Differential expression of immune related genes in high-grade ovarian serous carcinoma

Abstract

Objective.

To identify novel immunologic targets and biomarkers associated with overall survival (OS) in high-grade serous ovarian cancer (HGSC).

Methods.

In this retrospective study, microarray data from 51 HGSC specimens were analyzed (Affymetrix HG-U133A). A panel of 183 immune/inflammatory response related genes linked to 279 probe sets was constructed a priori and screened. Associations between gene expression and OS were assessed using logrank tests. Multiple testing was addressed within the False Discovery Rate (FDR) framework. For external validation, TCGA Ovarian dataset and five GSE publicly available HGSC datasets were evaluated.

Results.

In Duke data, 110 probe sets linked to 83 immunologic/inflammatory-related genes were differentially expressed in tumors from long versus short-term HGSC survivors (adjusted p < 0.05). In TCGA concordant with the results from the Duke discovery cohort, high expression of one probe (IL6R) demonstrated a consistent significance and concordant association with higher expression in long-term HGSC survivors (Duke q-value = 022) and improved OS in the TCGA dataset (p-value = 0.015, HR = 0.8). Thirteen genes in GSE14764 (N = 4) and GSE26712 (N = 9) datasets had significant p-values and consistent concordant with Duke Data. Despite the significant associations of gene expression and OS in the individual GSE datasets, in the GSE meta-analysis no genes were consistently concordant and significantly associated with survival.

Conclusions.

Evaluation of IL6R expression may be warranted based on higher expression in long-term survivors and association with improved survival in advanced HGSC. The other candidate genes may also be of worthy of further exploration to enhance immuno-oncology drug discovery.

1. Introduction

Ovarian cancer is the most lethal gynecologic malignancy. For 2019, the American Cancer Society estimates that in the United States an estimated 22,530 women will be diagnosed ovarian cancer, and an estimated 13,980 women will die from ovarian cancer. Worldwide the annual death due to ovarian cancer is estimated at 152,000 [1-3]. There is a significant clinical need for improved diagnosis and treatment of epithelial cancers of the ovary, fallopian tube, and peritoneum (collectively referred to as epithelial ovarian cancers (EOCs) in this manuscript). The majority of EOCs are high-grade serous cancers (HGSCs), an aggressive cancer that is associated with worse clinical outcome compared to other histologic subtypes [4,5]. Initial treatment involves cytoreductive surgery and platinum based chemotherapy. However, 75–80% of these patients recur and develop resistance to platinum based chemotherapy regimens. Ultimately, the majority of these patients will die of their disease with a median survival of 3–4 years [6]. Therefore, it is critical to identify novel therapeutic targets and regimens to improve survival for women with ovarian cancer.

Dramatic advances have been made in immunotherapy treatment for various types of cancers. In cancer immunotherapy the goal is increase the patienťs own immune function to destroy malignant cells [7]. Immunotherapy includes a number of different approaches, such as tumor directed vaccines, cytokines, cellular based therapies, and immune checkpoint inhibitors. Multiple novel antibodies including the antiCytotoxic T-lymphocyte antigen 4 (CTLA-4) antibody, ipilimumab; the anti PD-1 receptor antibodies, nivolumab and pembrolizumab; and the anti PD-L1 antibodies, avelumab, atezolizumab, and durvalumab, have been developed and studied in ovarian cancer [7-11]. The response rates in ovarian cancer, range from 10 to 15% and immunotherapies have not yielded the impressive anti-tumor effects seen in other malignancies [7,12]. However, some of the ovarian cancer responders have had long duration of disease control, suggesting at the promise of immune therapy in ovarian cancer as well as the importance of identifying patients most likely to benefit [7].

Identification of biomarkers for predicting effective therapies in individual patients will be important to maximize immunotherapy benefits and minimize toxicity. Therefore, our objective was to identify candidate genes that were differentially expressed in women with HGSC in order to establish novel immunologic targets and companion biomarkers for further exploration. We hypothesized that our results would both identify novel targets and prognostic/predictive immune biomarkers for the purpose of developing and directing future immunotherapies in HGSC.

2. Methods

2.1. Study population

RNA microarray data were retrospectively utilized from patients treated at Duke University Medical Center between 1988 and 2001 under a Duke Institutional Review Board (IRB) approved protocol (N = 51) [6]. The 51 patients had chemotherapy-naïve advanced HGSC (stage III (A, B, or C), or IV, and grade 2 or 3) [6,13]. As previously described [5,6,13] the original study used an extreme phenotype design, deliberately genetically profiling tumors from women who had either long or short-term overall survival (OS). Long-term survival was defined as OS >7 years vs. short-term survival defined as OS <3 years [5,6,13].

The 51 patients had chemotherapy-naïve advanced HGSC (stage III (A, B, or C), or IV, and grade 2 or 3) [6,13]. The patients underwent primary surgical debulking (Table 1). Of the 23 long-term survivors 10 (43%) patients were suboptimally debulked and 13 (57%) were optimally debulked (defined as <1 cm gross residual disease). Of the short term survivors 18 (64%) patients were suboptimally debulked and 10 (36%) were optimally debulked. Fifty-one were treated with platin-based therapy. Patients with long-term survival received the following: intravenous (IV) platinum/paclitaxel (n = 8); IV platinum/cyclophosphamide (n = 10) [2 of these patients also received intraperitoneal P32; and one, IV interferon); IV platinum/paclitaxel/etoposide (n = 1); intraperitoneal (IP) cisplatin/ IV cyclophosphamide (n = 1); IV carboplatin/ IP cisplatin/ IV paclitaxel n = 3). Patients with short-term survival received the following: IV platinum/paclitaxel (n = 14); IV platinum/cyclophosphamide (n = 11); IP cisplatin/ IV cyclophosphamide (n = 1); IV platin therapy (n = 2).

Table 1.

Patient characteristics.

	Overall survival statusa
	Short-term survivors OS < 3 years (n = 28)	Long-term survivors OS > 7 years (n = 23)
Median age (range) - year	60.0 (56.3–63.7)	59.1 (54.3–64.0)
	N (%)	N (%)
Stage
III	23 (82%)	23 (100%)
IV	5 (18%)	0
Race
White	18 (64%)	21 (91%)
Black	7 (25%)	1 (4%)
Other	3 (11%)	1 (4%)
Surgical debulking
Suboptimal	18 (64%)	10 (43%)
Optimalb	10 (36%)	13 (57%)
Chemotherapyc
IV platinumd/paclitaxel	14 (50%)	8 (35%)
IV platinumd/cyclophosphamide	11 (39%)	10 (43%)e
IV carboplatin/IP cisplatin/IV	–	3 (13%)
Paclitaxel
IV platind therapy	2 (7%)	–
IV	–	1 (4%)
carboplatinum/paclitaxel/etoposide
IP cisplatin/IV cyclophosphamide	1 (4%)	1 (4%)

^aPercentages may not total up to 100 due to rounding.

^bOptimal debulking defined as <1 cm of gross residual disease.

^cIntravenous (IV); intraperitoneal (IP).

^dPlatinum refers to either cisplatin or carboplatin.

^eTwo of these patients also received intraperitoneal P32; and one received, intravenous interferon.

At the time of our current retrospective data analyses, the study population had long-term surveillance of >10 years.

2.2. Microarray analysis

The microarray data retrospectively used in our analysis were from RNA from fresh frozen HGSC tumors. At initial surgery, the tumor specimens were collected and snap-frozen. The RNA was harvested from pretreatment biopsies with at least 60% invasive disease throughout the core sample. RNA was prepared, quality assured using an Agilent 2100 Bioanalyzer, probes generated, and used for hybridization to Affymetrix HG-U133A GeneChip arrays as described previously in detail [5,6].

2.3. Statistical analysis and validation of the association between IT genes expression and overall survival (OS) in publically available TCGA and GSE databases

We developed a list of 183 immune and inflammatory response related genes (please see Supplemental Table 1). The list was generated based on: literature review of known genes in the immune-related pathways of interest, and potential therapeutic targets for immune therapy (e.g. Cytokines, PD1, and PDL1 inhibitors) [14-20].

Not all genes of interest were represented on the Affymetrix HG-U133A GeneChip. For example PD1 (also known as PDCD1) was represented on the Affymetrix HG-U133A GeneChip but the PD-L1 (also known as CD274) probe was not represented. A panel of 279 probe sets linked to 183 unique immune response-related genes was screened. The link of each of the 279 probe sets to HUGO Gene Nomenclature Committee (HGNC) gene symbols was mapped using annotation data from public databases through a Bioconductor package biomaRt v2.31.0 [21-23]. The Robust Multi-chip Averaging (RMA) algorithm was used to pre-process the raw Affymetrix HG-U133A array data [24].

Association between gene expression of each probe set and OS was assessed using a logrank test [25]. Corresponding effect size was quantitatively assessed using a hazard ratio (HR) assuming proportional hazards for the TCGA-OV and GEO datasets. Conditional inference trees were used to find optimal cut-points [26]. As we have previously described [13], the association between gene expression and survival outcomes (long versus short) was presented as box plots for the Duke data and Kaplan–Meier (KM) survival curves for the TCGA and GEO datasets. The Duke survival data is presented as box plots instead of Kaplan–Meier (KM) plots due to the extreme phenotype design [13]. The 51 samples from the Duke study were pre-selected using a biased sampling approach by selecting patients who had an OS of ≤3 years or ≥7 years (short vs. long overall survival). Thus, the application of the KM method to estimate the survival distribution conditional on the gene expression value was not deemed appropriate. HRs are provided to inform direction of effect only and are not relevant for the Duke analysis given the study design [13,27,28]. Multiple comparisons were addressed within a framework of control of False Discovery Rate (FDR) framework. FDR adjusted p-values (Q-values) were calculated using the method by Storey [25,29,30]. The results presented for the external databases (conformation cohorts) are not adjusted for multiple testing. All analyses were performed using the R Statistical Environment [R] and extension packages from CRAN and the Bioconductor project [25,31].

2.4. Confirmation cohorts

For external confirmation, we used publicly available HGSC data from The Cancer Genome Atlas (TCGA-OV, N = 229) and datasets from the Gene Expression Omnibus (GEO) repository. TCGA data for ovarian cancer were retrieved from the Cancer Genomic Data Server (CGDS) through the Computational Biology Center Portal (cBio): http://www.cbioportal.org/. The cdgsr extension package v1.2.6 was used to execute the retrieval (https://CRAN.R-project.org/package=cgdsr) [32,33]. TCGA results presented here are based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga.

GSE mRNA data were obtained from 11 publicly available GSE datasets as described in the 2012 Bentink et al. paper [34]. Of the 11 datasets there were five (GSE26712 (N = 153) [35], GSE18520 (N = 53) [36], GSE14764 (N = 63) [37], GSE 19161 (N = 61) [38], and GSE 17260 [39]) that had information on HGSC patients and the numbers of HGSC cases matched to the Bentink et al. paper [34,40]. However, the GSE17260 was not used because none of the 110 probes could be found in the set.

3. Results

3.1. Duke database results

The Duke data demonstrates an association between OS and differential expression of a panel of 83 immune/inflammatory response genes (linked to 110 Affymetrix probe sets, adjusted p-value (q) <0.05). The complete list of the 83 gene is provided in the Supplemental Table 2. The association between gene expression and survival outcomes (long versus short survival) are presented as box plots (Figs. 1 and 2). For instance, interleukin 6 receptor (IL6R), was differentially expressed in women with short vs. long-term survival (Fig. 1 and Table 2). Women with longer survival had higher expression levels of IL6R compared to those with short-term survival (The Duke adjusted p-value (q) = 0.022) and denotes the association between gene expression and survival outcomes. The 83 genes were further evaluated in the TCGA and GSE databases.

Fig. 1.

Association between IL6R gene expression level and overall survival (OS) in the Duke database and TCGA database. The Duke data revealed that HGSC patients with long survival were more likely to have tumors that expressed higher levels of IL6R (OS ≥ 7 years) compared to those women with short survival (OS ≤ 3 years. (IL6R Duke q-value = 0.02). In the TCGA database, higher IL6R expression levels were associated with longer overall survival (IL6R TCGA p-value = 0.015; HR = 0.8). The distribution of the gene expression, y-axis, was estimated conditional on surviving less than three years and compared to that of the conditional distribution of surviving at least seven years. The Duke survival data is presented as a boxplot. Boxes represent 25th (Q1) and 75th (Q3) percentiles; Horizontal lines indicate the medians; Upper whiskers indicate min (max(x), Q3 + 1.5 * IQR); Lower whiskers indicate max (min(x), Q1−1.5 * IQR). Points indicate any observations outside the whiskers.

Fig. 2.

Association between gene expression levels and overall survival (OS) in Duke data and GSE database. The Duke data revealed that HGSC patients with long survival were more likely to have tumors that expressed higher levels of CXCL10 (OS ≥ 7 years) compared to those women with short survival (OS ≤ 3 years). These findings were externally confirmed in the GSE database (GSE 26712 p-value = 0.044, HR = 0.9, Duke q-value = 0.03). In contrast HGSC patients with lower expression levels of BCL2 has long survival compare to patients with higher BCL2 expression (GSE 26712 p-value = 0.002, HR = 1.9, Duke q-value = 0.02). In the meta-analysis of the GSE datasets no genes were consistently concordant.

Table 2.

Differentially expressed genes common in the Duke in TCGA and GSE databases.

Gene symbol	Probe	Duke extreme phenotype design cohort		External database
		Q-valuea	Gene expression in short-term survivors	Name of external database	p-Value	HR	95% CI
IL6R	205945_at	0.022	Low	TCGA	0.015	0.8	0.4, 0.9
CXCR1	207094_at	0.013	High	GSE14764	0.002	1.8	1.2, 2.8
IL4	207538_at	0.035	High	GSE14764	0.017	1.5	1.0, 2.3
IL9R	217212_s_at	0.027	High	GSE14764	0.023	2.2	1.1, 4.5
MBL2	207256_at	0.017	High	GSE14764	0.027	1.5	1.0, 2.0
BCL2	203684_s_at	0.017	High	GSE26712	0.002	1.9	1.2, 3.0
CCL15	210390_s_at	0.039	High	GSE26712	0.019	2.4	1.2, 5.0
CCL16	207354_at	0.039	High	GSE26712	0.033	1.4	1.0, 1.9
CCL21	204606_at	0.040	High	GSE26712	0.010	1.5	1.1, 2.02
CXCL10	204533_at	0.027	Low	GSE26712	0.044	0.9	0.7, 1.0
IFNA1	208344_x_at	0.039	High	GSE26712	0.023	1.9	1.2, 3.1
IL1A	210118_s_at	0.023	High	GSE26712	0.035	1.8	1.0, 3.21
MAPK8	210671_x_at	0.030	High	GSE26712	0.022	1.7	1.1, 2.5
	210477_x_at	0.008	High	GSE26712	0.027	1.5	1.0, 2.1
TLR8	220832_at	0.008	High	GSE26712	0.035	2.0	1.1, 3.4

^aQ-value: denotes adjusted p-value.

3.2. TCGA database results

In the TCGA database [43,44] of the 83 genes, one probe, IL6R, exhibited level of significance and consistent direction of association for OS. Specifically, high IL6R expression was associated with improved OS in the TCGA dataset (p-value = 0.015, HR = 0.80) (Table 1 and Fig. 1). In addition, after adjusting for age, race, grade, and residual disease (N = 200) in the TCGA dataset, the association between IL6R expression and OS remained statistically significant (p value = 0.023).

3.3. GSE analysis results

There were five datasets (GSE26712 [35], GSE18520 [36], GSE14764 [37], GSE 17260 [39], and GSE 19161 [38]) that had information on HGSC patients and the numbers of HGSC cases matched to the Bentink et al. paper [34,40]. Of the five GSE datasets, the GSE17260 dataset was not utilized in this analysis because none of our 110 probes of interest could be found. In GSE18520 dataset, no probe shows the level of significance and consistent direction of the association with OS. In the remaining GSE datasets, there were 13 genes (14 probe sets) that exhibited level of significance (p-value < 0.05) and concordant direction for their association with OS (Table 1, Fig. 2). Of these 13 genes, high expression of 12 genes was more likely associated with short-term survival compared to those with long survival (OS of ≤3 years or ≥7 years) in the Duke dataset. Moreover, high gene expression of these genes was also associated with worse OS in the GSE data. The 12 genes whose higher expression levels were associated with shorter survival included the following: C-X-C motif chemokine receptor 1 (CXCR1, Duke q-value = 0.013; GSE p-value = 0.002, HR = 1.8), interleukin 4 (IL4, GSE p-value = 0.017, HR = 1.5, Duke q-value = 0.035), interleukin 9 receptor (IL9R, GSE p = 0.023, HR = 2.2, Duke q-value = 0.027), mannan-binding lectin (MBL)-2 (MBL2, GSE p-value = 0.027, HR = 1.5, Duke q-value = 0.017), BCL2, apoptosis regulator (BCL2, GSE p-value = 0.002, HR = 1.9, Duke q-value = 0.017), C—C motif chemokine ligand 15 (CCL15, GSE p-value = 0.019, HR = 2.4, Duke q-value-0.039), C—C motif chemokine ligand 16 (CCL16, GSE p = 0.033, HR = 1.4, Duke q-value = 0.039), C—C motif chemokine ligand 21 (CCL21, GSE p = 0.01, HR = 1.5, Duke q-value = 0.04), interferon alpha 1 (IFNA1, GSE p-value = 0.023, HR = 1.9, Duke q-value = 0.039), interleukin 1 alpha (ILIA, GSE p-value = 0.035, HR = 1.8, Duke q-value = 0.023), mitogen-activated protein kinase 8 (MAPK8, GSE p-value = 0.022, HR = 1.7, Duke q-value = 0.03; and GSE p-value = 0.027, HR = 1.5, Duke q-value = 0.008), toll like receptor 8 (TLR8, GSE p-value = 0.035, HR = 2.0, Duke q-value = 0.008). In contrast, high expression of C-X-C motif chemokine ligand (CXCL10) was associated with longer OS (GSE p-value = 0.044, HR = 0.9, Duke q-value = 0.027). (Table 1 and Fig. 2) However, in the meta-analysis of the GSE datasets, no gene had significant p-values and was consistently concordant for association with OS.

4. Discussion

Our exploratory findings have identified 14 potential immunologic associated genes of interest that were differentially expressed in short vs long-term HGSC survivors. There are several genes of interest, however, further evaluation of IL6R expression may be warranted based on consistent direction of association in the Duke and TCGA datasets. Specifically, higher IL6R expression in long-term survivors in the Duke dataset and association with improved survival in patients with advanced HGSC that exhibited high IL6R expression in the TCGA database. Briefly, IL6 cytokines can cause increase immunosuppression and recruitment of inflammatory molecules resulting in infiltration of leukocytes in the tumor microenvironment. The leukocytes, including macrophages and dendritic cells, produce factors that promote tumor angiogenesis [41,42]. Recently we reported that plasma IL6 was both prognostic and predictive for bevacizumab efficacy in women with HGSC treated on GOG-0218 protocol [43]. In our present tissue based study, we observed that the IL6 signaling partners, IL6R was differentially expressed in women with short vs. long-term survival (Table 1). Women with longer survival had higher expression levels of IL6R compared to those with short-term survival. Moreover, higher IL6R expression was more likely in women with longer survival in the TCGA data. Similarly, increased sIL6R levels in EOC patients have been associated with a survival benefit (p = 0.03) [44]. However, in clear cell ovarian cancer higher expression levels of tissue IL6R and serum IL6R expression were correlated with poor patient survival [45,46]. Further evaluation is needed to understand the interaction between IL6 signaling pathway substrates in EOC as well as expression in blood versus tumor tissue.

In the GSE datasets, of the 83 genes that demonstrated significant association with OS in the Duke cohort, 13 genes (14 probe sets) exhibited level of significance and concordant direction for their association with OS (Table 1, Fig. 2). These genes may still be worthy of interest given our small sample size and limited power to investigate fully. Prognostic associations for some of the 13 Duke/GSE database candidate genes have been reported in the literature and showed associations with survival. For instance, high protein expression levels of CXCL10 in HGSC, was independently associated with increased overall survival compared to tumors with low expression which supports our findings [47]. CXCL10 enhances recruitment of tumor-infiltrating lymphocytes (TILS) and the increase of TILs in the microtumor environment may augment TIL-dependent immunotherapy strategies [47].

Immunohistochemistry assays for IL4R demonstrated that about 60% of ovarian tumor samples had high level of IL4R expression and 33% ovarian tumor samples demonstrated moderate levels of expression. In comparison, normal ovarian tissue has low expression or no expression of IL4R [48]. BCL2 is expressed in solid tumors including ovarian tumors. BCL2 expression had been reported to be involved in ovarian cancer progression and tumor infiltrating lymphocytes. IHC staining is reported as positive in both ovarian tumor and normal ovarian tissue. IHC staining for BCL2 in normal and benign ovarian tissues is stronger than in ovarian tumors [49]. BCL2 is considered a pro-survival protein and a major player in tumor pathogenesis, progression and resistance to treatment. BCL2 proteins are considered promising anticancer treatment targets [49-51]. MAPK8 (also known as JUN N-terminal kinase (JNK/JNK1)) inhibition has reported anti-tumor activity in ovarian cancer cell lines [52]. Activated MAPK8 (or pJNK) has been reported to be associated with decrease progression free survival in EOC [52].

To our knowledge, of the 14 common genes in the Duke, TCGA, and GSE databases, several of the genes of interest have been therapeutically targeted both in preclinical and clinical trials a total of 5 of the candidate genes have therapeutic agents available. Those included BCL2, CXCL10, CXCR1, IL6R and TLR8 (Supplemental Table 3). An IL6 pathway inhibitor, tocilizumab, is a novel monoclonal antibody that competitively inhibits the binding of IL to its receptor (IL6R) [53,54]. This immunosuppressive drug was studied in a phase I clinical trial in combination with carboplatin/doxorubicin in EOC. This combination chemo/immune therapy using tocilizumab has been recommended for a phase II clinical trial evaluation based on immune parameters [55].

Targeting TLR8 with VTX-2337, a TLR8 agonist, has been reported in a phase-1b trial (NCT01294293) as well as in a randomized phase II trial in recurrent or persistent ovarian epithelial, fallopian tube, or peritoneal cavity cancer [56,57]. The results demonstrated no improvement in PFS or OS in the unselected population. In a pre-specified subgroup analysis, those who experienced injection-site reactions had a lower risk of death when treated with VTX-2337 compared to those who did not have reactions. Additional biomarkers, such as tumor infiltrating lymphocytes, TLR8 single-nucleotide polymorphisms, and mutational BRCA and DNA repair gene status were not associated with OS or PFS [57]. Targeting the TLRs can be challenging as they have been reported to act as a “double-edged sword”, interacting with signaling pathways that can result in tumor-growth and tumor-suppression [58]. The findings from these clinical trials illustrate the importance of assessment of the therapeutic agents in tandem with rational biomarkers during drug development.

The strengths of our study are long-term surveillance and available survival data. Since none of the HGSC subjects in our Duke cohort was treated with immune therapy agents, the role of these candidate biomarkers could not be evaluated as predictive biomarkers for immune therapy treatment. Limitation inherent to the TCGA and GSE databases, include the short clinical follow-up and the variety of multiple platforms utilized to determine gene expression. Lack of access to additional biopsy samples from the Duke HGSC cohort also limited our ability to validate our RNA microarray results further using techniques such as qPCR and western blotting. While, our Affymetrix HG-U133A GeneChip microarray platform did include the probes for the PD1 (PDCD1) gene, a commonly targeted immunoncology strategy, it did not include the PDL1 (CD274). Other limitations include the small size of the Duke discovery cohort; small event data for survival analysis in genes such as BCL2-GSE26712 OS and the IL-4-GSE14764 (Fig. 2); tumor and data collection over a long time span (1988–2001); non-contemporaneous definition of optimal debulking; varied treatment regimens and modes of administration. While there may be concern regarding RNA degradation given the long time span of the study, only RNA with sufficient quantity and quality was utilized.

In conclusion, we have identified immune/inflammatory-related genes of interest that appear to be differentially expressed in short- and long-term survivors with advanced HGSC treated with platinum-based therapy that may be worthy of further investigation. Several of the genes have been reported in the literature to be associated with survival outcomes or poor prognostic clinicopathologic variables. There are several therapeutic agents that target genes of interest that are currently available that could be readily assessed in ovarian cancer, but further validation and preclinical studies are needed before targeting these genes and their respective proteins can be utilized in clinical trial development. The disappointing results noted with interferon and VTX-2337 highlight the importance of drug development in tandem with rationale biomarkers.

HIGHLIGHTS.

• Immunologic/inflammatory-related genes can be differentially expressed in HGSC.

• Differential expression of those genes can be associated with long versus short-term HGSC survival.

• High expression of IL6R demonstrated higher expression in long-term survivors in Duke extreme phenotype study database.

• High expression of the IL6R was associated with improved overall survival in the TCGA database.