Genomic profiling in locally advanced and inflammatory breast cancer and its link to DCE-MRI and overall survival

Abstract

Purpose: We have previously reported that dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) perfusion patterns obtained from locally advanced breast cancer (LABC) patients prior to neoadjuvant therapy predicted pathologic clinical response. Genomic analyses were also independently conducted on the same patient population. This retrospective study was performed to test two hypotheses: (1) gene expression profiles are associated with DCE-MRI perfusion patterns, and (2) association between long-term overall survival data and gene expression profiles can lead to the identification of novel predictive biomarkers. Methods: We utilised RNA microarray and DCE-MRI data from 47 LABC patients, including 13 inflammatory breast cancer (IBC) patients. Association between gene expression profile and DCE-MRI perfusion patterns (centrifugal and centripetal) was determined by Wilcoxon rank sum test. Association between gene expression level and survival was assessed using a Cox rank score test. Additional genomic analysis of the IBC subset was conducted, with a period of follow-up of up to 11 years. Associations between gene expression and overall survival were further assessed in The Cancer Genome Atlas Data Portal. Results: Differences in gene expression profiles were seen between centrifugal and centripetal perfusion patterns in the sulphotransferase family, cytosolic, 1 A, phenol-preferring, members 1 and 2 (SULT1A1, SULT1A2), poly (ADP-ribose) polymerase, member 6 (PARP6), and metastasis tumour antigen1 (MTA1). In the IBC subset our analyses demonstrated that differential expression of 45 genes was associated with long-term survival. Conclusions: Here we have demonstrated an association between DCE-MRI perfusion patterns and gene expression profiles. In addition we have reported on candidate prognostic biomarkers in IBC patients, with some of the genes being significantly associated with survival in IBC and LABC.

• Breast cancer
• DCE-MRI
• genomic
• imaging
• inflammatory breast cancer
• locally advanced breast cancer
• perfusion patterns
• survival

Introduction

Locally advanced breast cancer (LABC) and inflammatory breast cancer (IBC) have poor 5-year survival rates (55% and 33%, respectively) when compared to early stage breast cancer (80%) [1–3]. The purpose of neoadjuvant chemotherapy is to improve the treatment outcome of breast cancer patients by down-staging the tumour. Effective down-staging of the tumour before surgery will provide for more conservative surgery and potentially converting inoperable tumours to an operable state [4,5]. In breast cancer patients who receive neoadjuvant chemotherapy, 20–30% will achieve pathological clinical response (pCR). It has been shown that patients who achieve a pCR have a better prognosis. However, not all patients respond to neoadjuvant treatment, and thus there is a clear need for biomarkers that accurately predict response to treatment [5].

Between 2000 and 2004, 47 LABC patients were enrolled in an IRB-approved phase I/II clinical trial at Duke University Medical Center [6]. The patients enrolled in this clinical trial received a neoadjuvant combination of breast hyperthermia, liposomal doxorubicin (Evacet™), and paclitaxel. Hyperthermia was administered within 1 h of completing chemotherapy infusion. The hyperthermia prescription was to achieve 41–41.5 °C in greater than 90% of measured points for 60–70 min. The neoadjuvant treatment was followed by surgery, radiation therapy and adjuvant chemotherapy. Safety, tolerability and pathological response of this combination treatment have been reported previously [4,6].

RNA from pre-treatment ultrasound-guided core biopsies was assayed using Affymetrix GeneChip® Human Genome U133 Plus 2.0 arrays (Santa Clara, CA). Of the 47 patients enrolled in the clinical trial, a 20-patient subgroup in the phase II portion of the trial received the same paclitaxel and liposomal drug dose. Pre-treatment dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) studies were performed on this subset. The DCE-MRI enhancement features were analysed for association with treatment response [7]. DCE-MRI patterns can vary between tumours, based on factors such as the size of the extracellular space and vascularity of the tumour. Thus, the results can be indicative of angiogenesis, perfusion and/or hypoxic conditions in the tissue, all of which play vital roles in tumour development, progression and sensitivity to treatment [7–11]. In this patient dataset we observed two distinct patterns of image enhancement: (1) centripetal (CP), an inhomogeneous ring-type enhancement and (2) centrifugal (CF), characterised by more homogenous signal enhancement from centre to periphery (Figure 1) [7,9,10].

Figure 1. Wash-in parameter (WiP) maps generated from DCE-MRI in LABC patients. WiP maps are surrogates for perfusion/permeability [7]. The right image shows the CF enhancement pattern which is a uniform enhancement from centre to periphery. The left image is the CP enhancement pattern which is an inhomogeneous ring-type enhancement. The 10–250 scale represents the range of the WiP as previously described in detail [7].

In this current study we tested two hypotheses: (1) gene expression profiles are associated with the DCE-MRI perfusion patterns, and (2) association between long-term overall survival (OS) data and gene expression profiles can lead to identification of novel predictive biomarkers.

In order to test our hypotheses we analysed potential linkages between gene expression profiles and DCE-MRI parameters in LABC patients. In addition we analysed associations between long-term OS and gene expression profiles in the IBC subgroup. Our OS analysis results were further validated using The Cancer Genome Atlas (TCGA) Data Portal (https://tcga-data.nci.nih.gov/tcga/).

Materials and methods

Study population

Institutional review board (IRB) approval was obtained for this retrospective study (Protocol no. Pro00028021). The Duke Health System (DUHS) IRB for Clinical Investigations has determined the specific components above to be in compliance with all applicable Health Insurance Portability and Accountability Act (HIPAA) regulations. In addition, DUHS IRB has approved and provided us with a Waiver of Consent and HIPAA Authorization. We were not required to register this study on clinicaltrials.gov, because retrospective studies do not qualify as applicable clinical trial (ACT) studies.

The original selection of the study population has been described in detail elsewhere, including the patients’ treatment, clinical characteristics, and demographic data [4,6]. Briefly, between 2000 and 2004, 47 LABC patients (Stage IIB and III) were enrolled on an IRB-approved phase I/II clinical trial at Duke University Medical Center (Durham, NC, Protocol no. Pro1884-99-10). Patients were treated with neoadjuvant liposomal doxorubicin and paclitaxel in combination with breast hyperthermia, every 3 weeks for four cycles. The median patient age was 50 years (range 27–75 years). From the total of 47 patients enrolled in the clinical trial, pretreatment biopsies, collected under ultrasound guidance, were available from 37 patients, including 13 with a diagnosis of IBC. All 37 pretreatment biopsies were confirmed to have at least 60% invasive disease throughout the core sample before RNA harvesting. RNA was prepared, probe generated, and hybridised to the Affymetrix GeneChip^® Human Genome U133 Plus 2.0 array [4,6]. For the current analysis our study population had long-term surveillance of 11 years (median 8.4 years, range 0.3–11 years). All genomic data and samples were linked to clinical data.

DCE-MRI

The image analysis protocol has previously been described in detail for characterisation of the CP and CF perfusion patterns [7,12]. A 20-patient subgroup in the phase II portion of the trial was chosen for the DCE-MRI analysis from the 47 patients. These 20 patients were chosen because their breast DCE-MRI data was attained using the same protocol, and because all received the same dosage of neoadjuvant chemotherapy and hyperthermia. The DCE-MRI was done over 30 min following bolus injection of gadopentetate-based contrast agent (gadopentetate dimeglumine (Gd-DTPA). The contrast agent was administered through an indwelling intravenous catheter with a 2 cm³/s rate [7].

Data analysis and statistical considerations

The genomic data discussed in this paper has been deposited in the US National Center for Biotechnology Information (NCBI)’s Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/http://www.ncbi.nlm.nih.gov/geo/) [13]. The deposited data are accessible though the GEO series accession number GSE52322, ‘Novel linkages between DCE-MRI and genomic profiling in locally advanced and inflammatory breast cancer’ (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE52322).

Analysis of association between gene expression profiles and DCE-MRI perfusion patterns

LABC samples were assayed using the Affymetrix GeneChip® Human Genome U133 Plus 2.0 array, which interrogates mRNA expression, based on 54,675 probe sets. All statistical analyses were conducted using the R statistical environment [14,15]. Pre-processing was conducted on the basis of all available samples from the LABC study deemed to be of good quality, regardless of the availability of the imaging phenotype [16]. The arrays were pre-processed using the RMA algorithm [17]. Ordination methods (e.g. multi-dimensional scaling) were used to identify experimental artefacts, such as batch effects [18]. From the 20-patient subgroup who had the DCE-MRI, 15 had both imaging and genomic data available. Association between the gene expression profile and the DCE-MRI CP versus CF morphology was determined by the Wilcoxon rank sum test to prioritise the features (probe sets) according to statistical significance [16,19].

Analysis of gene expression profiles and patient survival

Association between feature expression level and survival was assessed using a Cox rank score test [16]. Corresponding effect size was quantitatively assessed using a hazard ratio assuming proportional hazards. Conditional inference trees were used to find optimal cut points [20]. As the feature selection process was considered to be hypothesis generating, the p values reported have not been adjusted for multiple testing. The associations between expression level and OS for the implicated genes were further assessed in a published cohort from TCGA (https://tcga-data.nci.nih.gov/tcga/) using the breast invasive carcinoma project dataset [21]. The TCGA mRNA data were retrieved from the Cancer Genomic Data Server (CGDS) through the Computational Biology Center Portal (http://www.cbioportal.org/). The Cancer Genomics Data Server R (cdgsr) extension package (cran.rproject.org/web/packages/cgdsr/) was used to execute the retrieval.

Results

Association between gene expression profiles and DCE-MRI perfusion patterns

Of the 15 patients with both genomic and imaging information, 10 patients had the CP perfusion pattern while five had the CF pattern (Figure 1). Following data analysis on RNA microarray results we identified genes that were differentially expressed in the CP versus CF pattern (Figure 2). The CP pattern was found to have higher expression of genes involved in a component of the sulphur metabolism pathway. In particular, genes included those in the sulphotransferase family: (1) sulphotransferase family, cytosolic, 1A, phenol-preferring, member 1 (SULT1A1), and (2) sulphotransferase family, cytosolic, 1A, phenol-preferring, member 2 (SULT1A2) (Figure 3A and B). In addition, CP had similarly higher RNA expression levels of poly (ADP-ribose) polymerase, member 6 (PARP6) (Figure 3C). The metastasis tumour antigen1 (MTA1) was also differentially expressed in the CP versus CF morphology. However, in contrast to the previous three genes, MTA1 was overexpressed in CF when compared to CP (Figure 3D). Even though MTA functions as an oestrogen receptor 1 (ESR1) co-repressor, we did not find a positive association between the CP and CF enhancement patterns and levels of ESR1 expression in our dataset. Only one of the nine probe sets representing ESR1 demonstrated a positive association (ESR1:205225_at, p = 0.028), with lower ESR1 expression in the CF cases compared to CP.

Figure 2. Heat map cluster plot of Affymetrix microarray data in LABC samples. This heat map plot records post-normalisation between-probes distances measured by their absolute median difference. The left diagram represents all the top probe sets associated with144 genes, which were differentially expressed in CF versus CP (p < 0.008); 80% of the probes had higher expression levels in the CP enhancement pattern and only 20% had higher expression levels in CF. The right diagram is the enlarged heat map for MTA1, PARP6, SULT1A1 and SULT1A2. In the table, test statistic (teststat) is the smaller of two sums of ranks of given sign. Negative test statistics represent higher expression level of a gene in the CF pattern, and positive test statistics represent higher expression level in the CP perfusion pattern.

Figure 3. Differential expression levels in the CF enhancement pattern compared to the CP. (A and B) Our analysis demonstrates an overexpression of SULTs (SULT1A1 and SULT1A2) in the CP enhancement compare to CF. (C) The CP enhancement pattern is associated with higher expression levels of PARP6. (D) MTA1 shows higher expression levels in the CF enhancement pattern compared to the CP.

Association between long-term overall survival data and gene expression profiles

Our follow-up time for the patients in this trial was 11 years and the median follow-up time was 8.4 years. We analysed the association between the genomic expression data and OS in the subgroup of 13 IBC patients. We limited the sample analysis to the IBC cases because they provided the largest subgroup out of the total 47 that had a uniform tumour type. In our survival analysis, overexpression of 45 genes was found to be associated with OS in IBC patients (p < 0.001). The genes identified were involved in the sulphur metabolism pathway and the focal adhesion kinase (FAK) pathway (Table 1 and supplementary data). Overexpression of two genes involved in the sulphur metabolism pathway was associated with shorter OS, these included carbohydrate (chondroitin 4) sulphotransferase 11 (CHST11, p = 7.88E-05, hazard ratio (HR) = 2.4) and carbohydrate (chondroitin 6) sulphotransferase 3 (CHST3, p = 5.84E-04, HR = 1.5). In addition, overexpression of four members of the FAK pathway were significantly associated with shorter OS in IBC patients. These included collagen type IV alpha 4 (COL4A4, p = 7.82E-04, HR = 1.6), protein kinase C alpha (PRKCA, p = 6.56E-04, HR = 1.6), talin 2 (TLN2, p = 5.25E-04, HR = 1.6) and tenascin C (TNC p = 9.01E-04, HR = 1.7). In our OS analysis we also noticed that our list included two of the genes associated with tumour invasiveness and treatment resistance: 1) hypoxia up-regulated protein 1 (HYOU1, p = 2.00E-04, HR = 1.7), and 2) apoptosis caspase activation inhibitor (AVEN, p = 1.03E-03, HR = 1.5).

Table 1. Association between gene expression patterns and overall survival in IBC patients.

Probe	Symbol	p Value	HR	Sctest p value	Cox HR
226914_at	ARPC5L	4.00E-04	1.8	4.42E-04	1.8
219366_at	AVEN	1.90E-03	1.5	1.03E-03	1.5
236331_at	CDKL2	1.50E-03	1.5	1.20E-03	1.5
226372_at	CHST11	0.00E + 00	2.4	7.88E-05	2.4
32094_at	CHST3	7.00E-04	1.5	5.84E-04	1.5
229779_at	COL4A4	9.00E-04	1.6	7.82E-04	1.6
213274_s_at	CTSB	6.00E-04	1.6	4.63E-04	1.6
200825_s_at	HYOU1	1.00E-04	1.7	1.99E-04	1.7
213093_at	PRKCA	1.00E-03	1.6	6.56E-04	1.6
209325_s_at	RGS16	1.00E-03	1.6	6.26E-04	1.6
212701_at	TLN2	5.00E-04	1.6	5.25E-04	1.6
201645_at	TNC	1.50E-03	1.7	9.01E-04	1.7

The complete list of the 45 genes associated with OS is presented in the supplementary data.

The HR and p value are computed based on the permutation resampling inference test for survival data. The ‘sctest p value’ and ‘cox HR’ columns are based on the log rank test.

We used the publicly available TCGA database for external confirmation of our survival analysis results. Currently TCGA does not provide access to imaging data to be used for CP versus CF classification; therefore we were not able to use the TCGA database for confirmation of linking DCE-MRI and gene expression profiles. We analysed the survival association between the expression level of the 45 genes of interest in TCGA’s LABC group with a sample size of N = 462 (we did not have the data available from TCGA to specifically select the IBC cases from the main LABC group). In TCGA’s LABC cohort, differential expression of four genes from our gene list exhibited the same significant concordance and association with overall survival as in IBC patients. The genes included 1) actin-related protein 2/3 complex, subunit 5-like (ARPC5L, Duke p = 4.42E-04, HR = 1.8, TCGA p = 1.20E-03, HR = 2.3), 2) regulator of G protein signalling 16 (RGS16, Duke p = 6.26E-04, HR = 1.6, TCGA p = 1.58E-03, HR = 2.8), 3) cyclin-dependent kinase-like 2 (CDKL2, Duke p = 1.50E-03, HR = 1.5, TCGA p = 9.66E-03, HR = 2.0), 4) cysteine protease cathepsin B (CTSB, Duke p = 4.63E-04, HR = 1.6, TCGA p = 1.95E-02 HR = 2.1) (Figures 4 and 5, Table 1).

Figure 4. Gene expression patterns and overall survival in IBC patients. ARPC5L and CDKL2 expression levels are significantly associated with long-term OS in IBC patients. Survival time represents years. Three of the patients from the Duke clinical trial who are alive are represented with a triangle symbol in the graph.

Figure 5. External confirmation for gene expression patterns and overall survival in LABC using TCGA database. In TCGA LABC cohort differential expression of four genes exhibited the same significant concordance and association with overall survival as in the IBC patients. The black lines in the graph represent high expression levels and the red lines represent low expression levels of the genes. The “sctest p-value” and “HR” (which is a cox HR) columns are based on the log rank test.

Discussion

Association between gene expression profiles and DCE-MRI perfusion patterns

Our group previously reported that DCE-MRI parameters are linked to treatment outcome in canine and human cancers [7,22]. We reported that DCE-MRI parameters were predictive of overall and metastasis-free survival in 37 dogs with soft tissue sarcomas that were treated with thermoradiotherapy. This was the first study to report a linkage between DCE-MRI parameters in sarcomas and long-term outcome [22]. In humans we previously reported that DCE-MRI perfusion patterns obtained from LABC patients prior to neoadjuvant and hyperthermia therapy were predictive of pCR [7].

Here we have analysed links in gene expression profiles and DCE-MRI parameters in LABC patients. Our results indicate that the two DCE-MRI enhancement patterns of CF and CP have distinct gene expression profiles. Out of the top 200 probe sets associated with 144 genes, which were differentially expressed in CF versus CP (p < 0.008), 80% of the probes had higher expression levels in the CP enhancement pattern and only 20% had higher expression levels in CF.

CP was found to have higher expression in genes involved in a component of sulphur metabolism pathway. Cytosolic sulphotransferase (SULTs) function as catalysts for biotransformation of steroid hormones, drugs and environmental toxins. Table 2 provides some background information and references for SULTs. Through sulphoconjugation, SULTs play a key role in the metabolism and inactivation of oestrogen. It is therefore theorised that overexpression of the SULTs genes in CP morphologies could decrease the levels of unconjugated oestrogens (the active form of oestrogen) in breast cancer tissue, and consequently block the proliferative effect of the hormone. In addition, oestrogen directly affects tumour microenvironment. Oestrogen has a direct effect on endothelial cells resulting in endothelial cell migration, proliferation and modulating angiogenesis [23]. This suggests that patients with the CP enhancement pattern with higher expressions of SULTs could have lower levels of active oestrogen. This observation potentially draws an important connection between genomics, imaging, and therapy, and thus merits further investigation.

Table 2. Overview of some of the biological processes associated with differential expression of our genes of interest.

Gene symbol	Gene function
A. Genes differentially expressed in the DCE-MRI enhancement patterns (centripetal vs. centrifugal)
MTA1	MTA1 is a potential predictor for breast tumour aggressiveness, involved in angiogenesis and is a predictor of early disease relapse [44,45]
	Can repress the ESR1 trans-activation function [46,47]
	Can inactivate the tumour suppressor TP53 [48,49]
	Can de-acetylate and stabilise the hypoxia-inducible factor-1 alpha (HIF-1α) which is a key regulator of angiogenic factors [50–52]
	MTA proteins, including MTA1, have been reported as a possible set of ‘master co-regulatory molecules’ involved in the carcinogenesis and progression of various malignancies [47]
PARP6	PARP superfamily is involved in the regulation of angiogenesis, differentiation, proliferation, cell death, tumour transformation and DNA damage recovery [53,54]
	PARP inhibitors can decrease angiogenesis since inhibition of PARP will affect VEGF-induced proliferation and migration [55,56]
SULT1A1, SULT1A2	Alterations in the expression of SULT enzymes and their genetic polymorphisms have been reported in lung, prostate, and breast cancer [57,58]
	Several members of the SULT family are responsible for the sulphation of oestrogens, therapeutic estrogenic compounds, and hormone replacement therapy agents; SULTs play a key role in the metabolism and inactivation of oestrogen [59]
B. Genes differentially expressed in the survival analysis for inflammatory breast cancer patients*
ARPC5L	ARPC5 family has been implicated in invasion and metastasis in human head and neck squamous cell carcinoma [38]
AVEN	Considered a regulator of apoptosis. Resistance to apoptosis is associated with carcinogenesis, tumour progression, and treatment resistance. Overexpression associated with poor prognosis in tumours such as childhood acute lymphoblastic leukaemia and acute myeloid leukaemia [60–62]
CDKL2	CDKL2 is a member of a large family of CDC2-related serine/threonine protein kinase. CDC2 overexpression in breast tumours has been associated with infiltrative tumour border pattern and histological high grade breast carcinomas [40]
CHST11, CHST3	CHST11 and CHST3 are involved in the sulphur metabolism pathway. CHST11 mediates 4-O sulphation of chondroitin, CHST3 mediates 4-O sulphation of chondroitin [63]
	Chondroitin sulphates are involved in breast tumour aggressiveness and metastasis and CHST11 and CHST3 overexpression have been reported to be associated with more aggressive breast tumours [35]
COL4A4	COL4A4 protein belongs to the type IV collagen family. Overexpression of type IV collagen family members has been associated with tumour size, higher grade, metastasis and invasion in malignancies such as breast, ovarian, cervical, colorectal and gastric cancers [64–66]
CTSB	Overexpression has been reported to be associated with tumour invasiveness and poor patient prognosis in melanoma, prostate and breast cancers [67]
	CTSB overexpression in inflammatory breast cancer has been suggested as a prognosis biomarker associated with poor survival and increased number of positive metastatic lymph nodes [41]
HYOU1	Associated with tumour invasiveness and resistance to therapy. Suppression also results in increased apoptosis [68–70]
PRKCA	A member of the protein kinase C family which can play an important role in cell signalling pathways by phosphorylating protein targets. High expression levels of PRKCA have been reported as a potential marker for poor prognosis and associated with breast cancer progression [71]
RGS16	A member of the RGS superfamily, a class of intracellular signalling regulators. Considered as therapeutic targets, including in breast cancer [42,43]
TLN2	In the FAK pathway, FAK and TLN are two key players in matrix-cell adhesion assembly required for cell migration which can result in tumour metastasis and progression [72,73]
TNC	Frequently up-regulated in breast cancer and promotes tumour invasion. Associated with promoting cell migration, angiogenesis and can act as a cell survival factor [39]

The complete list of the 45 genes associated with OS is presented in the supplementary data.

Here we have reported that the CP morphology is also associated with higher expression levels of PARP6 (Table 2). Recently PARP inhibitors have been used in clinical trials and suggested as a favourable target in breast cancer therapy [24–27]. Our findings suggest that an enhancement pattern of CP might benefit more from PARP inhibitor treatment. Recent studies in animal models of endometrioid carcinoma, the major subtype of human endometrial cancers, have demonstrated that high oestrogen levels interfered with the PARP inhibitor AZD2281 (olaparib). In lower serum oestrogen levels, the PARP inhibitors resulted in more reduction in tumour size [28]. In addition, as mentioned previously the higher levels of SULTs in the CP tumours could result in inactivation of oestrogen. The inactivation of estrogen potentially could also make tumours with CP enhancement a better target for PARP inhibitor treatment, and affect the patients’ pCR and prognosis.

Our imaging and gene expression analysis also demonstrated that MTA1 was overexpressed in CF tumours compare to CP. The metastasis tumour antigen (MTA) family of proteins (MTA1, MTA2, and MTA3) are transcriptional co-repressors. MTA1 is overexpressed in a wide range of human cancers, such as gastrointestinal, ovarian, prostate, and breast cancers [29–33]. MTA1 is considered a potential predictor for breast tumour aggressiveness (Table 2). We noted that MTA1 was overexpressed in CF tumours, which were characterised to have a more homogenised vascular network, and thus a better potential for drug delivery. The improved drug delivery characteristics could potentially be counteracted by increased aggressiveness in tumours with CF phenotypes which at the end could affect pCR rate.

Association between overall survival data and gene expression profiles

In the second part of our study we focused on the association between gene expression profiles and long-term survival of breast cancer patients. The long follow-up time of 11 years and availability of clinical data provided the unique opportunity to analyse the association between gene expression profiles and long-term OS. Limiting our analysis to the IBC cases provided the largest subgroup out of the total 47 patients with a more uniform tumour type. IBC is a rare subtype of LABC, and the most aggressive form of breast cancer. IBC survival rates are relatively poor, averaging 35% to 40%. IBC has no known aetiology and has no proven effective therapy. In spite of aggressive multimodality therapies over the past three decades, there has been relatively little improvement in the overall survival rate of IBC [2,3,34].

Here we were able identify candidate gene expression profiles that defined signalling pathways associated with survival in IBC patients. For the Duke IBC patient subgroup, overexpression of 45 genes was associated with poor survival; this included genes involved in the sulphur metabolism pathway and the FAK pathway. The expression patterns reflect themes common to the aggressive biology of breast cancer (e.g. metastasis, tumour invasiveness, cell migration) (Table 2).

Our results demonstrated that overexpression of CHST11 and CHST3, genes involved in the sulphur metabolism pathway, was associated with shorter OS in IBC patients. Chondroitin sulphates are involved in breast tumour aggressiveness and metastasis and CHST11 and CHST3 overexpression have been reported to be associated with more aggressive breast tumours [35]. This is in concordance with our findings since the patients with higher expression levels of CHST11 and CHST3, had shorter OS (HR of 2.4 and 1.5 respectively).

Activation of FAK signalling pathway in tumour cells, including breast tumours, can affect cellular processes such as metastasis, cell spreading, adhesion, proliferation, angiogenesis, progression and tumour survival. The FAK pathway can be a potential cancer therapeutic target to inhibit tumour invasion and progression [36,37]. Our results demonstrated that in IBC patients, overexpression of four FAK pathway genes: (1) COL4A4, (2) PRKCA, (3) TLN2, and (4) TNC were significantly associated with shorter OS.

When we analysed the association between survival and the expression level of the 45 genes of interest in the TCGA’s LABC cases (N = 462), four genes had common levels of significance and concordant direction for LABC patients. Those four common genes between IBC and LABC are: (1) ARPC5L, (2) RGS16, (3) CDKL2, and (4) CTSB.

The TCGA analysis provides an external confirmation for the importance of these four genes in LABC. These four genes have been implicated in other studies for their roles in tumour invasion and metastasis (ARPC5L and TNC) [38,39], infiltrative tumour border pattern and histologic high grade breast carcinomas (CDKL2) [40], as a potential prognosis biomarker associated with poor survival and increased number of positive metastatic lymph nodes in IBC patients (CTSB) [41], and as potential therapeutic targets in cancer therapy (RGS16) [42,43].

Conclusion

In conclusion, although these findings are preliminary we feel these results are especially important and merit further investigation and validation in independent datasets with a larger sample size and more homogenous tumour types. The major limitation of our study is the small sample size and lack of access to tumour samples from the patients enrolled in the original phase I/II clinical trial to perform additional validation experiments such as qPCR and western blot analysis. However, our preliminary results have shown unique links to changes in DCE-MRI and gene expression profiles. Here we reported on the differential genes expression profile between the CP and CF enhancement patterns. In addition we have reported on candidate prognostic biomarkers in IBC patients, with some of the genes being significantly associated with OS in IBC and LABC. While not all these genes have previously been reported to be associated with prognosis in IBC or LABC, they are consistent with other studies which associate these genes and gene products with aggressiveness and poor outcomes in other cancers.

Acknowledgement

We would like to acknowledge Salvatore Mungal, MS, from the Duke Cancer Institute Bioinformatics Shared Resource, for his help in submitting our data to the US National Center for Biotechnology Information’s Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/http://www.ncbi.nlm.nih.gov/geo/).

Declaration of interest

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.