molecular biomarkers of residual disease after surgical ... · erin k. crane 2, robert l. coleman ,...

Imaging, Diagnosis, Prognosis

Molecular Biomarkers of Residual Disease after SurgicalDebulking of High-Grade Serous Ovarian Cancer

Susan L. Tucker1, Kshipra Gharpure2, Shelley M. Herbrich1, Anna K. Unruh1, Alpa M. Nick2,Erin K. Crane2, Robert L. Coleman2, Jamie Guenthoer6, Heather J. Dalton2, Sherry Y. Wu2,Rajesha Rupaimoole2, Gabriel Lopez-Berestein3,5, Bulent Ozpolat3, Cristina Ivan2, Wei Hu2,Keith A. Baggerly1, and Anil K. Sood2,4,5

AbstractPurpose: Residual disease following primary cytoreduction is associated with adverse overall survival in

patients with epithelial ovarian cancer. Accurate identification of patients at high risk of residual disease has

been elusive, lacking external validity and prompting many to undergo unnecessary surgical exploration.

Our goal was to identify and validate molecular markers associated with high rates of residual disease.

Methods: We interrogated two publicly available datasets from chemona€�ve primary high-grade

serous ovarian tumors for genes overexpressed in patients with residual disease and significant at a 10%

false discovery rate (FDR) in both datasets. We selected genes with wide dynamic range for validation in

an independent cohort using quantitative RT-PCR to assay gene expression, followed by blinded

prediction of a patient subset at high risk for residual disease. Predictive success was evaluated using

a one-sided Fisher exact test.

Results: Forty-seven probe sets met the 10% FDR criterion in both datasets. These included FABP4 and

ADH1B, which tracked tightly, showed dynamic ranges >16-fold and had high expression levels associated

with increased incidence of residual disease. In the validation cohort (n ¼ 139), FABP4 and ADH1B were

again highly correlated. Using the top quartile of FABP4 PCR values as a prespecified threshold, we found 30

of 35 cases of residual disease in the predicted high-risk group (positive predictive value ¼ 86%) and 54 of

104 among the remaining patients (P ¼ 0.0002; OR, 5.5).

Conclusion: High FABP4 and ADH1B expression is associated with significantly higher risk of residual

disease in high-grade serous ovarian cancer. Patients with high tumoral levels of these genes may be

candidates for neoadjuvant chemotherapy. Clin Cancer Res; 20(12); 3280–8. �2014 AACR.

IntroductionSurgery is widely accepted as a critical component of

treatment for high-grade serous ovarian cancer (HGSOC),although the optimal timing of surgery has been debated.

Residual disease following a primary cytoreductive effortis negatively associated with clinical metrics, includingresponse to adjuvant chemotherapy, progression-free sur-vival, and overall survival. Initial reports found improvedsurvival among women left with <1.5 cm residual diseasecompared with larger volume disease (1–3). Since then, theyardstick for defining "optimal" debulking has variedbetween <2 and 0 cm (complete resection or R0). Recentreports suggest that the best metric is the distinctionbetween R0 and any residual disease (3, 4), which is alsomore reliably assessed and reproducible (external validity).

Accurate predictors of surgical outcome could substan-tially impact management of ovarian cancer by guidingpatients with highest likelihood of having residual diseaseto neoadjuvant chemotherapy with the potential for inter-val cytoreductive surgery later. While baseline rates ofleaving some extent of residual disease at upfront surgeryare high (65%–75%), identifiable factors are needed toindicate whether a specific patient is at significantly higherrisk. Some studies have attempted to define predictors ofoptimal cytoreduction, most often defined as <1 cm resid-ual disease (5). However, given the considerable variability

Authors' Affiliations: Departments of 1Bioinformatics and ComputationalBiology, 2Gynecologic Oncology andReproductiveMedicine, 3Experimen-tal Therapeutics, 4Cancer Biology, and 5Center for RNA Interference andNon-Coding RNA, The University of Texas MD Anderson Cancer Center,Houston, Texas; and 6Fred Hutchinson Cancer Research Center, Seattle,Washington

Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).

S.L. Tucker, K. Gharpure, and S.M. Herbrich contributed equally to thisarticle.

K.A. Baggerly and A.K. Sood share senior authorship of this article.

Corresponding Author: Anil K. Sood, The University of Texas MD Ander-son Cancer Center, 1155 Pressler St, Unit 1362, Houston, TX 77030.Phone: 713-792-4130; Fax: 713-792-7586; E-mail:[email protected]

doi: 10.1158/1078-0432.CCR-14-0445

�2014 American Association for Cancer Research.

ClinicalCancer

Research

Clin Cancer Res; 20(12) June 15, 20143280

on August 17, 2020. © 2014 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst April 22, 2014; DOI: 10.1158/1078-0432.CCR-14-0445

http://clincancerres.aacrjournals.org/

in definition and assessment of debulked disease, suchpredictors have not been particularly reliable. Other effortsto spare unnecessary primary debulking surgery by pre- orintraoperative assessment have abounded (6–8). Unfortu-nately, none have reached the level of external validitynecessary for incorporation into general practice.A predictor of residual disease with high sensitivity is

unlikely to exist, as incomplete resection sometimes occursowing to tumor location near critical organ structures. Inmany cases, though, residual disease is a consequence ofbiologic tumor characteristics, with wide dissemination ofdisease throughout the pelvis. We hypothesized that in thelatter case, a high likelihood of residual disease might bepredictable based on biomarkers assessed from tumor tis-sue. The goal of the present study, therefore, was to identifymolecular markers associated with a high likelihood ofresidual disease. We used 2 publicly available microarraydatasets that included residual disease information to dis-cover candidate gene markers and subsequently validatedour biomarkers in an independent clinical cohort. In thevalidation cohort, blinded predictions of residual diseaserisk based on candidate biomarker gene expression assayedusing quantitative real time polymerase chain reaction(qRT-PCR) were compared with actual surgical outcomes.

Materials and MethodsHere, we briefly outline our materials, experimental

procedures, and methods of analysis. Full details (includ-ing computer code) are given in the SupplementaryAppendix and at http://bioinformatics.mdanderson.org/Supplements/ResidualDisease.

Data for exploratory studiesFor biomarker discovery, we used 2 large publicly avail-

able Affymetrix microarray datasets involving patients withHGSOCs and providing associated clinical information,

including residual disease status. The first of these was theovarian cancer cohort from The Cancer Genome Atlas(TCGA; ref. 9). We downloaded CEL files (level 1 data) forthe ovarian samples (Affymetrix HT HG-U133A arrays, n¼598) on September 2, 2012; these represent the TCGAupdate that was current as of June 24, 2011 (revision1007). We downloaded the associated clinical data (n ¼576) onSeptember 14, 2012.Weomitted 4 samplesmarkedfor exclusion by TCGA and performed additional samplefiltration based on clinical annotation. We excluded sam-ples if they were from recurrent tumor, omental tumor, ornormal tissue. When there were multiple primary tumorsamples per patient, we retained data from just one sample.We also excluded cases if there was no information aboutresidual disease status, if the tumor was not high grade, orif the patient received neoadjuvant chemotherapy. Thesecond dataset was from the study of Tothill and colleagues(10).We downloaded CEL files (Affymetrix U133þ2 arrays,n ¼ 285) and clinical data from the Gene ExpressionOmnibus (GSE9891) on September 13, 2012. We excludedcases from this dataset if tumor samples were low grade, oflow malignant potential, non-serous histology, or non-ovarian or peritoneal origin. Cases were also excluded ifthe patient received neoadjuvant chemotherapy or if resid-ual disease status was not provided. Accordingly, biomarkerdiscovery was performed using only data from primarytumors of chemona€�ve patients with HGSOCs.

Following initial biomarker discovery, we considered 2additional datasets for qualitative checks on patterns ofexpression for genes of interest (Bonome and colleaguesand The Cancer Cell Line Encyclopedia; see SupplementaryAppendix for details; refs. 11, 12).Wenote that although thestudy of Bonome and colleagues includes clinical data, wecould not use it for validation because in that datasetresidual disease is scored as optimal versus suboptimal,whereas our method was developed to predict risk of anyresidual disease versus R0.

For each of the 4 datasets described above, we quantifiedexpression at the probeset level with R statistical software(version 2.15.1) using the robust multi-array average pro-cedure (13) as implemented in the subroutine justRMAin the R "affy" package (version 1.34.0). We only consider-ed probesets common to the array platforms used in thefirst 2 datasets. Except for the 4 samples marked for exclu-sion by TCGA, our sample filtrations described above wereperformed subsequent to the quantification step.

Validation studiesFollowing identification of candidate genes having a

wide dynamic range and with high expression levels asso-ciated with high risk of residual disease in exploratoryanalyses (see below), we performed validation studies inan independent cohort. Following Institutional ReviewBoard approval, we obtained primary ovarian tumor sam-ples from archived surgical material at The University ofTexas MD Anderson Cancer Center (Houston, TX; n ¼ 84)and the Pacific Ovarian Cancer Research Consortium atthe Fred Hutchinson Cancer Research Center (Seattle, WA;

Translational RelevancePrimary debulking surgery followed by adjuvant che-

motherapy has been the standard for treatment of wom-en with high-grade serous ovarian cancer. Residual dis-ease following primary cytoreduction is known to beassociated with poor overall and progression-free sur-vival, and patients at high risk of having residual diseasemay be candidates for neoadjuvant chemotherapy. Ourstudy used publicly available microarray data to identify2 biomarkers, FABP4 and ADH1B, which were subse-quently validated in an independent clinical cohort toidentify patients at high risk of residual disease afterprimary surgery. Use of these biomarkers could helpclinicians personalize treatment options for womensuspected of having ovarian cancer. Our study has ledto ongoing development of a CLIA-based assay for use ofthese biomarkers in a prospective validation trial.

Biomarkers of Residual Disease in Ovarian Cancer

www.aacrjournals.org Clin Cancer Res; 20(12) June 15, 2014 3281




n ¼ 55). We also included 41 samples of omental tumortissue fromMDAnderson for comparison. All samples wereselected from chemona€�ve tumors. Information aboutresidual disease status in the validation cohort was extractedfrom patient medical records by clinical members of theteam (A.M.Nick andA.K. Sood) andwas scored as R0 versusany residual disease. This information was kept blindedfrom all other team members until after predictions weremade concerning identification of a subset of patients athigh risk for residual disease. These predictions were madeon the basis of expression levels of a candidate gene ofinterest relative to 18S, assayed using qRT-PCR (ref. 14; seeSupplementary Appendix for additional details).

Statistical methodsExploratory data analyses. We used 2-sample t tests to

compare expression levels by residual disease status (R0 vs.any residual disease) in the TCGA and Tothill datasets,separately.We analyzed the resulting collections of nominalP values using themethod of Benjamini andHochberg (15)to identify probesets significant at a 10% or 5% falsediscovery rate (FDR) in each data set separately.We selectedprobesets meeting the specified criterion in both subsets.We produced density plots for these probesets, as well asheatmaps using hierarchical clustering, to illustrate pat-terns of expression. We examined the resulting plots toidentify probesets with (i) wide dynamic range and(ii) high levels of expression associated with particularlyhigh incidence of residual disease. We examined similarplots for the selected probesets in the Bonome and CCLEdatasets to see whether the qualitative expression patternswere also observed there.

Selecting a sample size and prediction threshold for thevalidation study. Following identification of genes forwhich high expression levels were associated with highrisk of residual disease, we wished to select an a prioridecision threshold for "calling" a patient in the validationcohort to be at high risk of residual disease based onelevated biomarker expression measured by qRT-PCR.Patients with expression of a candidate biomarker geneabove the selected threshold would be in the predictedhigh-risk group for residual disease. To select a decisionthreshold, we first computed the positive predictivevalues (PPV) in the TCGA and Tothill datasets at thresh-olds defined by varying quantiles of biomarker expres-sion. We assumed that similar PPVs would apply to thevalidation cohort and performed numerical simulationsto determine a decision threshold at which we would bemaximally powered to detect a difference in incidence ofresidual disease in the validation set. We chose a thresh-old with an estimated power of at least 80% to detect adifference in incidence of residual disease significant atP < 0.05 using a 1-sided Fisher exact test. We used similarsimulations for our initial sample size computationsbefore collecting the validation set.

Statistical assessment of success. We constructed a 2 � 2table showing our calls (high risk of residual disease vs.lower risk of residual disease) comparedwith actual surgical

outcomes (residual disease vs. R0) in the validation cohort.We used a 1-sided Fisher exact test to test the null hypothesisof equivalent rates of residual disease in the 2 groups againstthe alternative of increased incidence of residual disease inthe predicted high-risk group.

ResultsClinical outcomes in the exploratory cohorts

In the TCGA and Tothill datasets, 491 and 189 patients,respectively, met the criteria for inclusion in ouranalysis. Figure 1 shows overall survival in these cohortsby extent of residual disease after debulking surgery. Sur-vival is significantly better for patients with complete resec-tion (R0) compared with those with any residual disease,including those traditionally considered to be optimallydebulked (1–10 mm residual disease). This result, reportedpreviously for the TCGA data (9), is consistent with thefindings from other studies as well (3, 4). The percentagesof patients with any residual disease are 77% (n ¼ 378)and 74% (n ¼ 139) in the TCGA and Tothill cohorts,respectively.

Identification of genes associated with increasedincidence of residual disease

There are 47 probesets, representing 38 different genes(Supplementary Table S1), with P values meeting a 10%FDR criterion in both datasets. Lowering the FDR to 5%results in 8 probesets common to both datasets. Of note,the additional filtration imposed by requiring successin both datasets is severe; the 8 common probesetsrepresent the intersection of 149 probesets meeting the5% FDR threshold for TCGA and 81 meeting the samecriterion for the Tothill cohort. Heatmaps of standardizedexpression of the 8 probesets meeting the 5% FDR crite-rion in the TCGA and Tothill cohorts are shown in Fig. 2;heatmaps for the 47 probesets meeting the 10% FDRcriterion in both data sets are included in the Supple-mentary Appendix.

Figure 3 shows density plots of expression levels for 2 ofthe top 8 candidate probesets of interest, LUM and FABP4(plots for all 47probesets identified at the 10%FDR level areincluded in the Supplementary Appendix). While bothLUM and FABP4 show increased rates of residual diseaseat higher expression levels, the distributions of the 2 genesare qualitatively different. For FABP4, there is a level ofexpression above which nearly all patients have residualdisease. For LUM, in contrast, there is no clearly definedcutoff point above which the majority of patients haveresidual disease.

An additional probeset exhibiting the same type ofqualitative behavior as for FABP4 corresponds to ADH1B.For the FABP4 and ADH1B probesets, the distributionof expression values is bimodal, with a large cohorthaving low expression levels falling within a relativelynarrow range (the lower limit of detection for the assay),and a smaller cohort with substantially (up to orders ofmagnitude) higher expression levels. Moreover, the

Tucker et al.

Clin Cancer Res; 20(12) June 15, 2014 Clinical Cancer Research3282




heatmaps in Fig. 2 show that the highest expression levelsof FABP4 and ADH1B occur in a subset of patients with avery high incidence of residual disease. Figure 4A and B

show the highly correlated levels of FABP4 and ADH1Bin the TCGA and Tothill datasets, respectively. Usingcutoffs of 3.5 on both TCGA axes, 97 of 107 (90.6%)

Color keyand histogram

TCGA samples Tothill samples

–4 –2 0

060

01,

400

Row Z-score

DCN

DCN

CXCL12

CXCL12

CXCL12

GADD45B

GADD45B

CRISPLD2

CRISPLD2

LUM

LUMFABP4

FABP4ADH1B

ADH1BCXCL12

Cou

nt

2 4

Color keyand histogram

A B

–3 –1

010

025

0

Row Z-score

Cou

nt

1 3

Figure 2. Heatmaps showing standardized expression of the 8 probesets with false discovery rates <5% identified using t tests comparing patients withand without residual disease in the TCGA (A) and Tothill (B) cohorts. The colorbars at top indicate presence (red) or absence (blue) of residual disease foreach patient.

Figure 1. Overall survival by extent of residual disease after cytoreductive surgery for patients with HGSOC in the TCGA (A) and Tothill (B) cohorts. Categoriesfor amount of residual disease are labeled as in the original datasets: "No macroscopic disease" in A and "nil" in B correspond to no residual disease (R0);"1–10mm" in A corresponds to "<1" in B; "11–20mm" and ">20mm" in A correspond to ">1" in B; the size of the macroscopic tumor remaining is not knownfor 13 of the samples from Tothill and colleagues ("macro size NK"). P values are from log-rank comparisons of all groups per panel; P values from thecomparison of R0 versus any residual disease (3 lower groups combined) are P < 0.0001 and P ¼ 0.0022 for A and B, respectively.






patients with high expression of both genes have residualdisease versus 281 of 384 (73.2%) of the remainingpatients. Using cutoffs of 5.25 and 4.5 for FABP4 and

ADH1B, respectively, in the Tothill data, 59 of 63 (93.7%)patients with high expression of both genes have residualdisease, versus 80 of 126 (63.5%) of the others.

Figure 3. Dot and density plots forLUM, lumican (A), and FABP4 (B) inthe TCGA and Tothill datasets.Caseswith no residual disease (R0)are in blue, cases with residualdisease in red. Black densitycurves represent the entire cohort.

Tucker et al.





In the cohort of Bonome and colleagues, we again see asubset of patients with very high levels of both FABP4 andADH1B (Fig. 4C), but the correlation between the 2 is not astight as for the TCGA and Tothill datasets, possibly as aresult of the microdissection used in the study of Bonomeand colleagues. In the cell line data from CCLE (Fig. 4D),joint overexpression of FABP4 and ADH1B does not occuramong the ovarian cancer lines, although there is clearoverexpression of each gene in some cell lines. This suggeststhe linkage we see in clinical samples may be due to sometype of stromal interaction.

Validation studiesIn general, levels of FABP4 in the validation studies were

markedly higher in tumor samples from omentum thanfromprimary tumor (Supplementary Appendix). Therefore,we restricted our attention to the samples from primarytumor for consistencywith the datasets of TCGA and Tothilland colleagues. The paired FABP4 and ADH1B qRT-PCRvalues for the primary ovarian cancer samples are shownin Fig. 5. Expression levels of the 2 genes are highly corre-lated, and there is no difference in distribution between thevalues from samples contributed by the 2 institutions.

Figure 4. Plots ofFABP4 againstADH1B for the TCGA (A), Tothill (B), Bonome (C), andCCLE (D) datasets. Cutoffs separating "low" and "high" expressionwerechosen by eye and loosely correspond to dips in density estimates of the corresponding distributions. The genes track together in the TCGA and Tothillcohorts, and rates of residual disease are higher when both genes are elevated. Tracking is more diffuse in the (microdissected) Bonome samples andessentially absent in the CCLE samples, suggesting the linkage may be driven by the microenvironment.






Although the correlations between the microarrayvalues for FABP4 and ADH1B in our discovery cohortssuggested that we would not need both for successfulprediction of high residual disease risk, it was not certain,a priori, that either assay would work. In addition, wewere interested in exploring the nature of the associationbetween FABP4 and ADH1B further. Therefore, weassayed expression levels of both genes in our validationstudies. However, after observing the results in Fig. 5, andin view of the lack of elevation of ADH1B in the CCLEdata, we proceeded to predict high risk of residual diseasein our validation cohort based on FABP4 expressionlevels alone.

The bimodal distribution seen in the microarray valuesfor FABP4 was not seen with the qRT-PCR values in thevalidation cohort, pointing to a dynamic range limitationfor the former. Therefore, we chose to select a cutoff forpredicting high risk of residual disease based on a specificquantile of FABP4 expression (Supplementary Appendix).Specifically, we selected a prediction threshold correspond-ing to the 25% of patients with the highest levels of FABP4expression; we predicted that these patients had a signifi-cantly higher incidence of residual disease than the remain-ing patients.

After unblinding the clinical findings concerning resid-ual disease in the validation cohort, we observed thatamong the 35 patients predicted to be at high risk forresidual disease, 30 (PPV 86%) did have residual disease.In contrast, only 54 of the 104 patients with FABP4 valuesbelow the decision threshold (52%) had incompleteresections. This difference is highly significant using a1-sided Fisher exact test (P ¼ 0.0002) and corresponds to

an OR of 5.5. Our predictive method based on FABP4,therefore, correctly identified a cohort of patients withsignificantly increased rates of residual disease in anindependent test set. Further examination of the ADH1Bresults (performed after unblinding) showed that predic-tions using either ADH1B alone or in combinationwith FABP4 would have produced similar results (Fig.5 and Supplementary Appendix).

To obtain insight into the potential biological role ofFABP4 and ADH1B, we used the reverse-phase proteinarray (RPPA) data available from TCGA to investigatewhich proteins, if any, had levels correlated with FABP4and ADH1B expression in the TCGA samples. While allcorrelations were low, there were nonetheless more sig-nificant associations than could be explained by chance(Supplementary Appendix). Many of the positively asso-ciated proteins are related to metastasis or proliferation,and among the proteins negatively associated with bothgenes is E-cadherin, which might imply greater levels ofepithelial–mesenchymal transition and would also beprometastatic.

DiscussionWe have shown that high expression levels of FABP4

and ADH1B in primary tumor specimens are associatedwith high incidence of residual disease after primarydebulking surgery of high-grade serous ovarian cancer. Ifconfirmed in the biopsy setting, this discovery lays thefoundation for customized surgical treatment approachesbased on unbiased and objective measures from the tumoritself. Triage to neoadjuvant chemotherapy versus primarydebulking could be prospectively assessed to provide anoptimized treatment algorithm based on intrinsic tumorbiology. For such use, biopsy from the primary versusmetastatic sites would also have to be carefully considered.Consistent with a report of Nieman and colleagues (16),we observed higher levels of FABP4 in samples from theomentum compared with primary tumor and in 4 patientswith samples available from both sites, FABP4 levels werehigher in metastatic tissue (Supplementary Appendix).These results suggest that if omental levels of FABP4expression are also associated with risk of residual disease,the FABP4 threshold for calling patients to be at high riskof residual disease might depend on whether primary ormetastatic tissue is sampled.

Our identification of FABP4 and ADH1B, which encodefor fatty acid–binding protein 4 and alcohol dehydroge-nase 1B, respectively, leveraged large, publicly availabledatasets, providing further evidence that such datasets(with clinical annotation) may be relevant for discoveringapproaches to tailor therapy. This assertion, however,comes with several caveats. First, results from almost allsuch studies can be tainted by batch effects (17). Tominimize the potential for such effects, we sought asso-ciations occurring in 2 different datasets and looked forconfirmation of expression patterns in at least one addi-tional publicly available dataset, before proceeding to

Figure 5. Plot of FABP4 versus ADH1B qRT-PCR values for the ovarianvalidation samples. Each unit change corresponds to a doubling ofmeasured intensities. Caseswith residual disease are in red. The solid lineis the cutoff used in our blinded validation; the dashed line is the cutoffassociated with weighting the 2 genes equally (with performancechecked after unblinding). RD, residual disease.

Tucker et al.





validation. Second, by selecting genes whose expressionlevels exhibited a wide dynamic range, we focused ondifferences large enough to survive migration to a differ-ent assay. This is in keeping with the use of "bimodal"genes for classification (18, 19) and is similar in spirit tothe "barcode" approach (20). Third, to keep the testpractical, we restricted attention to a very small numberof genes (here 2). Fourth, we used a clinical endpoint thatshould be consistent across centers (i.e., R0 resectioncontrasted with any residual disease).Although the extent of cytoreduction is still most com-

monly scored as optimal (<1 cm) versus suboptimal(�1 cm) resection, we elected to seek biomarkers of anyresidual disease, instead of suboptimal debulking, for 2reasons. First, there is growing evidence that there is alarger difference in overall survival between patients withR0 resection versus any residual disease than betweenoptimal versus suboptimal resection, as shown in our Fig.1 and elsewhere (3, 4, 9). Second, the distinction betweenR0 resections and any residual disease is expected to bemore reliably assessed and reproducible between institu-tions. Having elected to investigate the distinctionbetween residual disease and no residual disease, wesought biomarkers associated with residual diseaseinstead of R0 because there is a subset of patients inwhom complete resection is not achievable because ofwidely disseminated disease. This suggests the existenceof a subset of patients at high risk of residual diseasebecause of biologic tumor characteristics, whereas failureto achieve R0may occur simply because of tumor locationnear critical organs, and other medical reasons unrelatedto biologic characteristics of the tumor. We, therefore,expect that a biologic predictor for likelihood of R0disease may not achieve high predictive accuracy. Con-sistent with this hypothesis, although we find a contin-uous trend in the risk of residual disease as a function ofFABP4 expression levels (Supplementary Appendix), theincidence of R0 resection is not very high (�60%) amongpatients with the lowest levels of FABP4 in their primarytumors.The present study did not address the biologic mecha-

nisms that might underlie the ability of FABP4 and/orADH1B to serve as biomarkers for high risk of residualdisease. The study of Nieman and colleagues suggestedthat FABP4 plays a role in widely metastatic disease(16), which could explain the associated high risk ofresidual disease, and several studies have reported a roleof FABP4 in angiogenesis (21, 22). We found thatprotein levels measured by RPPA in the TCGA data andcorrelated with high FABP4 or ADH1B expression alsosuggest increased metastasis and proliferation. We fur-ther note that several collagen genes are included amongthe 47 genes meeting the 10% FDR in our discoveryanalyses (Supplementary Appendix); this might indicatean ability of the tumor to remodel the local environ-ment and promote metastases. Additional studies in-vestigating both genes are currently underway in ourlaboratory.

While current front-line treatment for patients withHGSOC consists of tumor debulking surgery followedby platinum- and taxane-based chemotherapy (23), somestudies point to comparable outcome with neoadjuvantchemotherapy followed by interval debulking surgery(24). An accurate predictor could allow "personalized"surgical therapy, whereby patients unlikely to have com-plete debulking would forego upfront surgery and wouldreceive neoadjuvant chemotherapy instead. With such anapproach, patients with a high likelihood of residualdisease at the upfront surgery could be spared from themorbidity associated with such procedures. Previousattempts to predict resectability have included use ofcomputerized tomography, which was not found to behighly predictive (25) and preoperative serum levels ofCA-125, which was found to be a significant predictor ofsuboptimal disease but with an OR lower than the valuefound here for prediction of residual disease using FABP4(26). Consistent with our observation that low FABP4expression does not accurately identify patients withoutresidual disease, the latter study found that CA-125 lacksthe ability to predict optimal cytoreduction. An ongoingstudy at our institution is investigating the utility of alaparoscopy-based assessment of the potential to achieveR0 resection, which could potentially complement thebiomarkers proposed here for detecting high risk ofresidual disease. Development of a CLIA-compliant assayfor FABP4 is currently underway to enable prospectivevalidation of our proposed molecular biomarkers. Inaddition, ongoing biologic studies of FABP4 and ADH1Bmay elucidate the mechanisms behind some of the mostrefractory cases of HGSOC and point to new avenues fortherapeutic intervention.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

Authors' ContributionsConception and design: S.L. Tucker, A.K. Unruh, R.L. Coleman, S.Y. Wu,G. Lopez-Berestein, B. Ozpolat, K.A. Baggerly, A.K. SoodDevelopment of methodology: K.A. Baggerly, A.K. SoodAcquisition of data (provided animals, acquired and managedpatients, provided facilities, etc.): K. Gharpure, A.M. Nick, R.L. Cole-man, J. Guenthoer, H.J. Dalton, S.Y. Wu, R. Rupaimoole, W. Hu, A.K.SoodAnalysis and interpretation of data (e.g., statistical analysis, biosta-tistics, computational analysis): S.L. Tucker, K. Gharpure, S.M. Herbrich,A.K. Unruh, A.M.Nick, R.L. Coleman,H.J. Dalton, R. Rupaimoole, G. Lopez-Berestein, C. Ivan, K.A. Baggerly, A.K. SoodWriting, review, and/or revision of the manuscript: S.L. Tucker,K. Gharpure, S.M. Herbrich, A.K. Unruh, A.M. Nick, E.K. Crane, R.L. Cole-man, J. Guenthoer, H.J. Dalton, S.Y.Wu, R. Rupaimoole, G. Lopez-Berestein,B. Ozpolat, K.A. Baggerly, A.K. SoodAdministrative, technical, or material support (i.e., reporting or orga-nizing data, constructing databases): S.L. Tucker, S.M. Herbrich, R.L.Coleman, A.K. SoodStudy supervision: A.K. Sood

AcknowledgmentsThe authors thank Drs. Charles Drescher andMuneesh Tewari at the Fred

Hutchinson Cancer Research Center for helpful discussions and for provid-ing tumor samples.






Grant SupportThis work was supported in part by the National Cancer Institute at the

NIH (CA016672, CA101642, CA109298, CA128797, P50CA083639,P50CA098258, U54CA151668, U54CA96300, U54CA96297, CA177909,and U24 CA143883); the Cancer Prevention Research Institute of Texas(RP110595); the Department of Defense (BC085265, OC073399,W81XWH-10-1-0158); the Ovarian Cancer Research Fund (Program ProjectDevelopment Grant); the Chapman Foundation; the Betty Ann AscheMurrayDistinguished Professorship; theMeyer and IdaGordon Foundation,the Judi A. Rees Ovarian Cancer Research Fund, the Marcus Foundation; the

RGK Foundation; the Gilder Foundation; and the Blanton-Davis OvarianCancer Research Program.

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.

Received February 22, 2014; revised April 4, 2014; accepted April 8, 2014;published OnlineFirst April 22, 2014.

References1. Aletti GD, Dowdy SC, Gostout BS, Jones MB, Stanhope CR, Wilson

TO, et al. Aggressive surgical effort and improvedsurvival in advanced-stage ovarian cancer. Obstet Gynecol 2006;107:77–85.

2. Bookman MA, Brady MF, McGuire WP, Harper PG, Alberts DS,Friedlander M, et al. Evaluation of new platinum-based treatmentregimens in advanced-stage ovarian cancer: a phase III trial of thegynecologic cancer intergroup. J Clin Oncol 2009;27:1419–25.

3. Chang SJ, Bristow RE, Ryu HS. Impact of complete cytoreductionleaving no gross residual disease associated with radical cytoreduc-tive surgical procedures on survival in advanced ovarian cancer. AnnSurg Oncol 2012;19:4059–67.

4. du Bois A, Reuss A, Pujade-Lauraine E, Harter P, Ray-Coquard I,Pfisterer J. Role of surgical outcome as prognostic factor in advancedepithelial ovarian cancer: a combined exploratory analysis of 3 pro-spectively randomized phase 3 multicenter trials: by the Arbeitsge-meinschaft Gynaekologische Onkologie Studiengruppe Ovarialkarzi-nom (AGO-OVAR) and theGroupe d'Investigateurs Nationaux Pour lesEtudes desCancers de l'Ovaire (GINECO). Cancer 2009;115:1234–44.

5. Bristow RE, Tomacruz RS, Armstrong DK, Trimble EL, Montz FJ.Survival effect of maximal cytoreductive surgery for advanced ovariancarcinoma during the platinum era: a meta-analysis. J Clin Oncol2002;20:1248–59.

6. Ferrandina G, Sallustio G, Fagotti A, Vizzielli G, Paglia A, Cucci E, et al.Role of CT scan-based and clinical evaluation in the preoperativeprediction of optimal cytoreduction in advanced ovarian cancer: aprospective trial. Br J Cancer 2009;101:1066–73.

7. Rodriguez N, Rauh-Hain JA, Shoni M, Berkowitz RS, Muto MG,Feltmate C, et al. Changes in serum CA-125 can predict optimalcytoreduction to no gross residual disease in patients with advancedstage ovarian cancer treated with neoadjuvant chemotherapy. Gyne-col Oncol 2012;125:362–6.

8. Angioli R, Plotti F, Capriglione S, Aloisi A, Montera R, Luvero D, et al.Can the preoperative HE4 level predict optimal cytoreduction inpatients with advanced ovarian carcinoma? Gynecol Oncol2013;128:579–83.

9. Integrated genomic analyses of ovarian carcinoma. Nature 2011;474:609–15.

10. Tothill RW, Tinker AV, George J, Brown R, Fox SB, Lade S, et al. Novelmolecular subtypes of serous and endometrioid ovarian cancer linkedto clinical outcome. Clin Cancer Res 2008;14:5198–208.

11. Bonome T, Levine DA, Shih J, Randonovich M, Pise-Masison CA,Bogomolniy F, et al. A gene signature predicting for survival in sub-optimally debulked patients with ovarian cancer. Cancer Res 2008;68:5478–86.

12. Barretina J, Caponigro G, Stransky N, Venkatesan K,Margolin AA, KimS, et al. The Cancer Cell Line Encyclopedia enables predictive model-ling of anticancer drug sensitivity. Nature 2012;483:603–7.

13. Bolstad BM, Irizarry RA, Astrand M, Speed TP. A comparison ofnormalization methods for high density oligonucleotide array databased on variance and bias. Bioinformatics 2003;19:185–93.

14. Ruijter JM, Ramakers C, Hoogaars WMH, Karlen Y, Bakker O, van denHoffMJB, et al. Amplification efficiency: linking baseline and bias in theanalysis of quantitative PCR data. Nucleic Acids Res 2009;37:e45.

15. Benjamini Y, Hochberg Y. Controlling the false discovery rate: Apractical and power-ful approach to multiple testing. J R Stat Soc B1995;57:289–300.

16. Nieman KM, Kenny HA, Penicka CV, Ladanyi A, Buell-Gutbrod R,Zillhardt MR, et al. Adipocytes promote ovarian cancer metasta-sis and provide energy for rapid tumor growth. Nat Med 2011;17:1498–503.

17. Leek JT, Scharpf RB, BravoHC, SimchaD, LangmeadB, JohnsonWE,et al. Tackling the widespread and critical impact of batch effects inhigh-throughput data. Nat Rev Genet 2010;11:733–9.

18. Kernagis DN, Hall AH, Datto MB. Genes with bimodal expressionare robust diagnostic targets that define distinct subtypes of epi-thelial ovarian cancer with different overall survival. J Mol Diagn2012;14:214–22.

19. Wang J, Wen S, Symmans WF, Pusztai L, Coombes KR. The bimo-dality index: a criterion for discovering and ranking bimodal signaturesfrom cancer gene expression profiling data. Cancer Inform 2009;7:199–216.

20. Zilliox MJ, Irizarry RA. A gene expression bar code for microarray data.Nat Methods 2007;4:911–3.

21. Elmasri H, Karaaslan C, Teper Y, Ghelfi E,WengM, Ince TA, et al. Fattyacid binding protein 4 is a target of VEGF and a regulator of cellproliferation in endothelial cells. FASEB J 2009;23:3865–73.

22. Kajimoto K, Takayanagi S, Sasaki S, Akita H, Harashima H. RNAinterference-based silencing reveals the regulatory role of fattyacid-binding protein 4 in the production of IL-6 and vascularendothelial growth factor in 3T3-L1 adipocytes. Endocrinology2012;153:5629–36.

23. Vaughan S, Coward JI, Bast RC Jr, Berchuck A, Berek JS, Brenton JD,et al. Rethinking ovarian cancer: recommendations for improvingoutcomes. Nat Rev Cancer 2011;11:719–25.

24. Vergote I, Trope CG, Amant F, Kristensen GB, Ehlen T, Johnson N,et al. Neoadjuvant chemotherapy or primary surgery in stage IIIC or IVovarian cancer. N Engl J Med 2010;363:943–53.

25. Axtell AE, Lee MH, Bristow RE, Dowdy SC, Cliby WA, Raman S, et al.Multi-institutional reciprocal validation studyof computed tomographypredictors of suboptimal primary cytoreduction in patients withadvanced ovarian cancer. J Clin Oncol 2007;25:384–9.

26. Kang S, Kim TJ, NamBH, Seo SS, Kim BG, Bae DS, et al. Preoperativeserum CA-125 levels and risk of suboptimal cytoreduction in ovariancancer: a meta-analysis. J Surg Oncol 2010;101:13–7.


Tucker et al.




2014;20:3280-3288. Published OnlineFirst April 22, 2014.Clin Cancer Res Susan L. Tucker, Kshipra Gharpure, Shelley M. Herbrich, et al. of High-Grade Serous Ovarian CancerMolecular Biomarkers of Residual Disease after Surgical Debulking

Updated version

10.1158/1078-0432.CCR-14-0445doi:

Access the most recent version of this article at:

Material

Supplementary

http://clincancerres.aacrjournals.org/content/suppl/2014/04/22/1078-0432.CCR-14-0445.DC1

Access the most recent supplemental material at:

Cited articles

http://clincancerres.aacrjournals.org/content/20/12/3280.full#ref-list-1

This article cites 26 articles, 5 of which you can access for free at:

Citing articles

http://clincancerres.aacrjournals.org/content/20/12/3280.full#related-urls

This article has been cited by 5 HighWire-hosted articles. Access the articles at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://clincancerres.aacrjournals.org/content/20/12/3280To request permission to re-use all or part of this article, use this link



http://clincancerres.aacrjournals.org/lookup/doi/10.1158/1078-0432.CCR-14-0445

http://clincancerres.aacrjournals.org/content/suppl/2014/04/22/1078-0432.CCR-14-0445.DC1

http://clincancerres.aacrjournals.org/content/20/12/3280.full#ref-list-1

http://clincancerres.aacrjournals.org/content/20/12/3280.full#related-urls

http://clincancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

http://clincancerres.aacrjournals.org/content/20/12/3280


molecular biomarkers of residual disease after surgical ... · erin k. crane 2, robert l. coleman ,...

Documents