Proteome Profiling Uncovers an Autoimmune Response Signature That Reflects Ovarian Cancer Pathogenesis

Harnessing the immune response to tumor antigens in the form of autoantibodies, which occurs early during tumor development, has relevance to the detection of cancer at early stages. We conducted an initial screen of antigens associated with an autoantibody response in serous ovarian cancer using recombinant protein arrays. The top 25 recombinants that exhibited increased reactivity with cases compared to controls revealed TP53 and MYC, which are ovarian cancer driver genes, as major network nodes. A mass spectrometry based independent analysis of circulating immunoglobulin (Ig)-bound proteins in ovarian cancer and of ovarian cancer cell surface MHC-II bound peptides also revealed a TP53–MYC related network of antigens. Our findings support the occurrence of a humoral immune response to antigens linked to ovarian cancer driver genes that may have utility for early detection applications.


Introduction
A humoral immune response in the form of autoantibodies to tumor antigens occurs early during tumor development. Identification of antigens that induce a selective autoantibody response associated with a particular cancer type has translational relevance for cancer screening [1][2][3]. There is currently an ongoing search for biomarkers that have utility for ovarian cancer early detection. The overall five-year survival rate for this cancer is below 30%, as over 70% of patients are diagnosed with stages III or IV disease. However, subjects diagnosed with localized disease have a survival rate of 75-90% [4]. At present, cancer antigen 125 (CA125) is the most investigated early detection marker for ovarian cancer [5]. Sequential monitoring of subjects with ultrasound and for elevated circulating levels of CA125 can achieve moderate specificity [6], but with limited sensitivity. There remains a need for identification of additional markers for ovarian cancer early detection. Tumor associated autoantibodies may improve on the performance of CA125 alone as we recently described for the human epididymis protein 4 (HE4) antigen-autoantibody complexes as complementing CA125 for detecting early-stage ovarian cancer [7].
Multiple approaches are currently available for the discovery of tumor antigens that induce a humoral autoantibody response. No single approach allows a comprehensive assessment of the full repertoire of epitopes associated with an autoantibody response in cancer. cDNA expression libraries [8], phage display [9] and recombinant protein arrays [10][11][12] have been utilized to identify antigens associated with autoantibodies. Other approaches include natural protein arrays that utilize fractionated tumor cell lysates as the source of antigens to preserve post-translational modifications (PTMs) and other protein alterations associated with immune reactivity [13][14][15]. Recently we have reported on the use of whole-genome derived peptide arrays as an approach for identification of pre-diagnostic autoantibodies associated with lung cancer, which provides a comprehensive coverage of peptide epitopes encoded in the genome [16].
In this study we explored the relationship of the autoantibody response in ovarian cancer to disease pathogenesis. We first investigated the repertoire of antigens that induce a humoral immune response in ovarian cancer using recombinant protein arrays, which was followed by analysis of circulating antigen-antibody complexes in ovarian cancer using mass spectrometry. We also profiled using mass spectrometry ovarian cancer cell line MHC-II bound peptides as a potential source of epitopes associated with autoantibodies. Integrated data analyses yielded immune network signatures involving TP53 and MYC, which are major contributors to the pathogenesis of ovarian cancer.

Recombinant Protein Array-Based Ovarian Cancer Autoantibody Signature
We investigated the antibody reactivity of 20 serous ovarian cancer cases and 17 controls (Table S1) using recombinant protein arrays. The IgG reactivity of 75 recombinant proteins showed a statistically significant increase in ovarian cancer cases compared to controls (p < 0.05, Figure 1A and Table S2). Applying stricter criteria (p < 0.02, two-tailed Wilcoxon signed-rank test) narrowed the list to 25 recombinants (Table 1). The direct interaction network analysis using Ingenuity Pathway Analysis (IPA) for these top reactive proteins revealed TP53 and MYC to be the major central network nodes ( Figure 1B).

Circulating Immunoglobulin (Ig)-Bound Protein Signature in Ovarian Cancer
Released antigens may occur in circulation bound to Ig [15,[19][20][21]. We profiled ovarian cancer circulating Ig-bound proteins in ovarian cancer subjects compared to controls using tandem mass tag (TMT)-based liquid chromatography mass spectrometry (LCMS). For TMT labeling experiments (see methods), three sample sets (sample set-1, -2 and -3) were prepared; each sample set consisted of four pooled case samples (each pool was comprised of three cases, n = 36 in total) and two pooled control samples (each pool was comprised of 10 age-matched healthy controls). Patient information is provided in Table S1. Pooling strategies were as follows: Case pools 1, 2, 5 and 6 consisted of CA125 negative (defined as < 35 U/mL) cases, case pools 3, 4, 7 and 8 consisted of CA125 positive (≥ 35 U/mL) cases and case pools 9-12 were based on histology (Table S1). We filtered out abundant plasma proteins as contaminants and considered proteins as tumor-derived antigen candidates using the following criteria: (i) case/control ratios of Ig-bound proteins greater than 1.2 identified in at least two sets, (ii) protein products of genes expressed in ovarian cancer cells [22], which yielded 24 proteins (Table 2). Interestingly, IPA again revealed the top protein network as centered around TP53, MYC and ESR1 ( Figure 2A) with functions consisting of cell cycle, cell death and survival and organismal injury (Table S3).

Discussion
Using two proteomics platforms with independent subject samples, we investigated autoimmune response networks of antigenic proteins and peptides in ovarian cancer. We observed in the initial discovery set significant reactivity against 75 recombinants with ovarian cancer sera compared to controls.
Notably, it has previously been reported that autoantibodies against RALBP1, transcriptional adapter 3 (TALD3L), E3 ubiquitin-protein ligase CBL-B (CBLB) and serine/arginine-rich splicing factor 10 (FUSIP1) are statistically significantly elevated in sera of ovarian cancer patients in comparison to healthy controls [10]. Consistently, our independent analysis also indicated elevated autoantibody reactivity against these protein targets with corresponding AUCs of 0.767, 0.685, 0.653 and 0.653, respectively, for delineating ovarian cancer cases from healthy controls. Thus, our findings and those of others were validated [10].
Using IPA, the 25 top performers in the current study were part of a TP53 and MYC network. Given prior publications of autoantibodies in ovarian cancer using different platforms to search for autoantibodies, we performed similar IPA on data from other reports to determine associated networks ( Figure S2) [10,[25][26][27].
Consistent with our findings, we uncovered TP53 and MYC as major nodes for antigens associated with autoantibodies, suggesting an intrinsic relationship between established drivers of serous ovarian cancer pathogenesis and autoantibody targets [28]. We previously reported on a triple-negative breast cancer (TNBC) autoimmune response signature that was also mainly contributed by TP53 and MYC [15]. According to the Cancer Genome Atlas (TCGA), serous ovarian carcinoma and the basal type of breast cancer have molecular phenotype similarity that include MYC high expression and high frequency of TP53 inactivation [29]. Thus, a similarity in driver genes would account for similarity in the autoimmune response network between the two cancer types.
Autoantibodies to TP53 itself are known to be elevated in various types of cancer [30][31][32][33]. Shimada et al. reported positivity of TP53 autoantibody was detected in about 20% of cancer patients [34]. In ovarian cancer, Yang et al. reported the utility of TP53 autoantibody for early ovarian cancer detection combined with CA125 based on pre-diagnostic samples [6]. Additionally, MYC autoantibodies have been reported in ovarian cancer [26,35,36]. These results suggested that gene alternations such as amplification or mutation will trigger TP53 and MYC autoantibody production in ovarian cancer. TP53 and MYC were not part of the set of recombinants on the array we have utilized. Similarly, mass spectrometry-based detection of TP53 and MYC is often lacking because of sensitivity and/or post-translational modifications [37].
CSTF2, RALBP1 and its binding partner REPS1 were part of the TP53 and MYC signature and showed significant performance with an AUC = 0.958. CSTF2 was a member of the cleavage stimulation factor (CSTF) complex that is involved in the 3' end cleavage and in polyadenylation of pre-mRNAs [38]. Evidence suggests that regulation of polyadenylation may play an important role in cell growth control and tumor development [39]. The formation of a complex between CSTF, BARD1/BRCA1 and TP53 has been reported to repress mRNA polyadenylation following treatment of cells with DNA-damage-inducing agents, suggesting that CSTF may have a direct role in the development of ovarian cancer [40]. CSTF2 mRNA expression was low or absent in most normal tissues suggesting that the presence of autoantibodies to this protein was reflective of its dysregulated expression in ovarian cancer [41]. Moreover, RALBP1 binding was critical for the activation of Ral signaling in Ras-induced transformation and tumorigenesis of human cells [42]. Dysregulation of micro-143-3p and RALBP1 has been reported to contribute to the pathogenesis of ovarian cancer [43]. REPS1 is a binding partner of RALBP1 that was found to play a role in regulating EGF receptors and Ral-GTPases activity [17]. Collectively, these findings highlight that the target antigens identified in this study are related to the pathogenesis of ovarian cancer.
Mass spectrometry-based circulating Ig-bound protein analysis yielded concordant results with respect to TP53 and MYC driven network with both newly diagnosed and pre-diagnostic samples. Likewise, ovarian cancer cell surface MHC-II bound peptidome analysis showed clearly a TP53 and MYC centered signature. These results further reinforce the role of the driver genes TP53 and MYC in inducing proteins that trigger a humoral immune response.
There is increasing evidence for circulating immune complexes during tumor development that may serve as cancer biomarkers. We recently reported that human epididymis protein 4 (HE4) antigen-autoantibody complexes could significantly improve diagnostic performance in combination with CA125 compared with CA125 alone based on analysis of early stage ovarian cancer samples [7]. Other complexes notably involving cofilin 1 were found to be associated with pancreatic cancer [44].
We acknowledge that there is limited overlap between protein-autoantibody targets identified through the recombinant protein arrays with that of Ig-bound antigen complexes identified via mass spectrometry. There are a multitude of strategies available for discovery of tumor antigens directed autoantibodies in circulation. Each strategy targets a different repertoire of antigens and presents both advantages and disadvantages as we have previously outlined in a review [21]. The primary intent of this study is to explore the relationship of the autoantibody response in ovarian cancer to pathogenesis. Thus, we intentionally employed a multi-platform approach to uncover a diversity of autoantibodies with a goal to ascertain their relationship to disease pathogenesis given that different platforms would identify different autoantibodies but that may reflect the same underlying origin.
In conclusion, our data from this study as well as pathway analysis of other reported data is indicative of an autoimmune response targeting antigens regulated by driver genes in ovarian cancer. Further validation of autoantibodies against targets that exhibited high performance notably CSTF2, RALBP1 and REPS1 will be needed. If successful, such autoantibody targets may offer utility for early detection of ovarian cancer.

Recombinant Protein Array Analysis
For the autoantibody discovery analysis using recombinant protein arrays, blood samples were collected at the Fred Hutchinson Cancer Research Center following Institutional Review Board approval and informed consent (no ethic code and protocol numbers were assigned). The subjects were women diagnosed with serous ovarian cancer and controls consisting of apparently healthy women attending regular breast cancer screening exams and women undergoing gynecologic surgery for a variety of conditions but with normal ovarian pathology. Controls were matched to cases for age, race, family history of ovarian and breast cancer and collection date. Subject information is provided in Table S1.
Recombinant protein arrays containing 5005 recombinants arrayed in duplicate were utilized in the initial discovery phase (Thermo Fisher Scientific, Waltham, MA, USA). Alexa 647-labeled anti-human IgG (Thermo Fisher Scientific) was utilized for quantification of reactivity. Serum samples were assessed for IgG reactivity against arrayed proteins using a three-step indirect immunofluorescence protocol. All steps were done at 4 • C. Briefly, a blocking reaction for protein microarrays was done using a blocking buffer (PBS with 1% BSA and 0.1% Tween-20) for 1 h. Serum samples were diluted 1:150 in the probing/washing buffer (PBS with 1% BSA, 0.5 mM DTT, 5 mM MgCl 2 , 0.05% TritonX-100 and 5% glycerol) and applied onto the microarrays and incubated for 2.5 h. Following washing with the washing buffer for 3 × 10 min, microarrays were incubated with 1 µg/mL Alexa 647-labeled anti-human IgG antibody diluted in the washing buffer for 1 h. The washing buffer was subsequently applied for 3 × 10 min, followed by drying via spinning at 500 × g for 2 min. All microarrays were scanned with a GenePix 4200A scanner using the same settings. Scanned images were analyzed using GenePix 6.0 microarray analysis software. Local background subtracted median spot intensities were used for downstream statistical analysis.

Analysis of Circulating Ig-Bound Proteins in Ovarian Cancer
For mass spectrometry based circulating Ig-bound protein analysis, blood samples from ovarian cancer patients and from healthy controls who did not develop ovarian cancer were collected at the University of Texas M.D. Anderson Cancer Center Gynecologic Tissue Bank. All samples were collected following Institutional Review Board approval and informed consent. Subject information is available in Table S1. The study cohort at MD Anderson is MDACC-NROSS. The protocol number of the study at MD Anderson is ID01-022.
Detailed information regarding mass spectrometry-based analysis of Ig-bound protein complexes is described elsewhere [19]. Briefly, Ig-bound proteins from a total of 100 µL of plasma were extracted using NAb protein A/G spin columns (Thermo Fisher Scientific) according to the manufacturer's instructions. Columns were equilibrated twice with 400 µL binding buffer (phosphate buffered saline; PBS, pH 7.2) and then incubated for 10 min at room temperature (RT) with plasma samples diluted 1:2 in PBS, pH 7.2. Columns were washed three times with 400 µL of PBS, pH 7.2. Ig-bound proteins were eluted twice with 400 µL of 0.1 M glycine, pH 3. The flow-through was collected and then neutralized with 40 µL of PBS, pH 9. After each step, columns were centrifuged for 1 min at 5000× g. To reduce non-specific binding to the protein A/G spin columns, an additional low pH wash with 400 µL of PBS, pH 5, was performed before Ig-bound protein elution.
For mass spectrometry analysis, the collected proteins were treated with 25 mM TCEP for Cys reduction and subsequently alkylated with acrylamide. The samples were next fractionated at the protein level by reverse-phase chromatography followed by desalting for 5 min with 95% mobile phase A (0.1% TFA in 95% H2O). Proteins were eluted from the column and collected into 12 fractions, with a gradient elution that included an increase from 5% to 70% mobile phase B (0.1% TFA in 95% acetonitrile) over 25 min, 70% to 95% mobile phase B for 3 min, a wash step to hold at 95% mobile phase B for 2 min, followed by a re-equilibration step at 95% mobile phase A for 5 min.

Immunopeptidome Analysis
The OVCAR8 cell line was incubated with 50 ng/mL interferon gamma (IFN γ) for 24 h before collecting MHC-II bound peptides from 500 million cells. MHC-II bound peptides were eluted, processed and analyzed by LC-MS/MS and searched using our previously reported methodology [19,23].

Mass Spectrometry Analysis
For Ig-bound protein analysis, protein digestion and identification by LC-MS/MS was performed using our established protocol [19,45,46]. Briefly, a nanoAcquity UPLC system coupled in-line with WATERS SYNAPT G2-Si mass spectrometer was used for the separation of pooled digested protein fractions. The system was equipped with a Waters Symmetry C18 nanoAcquity trap-column (180 µm × 20 mm, 5 µm) and a Waters HSS-T3 C18 nanoAcquity analytical column (75 µm × 150 mm, 1.8 µm). Data were acquired in resolution mode with SYNAPT G2-Si using Waters Masslynx (version 4.1, SCN 851). The mass spectrometer was operated in V-mode with a typical resolving power of at least 20,000. All analyses were performed using positive mode ESI using a NanoLockSpray source. The lock mass channel was sampled every 60 s. Accurate mass LC-HDMSE data were collected in an alternating, low energy (MS) and high energy (MSE) mode of acquisition with mass scan range from m/z 50 to 1800. The spectral acquisition time in each mode was 1.0 s with a 0.1 s inter-scan delay. The acquired LC-HDMSE data were processed and searched against protein knowledge database (Uniprot and TruEMBL, 92,355 human protein sequences) through ProteinLynx Global Server (Version 3.0.2, Waters Company) with 4% FDR.

Immunoprecipitation (IP) and Western Blot Analysis
Two ovarian cancer cell lines (OVCAR8 and DOV13) were washed two times with PBS and treated with IP lysis buffer (Thermo Fisher Scientific) at 4 • C for 30 min. After centrifugation at 20,000× g for 30 min at 4 • C, the supernatant was collected for IP. To conjugate primary antibody, 2 ug of anti-RALBP1 antibody (clone 2A1, Abnova, Taipei, Taiwan), 2 uL of anti-REPS1 antibody (clone D6F4, Cell Signaling Technology, Danvers, MA, USA), 2 ug of mouse isotype control IgG (clone 20102, R&D Systems, Minneapolis, MN, USA) and 2 ug of rabbit isotype control IgG (clone DA1E, Cell Signaling Technology) were mixed with Dynabeads protein G (Thermo Fisher Scientific) for 30 min at room temperature. Following incubation with antibody-Dynabeads conjugate and 1 mg of cell lysate overnight at 4 • C, antibody-antigen complex was washed three times with PBS. Precipitated proteins were eluted using Laemmli's buffer (Bio-rad, Hercules, CA, USA). Western blotting was performed as previously described [46].

Ingenuity Pathway Analysis (IPA)
IPA (Version 49309495, Qiagen, Hilden, Germany) was utilized for network signature analysis with the following settings: (1) direct relationships, (2) excluded endogenous chemicals, (3) number of molecules per network was 35 and networks per analysis was 25 and (4) relationships considered were those experimentally observed and human.

Statistical Analysis
Recombinant protein array data were normalized with quantile normalization, and intensity measures for duplicate spots were averaged. A two-tailed Wilcoxon signed-rank test was applied to each recombinant protein to compare differences in mean intensity between cases and controls. Receiver operating characteristic (ROC) curve analysis was performed to assess the performance of biomarker candidates in distinguishing cases from controls. Model building was based on a logistic regression model. The AUC of the derived panel was determined by using the empirical ROC estimator of the linear combination corresponding to the model. The standard error (S.E.) and the corresponding 95% confidence intervals presented for the performance of each biomarker or biomarker panel were based on the bootstrap procedure in which we re-sampled with replacement separately for the controls and the diseased 1000 bootstrap samples. ROC curves and model building was performed using R statistical software version 3.3.1.

Conclusions
Our proteomics based data from this study as well as pathway analysis of other reported data is indicative of an autoimmune response targeting antigens regulated by driver genes such as TP53 and MYC in ovarian cancer.