Discovery of Prostate Cancer Biomarkers by Microarray Gene Expression Profiling

Karina Dalsgaard Sørensen; Torben Falck Ørntoft

Expert Rev Mol Diagn. 2010;10(1):49-64.

Microarray Gene Expression Profiling

Microarray expression profiling of tumor samples for biomarker discovery is based on the assumption that gene expression patterns are major determinants of cancer cell behavior. Using microarray technologies, it is possible to map, at a genome-wide scale, the complex molecular aberrations associated with cancer development. Findings can be correlated to clinical data in order to identify molecules, or molecular signatures, associated with a certain tumor stage, grade or clinical outcome.

A microarray is an ordered array of thousands of well-defined single-stranded DNA molecules (probes) immobilized on a glass slide. The following protocol refers to mRNA expression analyses, while profiling of noncoding miRNAs is described in a separate section. Typically, total RNA extracted from a biological sample is reverse transcribed into cDNA, labeled and hybridized to the microarray for determination of transcript abundance (single-color systems) or relative expression compared to a reference sample with a different label (two-color systems). Several microarray platforms using synthetic oligonucleotide probes are commercially available, and generally provide high-quality data, with superior reproducibility compared with earlier generation custom-spotted cDNA arrays.

The feasibility of measuring the expression of tens of thousands of genes in a single experiment has shifted the bottleneck from data generation to data analysis. There are two fundamental strategies for analyzing microarray datasets. The first approach, known as unsupervised classification (or hierarchical clustering), aims to identify subgroups of samples with similar gene expression profiles and is used to discover previously unknown relationships between samples (or genes).^[13,14] An advantage of this method is that additional clinical data are not required. The second approach, termed supervised classification (or nonhierarchical clustering), is utilized for identification of differentially expressed genes between predefined groups of samples; for example, recurrent versus nonrecurrent tumors. While one set of samples (the training set) is used to identify the optimal expression signature that can discriminate between the groups, another set (the validation or test set) is used to evaluate how accurately the signature can classify samples that have not been grouped.^[15] Due to the complexity of microarray datasets, there is an inherent risk of identifying irrelevant genes or clusters. Independent validation is therefore essential for assessment of data reliability. If an independent sample set is not available, internal validation methods based on resampling statistics are often used (e.g., cross-validation, bootstrapping and jack-knifing). A common approach is leave-one-out cross-validation (LOOCV), which repeatedly uses one sample from the original set for validation with the remaining samples as training set, until all samples have been used once for validation.

1 2 3 4 5 6 7 8 9 10 11 12

Next Section

Cite this: Discovery of Prostate Cancer Biomarkers by Microarray Gene Expression Profiling - Medscape - Jan 01, 2010.

Table 1. Microarray gene expression profiling studies of prostate cancer.
Study (year)	Microarray type (probes [n])	Samples included	Main objective	Main findings	Ref.
Dhanasekaran et al. (2001)	cDNA array (10K)	14 PC, 20 CRPC, 13 BPH, 9 NAP	Molecular classification	Molecular signature for PC. PIM1 and hepsin as prognostic markers for PSA recurrence	^[29]
Luo et al. (2001)	cDNA array (6.5K)	16 PC, 9 BPH	Molecular classification	Molecular signature for PC	^[16]
Magee et al. (2001)	Affymetrix Hu6800 (7K)	11 PC, 4 NAP	Molecular classification	Molecular signature for PC	^[17]
Welsh et al. (2001)	Affymetrix U95a (8.9K)	23 PC, 1 LNM, 9 NAP	Molecular classification	Molecular signature for PC	^[26]
LaTulippe et al. (2002)	Affymetrix U95 (63K)	23 PC, 9 MPC, 3 nonmalignant	Molecular classification	Molecular signature for PC	^[24]
Singh et al. (2002)	Affymetrix U95Av2 (12.6K)	52 PC, 50 NAP	Molecular classification and prognostic markers	5-gene signature for PSA recurrence. 29-gene signature for Gleason score prediction. Gene signatures for prediction of normal vs cancer	^[18]
Best et al. (2003)	cDNA array (6.4K)	13 PC, 13 NAP (pooled)	Molecular classification	Molecular classification of moderate- vs high-grade PC (21-gene signature)	^[19]
Henshall et al. (2003)	Affymetrix Eos Hu03 (60K)	72 PC	Prognostic markers	Cox proportional hazards survival analysis applied to microarray expression data identified Trp-p8 as marker for PSA recurrence	^[106]
^Ashida et al.* (2004)	cDNA array (23K)	20 PC, 10 HG-PIN	Molecular classification	Molecular signature for HG-PIN to PC transition	^[23]
Glinsky et al. (2004)	Affymetrix U95Av2 (12.6K)	79 PC	Prognostic markers	Three 4-/5-gene signatures for prediction of PSA recurrence, which combined into a single predictor algorithm improved the accuracy of a clinical nomogram for recurrence prediction	^[10]
Lapointe et al. (2004)	cDNA array (26.3K)	62 PC, 9 LNM, 41 NAP	Molecular classification	Clinically relevant subgroups of PC by unsupervised clustering. MUC1 and AZPG1 as prognostic markers for PSA recurrence	^[30]
Stuart et al. (2004)	Affymetrix U95Av2 (12.6K)	88 prostate tissue samples scored for relative content of PC, BPH and stromal cells	Molecular classification	Deliniation of cell type-specific expression profiles for PC cells, BPH cells, and PC-associated stromal cells by a new mathematical in silico dissection approach	^[107]
Yu et al. (2004)	Affymetrix U95 (37.7K)	66 PC, 60 NAP, 23 N	Prognostic markers	70-gene tumor aggressiveness predictor signature. 11-principal components model for classification of normal vs cancer	^[37]
Halvorsen et al. (2005)	cDNA array (40K)	29 PC, 23 NAP	Molecular classification	Molecular signature for PC. Multigene model for normal vs cancer classification	^[21]
^Kristiansen et al.* (2005)	cDNA array (4K)	42 PC, 42 NAP	Molecular classification	Molecular signature for PC. CD24 and CD166/MEMD as prognostic markers for PSA recurrence	^[20]
Stephenson et al. (2005)	Affymetrix U133A (22.3K)	79 PC	Prognostic markers	Two logistic regression models for prediction of PSA recurrence. An integrated 'nomogram plus gene signature' model with improved predictive power compared with the clinical nomogram alone	^[11]
^True et al.* (2006)	cDNA array (7.7K)	121 PC (Gleason pattern 3, 4, or 5) and matched NAP cells from 59 individuals	Molecular classification	Molecular signatures for PC and Gleason grade. 86-gene signature for prediction of low- vs high-grade PC	^[22]
^Richardson et al.* (2007)	Affymetrix U133 plus 2.0 (54K)	5 matched PC, NAP, benign stroma and PC-associated stroma cell populations	Molecular classification	Molecular signature for PC and PC-associated stroma	^[108]
^Tamura et al.* (2007)	cDNA array (7.7K)	25 CRPC, 10 HSPC cell populations from 18 and 10 individuals, respectively	Molecular classification	Molecular signature for CRPC by unsupervised clustering	^[32]
^Tomlins et al.* (2007)	cDNA array (20K)	101 cell populations (PC, CRPC, BPH, N, PIA, PIN and stroma) from 44 individuals	Bioinformatic modelling	An integrative molecular concept map model of PC progression from normal epithelium to CRPC	^[31]
^Knight et al.* (2008)	cDNA array (32K)	5 matched basal and luminal prostate epithelial cell populations from BPH patients	Molecular classification	TEAD1 and c-Cbl as prostate basal cell markers and prognostic markers for survival	^[42]
Thorsen et al. (2008)	Affymetrix HuEx 1.0 ST (229K)	15 PC, 10 NAP	Molecular classification	Molecular signature for PC. Identification of PC-specific mRNA splice variants	^[109]
^Dakhova et al.* (2009)	Agilent (41K)	17 matched reactive stroma (grade 3) and benign stroma cell populations from PC patients	Molecular classification	Molecular signature of PC-associated reactive stroma. Identification of deregulated pathways in reactive stroma by gene ontology analysis	^[101]

^*Studies using microdissected samples. All other studies used grossly dissected tissue samples. The list is nonexhaustive.
BPH: Benign prostatic hyperplasia; HG-PIN: High-grade prostatic intraepithelial neoplasia; CRPC: Castrate-resistant prostate cancer; LNM: Lymph node metastasis; MET: Metastasis; N: Normal prostate; NAP: Nonmalignant adjacent prostate; PC: Prostate cancer; PIA: Prostatic inflammatory atrophy; PIN: Prostatic intraepithelial neoplasia.

Table 2. Gene expression-based prognostic models for prostate cancer outcome prediction.
Study (year)	Prognostic model	End point	Independent validation (dataset)	Ref.
*Prostate cancer signatures*
Singh et al. (2002)	Five-gene signature	PSA recurrence	No (cross-validation only)	^[18]
Glinsky et al. (2004)	Three four- to five-gene signatures Combined predictor algorithm	PSA recurrence	Yes (independent set; n = 79)	^[10]
Lapointe et al. (2004)	23-gene signature	PSA recurrence	Yes (Singh et al. [2002]; n = 21)	^[30]
Yu et al. (2004)	70-gene signature	Aggressive PC	Yes (independent set; n = 23)	^[37]
Varambally et al. (2005)	50-gene signature	PSA recurrence	Yes (Glinsky et al. [2004]; n = 79)	^[56]
Stephenson et al. (2005)	Multigene signatures and integrated nomogram plus gene signature model	PSA recurrence	No (cross-validation only)	^[11]
Bismar et al. (2006)	12-gene signature	PSA recurrence	Yes (Glinsky et al. [2004]; n = 79)	^[55]
Bibikova et al. (2007)	16-gene signature	PSA recurrence	No (not done)	^[39]
Cheville et al. (2008)	Four-variable model (two-gene signature, ets-fusion status, and aneuplodi status)	Systemic progression	Yes (independent set; n = 57)	^[57]
Nakagawa et al. (2008)	17-gene signature	Systemic progression Cancer-specific death	Yes (independent set; n = 205)	^[58]
*Other expression signatures applied to prostate cancer*
Ramaswamy et al. (2003)	17-gene metastasis signature	PSA recurrence	Singh et al. (2002); n = 21 (significant)	^[59]
Glinsky et al. (2005)	11-gene stem cell-like signature	PSA recurrence	Singh et al. (2002); n = 21 (significant) Glinsky et al. (2004); n = 79 (significant)	^[60]
Saal et al. (2007)	246-gene PTEN signature	PSA recurrence	Glinsky et al. (2004); n = 79 (significant)	^[64]
Yu et al. (2007)	14-gene Polycomb repression signature	PSA recurrence	Yu et al. (2004); n = 61 (significant) Glinsky et al. (2004); n = 79 (significant)	^[61]

PC: Prostate cancer; PSA: Prostate-specific antigen.

Table 3. MicroRNA expression profiling datasets for prostate cancer.
Study (year)	Profiling platform	miRNAs (n)	RNA extraction	Samples included in study	Ref.
Lu et al. (2005)	Bead-based flow cytometry	217	Small RNAs (18–26 nt)^*	6 PC, 8 nonmalignant	^[82]
Mattie et al. (2006)	Microarray	< 200	Small RNAs (< 200 nt)^*	1 PC, 1 LNM, 1 NAP (pool), 1 TCM, 2 PC cell lines	^[110]
Volinia et al. (2006)	Microarray	228	Total RNA	56 PC, 7 nonmalignant	^[84]
Porkka et al. (2007)	Microarray	319	Small RNAs (< 300 nt)^*	5 PC, 4 CRPC, 4 BPH, 6 PC cell lines, 9 PC xenografts	^[87]
Ambs et al. (2008)	Microarray	329	Total RNA	60 PC, 16 NAP	^[89]
Ozen et al. (2008)	Microarray	480	Small RNAs (< 200 nt)^*	16 PC, 10 NAP	^[88]
Prueitt et al. (2008)	Microarray	235	Total RNA	57 PC (with/without PNI)	^[92]
Rosenfeld et al. (2008)	Microarray	< 600	Total RNA (FFPE)	5 PC, 2 BPH, 1 MET	^[83]
^‡Tong et al. (2008)	Bead-based flow cytometry	114	Total RNA (FFPE)	40 PC, 40 NAP (matched)	^[90]
Leite et al. (2009)	TaqMan-based qRT-PCR	14	Small RNAs (< 200 nt)^§	18 PC, 4 CRPC, 2 PC cell lines	^[111]
Lodes et al. (2009)	Microarray	547	Total RNA (serum)	Serum (6 PC, 8 healthy males)	^[98]
Spahn et al. (2009)	Microarray	665	Total RNA (FFPE)	4 PC and 4 LNM (matched), 4 BPH	^[91]

^*Only mature miRNAs are profiled. See text for further comments.
^‡This study used microdissection. All other studies used macrodissected tissue samples. Studies profiling prostate cancer xenografts or cell lines only are not listed.
^§TaqMan assays were designed to detect mature miRNAs only.
BPH: Benign prostatic hyperplasia; CRPC: Castrate-resistant prostate cancer; FFPE: Formalin-fixed paraffin-embedded; LNM: Lymph node metastasis; MET: Distant metastasis; NAP: Nonmalignant adjacent prostate; PC: Prostate cancer; PNI: Perineural invasion; qRT-PCR: Quantitative reverse-transcriptase polymerase chain reaction; TCM: Transitional cell metaplasia.

Table 4. MicroRNAs with reported functional roles in prostate cancer.
MicroRNA	Regulation in PC^*	Function(s)^‡	Target(s)^§	Ref.
miR-15a-miR-16–1 cluster	Down	Inhibits survival, proliferation and invasion in vitro and in vivo	BCL2, CCND1, WNT3a	^[112]
miR-20a	Up	Inhibits apoptosis	E2F2, E2F3	^[113]
miR-21	Up	Inhibits apoptosis, promotes migration and invasion	MARCKS, PDCD4, TPM1	^[114]
miR-34a	Down	Inhibits cell growth and chemoresistance. Induces cell cycle arrest	SIRT1	^[115]
miR-101	Down	Inhibits proliferation, migration, invasion and colony formation in soft agar	EZH2	^[116]
miR-125b	Down	Promotes CRPC phenotype	BAK1	^[117]
miR-126 ^*	Down	Inhibits migration and invasion	Prostein	^[118]
miR-146a	Down	Inhibits proliferation, invasion and metastasis	ROCK1	^[119]
miR-200b	Down	Inhibits migration, invasion, adhesion and other phenotypic characteristics of EMT	ZEB1, ZEB2	^[120]
miR-205	Down	Inhibits migration and invasion	PRKCE	^[121]
miR-221 and miR-222	Down	Induce cell cycle progression and stimulate colony formation in soft agar	p27(Kip1)	^[94]
	Down	Stimulate PC xenograft growth in vivo	p27(Kip1)	^[95]
	Down	Promote CRPC phenotype	p27(Kip1)	^[96]
miR-330	Down	Induces apoptosis, inhibits cell proliferation and focus-formation	E2F1	^[122]
miR-331–3p	Down	Blocks AR-signaling and PI3K/AKT survival pathway-signaling	ERBB2	^[123]
miR-449a	Down	Induces cell cycle arrest, apoptosis and a senescent-like phenotype	HDAC-1	^[124]
miR-521		Induces radiation sensitivity	CSA	^[125]

^*Compared with miRNA expression levels in nonmalignant prostate tissue, based on data from the studies listed in Table 3 and from the reference(s) listed here.
^‡Derived from in vitro phenotypic assays using cell culture, unless otherwise specified.
^§Functional link to miRNA-induced phenotype(s) not established in all cases.
AR: Androgen receptor; CRPC: Castrate-resistant prostate cancer; EMT: Epithelial–mesenchymal transition; PC: Prostate cancer.

Comments

Commenting is limited to medical professionals. To comment please Log-in.

Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.