A total of 31 HGSOC patients were included in this retrospective study. Patients were selected based on pathology and clinical diagnosis of HGSOC, which by definition refers to patients with serous pathology with grade 3 disease or greater. In order to target our investigation further, we focused only on stage III and IV disease, since HGSOC is most frequently diagnosed at a late stage. Matched formalin-fixed, paraffin-embedded (FFPE) tumor tissues from treatment naive biopsies and interval debulking surgery following exposure to NACT were obtained from each patient, with associated clinical information, for a total of 62 samples. All experiments were performed in accordance with the relevant guidelines and regulations of the Women and Infants Hospital Institutional Review Board committee. All patients received frontline carboplatin and paclitaxel, although some patients received additional therapies. Treatment regimens, along with complete patient clinical information is listed in Table 1 .

Pre-treatment and post-treatment cases were reviewed to select an optimal FFPE tissue block for each case. An optimal pre-treatment block contained maximum tumor cellularity (minimum of 20%) with no lymph node tissue present. An optimal post-treatment block contained maximum tumor cellularity (minimum of 20%) with no lymph node tissue present, and with evidence of a lymphocytic response. For each block, ten unstained sections of 4-5 µm thickness were cut and placed on Avantix uncharged slides. An eleventh slide was cut at 4-5 µm, stained with hematoxylin and eosin, cover-slipped and reviewed to confirm the appropriate tissue was still present. FFPE sections were scraped into tubes for RNA isolation using an RNeasy FFPE Kit (Qiagen, 73504) according to manufacturer’s instructions. RNA concentration and quality were measured by NanoDrop and 50 ng RNA was used for analysis with the nCounter PanCancer IO 360™ Panel (NanoString, XT-CSO-HIO360-12). The reporter code set and capture probe set tubes were removed from -80°C and thawed on ice prior to mixing by tapping and pulse centrifugation. 70 µl hybridization buffer was added to the tube containing the reporter probe set and mixed by tapping, followed by pulse centrifugation. 8 µl of the diluted reporter probe was added to a PCR tube. Each RNA sample was diluted in water to 25 ng/µl and 2 µl RNA was mixed with 3 µl water to 5 µl final volume for 50 ng total input per sample. 5 µl RNA sample was added to each tube containing the reporter probe set and mixed by tapping, followed by pulse centrifugation. 2 µl capture probe set was added to each tube, mixed by tapping, and the sample was collected by pulse centrifugation and then immediately placed in a pre-heated PCR machine, with the lid set to 70°C and the block set to 65°C. The hybridization was performed for 18 h at 65°C. Samples were cooled to 4°C and any condensation was collected by pulse centrifugation. Each sample was diluted with 18 µl hybridization buffer. An nCounter Sprint cartridge was calibrated to room temperature for 15 min prior to injection of 33 µl of each sample into one of the 12 injection ports of the cartridge, followed by the introduction of an air seal. The sample injection cartridge was sealed with provided tape and the reagent supply ports were unsealed prior to loading on the nCounter Sprint profiler. The analysis run was started immediately after loading. The resulting data file in RCC format was used for data analysis. RCC files were deposited in NCBI’s Gene Expression Omnibus (GEO) (25) and are accessible through GEO Series accession number GSE201600 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE201600).

PEA1/PEA2 and PEO1/PEO4 cells were obtained from Millipore Sigma and cultured in RPMI 1640 supplemented with 2 mM Glutamine, 2 mM Sodium Pyruvate, and 10% Fetal Bovine Serum (FBS). OVCAR4 cells were also purchased from Millipore Sigma and cultured in RPMI 1640 supplemented with 2 mM Glutamine, 0.25 U/ml Insulin (Millipore Sigma, 407709), and 10% FBS. OVCAR8 cells were originally purchased from American Type Culture Collection (ATCC) and cultured in DMEM with 10% FBS. OV90 cells were obtained from ATCC and cultured in a 1:1 mixture of MCDB 105 medium containing a final concentration of 1.5 g/L sodium bicarbonate and Medium 199 containing a final concentration of 2.2 g/L sodium bicarbonate, supplemented with 15% FBS. All cells were cultured in 1% penicillin/streptomycin and kept in a 37 °C/5% CO₂ humidified chamber. All cell lines were treated with 100 µM carboplatin (Santa Cruz Biotechnology, CAS 4157.5-94-4) and 10 nM paclitaxel (NIH Developmental Therapeutics Drug Cancer Panel) in combination, with control cells treated with DMSO (Sigma Aldrich, D54879) for 48 h. For chemotherapy treated cells, cells were spun down in order to retain detached and detaching cells.

RNA was isolated using Trizol extraction/LiCl high salt precipitation. Total RNA (500 ng) was reverse transcribed into cDNA using the iScript cDNA Synthesis Kit (Bio-Rad, 1708890) according the manufacturer’s protocol. All quantitative PCR was performed in triplicate by loading 1 μl of cDNA reaction, 5 μM primers, 10 μl of SYBR Green (New England Biolabs, M3003E), and 5 μl of RNAse-free water to each well. For validated BioRad primers a final 1X concentration was used with 10 μl of SYBR green (New England Biolabs, M3003E), and 8 μl of RNAse-free water added to each well. All samples were run on an ABI 7500 Fast-Real Time PCR System, and data was analyzed using the ΔΔ Ct method, with relative expression levels normalized to 18s rRNA. Validated primers were purchased from realtimeprimers.com (ATF3, NFATC2, DUSP1) or Bio-Rad (AREG, SGK1). Custom primer sequences (Invitrogen) are as follows:

18S rRNA—F-CCGCGGTTCTATTTTGTTGG

18S rRNA—R-GGCGCTCCCTCTTAATCATG

FFPE human ovarian cancer tissue slides were baked for 2 hours at 65°C. Slides were then washed in SafeClear xylene substitute, 100, 95, 70% ethanol, deoxygenated water, and FTA Hemagglutination Buffer. Antigen retrieval was performed using Antigen Retrieval Solution (1X) (Vector Laboratories, H-3300) and heated to 95°C for 20 min. Slides were then blocked with 5% horse serum in FTA Hemagglutination buffer and incubated overnight in primary antibody diluted in FTA buffer with 5% horse serum at 4°C. Anti-rabbit IgG Dylight 488 secondary antibody (Vector Laboratories, DI-488, 1:1000) was then applied to slides followed by incubation in the dark at room temperature for 1 hour. Slides were washed between each step using FTA Hemagglutination buffer and cover-slipped with DAPI containing mounting medium (Vector Laboratories, H-1200). For HGSOC cell lines, cells were cultured in chamber slides, fixed with 4% paraformaldehyde for 20 min, permeabilized in 0.1% Triton-X for 5 min, blocked in phosphate buffered saline-Tween 20 (PBS-T) with 5% horse serum for 30 min, then incubated overnight with primary antibody in the same blocking solution. Slides were washed the next day with PBS then incubated for 1 h with secondary antibody in PBS-T, washed again, and finally coverslipped with mounting media with DAPI.

Primary antibodies and dilutions used were as follows:

ATF3 (Novus Biologicals, NBP1-85816, [1:50])

AREG (Proteintech, 16036-1-AP, [1:50])

CD8 (Origene, TA802079, [1:50])

Images were obtained from a Zeiss Axio Imager M1 and were acquired using diode lasers 402, 488, and 561. To obtain images for CD8+ cell counting, five randomly selected fields per sample were selected based on DAPI staining and acquired using a 40x objective. For AREG and ATF3, three randomly selected fields per sample were selected based on DAPI staining and acquired using a 20x objective. Each wavelength was acquired separately and an RGB image was created.

Image processing and analysis was performed using Image J. Image analysis was performed on grayscale 8-bit images that were thresholded for specific staining, and mean and maximum intensity, along with integrated optical density was calculated. Representative images were taken using a 40x objective.

The co-expression feature was utilized from U133 microarray (n=310) data obtained from the Ovarian Cancer, TCGA Firehose Legacy cohort at cBioPortal.org (26, 27).

The ovarian cancer Kaplan-Meier plotter was accessed at https://kmplot.com/analysis/index.php?p=service&cancer=ovar (28) to determine the association of ATF3 and EGR1 with progression free survival (PFS) in stage III-IV, grade 3 serous ovarian cancer, using upper quartile as a cutoff. The probe sets were selected based on the recommended best probe set defined by Jetset algorithm.

Data was analyzed in nSolver Advanced Analysis Software. Raw data was uploaded to nSolver for automated normalization, background subtraction, and quality control (QC) check. All samples passed QC. Paired pre- and post-NACT samples were used to construct two groups of patient data to which an unpaired t-test was run to generate the data in the volcano plot. By performing the analysis this way the variability seen in patient data is mitigated and more robust data surrounding differentially expressed genes linked to post treatment status can be visualized. Differential expression was determined with p-values and Benjamini-Yekutieli adjusted p-values. Pathway scores are generated in nSolver as a summarization of expression level changes of biologically related groups of genes. Pathway scores are derived from the first Principle Components Analysis (PCA) scores (1^st eigenvectors) for each sample based on the individual gene expression levels for all the measured genes within a specific pathway. Expression levels of multiple genes comprise this first PC, with some genes having higher weight applied depending on their contribution to data variability. Cell type profiling scores are generated for immune cell types using expression levels of cell-type specific mRNAs as described in the literature (29). The cell type score itself is calculated as the mean of the log₂ expression levels for all the probes included in the final calculation for that specific cell type. An additional level of quality control is by default performed, and markers that do not correlate with other cell type specific markers are discarded from the estimates of abundance. The software also utilizes a resampling technique to generate a significance level for confidence in the individual cell type scores, with lower p-values considered higher confidence. For our analysis, we excluded T_regs, NK cells, and Th1 cells since there was low confidence in the accuracy of those cell type scores.

Pathway scores, cell type scores, and log transformed, normalized mRNA expression data was exported and utilized for further analyses in GraphPad Prism. Significant differences in median pathway and cell type scores between matched pre- and post-NACT samples were determined using 2-tailed Wilcoxon matched-pairs signed rank test. Significant differences in median pathway and cell type scores between ≤12 mo PFI and >12 mo PFI were determined using unpaired Mann Whitney test. The relationship between PFI and cell types, pathways, or gene expression was determined in Prism using Cox proportional hazards regression, with hazard ratios, 95% confidence intervals, and p-values reported. Pearson r-values with 2-tailed p-values were also generated in GraphPad Prism. Pathway and cell type scores do not indicate abundance of one cell type/pathway relative to another, but changes in scores between comparison groups can be relatively compared.

Thirty-one HGSOC patients were analyzed in this study, with matched tissue pre- and post-NACT. All patients received frontline therapy with carboplatin and paclitaxel in the neoadjuvant setting, and some patients received additional therapies or were enrolled in clinical trials. Twenty-seven patients had recurred at time of analysis (4 non-recurrent), and 17 patients were deceased (14 living), with at least 12 months follow-up time ( Table 1 ).

In order to determine immunologic changes resulting from NACT exposure, we performed NanoString PanCancer IO360 analysis on patient samples pre- and post-NACT. DUSP1, EGR1, ATF3, SGK1, NFATC2, NFIL3, DUSP5, CDKN1A, CCL4, and CCL3/L1 are among the top upregulated differentially expressed genes (DEGs) post-NACT relative to pre-NACT, while CEP55, TPI1, HMGA1, RRM2, ANLN, CENPF, MK167, HELLS, H2AFX, and TNFSF12 are among the top downregulated DEGs post-NACT relative to pre-NACT ( Figures 1A, B ; Supplementary Tables 1 , 2 ). Furthermore, we examined NanoString pathway changes following NACT, finding significant decreases in “Cell Proliferation”, “DNA Damage Repair”, and “Epigenetics”, and significant increases in all other pathways except for “Cytotoxicity” and “Interferon Signaling” ( Figure 1C ; Supplementary Table 3 ).

We next examined gene changes associated with these individual pathways, which we grouped into three categories: 1) Proliferation and Stress Response ( Figure 2A ); 2) Pro-tumorigenic Signaling Pathways ( Figure 2B ); and 3) Immunoregulatory Pathways ( Figure 2C ). Pathways with fewer gene changes can be seen in Supplementary Figure 1 . Each of these pathways displayed some overlap with other pathways. Of note, CDKN1A gene, which codes for the tumor suppressor protein p21, was the top upregulated gene in the “Cell Proliferation” pathway, while most other proliferation related genes were downregulated. In addition, mitogen-activated protein kinase (MAPK) signaling genes such as DUSP1/2/5, and KIT were strongly upregulated, as were Phosphoinositide 3-kinase (PI3K) signaling genes SGK1 and IL6. Finally, in the Immunoregulatory Pathways category, robust increases were observed in cytokines such as CCL4, CCL3L1, and IL8, as well as transcription factors ATF3 and EGR1.

These findings suggest that chemotherapy may introduce changes that have both immunostimulatory and immunosuppressive effects. Our data demonstrate that chemotherapy regulates genes and pathways important for mediating its cytotoxic effects, such as through regulation of proliferation and DNA damage, but also demonstrates upregulation of genes and pathways that may impede its efficacy or contribute to the development of chemoresistance, such as through regulation of pro-tumorigenic pathways. In addition, we observed that two of the top regulated genes are EGR1 and ATF3, which are both master regulating transcription factors that may play a key role in mediating many chemotherapy-induced effects. Interestingly, while both EGR1 and ATF3 are well studied as stress-responsive transcription factors, ATF3 emerged as a gene of interest since it is a master transcription factor that regulates immunity and inflammation, but little is known about its role in ovarian cancer.

We examined a subset of top DEGs (ATF3, AREG, DUSP1, SGK1, and NFATC2) using HGSOC cell lines treated with carboplatin and paclitaxel chemotherapy for 48 h ( Figure 3 ). Cell lines examined included the matched platinum-sensitive/resistant pairs PEA1/PEA2 and PEO1/PEO4, as well as OVCAR8, OVCAR4, and OV90 cells. Quantitative PCR revealed consistent increases in ATF3 and AREG mRNA levels with chemotherapy exposure in all cell lines, except OV90 cells for AREG. Moreover, PEA2 chemoresistant cells displayed significantly higher ATF3 and AREG levels than their chemosensitive counterpart PEA1. DUSP1 and SGK1 also displayed increases with chemotherapy exposure in certain cell lines, although these genes were less consistent than ATF3 and AREG. Finally, NFATC2 levels were relatively low in all cell lines except OV90, and the increase in NFATC2 levels observed in human tumors with chemotherapy exposure was not observed in cell lines, which is consistent with its role as a T cell transcription factor.

We next validated the increase in ATF3 and AREG genes at the protein level using fluorescent immunohistochemistry (IHC) in a subset of patients from our NanoString cohort. Interestingly, we observed a robust increase in cytoplasmic and nuclear ATF3 in patients’ post-NACT, which was highly regional in occurrence ( Figure 4A ). Mean and maximum ATF3 intensity were significantly increased with NACT exposure ( Figures 4B, C ). Maximum ATF3 intensity was furthermore significantly correlated to NanoString mRNA levels for each patient, indicating the robustness of the NanoString data (r = 0.49, p = 0.002)( Figure 4D ). The regional occurrence of ATF3 upregulation strongly suggests a highly heterogeneous chemotherapy response, which should be further examined using spatial profiling approaches to understand the localized responses and correlates of chemotherapy response. To further query the unexpected cytoplasmic staining observed, we performed fluorescent IHC in PEA1 and PEA2 HGSOC cell lines. In PEA1 cells, we observed the expected nuclear localization of ATF3, indicating the specificity of the antibody. However, in PEA2 cells, we observed more strong cytoplasmic expression, indicating that ATF3 can be found in the cytoplasm in addition to the nucleus, although the significance of this finding is not yet known ( Supplementary Figure 2 ).

We also validated the increase in AREG gene expression with NACT exposure at the protein level using IHC. AREG integrated optical density (IOD) was significantly increased following NACT, however this appeared less robust than for ATF3 ( Figures 5A, B ). AREG mRNA levels and AREG IOD were also significantly correlated, although the correlation was weaker and did not reach the level of significance, with an r-value of 0.2805 (p = 0.09) , which could be due to the fact that it is a secreted protein ( Figure 5C ), which could be due to the fact that it is a secreted protein.

We next went on to examine ATF3 in the context of other top DEGs. We utilized The Cancer Genome Atlas (TCGA) Firehose Legacy ovarian cancer cohort to determine which genes ATF3 is correlated with in a large HGSOC cohort. Interestingly, many of the top ATF3 correlated genes in TCGA were among the top upregulated genes following NACT in our cohort of human HGSOC, suggesting that ATF3 is coregulated with these genes or plays a critical role in their transcriptional regulation. These genes included DUSP1, EGR1, SGK1, CDKN1A, NFIL3, AREG, DUSP2, DUSP5, CXCL2, IL6, and THBD ( Figure 6A ). Nonetheless, despite their strong correlations and concordant upregulation with NACT, it is important to note their potential differing functions, even among the two top upregulated transcription factors, ATF3 and EGR1. As a point of comparison, we observed the relationship of gene expression levels of these two transcription factors with patient PFS using Gene Expression Omnibus (GEO) Series (GSE) and TCGA ovarian cancer data in the KMplotter (28). While ATF3 was significantly associated with worse PFS in naïve to treatment HGSOC tumors, EGR1 had no relationship with PFS ( Figure 6B ). While the role of these transcription factors in a stress-inducible state likely differs from their role in untreated tumors, this result demonstrates that NACT-induced ATF3 and EGR1 may also have very different, unique roles within the tumor microenvironment in a post-NACT setting as well.

Next, we went on to examine immune cell infiltration changes with NACT exposure. NanoString cell type scores for all immune cell types were significantly increased following NACT, except for natural killer (NK) CD56_dim cells ( Figure 7A , Supplementary Table 4 ). An examination of individual patient scores pre- and post-NACT also revealed increases for all immune cell subsets in most patients ( Figure 7B ). A correlation analysis of top DEGs with immune cell scores revealed that the cytokines CCL4 and CCL3L1 were strongly and significantly correlated with cytotoxic cells, CD8+ cells, exhausted CD8+ cells, macrophages, and mast cells, suggesting that these cytokines may be key mediators of immune cell infiltration post-NACT ( Figure 7C ). Finally, we validated increases in CD8+ T cells by fluorescent IHC and found significant increases in CD8+ T cell counts post-NACT ( Figures 7D, E ). We also noted a strong positive correlation between CD8+ T cell counts by IHC and NanoString CD8 cell type scores ( Figure 7F ). In sum, there is a net increase in immune cell infiltration with chemotherapy, as others have previously reported, and CCL3L1 and CCL4 could play a role in this immune cell recruitment.

Next, we performed Cox proportional hazards regression analysis of immune cell subsets with PFI, using pre- and post-NACT scores. There were no significant results with pre-NACT scores, but there was a significantly lower risk of earlier recurrence in patients with higher levels of post-NACT exhausted CD8+ cells, dendritic cells, and mast cells, while NK CD56_dim cells trended toward significance ( Figures 8A, B ).

When examining pre-NACT pathway scores, only the pathway “Cytotoxicity” was significantly related to lower risk of earlier recurrence. In addition, “Interferon Signaling” almost reached significance for lower risk of recurrence. On the other hand, several post-NACT pathways were significantly related to PFI. Higher scores in the categories “Angiogenesis”, “Autophagy”, Hedgehog Signaling”, “Hypoxia”, “JAK-STAT Signaling”, “MAPK Signaling”, “Matrix Remodeling and Metastasis”, “Metabolic Stress”, “Notch Signaling”, “PI3K-Akt Signaling”, “Transforming growth factor-beta (TGF-β) Signaling”, and “Wnt Signaling” were significantly associated with lower risk of recurrence. Conversely, higher scores in the pathways “Cell Proliferation” and “Epigenetic Regulation” were associated with a greater risk of earlier recurrence ( Figure 8C ).

When we stratified pre- and post-NACT samples by 12-mo PFI, we found no significant differences in median immune cell type scores ( Figures 9A, B ; Supplementary Figures 3A, B ), despite the relationship between PFI and certain immune cell subsets observed in Figure 8B . Only post-NACT mast cells showed a trend toward increased levels in patients with PFI >12 mo (p = 0.07). Likewise, pre-NACT pathway scores were not significantly different between patients with ≤12 mo PFI versus >12 mo PFI ( Figure 9C ; Supplementary Figure 3C ). However, post-NACT pathway scores were significantly different between patients with ≤12 mo and >12 mo PFI, namely for “Angiogenesis”, “Autophagy”, “Cell Proliferation”, “Epigenetic Regulation”, “Metabolic Stress”, “TGF-β Signaling”, and “Wnt Signaling”, which agrees with the results of the Cox proportional hazards regression analysis in Figure 8D ( Figure 9D ; Supplementary Figure 3D ).

Finally, we performed differential expression analysis for post-NACT samples stratified by 12-mo PFI ( Figures 10A, B ; Supplementary Table 5 ). While this analysis did not reveal any changes that met false discovery rate correction, the genes that were differentially expressed according to p-value < 0.05 revealed a trend toward higher levels of proliferation and DNA damage repair genes in patients with PFI ≤12 mo (UBE2C, CCNE1, CCNB1, BIRC5, CENPF, RRM2, ANLN, MK167, CEP55, RAD51, and FANCA), and lower levels of expression of genes related to adhesion, angiogenesis, and epithelial-to-mesenchymal transition (ANGPT1, ZEB1, ITGA1, SNAI1, FLT1, CDH5, PDGFRB, ICAM2, PRR5, COL17A1, VEGFC, PTGS2) in patients with ≤12 mo PFI. Most of the genes identified through differential expression were also found to be related to PFI according to Cox proportional hazards regression ( Figure 10C ).

Together, the analysis of patient immune genes/pathways and cell types with PFI shows that patients with shorter PFI may have less robust immune infiltration of certain subsets post-NACT, although the differences are not very striking. More strikingly, post-NACT pathway scores for several typically pro-tumorigenic and immunoregulatory pathways were significantly associated with longer PFI, suggesting that these pathways may actually play a role in tumor responsiveness to chemotherapy. Finally, the trend toward higher expression of a cluster of proliferative genes, in particular UBE2C, CCNE1, and CCNB1, in patients with shorter PFI demonstrates the importance of chemotherapy-induced regulation of these genes in mediating responses.