post-transcriptional level. A similar interactive effect was seen between RAD50 and BRCA1 deletions, but could not be replicated at the transcript level independently of CNV. While low INPP4B and, to a lesser degree, RAD50 expression appeared to associate with increased hazard when the expression of the other gene was high, the hazard ratio for the interaction term (concomitant low expression of both genes) indicated a protective effect. The findings at the transcript level were replicated in a second data set consisting of 1,596 cases, using RFS as the end point of the analysis. The favorable prognosis of the low:low category may seem counterintuitive, as these tumors are more commonly ER negative and thus expected to have worse prognosis. Furthermore, co-deletion and correlated low expression of the two genes have been reported to occur most frequently
ns
in basal-like tumors, a subtype typically characterized by aggressive behavior [11]. The association of the RAD50(high):INPP4B(high) group with a relatively favorable prognosis is, on the other hand, somewhat anticipated, given that this is a strongly ER-positive subgroup. Nevertheless, it is clear that the differences in prognosis are not dependent on ER status. While both genes are very strongly correlated with ER status, the interactive prognostic effect was evident after stratification by ER, and indeed even when the analysis was restricted to ER-positive tumors alone. It is notable that the cases with RAD50(low):INPP4B(low) tumors have a somewhat reduced frequency of lymph node metastasis at the time of diagnosis, which may indicate a difference in metastatic potential, but this effect appears to be fairly modest. Compared to the high:high and low:low categories, tumors with low (below median) INPP4B expression combined with high RAD50 expression associated with particularly poor OS in the TCGA data. INPP4B is a tumor suppressor gene involved in the PI3K/Pten pathway [27]. It is selectively lost in a large subset of basal-like breast carcinomas, a typically highly aggressive tumor subtype with poor prognosis [28]. Reduced expression of INPP4B, as well as loss of heterozygosity at the INPP4B locus, has been reported to associate with adverse prognosis [11, 27]. While these findings can be considered consistent with ours, we did not see clear evidence of INPP4B as a univariate prognostic marker, but this may be because our decision to binarize expression data at the median may not be the most informative categorization. Weigman et al. split gene expression data into tertiles [11], which could increase sensitivity, but we did not consider this a viable approach for our interaction analyses due to statistical power issues.
123
370
Based on published literature on these two genes, we hypothesize that the survival pattern observed in the Cox models may arise from variations in treatment response. PTEN modulates cellular sensitivity to genotoxic stress [29], which may imply a wider role for the PI3K/PTEN pathway in cellular response to genotoxic chemotherapy. While depletion of INPP4B has been shown to confer clinically relevant phenotypic features, such as proliferation, Akt activation, and anchorage-independent growth, in breast cancer cells, it has also been reported that high INPP4B expression associates with resistance to anticancer treatment, and that depletion of the gene confers increased sensitivity [30, 31]. Furthermore, RAD50 expression has been reported to correlate negatively with cancer cell sensitivity to platinum-based agents and commonly used nucleoside analogs [32, 33], and RAD50/RAD17 doubleknockdown cells have been shown to be sensitive to carboplatin and the PARP inhibitor ABT-888, although RAD50 knockdown alone was reported to result in paradoxical carboplatin resistance [11]. It can therefore be speculated that RAD50 and INPP4B may interactively modulate treatment response to some degree. Unfortunately, we were unable to evaluate this hypothesis further in this study (e.g., by conducting analyses in specific treatment groups) due to limitations in sample size and available information in the TCGA and KM-Plotter data sets. It is worth noting that genomic loss of RAD50 commonly occurs as part of a larger deletion in the 5q11–35 chromosomal region [5]. This region contains a large number of genes, including other central DNA repair genes such as RAD17 and RAP80, of which the co-deletion of RAD17 in particular appears to be of biological significance in breast cancer [11]. We cannot therefore conclude with certainty that deletions of the RAD50 gene, specifically, are causal to the interactive effect seen in the CNV data. Our CNVadjusted analyses of the transcript and protein levels suggest a relevant role for RAD50 expression itself, however, either as a causative agent or as a marker of wider deregulation in the double-strand break repair pathway. Another possible caveat is that protein and transcript levels measured from cell lysates may not reflect the underlying biological mechanisms very well, and our findings may therefore be partially confounded by effects undetected using this approach (e.g., aberrant localization, functional mutations). It should also be noted that our selection of cases from the TCGA database was performed with maximal statistical power in mind, that is, the analysis was not restricted to cases with both gene expression and protein data available. Thus, while largely overlapping, the TCGA sample sets at the CNV, transcript, and protein levels are not exactly the same. Some caution is therefore advisable when making comparisons between the different analyses. The validation data set, in turn, consists of a number of gene expression
123
Breast Cancer Res Treat (2015) 149:363–371
data sets, obtained from different populations using different selection criteria. Here we have relied on study (GEO data set) stratification to alleviate possible studyspecific biases within the KM-plotter data set. It might also have been possible to include additional publicly available data sets, gene expression data in particular, that have currently not been utilized in this study. However, we have deliberately restricted our analyses to these previously defined and published data sets, obtained using specific experimental platforms that have already been analyzed together in TCGA studies [25]. In conclusion, we report that RAD50 and INPP4B associate with breast cancer survival in an interactive manner in two independent data sets, an effect that could possibly be explained by a synergistic influence on treatment response. These findings are hypothesis generating in nature, and warrant further studies, but may be of clinical significance pending further research. Acknowledgments This work was supported by the Helsinki University Central Hospital Research Fund, the Finnish Cancer Society, the Sigrid Juselius Foundation, and the Academy of Finland (266528). The results shown in this study are in part based upon data generated by the TCGA Research Network (http://cancergenome.nih.gov/) and the Kaplan–Meier Plotter database (http://kmplot.com/). We thank the specimen donors and research groups who have provided these data and made this study possible. Conflicts of interest
All authors declare no conflicts of interest.
References 1. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie T, Eisen MB, van de Rijn M, Jeffrey SS, Thorsen T, Quist H et al (2001) Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A 98:10869–10874 2. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, Fluge O, Pergamenschikov A et al (2000) Molecular portraits of human breast tumours. Nature 406:747–752 3. Chin K, DeVries S, Fridlyand J, Spellman PT, Roydasgupta R, Kuo WL, Lapuk A, Neve RM, Qian Z, Ryder T, Chen F, Feiler H et al (2006) Genomic and transcriptional aberrations linked to breast cancer pathophysiologies. Cancer Cell 10:529–541 4. Wood LD, Parsons DW, Jones S, Lin J, Sjoblom T, Leary RJ, Shen D, Boca SM, Barber T, Ptak J, Silliman N, Szabo S et al (2007) The genomic landscapes of human breast and colorectal cancers. Science 318:1108–1113 5. Bergamaschi A, Kim YH, Wang P, Sorlie T, Hernandez-Boussard T, Lonning PE, Tibshirani R, Borresen-Dale AL, Pollack JR (2006) Distinct patterns of DNA copy number alteration are associated with different clinicopathological features and geneexpression subtypes of breast cancer. Genes Chromosomes Cancer 45:1033–1040 6. Van Loo P, Nordgard SH, Lingjaerde OC, Russnes HG, Rye IH, Sun W, Weigman VJ, Marynen P, Zetterberg A, Naume B, Perou CM, Borresen-Dale AL et al (2010) Allele-specific copy number analysis of tumors. Proc Natl Acad Sci U S A 107:16910–16915
Breast Cancer Res Treat (2015) 149:363–371 7. Russnes HG, Vollan HK, Lingjaerde OC, Krasnitz A, Lundin P, Naume B, Sorlie T, Borgen E, Rye IH, Langerod A, Chin SF, Teschendorff AE et al (2010) Genomic architecture characterizes tumor progression paths and fate in breast cancer patients. Sci Transl Med 2:38–47 8. Johannsdottir HK, Jonsson G, Johannesdottir G, Agnarsson BA, Eerola H, Arason A, Heikkila P, Egilsson V, Olsson H, Johannsson OT, Nevanlinna H, Borg A et al (2006) Chromosome 5 imbalance mapping in breast tumors from BRCA1 and BRCA2 mutation carriers and sporadic breast tumors. Int J Cancer 119:1052–1060 9. Heikkinen K, Rapakko K, Karppinen SM, Erkko H, Knuutila S, Lundan T, Mannermaa A, Borresen-Dale AL, Borg A, Barkardottir RB, Petrini J, Winqvist R (2006) RAD50 and NBS1 are breast cancer susceptibility genes associated with genomic instability. Carcinogenesis 27:1593–1599 10. Tommiska J, Seal S, Renwick A, Barfoot R, Baskcomb L, Jayatilake H, Bartkova J, Tallila J, Kaare M, Tamminen A, Heikkila P, Evans DG et al (2006) Evaluation of RAD50 in familial breast cancer predisposition. Int J Cancer 118:2911–2916 11. Weigman VJ, Chao HH, Shabalin AA, He X, Parker JS, Nordgard SH, Grushko T, Huo D, Nwachukwu C, Nobel A, Kristensen VN, Borresen-Dale AL et al (2012) Basal-like Breast cancer DNA copy number losses identify genes involved in genomic instability, response to therapy, and patient survival. Breast Cancer Res Treat 133:865–880 12. Miki Y, Swensen J, Shattuck-Eidens D, Futreal PA, Harshman K, Tavtigian S, Liu Q, Cochran C, Bennett LM, Ding W (1994) A strong candidate for the breast and ovarian cancer susceptibility gene BRCA1. Science 266:66–71 13. Zhong Q, Chen CF, Li S, Chen Y, Wang CC, Xiao J, Chen PL, Sharp ZD, Lee WH (1999) Association of BRCA1 with the hRad50-hMre11-p95 complex and the DNA damage response. Science 285:747–750 14. Herschkowitz JI, Simin K, Weigman VJ, Mikaelian I, Usary J, Hu Z, Rasmussen KE, Jones LP, Assefnia S, Chandrasekharan S, Backlund MG, Yin Y et al (2007) Identification of conserved gene expression features between murine mammary carcinoma models and human breast tumors. Genome Biol 8:R76 15. Cressman VL, Backlund DC, Hicks EM, Gowen LC, Godfrey V, Koller BH (1999) Mammary tumor formation in p53- and BRCA1-deficient mice. Cell Growth Differ 10:1–10 16. Jiang Z, Deng T, Jones R, Li H, Herschkowitz JI, Liu JC, Weigman VJ, Tsao MS, Lane TF, Perou CM, Zacksenhaus E (2010) Rb deletion in mouse mammary progenitors induces luminal-B or basal-like/EMT tumor subtypes depending on p53 status. J Clin Investg 120:3296–3309 17. Lopez-Knowles E, O’Toole SA, McNeil CM, Millar EK, Qiu MR, Crea P, Daly RJ, Musgrove EA, Sutherland RL (2010) PI3K pathway activation in breast cancer is associated with the basallike phenotype and cancer-specific mortality. Int J Cancer 126:1121–1131 18. Jackson SP, Bartek J (2009) The DNA-damage response in human biology and disease. Nature 461:1071–1078 19. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, Jacobsen A, Byrne CJ, Heuer ML, Larsson E, Antipin Y, Reva B et al (2012) The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer Discov 2:401–404 20. Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, Sun Y, Jacobsen A, Sinha R, Larsson E, Cerami E, Sander C et al
371
21.
22.
23.
24.
25. 26. 27.
28.
29.
30.
31.
32.
33.
(2013) Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci Signal 6:pl1 Beroukhim R, Getz G, Nghiemphu L, Barretina J, Hsueh T, Linhart D, Vivanco I, Lee JC, Huang JH, Alexander S, Du J, Kau T et al (2007) Assessing the significance of chromosomal aberrations in cancer: methodology and application to glioma. Proc Natl Acad Sci U S A 104:20007–20012 Hennessy BT, Lu Y, Gonzalez-Angulo AM, Carey MS, Myhre S, Ju Z, Davies MA, Liu W, Coombes K, Meric-Bernstam F, Bedrosian I, McGahren M et al (2010) A technical assessment of the utility of reverse phase protein arrays for the study of the functional proteome in non-microdissected human breast cancers. Clin Proteomics 6:129–151 The Cancer Genome Atlas Research Network (2012) Comprehensive molecular portraits of human breast tumours. Nature 490:61–70 Gyorffy B, Lanczky A, Eklund AC, Denkert C, Budczies J, Li Q, Szallasi Z (2010) An online survival analysis tool to rapidly assess the effect of 22,277 genes on breast cancer prognosis using microarray data of 1,809 patients. Breast Cancer Res Treat 123:725–731 The Cancer Genome Atlas Research Network (2011) Integrated genomic analyses of ovarian carcinoma. Nature 474:609–615 Smit AFA, Hubley R, Green P. RepeatMasker Open-4.0. Available at: http://www.repeatmasker.org. Accessed 15 Aug 2014 Gewinner C, Wang ZC, Richardson A, Teruya-Feldstein J, Etemadmoghadam D, Bowtell D, Barretina J, Lin WM, Rameh L, Salmena L, Pandolfi PP, Cantley LC (2009) Evidence that inositol polyphosphate 4-phosphatase type II is a tumor suppressor that inhibits PI3K signaling. Cancer Cell 16:115–125 Fedele CG, Ooms LM, Ho M, Vieusseux J, O’Toole SA, Millar EK, Lopez-Knowles E, Sriratana A, Gurung R, Baglietto L, Giles GG, Bailey CG et al (2010) Inositol polyphosphate 4-phosphatase II regulates PI3K/Akt signaling and is lost in human basal-like breast cancers. Proc Natl Acad Sci U S A 107:22231–22236 Bassi C, Ho J, Srikumar T, Dowling RJ, Gorrini C, Miller SJ, Mak TW, Neel BG, Raught B, Stambolic V (2013) Nuclear PTEN controls DNA repair and sensitivity to genotoxic stress. Science 341:395–399 Min JW, Kim KI, Kim HA, Kim EK, Noh WC, Jeon HB, Cho DH, Oh JS, Park IC, Hwang SG, Kim JS (2013) INPP4B-mediated tumor resistance is associated with modulation of glucose metabolism via hexokinase 2 regulation in laryngeal cancer cells. Biochem Biophys Res Commun 440:137–142 Kim JS, Yun HS, Um HD, Park JK, Lee KH, Kang CM, Lee SJ, Hwang SG (2012) Identification of inositol polyphosphate 4-phosphatase type II as a novel tumor resistance biomarker in human laryngeal cancer HEp-2 cells. Cancer Biol Ther 13:1307–1318 Ewald B, Sampath D, Plunkett W (2008) ATM and the Mre11Rad50-Nbs1 complex respond to nucleoside analogue-induced stalled replication forks and contribute to drug resistance. Cancer Res 68:7947–7955 Abuzeid WM, Jiang X, Shi G, Wang H, Paulson D, Araki K, Jungreis D, Carney J, O’Malley BW Jr, Li D (2009) Molecular disruption of RAD50 sensitizes human tumor cells to cisplatinbased chemotherapy. J Clin Investg 119:1974–1985
123
PRECLINICAL STUDY
INPP4B and RAD50 have an interactive effect on survival after breast cancer Xiaofeng Dai • Rainer Fagerholm • Sofia Khan Carl Blomqvist • Heli Nevanlinna
•
Received: 19 September 2014 / Accepted: 10 December 2014 / Published online: 21 December 2014 Ó Springer Science+Business Media New York 2014
Abstract Genes sharing similar genomic landscape have the potential to interactively orchestrate certain clinicopathological features of a disease. Deletion of the RAD50 gene is a common event particularly in basal-like breast cancer, and often occurs together with deletions of BRCA1, RB1, TP53, PTEN, and INPP4B. In this study, we investigate whether these co-deleted genes have interactive effects on survival in breast cancer. Using publicly available TCGA data, we employed Cox’s proportional hazards models to test whether genomic deletions of these genes, or reduced protein or transcript levels associate with breast cancer patient survival in an interactive manner. Further validation was obtained at the transcriptional level by including 1,596 additional cases from 13 publicly available gene expression data sets from the KM-plotter database. Our results indicate that RAD50 and INPP4B associate interactively with breast
Xiaofeng Dai and Rainer Fagerholm have contributed equally to this work.
Electronic supplementary material The online version of this article (doi:10.1007/s10549-014-3241-y) contains supplementary material, which is available to authorized users. X. Dai R. Fagerholm S. Khan H. Nevanlinna (&) Department of Obstetrics and Gynecology, University of Helsinki and Helsinki University Central Hospital, PO Box 700, 00029 HUS Helsinki, Finland e-mail: [email protected] X. Dai National Engineering Laboratory for Cereal Fermentation Technology, School of Biotechnology, JiangNan University, Wuxi 214122, China C. Blomqvist Department of Oncology, University of Helsinki and Helsinki University Central Hospital, Helsinki, Finland
cancer survival at the transcriptional, translational, and genomic levels in the TCGA data set (p(interaction) \ 0.05). While neither of the genes was independently prognostic on its own, low INPP4B levels in combination with above median RAD50 abundance associated with increased hazard, both at the mRNA (HR 2.39, 95 % CI 1.20–4.76) and protein (HR 2.92, 95 % CI 1.42–6.00) levels, whereas concomitant deletion or low expression of both genes associated with unexpectedly improved survival. A similar pattern was observed in the KM-plotter data set (p(interaction) = 0.0067). We find that RAD50 and INPP4B expression levels have a synergistic influence on breast cancer survival, possibly through their effects on treatment response. Keywords Breast cancer RAD50 INPP4B Survival Interaction Abbreviations TCGA The Cancer Genome Atlas CNV Copy number variation HR Hazard ratio CI Confidence interval RPPA Reverse-phase protein microarray OS Overall survival RFS Relapse-free survival ER Estrogen receptor T Tumor size N Lymph node metastasis
Introduction Breast cancer is the most common type of cancer among women worldwide and one of the leading causes of death. It is a morphologically and molecularly heterogeneous disease
123
364
that can be categorized into a number of biologically distinct, intrinsic subtypes based on gene expression profiles [1, 2]. Tumors belonging to the same intrinsic subtype often share similar patterns of DNA copy number variation (CNV) [3–7]. Thus, genes in the frequently altered regions may have interactive effects on the clinical and pathological features of cancer. It could be further hypothesized that the expression levels of such combinations of genes may associate with cancer phenotype and prognosis independently of CNV, reflecting possible deregulation at the transcriptional or posttranscriptional levels. RAD50 is a highly conserved DNA double-strand break repair gene involved in DNA repair and frequently deleted in basal-like tumors [8]. It plays an essential role in maintaining genomic integrity and preventing tumorigenesis, and has been suggested to be a breast cancer susceptibility gene associated with genomic instability, although results vary between studies [9, 10]. The genomic locus containing RAD50 has been reported to be deleted in a large subset of breast carcinomas, particularly in the basal-like subtype, where a copy of RAD50 is lost in roughly 50 % of tumors. Loss of RAD50 commonly co-occurs with the deletion of one or more of the tumor suppressor genes BRCA1, INPP4B, PTEN, RB1, and TP53 [11]. BRCA1 is a major breast cancer susceptibility gene that is centrally involved in DNA homologous recombination repair, cooperatively with RAD50 [12, 13]. RB1 and TP53 are cell cycle regulators with crucial roles in breast cancer subtype differentiation and epithelial-to-mesenchymal transition [14–16]. The tumor suppressors INPP4B and PTEN are members of the PI3K/Pten/mTOR pathway, a complex network that controls proliferation and cell survival and is deregulated in over 70 % of breast carcinomas [17]. Concomitant loss of expression of several DNA damage response genes, particularly genes in separate pathways, may have complex prognostic and predictive implications in cancer, as defective DNA repair is not only a driver of tumorigenesis, but can also sensitize tumors to treatment [18]. We therefore hypothesized that concomitant loss of expression in two such genes may influence breast cancer survival in an interactive manner. To evaluate this hypothesis, we have retrieved and analyzed publicly available gene expression, protein expression, and CNV data to test for pairwise interactions between RAD50 and the genes commonly co-deleted with it in association with breast tumor characteristics and patient survival.
Materials and methods The TCGA data set The Cancer Genome Atlas (TCGA) is a publicly accessible database of molecular data in various types of cancer. The molecular marker data consist of (I) genome-wide SNP-based
123
Breast Cancer Res Treat (2015) 149:363–371
CNV data as determined by GISTIC 2.0 software [21]; (II) reverse-phase protein microarray (RPPA) data on 171 protein markers, with an emphasis on markers currently used for breast cancer classification due to their value in treatment decisions, markers implicated in breast cancer pathophysiology, and markers implicated in the pathophysiology of other cancer lineages [22, 23]; and (III) lowess-normalized microarray-based mRNA expression data from an Agilent 244K Custom Gene Expression G4502A-07-3 platform [23]. All TCGA data used in this study originate from primary solid breast tumors, and were retrieved using the TCGA cBioPortal interface and the associated R package ‘cgdsr’ [19, 20] which provides a programmable interface to the database (see http://www.cbioportal.org/public-portal/cgds_r.jsp for detailed documentation). We included all available tumor samples, defined by the TCGA case list brca_tcga_all (‘‘All Tumors’’) within the cancer study brca_tcga [‘‘Breast Invasive Carcinoma (TCGA, Provisional)’’]. The tumor marker data sets for all genes of interest (RAD50, BRCA1, INPP4B, PTEN, RB1, and TP53) were obtained from the following TCGA Genetic Profiles: brca_tcga_gistic (‘‘Putative copynumber alterations from GISTIC’’), brca_tcga_mrna [‘‘mRNA expression (microarray)’’], and brca_tcga_RPPA_protein_level [‘‘protein/phosphoprotein level (RPPA)’’]. Clinical data for the TCGA breast cancer cases were downloaded as a data matrix from https://tcga-data.nci.nih.gov/ tcga/dataAccessMatrix.htm. From this data set, we employed information on patient survival, as well as data on age at diagnosis, estrogen receptor (ER) status, HER2 immunohistochemistry, tumor size (T), and nodal status (N). The use of these parameters in the statistical analyses is explained in detail below. The experimental procedures, as well as data acquisition, normalization, and quality control for the TCGA data, have been previously described [23]. In total, 526–892 TCGA cases were available for the different stages of the study; see Table 1 for a more detailed summary.
Validation data set To validate our findings at the gene expression level, we downloaded a data set of 1,809 normalized breast cancer microarrays from the Kaplan–Meier Plotter database [24], consisting of the following GEO data sets: GSE11121, GSE12093, GSE12276, GSE1456, GSE16391, GSE2034, GSE2990, GSE3494, GSE4922, GSE5327, GSE6532, GSE7390, and GSE9195. These data sets have been obtained with Affymetrix HG-U133A, HG-U133 Plus 2.0, and HG-U133A 2.0 platforms, shown to be compatible for analysis with the Agilent-based TCGA gene expression data [25]. In total, the data set [with available relapse-free survival (RFS) information] consisted of 1,596 cases (Table 1). The gene expression and clinical information for
Breast Cancer Res Treat (2015) 149:363–371 Table 1 Descriptive statistics of the data sets used in this study
365
TCGA CNV
TCGA Protein
TCGA mRNA
Validation mRNA
Cases
892
729
526
1,596
Events
92
78
65
664
Follow-up endpoint
OS
OS
OS
RFS
Median
19.4
19.6
24.8
77.3
MADa
25.4
26.7
24.8
61.9
59
58
58
57
13.34
13.34
13.34
14.83
Negative Positive
191 (21 %) 660 (74 %)
167 (23 %) 520 (71 %)
117 (22 %) 402 (76 %)
259 (16 %) 964 (60 %)
Missing
41 (5 %)
42 (6 %)
7 (1.3 %)
373 (23 %)
T1
231 (26 %)
177 (24 %)
133 (25 %)
–
T2
521 (58 %)
433 (59 %)
311 (59 %)
–
T3
102 (11 %)
86 (12 %)
59 (11 %)
–
T4
35 (4 %)
32 (4 %)
20 (4 %)
–
Missing
3 (0.3 %)
1 (0.1 %)
3 (0.6 %)
1,596 (100 %)
N0
420 (47 %)
338 (46 %)
255 (48 %)
973 (61 %)
N1?
458 (51 %)
381 (52 %)
260 (49 %)
187 (12 %)
Missing
14 (1.6 %)
10 (1.4 %)
11 (2.1 %)
436 (27 %) –
Follow-up time (months)
Age at diagnosis Median MAD
a
ER status
Tumor size category (T)
Lymph node metastasis (N)
Metastasis at diagnosis (M)
a
Median absolute deviation
M0
788 (88 %)
651 (89 %)
497 (94 %)
M1
15 (2 %)
15 (2 %)
14 (3 %)
–
Missing
89 (10 %)
63 (9 %)
15 (3 %)
1,596 (100 %)
these data sets were downloaded from http://kmplot.com/ analysis/index.php?p=download. Survival interaction analysis Cox proportional hazards models were constructed for the purpose of evaluating the independent and interactive effects of INPP4B and RAD50 on breast cancer survival. The analysis was conducted at three levels, using gene expression, protein expression, and CNV data. The expression levels were first binarized at the median, and fitted into two Cox models: one with both genes represented individually, and one that included an interaction term between the two. A likelihood ratio test was then conducted to examine whether the interaction model is a better fit for the prognostic data. Any observed interaction was approximately visualized using Kaplan–Meier plots representing different RAD50:INPP4B combinations adjusted to the mean of any covariates included in the analyses (e.g., age at diagnosis, ER status). Given the information available in each data set, patient survival
refers to overall survival (OS) in TCGA, and RFS in the validation data set (in which there are only 357 cases with OS data, and 96 events, compared to 1,592 cases and 689 events with RFS data). OS analyses were adjusted for patient age at diagnosis. In the case of gene expressionrelated survival, combined meta-analysis of the two data sets was conducted using the R package ‘rmeta’. In addition to the main analysis, parallel survival analyses were conducted using ER status as a stratifier in the TCGA data. The validation analyses were stratified by study and, to avoid further splitting the data, ER status was used as a covariate instead of a stratifier. Initially, we tested for interaction using TCGA CNV data. Here, cases with genomic amplifications were removed, and hazard ratios were calculated for deletions (hemizygous and homozygous combined) versus normal copy number. Five pairs were analyzed in total: RAD50 together with BRCA1, INPP4B, PTEN, RB1, and TP53. If a statistically significant (p \ 0.05) interaction was detected for a pair of genes, the pair was further analyzed at the protein and gene expression level.
123
366
Gene expression and protein expression data were binarized at the median. For the purpose of comparison of the results with the CNV analysis, where hazard ratios were calculated for deletion versus normal copy number, ‘‘high’’ (above median) expression was set as the reference category for the survival analyses. In order to assess the independent effect of expression levels, genomic copy number was included as a covariate in these analyses. The Agilent (TCGA) mRNA expression data of INPP4B and RAD50 were obtained as the median signal of all probes targeting each gene. On the Affymetrix (validation) arrays, INPP4B was represented by a single probe set, whereas there are two RAD50 probe sets: 208393_s_at and 209349_at. Affymetrix 208393_s_at was chosen to represent RAD50, as the 209349_at probe set gives non-specific BLAST hits across the genome, and 47 % of its target sequence is flagged by RepeatMasker [26]. Other statistical analyses To evaluate the connection between tumor histopathology and the genes of interest, a histopathological association analysis was conducted. Samples were labeled with high and low expression of INPP4B and RAD50 as described above, and analyzed for association between the expression levels of INPP4B and RAD50 and histopathological markers including ER, PR, HER2, T, N, and TNM stage. The statistical significance was assessed by a Chi-square test. The spearman correlation test was used to detect correlation between CNV, protein, and gene expression levels.
Results RAD50:INPP4B and RAD50:BRCA1 deletions have an interactive effect on survival after breast cancer No interaction pertaining to OS was detected between RAD50 and RB1, PTEN, or TP53 deletions (Supplementary Table 1). Deletions in RAD50 and INPP4B did associate with OS in an interactive manner, however (p(interaction) = 0.0129; p = 0.0086 when stratified by ER status) (Table 2). While RAD50 and INPP4B deletions associated with increased hazard when the other gene had normal copy number (HR 2.11 and 1.81, respectively), concomitant deletion of both appeared to have a protective effect (HR 0.24, 95 % CI 0.08–0.73). The survival curves for different CNV combinations between RAD50 and INPP4B are displayed in Fig. 1a. A similar interactive effect was observed between RAD50 and BRCA1 deletions (p(interaction) = 0.0247): BRCA1 deletions associated with adverse prognosis when RAD50 copy number was normal (HR 2.49, 95 % CI
123
Breast Cancer Res Treat (2015) 149:363–371
1.36–4.57), but concomitant deletion of both genes reduced the hazard (HR 0.24, 95 % CI 0.08–0.78) (Supplementary Table 1). RAD50 and INPP4B expression levels have an interactive effect on survival, independently of copy number variation Next, we investigated whether the RAD50:INPP4B and RAD50:BRCA1 interactions seen at the CNV level could be further seen at the transcriptional and translational levels. Since CNV at these loci associates with gene expression [11], we included copy number as a covariate in the Cox models in order to assess the independent effect of gene expression. The tests for interaction between RAD50 and INPP4B were statistically significant in both the protein and transcript level analyses (p(interaction) = 0.0420 for protein expression; p = 0.0217 at the transcript level), and the pattern of hazard ratios was similar: concomitant low expression of both genes associated with reduced hazard, whereas low INPP4B expression in particular associated with increased hazard when RAD50 expression was high (Table 2; Fig. 1b, c). The transcript and protein levels of RAD50 and INPP4B were not prognostic on their own in models without interaction terms. At the transcript level, we could detect no interaction between RAD50 and BRCA1 in the TCGA data set (p(interaction) = 0.7894; Supplementary Table 1). Further analysis at the protein level was not possible due to lack of BRCA1 protein data at TCGA. To further evaluate our findings regarding RAD50 and INPP4B, we repeated the interaction analysis in the validation data set of 1,596 gene expression samples, and the results were roughly similar (p(interaction) = 0.0067; p = 0.0211 when stratified by ER status) (Table 2; Fig. 1d). The interactive effect of concomitant low expression of both genes was again protective, although here the hazard ratios were somewhat more modest than in the TCGA data set, and the INPP4B(low):RAD50(high) category did not as clearly associate with increased hazard. It must be noted that instead of overall survival, RFS was used as the end point in this analysis, due to data availability. INPP4B and RAD50 protein expression levels are associated with hormone receptor status and lymph node metastasis As was expected based on existing literature, we observed some correlation between INPP4B and RAD50 copy numbers (r = 0.148, p \ 0.0001). A somewhat stronger correlation was also seen at the transcript level (r = 0.352, p \ 0.0001) and at the protein level (r = 0.228, p \ 0.0001).
Breast Cancer Res Treat (2015) 149:363–371
367
Table 2 Survival analysis and interaction tests between combinations of RAD50 and INPP4B in the TCGA and validation data sets. Interaction terms between RAD50 and INPP4B are indicated by an asterisk (*) I. Genomic deletion (TCGA)a
Main analysis
Stratified by ER status
N = 653 (65 events)
N = 626 (60 events)
HRc
95% CI
p
HRc
95% CI
1.19
(0.66–2.14)
0.57
0.94
(0.49–1.83)
0.86
1.08
(0.6–1.92)
0.8
1.00
(0.54–1.84)
1
RAD50_deletion
2.11
(1.09–4.06)
0.026
1.75
(0.86–3.58)
0.12
INPP4B_deletion
1.81
(0.95–3.45)
0.07
1.80
(0.92–3.53)
0.087
RAD50_del*INPP4B_del
0.24
(0.08–0.73)
0.012
0.21
(1.5–15.16)
0.0082
No interaction RAD50_deletion INPP4B_deletion Interaction term included
II. Protein level (TCGA)a
p(interaction) = 0.0129
p(interaction) = 0.0086
Main analysis
Stratified by ER status
N = 715 (74 events)
N = 674 (68 events)
p
No interaction
HRd
95% CI
p
HRd
95% CI
p
RAD50_low
1.06
(0.64–1.77)
0.82
1.13
(0.7–1.83)
0.62
INPP4B_low
1.56
(0.89–2.75)
0.12
1.57
(0.97–2.53)
0.066
RAD50_low
2.04
(0.96–4.3)
0.062
1.94
(0.97–3.88)
0.06
INPP4B_low
2.39
(1.2–4.76)
0.014
2.35
(1.27–4.37)
0.0068
RAD50_low*INPP4B_low
0.32
(0.12–0.88)
0.027
0.38
(0.15–0.96)
0.041
Interaction term included
III. Transcript level (TCGA)a
p(interaction) = 0.0420
p(interaction) = 0.0276
Main analysis
Stratified by ER status
N = 511 (61 events)
N = 505 (59 events)
No interaction
HRd
95% CI
p
HRd
95% CI
p
RAD50_low
0.69
(0.38–1.24)
0.21
0.79
(0.45–1.38)
0.4
INPP4B_low
1.80
(0.99–3.29)
0.055
1.80
(1.04–3.11)
0.036
RAD50_low
1.49
(0.65–3.42)
0.34
1.49
(0.68–3.28)
0.32
INPP4B_low
2.92
(1.42–6)
0.0036
2.89
(1.44–5.79)
0.0028
RAD50_low*INPP4B_low
0.27
(0.09–0.81)
0.019
0.33
(0.12–0.93)
0.036
Interaction term included
IV. Transcript level (Validation)b
p(interaction) = 0.02167
p(interaction) = 0.0389
Main analysis
Stratified by ER status
N = 1,594 (663 events)
N = 1,221 (417 events)
No interaction
HRd
95% CI
p
HRd
95% CI
p
rad50_low
0.94
(0.79–1.11)
0.46
0.99
(0.79–1.24)
0.93
inpp4_low
1.10
(0.94–1.29)
0.24
1.00
(0.81–1.23)
0.97
rad50_low
1.18
(0.93–1.48)
0.17
1.27
(0.94–1.71)
0.12
inpp4b_low rad50_low*inpp4b_low
1.34 0.65
(1.08–1.66) (0.47–0.89)
0.0066 0.0065
1.25 0.61
(0.95–1.64) (0.41–0.91)
0.11 0.015
Interaction term included
p(interaction) = 0.0067 a
p(interaction) = 0.0211
Age at diagnosis was included as a covariate in all TCGA analyses but is omitted from the table for space and clarity
b
The validation analyses were stratified by data set identity (GEO)
c
Hazard ratio compared to normal copy number
d
Hazard ratio compared to above median expression
123
368
Breast Cancer Res Treat (2015) 149:363–371
Fig. 1 Kaplan–Meier plots illustrating the cumulative survival among different RAD50:INPP4B combinations of a genomic deletions in TCGA data, b median split protein expression in TCGA data, c median split gene expression in TCGA microarray data and, d median split gene expression in the validation (KM-plotter) data set.
The curves have been adjusted for ER status and, in the case of TCGA protein and gene expression, for copy number variation at the corresponding genomic loci. The x-axes indicate follow-up time in years, while cumulative survival is shown on the vertical axes
As seen in Table 3, the protein expression levels of INPP4B and, to a lesser degree, RAD50, were associated with ER status (p \ 0.0001). The combined RAD50(high):INPP4B(high) category was predominantly ER positive (96.9 %), and so was the RAD50(low):INPP4B(high) group (93.3 %). The expression levels of the two proteins were also somewhat correlated with lymph node metastasis (N; p = 0.0102), with the RAD50(low):INPP4B(low) group unexpectedly associating with the lowest frequency of positive lymph nodes.
confounder in these analyses. We therefore repeated all interaction analyses with ER status as a stratifier in the Cox models (Table 2). The statistics were nearly identical to the ER-independent analyses across all stages of the study, and the tests for interaction remained significant after ER stratification despite the loss of some cases due to missing data (especially in the validation data set). Additionally, to determine if the pattern of hazard ratios differs between ER-positive and ER-negative cases, we ran the gene expression interaction analysis separately in ERpositive and ER-negative subgroups (Supplementary Table 2). Low INPP4B transcript abundance combined with above median RAD50 level associated with particularly high hazard in the ER-positive subgroup (HR 3.58, 95 % CI 1.66–7.71), and INPP4B was in fact prognostic even in a model without an interaction term (HR 2.14,
The interaction between RAD50 and INPP4B is not dependent on ER status Due to the strong association of both RAD50 and INPP4B with ER status, it was necessary to rule out ER as a
123
Breast Cancer Res Treat (2015) 149:363–371 Table 3 Association analysis between different combinations of RAD50:INPP4B protein levels and key tumor characteristics in the TCGA data set
369
RAD50:INPP4B protein level All
High:high
Low:high
High:low
Low:low
p value
\0.0001
ER status Negative
166 (24.2 %)
6 (3.1 %)
10 (6.7 %)
48 (31.4 %)
102 (53.4 %)
Positive
519 (75.8 %)
185 (96.9 %)
140 (93.3 %)
105 (68.6 %)
89 (46.6 %)
Negative
240 (54.4 %)
67 (55.4 %)
43 (45.7 %)
60 (61.2 %)
70 (54.7 %)
Positive
201 (45.6 %)
54 (44.6 %)
51 (54.3 %)
38 (38.8 %)
58 (45.3 %)
HER2 status ns
Nodal metastasis (N) Negative
338 (47.1 %)
90 (45.0 %)
61 (38.4 %)
77 (47.8 %)
110 (55.8 %)
Positive
379 (52.9 %)
110 (55.0 %)
98 (61.6 %)
84 (52.2 %)
87 (44.2 %)
0.0102
Tumor size category (T) 1 2
177 (24.4 %) 431 (59.4 %)
61 (30.0 %) 112 (55.2 %)
39 (24.2 %) 100 (62.1 %)
36 (22.4 %) 97 (60.2 %)
41 (20.4 %) 122 (60.7 %)
3
86 (11.8 %)
20 (9.9 %)
15 (9.3 %)
21 (13.0 %)
30 (14.9 %)
4
32 (4.4 %)
10 (4.9 %)
7 (4.3 %)
7 (4.3 %)
8 (4.0 %)
95 % CI 1.16–3.97, p = 0.015). The test for interaction was statistically significant in the ER-positive subgroup (p(interaction) = 0.0363). The low power in the ER-negative subgroup (17 events) rendered the analysis statistically uninformative, but the protective effect of the interaction term could be observed in this subgroup as well.
Discussion In this study, RAD50 and INPP4B were found to associate with breast cancer survival in an interactive manner. We used TCGA CNV data as a starting point, and discovered that the interaction detected at the CNV level can also be seen at the transcript and protein levels, independently of copy number, possibly reflecting further deregulation at the transcriptional or post-transcriptional level. A similar interactive effect was seen between RAD50 and BRCA1 deletions, but could not be replicated at the transcript level independently of CNV. While low INPP4B and, to a lesser degree, RAD50 expression appeared to associate with increased hazard when the expression of the other gene was high, the hazard ratio for the interaction term (concomitant low expression of both genes) indicated a protective effect. The findings at the transcript level were replicated in a second data set consisting of 1,596 cases, using RFS as the end point of the analysis. The favorable prognosis of the low:low category may seem counterintuitive, as these tumors are more commonly ER negative and thus expected to have worse prognosis. Furthermore, co-deletion and correlated low expression of the two genes have been reported to occur most frequently
ns
in basal-like tumors, a subtype typically characterized by aggressive behavior [11]. The association of the RAD50(high):INPP4B(high) group with a relatively favorable prognosis is, on the other hand, somewhat anticipated, given that this is a strongly ER-positive subgroup. Nevertheless, it is clear that the differences in prognosis are not dependent on ER status. While both genes are very strongly correlated with ER status, the interactive prognostic effect was evident after stratification by ER, and indeed even when the analysis was restricted to ER-positive tumors alone. It is notable that the cases with RAD50(low):INPP4B(low) tumors have a somewhat reduced frequency of lymph node metastasis at the time of diagnosis, which may indicate a difference in metastatic potential, but this effect appears to be fairly modest. Compared to the high:high and low:low categories, tumors with low (below median) INPP4B expression combined with high RAD50 expression associated with particularly poor OS in the TCGA data. INPP4B is a tumor suppressor gene involved in the PI3K/Pten pathway [27]. It is selectively lost in a large subset of basal-like breast carcinomas, a typically highly aggressive tumor subtype with poor prognosis [28]. Reduced expression of INPP4B, as well as loss of heterozygosity at the INPP4B locus, has been reported to associate with adverse prognosis [11, 27]. While these findings can be considered consistent with ours, we did not see clear evidence of INPP4B as a univariate prognostic marker, but this may be because our decision to binarize expression data at the median may not be the most informative categorization. Weigman et al. split gene expression data into tertiles [11], which could increase sensitivity, but we did not consider this a viable approach for our interaction analyses due to statistical power issues.
123
370
Based on published literature on these two genes, we hypothesize that the survival pattern observed in the Cox models may arise from variations in treatment response. PTEN modulates cellular sensitivity to genotoxic stress [29], which may imply a wider role for the PI3K/PTEN pathway in cellular response to genotoxic chemotherapy. While depletion of INPP4B has been shown to confer clinically relevant phenotypic features, such as proliferation, Akt activation, and anchorage-independent growth, in breast cancer cells, it has also been reported that high INPP4B expression associates with resistance to anticancer treatment, and that depletion of the gene confers increased sensitivity [30, 31]. Furthermore, RAD50 expression has been reported to correlate negatively with cancer cell sensitivity to platinum-based agents and commonly used nucleoside analogs [32, 33], and RAD50/RAD17 doubleknockdown cells have been shown to be sensitive to carboplatin and the PARP inhibitor ABT-888, although RAD50 knockdown alone was reported to result in paradoxical carboplatin resistance [11]. It can therefore be speculated that RAD50 and INPP4B may interactively modulate treatment response to some degree. Unfortunately, we were unable to evaluate this hypothesis further in this study (e.g., by conducting analyses in specific treatment groups) due to limitations in sample size and available information in the TCGA and KM-Plotter data sets. It is worth noting that genomic loss of RAD50 commonly occurs as part of a larger deletion in the 5q11–35 chromosomal region [5]. This region contains a large number of genes, including other central DNA repair genes such as RAD17 and RAP80, of which the co-deletion of RAD17 in particular appears to be of biological significance in breast cancer [11]. We cannot therefore conclude with certainty that deletions of the RAD50 gene, specifically, are causal to the interactive effect seen in the CNV data. Our CNVadjusted analyses of the transcript and protein levels suggest a relevant role for RAD50 expression itself, however, either as a causative agent or as a marker of wider deregulation in the double-strand break repair pathway. Another possible caveat is that protein and transcript levels measured from cell lysates may not reflect the underlying biological mechanisms very well, and our findings may therefore be partially confounded by effects undetected using this approach (e.g., aberrant localization, functional mutations). It should also be noted that our selection of cases from the TCGA database was performed with maximal statistical power in mind, that is, the analysis was not restricted to cases with both gene expression and protein data available. Thus, while largely overlapping, the TCGA sample sets at the CNV, transcript, and protein levels are not exactly the same. Some caution is therefore advisable when making comparisons between the different analyses. The validation data set, in turn, consists of a number of gene expression
123
Breast Cancer Res Treat (2015) 149:363–371
data sets, obtained from different populations using different selection criteria. Here we have relied on study (GEO data set) stratification to alleviate possible studyspecific biases within the KM-plotter data set. It might also have been possible to include additional publicly available data sets, gene expression data in particular, that have currently not been utilized in this study. However, we have deliberately restricted our analyses to these previously defined and published data sets, obtained using specific experimental platforms that have already been analyzed together in TCGA studies [25]. In conclusion, we report that RAD50 and INPP4B associate with breast cancer survival in an interactive manner in two independent data sets, an effect that could possibly be explained by a synergistic influence on treatment response. These findings are hypothesis generating in nature, and warrant further studies, but may be of clinical significance pending further research. Acknowledgments This work was supported by the Helsinki University Central Hospital Research Fund, the Finnish Cancer Society, the Sigrid Juselius Foundation, and the Academy of Finland (266528). The results shown in this study are in part based upon data generated by the TCGA Research Network (http://cancergenome.nih.gov/) and the Kaplan–Meier Plotter database (http://kmplot.com/). We thank the specimen donors and research groups who have provided these data and made this study possible. Conflicts of interest
All authors declare no conflicts of interest.
References 1. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie T, Eisen MB, van de Rijn M, Jeffrey SS, Thorsen T, Quist H et al (2001) Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A 98:10869–10874 2. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, Fluge O, Pergamenschikov A et al (2000) Molecular portraits of human breast tumours. Nature 406:747–752 3. Chin K, DeVries S, Fridlyand J, Spellman PT, Roydasgupta R, Kuo WL, Lapuk A, Neve RM, Qian Z, Ryder T, Chen F, Feiler H et al (2006) Genomic and transcriptional aberrations linked to breast cancer pathophysiologies. Cancer Cell 10:529–541 4. Wood LD, Parsons DW, Jones S, Lin J, Sjoblom T, Leary RJ, Shen D, Boca SM, Barber T, Ptak J, Silliman N, Szabo S et al (2007) The genomic landscapes of human breast and colorectal cancers. Science 318:1108–1113 5. Bergamaschi A, Kim YH, Wang P, Sorlie T, Hernandez-Boussard T, Lonning PE, Tibshirani R, Borresen-Dale AL, Pollack JR (2006) Distinct patterns of DNA copy number alteration are associated with different clinicopathological features and geneexpression subtypes of breast cancer. Genes Chromosomes Cancer 45:1033–1040 6. Van Loo P, Nordgard SH, Lingjaerde OC, Russnes HG, Rye IH, Sun W, Weigman VJ, Marynen P, Zetterberg A, Naume B, Perou CM, Borresen-Dale AL et al (2010) Allele-specific copy number analysis of tumors. Proc Natl Acad Sci U S A 107:16910–16915
Breast Cancer Res Treat (2015) 149:363–371 7. Russnes HG, Vollan HK, Lingjaerde OC, Krasnitz A, Lundin P, Naume B, Sorlie T, Borgen E, Rye IH, Langerod A, Chin SF, Teschendorff AE et al (2010) Genomic architecture characterizes tumor progression paths and fate in breast cancer patients. Sci Transl Med 2:38–47 8. Johannsdottir HK, Jonsson G, Johannesdottir G, Agnarsson BA, Eerola H, Arason A, Heikkila P, Egilsson V, Olsson H, Johannsson OT, Nevanlinna H, Borg A et al (2006) Chromosome 5 imbalance mapping in breast tumors from BRCA1 and BRCA2 mutation carriers and sporadic breast tumors. Int J Cancer 119:1052–1060 9. Heikkinen K, Rapakko K, Karppinen SM, Erkko H, Knuutila S, Lundan T, Mannermaa A, Borresen-Dale AL, Borg A, Barkardottir RB, Petrini J, Winqvist R (2006) RAD50 and NBS1 are breast cancer susceptibility genes associated with genomic instability. Carcinogenesis 27:1593–1599 10. Tommiska J, Seal S, Renwick A, Barfoot R, Baskcomb L, Jayatilake H, Bartkova J, Tallila J, Kaare M, Tamminen A, Heikkila P, Evans DG et al (2006) Evaluation of RAD50 in familial breast cancer predisposition. Int J Cancer 118:2911–2916 11. Weigman VJ, Chao HH, Shabalin AA, He X, Parker JS, Nordgard SH, Grushko T, Huo D, Nwachukwu C, Nobel A, Kristensen VN, Borresen-Dale AL et al (2012) Basal-like Breast cancer DNA copy number losses identify genes involved in genomic instability, response to therapy, and patient survival. Breast Cancer Res Treat 133:865–880 12. Miki Y, Swensen J, Shattuck-Eidens D, Futreal PA, Harshman K, Tavtigian S, Liu Q, Cochran C, Bennett LM, Ding W (1994) A strong candidate for the breast and ovarian cancer susceptibility gene BRCA1. Science 266:66–71 13. Zhong Q, Chen CF, Li S, Chen Y, Wang CC, Xiao J, Chen PL, Sharp ZD, Lee WH (1999) Association of BRCA1 with the hRad50-hMre11-p95 complex and the DNA damage response. Science 285:747–750 14. Herschkowitz JI, Simin K, Weigman VJ, Mikaelian I, Usary J, Hu Z, Rasmussen KE, Jones LP, Assefnia S, Chandrasekharan S, Backlund MG, Yin Y et al (2007) Identification of conserved gene expression features between murine mammary carcinoma models and human breast tumors. Genome Biol 8:R76 15. Cressman VL, Backlund DC, Hicks EM, Gowen LC, Godfrey V, Koller BH (1999) Mammary tumor formation in p53- and BRCA1-deficient mice. Cell Growth Differ 10:1–10 16. Jiang Z, Deng T, Jones R, Li H, Herschkowitz JI, Liu JC, Weigman VJ, Tsao MS, Lane TF, Perou CM, Zacksenhaus E (2010) Rb deletion in mouse mammary progenitors induces luminal-B or basal-like/EMT tumor subtypes depending on p53 status. J Clin Investg 120:3296–3309 17. Lopez-Knowles E, O’Toole SA, McNeil CM, Millar EK, Qiu MR, Crea P, Daly RJ, Musgrove EA, Sutherland RL (2010) PI3K pathway activation in breast cancer is associated with the basallike phenotype and cancer-specific mortality. Int J Cancer 126:1121–1131 18. Jackson SP, Bartek J (2009) The DNA-damage response in human biology and disease. Nature 461:1071–1078 19. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, Jacobsen A, Byrne CJ, Heuer ML, Larsson E, Antipin Y, Reva B et al (2012) The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer Discov 2:401–404 20. Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, Sun Y, Jacobsen A, Sinha R, Larsson E, Cerami E, Sander C et al
371
21.
22.
23.
24.
25. 26. 27.
28.
29.
30.
31.
32.
33.
(2013) Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci Signal 6:pl1 Beroukhim R, Getz G, Nghiemphu L, Barretina J, Hsueh T, Linhart D, Vivanco I, Lee JC, Huang JH, Alexander S, Du J, Kau T et al (2007) Assessing the significance of chromosomal aberrations in cancer: methodology and application to glioma. Proc Natl Acad Sci U S A 104:20007–20012 Hennessy BT, Lu Y, Gonzalez-Angulo AM, Carey MS, Myhre S, Ju Z, Davies MA, Liu W, Coombes K, Meric-Bernstam F, Bedrosian I, McGahren M et al (2010) A technical assessment of the utility of reverse phase protein arrays for the study of the functional proteome in non-microdissected human breast cancers. Clin Proteomics 6:129–151 The Cancer Genome Atlas Research Network (2012) Comprehensive molecular portraits of human breast tumours. Nature 490:61–70 Gyorffy B, Lanczky A, Eklund AC, Denkert C, Budczies J, Li Q, Szallasi Z (2010) An online survival analysis tool to rapidly assess the effect of 22,277 genes on breast cancer prognosis using microarray data of 1,809 patients. Breast Cancer Res Treat 123:725–731 The Cancer Genome Atlas Research Network (2011) Integrated genomic analyses of ovarian carcinoma. Nature 474:609–615 Smit AFA, Hubley R, Green P. RepeatMasker Open-4.0. Available at: http://www.repeatmasker.org. Accessed 15 Aug 2014 Gewinner C, Wang ZC, Richardson A, Teruya-Feldstein J, Etemadmoghadam D, Bowtell D, Barretina J, Lin WM, Rameh L, Salmena L, Pandolfi PP, Cantley LC (2009) Evidence that inositol polyphosphate 4-phosphatase type II is a tumor suppressor that inhibits PI3K signaling. Cancer Cell 16:115–125 Fedele CG, Ooms LM, Ho M, Vieusseux J, O’Toole SA, Millar EK, Lopez-Knowles E, Sriratana A, Gurung R, Baglietto L, Giles GG, Bailey CG et al (2010) Inositol polyphosphate 4-phosphatase II regulates PI3K/Akt signaling and is lost in human basal-like breast cancers. Proc Natl Acad Sci U S A 107:22231–22236 Bassi C, Ho J, Srikumar T, Dowling RJ, Gorrini C, Miller SJ, Mak TW, Neel BG, Raught B, Stambolic V (2013) Nuclear PTEN controls DNA repair and sensitivity to genotoxic stress. Science 341:395–399 Min JW, Kim KI, Kim HA, Kim EK, Noh WC, Jeon HB, Cho DH, Oh JS, Park IC, Hwang SG, Kim JS (2013) INPP4B-mediated tumor resistance is associated with modulation of glucose metabolism via hexokinase 2 regulation in laryngeal cancer cells. Biochem Biophys Res Commun 440:137–142 Kim JS, Yun HS, Um HD, Park JK, Lee KH, Kang CM, Lee SJ, Hwang SG (2012) Identification of inositol polyphosphate 4-phosphatase type II as a novel tumor resistance biomarker in human laryngeal cancer HEp-2 cells. Cancer Biol Ther 13:1307–1318 Ewald B, Sampath D, Plunkett W (2008) ATM and the Mre11Rad50-Nbs1 complex respond to nucleoside analogue-induced stalled replication forks and contribute to drug resistance. Cancer Res 68:7947–7955 Abuzeid WM, Jiang X, Shi G, Wang H, Paulson D, Araki K, Jungreis D, Carney J, O’Malley BW Jr, Li D (2009) Molecular disruption of RAD50 sensitizes human tumor cells to cisplatinbased chemotherapy. J Clin Investg 119:1974–1985
123
![or metastatic breast cancer].](/assets/img/hold.png)
