Volume 10 Number 11 November 2008 pp. 1285–1294 1285 www.neoplasia.com

† Sooryanarayana Varambally*, , Golgi GOLM1 Is a ,† Bharathi Laxman*, Rohit Mehra* , Qi Cao*, Tissue and Urine Biomarker Saravana M. Dhanasekaran*, Scott A. Tomlins*, 1,2 Jill Granger*, Adaikkalam Vellaichamy*, ‡ ‡ of Arun Sreekumar*, Jianjun Yu , Wenjuan Gu , ‡ ‡ Ronglai Shen , Debashis Ghosh , Lorinda M. Wright§, Raleigh D. Kladney¶, Rainer Kuefer#, Mark A. Rubin**, Claus J. Fimmel§, † †† and Arul M. Chinnaiyan*, ,

*Michigan Center for Translational Pathology, Department of Pathology, University of Michigan Medical School, Ann Arbor, MI 48109, USA; †Comprehensive Cancer Center, University of Michigan Medical School, Ann Arbor, MI 48109, USA; ‡Department of Biostatistics, University of Michigan, Ann Arbor, MI 48109, USA; §Edward Hines VA Medical Center, Hines, IL and Division of Gastroenterology, Hepatology and Nutrition, Loyola University, Stritch School of Medicine, Maywood, IL 60153, USA; ¶Division of Urologic Surgery, Washington University School of Medicine, St. Louis, MO 63110, USA; #Department of Urology, University of Ulm, Ulm, Germany; **Department of Pathology and Laboratory Medicine, Cornell University, New York, NY 10021, USA; ††Department of Urology, University of Michigan Medical School, Ann Arbor, MI 48109, USA

Abstract Prostate cancer is the most common type of tumor found in American men and is the second leading cause of cancer death in males. To identify biomarkers that distinguish prostate cancer from normal, we compared multiple expression profiling studies. Through meta-analysis of expression array data from multiple prostate cancer studies, we identified GOLM1 (Golgi membrane protein 1, Golm 1) as consistently up-regulated in clinically local- ized prostate cancer. This observation was confirmed by reverse transcription–polymerase chain reaction (RT-PCR) and validated at the protein level by immunoblot assay and immunohistochemistry. Prostate epithelial cells were identified as the cellular source of GOLM1 expression using laser capture microdissection. Immunohistochemical staining localized the GOLM1 signal to the subapical cytoplasmic region, typical of a Golgi distribution. Surprisingly,

Abbreviations: Golm1, Golgi membrane protein 1; PSA, prostate-specific antigen; PIN, prostatic intraepithelial neoplasia; BPH, benign prostatic hyperplasia; LCM, laser cap- ture microdissection Address all correspondence to: Arul M. Chinnaiyan, MD, PhD, Department of Pathology, University of Michigan Medical School, 1400 E. Medical Center Dr. 5316 CCGC, Ann Arbor, MI 48109-0602. E-mail: [email protected]; and Sooryanarayana Varambally, PhD, Department of Pathology, University of Michigan Medical School, 1400 E. Medical Center Dr. 5430 CCGC, Ann Arbor, MI 48109. E-mail: [email protected] 1This research was supported in part by the National Institutes of Health Prostate SPORE P50 CA69568 (to A.M.C.); Department of Defense Grants PC040517 (to R.M.), PC051081 (to A.M.C. and S.V.). C.J.F. was supported by a VA Merit Review. 2This article refers to supplementary materials, which are designated by Tables W1 and W2 and are available online at www.neoplasia.com. Received 4 August 2008; Revised 2 September 2008; Accepted 2 September 2008 Copyright © 2008 Neoplasia Press, Inc. All rights reserved 1522-8002/08/$25.00 DOI 10.1593/neo.08922 1286 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. Neoplasia Vol. 10, No. 11, 2008

GOLM1 immunoreactivity was detected in the supernatants of prostate cell lines and in the urine of patients with prostate cancer. The mechanism by which intact GOLM1 might be released from cells has not yet been elucidated. GOLM1 transcript levels were measured in urine sediments using quantitative PCR on a cohort of patients present- ing for biopsy or radical prostatectomy. We found that urinary GOLM1 mRNA levels were a significant predictor of prostate cancer. Further, GOLM1 outperformed serum prostate-specific antigen (PSA) in detecting prostate can- cer. The area under the receiver-operating characteristic curve was 0.622 for GOLM1 (P = .0009) versus 0.495 for serum PSA (P = .902). Our data indicating the up-regulation of GOLM1 expression and its appearance in pa- tients’ urine suggest GOLM1 as a potential novel biomarker for clinically localized prostate cancer.

Neoplasia (2008) 10, 1285–1294

Introduction tients with established prostate cancer. The secretion from cell lines Prostate cancer is the most commonly diagnosed malignancy and is could be inhibited by treating cells with the protein transport inhib- a leading cause of cancer-related death in the Western male popu- itor brefeldin A [20,21]. Our observation suggests that GOLM1 has lation [1,2]. Early diagnosis is critical for the effective treatment of potential as a noninvasive biomarker of localized prostate cancer. malignant tumors. High-throughput technologies such as DNA and protein microarray have enabled the identification of and their corresponding that are differentially regulated in malignant Materials and Methods conditions [3,4]. These high throughput studies have offered re- searchers a better understanding of the disease and the molecular Quantitative Real-time Polymerase Chain Reaction circuitries that are dysregulated in cancer. Our previous DNA mi- To validate GOLM1 overexpression observed in multiple gene ex- croarray studies using prostate cancer tissue RNA have identified pression profiling studies, we performed quantitative real-time poly- multiple genes, including alpha methylacyl coenzyme A racemase merase chain reaction (qPCR) for GOLM1 expression using SYBR (AMACR), Enhancer of Zeste Homolog 2 (EZH2), tumor protein green [6,22]. Briefly, cDNA was made with the total RNA isolated TPD52, and ERG, as dysregulated in localized and metastatic pros- from 11 benign prostatic hyperplasia (BPH), 27 localized prostate tate cancer [5–8]. Many of these findings provided new insights cancers, and 8 metastatic prostate cancer samples. The quantification into the biology of prostate carcinogenesis and have identified the of cDNA in each sample was performed by interpolating a C t value next generation of candidate biomarkers and potential therapeutic from a standard curve of C t values obtained from serially diluted, targets [9–12]. In the case of AMACR, a peroxisomal fatty acid– commercially prepared cDNA pooled normal prostate samples metabolizing enzyme, the up-regulated protein was detectable in (Clontech, Palo Alto, CA). This calculated quantity of GOLM1 patients’ urine, which indicated its potential use for noninvasive di- was then normalized against the quantity of the housekeeping gene agnostic studies [13]. Furthermore, our studies indicated an immune glyceraldehyde-3 phosphate dehydrogenase (GAPDH) expressed in response to AMACR in prostate cancer patients and autoantibodies that sample. “No–reverse transcription” controls were included to ex- directed against AMACR that could be detected in the patient’s se- clude amplification of the genomic DNA. The primer sequences rum [14]. used for GOLM1 were: 5′-CAGCGTGAAAAGCGGAATC-3′ and GOLM1 (Golm 1, NM_016548) is a resident cis-Golgi membrane 5′-TCGGCCCTGTTGTGAAATA-3′. GAPDH primers were pre- protein of unknown function. The first evidence of its up-regulation pared as described by Vandesompele et al. [23]. was shown in the hepatocytes of patients with acute and chronic Urine samples were obtained from 333 patients with informed forms of hepatitis and hepatocellular cancer [15]. GOLM1 has a sin- consent after a digital rectal examination before either needle biopsy gle N-terminal transmembrane domain and an extensive C-terminal, (n = 269) or radical prostatectomy (n = 64) at the University of coiled-coil domain that faces the luminal surface of the Golgi appa- Michigan Health System with institutional review board approval ratus. N-terminal cleavage by a furin proprotein convertase resulted as described earlier (Table W1) [19]. Isolation of RNA from urine in the release of the C-terminal ectomain and its appearance in serum and Transplex Whole Transcriptome Amplification (WTA) were as [16]. The cleaved form of GOLM1 was detectable in the serum of described [24]. Quantitative polymerase chain reaction was used to patients with hepatocellular cancer, a finding that may have diagnos- detect GOLM1 and the control transcripts PSA and GAPDH from tic value [17]. Initial array studies in our laboratory, WTA amplified cDNA essentially as described [24]. The primer and by others [3,18] suggested increased expression levels of GOLM1 sequences for GOLM1 [19], GAPDH [23], and PSA [25] were mRNA in prostate cancer tissues. We subsequently detected GOLM1 previously described. Quantitative polymerase chain reaction was transcripts in patient urine samples. By multiplexing with other urine performed on WTA cDNA from urine collected from 129 biopsy- markers, we demonstrated that GOLM1 mRNA levels can serve as negative patients and 204 patients with prostate cancer (140 biopsy- significant predictors of prostate cancer [19]. Our study confirms positive patients and 64 prostatectomy patients). Samples that had – the epithelial cell specific expression of GOLM1 in prostate cancer PSA C t values > 28 were excluded to ensure sufficient prostate cell tissues. Furthermore, we show that the GOLM1 protein is released collection, leading to 124 biopsy-negative and 195 samples from pa- from prostate cell lines in vitro and is detectable in the urine of pa- tients with prostate cancer in the analysis. This resulted in a final data Neoplasia Vol. 10, No. 11, 2008 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. 1287 set of samples from 195 patients with prostate cancer (133 positive and pathology data from patients with biopsy-proven clinically local- needle biopsy and 62 radical prostatectomy) and 124 biopsy-negative ized prostate cancer used in this study is provided in Table W2. patients. Threshold levels were set during the exponential phase No treatment was administered to the subjects in the interval be- of the qPCR reaction using Sequence Detection Software version tween biopsy and urine sample collection. As controls, urine samples 1.2.2 (Applied Biosystems, Foster City, CA), with the same baseline from 50 male subjects (mean ± SD, 57.9 ± 0.1 years) with no known and threshold set for each plate, to generate threshold cycle (C t) va- history of prostate cancer were collected from the University of lues for all genes for each sample. We adjusted GOLM1 was against Michigan Clinical Pathology Laboratories. Within 2 days of collec- its mean urine PSA and GAPDH values (2(CtPSA + C tGAPDH ) /2− tion, the urine samples were concentrated at 4°C using a Biomax C tGOLM1) × 1000). Ultrafree concentrator (5000 MW cutoff; Millipore Corporation, Bedford, MA). The protein content of each sample was estimated ’ Statistical Analysis with Bradford reagent following the manufacturer s instructions The receiver-operating characteristic (ROC) curve and a Box plot (Bio-Rad Laboratories, Hercules, CA). The concentrated urine sam- − for GOLM1 were generated using Statistical Package for the Social ples were frozen at 20°C until use. Twenty micrograms of total pro- Sciences 11.5 (SPSS, Inc., Chicago, IL). The area under the ROC tein from each urine sample was electrophoresed and analyzed for curve (AUC-ROC) was then calculated for GOLM1 and serum GOLM1 protein expression by immunoblot analysis as described PSA, respectively. above. The intensity of the 74-kDa GOLM1 band in each sample was scored visually by a researcher who was blinded to the diagnosis of samples and the pattern of loading in the gels. A GOLM1-positive Laser Capture Microdissection control sample (LNCaP total cell extract) was used for immunoblot Normal and cancerous prostate epithelial cells were collected using experiments. While scoring, the band in the sample lanes with high- laser capture microdissection (LCM) as described previously [26]. est reactivity was assigned a score of 4 (highest level of reactivity), μ Briefly, for LCM, the SL Microtest device with CUT software whereas absence of a band was scored 0 (non reactivity). Intermediate was applied (MMI GmbH, Heidelberg, Germany). A total area of 2 band intensities were assigned as follows: weak (score = 1), interme- 6mm was cut for all of the samples and was collected with the diate (score = 2), and high (score = 3). The mean values for GOLM1 aid of the adhesive surface lid from a specially manufactured tube reactivity were presented as population mean values with 95% con- (MMI GmbH). Approximately 10,000 cells collected from each fidence intervals. Student’s t test (2-sided) was used to test for statis- μ sample were lysed with 30 l of 5× SDS-reducing sample buffer tically significant differences in GOLM1 reactivity between patients and clarified by centrifugation at 12,000g for 10 minutes. The sam- with prostate cancer and control subjects. No adjustment for multi- ples were then loaded onto a gel for immunoblot analysis. ple testing was made during the analysis. P values less than or equal to 0.05 were considered statistically significant. Receiver-operating Immunoblot Analysis characteristic curves were used to assess the sensitivity and specificity Normal and prostate cancer tissues were homogenized in NP-40 of urine-associated GOLM1 to detect prostate cancer. lysis buffer containing 50 mM Tris-HCl, pH 7.4, 1% Nonidet P-40 To test the secretion of the GOLM1 protein, a flag-tagged GOLM1 (Sigma, St. Louis, MO) and complete protease inhibitor cocktail construct was overexpressed in HEK293 Phoenix cells. Cell culture (Roche, Indianapolis, IN). Fifteen micrograms of protein extract was supernatants were collected 12 and 24 hours after transfection. For mixed with SDS sample buffer and run on a 10% SDS–polyacrylamide the analysis, 5 μl of culture supernatant from GOLM1-transfected, gel under reducing conditions. The separated proteins were transferred vector-transfected, and control protein-transfected cells were run on onto polyvinylidene fluoride membranes (Amersham Pharmacia Bio- SDS-PAGE. Immunoblots were probed with either flag or GOLM1 tech, Piscataway, NJ), which were incubated for 1 hour in blocking antibodies. Likewise, the presence of GOLM1 was also assessed in buffer [Tris-buffered saline with 0.1% Tween (TBS-T) and 5% nonfat the culture supernatants of LNCaP (prostate cell line) and DU145 dry milk]. GOLM1 rabbit polyclonal antibody (raised against amino (prostate cancer cell line). Prostate cancer cell lines were cultured acids 41-400) was applied at a 1:50,000 dilution in blocking buffer in RPMI medium containing 10% FBS, and the medium was overnight at 4°C. After washing with TBS-T buffer, the membrane changed to 5% FBS 12 hours before the collection of the superna- was incubated with horseradish peroxidase–linked donkey, anti–rabbit tant to reduce the bovine serum protein content. Culture superna- IgG antibody (GE Healthcare Life Sciences, Piscataway, NJ) at 1:5000 tants were collected at 24 hours and analyzed by SDS-PAGE and for 1 hour at room temperature. The signals were visualized with the immunoblot analysis with GOLM1-specific antibody. To investi- ECL detection system (GE Healthcare) and autoradiography. For gate the specificity of the secretion, the GOLM1 antibody was pre- GAPDH and actin Western blots, the GOLM1 antibody probed treated with 5 μg of recombinant GOLM1 protein [15]. Cell-free membranes were stripped with western reprobe buffer (Geno-tech, medium served as the negative control. St. Louis, MO), blocked in TBS-T containing 5% nonfat dry milk, and incubated with either rabbit anti-GAPDH antibodies (1:50,000 In Vitro Overexpression of GOLM1 dilution;Abcam,Cambridge,MA)orrabbitanti–actin antibodies A mammalian expression construct of GOLM1 was generated by (1:5000 dilution; Sigma) for 2 hours. For laser capture microdissected subcloning the PCR product into the pACRSVpLpA(−)loxP-SSP prostate epithelial cell immunoblot, the cells were lysed in sample vector (UMICH vector core). Primers used were GOLM1-KpnI-6- buffer and analyzed for GOLM1 protein expression by immunoblot His-F: 3′-GCGGGTACCATGCACCACCACCACCACCACGG- analysis as described above. CTTGGGAAACGGGCGTCGCAGC-5′ and GOLM1-XbaI- At the time of initial diagnosis, urine samples (n = 52) from biopsy- FLAG-R: 5′-GCGTCTAGATCAAAGCTTGTCGTCATCGTCT- proven, clinically localized prostate cancer patients (mean ± SD age, TTGTAGTCGAGTGTATGATTCCGCTTTTC-3′. HEK293 58.1 ± 0.60 years) were collected with informed consent. Clinical Phoenix cells were transfected with the flag-GOLM1 construct as 1288 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. Neoplasia Vol. 10, No. 11, 2008 well as vector control using FuGENE6 (Roche) transfection reagent. TMA construction. All TMAs were assembled using the manual tis- Culture supernatants were collected 12 and 24 hours after transfec- sue arrayer (Beecher Instruments, Silver Spring, MD). At least three tion for immunoblot analysis. tissue cores were sampled from each donor block. Histologic diagno- sis of the tissue cores was verified by standard hematoxylin and eosin staining of the initial TMA slide. This radical prostatectomy series is Immunohistochemistry part of the University of Michigan Prostate Cancer Specialized Pro- For the development of the tissue microarray (TMA), prostate gram of Research Excellence Tissue Core, with informed consent of cancer and normal prostate tissues were embedded in paraffin. Study the patients and prior institutional review board approval. Standard pathologists (M.A.R. and R.M.) reviewed the slides of all cases and biotin-avidin complex immunohistochemistry was performed using designated areas of interest. These slides were used as a template for a polyclonal anti-GOLM1 antibody. Digital images were acquired

Figure 1. GOLM1 expression in prostate cancer. (A) GOLM1 transcript levels were collected from DNA microarray analysis of 101 pros- tate samples. Cy3/Cy5 ratios indicate the expression of GOLM1 in prostate tissue RNA compared to RNA from normal prostate pool. BPH indicates benign prostatic hyperlasia; MET, metastatic prostate cancer; NAP, normal adjacent prostate; PCA, prostate cancer. (B) GOLM1 expression in prostate cancer profiling studies from different publicly available cancer microarray data sets in Oncomine [3,18,26,40] and Yang et al. (unpublished data). Expression array analysis of multiple prostate cancer microarray data sets were collected and analyzed, and statistical significance was calculated. (For details visit: www.oncomine.org). Class 1 represents the GOLM1 expres- sion in normal tissues and class 2 represents GOLM1 expression in cancer. (C) Quantitative polymerase chain reaction confirmation of increased expression of GOLM1 transcripts in prostate cancer. Quantitative polymerase chain reaction was done using RNA from benign, prostate cancer, and metastatic prostate cancer tissue samples. Quantitative polymerase chain reaction of glyceraldehyde-3-phosphate dehydrogenase served as the internal control. Neoplasia Vol. 10, No. 11, 2008 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. 1289

Figure 2. Immunoblot analyses of the GOLM1 protein in prostate tissue lysates. (A) Immunoblot analysis of GOLM1 protein expression using benign and prostate cancer tissue lysates. Total prostate tissue lysates (10-μg total protein) were analyzed using anti–GOLM1 rabbit polyclonal antibody. Ten micrograms of total tissue lysate from benign prostate tissue, prostate cancer, and metastatic prostate cancer tissues were analyzed by immunoblot. The blot was reprobed with GAPDH antibody to confirm equal loading. (B) The prostate epithelial cells were isolated by LCM technique and were analyzed by immunoblot analysis. The cells from benign adjacent tissue, PIN, prostate cancer, and metastatic cancer were used for this analysis. The blots were reprobed with anti–β-actin antibody.

using the BLISS Imaging System (Bacus Laboratory, Lombard, IL) up-regulation of GOLM1 expression was corroborated at the protein and evaluated using a previously validated Web-based tool (TMA level by immunoblot analysis with lysates from prostate tissues Profiler, University of Michigan, Ann Arbor, MI) [27]. Staining was (Figure 2A). There is a marked up-regulation of GOLM1 protein scored as negative (score = 1), weak (score = 2), moderate (score = in prostate cancer tissues compared to benign adjacent specimens, 3), and strong (score = 4) based on the intensity of staining of tumor whereas metastatic tissue displayed a moderate up-regulation. Further cells using a similar system that has been validated previously [28]. verification and data on cell type–specific expression was obtained by performing immunoblot analysis on laser capture microdissected Immunofluorescence Staining and Confocal Microscopy samples. Epithelial cells from BPH, prostatic intraepithelial neoplasia Prostate tissue sections were soaked in xylene for 1 hour to remove (PIN), localized prostate cancer (PCA), and hormone refractory met- paraffin. Antigen retrieval was achieved by heating the slides in citrate astatic prostate cancer (MET) obtained by LCM were analyzed by buffer (pH 6.0) for 15 minutes in a pressure cooker. The slides were Western blot analysis with anti-GOLM1 and subsequently with then blocked in PBS-T containing 5% normal donkey serum for anti–β-actin (loading control) antibodies (Figure 2B). Laser capture 1 hour. Slides were incubated overnight at 4°C with a mixture of microdissected samples showed up-regulation of GOLM1 protein in mouse anti–E-cadherin (BD Biosciences, San Diego, CA) and rabbit PCA and MET samples, demonstrating that the up-regulation ob- GOLM1 antibodies at 1:200 and 1:10,000 dilutions, respectively, served in whole tissue lysates (Figure 2A) was caused by an increase washed, and followed by secondary antibodies (anti–mouse Alexa in GOLM1 expression within the cancer epithelia. The apparent de- 555 and anti–rabbit Alexa 488 at 1:1000 dilutions) for 1 hour. Slides crease in GOLM1 expression in MET compared to PCA was intrigu- were mounted using Vectashield mounting medium containing DAPI ing and was previously noted by Chandran et al. [30]. (Vector Laboratories, Burlingame, CA) after washing with PBS-T and Immunofluorescence staining and confocal imaging of prostate tis- PBS. Confocal images were taken with a Zeiss LSM510 META (Carl sue section stained with GOLM1-specific antibody not surprisingly Zeiss, Göttingen, Germany) imaging system using an ultraviolet, ar- revealed a Golgi staining pattern. There was a clear increase in gon, and helium neon 1 light source. The triple color images were GOLM1 staining in prostate cancer epithelia compared to normal exported as TIFF images. epithelial cells (Figure 3A, pink arrows). High-density TMA analyses indicated moderate-to-strong GOLM1 protein expression in clini- cally localized prostate cancer samples with predominant cytoplasmic Results localization (Figure 3B). Expression levels of GOLM1 protein in Gene expression profiling studies from our group using prostate malignant epithelia were greater compared to those of benign tissue. tissue RNA demonstrated that GOLM1, a type II Golgi membrane Scoring of the TMAs indicated that the maximum staining (highest protein–coding gene, was up-regulated and consistently overexpressed score of 4) was observed in prostate cancer, whereas PIN demonstrated in prostate cancer (Figure 1A). Oncomine [29] (www.oncomine.org) moderate increases when compared to normal prostate analysis of various prostate cancer microarray data sets revealed up- (Figure 3C). regulation of GOLM1 in prostate cancer samples compared to normal We next sought to detect GOLM1 protein in prostate cell culture (Figure 1B). Quantitative real-time PCR revealed consistent up- supernatant to probe whether it is secreted by prostate cancer cells. A regulation of GOLM1 transcripts in prostate cancer samples confirm- previous study that detected the presence of GOLM1 in serum of ing the gene expression profiling study findings (Figure 1C). The hepatocellularcancerpatients[17]supportedthisnotion.Hence, 1290 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. Neoplasia Vol. 10, No. 11, 2008

Figure 3. Immunofluorescence and immunohistochemical analyses of GOLM1 expression. (A) GOLM1 staining and confocal imaging of prostate cancer epithelial cells. Representative GOLM1 staining is shown in green. E-cadherin (red) was used for costaining, and nuclei were stained with DAPI (blue). Pink arrows indicate the overexpressed GOLM1 protein stained in green. The white arrow indicates a lower level of GOLM1 expression in normal epithelium. (B) The TMA analyses of prostate tissues using anti–GOLM1 antibody. Intense staining of GOLM1 is seen in cancerous prostate epithelial cells. (C) Correlations of GOLM1 staining intensity in benign, PIN, and pros- tate cancer. Neoplasia Vol. 10, No. 11, 2008 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. 1291 the presence of a specific band in immunoblots of culture superna- coimmunostaining with Golgi marker Golgin 97 confirmed this ob- tant from HEK293 Phoenix cells that overexpressed Flag-tagged servation (Figure 4D, panel iii). However, cells treated with brefeldin GOLM1, with either FLAG antibody or GOLM1-specific antibody A showed diffuse staining (Figure 4D, panels iv and v) and colocali- (Figure 4A), suggested the presence of a secretory form of GOLM1. zation with the endoplasmic reticulum marker protein disulfide This signal was abrogated when GOLM1 antibody was preincubated isomerase (Figure 4D, panel vi). This observation suggested interfer- with purified recombinant protein, showing the specificity of this sig- ence with anterograde transport and increased retrograde transport nal (Figure 4B). In addition, immunoblots also revealed secretion of [20,21] and redistribution of GOLM1 from the Golgi apparatus endogenous GOLM1 by DU145 and LNCaP prostate cell lines, to the endoplasmic reticulum, which explains the reduced GOLM1 which was inhibited by treatment with 0.5 μMbrefeldinA(Fig- in the culture supernatant after brefeldin A treatment. ure 4C), a fungal lactone and inhibitor of protein trafficking. This Having established the presence of an endogenous secretory form of result suggests secretion of Golm1 through the anterograde secretory GOLM1, we further investigated the possibility of detecting GOLM1 pathway [20,21]. Brefeldin also blocked the secretion of prostate- in patient serum and urine samples. Immunoblot analysis of urine specific antigen (PSA) by LNCaP cells (data not shown). samples from prostate cancer patients showed a specific band To determine the subcellular localization of GOLM1 in prostate corresponding to GOLM1 comparable in molecular weight to cells, we performed fluorescence immunostaining with GOLM1 LNCaP lysate. A representative immunoblot is depicted in Fig- antibody using LNCaP cell line. The immunostaining indicated spe- ure 5A. Of note, inhibition of GOLM1 reactivity in the urine cific Golgi staining as anticipated (Figure 4D, panels i and ii), and was observed after preincubation of GOLM1 antibody with the

Figure 4. Detection of GOLM1 protein in cell culture medium. (A) Culture supernatants were collected from HEK293 Phoenix cells over- expressing Flag-GOLM1 construct and immunoblotted with either anti–Flag mouse monoclonal antibody or GOLM1 rabbit polyclonal antibody to detect the presence of GOLM1. Total protein was stained with Ponceau S showing equal loading of culture supernatant. Vector-transfected cells were used as controls. Un-T is culture supernatant from untransfected cells. (B) Immunoblot using anti–GOLM1 antibody that was preincubated with recombinant GOLM1 protein. For immunoblot, culture supernatants from LNCaP and DU145 cell lines were used. (C) Culture supernatants from Du145 and LNCaP prostate cancer cell lines were collected and analyzed by immunoblot. Addition of secretory blocker brefeldin A reduced the GOLM1 protein in the culture supernatant. Representative immunoblots from multiple experiments are shown here. (D) Immunostaining of GOLM1. Panels i and ii show untreated LNCaP cells stained with GOLM1 (green) and E-cadherin (red) antibodies. Nuclei were stained with DAPI (blue). Panel iii shows colocalization of GOLM1 with a Golgi marker protein, Golgin 97 (red). When treated with brefeldin A, GOLM1 shows diffused staining as shown in panels iv and v. Panel vi shows staining of protein disulfide isomerase (red), an endoplasmic reticulum enzyme distribution in brefeldin A–treated cells. 1292 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. Neoplasia Vol. 10, No. 11, 2008 recombinant GOLM1 protein, similar to the results obtained for in the range of 2 to 4 in contrast to controls (28%; Figure 5B). An culture supernatants (Figure 4B; data not shown). The mean score ROC curve was generated for GOLM1 reactivity, and an optimum for urine-associated GOLM1 reactivity in prostate cancer patients cutoff point was selected at the region where the slope of the curve (mean = 2.77) was significantly greater (P < .0001) than in the had the highest value (Figure 5C). At this cutoff point, GOLM1 control subjects (mean = 0.96). Significantly greater percentage of had the best discriminatory power in distinguishing between urine prostate cancer urine samples (75%) had GOLM1 reactivity score from prostate cancer patients and control populations (AUC = Neoplasia Vol. 10, No. 11, 2008 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. 1293

0.785, 95% confidence interval = 0.693-0.876, P < .0001) repre- Larger-scale prospective studies will be necessary to further explore this senting a sensitivity and specificity of 75% and 72%, respectively possibility. Recently, our group has shown that increased GOLM1 (Figure 5C). Overall, 39 of 52 urine samples from patients with transcript in urine, along with SPINK1 and PCA3 transcript expres- clinically localized prostate cancer and 14 of 50 control samples sion, and TMPRSS2:ERG fusion status were able to predict prostate were considered positive for GOLM1 reactivity. Immunoblot anal- cancer outcome. Multivariate regression analysis showed that a multi- ysis of prostate cancer patient sera did not indicate the presence of plex of these biomarkers outperformed serum PSA or PCA3 alone in GOLM1 protein (data not shown). Furthermore, GOLMI mRNA detecting prostate cancer [19]. transcripts showed significant association for discriminating patients Although its function remains uncharacterized, the differential ex- with prostate cancer from the negative needle biopsy cohort, as re- pression of GOLM1 seems to be a common feature of malignancy ported in our earlier study [19] (Figure 5D). We assayed a total of and has great potential to be developed as a cancer biomarker. For 333 samples, a much larger sample size than previously reported example, Lu et al. [34] identified GOLM1 as one of 187 genes that [19] and tested for the ability to detect prostate cancer based on were consistently dysregulated in 20 common types of cancers, in- the ROC curves, GOLM1 (AUC = 0.622, P = .0009) outperformed cluding bladder, breast, colon, endometrium, kidney, liver, lung, mel- serum PSA (AUC = 0.495, P = .902; Figure 5E), suggesting the use anoma, lymphoma, pancreatic, prostate, and thyroid cancers. In of urine-based GOLM1 mRNA measurements for the noninvasive cultured cells, adenovirus infection induces GOLM1 expression detection of prostate cancer. The sensitivity, specificity, and positive through the CtBP domain of the adenoviral E1A protein [35]. Inter- and negative predictive values for the detection of GOLM1 in urine estingly, recent cell culture studies have demonstrated estrogen- were 0.594, 0.709, 0.732, and 0.490, respectively. induced changes in GOLM1 expression in the mouse uterus [36] and in androgen-resistant prostate cancer cell lines [37], suggesting a possible hormone regulation in prostate neoplasia. A recent study Discussion indicated that there is cancer cell–specific alterations of GOLM1 and Gene expression analysis facilitated the identification of gene sig- MYO6 in prostate cancer [38]. natures that are dysregulated in a given cancer. Multiple microarray In the absence of any obvious protein domains that would indicate studies identified GOLM1 (GP73 or GOLPH2), a resident Golgi the normal biologic function of GOLM1, gene knock-out studies protein, to be transcriptionaly up-regulated in prostate cancer. Here, may be necessary to determine its role in cellular processes. Prelim- we demonstrate an up-regulation of GOLM1 in prostate cancer inary in vivo studies in mouse model with a C-terminal truncation epithelial cells. Further, a secretory form of this protein was iden- suggest a decreased survival and severe renal abnormalities in ho- tified in culture supernatants of prostate cell lines. On the basis mozygotes, but no obvious defects in prostate morphology [39]. of the detection of the N-terminal FLAG tag in luminal GOLM1, Cell-specific knockout studies of GOLM1 are currently in progress it seems that the release of GOLM1 into the lumen does not in- (Fimmel, in preparation) and may shed light on the function of the volve N-terminal cleavage at the proprotein convertase cleavage site protein and its role in prostate and other cancers. (R52VRR55) [16]. This suggests that the release mechanism used The membrane localization/secretory form of GOLM1 makes it by prostate cancer cells may differ from hepatocytes. The secretory an ideal target for immunotherapy approaches. Further, a better un- mechanism of this apparent “full-length GOLM1” is intriguing and derstanding of its function and processing may provide new insights remains to be determined. The urinary GOLM1 protein may also into the transport, sorting, and processing of resident Golgi and se- be derived from sloughed prostate cancer cells, although such cells cretory proteins. Of particular interest will be the possible role, if any, are typically only encountered in advanced stages of the disease and played by dysregulated GOLM1 expression in PSA secretion through after prostate massage [31]. This, however, does not explain the release the ER-Golgi-membrane transport pathway. of intact protein from cultured cells. GOLM1 release from prostate cell lines could be inhibited by brefeldin A, a protein transport inhibitor [20,21], which suggests that this process could involve secretion. Al- Acknowledgments ternatively, GOLM1 could be released within exosomes [32] or per- The authors thank Venkatesh Basrur and for helpful discussions. haps microvesicles [33]. Regardless of the mechanism, the value of The authors thank the staff of the Microscopy and Image Analyses urine GOLM1 measurements as a supplement to serum PSA screening laboratory at the University of Michigan for their assistance in the and digital rectal examination is promising and needs to be evaluated. microscopic analyses used in this study.

Figure 5. Evaluation of GOLM1 protein and transcripts in urine. (A) Detection of GOLM1 in urine of prostate cancer patients. Urine from healthy normal controls and confirmed prostate cancer patients were analyzed by SDS-PAGE and immunoblot analysis using anti– GOLM1 antibody to test for the presence of GOLM1 protein in the urine. Twenty representative samples are shown out of a total of 102 that were analyzed. LNCaP, prostate cancer cell line extract, was used as a positive control for GOLM1 immunoreactivity. (B) Plot based on the mean score for urine-associated GOLM1 reactivity. (C) Receiver-operating characteristic curve for the detection of GOLM1 in urine as assessed by immunoblot analysis (n = 52 sera from clinically localized prostate cancer patients and n = 50 control subjects). Area under the curve for GOLM1 is 0.785. (D) Quantitative PCR for the detection of GOLM1 transcripts in urine. Quantitative PCR was performed on WTA cDNA from urine obtained from patients presenting for needle biopsy or prostatectomy. GOLM1 expression in pa- tients with negative needle biopsies (green) or patients with prostate cancer (positive needle biopsy or prostatectomy; red) are shown. Normalization was performed using −ΔCt, with PCA3 normalized to the average of urine sediment PSA and GAPDH expression. (E) Receiver-operating characteristic curves for GOLM1 for the diagnosis of prostate cancer. Areas under the curve for GOLM1 and serum PSA are 0.622, and 0.495, respectively. 1294 GOLM1 Is a Tissue Biomarker of Prostate Cancer Varambally et al. Neoplasia Vol. 10, No. 11, 2008

References grade transport from the Golgi apparatus to the endoplasmic reticulum. J Biol [1] Abate-Shen C and Shen MM (2000). Molecular genetics of prostate cancer. Chem 272, 20828–20834. Genes Dev 14, 2410–2434. [22] Kleer CG, Cao Q, Varambally S, Shen R, Ota I, Tomlins SA, Ghosh D, Sewalt [2] Gronberg H (2003). Prostate cancer epidemiology. Lancet 361, 859–864. RG, Otte AP, Hayes DF, et al. (2003). EZH2 is a marker of aggressive breast [3] Dhanasekaran SM, Barrette TR, Ghosh D, Shah R, Varambally S, Kurachi K, cancer and promotes neoplastic transformation of breast epithelial cells. Proc Pienta KJ, Rubin MA, and Chinnaiyan AM (2001). Delineation of prognostic Natl Acad Sci USA 100, 11606–11611. biomarkers in prostate cancer. Nature 412, 822–826. [23] Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Paepe A, and [4] Emmert-Buck MR, Gillespie JW, Paweletz CP, Ornstein DK, Basrur V, Appella Speleman F (2002). Accurate normalization of real-time quantitative RT-PCR E, Wang QH, Huang J, Hu N, Taylor P, et al. (2000). An approach to proteo- data by geometric averaging of multiple internal control genes. Genome Biol 3, mic analysis of human tumors. Mol Carcinog 27, 158–165. RESEARCH 0034. [5] Rubin MA, Zhou M, Dhanasekaran SM, Varambally S, Barrette TR, Sanda MG, [24] Laxman B, Tomlins SA, Mehra R, Morris DS, Wang L, Helgeson BE, Shah RB, Pienta KJ, Ghosh D, and Chinnaiyan AM (2002). alpha-Methylacyl coenzyme A Rubin MA, Wei JT, and Chinnaiyan AM (2006). Noninvasive detection of racemase as a tissue biomarker for prostate cancer. JAMA 287, 1662–1670. TMPRSS2: ERG fusion transcripts in the urine of men with prostate cancer. [6] Rubin MA, Varambally S, Beroukhim R, Tomlins SA, Rhodes DR, Paris PL, Neoplasia 8, 885–888. Hofer MD, Storz-Schweizer M, Kuefer R, Fletcher JA, et al. (2004). Overex- [25] Specht K, Richter T, Muller U, Walch A, Werner M, and Hofler H (2001). pression, amplification, and androgen regulation of TPD52 in prostate cancer. Quantitative gene expression analysis in microdissected archival formalin-fixed Cancer Res 64, 3814–3822. and paraffin-embedded tumor tissue. Am J Pathol 158, 419–429. [7] Tomlins SA, Rhodes DR, Perner S, Dhanasekaran SM, Mehra R, Sun XW, [26] Tomlins SA, Mehra R, Rhodes DR, Cao X, Wang L, Dhanasekaran SM, Kalyana- Varambally S, Cao X, Tchinda J, Kuefer R, et al. (2005). Recurrent fusion of Sundaram S, Wei JT, Rubin MA, Pienta KJ, et al. (2007). Integrative molecular TMPRSS2 and ETS transcription factor genes in prostate cancer. Science 310, concept modeling of prostate cancer progression. Nat Genet 39,41–51. 644–648. [27] Manley S, Mucci NR, De Marzo AM, and Rubin MA (2001). Relational data- [8] Mertz KD, Setlur SR, Dhanasekaran SM, Demichelis F, Perner S, Tomlins S, base structure to manage high-density tissue microarray data and images for pa- Tchinda J, Laxman B, Vessella RL, Beroukhim R, et al. (2007). Molecular char- thology studies focusing on clinical outcome: the prostate specialized program of acterization of TMPRSS2-ERG gene fusion in the NCI-H660 prostate cancer research excellence model. Am J Pathol 159, 837–843. cell line: a new perspective for an old model. Neoplasia 9, 200–206. [28] Mehra R, Varambally S, Ding L, Shen R, Sabel MS, Ghosh D, Chinnaiyan AM, [9] Varambally S, Dhanasekaran SM, Zhou M, Barrette TR, Kumar-Sinha C, Sanda and Kleer CG (2005). Identification of GATA3 as a prognostic MG, Ghosh D, Pienta KJ, Sewalt RG, Otte AP,et al. (2002). The polycomb group marker by global gene expression meta-analysis. Cancer Res 65, 11259–11264. protein EZH2 is involved in progression of prostate cancer. Nature 419, 624–629. [29] Rhodes DR, Yu J, Shanker K, Deshpande N, Varambally R, Ghosh D, Barrette [10] Tomlins SA, Laxman B, Varambally S, Cao X, Yu J, Helgeson BE, Cao Q, T, Pandey A, and Chinnaiyan AM (2004). ONCOMINE: a cancer microarray Prensner JR, Rubin MA, Shah RB, et al. (2008). Role of the TMPRSS2-ERG database and integrated data-mining platform. Neoplasia 6,1–6. gene fusion in prostate cancer. Neoplasia 10, 177–188. [30]ChandranUR,MaC,DhirR,BiscegliaM,Lyons-WeilerM,LiangW, [11] Menschikowski M, Hagelgans A, Gussakovsky E, Kostka H, Paley EL, and Michalopoulos G, Becich M, and Monzon FA (2007). Gene expression profiles Siegert G (2008). Differential expression of secretory phospholipases A2 in nor- of prostate cancer reveal involvement of multiple molecular pathways in the mal and malignant prostate cell lines: regulation by cytokines, cell signaling metastatic process. BMC Cancer 7, 64. pathways, and epigenetic mechanisms. Neoplasia 10, 279–286. [31] Hessels D, Klein Gunnewiek JM, van Oort I, Karthaus HF, van Leenders GJ, [12] Dong Z, Saliganan AD, Meng H, Nabha SM, Sabbota AL, Sheng S, Bonfil RD, van Balken B, Kiemeney LA, Witjes JA, and Schalken JA (2003). DD3(PCA3)- and Cher ML (2008). Prostate cancer cell–derived urokinase-type plasminogen based molecular urine analysis for the diagnosis of prostate cancer. Eur Urol 44, activator contributes to intraosseous tumor growth and bone turnover. Neoplasia 8–15; discussion 15–16. 10, 439–449. [32] Johnstone RM (2006). Exosomes biological significance: a concise review. Blood [13] Rogers CG, Yan G, Zha S, Gonzalgo ML, Isaacs WB, Luo J, De Marzo AM, Cells Mol Dis 36, 315–321. Nelson WG, and Pavlovich CP (2004). Prostate cancer detection on urinalysis [33] MacKenzie A, Wilson HL, Kiss-Toth E, Dower SK, North RA, and Surprenant for alpha methylacyl coenzyme a racemase protein. J Urol 172, 1501–1503. A (2001). Rapid secretion of interleukin-1beta by microvesicle shedding. Immu- [14] Sreekumar A, Laxman B, Rhodes DR, Bhagavathula S, Harwood J, Giacherio nity 15, 825–835. D, Ghosh D, Sanda MG, Rubin MA, and Chinnaiyan AM (2004). Humoral [34] Lu Y, Yi Y, Liu P, Wen W, James M, Wang D, and You M (2007). Common immune response to alpha-methylacyl-CoA racemase and prostate cancer. J Natl human cancer genes discovered by integrated gene-expression analysis. PLoS Cancer Inst 96, 834–843. ONE 2, e1149. [15] Kladney RD, Bulla GA, Guo L, Mason AL, Tollefson AE, Simon DJ, Koutoubi [35] Kladney RD, Tollefson AE, Wold WS, and Fimmel CJ (2002). Upregulation of Z, and Fimmel CJ (2000). GP73, a novel Golgi-localized protein upregulated the Golgi protein GP73 by adenovirus infection requires the E1A CtBP inter- by viral infection. Gene 249,53–65. action domain. Virology 301, 236–246. [16] Bachert C, Fimmel C, and Linstedt AD (2007). Endosomal trafficking and pro- [36] Moggs JG, Tinwell H, Spurway T, Chang HS, Pate I, Lim FL, Moore DJ, protein convertase cleavage of cis Golgi protein GP73 produces marker for he- Soames A, Stuckey R, Currie R, et al. (2004). Phenotypic anchoring of gene patocellular . Traffic (Copenhagen, Denmark) 8, 1415–1423. expression changes during estrogen-induced uterine growth. Environ Health Per- [17] Marrero JA, Romano PR, Nikolaeva O, Steel L, Mehta A, Fimmel CJ, Comunale spect 112, 1589–1606. MA, D’Amelio A, Lok AS, and Block TM (2005). GP73, a resident Golgi glycopro- [37] Coleman IM, Kiefer JA, Brown LG, Pitts TE, Nelson PS, Brubaker KD, tein, is a novel serum marker for . JHepatol43,1007–1012. Vessella RL, and Corey E (2006). Inhibition of androgen-independent prostate [18] Lapointe J, Li C, Higgins JP, van de Rijn M, Bair E, Montgomery K, Ferrari M, cancer by estrogenic compounds is associated with increased expression of Egevad L, Rayford W, Bergerheim U, et al. (2004). Gene expression profiling immune-related genes. Neoplasia 8, 862–878. identifies clinically relevant subtypes of prostate cancer. Proc Natl Acad Sci USA [38] Wei S, Dunn TA, Isaacs WB, De Marzo AM, and Luo J (2008). GOLPH2 and 101, 811–816. MYO6: putative prostate cancer markers localized to the Golgi apparatus. Pros- [19] Laxman B, Morris DS, Yu J, Siddiqui J, Cao J, Mehra R, Lonigro RJ, Tsodikov A, tate 68, 1387–1395. Wei JT, Tomlins SA, et al. (2008). A first-generation multiplex biomarker analysis [39] Wright LM, Yong S, Picken MM, Rockey D, and Fimmel CJ (2009). Decreased of urine for the early detection of prostate cancer. Cancer Res 68, 645–649. survival and hepato-renal pathology in mice with C-terminally truncated GP73 [20] Strous GJ, van Kerkhof P, van Meer G, Rijnboutt S, and Stoorvogel W (1993). (GOLPH2). Int J Clin Exp Pathol 2,34–47. Differential effects of brefeldin A on transport of secretory and lysosomal pro- [40] Varambally S, Yu J, Laxman B, Rhodes DR, Mehra R, Tomlins SA, Shah RB, teins. J Biol Chem 268, 2341–2347. Chandran U, Monzon FA, Becich MJ, et al. (2005). Integrative genomic and [21] Ivessa NE, Gravotta D, De Lemos-Chiarandini C, and Kreibich G (1997). proteomic analysis of prostate cancer reveals signatures of metastatic progression. Functional protein prenylation is required for the brefeldin A–dependent retro- Cancer Cell 8, 393–406. Table W1. Clinical and Pathology Characteristics for the Patients Presenting for Prostate Biopsy or Table W2. Clinical and Pathologic Characteristics for Prostate Cancer Patients Used for Urine Radical Prostatectomy for the Urines Used for Detection of GOLM1 Transcripts. Analysis of GOLM1 Protein.

Age, mean ± SE (n = 332) 62.7 ± 8.7 Characteristic Value Gland weight, mean ± SE (n = 70) 33.6 ± 19.3 Tumor size, mean ± SE (n = 69) 1.2 ± 0.7 Age,* mean ± SEM (years) 58.1 ± 0.7 PSA, mean ± SE (n = 271) 8.6 ± 10.2 Gland weight,* mean ± SEM (g) 49.1 ± 2.4 0-2.5 ng/ml (%) 8.5% Max dimension of largest tumor,* mean ± SEM (cm) 1.4 ± 0.1 † 2.6-10 ng/ml (%) 71.9% PSA, mean ± SEM (ng/ml) 8.51 ± 1.7 4-10 ng/ml (%) 53.8% 0-2.5 ng/ml (%) 11.1 >10 ng/ml (%) 19.6% 2.6-10 ng/ml (%) 66.7 Gleason Sum (n = 169) 4-10 ng/ml (%) 55.6 ≤6 (%) 43.8% >10 ng/ml 22.2 ‡ >7 (%) 56.2% Gleason grade Stage (n = 151) ≤6 (%) 39.6 T1 79.4% ≥7 (%) 60.4 § T2 19.9% Pathologic tumor staging T3 0.7% T2 (%) 82.0 T3 (%) 14.0 T4 (%) 4

*Values available for 50 samples only. † Values available for 45 samples only. ‡ Values available for 48 samples only. §Values available for 50 samples only.