Serum Proteomics on the Basis of Discovery of Predictive Biomarkers of Response to Androgen Deprivation Therapy in Advanced Prostate Cancer

Original Study

Serum Proteomics on the Basis of Discovery of Predictive Biomarkers of Response to Androgen Deprivation Therapy in Advanced Prostate Cancer Manish Kohli,1 Ann L. Oberg,2 Douglas W. Mahoney,2 Shaun M. Riska,2 Robert Sherwood,3 Yuzi Zhang,2 Roman M. Zenka,4 Deepak Sahasrabudhe,5 Rui Qin,6 Sheng Zhang3 Abstract A global serum proteomic analysis was performed using liquid chromatography coupled with tandem mass spectrometry to identify predictive biomarkers in hormone-sensitive prostate cancer patients undergoing androgen deprivation therapy. We identified 47 candidates and then performed network enrichment analysis, which implicated b estradiol, nuclear factor kappa-light-chain enhancer of activated B cells complex, and P38 mitogen-activated protein kinases complex pathways as candidate pathways with potential predictive biomarkers of androgen deprivation. Background: We investigated the serum proteome of hormone-sensitive prostate cancer patients to determine candidate biomarkers associated with androgen deprivation therapy (ADT) efficacy. Patients and Methods: Serum proteomes generated using isobaric mass tags for relative and absolute quantitation were analyzed using reversephase liquid chromatography coupled to tandem mass spectrometry. The advanced hormone-sensitive prostate cancer cohorts studied were: (1) untreated “paired” pre-ADT and 4-month post-ADT hormone-sensitive patients (n ¼ 15); (2) “early ADT failure” patients (n ¼ 10) in whom ADT treatment failed within a short period of time; and (3) “late ADT failure” patients (n ¼ 10) in whom ADT treatment failed after a prolonged response time. Differential abundance was assessed, and ingenuity pathway analysis (IPA) was used to identify interaction networks in selected candidates from these comparisons. Results: Between “post-ADT” and combined “early” and “late” ADT failure groups 149 differentially detected candidates were observed, and between “early” and “late” ADT failure groups 98 candidates were observed; 47 candidates were common in both comparisons. IPA network enrichment analysis of the 47 candidates identified 3 interaction networks (P < .01) including 17-b-estradiol, nuclear factor kappa-light-chain enhancer of activated B cells complex, and P38 mitogen-activated protein kinases as pathways with potential markers of response to ADT. Conclusion: A global proteomic analysis identified pathways with markers of ADT response, which will need validation in independent data sets. Clinical Genitourinary Cancer, Vol. 17, No. 4, 248-53 ª 2019 Elsevier Inc. All rights reserved. Keywords: Hormonal therapy, Predictive factors, Prostate cancer, Proteomics, Serum

Introduction Prostate cancer is a leading cause of cancer-related morbidity and mortality in US men.1 Androgen deprivation therapy (ADT) is a standard systemic hormonal therapy used for treating advanced 1

Department of Medical Oncology, Mayo Clinic, Rochester, MN Department of Health Sciences Research, Mayo Clinic, Rochester, MN Institute for Biotechnology and Life Science Technologies, Cornell University, Ithaca, NY 4 Proteomcis Core, Mayo Clinic, Rochester, MN 5 Department of Medicine, University of Rochester, NY 6 Clinical Biostatistics, Janssen Pharmaceuticals, Raritan, NJ 2 3

248

-

Clinical Genitourinary Cancer August 2019

prostate cancer and also for nonmetastatic stage patients with poor prognosis in the presence of high-risk features for early clinical progression. ADT slows disease progression but the duration of ADT response can range from a few months to several years2-6

Submitted: Dec 28, 2018; Revised: Mar 11, 2019; Accepted: Mar 11, 2019; Epub: Mar 15, 2019 Addresses for correspondence: Manish Kohli, MD, Department of Genitourinary Oncology, H. Lee Moffitt Cancer Center, 12902 Magnolia Dr, Tampa, FL 33612; or Sheng Zhang, PhD, Institute for Biotechnology and Life Science Technologies, Cornell University, 526 Campus Rd, Ithaca, NY 14853 E-mail contact: Manish.Kohli@moffitt.org; [email protected]

1558-7673/$ - see frontmatter ª 2019 Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.clgc.2019.03.006

before a state of castration resistance emerges. Currently there are no validated serum-based predictive markers available for application in this clinical setting.7 Therapeutic advances in the treatment of hormone-sensitive prostate cancer have led to upfront combinations with ADT for increasing the overall efficacy of primary ADT.8,9 These combinations work best for patients with clinical poor-risk features or with high-volume disease; biomarkers are not yet used to decide whether these patients receive ADT alone or these combinations. The most well known biomarker in prostate cancer, serum prostatic-specific antigen (PSA), lacks evidence as a predictive factor for defining ADT response duration.10-12 PSA does not adequately reflect underlying heterogeneity in tumor biology. Beyond PSA, other clinical and tumor factors (such as Gleason score), androgen levels, germline markers, which have been evaluated as predictors of hormonal therapy response are yet to be validated.13-17 We hypothesized that serum-based candidates of ADT response could be identified by comparing global serum proteomic signatures in advanced prostate cancer patients initiating ADT who experience treatment failure after variable durations of response. We profiled serum samples taken from multiple advanced-stage hormone-sensitive prostate cancer patient cohorts to test this hypothesis.

Patients and Methods Study Population Whole blood was collected from patients visiting a tertiary level hospital after obtaining written consent for participation in an advanced prostate cancer biomarker registry at the University of Rochester, New York. Institutional review board (IRB)-approved written consents were used and IRB approval was received for research on specimens used. Specimens from 3 distinct patient cohorts of nonlocalized stage prostate cancer were identified for serum proteomic profiling. The first cohort consisted of 15 patients with untreated advanced hormone-sensitive prostate cancer with “paired” “pre-ADT” initiation and a 4-month “post-ADT” serum specimen (a total of 30 serum specimens) between 2007 and 2008. The second cohort included serum collected from 10 patients with disease progression during ADT for nonlocalized hormone-sensitive prostate cancer within a short (<2 years) period of time (“early ADT failure”). The third cohort consisted of 10 additional patients with disease progression during ADT after a long (>3 years) response duration (“late ADT failure”). The definition of ADT failure for metastatic prostate cancer and progression to castrate resistance stage was defined as development of new metastatic lesions on imaging or biochemical progression (2 serially rising PSAs at least 1 week apart) during continuous androgen deprivation. Serum was collected using a uniform processing protocol for all specimens as described in Supplemental Appendix A in the online version.

Serum Proteome Generation and Mass Spectrometry Methods Removal of high abundance proteins in serum was performed using a Hu-14 Multiple Affinity Removal Column, 4.6  100 mm (product number 5188-6558; Agilent Technologies, Newcastle, DE), attached to an Agilent 1100 binary gradient high performance liquid chromatography system (Agilent

Technologies, Inc, Santa Clara, CA). Subsequently, a total of 50

mg protein of each sample was used for isobaric mass tags for

relative and absolute quantitation (iTRAQ) labeling. This was followed by high pH reverse phase fractionation. After fractionation of serum samples we performed tandem mass spectrometry with nanoscale reverse phase chromatography and tandem mass spectrometry analysis using an LTQ-Orbitrap Velos (ThermoFisher Scientific, San Jose, CA) mass spectrometer equipped with a “Plug and Play” nano ion source (CorSolutions, LLC, Ithaca, NY) using high energy collision dissociation (HCD). Data processing after mass spectrometry and identification of proteins was performed using Proteome Discoverer 1.1 software. Detailed methods are included in Supplemental Appendix A in the online version.

Statistical Methods Samples were randomly allocated to a mass spectrometry experiment and iTRAQ label via randomized block methods.18 Two samples were replicated for quality control purposes. Thus, a total of 52 samples from 35 subjects were assayed in 13 mass spectrometry experiments with the 4-plex iTRAQ system (see Supplemental Figure 1 in the online version). In addition to the laboratory quality control assessments, global quality and bias were assessed graphically.19,20 The natural log peptide abundance values were normalized using a linear model to remove global experimental effects.21 The residual values from the model fit were used as normalized values in further analyses. Per-protein linear mixed effects models with contrast statements were used to assess differential abundance. Three comparisons were conducted: pre-ADT versus post-ADT (15 paired samples), early versus late failure (10 samples each), and post-ADT versus any failure (early plus late failure cohorts). A random effect for subject was used to account for correlation between paired samples for the pre-ADT versus post-ADT analysis. Only the first replicates from the 2 duplicated samples were included in differential abundance analyses. Statistical significance was assessed using a false discovery rate (FDR) of 0.20 for each comparison.22 Volcano plots for each comparison were created, with fold changes being reported on the log2 scale for ease of interpretation. Differentially abundant proteins in each of these comparisons are presented in Venn diagrams illustrating the overlap of candidates in these comparisons (at an FDR of 0.20). Experimentally observed candidates with differential abundance in these comparisons were then grouped using ingenuity pathway analysis (IPA) as described in Supplemental Appendix A in the online version.

Results Study Population Description Three of 15 patients were diagnosed initially in metastatic stage and proceeded to receive ADT as first treatment without undergoing any primary treatment to the prostate. None of the 3 initially metastatic patients had high-volume metastatic disease as in 3 metastatic sites. All 15 patients received ADT alone as initial treatment for hormone-sensitive disease. The early and late ADT failure cohorts (n ¼ 10 each) included subjects selected after failure of ADT. For the castration-resistant states all patients met the

Clinical Genitourinary Cancer August 2019

- 249

Proteomics in Hormonal Therapy of Prostate Cancer Table 1 Patient Demographic Characteristics Including Age, Initial Treatments at Diagnosis, Time Between Initial Diagnosis and Start of ADT for Metastatic Hormone-Sensitive Disease, and Wherever Appropriate PSA and Time of ADT Early (n [ 10)

Late (n [ 10)

Pair (n [ 15)

Total (n [ 35)

Age n Mean (SD)

10

10

15

35

73.6 (9.2)

76.6 (8.0)

65.7 (9.2)

71.1 (9.9)

Median

71.5

75.5

65.0

70.0

Q1, Q3

68.0, 81.0

71.0, 83.0

61.0, 71.0

65.0, 80.0

Range

(59.0-87.0)

(67.0-90.0)

(49.0-87.0)

(49.0-90.0)

Gleason Score at Initial Diagnosis n Mean (SD)

10

7

14

31

8.3 (1.2)

6.7 (2.1)

7.1 (0.9)

7.4 (1.4)

Median

8.0

7.0

7.0

7.0

Q1, Q3

7.0, 9.0

5.0, 9.0

7.0, 8.0

7.0, 8.0

Range

(7.0-10.0)

(3.0-9.0)

(6.0-9.0)

(3.0-10.0)

ADT

6 (60.0%)

0 (0.0%)

1 (6.7%)

7 (20.0%)

ADTC

1 (10.0%)

0 (0.0%)

1 (6.7%)

2 (5.7%)

ADTX

1 (10.0%)

0 (0.0%)

1 (6.7%)

2 (5.7%)

S

0 (0.0%)

3 (30.0%)

8 (53.3%)

11 (31.4%)

SADTX

1 (10.0%)

0 (0.0%)

0 (0.0%)

1 (2.9%)

SX

0 (0.0%)

0 (0.0%)

2 (13.3%)

2 (5.7%)

SXADT

0 (0.0%)

1 (10.0%)

0 (0.0%)

1 (2.9%)

X

1 (10.0%)

6 (60.0%)

2 (13.3%)

9 (25.7%)

Initial Prostate Treatments

Time Between Diagnosis and Start of ADT Treatments, Months n Mean (SD)

10

10

15

35

8.5 (23.8)

54.0 (40.2)

43.2 (60.8)

36.4 (49.4)

Median

1.0

54.0

22.3

21.7

Q1, Q3

0.0, 1.0

24.0, 73.1

6.8, 45.1

1.0, 48.0

Range

(0.0-76.0)

(1.9 to 132.5)

(0.0-200.1)

(1.9 to 200.1)

16.5 (1-23)

52 (39-246)

NA

NA

NA

15

15

27.4 (47.1)

27.4 (47.1)

Median Time of ADT for Hormone-Sensitive Metastatic Disease Before Treatment Failure (Range), Months PSA at the Start of ADT for Paired Samples, ng/mL n Mean (SD) Median

3.2

3.2

Q1, Q3

1.0, 35.0

1.0, 35.0

Range

(0.1-152.0)

(0.1-152.0)

15

15

1.0 (1.7)

1.0 (1.7)

Four-Month Mean PSA After ADT Initiation, ng/mL n

NA

NA

Mean (SD) Median

0.3

0.3

Q1, Q3

0.0, 1.1

0.0, 1.1

Range

(0.0-6.2)

(0.0-6.2)

NA

20

PSA at the Time of ADT Failure, ng/mL n Mean (SD)

250

-

10

10

39.9 (48.8)

16.7 (24.4)

Median

27.4

4.3

28.3 (39.4) 9.6

Q1, Q3

2.9, 53.1

1.0, 26.9

1.4, 42.0

Range

(0.7-152.1)

(0.5-76.6)

(0.5-152.1)

Abbreviations: ADT ¼ androgen deprivation therapy; ADTC ¼ androgen deprivation therapy with chemotherapy; ADTX ¼ androgen deprivation therapy with radiation; PSA ¼ prostate specific antigen; Q1 ¼ 25th percentile; Q3 ¼ 75th percentile; S ¼ surgery/radical prostatectomy; SADTX ¼ surgery with androgen deprivation therapy followed by radiation; SX ¼ surgery followed by radiation; SXADT ¼ surgery followed by radiation followed by androgen deprivation therapy; X ¼ primary radiation therapy to prostate.

Clinical Genitourinary Cancer August 2019

Manish Kohli et al criteria of PSA progression and did not have any new metastases. Patient demographic characteristics including age, initial treatments at diagnosis, time between initial diagnosis and start of ADT for metastatic hormone-sensitive disease, and wherever appropriate PSA and time of ADT are summarized in Table 1.

Description of Proteomic Analyses A total of 586 proteins/peptides were detected after restricting to the nonreplicate experiment set (n ¼ 50), with 153 detected in all 13 mass spectrometry experiments and 208 detected in only 1 mass spectrometry experiment (see Supplemental Figure 2 in the online version). To identify predictors of ADT response we were interested in candidate proteins/peptides that changed significantly during ADT in the comparisons between the 4-month post ADT initiation cohort proteome and any failure, as well as between the proteomes of “early” and “late” ADT failure cohorts. This was on the basis of the assumption that to be considered as circulatory predictive markers of response to ADT, at the very least, candidates would have to show change in abundance with ADT initiation and again at the time of ADT failure. Volcano plots for each comparison are presented in Supplemental Figure 3 in the online version. Venn diagrams illustrating the overlap of proteins/peptides with differential abundance in these comparisons according to a FDR of 0.20 or less are presented in Figure 1. At a FDR level of 0.20, 19 proteins/peptides were found to be differentially abundant in the “pre” Figure 1 Venn Diagram Showing the Number of Proteins Significant With False Discovery Rate £0.2 and the Number in Common Between Each of the 3 Comparisons Performed. “Pre” Refers to Patients in Cohort 1 Before Initiation of ADT. “Post” Refers to the Paired 4-Month Sample After Initiating ADT. “Early” Indicates the “Early-ADT Failure” Cohort and “Late” Indicates the “Late-ADT Failure” Cohort

versus “post” ADT initiation comparison in the first cohort; 98 in the “early-ADT failure” versus “late-ADT” comparison, and 149 in the “post ADT initiation” versus all failure comparisons. There were 47 proteins (regions CþG in Figure 1) that were differentially abundant in the early versus late and post versus failure comparisons (Supplemental Appendix B in the online version provides identity of various proteins on the basis of these comparisons). These candidates were analyzed further in network analyses.

Description of Pathway Analysis In total, 47 candidates were considered to be significantly altered for pathway level enrichment analysis. The genes corresponding to these proteins were overlaid onto the molecular network collected by the IPA database. Only experimentally observed interactions were included in this analysis, assuring the high confidence of network knowledge. A representative gene network identified in the candidate proteins is provided in Figure 2. This network was centered on b-estradiol and the genes within the network correspond to organismal injury and abnormalities, respiratory disease, and cellular compromise (significance score is 29, corresponding to P value less than 1E-29). A network legend, and the corresponding gene lists are provided in Supplemental Appendix A in the online version. Two other networks were also identified as significant hits in the analysis. One was centered on the nuclear factor kappa-lightchain enhancer of activated B cells (NfKB) complex (see Supplemental Figure 4 in the online version), and the other was

Figure 2 Pathway Enrichment Network Results of 47 Proteins After Comparing Post-ADT Proteome With Failure Proteome (From the Venn Diagram in Figure 1 (Areas G D C) Showing b Estradiol (Red Box) as a Key Gene Implicated in the Pathway Network Analysis (P Value Less Than 10E-30 Considered Significant). Green Signifies a Decrease in Protein Abundance in the Comparison. Red Signifies an Increase in Protein Abundance

Clinical Genitourinary Cancer August 2019

- 251

Proteomics in Hormonal Therapy of Prostate Cancer centered on the P38 mitogen-activated protein kinases (P38MAPK) complex (see Supplemental Figure 5 in the online version).

Discussion

252

-

Because no serum-based predictive biomarkers for ADT response are known we performed a serial global analysis of the serum proteome in patients who started ADT for advanced stage disease. To generate serum proteomic profiles we used iTRAQ analyzed using reverse-phase liquid chromatography coupled to tandem mass spectrometry and undertook a strategy in multiple “discovery level” cohorts and then followed by conducting a network analysis of 47 candidate biomarkers. Of the 3 networks implicated on the basis of biological relevance in this hormonally driven tumor we focused on one, a hormonal-based network in an independent cohort of patients. Variation in sex steroids might in fact have some relevance in predicting response to hormonal ablation23 but this observation will need further evaluation in independent cohorts. Post ADT initiation measurements of PSA and testosterone as prognostic and predictive biomarkers have been investigated in advanced prostate cancer24,25 although are not validated yet for clinical practice. A serially rising serum PSA is a sensitive measure of progressive disease, but its pretreatment level as a predictive marker has not been established and is more likely to correlate with tumor burden. A 7month post ADT treatment serum PSA decline has been observed to have a level of association with long-term response, but has not been validated as a predictive biomarker. Eisenberger et al2 reported that 76% of patients with metastatic prostate cancer treated with ADT had a PSA decrease to <4.0 ng/mL. In general, a nadir level is reached 2.5 to 3 months after starting ADT. The same investigators reported that the median time to PSA <4 ng/mL was <82 days, whereas others have reported 99% of patients reaching nadir levels by 3 to 4 months.2 Normalization of a rising PSA and PSA kinetics (PSA doubling time; PSA velocity) have been evaluated in clinical studies and appear to have some prognostic significance, but there is no consensus on either absolute PSA measurements or PSA kinetics to demonstrate predictive value to ADT. By using emerging proteomic-based technologies it has become possible to perform this type of large-scale study of protein abundance, in which thousands of protein candidates can be identified in a single experiment as candidates for prospective evaluation. To our knowledge, our study is the first of its kind to perform a global proteome analysis using an iTRAQ approach and we were able to identify 47 potential candidates and at least 2 other networks that might serve as biomarkers of response. The need to investigate these in future is relevant because currently no serum-based proteomic biomarkers are available to identify ADT response. Comparative proteomic approaches using 2-dimensional gel electrophoresis (2DE) in conjunction with mass spectrometry (MS) have provided valuable information for human plasma proteome projects.26-28 However, 2-DE has well recognized limitations such as its limited applicability to hydrophobic proteins and those of extreme molecular weight or isoelectric point. Furthermore, 2-DE suffers from a limited dynamic range.29 For these reasons, solution-based “shotgun” techniques on the basis of 2-dimensional (2D) liquid chromatography (LC) are widely used as an alternative approach, in which low pH reverse phase LC is always used as the second chromatographic dimension, coupled directly with tandem MS

Clinical Genitourinary Cancer August 2019

(MS/MS) detection.30,31 Importantly, 2D-LC is highly compatible with a stable isotope label-based quantitative technique using iTRAQ, allowing multiplexed (high throughput) quantitative analysis32 across multiple biological samples. We used this technique for identifying serum-based candidates of ADT response and then used pathway enrichment tools from a manually curated repository containing thousands of biological interactions and functional annotations extracted from published literature to identify clinically relevant candidates for prospective testing.

Conclusion Despite the sensitive and global proteome detection techniques used in this hypothesis-generating study, there are limitations. Our study population was from a single institution and limited to 35 advanced prostate cancer patients. The identified candidates were observed to have potential to be pursued on the basis of a liberal statistical cutoff value of a FDR of 0.2. This might be a result of a limited cohort size and event rate, and will need further follow-up and evaluation in larger cohorts with careful and uniform collection and processing protocols to avoid preanalytic bias. A serum proteome-based signature for ADT response on the basis of other networks identified in our analysis (NfKB complex and P38-MAPK complex) or the individual 47 candidates might also need to be included beyond hormonal pathways. An independent validation set of pre- and post ADT initiation serum estrogen levels will be needed to determine whether estrogen levels can serve as predictive biomarkers of ADT efficacy in advanced prostate cancer.

Clinical Practice Points  The most well-known biomarker in prostate cancer, serum PSA

 







 

lack the evidence as a predictive factor for defining ADT response duration. Serum-based circulatory predictive biomarkers of response to ADT are not known or used in clinical practice. We investigated global serum proteomic signatures using emerging proteomic-based technologies in multiple independent advanced prostate cancer patient cohorts including patients initiating ADT and patients in whom ADT has failed after variable durations of response to determine candidate biomarkers associated with ADT efficacy. The multiple serum proteomes generated from these independent cohorts used iTRAQ analyzed using reverse-phase LC-MS/ MS. We identified 47 candidates after comparisons of the proteomic signatures generated from patients who started ADT; those in whom ADT had failed in a short time period, and those in whom ADT had failed after a prolonged period of efficacy and then performed an IPA-based network enrichment analysis of these candidates. The IPA analysis identified b-estradiol, NfKB complex, and P38MAPK complex pathways as candidate pathways with potential predictive biomarkers of androgen deprivation. Serum-based candidates of ADT response are critically needed in the practice of treating advanced prostate cancer patients. In patients who start ADT, if estradiol pathway measurements (such as b-estradiol and estrone levels) can be validated to predict

Manish Kohli et al duration of treatment efficacy, it will affect management of hormone-sensitive disease with ADT alone vs. ADT based-combination treatments.

Acknowledgments This work was supported by the National Institutes of Health, National Cancer Institute (5R21CA133536; M.K. and S.Z.).

Disclosure The authors have stated that they have no conflicts of interest.

Supplemental Data Supplemental material and figures accompanying this article can be found in the online version at https://doi.org/10.1016/j.clgc. 2019.03.006.

References 1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2018. CA Cancer J Clin 2018; 68:7-30. 2. Eisenberger MA, Blumenstein BA, Crawford ED, et al. Bilateral orchiectomy with or without flutamide for metastatic prostate cancer. N Engl J Med 1998; 339:1036-42. 3. Crawford ED, Eisenberger MA, McLeod DG, et al. A controlled trial of leuprolide with and without flutamide in prostatic carcinoma. N Engl J Med 1989; 321:419-24. 4. Denis LJ, Keuppens F, Smith PH, et al. Maximal androgen blockade: final analysis of EORTC phase III trial 30853. EORTC Genito-Urinary Tract Cancer Cooperative Group and the EORTC Data Center. Eur Urol 1998; 33:144-51. 5. Newling DW, Denis L, Vermeylen K. Orchiectomy vs. goserelin and flutamide in the treatment of newly diagnosed metastatic prostate cancer. Analysis of the criteria of evaluation used in the European Organization for Research and Treatment of Cancer–Genitourinary Group Study 30853. Cancer 1993; 72(12 suppl):3793-8. 6. Prostate Cancer Trialists Collaborative Group. Maximum androgen blockade in advanced prostate cancer: an overview of the randomised trials. Prostate Cancer Trialists’ Collaborative Group. Lancet 2000; 355:1491-8. 7. Grivas PD, Robins DM, Hussain M. Predicting response to hormonal therapy and survival in men with hormone sensitive metastatic prostate cancer. Crit Rev Oncol Hematol 2013; 85:82-93. 8. Fizazi K, Tran N, Fein L, et al. Abiraterone plus prednisone in metastatic, castration-sensitive prostate cancer. N Engl J Med 2017; 377:352-60. 9. Sweeney CJ, Chen YH, Carducci M, et al. Chemohormonal therapy in metastatic hormone-sensitive prostate cancer. N Engl J Med 2015; 373:737-46. 10. Vicini FA, Vargas C, Abner A, Kestin L, Horwitz E, Martinez A. Limitations in the use of serum prostate specific antigen levels to monitor patients after treatment for prostate cancer. J Urol 2005; 173:1456-62. 11. Smith JA Jr, Lange PH, Janknegt RA, Abbou CC, deGery A. Serum markers as a predictor of response duration and patient survival after hormonal therapy for metastatic carcinoma of the prostate. J Urol 1997; 157:1329-34. 12. Small EJ, Roach M 3rd. Prostate-specific antigen in prostate cancer: a case study in the development of a tumor marker to monitor recurrence and assess response. Semin Oncol 2002; 29:264-73.

13. Ross RW, Xie W, Regan MM, et al. Efficacy of androgen deprivation therapy (ADT) in patients with advanced prostate cancer: association between Gleason score, prostate-specific antigen level, and prior ADT exposure with duration of ADT effect. Cancer 2008; 112:1247-53. 14. Quinn DI, Henshall SM, Sutherland RL. Molecular markers of prostate cancer outcome. Eur J Cancer 2005; 41:858-87. 15. Kohli M, Riska SM, Mahoney DW, et al. Germline predictors of androgen deprivation therapy response in advanced prostate cancer. Mayo Clin Proc 2012; 87:240-6. 16. Geller J, Albert JD, Nachtsheim DA, Loza D. Comparison of prostatic cancer tissue dihydrotestosterone levels at the time of relapse following orchiectomy or estrogen therapy. J Urol 1984; 132:693-6. 17. Perachino M, Cavalli V, Bravi F. Testosterone levels in patients with metastatic prostate cancer treated with luteinizing hormone-releasing hormone therapy: prognostic significance? BJU Int 2010; 105:648-51. 18. Oberg AL, Vitek O. Statistical design of quantitative mass spectrometry-based proteomic experiments. J Proteome Res 2009; 8:2144-56. 19. Cunningham JM, Oberg AL, Borralho PM, et al. Evaluation of a new highdimensional miRNA profiling platform. BMC Med Genomics 2009; 2:57. 20. Oberg AL, Mahoney DW. Statistical methods for quantitative mass spectrometry proteomic experiments with labeling. BMC Bioinformatics 2012; 13(suppl 16):S7. 21. Oberg AL, Mahoney DW, Eckel-Passow JE, et al. Statistical analysis of relative labeled mass spectrometry data from complex samples using ANOVA. J Proteome Res 2008; 7:225-33. 22. Storey JD, Tibshirani R. Statistical significance for genomewide studies. Proc Natl Acad Sci U S A 2003; 100:9440-5. 23. Qin X, Zhang H, Ye D, Yao X, Zhang S, Dai B. Variations in circulating sex steroid levels in metastatic prostate cancer patients with combined androgen blockade: observation and implication. Andrology 2013; 1:512-6. 24. Harshman LC, Chen YH, Liu G, et al. Seven-month prostate-specific antigen is prognostic in metastatic hormone-sensitive prostate cancer treated with androgen deprivation with or without docetaxel. J Clin Oncol 2018; 36:376-82. 25. Petrylak DP, Ankerst DP, Jiang CS, et al. Evaluation of prostate-specific antigen declines for surrogacy in patients treated on SWOG 99-16. J Natl Cancer Inst 2006; 98:516-21. 26. Choi KS, Song L, Park YM, et al. Analysis of human plasma proteome by 2DE- and 2D nanoLC-based mass spectrometry. Prep Biochem Biotechnol 2006; 36:3-17. 27. Rai AJ, Gelfand CA, Haywood BC, et al. HUPO Plasma Proteome Project specimen collection and handling: towards the standardization of parameters for plasma proteome samples. Proteomics 2005; 5:3262-77. 28. Echan LA, Tang HY, Ali-Khan N, Lee K, Speicher DW. Depletion of multiple high-abundance proteins improves protein profiling capacities of human serum and plasma. Proteomics 2005; 5:3292-303. 29. Gygi SP, Corthals GL, Zhang Y, Rochon Y, Aebersold R. Evaluation of twodimensional gel electrophoresis-based proteome analysis technology. Proc Natl Acad Sci U S A 2000; 97:9390-5. 30. Peng J, Elias JE, Thoreen CC, Licklider LJ, Gygi SP. Evaluation of multidimensional chromatography coupled with tandem mass spectrometry (LC/LCMS/MS) for large-scale protein analysis: the yeast proteome. J Proteome Res 2003; 2:43-50. 31. Yang Y, Qiang X, Owsiany K, Zhang S, Thannhauser TW, Li L. Evaluation of different multidimensional LC-MS/MS pipelines for isobaric tags for relative and absolute quantitation (iTRAQ)-based proteomic analysis of potato tubers in response to cold storage. J Proteome Res 2011; 10:4647-60. 32. Ross PL, Huang YN, Marchese JN, et al. Multiplexed protein quantitation in Saccharomyces cerevisiae using amine-reactive isobaric tagging reagents. Mol Cell Proteomics 2004; 3:1154-69.

Clinical Genitourinary Cancer August 2019

- 253

Proteomics in Hormonal Therapy of Prostate Cancer Supplemental Data Supplemental Appendix A: Supplemental Methods Serum Sample Preparation Processing and handling of serum specimens followed Human Proteome Organization/Human Plasma Proteome Project standardization methods.1 For serum processing, blood was collected in Becton Dickinson serum separator tubes 6.0 mL vacutainers for serum processing on a single occasion. All specimens underwent centrifugation at 3000 rpm for 10 minutes to generate serum and were stored at 70 C within 30 to 45 minutes of sample collection from patients. After the initial centrifugation a protease inhibitor cocktail was added, the ingredients of which included 10 mL phosphate-buffered saline (Invitrogen number 14190300); 1 tablet complete of mini, ethylenediaminetetracetic acid-free protease inhibitor (Roche number 11 836 170 001); sodium vanadate Na3VO4 and phenylmethylsulphonyl fluoride (Sigma number P7626-5G). No serum specimen underwent any freeze-thaw cycles before performing affinity depletion and preparation for isobaric mass tags for relative and absolute quantitation (iTRAQ) labeling.

Methods for Affinity Depletion of High Abundant Proteins From Human Sera and Protein Digestion/iTRAQ Labeling

appliedbiosystems.com/search.taf; Applied Biosystems, Foster City, CA). A total of 13 sets of iTRAQ 4-plex tagged samples were prepared and designed as shown in Figure 1 covering all 50 individually depleted samples; paired samples 100059.early and 100037.late were duplicated in mass spectrometry (MS) experiment 4 and MS experiment 5. After labeling, the 4 samples were pooled in each of the 13 MS experiments and subjected to cation exchange chromatography using an Applied Biosystems cation exchange cartridge system. The pooled iTRAQ-labeled tryptic peptides (approximately 400 mL) samples were evaporated completely in a SpeedVac concentrator and reconstituted with 1 mL of loading buffer (10 mM potassium phosphate pH 3.0; 25% acetonitrile [ACN]). The pH of the sample was adjusted to 3.0 with formic acid before cartridge separation. The Strong-cation exchange (SCX) cartridge (10 mm id x 14 mm) is PolySULFOETHYL Aspartamide material with particle size at 12 mm and 300 Å pore size. After conditioning of the SCX cartridge with loading buffer, the sample (approximately 200 mg) was loaded and washed with an additional 2 mL of loading buffer. The peptides were eluted in 1 step by 1 mL of loading buffer containing 500 mM KCl. Desalting of SCX fractions was carried out using solid phase extraction on Sep-Pak Cartridges (Waters, Milford, MA) and ready for the first dimensional LC fractionation via a high pH reverse phase chromatography as described in the following section.

High pH Reverse Phase Fractionation

We obtained high performance liquid chromatography (HPLC) buffers from the column manufacturer and the separation followed the manufacturer’s recommended protocol with UV monitoring at 280 nm. Each serum sample (40 mL) was diluted to 200 mL with HPLC Buffer A (Equilibrate/Load/Wash) and filtered through a 0.22-mm cellulose acetate spin filter (Agilent Technologies) before injection. Duplicate 95-mL injections were run for each sample, unbound (low abundance proteins) fractions and bound (high abundance proteins) were collected as separate fractions. The unbound fractions for each sample were pooled, concentrated, and buffer exchanged to 500 mM triethylammonium bicarbonate pH 8.5 using BioMax 5K MWCO spin ultrafiltration cartridges (Millipore Corp, Bedford, MA). The protein concentration for each of the depleted samples was determined using Bradford assay using bovine serum albumin as the calibrant, and further quantified by running sodium dodecyl sulfate (SDS)-mini gel followed by ImageQuant Software version 5.2 (GE Healthcare).

High pH reverse phase fractionation was completed using a Dionex UltiMate3000 HPLC system with built-in micro fraction collection option in its autosampler and UV detection (Sunnyvale, CA). The iTRAQ-tagged tryptic peptides were reconstituted in buffer A (20 mM ammonium formate pH 9.5 in water), and loaded onto a hybrid silica column XTerra MS C18 (2.1 mm  150 mm) from Waters. Buffer B consists of 20 mM ammonium formate pH 9.5 with 90% ACN in water. The high pH reverse phase (RP) chromatography separation was performed at a flow rate of 200 mL per minute using a gradient of 5% B for 3 minutes, 5% to 45% B in 30 minutes, and ramp up to 90% B in 5 minutes. During gradient elution, 48 fractions were collected at 1 minute per fraction. On the basis of the UV trace at 214 nm, several fractions were pooled to yield a total of 19 high pH RP fractions, which were dried and reconstituted in 2% ACN 0.5% formic acid for subsequent nanoscale reverse phase chromatography and tandem MS analysis.

Protein Digestion and iTRAQ Labeling

Nanoscale Reverse Phase Chromatography and Tandem MS

A total of 50 mg protein of each sample was denatured with 1% SDS, reduced with 5 mM tris-(2-carboxylethyl) phosphine at 60 C for 1 hour and the cysteine residues were blocked with 8 mM methyl methanethiosulfonate for 10 minutes at room temperature. Protein samples were digested with 4 mg of sequencing-grade modified trypsin at 37 C for 16 hours. Tryptic peptides from 4 different randomized (see the Statistical Methods section) samples were each labeled with iTRAQ reagents 114, 115, 116, and 117, respectively, according to the manufacturer’s protocols (document 4351918A and 4350831C downloaded from http://docs.

The Orbitrap was interfaced with an UltiMate3000 multiple dimensional liquid chromatography system (Dionex, Sunnyvale, CA). An aliquot of high pH RP peptide fractions (2.0-10.0 mL) was injected onto a PepMap C18 trap column (5 mm, 300 mm  5 mm, Dionex) at 20 mL per minute flow rate for rapid sample loading and then separated on a PepMap C-18 RP nano column (3 mm, 75 mm  15 cm), and eluted using a 60-minute gradient of 5% to 35% ACN in 0.1% formic acid at 300 nL per minute, followed by a 3-minute ramping to 95% ACN-0.1% formic acid and a 5-minute holding at 95% ACN-0.1% FA. The column was re-equilibrated

253.e1

-

Clinical Genitourinary Cancer August 2019

Manish Kohli et al with 2% ACN-0.1% FA for 20 minutes before the next run. The eluted peptides were detected by Orbitrap through a nano ion source containing a 10-mm analyte emitter (NewObjective, Woburn, MA). The Orbitrap Velos was operated in positive ion mode with nano spray voltage set at 1.5 kV and source temperature at 225 C. Either internal calibration using the background ion signal at mass to charge ratio (m/z) 445.120025 as a lock mass or external calibration for fourier transform (FT) mass analyzer was performed. The MS and MS/MS data were acquired at data-dependent acquisition (DDA) mode using an FT mass analyzer for 1 survey MS scan for precursor ions followed by 10 datadependent HCD-MS/MS scans for precursor peptides with multiple charged ions above a threshold ion count of 5000 with normalized collision energy of 45%. MS survey scans at a resolution of 30,000 (full width at half maximum at m/z 400), for the mass range of m/z 400 to 1400, and MS/MS scans at 7500 resolution for the mass range m/z 100 to 2000. All data were acquired using Xcalibur 2.1 operation software (Thermo-Fisher Scientific).

Mass Spectrometry Data Processing, Protein Identification, and Analysis All MS and MS/MS raw spectra from iTRAQ experiments were processed using Proteome Discoverer 1.1 (PD1.1, Thermo) and the spectra from each DDA file were output as an mascot generic format file for subsequent database search using in-house license Mascot (Matrix Science, London, UK; version 2.2 ), Sequest (Thermo Fisher Scientific, San Jose, CA; version 27, rev 12), X!Tandem (The GPM, thegpm.org; version 2009.09.15.3), and Scaffold (version 3.0, Proteome Software, Inc. Portland, OR). The SwissProt Human 2010_05 database containing 20,277 sequence entries downloaded from Uniprot (http://uniprot.org) was used for database query. The default search settings used for 4-plex iTRAQ quantitative processing and protein identification in Scaffold were: 1 miscleavage for full trypsin with fixed methyl methanethiosulfonate modification of cysteine, fixed 4-plex iTRAQ modifications on lysine and N-terminal amines and variable modifications of methionine oxidation and 4-plex iTRAQ on tyrosine. The peptide mass tolerance and fragment mass tolerance values were 10 ppm and 50 mDa, respectively. To estimate the false discovery rate (FDR) for a measure of identification certainty in each replicate set, a decoy database search was performed by augmenting the database with a set of reverse sequences. A filter was applied using the Scaffold software, requiring peptide and protein probability at >95% and at least 2 peptides per protein. The estimated FDR for these thresholds was <1%. Intensities of the reporter ions from iTRAQ tags upon fragmentation were used for quantification.

Statistical Methods for Proteomic Analyses To ensure that sample groups of interest were well balanced over potential experimental effects such as time, iTRAQ labeling tag and MS experiment, randomized block design principles2,3 were used to create 2 randomization schemes to determine: (1) the order of

sample processing; and (2) allocation to the iTRAQ 4-plex experiments. The paired pre-/post ADT samples were forced to be in the same MS experiment to minimize variability for the paired comparison. One matched pair from the early/late group was replicated for quality control purposes. Thus, a total of 52 samples from 35 subjects were assayed in 13 MS experiments with the 4-plex iTRAQ system (see Supplemental Figure 1 in the online version). In addition to the laboratory quality control assessments, global quality and bias were assessed graphically.3,4 Box and whisker plots were used to assess global shifts in the distribution of abundance across samples and experiments. Plots of coefficients of variation (CV) versus the mean were used to assess the mean variance relationship, which was not found to be a function of the mean in these data (see Supplemental Figure 6 in the online version: “CV vs. mean”). Finally, coverage plots were used to assess the number of identified proteins and peptides across iTRAQ experiments. On the basis of these quality control plots, 2 MS experiments were repeated because of incorrect machine settings. Subsequently, all experiments and samples were deemed of good quality and all 52 specimens were retained for further analysis. The natural log peptide abundance values of the 52 samples were normalized using a linear model to remove global experimental effects.5 The residual values from the model fit were used as normalized values and kept for further analysis. Box-and-whisker plots were used to confirm that normalization was successful (see Supplemental Figure 7 in the online version). Per-protein linear mixed effects models with contrast statements were used to assess differential abundance using the normalized data. Three comparisons were conducted: pre-ADT versus post-ADT (15 paired samples), early versus late failure (10 samples each), and post-ADT versus any failure (early plus late failure cohorts). A random effect for subject was used to account for correlation between paired samples for the pre-ADT versus post-ADT analysis. Only the first replicates from the 2 duplicated samples were included in differential abundance analyses. Statistical significance was assessed using an FDR, with a threshold of 0.20 for each comparison.6 Volcano plots for each comparison were created, with fold changes being reported on the log2 scale for ease of interpretation. Differentially abundant proteins/peptides in each of these comparisons are presented in Venn diagrams illustrating the overlap of candidates in these comparisons (Figure 1).

Pathway Enrichment Methods Experimentally observed candidates with differential abundance in these comparisons were then grouped using pathway network analyses to identify key genes and their translational serum products for prospective validation. Using the ingenuity pathway analysis (IPA) tool, genes that correspond to candidate proteins were mapped to relevant pathways and networks on the basis of their functional annotation and known molecular interactions in Ingenuity Knowledge Base (IKB). IKB is a manually curated repository containing thousands of biological interactions and functional annotations extracted from published literature.7 A molecular network of direct or indirect physical, transcriptional, and

Clinical Genitourinary Cancer August 2019

- 253.e2

Proteomics in Hormonal Therapy of Prostate Cancer enzymatic interactions between mammalian orthologs was computed from IKB. The corresponding gene names were uploaded into IPA along with the gene identifiers and corresponding protein fold change values. The genes were then overlaid onto a global molecular network in IKB. In the network analysis, networks of these genes are then algorithmically generated on the basis of their

253.e3

-

Clinical Genitourinary Cancer August 2019

connectivity. Two genes are considered to be connected if there is a path in the network between them. Highly interconnected networks likely represent significant biological function. IPA construct networks that optimize for interconnectivity and number of focus genes (defined on the basis of triangular connectivity) under the constraint of maximal network size (default is 35 genes).

Manish Kohli et al Supplemental Figure 1 Study Design for the Mass Spectrometry Experiments

Supplemental Figure 2 Proteins and Peptides Coverage Plots. The Left Vertical Axis Indicates Mass Spectrometry (MS) Experiment. The Right Vertical Axis Indicates the Number of Peptides or Proteins Detected in That MS Experiment. For Example, 3259 Peptides Were Detected in MS Experiment Number 1. The Bottom Vertical Axis Is the Peptide or Protein Rank According to Prevalence. The Top Vertical Axis Indicates the Number of Peptides or Proteins Detected in 1, 2, etc, up to All MS Experiments. For Example, 2057 Peptides Were Detected in Only 1 MS Experiment, Whereas 1587 Peptides Were Detected in All 13 MS Experiments. A Vertical Gray Line Within the Plot Indicates That Peptide or Protein Was Detected in That MS Experiment. Thus, a Peptide or Protein Detected in All 13 MS Experiments Would Show as a Vertical Gray Line Over the Entire Plot. The Red Line Indicates the Number of Peptides or Proteins Detected for a Given Number of MS Experiments. For Example, the First Step Indicates the 2057 Peptides Detected in Only 1 MS Experiment

Clinical Genitourinary Cancer August 2019

- 253.e4

Proteomics in Hormonal Therapy of Prostate Cancer Supplemental Figure 3 Volcano Plots. The Horizontal Axis Indicates Fold Change on the Log2 Scale. The Vertical Axis Indicates eLog10 (P Value). Horizontal Lines Indicate Various Levels of Statistical Significance. Points in the Top Right and Left Corners of the Plots Are the Most Statistically Significant and Have the Greatest Fold Changes. Lines Indicating 2-Fold Change Are Included for Reference Only

Supplemental Figure 4 Results of Ingenuity Pathway Analysis Networks Identified as Significant Hits in the Analysis Centered on Nuclear Factor Kappa-Light-Chain-Enhancer of Activated B Cells Complex

253.e5

-

Clinical Genitourinary Cancer August 2019

Manish Kohli et al Supplemental Figure 5 Results of Ingenuity Pathway Analysis Networks Identified as Significant Hits in the Analysis Centered on P38Mitogen Activated Protein Kinase Complex

Supplemental Figure 6 Coefficient of Variation (CV) Versus Mean. The Vertical Axis Is the Smoothed (Via Loess) Within MS Experiment CV. This Is Calculated as the SD Between the Abundance Values on the Natural Log Scale for Each Tandem Mass Spectrometry (MS/MS) Observation of a Peptide

Abbreviation: NfKB ¼ nuclear factor kappa-light-chain-enhancer of activated B cells.

Clinical Genitourinary Cancer August 2019

- 253.e6

Proteomics in Hormonal Therapy of Prostate Cancer Supplemental Figure 7 Box and Whisker Plots. Pre- (Left Panel) and Postnormalization (Right Panel) Box and Whisker Plots Showing the Global Peptide Abundance Distribution. The Bottom, Middle Bar, and Top of the Boxes Indicate the 25th, 50th, and 75th Percentiles, Respectively. The Whiskers (Vertical Dashed Lines) Extend to 1.5 Times the Interquartile Range (ie, 1.5 3 [50th Percentile L 25th Percentile]) or the Maximum Value, Whichever Is Smaller. The X-Axis Indicates MS Experiment; Within an Experiment Boxes Are Sorted According to Tag. The Y-Axis Is Peptide Abundance on the Log10 Scale With Labels Indicating Raw Scale. Global Shifts Before Normalization Indicate the Need for Normalization, and the Absence of these Shifts After Normalization Indicates Successful Normalization

253.e7

-

Clinical Genitourinary Cancer August 2019

References 1. Rai AJ, Gelfand CA, Haywood BC, et al. HUPO Plasma Proteome Project specimen collection and handling: towards the standardization of parameters for plasma proteome samples. Proteomics 2005; 5:3262-77. 2. Oberg AL, Vitek O. Statistical design of quantitative mass spectrometry-based proteomic experiments. J Proteome Res 2009; 8(5):2144-56. 3. Oberg AL, Mahoney DW. Statistical methods for quantitative mass spectrometry proteomic experiments with labeling. BMC bioinformatics 2012; 13(Suppl 16):S7. 4. Cunningham JM, Oberg AL, Borralho PM, et al. Evaluation of a new highdimensional miRNA profiling platform. BMC Med Genomics 2009; 2:57. 5. Oberg AL, Mahoney DW, Eckel-Passow JE, et al. Statistical analysis of relative labeled mass spectrometry data from complex samples using ANOVA. J Proteome Res 2008; 7:225-33. 6. Storey JD, Tibshirani R. Statistical significance for genomewide studies. Proc Natl Acad Sci U S A 2003; 100:9440-5. 7. Calvano SE, Xiao W, Richards DR, et al. A network-based analysis of systemic inflammation in humans. Nature 2005; 437:1032-7.