Purpose:

Ovarian cancer has one of the highest deaths to incidence ratios across all cancers. Initial chemotherapy is effective, but most patients develop chemoresistant disease. Mechanisms driving clinical chemo-response or -resistance are not well-understood. However, achieving optimal surgical cytoreduction improves survival, and cytoreduction is improved by neoadjuvant chemotherapy (NACT). NACT offers a window to profile pre- versus post-NACT tumors, which we used to identify chemotherapy-induced changes to the tumor microenvironment.

Experimental Design:

We obtained matched pre- and post-NACT archival tumor tissues from patients with high-grade serous ovarian cancer (patient, n = 6). We measured mRNA levels of 770 genes (756 genes/14 housekeeping genes, NanoString Technologies), and performed reverse phase protein array (RPPA) on a subset of matched tumors. We examined cytokine levels in pre-NACT ascites samples (n = 39) by ELISAs. A tissue microarray with 128 annotated ovarian tumors expanded the transcriptional, RPPA, and cytokine data by multispectral IHC.

Results:

The most upregulated gene post-NACT was IL6 (16.79-fold). RPPA data were concordant with mRNA, consistent with elevated immune infiltration. Elevated IL6 in pre-NACT ascites specimens correlated with a shorter time to recurrence. Integrating NanoString (n = 12), RPPA (n = 4), and cytokine (n = 39) studies identified an activated inflammatory signaling network and induced IL6 and IER3 (immediate early response 3) post-NACT, associated with poor chemo-response and time to recurrence.

Conclusions:

Multiomics profiling of ovarian tumor samples pre- and post-NACT provides unique insight into chemo-induced changes to the tumor microenvironment. We identified a novel IL6/IER3 signaling axis that may drive chemoresistance and disease recurrence.

Translational Relevance

Patients with ovarian cancer often respond to initial chemotherapy, however, most patients will recur with chemo-resistant disease. Biomarkers or factors that are able to predict disease recurrence are lacking. We hypothesize that the extent of chemotherapy-induced remodeling of the tumor immune microenvironment directly contributes to disease recurrence. We performed multiomics analysis of tumors pre- and postchemotherapy. Cytokine profiles and tumor microenvironments were then characterized in a large cohort of pre- and postchemotherapy, and recurrent tumors. We discovered that the capacity of the tumor microenvironment to induce an inflammatory response correlates to a shorter disease-free interval. Predicting recurrence by defining chemotherapy-induced transcriptional and proteomic changes within a patient's tumor provides a precision medicine–based approach to improve disease monitoring, to hasten response to disease recurrence, and tailor future therapies.

High-grade serous ovarian carcinoma (HGSOC) accounts for nearly 75% of all ovarian cancers and is the deadliest histotype, driven by the propensity of disease recurrence and therapy resistance (1). Most patients are treated by cytoreductive surgery and combination platinum/taxane-based chemotherapy. The understanding of clinical chemoresistance in HGSOC is limited, as access to treated tumor tissue after the initial cytoreductive surgery and subsequent chemotherapy is uncommon. This lack of access to treated tumor hinders the ability to define how tumors respond and adapt to therapy, and to identify predictors and drivers of disease progression.

Although the mechanistic understanding of chemo-response is limited, successful surgical removal of tumor remains one of the best predictors of survival. On the basis of this, patients with HGSOC undergo a diagnostic laparoscopy to determine the extent of disease burden and likelihood for optimal cytoreduction (2, 3). Unresectable disease is associated with poor patient outcomes; however, tumor resectability is improved by the use of neoadjuvant chemotherapy (NACT) with interval cytoreduction. NACT consists of three cycles of carboplatin/paclitaxel, then an interval surgical cytoreduction of remaining disease, followed by another three carboplatin/paclitaxel cycles. NACT improves rates of optimal surgical cytoreduction in patients with advanced disease (4–6). Importantly, the carboplatin/paclitaxel chemotherapies used in NACT remain the standard upfront care for adjuvant therapy. The NACT paradigm provides access to the tumor site before and after chemotherapy, which presents an opportunity to investigate the mechanisms of tumor response to chemotherapy.

Studies of HGSOC tumor response to NACT are emerging, and have identified that immune/inflammatory signaling plays a significant role in chemo-response and -resistance. Böhm and colleagues reported that following NACT, an increase in tumor infiltration of immunosuppressive T regulatory cells (Tregs) and elevated T-cell expression of checkpoint receptors correlate to a poorer prognosis (7). Lo and colleagues reported a similar finding, while chemotherapy promotes an immune response, this immune response does not circumvent an established immunosuppressive microenvironment (8). Beyond NACT, the cytokine IL6 significantly contributes to the immunosuppressive microenvironment, and is associated with disease progression and chemotherapy response (9–13). These observations suggest a key role for cytokines in modulating tumor response to chemotherapy. Greater mechanistic understanding of the tumor immune microenvironment may allow us to overcome chemo-resistance and improve patient outcomes.

We hypothesized that integrating multiple profiling methods of both the tumor and tumor microenvironment pre- and post-NACT would identify predictors and drivers of chemotherapy response. To address this, we identified a series of matched archival tumor samples pre- and post-NACT from six patients with HGSOC. Using these samples, we defined the transcriptomic effects of NACT using the NanoString PanCancer Immuno-Oncology (IO) 360 panel. We then identified signaling events linked to these NACT-driven transcriptomic changes by using reverse phase protein array (RPPA) analyses. We examined the cytokine milieu in pretreatment ascites from 39 patients. Integrating these multiomics analyses via network analysis and comparing data against different platforms, we identified an immune microenvironment signaling pathway associated with decreased time to recurrence. These data suggest that the magnitude of chemotherapy-induced remodeling of the tumor microenvironment provides a unique predictor of recurrence and that multiomics profiling of pre- and post-NACT tumors is a powerful tool for understanding HGSOC.

Gynecologic tumor and fluid bank

Patients in this study were consented under the gynecologic tumor and fluid bank (GTFB) protocol at the University of Colorado [Aurora, CO, Colorado Medical Institutional Review Board (COMIRB) #07-935]. The protocol was approved and conducted in accordance with the University of Colorado (Aurora, CO) ethical policy. All patients signed a written consent prior to enrollment. Specimens were deidentified and processed into formalin-fixed, paraffin embedded (FFPE) blocks. Tumors were also flash-frozen and stored at −80°C. Ascites was centrifuged at 200 × g for 7 minutes and the fluid and cellular components were stored at −80°C.

Multiplex ELISA

The concentration of cytokines/chemokines was assessed using the V-PLEX Proinflammatory Panel 1 Kit (Meso Scale Discovery). Ascites fluid was diluted twofold in diluent 2, incubated for 2 hours while shaking (600 rpm) at room temperature, and then incubated for 2 hours with detection antibody in diluent 3 while shaking (600 rpm) at room temperature. The plates were washed using an automated Plate Washer (Biotek Instruments). Electrochemiluminescence values were read on the Meso QuickPlex SQ 120 (Meso Scale Discovery). Sample concentrations were extrapolated using the standard curve of diluted calibrators.

RNA extraction

FFPE tissue blocks were acquired (COMIRB#18-0119) and a board-certified pathologist confirmed tumor tissue (Supplementary Table S1). This protocol was deemed exempt, as it used previously collected data, and the information was not recorded in a manner that is identifiable. FFPE tissue blocks were sectioned into 10-μm tissue-containing paraffin scrolls. RNA was extracted using the High Pure FFPET RNA Isolation Kit (Roche). RNA quantity and quality were assessed using a RNA ScreenTape on a TapeStation 4150 (Agilent Technologies). RNA concentration was determined by comparison with the RNA ladder and the percentage of RNA fragments greater than 200 bp was calculated (average 70.6%, all samples >55%).

NanoString PanCancer IO 360

RNA (150 ng) was combined with hybridization buffer and the reporter CodeSet for the PanCancer IO 360 Panel (NanoString Technologies) and incubated for 20 hours at 65°C. The hybridized reaction was analyzed on an nCounter SPRINT Profiler (NanoString Technologies). nSolver calculated normalization factors for each sample using raw gene counts and 14 housekeeping genes (Supplementary Table S2). Differential gene expression was calculated from normalized gene counts data and an FDR with a Benjamini–Hochberg multicomparison test. The average count for the negative control probes was used for thresholding “positive” genes. After thresholding (<20 counts), a total of 666 genes were subsequently used for downstream analysis. The heatmap was generated using Clustergrammer (14). Raw gene counts were normalized using the logCPM method, filtered by selecting the genes with most variable expression, and transformed using the z-score method.

nSolver advanced analysis tool was used to generate a pathway score for 25 different pathways (e.g., hypoxia). The pathways scores were grouped and compared on the basis of pre- versus post-NACT. Genetic signature analysis was performed by NanoString Technologies as described in (15–19). The log2-transformed gene expression values were multiplied by predefined weighted coefficient (16) and the sum of these values within each gene set was defined as the signature score.

RPPA

Fresh snap-frozen tumor tissue was processed and analyzed by MD Anderson Functional Proteomics RPPA Core Facility (Houston, TX; ref. 20). The RPPA (set167) included 449 antibodies, which were analyzed by Array-Pro Analyzer 6.3 then by SuperCurve 1.5.0 via SuperCurveGUI 2.1.1. The data were normalized on the basis of the median and centered on the basis of the medians normalized to log2 intensities.

Transcriptional factor analysis

With the genes (n = 18) that had a significant correlation with time to recurrence, we utilized PathwayNet Analysis to predict transcription factor (TF) relationships (21). Of these 18 query genes, five had no associated TFs; the remaining 13 query genes had 371 associated TFs. From this bipartite network, we derived a weighted projection onto the 13 query genes such that two nodes in the projection were connected by an edge weighted by the number of associated TFs they have in common. Under this projection, all TFs that associated with only one query gene were effectively removed, and higher weight edges could be interpreted as more reliable associations. Each node was then annotated by its “strength,” defined as the total weight of edges incident to it in the projection, that is, the number of shared TFs with any other query gene. This operation produced a network with 13 nodes and 54 edges, accounting for 250 shared TFs.

Tissue microarray

A described (22) tissue microarray (TMA) comprising serous tumors (COMIRB #17-7788) was used. This protocol was deemed exempt, as it used previously collected data, and the information was not recorded in a manner that is identifiable. The TMA consisted of 109 primary tumors collected prechemotherapy, 19 primary tumors collected postchemotherapy, and 28 recurrent tumors. The time range of recurrence was 8–81 months.

Multispectral IHC

Multispectral IHC analyses were performed using Vectra Automated Quantitative Pathology Systems (Akoya Biosciences) as described previously (23). For Fig. 4CH, the TMA slides were sequentially stained with antibodies specific for CD4 (T cells, 4B12, Leica Biosystems), CD8 (T cells, C8/144B, Agilent Technologies), FOXP3 (Tregs, 236A/E7, Abcam), Granzyme B (effector T cells, GRB-7, Thermo Fisher Scientific), CD68 (macrophages, KP1, Agilent Technologies), and cytokeratin (tumor cells, AE1/AE3, Agilent Technologies). For Fig. 4I and J, the TMA slides were sequentially stained with antibodies specific for CD3 (T cells, LN10, Leica), CD19 (B cells, BT51E, Leica), IER3 (NBP1-76802, Novus Biologicals), Ki67 (SP6, Thermo Fisher Scientific), and CD8, CD68, and cytokeratin as above. Image analysis was performed using inForm Software version 2.3 (Akoya), including tissue segmentation, cell segmentation, and phenotyping to assign each cell to a phenotypic category. The TMA slides were stained with IER3 antibody, imaged on the Vectra Polaris Microscope (Akoya), and analyzed with inForm Software version 2.8 (Akoya) using a 4-bin algorithm to calculate histologic scores (H-scores).

cBioPortal Cancer Cell Line Encyclopedia and Gene Expression Omnibus

Cancer Cell Line Encyclopedia (CCLE) data (Broad 2019 dataset update; ref. 24) were downloaded from the cBioPortal in May 2020. Ovarian cancer cell lines were selected from the “ovarian” tumor type category; variable n's in the indicated subsets in Fig. 3E and F represent mixed availability of expression and drug response data available per cell line. IL6/IER3-high versus -low expression was defined by a combined z-score cutoff of 2. GSE117765 (25) of olaparib-sensitive and -resistant cells was used to assess transcript counts of IL6 and immediate early response 3 (IER3).

Cell culture

OVCAR4 (CVCL_1627) and OVCAR8 (CVCL_1629) cell lines were authenticated using small tandem repeat analysis (The University of Arizona Genetics Core, Tucson, AZ) and tested for Mycoplasma (MycoLookOut, Sigma). Cells were obtained from the GTFB. Mycoplasma testing was last performed on 28 May 2020. Cells were cultured up to 20 passages in RMPI1640 (Gibco) medium supplemented with 1% penicillin–streptomycin and 10% FBS (Access) and maintained in 5% CO2 at 37°C.

siRNA knockdown

Cells were transfected using Lipofectamine RNAiMAX Reagent (Invitrogen #13778-075) according to the manufacturer's protocol. The siRNAs used were siIER3 (Ambion Silencer Select #4392421, ID s16941) or Negative Control siRNA (Ambion #AM4613).

Cisplatin dose response

Cisplatin was purchased from SelleckChem and resuspended in 0.9% sodium chloride/ddH2O. Cells were plated in 96-well plates in cisplatin (100 nmol/L, 1 μmol/L, 10 μmol/L, or 50 μmol/L). Cells were imaged for 72 hours on an IncuCyte S3 live cell imaging system. Cell confluence was analyzed using IncuCyte software GUI version 2019B Rev2.

Quantitative reverse-transcription PCR

Quantitative reverse-transcription PCR was performed as described in (25). The Luna Universal One-step RT-qPCR Kit (New England Biolabs) was used on a Bio-Rad CFX96 Thermocycler. The following primers were used for IER3: forward- AGCCGCAGGGTTCTCTA and reverse- GATGGTGAGCAGCAGAAAGA; and GAPDH, as an internal control, forward- GTCTCCTCTGACTTCAACAGCG and reverse- ACCACCCTGTTGCTGTAGCCAA.

Statistical analysis

Statistical analysis was performed in GraphPad Prism (v8). Two-tailed Student t test or ANOVA was utilized to calculate a P value. An adjusted P value of less than 0.20 was considered significant. Paired or unpaired t test was used on the basis of context and noted in figure legend. All quantitative data are graphed as mean with SEM. HR was calculated with Mantel–Haenszel test. The power analysis is based on the two-sample independent t tests adjusting for the FDR at a 0.20 level. Assuming that actually 10% of the biomarkers were truly differentially expressed, then minimum Cohen D effect size, in which the effect size (i.e., the fold change) was standardized by the pooled SDs, was calculated for sample size range from six to 20 patients (Supplementary Table S2).

Interrogation of chemotherapy-induced changes to the ovarian tumor transcriptome

Patients with advance-stage HGSOC and extensive disease burden were treated with NACT, which provides a unique opportunity to investigate genetic and molecular change induced by carboplatin and paclitaxel (Fig. 1A). Therefore, we examined the transcriptional profile of matched tumor samples from six patients pre- and post-NACT (total N = 12). All six patients had extensive disease burden upon clinical presentation and were all treated with NACT with interval cytoreduction. These cases were selected for further analysis on the basis of the availability of FFPE blocks, snap-frozen tumor samples, and clinical information. The average time to disease recurrence was 243.5 days (range, 25–505 days); additional clinical attributes considered included genetic characteristics such as BRCA and homologous recombination status (Supplementary Fig. S1A). During the interval debulking, the number of tumor sites was quantified and ranged from four to 15 (Supplementary Data). However, based on CA125 levels at the time of diagnosis, there was not a clear relationship between disease burden and time to recurrence (Supplementary Table S1).

Figure 1.

Chemotherapy-induced transcriptional dynamics. A, Graphic representation of NACT. Large arrowheads, surgeries; small black arrows, cycles of carboplatin/paclitaxel. B, Principal component (PC) analysis of the transcriptome of 770 genes (14 housekeeping genes not included) in HGSOC tumors pre- (blue) and post- (red) chemotherapy. C, Volcano plot of differentially regulated genes. P value calculated with paired t test with Benjamin–Hochberg multicomparison correction (Padj value). Yellow dots, log2 FC > 0 and Padj < 0.2. Pink dots, log2 FC < 0 and Padj < 0.2. D, Heatmap of the 74 differentially expressed genes, generated with Clustergrammer from NanoString data. Normalized counts are transformed with z-score method. E,MKI67 expression changes between pre- and post-NACT. F,IL6 expression changes between pre- and post-NACT. G, TF relationship of 18 genes associated with disease recurrence. Generated via PathwayNet data portal (21). H, Utilizing the gene expression data genetic signature analyses was performed on all 12 samples. Shown are representative images of the genetic signature analysis from two sets of pre- and postchemotherapy tumors (GTFB974 and GTFB1064). Each spoke on the circular plot indicates a unique genetic signature, such as proliferation, antigen-processing and presentation machinery loss, and glycolysis. Genetic signatures are grouped by tumor immune microenvironment and characteristics of the tumor microenvironment. The tumor inflammation score is indicated at the center of the circular plot. I, Log2 fold change forest plot of genetic signatures when grouped according to pre- and postchemotherapy. Orange asterisks, P < 0.05 and FDR < 0.20.

Figure 1.

Chemotherapy-induced transcriptional dynamics. A, Graphic representation of NACT. Large arrowheads, surgeries; small black arrows, cycles of carboplatin/paclitaxel. B, Principal component (PC) analysis of the transcriptome of 770 genes (14 housekeeping genes not included) in HGSOC tumors pre- (blue) and post- (red) chemotherapy. C, Volcano plot of differentially regulated genes. P value calculated with paired t test with Benjamin–Hochberg multicomparison correction (Padj value). Yellow dots, log2 FC > 0 and Padj < 0.2. Pink dots, log2 FC < 0 and Padj < 0.2. D, Heatmap of the 74 differentially expressed genes, generated with Clustergrammer from NanoString data. Normalized counts are transformed with z-score method. E,MKI67 expression changes between pre- and post-NACT. F,IL6 expression changes between pre- and post-NACT. G, TF relationship of 18 genes associated with disease recurrence. Generated via PathwayNet data portal (21). H, Utilizing the gene expression data genetic signature analyses was performed on all 12 samples. Shown are representative images of the genetic signature analysis from two sets of pre- and postchemotherapy tumors (GTFB974 and GTFB1064). Each spoke on the circular plot indicates a unique genetic signature, such as proliferation, antigen-processing and presentation machinery loss, and glycolysis. Genetic signatures are grouped by tumor immune microenvironment and characteristics of the tumor microenvironment. The tumor inflammation score is indicated at the center of the circular plot. I, Log2 fold change forest plot of genetic signatures when grouped according to pre- and postchemotherapy. Orange asterisks, P < 0.05 and FDR < 0.20.

Close modal

FFPE tumor tissue was subsequently identified and utilized for RNA isolation and transcriptomic analyses. We employed the NanoString platform to examine the expression of 770 highly annotated genes, including housekeeping genes, in the PanCancer IO 360 panel, which targets genes expressed by tumor cells and immune cells in the tumor microenvironment (19). Linear regression of housekeeping genes between matched pre- and post-NACT tumors showed a significant correlation in five of six samples (Supplementary Fig. S1B), confirming that housekeeping genes were comparable and could be used for normalization. While still significantly correlated, GTFB1064 samples demonstrated the highest degree of housekeeping gene variation between pre- and postchemotherapy. On the basis of these successful quality control benchmarks, we compared gene expression between matched samples, and across pre- versus post-NACT samples. A total of 666 target genes passed initial filtering (i.e., removal of genes with raw counts <20) and were subsequently used for downstream analysis.

Principal component analysis of the 12 samples revealed a clear delineation between the pre- and post-NACT tumors (Fig. 1B; Supplementary Fig. S1C). This delineation suggests that there is a common or shared effect of chemotherapy on the tumor and tumor microenvironment (rather than chemotherapy causing primarily tumor-specific transcriptional changes). In total, 69 of 666 genes were differentially expressed in paired analyses of pre- versus post-NACT samples (FDR P < 0.2; Fig. 1C and D; Supplementary Table S3). One of most strongly downregulated genes following NACT was MIK67 (Fig. 1E; Supplementary Fig. S1D). Downregulation of MIK67 (Ki-67, a proliferation marker) after NACT is consistent with chemotherapy targeting the dividing cells and causing cell-cycle arrest. We identified post-NACT changes in gene expression signatures for each tumor pair. Across this samples series, signatures for cell proliferation, DNA damage repair, hypoxia, MAPK, PI3K/AKT, and JAK/STAT were the most significantly enriched (Table 1). We considered that an inflammatory response may mediate many of these identified post-NACT gene expression changes, and supporting this, IL6 was the most upregulated gene in the post-NACT tumors (16.79-fold change; P = 0.0812; Fig. 1F). Although this did not reach statistical significance in paired analyses, this observation parallels reports that IL6 is induced following chemotherapy (10, 26). Consistent with the activated JAK/STAT signature as a downstream effector of IL6, we noted that JAK1/2/3 expressions were all significantly elevated in the post-NACT compared with pre-NACT (Supplementary Fig. S1F). These data highlight that at the transcriptomic level, chemotherapy is successfully targeting proliferating cells, but concomitantly activating an inflammatory response.

Table 1.

Pathway analysis of transcriptional profiles.

PathwayAverage signature score (relative to pre-NACT)PPadj
Cell proliferation −2.9001 0.0014 0.0084 
DNA damage repair −1.5427 0.0020 0.0084 
Hypoxia 1.8727 0.0015 0.0084 
MAPK 2.3865 0.0014 0.0084 
Metabolic stress 2.7211 0.0018 0.0084 
PI3K-Akt 2.7639 0.0012 0.0084 
Wnt signaling 1.6344 0.0035 0.0126 
Epigenetic regulation −1.1595 0.0064 0.0199 
Apoptosis −1.4862 0.0085 0.0235 
Hedgehog signaling 1.1389 0.0107 0.0253 
TGFβ signaling 1.1563 0.0111 0.0253 
JAK-STAT signaling 1.8937 0.0191 0.0399 
PathwayAverage signature score (relative to pre-NACT)PPadj
Cell proliferation −2.9001 0.0014 0.0084 
DNA damage repair −1.5427 0.0020 0.0084 
Hypoxia 1.8727 0.0015 0.0084 
MAPK 2.3865 0.0014 0.0084 
Metabolic stress 2.7211 0.0018 0.0084 
PI3K-Akt 2.7639 0.0012 0.0084 
Wnt signaling 1.6344 0.0035 0.0126 
Epigenetic regulation −1.1595 0.0064 0.0199 
Apoptosis −1.4862 0.0085 0.0235 
Hedgehog signaling 1.1389 0.0107 0.0253 
TGFβ signaling 1.1563 0.0111 0.0253 
JAK-STAT signaling 1.8937 0.0191 0.0399 

Note: Pathway name, average signature score relative to pre-NACT. Pre (n = 6) and post (n = 6). NanoString transcriptomic data were used as input to assess pathway signature. Two-tailed paired t test, Benjamini–Hochberg Padj value.

To identify transcriptional changes associated with disease recurrence, we assessed whether the measured transcriptional changes between pre- and post-NACT samples correlated with time to recurrence. We discovered that fold change between pre- and post-NACT of 18 genes significantly correlated with time to recurrence (|r| > 0.8; Table 2). For instance, the magnitude of NACT-induced expression of the FGFR ligand, FGF9, correlated with time to recurrence (r = −0.990; i.e., high FGF9 expression correlates with short time to recurrence). Signaling activated by FGF9 was consistent with the elevated PI3K/Akt and MAPK signaling observed in the pathway analyses. Taken together, these transcriptional data are suggestive of a specific stress-like response to the post-NACT tumor microenvironment.

Table 2.

Top 18 genes that correlated with time to recurrence.

Log2 FC post/pre for each patient
Gene symbolGene descriptionGTFB 1002GTFB 1064GTFB 1066GTFB 1046GTFB 722GTFB 974Correlation to recurrence (r, Pearson P < 0.05)
FGF9 Fibroblast growth factor 9 −1.36 −0.66 −0.38 −0.37 0.87 0.77 −0.990 
CXCL5 C-X-C motif chemokine 5 −1.41 0.08 −0.34 1.54 1.58 2.60 −0.921 
IER3 Immediate early response 3 0.20 −0.12 0.52 0.55 1.30 1.91 −0.885 
AXIN1 Axin-1 −0.29 −0.23 0.00 −0.07 0.29 0.04 −0.869 
SPRY4 Sprouty homolog 4 −1.11 −1.20 −0.62 −1.16 0.77 0.58 −0.869 
THBS1 Thrombospondin-1 0.07 −0.18 1.14 1.37 1.32 2.37 −0.853 
CASP3 Caspase-3 −0.67 −0.17 −0.50 −0.37 0.24 1.20 −0.839 
ITGAV Integrin alpha-V −0.10 −0.01 −0.05 −0.32 1.11 1.14 −0.823 
CEBPB CCAAT/enhancer-binding protein beta −0.01 0.29 0.16 1.00 1.28 0.81 −0.816 
OASL 2′-5′-Oligoadenylate synthetase like 0.83 −0.88 −1.72 −2.20 −1.79 −2.04 0.817 
IFITM1 Interferon induced transmembrane protein 1 1.23 1.83 1.33 −0.83 −1.00 −1.40 0.820 
CST2 Cystatin SA 2.32 −0.26 −0.69 −0.43 −0.65 −1.21 0.832 
HNF1A Hepatocyte nuclear factor 1-alpha 1.64 0.36 0.01 0.23 −0.10 −0.04 0.839 
IRF7 Interferon regulatory factor 7 0.35 −0.14 −0.52 −0.59 −0.86 −0.51 0.855 
VTCN1 V-Set domain containing T cell activation inhibitor 1 0.07 −2.44 −4.05 −1.47 −4.21 −5.44 0.867 
MX1 MX dynamin-like GTPase 1 1.07 0.40 −1.09 −2.00 −1.46 −2.46 0.873 
CCNO Cyclin O 0.48 −0.19 −1.56 −2.80 −2.73 −3.00 0.903 
HERC6 HECT and RLD domain containing E3 ubiquitin ligase 6 1.00 0.68 0.19 −1.23 −1.68 −1.58 0.919 
Time to recurrence (days) 505 347 279 245 60 25  
Log2 FC post/pre for each patient
Gene symbolGene descriptionGTFB 1002GTFB 1064GTFB 1066GTFB 1046GTFB 722GTFB 974Correlation to recurrence (r, Pearson P < 0.05)
FGF9 Fibroblast growth factor 9 −1.36 −0.66 −0.38 −0.37 0.87 0.77 −0.990 
CXCL5 C-X-C motif chemokine 5 −1.41 0.08 −0.34 1.54 1.58 2.60 −0.921 
IER3 Immediate early response 3 0.20 −0.12 0.52 0.55 1.30 1.91 −0.885 
AXIN1 Axin-1 −0.29 −0.23 0.00 −0.07 0.29 0.04 −0.869 
SPRY4 Sprouty homolog 4 −1.11 −1.20 −0.62 −1.16 0.77 0.58 −0.869 
THBS1 Thrombospondin-1 0.07 −0.18 1.14 1.37 1.32 2.37 −0.853 
CASP3 Caspase-3 −0.67 −0.17 −0.50 −0.37 0.24 1.20 −0.839 
ITGAV Integrin alpha-V −0.10 −0.01 −0.05 −0.32 1.11 1.14 −0.823 
CEBPB CCAAT/enhancer-binding protein beta −0.01 0.29 0.16 1.00 1.28 0.81 −0.816 
OASL 2′-5′-Oligoadenylate synthetase like 0.83 −0.88 −1.72 −2.20 −1.79 −2.04 0.817 
IFITM1 Interferon induced transmembrane protein 1 1.23 1.83 1.33 −0.83 −1.00 −1.40 0.820 
CST2 Cystatin SA 2.32 −0.26 −0.69 −0.43 −0.65 −1.21 0.832 
HNF1A Hepatocyte nuclear factor 1-alpha 1.64 0.36 0.01 0.23 −0.10 −0.04 0.839 
IRF7 Interferon regulatory factor 7 0.35 −0.14 −0.52 −0.59 −0.86 −0.51 0.855 
VTCN1 V-Set domain containing T cell activation inhibitor 1 0.07 −2.44 −4.05 −1.47 −4.21 −5.44 0.867 
MX1 MX dynamin-like GTPase 1 1.07 0.40 −1.09 −2.00 −1.46 −2.46 0.873 
CCNO Cyclin O 0.48 −0.19 −1.56 −2.80 −2.73 −3.00 0.903 
HERC6 HECT and RLD domain containing E3 ubiquitin ligase 6 1.00 0.68 0.19 −1.23 −1.68 −1.58 0.919 
Time to recurrence (days) 505 347 279 245 60 25  

Note: Log2 fold change (FC, n = 12) correlated with time to disease recurrence (last row), bold numbers. NanoString transcriptomic data were used. Pearson correlation analysis (|r| > 0.8; P < 0.05).

Transcription network and gene signature analyses of transcriptomes

We hypothesized that we could identify stress response factors driving the gene expression pattern changes post-NACT, and to do so, we used the 18 genes correlated with time to recurrence to construct a transcriptional regulatory network of post-NACT gene expression. In this network analysis, CEBPB (CCAAT enhancer binding protein beta; C/EBPβ) emerged as a central transcriptional hub driving a downstream network consisting of THBS1, IER3, OASL, IRF7, and CXCL5 (Fig. 1G). This result implicates C/EBPβ as a central integrator of immune, inflammatory, and stress signaling in the ovarian epithelial cancer tumor microenvironment during NACT, which drives transcriptional programs mediating poor response to chemotherapy. Importantly, pathways activating C/EBPβ include growth factor signaling and IL6 signaling (27), suggesting that the genes and pathways identified above as upregulated in response to NACT converge on C/EBPβ.

We then used a series of predefined genetic signatures developed by NanoString Technologies to extrapolate tumor microenvironment conditions based on gene expression, for example, immune and stromal cell components, inflammatory signaling (via a tumor inflammation score), and immune activation state (Fig. 1H and I; ref. 15–19). Comparing pre- versus post-NACT gene expression using these genetic signatures identified a significant decrease (P < 0.05) in “proliferation” and “apoptosis” signatures following chemotherapy. In contrast, signatures that were significantly upregulated (P < 0.05) following chemotherapy included antigen-processing and presentation machinery loss, IL10, mast cells, dendritic cells, and endothelial cells (Fig. 1I). These data support our pathway analysis and suggest that increased inflammatory signaling driven by remodeling of the tumor immune infiltrate alters tumor cell signaling and ultimately drives chemotherapy resistance.

Alignment of the transcriptomic with chemotherapy-induced protein changes

Because transcriptional analyses identified critical signaling pathways activated by chemotherapy, we directly assessed the activation of protein signaling cascades by RPPA. A limitation of this approach is the requirement for fresh-frozen tissue samples, which were available from only two of the six patients treated with NACT (GTFB1064 and GTFB1066). A pre- and post-NACT fresh-frozen tumor was analyzed from both GTFB1064 and GTFB1066 (n = 4). Despite this limitation, RPPA data closely aligned with NanoString gene expression data for the 110 targets overlapping between the PanCancer IO 360 panel and MD Anderson's standard RPPA platform (Fig. 2A; Supplementary Table S3). Among the 110 overlapping targets, there was a strong positive correlation in both GTFB1064 and GTFB1066 between mRNA- and protein-level fold changes (post/pre; Fig. 2B and C). Signaling changes post-NACT in the RPPA also supported signatures identified in transcriptional analyses. Consistent with suppression of proliferation by chemotherapy, CCNB1 downregulation (log2 fold change = −2.24; P = 0.013; FDR 0.181) was paralleled by decreased Cyclin B1 levels in RPPA (log2 fold change = −2.43; Fig. 2D and E). Consistent with activated PI3K/Akt signaling, SGK1 mRNA and SGK1 protein were upregulated post-NACT (mRNA log2 fold change = 2.26, FDR 0.20; RPPA log2 fold change = 0.76; Fig. 2F and G). Similarly, IL6 upregulation was paralleled by increased IL6 protein levels (RPPA log2 fold change = 0.43; Fig. 2H and I). RPPA analyses also supported the activation of IL6 and receptor tyrosine kinase (RTK) signaling via increased levels of phosphorylation of key signaling proteins. Levels of phosphorylated STAT1 (pY702), c-ABL (pY412), EGFR (pY1173), HER3 (pY1289), and IGFR (pY1135/36) were elevated post-NACT (Fig. 2J and K). Notably, chemotherapy did not activate all RTKs; phosphorylated HER2 (pY1248) and c-Met (pY1234/35) were unchanged following NACT (Supplementary Table S3). These RPPA data closely parallel our transcriptomic data, together identifying increased IL6 and activation of specific RTKs post-NACT. Importantly, the RPPA data directly identified signaling cascades activated post-NACT, in particular JAK/STAT, which is a key downstream target of IL6.

Figure 2.

RPPA of pre- and postchemotherapy-treated HGSOC tumors. A, Venn diagram indicating the overlap in targets from NanoString and RPPA. B, Scatter plot of log2 fold change (FC, post/pre) of the overlapping targets of the RPPA (y-axis) and NanoString (x-axis) from GTFB1064. Pearson correlation was calculated. C, Same as B, but with expression data from GTFB1066. D, Transcript counts of CCNB1 for GTFB1064 and GTFB1066. E, RPPA data of cyclin B1 protein from pre- and post-treated tumors. F, Transcript counts of SGK1 for GTFB1064 and GTFB1066. G, RPPA data of SGK1 protein from pre- and post-treated tumors. H, Transcript counts of IL6 for GTFB1064 and GTFB1066. I, RPPA data of IL6 protein for pre and post-treated tumors. J, RPPA data of phosphoSTAT1 and phospho-c-Abl. Connecting lines indicate matched tumors. K, RPPA data of phosphorylated RTKs, EGFR (pY1173), HER3 (pY1289), and IGF1R (pY1135/36).

Figure 2.

RPPA of pre- and postchemotherapy-treated HGSOC tumors. A, Venn diagram indicating the overlap in targets from NanoString and RPPA. B, Scatter plot of log2 fold change (FC, post/pre) of the overlapping targets of the RPPA (y-axis) and NanoString (x-axis) from GTFB1064. Pearson correlation was calculated. C, Same as B, but with expression data from GTFB1066. D, Transcript counts of CCNB1 for GTFB1064 and GTFB1066. E, RPPA data of cyclin B1 protein from pre- and post-treated tumors. F, Transcript counts of SGK1 for GTFB1064 and GTFB1066. G, RPPA data of SGK1 protein from pre- and post-treated tumors. H, Transcript counts of IL6 for GTFB1064 and GTFB1066. I, RPPA data of IL6 protein for pre and post-treated tumors. J, RPPA data of phosphoSTAT1 and phospho-c-Abl. Connecting lines indicate matched tumors. K, RPPA data of phosphorylated RTKs, EGFR (pY1173), HER3 (pY1289), and IGF1R (pY1135/36).

Close modal

Examination of the cytokine-associated ascites microenvironment

JAK/STAT signaling is initiated through the interaction of a cytokine receptor (e.g., IL6R) with its cognate ligand (e.g., IL6), the latter of which may be secreted from the tumor or components of the tumor microenvironment, such as macrophages. As a surrogate for the tumor and associated microenvironment, we examined cytokine levels in ascites fluid from patients with HGSOC (most patients with advanced HGSOC present with ascites accumulation within the peritoneal cavity). Given that we found increased IL6 levels and JAK/STAT signaling activity post-NACT, we examined whether cytokine levels prior to chemotherapy predicted time to recurrence. Using primary ascites from 39 patients (prechemotherapy, ascites fluid is typically not available posttherapy), we used a multiplexed ELISA method to measure levels of the cytokines IFNγ, IL10, IL12p70, IL13, IL-1β, IL2, IL4, IL6, IL8, and TNFα. Overall, IL6, IL8, IL10, and IFNγ had the largest range of concentration (Fig. 3A). The samples used for RPPA, GTFB1064 and GTFB1066, both had elevated IL6 concentrations, 1,576 and 1,820 pg/mL, respectively. We noted that IL12p70 and IL4 were significantly correlated (Fig. 3B; Pearson r = 0.9870; P < 0.0001). All of the cytokines were correlated with time to disease recurrence, which ranged from 1 to 66 months. Increasing concentrations of IL6 or IL10 in ascites correlated with a shorter time to disease recurrence (Fig. 3C and D), suggesting that elevated cytokine signaling prior to or in response to chemotherapy may mediate resistance. Notably, there were several patients with low levels of IL6 and IL10 that recurred within 10 months of the completion of primary chemotherapy (Fig. 3C and D, red dots). These data demonstrate that while IL6 (HR, 2.12) and IL10 (HR, 1.52) concentrations predict time to disease recurrence, there is still a need to further elucidate the mechanisms underlying the tumor and immune microenvironment response to cytokine stimulation, especially in the context of chemotherapy.

Figure 3.

Cytokine profiling of primary patient-derived ascites and transcriptional changes within matched tumors predict tumor recurrence. A, Ascites was collected from patients with HGSOC and was used for multiplex ELISA from the indicated cytokines. Red bars, mean. Left y-axis include = IFNγ, IL10, IL12p70, IL13, IL1β, IL2, IL4, and TNFα. The right y-axis includes IL6 and IL8. B, Correlation between the indicated cytokines. Red, 1 or strong correlation; blue, 0 or weak correlation (n = 39). Outline boxes indicate Spearman correlation P > 0.01. C, Correlation between time to disease recurrence and IL10 concentration. HR calculated by stratifying data on the basis of the median IL10 concentration and using the Mantel–Haenszel test. D, Correlation between time to disease recurrence and IL6 concentration. Red dots, short time to recurrence with low IL6 concentrations. HR calculated by stratifying data on the basis of the median IL6 concentration and using the Mantel–Haenszel test. E, Correlation of IL6 and IER3 expression from pre- and post-NACT samples. F,IL6 and IER3 expression in 489 HGSOC tumors within TCGA. G, Change in IL6 expression in pre- and post-treated tumors. Connecting lines indicate matching tumors. H, Change in IER3 expression in pre- and post-treated tumors. Connecting lines indicate matching tumors. I, Matched pre- and post-NACT tumors (n = 12) from NanoString analysis were used for multispectral IHC against IER3 and CK. The H-score for IER3 is graphed for CK+, CK, and total cells. J,IER3 and IL6 z-scores derived from ovarian cancer cell lines (n = 50) in the cBioPortal CCLE dataset (accessed in June 2020). A combined z-score of greater than 2 was set to identify IER3/IL6 high (blue dots) and low (yellow) expressing cell lines. K, Drug sensitivity was (AUC) assessed for IER3/IL6 high (blue dots) and low (yellow dots) expressing cell lines. Fisher LSD are from two-way ANOVA (no multiple correction). Red line, median AUC for group. L, siRNA-mediated IER3 knockdown in OVCAR4 and OVCAR8 cells. IER3 mRNA expression in siControl (siCtrl) and siIER3 cells. Internal control, GAPDH. Statistical analysis, unpaired t test. M, siCtrl and siIER3 OVCAR4 and OVCAR8 cells were dosed with cisplatin and imaged over 72 hours. The growth rate was calculated via linear regression for every conditions and a dose–response curve was graphed. Statistical analysis, nonlinear regression fit. Error bars, SD. N,IL6 and IER3 transcript counts per million of an olaparib sensitive (sens.) and resistant (res.) PEO1 cell line (GSE117765). Statistical analyses, unpaired t test (P value) and Padj, Benjamini–Hochberg.

Figure 3.

Cytokine profiling of primary patient-derived ascites and transcriptional changes within matched tumors predict tumor recurrence. A, Ascites was collected from patients with HGSOC and was used for multiplex ELISA from the indicated cytokines. Red bars, mean. Left y-axis include = IFNγ, IL10, IL12p70, IL13, IL1β, IL2, IL4, and TNFα. The right y-axis includes IL6 and IL8. B, Correlation between the indicated cytokines. Red, 1 or strong correlation; blue, 0 or weak correlation (n = 39). Outline boxes indicate Spearman correlation P > 0.01. C, Correlation between time to disease recurrence and IL10 concentration. HR calculated by stratifying data on the basis of the median IL10 concentration and using the Mantel–Haenszel test. D, Correlation between time to disease recurrence and IL6 concentration. Red dots, short time to recurrence with low IL6 concentrations. HR calculated by stratifying data on the basis of the median IL6 concentration and using the Mantel–Haenszel test. E, Correlation of IL6 and IER3 expression from pre- and post-NACT samples. F,IL6 and IER3 expression in 489 HGSOC tumors within TCGA. G, Change in IL6 expression in pre- and post-treated tumors. Connecting lines indicate matching tumors. H, Change in IER3 expression in pre- and post-treated tumors. Connecting lines indicate matching tumors. I, Matched pre- and post-NACT tumors (n = 12) from NanoString analysis were used for multispectral IHC against IER3 and CK. The H-score for IER3 is graphed for CK+, CK, and total cells. J,IER3 and IL6 z-scores derived from ovarian cancer cell lines (n = 50) in the cBioPortal CCLE dataset (accessed in June 2020). A combined z-score of greater than 2 was set to identify IER3/IL6 high (blue dots) and low (yellow) expressing cell lines. K, Drug sensitivity was (AUC) assessed for IER3/IL6 high (blue dots) and low (yellow dots) expressing cell lines. Fisher LSD are from two-way ANOVA (no multiple correction). Red line, median AUC for group. L, siRNA-mediated IER3 knockdown in OVCAR4 and OVCAR8 cells. IER3 mRNA expression in siControl (siCtrl) and siIER3 cells. Internal control, GAPDH. Statistical analysis, unpaired t test. M, siCtrl and siIER3 OVCAR4 and OVCAR8 cells were dosed with cisplatin and imaged over 72 hours. The growth rate was calculated via linear regression for every conditions and a dose–response curve was graphed. Statistical analysis, nonlinear regression fit. Error bars, SD. N,IL6 and IER3 transcript counts per million of an olaparib sensitive (sens.) and resistant (res.) PEO1 cell line (GSE117765). Statistical analyses, unpaired t test (P value) and Padj, Benjamini–Hochberg.

Close modal

Integration of the transcriptomic, proteomic, and cytokine profiling platforms

Transcriptomic and proteomic analysis of matched HGSOC samples pre- and post-NACT, coupled with cytokine profiling of pretreatment HGSOC ascites, suggest elevated cytokine and/or RTK signaling prior to or in response to chemotherapy may drive resistance. In particular, our data implicate IL6, consistent with previous studies (10, 26). However, we noted that while IL6 protein levels were associated with time to recurrence, a similar trend was not observed with IL6 mRNA expression. On the basis of this, we hypothesized that IL6 mRNA or protein levels alone may not sufficiently define activation of downstream cytokine signaling. Therefore, we compared the 18 genes that correlated with time to recurrence (Table 1) with IL6 expression within the NanoString dataset and The Cancer Genome Atlas (TCGA) ovarian cancer dataset (Supplementary Table S4). Between the two datasets, we observed that expression of the IER3 was the most positively correlated (Fig. 3E and F) with IL6. Furthermore, Fig. 3G demonstrates that the magnitude of IL6 induction following chemotherapy modestly correlated with time to disease recurrence. In contrast, the magnitude of IER3 induction post-NACT strongly correlated with time to disease recurrence (Fig. 3H). Using IHC with the NanoString matched tumor tissues, we examined IER3 expression in parallel with an epithelial tumor cell marker (cytokeratin, CK). Changes in IER3 protein expression were consistent with the changes in transcript levels and IER3 was predominantly expressed in CK+ cells (Fig. 3I).

Stress induces IER3, which serves an antiapoptotic function in a range of tumor types, and high IER3 is typically associated with disease progression and poor prognosis (28–30). We hypothesized that degree of IER3 induction compared with untreated tumors in ovarian cancer would be associated with therapy resistance. Using the cBioPortal CCLE dataset, we assessed the relationship of IER3/IL6 expression with response to chemotherapy across 50 ovarian cancer lines (24). IL6 and IER3 gene expression were strongly correlated, and using a combined IL6 and IER3 z-score, we identified 17 cell lines with high IL6/IER3 expression (Fig. 3J). Ovarian cancer cell lines with high IL6/IER3 were not resistant to single-agent cisplatin or docetaxel (Fig. 3K); responses to these agents in CCLE were minimal/modest. However, high IL6/IER3 expression did significantly correlate with resistance to second-line chemotherapies, doxorubicin and etoposide (Fig. 3K). To directly assess IER3's contribution to chemotherapy response, IER3 was knocked down in OVCAR4 and OVCAR8 HGSOC cells lines and the inhibitory concentration (IC50) for cisplatin was determined. IER3 knockdown was approximately 50% in both cell lines (Fig. 3L). In OVCAR4 cells, IER3 knockdown did not alter cisplatin sensitivity. In contrast, loss of IER3 in OVCAR8 cells significantly reduced the IC50 of cisplatin by 4.13-fold (Fig. 3M). Next, we examined IER3 expression in olaparib-resistant HGSOC cells (22, 25). PARP inhibitors (e.g., olaparib) are FDA approved for HGSOC and their use in the clinic is expanding. Both IL6 and IER3 were significantly upregulated, 3.7-fold and 55-fold, in the olaparib-resistant versus -sensitive (parental) cells, respectively (Fig. 3N). These functional data strongly support that IL6/IER3 signaling can mediate resistance to therapy in ovarian cancer.

IER3 in the immune tumor microenvironment

To examine the relationship between IER3 and the tumor microenvironment, we examined an independent cohort of HGSOC tumors by TMA, including pre- and post-NACT primary tumors and recurrent tumors (TMA; n = 119; ref. 22). When comparing IER3 levels by IHC (H-score) in pre- and post-NACT tumors and recurrent tumors, IER3 was specifically elevated in post-NACT tumors (P = 0.0332; Fig. 4A and B). On the basis of the median IER3 H-score, tumors were identified as having low (n = 80) or high (n = 39) IER3 expression. Because tumor infiltration of T cells is a positive prognostic indicator (31, 32), we next used multispectral IHC to identify correlations between IER3 expression and immune cells in the tumor microenvironment. Specifically, we evaluated IER3 expression in relation to tumor-associated CD4+ T cells, CD8+ T cells, CD4+ FOXP3+ Tregs, CD68+ macrophages, pan-CK+ tumor cells, and Granzyme B+ T cells (Fig. 4C). While tumor-associated macrophages and Tregs did not correlate with IER3 expression (Fig. 4D and E), low IER3 expression significantly correlated with tumor-associated CD4+ T cells (P = 0.0087), CD8+ T cells (P = 0.0502), and CD8/granzyme B+ T cells (P = 0.0386; Fig. 4FH). We determined that in the tumor compartment, IER3 is expressed in CD4 and CD8 T cells and CK+ cells with minimal expression in B cells and macrophages (Fig. 4I). Albeit IER3 expression was lower, the stromal compartment demonstrated a similar expression pattern to the tumor regions (Fig. 4J). Together, these data suggest that regulation of IER3 and the composition of tumor immune microenvironment are interconnected.

Figure 4.

Evaluating associations between IER3 expression and the tumor microenvironment. A, Representative images of IER3 IHC with associated histology score (H-score). Scale bars, 100 μm. B, IER3 histology score (H-score) for pre- (n = 85) and post-NACT (n = 19) and recurrent (n = 24) human HGSOC tumors (P value, one way ANOVA). C, Representative images of multispectral IHC of HGSOC tumors. Scale bars, 100 μm. On the basis of the median IER3 expression tumors were defined as “low” (n = 80) or “high” (n = 39). The percentage of tumor-associated macrophages (D), Tregs (E), CD4+ T cells (F), CD8+ T cells (G), and CD8/Granzyme B (Grb)+ T cells (H) was correlated to IER3 expression. Statistical analysis, unpaired t test with Welch correction. Error bars, SEM. Note: difference in n between G and H–L due to missing tissue cores. I, Multispectral analysis of IER3, CK+, CD19+ B cells, CD8 T cells, CD4 T cell, and CD68+ macrophages (macro.) J, IER3 expression in indicated cells from the tumor compartment (CK+) and stromal compartment. Models for signal integration (K) and magnitude of signal intensity (L).

Figure 4.

Evaluating associations between IER3 expression and the tumor microenvironment. A, Representative images of IER3 IHC with associated histology score (H-score). Scale bars, 100 μm. B, IER3 histology score (H-score) for pre- (n = 85) and post-NACT (n = 19) and recurrent (n = 24) human HGSOC tumors (P value, one way ANOVA). C, Representative images of multispectral IHC of HGSOC tumors. Scale bars, 100 μm. On the basis of the median IER3 expression tumors were defined as “low” (n = 80) or “high” (n = 39). The percentage of tumor-associated macrophages (D), Tregs (E), CD4+ T cells (F), CD8+ T cells (G), and CD8/Granzyme B (Grb)+ T cells (H) was correlated to IER3 expression. Statistical analysis, unpaired t test with Welch correction. Error bars, SEM. Note: difference in n between G and H–L due to missing tissue cores. I, Multispectral analysis of IER3, CK+, CD19+ B cells, CD8 T cells, CD4 T cell, and CD68+ macrophages (macro.) J, IER3 expression in indicated cells from the tumor compartment (CK+) and stromal compartment. Models for signal integration (K) and magnitude of signal intensity (L).

Close modal

Understanding how chemotherapy remodels the tumor and immune microenvironment is critical in predicting disease progression. To address this in HGSOC biology, we utilized tumor tissue from patients that had received NACT, and evaluated treatment-induced changes in the tumor microenvironment. Chemotherapy led to activation of several RTKs and associated downstream signaling pathways associated with stress response and cell survival. Supporting the transcriptional and signaling network identified in our integrated analyses, expression levels of multiple CEBP/β target genes, in particular IER3, predicted disease recurrence. Taken together, chemotherapy in HGSOC is paradoxically involved in promoting a protumor growth microenvironment. The implication is that dampening the chemotherapy-dependent inflammation could extend disease-free interval.

Chemotherapy results in significant tissue and cellular stress, which can lead to the activation of a multitude of effectors, including CEBP/β. The regulation of CEBP/β activity is highly complex, with multiple inhibitory and activating pathways (reviewed in ref. 27). Importantly, our analyses identified a positive feedback loop consistent with CEBP/β activation and upregulation of target genes. As depicted in Fig. 4K, although the response to surgery and chemotherapy is complex, other studies and our data suggest a convergence on CEBP/β activity. CEBP/β is a prognostic factor and is linked to epigenetic regulation of multiple drug resistance genes (33). On the basis of CEBP/β activity in HGSOC, combination therapies that drive CEBP/β dependence and inhibit CEBP/β transcriptional function are potential approaches to improve the antitumor effect of chemotherapy (34–38).

By characterizing the tumor microenvironment over the course of treatment, unique information was obtained regarding the extent and magnitude of specific molecular changes. Previous studies that evaluated HGSOC tumors pre- and postchemotherapy, also noted that chemotherapy promoted an inflammatory response, but the immunosuppressive environment prevented activation of antitumor immunity, suggesting a balance between inflammation and antitumor immunity. For instance, IL6 and IL10 cooperate to establish an immunosuppressive environment through expansion of myeloid-derived suppressor cells (39). Interestingly, in the ascites samples prior to chemotherapy, there was a subset of patients that had low IL6 and IL10 and a short disease-free interval, suggesting an immunologically “cold” tumor microenvironment. Therefore, understanding the chemotherapy-induced effects that contribute to the maintenance of an immunosuppressive environment is critical. For instance, while IL6 was upregulated in all of the treated tumors, IER3 was not, suggesting that tumors that upregulated IER3 are those competent to respond to IL6. In recurrent HGSOC tumors, IER3 expression was reduced compared with pre- and post-NACT, which is likely a reflection of IER3 regulation during chemotherapy. Consistent with differential IER3 induction, the magnitude of chemotherapy-induced IER3 upregulation was a superior predictor of disease recurrence compared with IL6 (Fig. 4L). IER3 regulates extrinsic apoptosis via inhibiting Fas and TNFα receptor activity (40). IER3 also potentiates MAPK signaling to attenuate apoptosis (41). In osteosarcoma, knockdown of IER3 resensitized cells to chemotherapy, an alkylating agent (42). Furthermore, in an inflammation-induced colon cancer model, Ier3 knockout led to an exacerbated inflammatory environment (43). In in vitro experiments with HGSOC cell lines, IER3 knockdown had a variable effect on cisplatin response, which could be attributed to several factors including the lack of a heterogeneous microenvironment.

We observed significant correlations between IER3 expression and the tumor immune microenvironment, specifically that low IER3 correlated with increased CD4 and CD8 T-cell infiltration. Because both IL6 and IER3 are key CEBP/β target genes, the relationship between IL6 and IER3 suggests that increased levels of multiple genes activated by inflammatory/stress signaling post-NACT more accurately reflect pathway activation and therapy resistance. Although there is currently no pharmacologic approach to target IER3, several clinical trials have investigated targeting IL6 (e.g., NCT00841191). While IL6 likely plays a direct role in tumor progression, our integrated analyses suggest that tumor and immune microenvironments more poised to integrate and manage stress signals tend to be associated with a more aggressive and therapy-resistant disease.

There are several limitations to this study. The analysis only included six patients that qualified for NACT, which inherently suggests an advanced disease stage. HGSOC frequently disseminates throughout the peritoneal cavity, which results in multiple sites of tumor colonization, and unique molecular signatures are observed on the basis of tumor sites (44). Accounting for these limitations, our data and conclusions are still consistent and in close agreement with previous reports, and importantly, our study further expands on these reports by demonstrating how dynamic aspects of the microenvironment can be exploited to improve patient outcomes. Specifically, we identified a novel, targetable IL6/IER3 signaling axis induced by chemotherapy that promotes inflammatory signaling that may inform subsequent therapy or help predict time to recurrence.

B.G. Bitler reports grants from NIH/NCI (R00CA194318-03), DOD (OC170228), and The University of Colorado Cancer Center Developmental Therapeutics Program during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

K.R. Jordan: Conceptualization, resources, formal analysis, funding acquisition, investigation, methodology, writing-original draft, writing-review and editing. M.J. Sikora: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing-original draft, writing-review and editing. J.E. Slansky: Conceptualization, resources, data curation, supervision, methodology. A. Minic: Data curation, formal analysis, methodology. J.K. Richer: Conceptualization, data curation, writing-review and editing. M.R. Moroney: Investigation, writing-original draft. J. Hu: Formal analysis, investigation, writing-review and editing. R.J. Wolsky: Formal analysis, supervision, visualization, pathological analysis. Z.L. Watson: Data curation, investigation, writing-review and editing. T.M. Yamamoto: Data curation, investigation. J.C. Costello: Conceptualization, software, methodology. A. Clauset: Conceptualization, software, investigation, visualization. K. Behbakht: Conceptualization, resources, funding acquisition, validation, investigation, project administration. T.R. Kumar: Conceptualization, funding acquisition. B.G. Bitler: Conceptualization, resources, data curation, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

We acknowledge Dr. Carol Goldstein for her contribution in the fight against ovarian cancer. We acknowledge Dr. Philip Owens for his assistance with the NanoString platform. We thank the Human Immune Monitoring Shared Resource of the University of Colorado Human Immunology and Immunotherapy Initiative for their expert assistance with the NanoString and Vectra platforms. We acknowledge funding from The University of Colorado OB/GYN Academic Enrichment Fund (to K. Behbakht and T.R. Kumar), Libations for Life (to K. Behbakht and T.R. Kumar), and The University of Colorado Cancer Center Developmental Therapeutics Program (to B.G. Bitler and K. Behbakht). Z.L. Watson was supported by Cancer League of Colorado research grant 193527-ZW and Colorado Clinical & Translational Sciences Institute CO-Pilot grant CO-J-20-006. This work was supported by grants from the NIH/NCI (R00CA194318-03 to B.G. Bitler; R00CA193734 to M.J. Sikora; and U01CA231978 to J.C. Costello), and Department of Defense awards (OC130212 to J.K. Richer and OC170228 to B.G. Bitler). K. Behbakht was also supported by the Emily McClintock-Addlesperger Endowed Chair in Ovarian Cancer Research. The Functional Proteomics RPPA Core facility was supported by MD Anderson Cancer Center Support grant # 5 P30 CA016672-40. Support of Core Facilities was provided by the University of Colorado Cancer Center Support grant (P30CA046934).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Siegel
RL
,
Miller
KD
,
Jemal
A
. 
Cancer statistics, 2019
.
CA Cancer J Clin
2019
;
69
:
7
34
.
2.
Fagotti
A
,
Ferrandina
G
,
Fanfani
F
,
Ercoli
A
,
Lorusso
D
,
Rossi
M
, et al
A laparoscopy-based score to predict surgical outcome in patients with advanced ovarian carcinoma: a pilot study
.
Ann Surg Oncol
2006
;
13
:
1156
61
.
3.
Petrillo
M
,
Vizzielli
G
,
Fanfani
F
,
Gallotta
V
,
Cosentino
F
,
Chiantera
V
, et al
Definition of a dynamic laparoscopic model for the prediction of incomplete cytoreduction in advanced epithelial ovarian cancer: proof of a concept
.
Gynecol Oncol
2015
;
139
:
5
9
.
4.
Bian
C
,
Yao
K
,
Li
L
,
Yi
T
,
Zhao
X
. 
Primary debulking surgery vs. neoadjuvant chemotherapy followed by interval debulking surgery for patients with advanced ovarian cancer
.
Arch Gynecol Obstet
2016
;
293
:
163
8
.
5.
May
T
,
Comeau
R
,
Sun
P
,
Kotsopoulos
J
,
Narod
SA
,
Rosen
B
, et al
A comparison of survival outcomes in advanced serous ovarian cancer patients treated with primary debulking surgery versus neoadjuvant chemotherapy
.
Int J Gynecol Cancer
2017
;
27
:
668
74
.
6.
Mueller
JJ
,
Zhou
QC
,
Iasonos
A
,
O'Cearbhaill
RE
,
Alvi
FA
,
El Haraki
A
, et al
Neoadjuvant chemotherapy and primary debulking surgery utilization for advanced-stage ovarian cancer at a comprehensive cancer center
.
Gynecol Oncol
2016
;
140
:
436
42
.
7.
Bohm
S
,
Montfort
A
,
Pearce
OM
,
Topping
J
,
Chakravarty
P
,
Everitt
GL
, et al
Neoadjuvant chemotherapy modulates the immune microenvironment in metastases of tubo-ovarian high-grade serous carcinoma
.
Clin Cancer Res
2016
;
22
:
3025
36
.
8.
Lo
CS
,
Sanii
S
,
Kroeger
DR
,
Milne
K
,
Talhouk
A
,
Chiu
DS
, et al
Neoadjuvant chemotherapy of ovarian cancer results in three patterns of tumor-infiltrating lymphocyte response with distinct implications for immunotherapy
.
Clin Cancer Res
2017
;
23
:
925
34
.
9.
Chudecka-Glaz
AM
,
Cymbaluk-Ploska
AA
,
Menkiszak
JL
,
Pius-Sadowska
E
,
Machalinski
BB
,
Sompolska-Rzechula
A
, et al
Assessment of selected cytokines, proteins, and growth factors in the peritoneal fluid of patients with ovarian cancer and benign gynecological conditions
.
Onco Targets Ther
2015
;
8
:
471
85
.
10.
Cohen
S
,
Bruchim
I
,
Graiver
D
,
Evron
Z
,
Oron-Karni
V
,
Pasmanik-Chor
M
, et al
Platinum-resistance in ovarian cancer cells is mediated by IL-6 secretion via the increased expression of its target cIAP-2
.
J Mol Med
2013
;
91
:
357
68
.
11.
Kato
T
,
Noma
K
,
Ohara
T
,
Kashima
H
,
Katsura
Y
,
Sato
H
, et al
Cancer-associated fibroblasts affect intratumoral CD8(+) and FoxP3(+) T cells via IL6 in the tumor microenvironment
.
Clin Cancer Res
2018
;
24
:
4820
33
.
12.
Lane
D
,
Matte
I
,
Rancourt
C
,
Piche
A
. 
Prognostic significance of IL-6 and IL-8 ascites levels in ovarian cancer patients
.
BMC Cancer
2011
;
11
:
210
.
13.
Moradi
MM
,
Carson
LF
,
Weinberg
B
,
Haney
AF
,
Twiggs
LB
,
Ramakrishnan
S
. 
Serum and ascitic fluid levels of interleukin-1, interleukin-6, and tumor necrosis factor-alpha in patients with ovarian epithelial cancer
.
Cancer
1993
;
72
:
2433
40
.
14.
Fernandez
NF
,
Gundersen
GW
,
Rahman
A
,
Grimes
ML
,
Rikova
K
,
Hornbeck
P
, et al
Clustergrammer, a web-based heatmap visualization and analysis tool for high-dimensional biological data
.
Sci Data
2017
;
4
:
170151
.
15.
Cristescu
R
,
Mogg
R
,
Ayers
M
,
Albright
A
,
Murphy
E
,
Yearley
J
, et al
Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy
.
Erratum. Science
2019
1
;363(6430):
eaax1384
.
doi: 10.1126/science.aax1384
.
16.
Ayers
M
,
Lunceford
J
,
Nebozhyn
M
,
Murphy
E
,
Loboda
A
,
Kaufman
DR
, et al
IFN-gamma-related mRNA profile predicts clinical response to PD-1 blockade
.
J Clin Invest
2017
;
127
:
2930
40
.
17.
Cristescu
R
,
Mogg
R
,
Ayers
M
,
Albright
A
,
Murphy
E
,
Yearley
J
, et al
Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy
.
Science
2018
;
362
.
18.
Danaher
P
,
Warren
S
,
Lu
R
,
Samayoa
J
,
Sullivan
A
,
Pekker
I
, et al
Pan-cancer adaptive immune resistance as defined by the tumor inflammation signature (TIS): results from The Cancer Genome Atlas (TCGA)
.
J Immunother Cancer
2018
;
6
:
63
.
19.
Voorwerk
L
,
Slagter
M
,
Horlings
HM
,
Sikorska
K
,
van de Vijver
KK
,
de Maaker
M
, et al
Immune induction strategies in metastatic triple-negative breast cancer to enhance the sensitivity to PD-1 blockade: the TONIC trial
.
Nat Med
2019
;
25
:
920
8
.
20.
Tibes
R
,
Qiu
Y
,
Lu
Y
,
Hennessy
B
,
Andreeff
M
,
Mills
GB
, et al
Reverse phase protein array: validation of a novel proteomic technology and utility for analysis of primary leukemia specimens and hematopoietic stem cells
.
Mol Cancer Ther
2006
;
5
:
2512
21
.
21.
Park
CY
,
Krishnan
A
,
Zhu
Q
,
Wong
AK
,
Lee
YS
,
Troyanskaya
OG
. 
Tissue-aware data integration approach for the inference of pathway interactions in metazoan organisms
.
Bioinformatics
2015
;
31
:
1093
101
.
22.
Watson
ZL
,
Yamamoto
TM
,
McMellen
A
,
Kim
H
,
Hughes
CJ
,
Wheeler
LJ
, et al
Histone methyltransferases EHMT1 and EHMT2 (GLP/G9A) maintain PARP inhibitor resistance in high-grade serous ovarian carcinoma
.
Clin Epigenetics
2019
;
11
:
165
.
23.
Lang
J
,
Capasso
A
,
Jordan
KR
,
French
JD
,
Kar
A
,
Bagby
SM
, et al
Development of an adrenocortical cancer humanized mouse model to characterize anti-PD1 effects on tumor microenvironment
.
J Clin Endocrinol Metab
2020
;
105
:
dgz014
.
24.
Ghandi
M
,
Huang
FW
,
Jane-Valbuena
J
,
Kryukov
GV
,
Lo
CC
,
McDonald
ER
 III
, et al
Next-generation characterization of the Cancer Cell Line Encyclopedia
.
Nature
2019
;
569
:
503
8
.
25.
Yamamoto
TM
,
McMellen
A
,
Watson
ZL
,
Aguilera
J
,
Ferguson
R
,
Nurmemmedov
E
, et al
Activation of Wnt signaling promotes olaparib resistant ovarian cancer
.
Mol Carcinog
2019
;
58
:
1770
82
.
26.
Wang
Y
,
Zong
X
,
Mitra
S
,
Mitra
AK
,
Matei
D
,
Nephew
KP
. 
IL-6 mediates platinum-induced enrichment of ovarian cancer stem cells
.
JCI Insight
2018
;
3
:
e122360
.
27.
Huber
R
,
Pietsch
D
,
Panterodt
T
,
Brand
K
. 
Regulation of C/EBPbeta and resulting functions in cells of the monocytic lineage
.
Cell Signal
2012
;
24
:
1287
96
.
28.
Liu
Z
,
Wang
XM
,
Jia
TF
,
Zhai
Y
,
Sun
LY
,
Cheng
YP
, et al
Expression of IER3 in primary hepatocarcinoma: correlation with clinicopathological parameters
.
Asian Pac J Cancer Prev
2015
;
16
:
679
82
.
29.
Wu
MX
. 
Roles of the stress-induced gene IEX-1 in regulation of cell death and oncogenesis
.
Apoptosis
2003
;
8
:
11
8
.
30.
Ye
J
,
Zhang
Y
,
Cai
Z
,
Jiang
M
,
Li
B
,
Chen
G
, et al
Increased expression of immediate early response gene 3 protein promotes aggressive progression and predicts poor prognosis in human bladder cancer
.
BMC Urol
2018
;
18
:
82
.
31.
Nielsen
JS
,
Sahota
RA
,
Milne
K
,
Kost
SE
,
Nesslinger
NJ
,
Watson
PH
, et al
CD20+ tumor-infiltrating lymphocytes have an atypical CD27- memory phenotype and together with CD8+ T cells promote favorable prognosis in ovarian cancer
.
Clin Cancer Res
2012
;
18
:
3281
92
.
32.
Zhang
L
,
Conejo-Garcia
JR
,
Katsaros
D
,
Gimotty
PA
,
Massobrio
M
,
Regnani
G
, et al
Intratumoral T cells, recurrence, and survival in epithelial ovarian cancer
.
N Engl J Med
2003
;
348
:
203
13
.
33.
Liu
D
,
Zhang
XX
,
Li
MC
,
Cao
CH
,
Wan
DY
,
Xi
BX
, et al
C/EBPbeta enhances platinum resistance of ovarian cancer cells by reprogramming H3K79 methylation
.
Nat Commun
2018
;
9
:
1739
.
34.
Coward
J
,
Kulbe
H
,
Chakravarty
P
,
Leader
D
,
Vassileva
V
,
Leinster
DA
, et al
Interleukin-6 as a therapeutic target in human ovarian cancer
.
Clin Cancer Res
2011
;
17
:
6083
96
.
35.
Duan
Z
,
Foster
R
,
Bell
DA
,
Mahoney
J
,
Wolak
K
,
Vaidya
A
, et al
Signal transducers and activators of transcription 3 pathway activation in drug-resistant ovarian cancer
.
Clin Cancer Res
2006
;
12
:
5055
63
.
36.
Kandala
PK
,
Srivastava
SK
. 
Diindolylmethane suppresses ovarian cancer growth and potentiates the effect of cisplatin in tumor mouse model by targeting signal transducer and activator of transcription 3 (STAT3)
.
BMC Med
2012
;
10
:
9
.
37.
Momeny
M
,
Yousefi
H
,
Eyvani
H
,
Moghaddaskho
F
,
Salehi
A
,
Esmaeili
F
, et al
Blockade of nuclear factor-kappaB (NF-kappaB) pathway inhibits growth and induces apoptosis in chemoresistant ovarian carcinoma cells
.
Int J Biochem Cell Biol
2018
;
99
:
1
9
.
38.
Rosell
R
,
Robinet
G
,
Szczesna
A
,
Ramlau
R
,
Constenla
M
,
Mennecier
BC
, et al
Randomized phase II study of cetuximab plus cisplatin/vinorelbine compared with cisplatin/vinorelbine alone as first-line therapy in EGFR-expressing advanced non-small-cell lung cancer
.
Ann Oncol
2008
;
19
:
362
9
.
39.
Wu
L
,
Deng
Z
,
Peng
Y
,
Han
L
,
Liu
J
,
Wang
L
, et al
Ascites-derived IL-6 and IL-10 synergistically expand CD14(+)HLA-DR(-/low) myeloid-derived suppressor cells in ovarian cancer patients
.
Oncotarget
2017
;
8
:
76843
56
.
40.
Morinobu
A
,
Tanaka
S
,
Nishimura
K
,
Takahashi
S
,
Kageyama
G
,
Miura
Y
, et al
Expression and functions of immediate early response gene X-1 (IEX-1) in rheumatoid arthritis synovial fibroblasts
.
PLoS One
2016
;
11
:
e0164350
.
41.
Garcia
J
,
Ye
Y
,
Arranz
V
,
Letourneux
C
,
Pezeron
G
,
Porteu
F
. 
IEX-1: a new ERK substrate involved in both ERK survival activity and ERK activation
.
EMBO J
2002
;
21
:
5151
63
.
42.
Bruheim
S
,
Xi
Y
,
Ju
J
,
Fodstad
O
. 
Gene expression profiles classify human osteosarcoma xenografts according to sensitivity to doxorubicin, cisplatin, and ifosfamide
.
Clin Cancer Res
2009
;
15
:
7161
9
.
43.
Sina
C
,
Arlt
A
,
Gavrilova
O
,
Midtling
E
,
Kruse
ML
,
Muerkoster
SS
, et al
Ablation of gly96/immediate early gene-X1 (gly96/iex-1) aggravates DSS-induced colitis in mice: role for gly96/iex-1 in the regulation of NF-kappaB
.
Inflamm Bowel Dis
2010
;
16
:
320
31
.
44.
Zhang
AW
,
McPherson
A
,
Milne
K
,
Kroeger
DR
,
Hamilton
PT
,
Miranda
A
, et al
Interfaces of malignant and immunologic clonal dynamics in ovarian cancer
.
Cell
2018
;
173
:
1755
69
.