Transcriptomic and functional pathway features were associated with survival after pathological complete response to neoadjuvant chemotherapy in breast cancer

Abstract

Pathological complete response (pCR) after neoadjuvant chemotherapy (NAC) has been proposed as a surrogate endpoint for the prediction of long-term survival in breast cancer (BC); however, an increased pCR rate has not clearly correlated with improved survival. We hypothesized that some transcriptomic and functional pathway features correlate with survival after pCR in BC. We utilized 2 published NAC cohorts, 105 women with gene expression data before, “Baseline”, and that changed during NAC, “Delta”, and TCGA database with 1068 BC patients to investigate the relationship between the efficacy of NAC and survival utilizing differentially expressed-mRNAs, construction and analysis of the mRNA-hub gene network, and functional pathway analysis. In mRNA expression profiling, S100A8 was a gene involved in survival after pCR in Baseline and NDP was a gene involved in recurrence after pCR in Delta. In functional pathway analysis, we found multiple pathways involved in survival after pCR. In mRNA-hub gene analysis, HSP90AA1, EEF1A1, APP, and HSPA4 were related to recurrence in BC patients with pCR due to NAC. TP53, EGFR, CTNNB1, ERBB2, and HSPB1 may play a significant role in survival for patients with pCR. Interestingly, high HSP90AA1, HSPA4, S100A8, and TP53, and low EEF1A1, EGFR, and CTNNB1 expressing tumors have significantly worse overall survival in TCGA BC cohort. We demonstrated the genes and functional pathway features associated with pCR and survival utilizing the bioinformatics approach to public BC cohorts. Some genes involved in recurrence after pCR due to NAC also served as prognostic factors in primary BC.

Introduction

Breast cancer (BC) is the most commonly diagnosed cancer and the second leading cause of cancer deaths among American women, and thus has been identified as a public health priority in the United States. The lifetime risk of developing BC today is one in every eight women [1]. Currently, surgery, radiotherapy, and chemo-/endocrine-therapy are the major treatment options for BC. Neoadjuvant chemotherapy (NAC), which is systemic therapy delivered before definitive BC surgery, has been widely applied for the following three reasons. First, NAC reduces the size and extent of locally advanced tumors, which allows for breast conserving surgery [2]. Second, NAC allows for early identification of unresponsive tumors and provides an opportunity to terminate the ineffective therapy and/or to switch to an alternative regimen [3]. Indeed, NAC trials have been used for the rapid assessment of drug efficacy that sped up the development and approval of drugs for early BC during the last two decades [4]. For example, the GeparTrio study showed that the regimen based upon NAC response was significantly better in prolonging disease free survival (DFS) and overall survival (OS) than a non-individualized approach with a fixed number of cycles, especially among patients with hormone receptor (HR)-positive tumors [5-7]. Third, response to NAC is used as an early predictive indicator of the prognosis of BC patients.

In general, pathological complete response (pCR) has been used as a surrogate endpoint for the prediction of long-term survival such as DFS and OS [2]; however, this notion has recently been challenged. First, responses to conventional NAC differ by the BC subtypes, complicating the investigation of the predictive value of biomarkers. Thus, pCR is currently utilized as a “surrogate marker” for accelerated drug registration only in aggressive BC subtypes such as triple negative (TN) or human epidermal growth factor receptor 2 (HER2)-positive cases [8-10]. Second, BC cells may remain in dormancy and survive in patients that achieved pCR to NAC. Multiple mechanisms have been proposed to explain how cancer cells become dormant, and how they become reactivated and exit dormancy [11].

To this end, further elucidation of the relationship between pCR and survival should enhance the role of pCR after NAC as a surrogate marker for survival for the BC patients. One of the approaches to exploit the full potential of NAC is to identify the key genes that are expressed prior to and that changed during the treatment and correlate them with survival. We hypothesized that some transcriptomic and functional pathway features correlate with pCR to NAC and survival in BC cohorts. To test this hypothesis we utilized mRNA expression profiles, construction and analysis of the mRNA-hub gene network, and functional and pathway enrichment analysis.

Materials and methods

Neoadjuvant chemotherapy cohorts

Two Gene Expression Omnibus (GEO) datasets, GSE32603 and GSE87455, were used to examine the association between response to anthracycline-based chemotherapy and survival in patients who underwent NAC (Figure S1). Microarray gene expression data in GEO datasets (http://www.ncbi.nlm.nih.gov/geo) were queried from the National Center for Biotechnology Information. In the GSE32603 cohort, out of 46 primary BC patients treated with anthracycline based chemotherapy (AC) followed by optimal taxane based chemotherapy (OTC), 36 women who had both gene expression data for before (T1) and during (T2) AC were analyzed [12]. In the GSE87455 cohort, out of 150 primary BC patients treated with epirubicin + docetaxel + bevacizumab (EDB), 69 women who had both gene expression data for T1 and T2 were analyzed [13]. Both cohorts were used to support the authenticity of the association between the effect of NAC and clinical outcomes. We defined the gene expression profile in T1 as “Baseline”, and the change from T1 to T2 as “Delta”.

Screening for differentially expressed mRNAs

The Student’s t-test was used to compare the difference between binary variables. Top 10 differentially expressed-mRNAs were selected based on |log2[fold change (FC)]|. P-value <0.05 and |log2FC| >0.17 were set as the thresholds for screening differentially expressed-mRNAs. This screening method was referred to as “Previous analysis” in prior publication [14].

Gene Ontology (GO) annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis

GO annotation and KEGG pathway enrichment analyses for the predicted target genes of the top 20 differentially expressed-mRNAs were conducted by using the R package “clusterProfiler”.

Construction and analysis of mRNA hub gene network

To assess the interactive relationship among the predicted target genes in each category, the STRING database (http://string-db.org) was used to construct the mRNA-hub gene network as follows; first, we chose 10 genes with the largest difference in expression using mRNA data. Next, we built the network using these top 10 genes and their interactors (genes that interact with them). Finally, we screened out the top 25 hub genes through the resulting network.

mRNA expression data from the cancer genome atlas (TCGA)

TCGA was supervised by the National Cancer Institute (NCI) and the National Human Genome Research Institute [15]. The gene expression levels (mRNA expression z-score from RNA-sequence) from Genomic Identification of Significant Targets in Cancer for TCGA cohort was downloaded through cBioportal (TCGA, PanCancer Atlas) [16,17] as described before [18,19]. The expression levels of potential mRNAs were extracted from the downloaded mRNA information.

Statistical analysis

All statistical analyses were performed using R software (http://www.r-project.org/) and Bioconductor (http://bioconductor.org/). The student-t test was used to assess baseline differences between binary variables. In the analysis of OS, the Kaplan-Meier method was used to estimate survival rates, and differences between survival curves were evaluated by the log-rank test. Differences were considered significant with a P-value <0.05.

Results

Identification of mRNA expression profile involved in recurrence after pCR in NAC cohorts

To clarify the relationship between pCR and survival, we divided BC patients into two groups in two ways in each cohort (Figure S2): Category 1, we divided BC patients into two groups with clinical integrity, clinical concordance versus discordance. Clinical concordance was defined as a group of patients who achieved pCR/not relapsed for 5 years, and non-pCR/relapsed within 5 years, after NAC. On the other hand, clinical discordance was defined as a group of patients who achieved pCR/relapsed and non-pCR/not relapsed, after NAC (Figure S2A). Category 2, we divided NAC treated BC patients according to whether they relapsed in the group that achieved pCR (Figure S2B). By using Category 1 as supporting data for Category 2, the genes associated with recurrence in pCR could be determined more accurately (Figure S2). Two gene expression microarray data sets GSE32603 and GSE87455 were downloaded from the GEO database [12,13]. The Student’s t-test was used to compare the difference between binary variables (Category 1: clinical concordance to discordance; Category 2: recurrence to no recurrence among pCR patients). The data were processed by unpaired t-test (P<0.05, |log2FC| >0.17). Here, we explored what kind of genes were up- or down-regulated in the course of treatment with chemotherapy. In Baseline analysis (analysis of T1 in Figure S1), the top 10 mRNAs with more or less mRNA expression at baseline are listed in Table 1. There was no common gene as Category 1 between GSE32603 and GSE87455. In Category 2, S100A8 was the common gene between GSE32603 and GSE87455. Since S100A8 decreased (no recurrence after pCR) in GSE32603 and increased (recurred after pCR) in GSE87455, it was suggested that S100A8 may behave differently in each regimen. In Delta analysis (analysis of the gene expression difference between T1 and T2 in Figure S1), in Category 1, 140 mRNAs (62 up-regulated and 78 down-regulated mRNAs) were screened out in the GSE32603 and 9 mRNAs (2 up-regulated and 7 down-regulated mRNAs) were screened out in the GSE87455. In Category 2, 1302 mRNAs (429 up-regulated and 873 down-regulated mRNAs) were screened out in the GSE32603 and 279 mRNAs (139 up-regulated and 140 down-regulated mRNAs) were screened out in the GSE87455. The top 10 mRNAs with a remarkable difference between T1 and T2 are listed in Table 2. NDP was recognized in Category 1 and 2 of GSE87455. Since NDP was decreased (clinical discordance) in Category 1 and increased (recurrence after pCR) in Category 2, it was speculated to be a specific gene involved in recurrence after pCR.

Table 1.

Identification of top 10 differentially expressed mRNAs in each category of each NAC cohort in Baseline

Category 1						Category 2
GSE32603			GSE87455			GSE32603			GSE87455
mRNA	logFC	P-Value	mRNA	logFC	P-Value	mRNA	logFC	P-Value	mRNA	logFC	P-Value
C4A	-1.574	0.002	ANKRD30A	-2.154	0.001	LITAF	3.450	0.011	MUCL1	3.597	0.022
GTF2IRD1	1.436	0.005	AGR2	-2.043	0.000	LTF	-3.277	0.006	S100P	3.176	0.015
NTN4	-1.404	0.020	FOXA1	-2.023	0.000	UBD	-3.025	0.010	MUC1	2.742	0.004
C1orf35	1.218	0.003	TFF3	-1.992	0.002	CDH2	-2.887	0.006	S100A7	2.720	0.000
F2RL2	-1.145	0.025	CEACAM6	-1.877	0.004	LYZ	-2.695	0.007	S100A9	2.708	0.006
RARRES3	-1.127	0.035	GABRP	1.581	0.003	MAOB	-2.650	0.008	S100A8	2.459	0.044
CSK	1.119	0.029	PIP	-1.533	0.030	S100A8	-2.559	0.019	HBA2	-2.249	0.031
TRIM29	-1.082	0.048	MLPH	-1.491	0.001	EFHD1	2.511	0.037	HBB	-2.243	0.030
KRT19	-1.076	0.038	TFF1	-1.447	0.022	RMND5B	2.354	0.002	VTCN1	2.213	0.000
SLC40A1	-1.054	0.001	SRARP	-1.409	0.004	KRT18	-2.342	0.021	HBA1	-2.175	0.033

Abbreviations: NAC, neoadjuvant chemotherapy; FC, fold change.

Table 2.

Identification of top 10 differentially expressed mRNAs in each category of each NAC cohort in Delta

Category 1						Category 2
GSE32603			GSE87455			GSE32603			GSE87455
mRNA	logFC	P-Value	mRNA	logFC	P-Value	mRNA	logFC	P-Value	mRNA	logFC	P-Value
CYC1	1.558	0.031	CPB1	-1.106	0.020	LITAF	-4.238	0.009	ALB	-2.651	0.028
C1orf35	-1.557	0.009	NDP	-0.924	0.009	ACRV1	-3.879	0.044	NDP	2.247	0.032
CXCL9	1.505	0.020	SCGB2A1	-0.897	0.018	DNAH14	3.681	0.002	GJA1	2.157	0.030
SLC4A4	1.398	0.025	WFDC2	-0.865	0.010	CPB1	3.526	0.002	MOXD1	2.131	0.013
MMP7	1.350	0.016	CRABP1	-0.827	0.003	JAM2	3.173	0.011	CRYAB	-1.939	0.006
CXCL10	1.349	0.026	APOC1	0.719	0.016	PVALB	3.016	0.009	CTXN1	1.925	0.002
FGFR1	-1.318	0.048	MSLN	-0.677	0.021	CEP55	-2.902	0.010	C1orf115	-1.823	0.0002
SIVA1	1.270	0.018	HDC	-0.670	0.021	ISG15	2.838	0.002	THBS4	-1.771	0.036
CNN3	1.223	0.006	DCD	0.646	0.015	ZCCHC9	-2.800	0.011	VASH2	1.638	0.049
FAM26F	1.203	0.003	GRIA2	-0.573	0.010	LRRC2	2.799	0.005	ANGPTL8	-1.632	0.017

Abbreviations: NAC, neoadjuvant chemotherapy; FC, fold change.

Construction and analysis of mRNA hub gene network

The construction of the mRNA-hub gene network is considerably helpful for us to identify the most potentially functional mRNAs [20]. Here, we explored the relationship between mRNAs in each category of each GEO. After analyzing the data from STRING using Cytoscape software, we first screened out the top 25 hub nodes according to degree related with the top 10 mRNA with more changes in each NAC cohort in Category 1 and 2, in the same way as published in [14] (Tables 3 and 4). For better visualization of the interactions of these hub genes, additionally, we constructed networks based on the screened top 25 hub genes related with the top 10 mRNAs with more changes, as presented in Figures 1 and 2. In Baseline analysis, the top 10 hub genes were SRC, HSP90AA1, CSK, JUN, FYN, SHC1, YWHAE, PXN, UBE2I, and HIST1H4F, among which SRC showed the highest node degree (degree = 43) in Category 1 of GSE32603. In Category 1 of GSE87455, the top 10 hub genes were GAPDH, HSPA8, EEF2, HSP90AA1, RPS3, RPL4, RPLP0, RPS2, EEF1A1, and EIF4A3, among which GAPDH showed the highest node degree (degree = 305). In Category 2 of GSE32603, the top 10 hub genes were TP53, EGFR, HSP90AA1, MUC1, ERBB2, SRC, ESR1, JUN, HSPA4, and CTNNB1, among which TP53 showed the highest node degree (degree = 37). In GSE87455, the top 10 hub genes were UBB, TP53, UBD, EGFR, CTNNB1, CDH2, SKP1, UBQLN1, ITCH, and PSMD4, among which UBB showed the highest node degree (degree = 50) (Table 3). Of note, EEF1A1 in GSE32603 and HSP90AA1 and HSPA4 in GSE87455 were genes common to Category 1 and 2. It was suggested that EEF1A1 may have a role in BC recurrence after pCR due to AC followed by OTC. On the other hand, HSP90AA1 and HSPA4 may have a role in BC recurrence after pCR due to EDB. TP53, EGFR, CTNNB1, ERBB2, and APP were genes common to GSE32603 and GSE87455 in Category 2, suggesting that these mRNAs may play a significant role in survival for patients with pCR (Figure 1).

Table 3.

Identification of top 25 hub nodes according to degree related with the top 10 mRNA with more changes in each category of each NAC cohort in Baseline

Category 1				Category 2
GSE32603		GSE87455		GSE32603		GSE87455
Gene Symbol	Degree	Gene Symbol	Degree	Gene Symbol	Degree	Gene Symbol	Degree
SRC	43	GAPDH	305	UBB	50	TP53	37
HSP90AA1	41	HSPA8	260	TP53	44	EGFR	31
CSK	33	EEF2	236	UBD	37	HSP90AA1	30
JUN	29	HSP90AA1	225	EGFR	36	MUC1	25
FYN	26	RPS3	225	CTNNB1	31	ERBB2	25
SHC1	24	RPL4	222	CDH2	30	SRC	25
YWHAE	23	RPLP0	215	SKP1	26	ESR1	25
PXN	22	RPS2	214	UBQLN1	26	JUN	23
UBE2I	19	EEF1A1	212	ITCH	25	HSPA4	21
HIST1H4F	18	EIF4A3	211	PSMD4	25	CTNNB1	21
AKT2	18	RPS16	208	EEF1A1	24	APP	21
IKBKG	18	RPS20	207	WWP1	22	ABL1	20
EIF4E	18	CCT2	205	NEDD4	22	GRB2	19
PARP1	18	EPRS	202	NEDD4L	22	SRSF7	19
PTK2	17	RPS6	201	PSMA6	21	SRSF11	19
CAV1	17	GNB2L1	201	RPL8	20	U2AF2	18
EEF1A1	16	RPS14	201	PSMC3	20	TOP1	17
PDPK1	16	RPS9	200	CDC34	19	SRSF5	16
ZAP70	16	NOP56	200	RPS16	19	SRSF3	15
IGF1R	16	RPL8	198	CCNA2	19	GSK3B	15
H1F0	16	RPL5	196	TSG101	18	SAP18	15
HNRNPK	15	RPL7	196	RPL3	17	TBP	15
ILK	15	HSPA4	194	ERBB2	17	PPARG	15
HSPA5	15	HNRNPA1	194	APP	17	TRA2B	14
TRAT1	15	RPL11	194	HGS	16	PRKCD	14

Abbreviations: NAC, neoadjuvant chemotherapy.

Table 4.

Identification of top 25 hub nodes according to degree related with the top 10 mRNA with more changes in each category of each NAC cohort in Delta

Category 1				Category 2
GSE32603		GSE87455		GSE32603		GSE87455
Gene symbol	Degree	Gene symbol	Degree	Gene symbol	Degree	Gene symbol	Degree
AKT1	82	GRIA2	11	GAPDH	87	ALB	76
TP53	70	GRIA1	8	HSPA8	74	TP53	68
SRC	67	MSLN	8	HSP90AA1	72	INS	60
FGFR1	58	APP	7	RPS27A	63	AKT1	59
PIK3R1	57	GRIP1	6	HSPA4	61	FN1	51
CTNNB1	51	RAB11A	6	UBB	57	MAPK1	49
PDGFRB	49	NDUFB5	5	ENO1	52	APP	47
HSP90AA1	47	MRPL28	4	EEF2	47	CASP3	46
FGF2	47	PICK1	4	VCP	45	VEGFA	44
PTPN11	46	ATP5F1	4	ISG15	44	CTNNB1	41
SOS1	44	MYO5A	4	HNRNPA2B1	44	CCND1	36
CYC1	41	GAPDH	4	GNB2L1	42	CDH1	36
CDH1	40	NDUFB9	4	HSPD1	41	CRYAB	35
STAT3	40	MRPL16	4	HSP90AB1	41	APOE	35
PIK3R2	39	NDUFV3	3	CCT2	39	UBC	35
ERBB3	38	RALA	3	PKM	38	GJA1	31
UBC	37	SDCBP	3	HNRNPK	37	F2	30
PLCG1	35	GRN	3	PGK1	36	SNCA	29
GAB1	35	MRPL1	3	ELAVL1	36	APOA1	29
JAK2	35	EIF1AX	2	EEF1G	35	FGF2	28
SHC1	35	RNMTL1	2	TSG101	35	FGA	28
MDM2	31	GRID2	2	HSPB1	34	HSPB1	27
CRK	31	PTPRF	2	CCT4	34	HP	27
VDAC1	31	AGAP2	2	ANXA2	33	TJP1	26
VAV1	31	TRMT6	1	EIF4G1	33	BCL2L1	26

Abbreviations: NAC, neoadjuvant chemotherapy.

Figure 1.

The top 25 mRNA-hub genes in baseline analysis. (A) mRNA hub gene network and (B) Venn diagram. Category 1, we divided BC patients into two groups with clinical integrity, clinical concordance versus discordance. Clinical concordance was defined as a group of patients who were pCR/not relapsed and non-pCR/relapsed, according to NAC. On the other hand, clinical discordance was defined as a group of patients who were pCR/relapsed and non-pCR/not relapsed, according to NAC. Category 2, we divided NAC treated BC patients according to whether they relapsed in the group having pCR. Two microarray data sets GSE32603 and GSE87455 were downloaded from the GEO database. The data were processed by unpaired t-test (P<0.05, |log2FC| >0.17). Abbreviations: BC, breast cancer; pCR, pathological complete response; NAC, neoadjuvant chemotherapy; GEO, Gene Expression Omnibus.

Figure 2.

The top 25 mRNA-hub genes in delta analysis. (A) mRNA hub gene network and (B) Venn diagram. Category 1, we divided BC patients into two groups with clinical integrity, clinical concordance versus discordance. Clinical concordance was defined as a group of patients who were pCR/not relapsed and non-pCR/relapsed, according to NAC. On the other hand, clinical discordance was defined as a group of patients who were pCR/relapsed and non-pCR/not relapsed, according to NAC. Category 2, we divided NAC treated BC patients according to whether they relapsed in the group having pCR. Two microarray data sets GSE32603 and GSE87455 were downloaded from the GEO database. The data were processed by unpaired t-test (P<0.05, |log2FC| >0.17). Abbreviations: BC, breast cancer; pCR, pathological complete response; NAC, neoadjuvant chemotherapy; GEO, Gene Expression Omnibus.

In Delta analysis, the top 10 hub genes were AKT1, TP53, SRC, FGFR1, PIK3R1, CTNNB1, PDGFRB, HSP90AA1, FGF2, and PTPN11, among which AKT1 showed the highest node degree (degree = 82) in Category 1 of GSE32603. In Category 1 of GSE87455, the top 10 hub genes were GRIA2, GRIA1, MSLN, APP, GRIP1, RAB11A, NDUFB5, MRPL28, PICK1, and ATP5F1, among which GRIA2 showed the highest node degree (degree = 11). In Category 2 of GSE32603 the top 10 hub genes were GAPDH, HSPA8, HSP90AA1, RPS27A, HSPA4, UBB, ENO1, EEF2, VCP, and ISG15, among which GAPDH showed the highest node degree (degree = 87). In GSE87455 the top 10 hub genes were ALB, TP53, INS, AKT1, FN1, MAPK1, APP, CASP3, VEGFA, and CTNNB1, among which ALB showed the highest node degree (degree = 76) (Table 4). Of note, HSP90AA1 in GSE32603 and APP in GSE87455 were genes common to Category 1 and 2. It was suggested that HSP90AA1 may have a role in BC recurrence of BC after pCR due to AC followed by OTC. On the other hand, APP may have a role in BC recurrence after pCR due to EDB. HSPB1 was the gene common to GSE32603 and GSE87455 in Category 2, suggesting that these mRNAs may play a significant role in BC patient survival after pCR. In summary, HSP90AA1 was identified by both analyses as a common gene for Category 1 and 2 and is suggested to strongly correlate with BC recurrence after NAC.

Functional and pathway enrichment analyses

To further explore the systematic features and biological functions of the identified genes, GO functional annotation and KEGG pathway enrichment analyses were performed by R package, “clusterProfiler”. In Baseline analysis, the GO terms of the identified genes of the top 10 mRNAs with more changes are shown in Figure 3A. In Category 1, there was no significant pathway in the GSE32603 and GSE87455. In Category 2, there was no significant pathway in the GSE32603, but the GSE87455 shows RAGE receptor binding, Haptoglobin binding, Organic acid binding, Oxygen carrier activity, Antioxidant activity, Oxygen binding, Molecular carrier activity, Peroxidase activity, Oxidoreductase activity, acting on peroxide as receptor, and Toll-like receptor binding.

Figure 3.

GO annotation analysis for the target genes of the top 10 most downregulated miRNAs. (A) Baseline analysis and (B) delta analysis. Category 1, we divided BC patients into two groups with clinical integrity, clinical concordance versus discordance. Clinical concordance was defined as a group of patients who were pCR/not relapsed and non-pCR/relapsed, according to NAC. On the other hand, clinical discordance was defined as a group of patients who were pCR/relapsed and non-pCR/not relapsed, according to NAC. Category 2, we divided NAC treated BC patients according to whether they relapsed in the group having pCR. Two microarray data sets GSE32603 and GSE87455 were downloaded from the GEO database. The data were processed by unpaired t-test (P<0.05, |log2FC| >0.17). Abbreviations: GO, Gene Ontology. BC, breast cancer; pCR, pathological complete response; NAC, neoadjuvant chemotherapy; GEO, Gene Expression Omnibus.

In Delta analysis, the GO terms of the target genes of the top 10 mRNAs with more changes are shown in Figure 3B. In Category 1, several pathways (CXCR chemokine receptor binding, Heparin binding, Glycosaminoglycan binding, Sulfur compound binding, Chemokine activity, and Cytokine receptor binding in the GSE32603 and Menocarboxylic acid binding alone in the GSE87455) were significant, but no pathway was common in Category 2. In Category 2 GSE32603 demonstrated Protein tag, ATP-dependent microtubule motor activity, minus-end-directed, Dynein intermediate chain binding, Dynein light intermediate chain binding, Metallocarboxypeptidase activity, WW domain binding, ATP-dependent microtubule motor activity, and Carboxypeptidase activity. The GSE87455 showed Copper ion binding, Receptor ligand activity, and Growth factor activity. There was no significant pathway in KEGG pathway analysis in both categories of both cohorts. Thus, in both Baseline and Delta analyses, there was no common pathway between Category 1 and 2.

Some genes involved in recurrence after pCR to NAC were also useful as prognostic factors in primary BC

Next, we explored the prognostic relevance of genes extracted by the above analysis (Gene common to Category 1 and 2 in Baseline and Delta analysis: HSP90AA1, genes common to Category 1 and 2 in Baseline or Delta analysis: EEF1A1, NDP, APP, and HSPA4, genes identified only in Category 2 in Baseline or Delta analysis: S100A8, TP53, EGFR, CTNNB1, ERBB2, and HSPB1) utilizing a large BC cohort, TCGA (Figure 4). A total of 150 (14%) of 1068 BC patients in TCGA died, which were regarded as events when analyzing OS. Patients with a high expression of HSP90AA1 (P<0.001), HSPA4 (P = 0.022), S100A8 (P = 0.0017), and TP53 (P = 0.012) mRNA and patients with low expression of EEF1A1 (P<0.001), EGFR (P = 0.029), and CTNNB1 (P = 0.026) mRNA were significantly associated with worse OS, as evaluated by the Kaplan-Meier method and verified by the log-rank test. There was no statistically significant correlation between NDP, APP, ERBB2, and HSPB1 and OS in TCGA. These data suggest that some genes involved in recurrence after pCR due to NAC were also useful as prognostic factors in primary BC.

Figure 4.

Kaplan-Meier plots of the association of the presence of (A) gene common to Category 1 and 2 in baseline and delta analysis: HSP90AA1, (B) gene common to Category 1 and 2 in baseline analysis: EEF1A1, and (C) in delta analysis: NDP, APP, and HSPA4, (D) genes identified only in Category 2 in baseline: S100A8, TP53, EGFR, CTNNB1, and ERBB2 and (E) in delta analysis: HSPB1, with OS in TCGA BC cohort. The cutoff of each mRNA expression was defined as optimal cutoff to show the difference of OS. Category 1, we divided BC patients into two groups with clinical integrity, clinical concordance versus discordance. Clinical concordance was defined as a group of patients who were pCR/not relapsed and non-pCR/relapsed, according to NAC. On the other hand, clinical discordance was defined as a group of patients who were pCR/relapsed and non-pCR/not relapsed, according to NAC. Category 2, we divided NAC treated BC patients according to whether they relapsed in the group having pCR. Abbreviations: OS, overall survival; TCGA, The Cancer Genome Atlas; BC, breast cancer; NS, not significant.

Discussion

NAC is widely used to treat early-stage BC and provides an improvement in the breast conservation rate by tumor volume reduction and the early identification of sensitivity to treatment. While the achievement of pCR became the goal of NAC with the expectation of a concomitant improvement in patient survival [2], the predictive value of pCR for patient outcomes remains controversial due to variances according to the different biological subtypes [8-10] and because even after pCR after NAC cancer cells killed were not completely eradicated but instead merely lay dormant until BC recurrence [11]. We identified the genes and the pathways involved in the relationship between pCR and survival in BC cohorts by identifying differentially expressed mRNAs, by construction and analysis of the mRNA-hub gene network, and by functional and pathway enrichment analysis.

This study generated three interesting results with clinical implications. First, in Baseline analysis and Delta analysis, we identified genes involved in the relationship between pCR and survival. In mRNA expression profiling, NDP was a gene involved in recurrence after pCR, and it was related to clinical discordance and recurrence after pCR in Delta analysis (Table 2). NDP gene encodes a protein called Norrin that plays a role in Wnt signaling, one of the key signaling pathways for cell proliferation, adhesion, migration, and many other cellular activities including cancer stem cell biology [21]. We also found that S100A8 may behave differently in each NAC regimen since S100A8 related to no recurrence after pCR in GSE32603, but related to recurrence after pCR in GSE87455 in Baseline analysis (Table 1). S100A8 gene encodes S100 calcium-binding protein A8, which is involved in the regulation of several cellular processes such as cell cycle progression and differentiation. Yang et al reported that S100A8 was associated with BC drug resistance by proteomics/bioinformatics approach [22]. In the analysis of the mRNA hub gene network, HSP90AA1 was involved in recurrence after pCR due to EDB in baseline and delta analyses (Tables 3 and 4; Figures 1 and 2). HSP90AA1 (Heat Shock Protein 90 Alpha Family Class A Member 1) is a Protein Coding gene. HSP90 is required for the stabilization of many proteins in pathways that play key roles in BC growth and survival, such as estrogen receptor (ER), progesterone receptor (PR), essential components of HER2 signaling (HER2, AKT, c-SRC, RAF and HIF-1α), and EGFR [23]. Only in Baseline analysis, EEF1A1 was involved in recurrence after pCR to AC followed by OTC (Table 3 and Figure 1). EEF1A1 gene encodes an isoform of the alpha subunit of the eukaryotic elongation factor 1 (EEF1) complex, which is responsible for the enzymatic delivery of aminoacyl tRNAs to the ribosome [24]. Only in Delta analysis, APP and HSPA4 were involved in recurrence after pCR to EDB (Table 4 and Figure 2). High expression of APP (Amyloid-β precursor protein) mRNA is causally linked to tumorigenicity as well as the invasion of aggressive BC [25]. Cao and colleagues reported that HSPA4 indirectly promoted lymph node metastasis by targeting pathogenic IgG produced by B cells [26]. To our knowledge this is the first report that NDP, S100A8, HSP90AA1, EEF1A1, APP, and HSPA4 are involved in recurrence after pCR due to NAC. In addition, TP53, EGFR, ERBB2, CTNNB1, and HSPB1 may play a significant role in the survival of patients after pCR. Only in Category 2, TP53, EGFR, ERBB2, and CTNNB1 were recognized only in Baseline analysis (Table 3 and Figure 1) and HSPB1 was recognized only in Delta analysis (Table 4 and Figure 2). TP53, EGFR, and ERBB2 are genes that are deeply involved in BC development, biology, and BC treatment. A high level of expression of CTNNB1 (the gene that codes β-catenin) mRNA is a strong predictor for a favorable prognosis in gastric carcinoma, without any reported clinical role for CTNNB1 expression in BC [27]. HSPB1 (Heat Shock Protein Family B (Small) Member 1) upregulation is associated with tumor growth and resistance to chemo- and radio-therapeutic treatments. Interestingly, Gibert and colleagues demonstrated that HSPB1 silencing led to reduced cell migration and invasion in vitro and that in vivo it correlated with a decreased ability of BC cells to metastasize and grow in the skeleton [28].

Second, some genes involved in recurrence after pCR due to NAC were also useful as prognostic factors in primary BC. Patients with a high level of expression of HSP90AA1 (P<0.001), HSPA4 (P = 0.022), S100A8 (P = 0.0017), and TP53 (P = 0.012) mRNA and patients with low level of expression of EEF1A1 (P<0.001), EGFR (P = 0.029), and CTNNB1 (P = 0.026) mRNA were significantly associated with worse OS (Figure 4). High-level expression of HSP90AA1, one of two cytoplasmic HSP90 isoforms, was driven by chromosome coding region amplifications and was ab independent factors that led to death from BC among patients with TN and HER2-/ER+ subtypes, respectively [29]. High serum anti-HSPA4 IgG was correlated with high tumor HSPA4 expression and a poor prognosis for BC subjects [26]. Thus, HSP90AA1 and HSPA4 were compatible with previous reports. On the other hand, high EEF1A2 expression, one of two EEF1A isoforms, was a marker for good outcome in BC [30]. However, it has never been examined whether EEF1A1 expression has any prognostic value in BC.

Third, in functional and pathway enrichment analysis, we found multiple pathways involved in survival after pCR to NAC. In both Baseline and Delta analyses, there was no common pathway between Category 1 and 2, and some significant pathways in Category 2 were associated with survival after pCR to NAC. In Baseline analysis, in the GSE87455, RAGE receptor binding, Haptoglobin binding, Organic acid binding, Oxygen carrier activity, Antioxidant activity, Oxygen binding, Molecular carrier activity, Peroxidase activity, Oxidoreductase activity, acting on peroxide as receptor, and Toll-like receptor binding may be the pathways involved in clinical outcome after pCR (Figure 3A). Of note, Peroxidase activity and Oxidoreductase activity have been demonstrated to be associated with chemotherapy resistance in the course of cancer treatment [31,32]. In Delta analysis, in the GSE32603, Protein tag, ATP-dependent microtubule motor activity, minus-end-directed, Dynein intermediate chain binding, Dynein light intermediate chain binding, Metallocarboxypeptidase activity, WW domain binding, ATP-dependent microtubule motor activity, and Carboxypeptidase activity and in the GSE87455, Copper ion binding, Receptor ligand activity, and Growth factor activity may be the pathways involved in survival after pCR (Figure 3B). All these findings suggested that the functional pathways extracted by delta analysis have never been demonstrated to be associated with chemotherapy resistance in the course of cancer treatment and they are worth studying as new therapeutic targets.

Although the study demonstrates promising results, it has limitations. First, this is a retrospective study utilizing publicly available datasets (GSE32603, GSE87455, and TCGA), thus it is prone to selection bias. Second, this study does not include any in vitro or in vivo experiments that proves the mechanism of our results to further understand the correlations reported. Third, due to the small number of patients in NAC cohorts, we were unable to evaluate the data by subtype. It is known that achievement of pCR strongly predicted improved survival in TNBC and HER2-enriched BC subtypes, while data remain controversial for the luminal subtypes. Fourth, our dataset allowed us to evaluate only a single point of gene expression during NAC. Liquid biopsy, which is a non-invasively conducted genetic test using genes extracted from body fluids such as blood or urine, has been developed as a way of providing equivalent or better information obtained from genes in tumor tissue as previously demonstrated [33-37]. If transcriptomes can be monitored by liquid biopsy, it is expected to deepen the understanding of the relationship between drug efficacy and clinical outcomes in BC patients.

In conclusion, we demonstrated the genes and functional pathways involved in survival after pCR to NAC utilizing collected data from public BC cohorts with a bioinformatics approach. We found the genes involved in the relationship between pCR and survival utilizing Baseline and Delta analysis, some of which genes were also useful as prognostic factors in primary BC. Based on these reported results, we anticipate that further research can be conducted to establish a greater understanding of the relationship between the effect of NAC and survival.

Disclosure of conflict of interest

None.

Abbreviations

BC

breast cancer

NAC

neoadjuvant chemotherapy

DFS

disease free survival

OS

overall survival

HR

hormone receptor

pCR

pathological complete response

TN

triple negative

HER2

human epidermal growth factor receptor 2

GEO

Gene Expression Omnibus

AC

anthracycline based chemotherapy

OTC

optimal taxane based chemotherapy

EDB

epirubicin + docetaxel + bevacizumab

FC

fold change

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

TCGA

the cancer genome atlas

NCI

National Cancer Institute

ER

estrogen receptor

PR

progesterone receptor

EEF1

eukaryotic elongation factor 1

APP

Amyloid-β precursor protein

HSPB1

Heat Shock Protein Family B (Small) Member 1