This article is mentioned in:

Introduction

Screening of human tissues for cancer biomarkers is an important task in cancer diagnosis and treatment, which is hindered by the complexity of the sample systems studied. A less complex system such as urine is a preferred medium to screen for protein or peptide biomarkers due to the non-invasive sampling of patients, ease of sampling and the unrestricted quantities obtainable. Urine is relatively stable in terms of protein/peptide composition and fragmentation compared with other bodily fluids such as serum, where proteolytic degradation by endogenous proteases has been shown to occur during or after sample collection (1).

Several investigations have been published describing the urinary peptidome and proteome (as well as biomarker discoveries for several diseases) using methodologies ranging from traditional 2D gel electrophoresis alone (2), or coupled with mass spectrometry (2-DE-MS) (3), immunohistochemistry (4), liquid chromatography mass spectrometry (LC-MS) (5), and surface enhanced laser desorption ionisation-time of flight mass spectrometry (SELDI-TOF-MS) (6–9).

The proteomic screening of urine for potential cancer markers has shown several proteins to be differentially present in ovarian cancer (10). Bladder cancer biomarkers constitute a different non-overlapping set of molecules (11–13), as do potential biomarkers for upper gastrointestinal cancers (9). An improvement in the reliability of diagnostic tests is to employ more than one biomarker synchronously (9,14). For example, one previous study employed an antibody-based array of 810 different antibodies to define peptide patterns in urine associated with cancer (15). A different approach was used successfully in recent years, combining urinary mass spectroscopy with protein/peptide pattern analysis to identify kidney disease (16).

There is a clear need to collect and cross-correlate the wealth of data published in the scientific literature. Currently, there are a number of urinary databases available. The majority consist of lists of identified proteins derived from tryptic digests analysed by liquid chromatography tandem mass spectrometry (LC-MS/MS), such as the Max-Planck Unified Proteome Database (MAPU) (17) and Sys-BodyFluid (18). More recently, a urinary database combining chromatographic reverse-phase retention times and m/z values has been established (19).

However, there is no database available which integrates all of the data. In order to fill this gap, we have assembled datasets from 100 urinary proteomic studies in our novel proteomic database termed the Large Scale Screening Resource (LSSR). LSSR is accessible and downloadable through the Proteomic Analysis DataBase (PADB) portal at www.PADB.org.

In this study, we explore the possibility of discovering novel cancer-associated molecular markers in human urine by subtractive analysis using a novel dataset of the human cancer urinary proteome [derived from patients with upper gastrointestinal (GI) cancer] and comparing it to non-cancer urinary datasets.

Materials and methods

Materials

Tris/Tricine peptide gels, gel-running buffers, CM and IMAC resins, and chromatography buffers were from Bio-Rad (Hemel Hempstead, UK). All other chemicals were obtained from Sigma-Aldrich (Gillingham, UK).

Sample collection

Urine samples were obtained from upper GI cancer patients (n=41) and non-cancer controls (n=21) as described previously (9). Summary participant demographics are shown in Table I. Participant age ranged between 21 and 84 (control group), and 43 and 82 (cancer group). Random morning urine samples were collected over a time period of 2 years. Cancer urine samples were collected prior to surgery if the patient was being considered for resection. All procedures were approved by the local research ethics committee, and written informed consent was obtained. The study conformed to the standards set by the Declaration of Helsinki. All urine samples were kept at −40°C for short-term or −80°C for long-term storage.

Table I Demographics of the study cohort.

Table I

Demographics of the study cohort.

	Cancer (n=41)	Control (n=21)	Entire cohort (n=62)
Age (years)	64 (9.5)	62.1 (23.5)	63.4 (15.6)
Male (M:F)	26:15	17:04	43:19
Primary tumor origin
Pancreas	15	N/A
Oesophagus	9
OGJ	7
Stomach	5
Duodenum	1
Unknown	4
Histology
Adenocarcinoma	34	N/A
Squamous carcinoma	3
Unknown	4

[i] Urine specimens were analysed from cancer patients (n=41) and healthy controls (n=21). Data are presented as means with standard deviations in brackets. OGJ, oesophago-gastric junction.

Chromatographic enrichment of urine proteins and peptides, and sample preparation

Aliquots of 0.5 ml from individual cancer or control urine samples was added to either 30 μl CM10 (n=33 cancer urines, n=8 control urines) or 30 μl IMAC30 (Cu2+-chelated) (n=21 cancer urines, n=19 control urines) spin column resin (Bio-Rad) and 0.75 ml binding buffer (either 0.1 M NaH3C2O2 pH 4.0 for CM resin, or 0.1 M NaHPO4 pH 7.0 including 0.5 M NaCl for IMAC30 resin) and incubated for 1 h at room temperature under constant agitation. Sample and resin combinations were chosen based on independent analyses using peak stratification by SELDI mass spectrometry (9). Unbound material was removed and the resin washed four times with 0.3 ml binding buffer. Bound material was separated by electrophoresis on a 16.5% Tris-Tricine gel (Bio-Rad), and gel bands in the region of 2–10 kDa were excised after Coomassie staining (BioSafe Coomassie; Bio-Rad). The molecular mass range of 2–10 kDa was selected since many urinary proteins are derived from proteolytic processing and urinary shedding as described (20). Additionally, we previously observed potential urinary cancer markers in this mass range (9).

LC-MS/MS mass spectrometry

Proteins and peptides from gel bands were digested in situ with trypsin. The resulting peptides were eluted with acetonitrile (ACN), and analysed by LC-MS/ MS (21). The LC-MS system consisted of an Agilent 1200 Series HPLC (Agilent Technologies, Yarnton, UK) with a Kasil sealed fused silica pre-column (Next Advance, New York, NY, USA) packed to a length of ~3 cm with Pursuit C18, 5 μm particle size (Varian, Crawley, UK) and PicoTip Emitter analytical column PF 360-75-15-N-5 (New Objective, Woburn, MA, USA) packed to a length of ~20 cm with Pursuit C18, 5 μm particle size (Varian). The column was equilibrated with solvent A (0.1% formic acid in 2.5% acetonitrile) and eluted with a linear gradient from 0 to 10% over 6 to 8 min; from 8 to 60% over 8 to 35 min; from 60 to 100% over 35 to 40 min; solvent B (0.1% formic acid, 0.025% TFA in 90% acetonitrile) over 45 min at a flow rate of 5 μl/min. The LTQ mass spectrometer (Thermo Scientific, Epsom, UK) was fitted with a NanoLC ESI source. Data-dependent acquisition was controlled by XCalibur software. Fragmentation spectra were then processed by XCalibur and BioWorks software (Thermo Fisher Scientific, Loughborough, UK) and submitted to the Mascot search engine (Matrix Science, London, UK) using UniProt/SwissProt (release May 2011, Homo sapiens, 18055 sequences) as the reference database. Mascot search parameters were: enzyme specificity trypsin, maximum missed cleavage 1, fixed modifications cysteine carbamidomethylation, variable modification methionine oxidation, precursor mass tolerance +/−3 kDa, fragment ion mass tolerance +/−0.4 kDa. Only Mascot hits with a false discovery rate (FDR) ≤0.05 were taken into consideration.

Meta-analysis and subtractive data analysis

Proteins with at least two peptide matches were analysed further by comparing molecules that were only observed in urine samples from cancer patients with a database consisting of proteins found by other studies in urine, blood and kidney. This database was assembled from 136 publications, listing 146 tissue-specific datasets. The blood datasets covered plasma, serum and erythrocytes; the kidney studies were derived from analyses of cortex, medulla, epithelium, glomerulus, inner medullary collecting duct, mesangium, parenchyma, peroxisomal membrane, peroxisome, basolateral membrane vesicles, brush border membrane vesicles, urothelial mucosa and whole kidney; and urine datasets described either the whole or exosomal proteomes. All entries were then matched to the UniProt database, followed by clustering to individual (unique) entries by annotating splice and variant entries to common parent molecules and ultimately assigning each unique cluster an in-house specific accession number. Additionally, all proteins mapping to immunoglobulins were clustered into one generic cluster, as well as all proteins belonging to the Major Histocompatibility Complex (MHC). Merging and subtraction analysis was done using software written in-house. We also manually added our own functional classification tags to each molecular cluster, based on known properties of each molecule, giving an abridged view of proteome compositions.

Results

Urine samples were extracted from 21 healthy non-cancer controls and 41 patients with upper GI cancer (n=41) (Table I). Of the 41 cancer patients, staging investigations demonstrated that at least 29 (70.7%) had nodal or metastatic disease. We analysed all 62 urine samples by LC-MS/MS in the region of 2–10 kDa by chromatographic enrichment using either CM10, IMAC30, or both resin types individually, resulting in a total of 81 chromatographic enrichments, followed by gel analysis, tryptic digestion and mass spectrometry. All molecular weight regions cut from gels were identical in at least three samples from each cohort group, thus also allowing comparison of identified molecules on a gel-region by gel-region basis After data extraction by Mascot searching (resulting in 35,801 peptides covering 7,639 proteins) and applying discovery criteria of a FDR ≤0.05 and a minimal Mascot score of 13, the resulting 81 datasets were further analysed by merging all protein lists. This yielded 1,228 unique non-redundant entries (data not shown). Additionally, all molecules relating to either immunoglobulins or MHC were also merged into two individual clusters since members of these two families are well known to show a great degree of hypervariability, and therefore they may skew any analysis towards single entries from those classes, since they are not expected to show any duplications across the datasets analysed in this study. The final list consisted of 1,161 molecular clusters. Furthermore, we re-classified all molecules in the datasets available by manually annotating every protein with a single molecular property or functionality tag as listed in the legend of Fig. 1. The properties or functionalities were assigned based on known properties of each individual protein, either from original publications or derived from database annotations, such as enzyme nomenclatures, sequence homologies and domain analysis. The compositional analysis of the merged datasets of blood, urine and kidney proteomes, as well as our urinary dataset is shown in Fig. 2. It was clear that all merged datasets consist of ~25% enzymes, 10% cell-shape molecules, 10% transcriptional or translational elements and 10% transport molecules. However, our novel dataset appeared to contain more cell-shape and transcriptional/translational proteins and less transport molecules, which may reflect an association with disease, rather than a general breakdown of cellular components.

Figure 1 Flow-diagram of the steps involved to elucidate potential novel cancer markers.

Figure 2 Composition of blood, urine, kidney and our datasets used in this study based on functional classifications. All merged datasets were analysed based on the functional description of each molecule assigned by our database and depicted as percentage pie-charts. The legend listing all possible classes is displayed on the right.

The 1,161 molecules were then split into groups depending on whether they were observed in cancer urine samples, or urine from healthy individuals (Fig. 3A). The 745 proteins only found in cancer urine samples were then tagged and the entire dataset compared to data of 31,743 unmerged entries derived from 146 tissue-specific datasets from 137 publications (data not shown). This external data consisted of 9,707 merged entries, covering proteomic studies from urine, kidney and blood (Table II). A comparative analysis of our dataset with the three largest urinary proteome profiling datasets showed a 46% overlap of our data with the dataset from Kentsis et al (22), a 41% overlap with the study by Adachi et al (23), and a 21% with the urinary exosome dataset from Gonzales et al (24) (Fig. 3B). A global comparison between proteomes from urine, kidney and blood (Fig. 3C) demonstrated a slightly larger overlap of the urinary proteome with the kidney proteome than the blood proteome.

Figure 3 Venn diagrams of the meta-analysis to define potential cancer-associated molecules. Our dataset was analysed to define the overlap with datasets described in the literature. (A) Our dataset split into potential cancer markers by subtraction of molecules found in urine samples from healthy subjects. (B) Venn diagram of the four largest urinary datasets listed in the LSSR database, including the number of overlapping entries, and the number of questionable entries based on single peptide identification in brackets. (C) Overlap of all proteins found in urine with those found in blood and kidney, based on the datasets listed in the LSSR database.

Table II Number of entries listed in the LSSR database for analysed samples derived from blood, urine and kidney.

Table II

Number of entries listed in the LSSR database for analysed samples derived from blood, urine and kidney.

	No. of entries prior to merge	Merged entries	No. of studies
Urine	13,635	5,868	101
Blood	4,433	3,660	12
Kidney	13,675	4,964	34

[i] The number of entries by tissue type is given either as numbers derived directly from the studies analysed, or after merging all datasets based on unique identifiers assigned by our database.

We then performed subtractive analysis on our urinary proteome data by eliminating any potential cancer candidate molecule if it was found in any of the urinary datasets unrelated to cancer. This reduced dataset of 268 proteins (data not shown) was further condensed by removing any entries which did not have a spectral count of at least two, resulting in 44 proteins, of which 24 were found uniquely in our study (in comparison to all other datasets), and 20 which were also found in the other tissues (Table III). All 44 of these proteins were then analysed by searching the Online Mendelian Inheritance in Man (OMIM) database for publications where these molecules were reported to be directly associated with human disease or cancer. Fourteen proteins were annotated in OMIM to be causative for a disease, and 30 were known to be involved in cancer.

Table III List of potential cancer candidate markers from human urine.

Table III

List of potential cancer candidate markers from human urine.

Peptide count	Spectral count	Gene	Protein	OMIM disease	PADB classification	Tissue	Molecular function	Cancer type	PubMed (cancer association)
Only detected in cancer patient urine, high confidence dataset
10	11	POTEKP	Putative β-actin-like protein 3		CS: Cell shape	Urine	Actin filament de-/re-polymerization	Hepatocellular carcinoma	16824795
194	3	DCP1A	mRNA-decapping enzyme 1A		ENZ: enzyme, enzymatic properties	Urine	Transcriptional co-activator	Gastric cancer	23932921
93	3	NAV1	Neuron navigator 1		DEV: development	Urine	neuronal migration
80	3	ZFYVE20	Rabenosyn-5		TP: transport, storage, endocytosis, exocytosis, vesicles	Urine	endosomal transport
77	3	PLA1A	Phospholipase A1 member A		ENZ: enzyme, enzymatic properties	Urine	Lipid metabolism	Prostate cancer	22904677
5	3	GLB1L	β-galactosidase-1-like protein		ENZ: enzyme, enzymatic properties	Urine	Glycosyl hydrolase, carbohydrate metabolism
32	2	COX4I2	Cytochrome c oxidase subunit 4 isoform 2, mitochondrial	Exocrine pancreatic insufficiency, dyserythropoietic anemia, calvarial hyperostosis	TP: transport, storage, endocytosis, exocytosis, vesicles	Urine	Mitochondrial electron transport	General (Warburg effect)	22320183
20	2	SOS2	Son of sevenless homolog 2		MOD: modulator, regulator	Urine	Guanine-nucleotide releasing factor
20	2	GALNT6	Polypeptide N-acetylgalactosa minyltransferase 6		ENZ: enzyme, enzymatic properties	Urine	Post-translational protein O-linked glycosylation	Breast cancer	20215525
17	2	CCDC88C	Protein Daple	Autosomal recessive nonsyndromic hydrocephalus HYC1	SIG: signaling	Urine	Negative regulator of canonical Wnt signaling pathway	Breast cancer	23593120
15	2	TTI1	TEL2-interacting protein 1 homolog		MOD: modulator, regulator	Urine	Regulator of DNA damage response	Multiple myeloma	23263282
13	2	RPGRIP1	X-linked retinitis-pigmentosa GTPase regulator interacting protein 1	Leber congenital amaurosis 6	CS: Cell shape	Urine	Sensory transduction
13	2	GBP4	Guanylate-binding protein 4		DEV: development	Urine	GTP hydrolysis
11	2	MTTP	Microsomal triglyceride transfer protein large subunit	Aβ lipoproteinemia	TP: transport, storage, endocytosis, exocytosis, vesicles	Urine	Lipid transport, plasma lipoprotein secretion	Small intestinal cancer	12630961
11	2	ERBB2	Receptor tyrosine-protein kinase erbB-2	Glioma susceptibility 1; ovarian cancer; lung cancer; gastric cancer	ENZ: enzyme, enzymatic properties	Urine	Protein tyrosine kinase involved in transcriptional regulation	Multiple	22014070
9	2	PLEKHG2	Pleckstrin homology domain-containing family G member 2		MOD: modulator, regulator	Urine	Guanine-nucleotide releasing factor	Pancreatic cancer	24041470
7	2	POLA2	DNA polymerase α subunit B		TF: transcription and translation	Urine	DNA replication and cell proliferation	Melanoma	24987109
6	2	GPSM2	G-protein-signaling modulator 2	Deafness, autosomal recessive 82	CC: cell cycle (turnover, mitosis, meiosis)	Urine	G-protein coupled receptor signaling pathway, spindle pole orientation	Breast cancer	20589935
6	2	GDNF	Glial cell line-derived neurotrophic factor	Central hypoventilation syndrome; Hirschsprung disease, susceptibility to, 3	SIG: signaling	Urine	neurotrophic factor	Pancreatic cancer	20960036
5	2	DDHD2	Phospholipase DDHD2		ENZ: enzyme, enzymatic properties	Urine	Lipid degradation and metabolism	Breast cancer	20940404
4	2	TEAD2	Transcriptional enhancer factor TEF-4		TF: transcription and translation	Urine	Transcription regulation	Prostate cancer	19478945
3	2	SARM1	Sterile α and TIR motif-containing protein 1		MOD: modulator, regulator	Urine	Regulator of Toll-like receptor signaling pathway	Colorectal cancer	20426761
3	2	DOK7	Downstream of tyrosine kinase 7	Myasthenia, limb-girdle	SIG: signaling	Urine	Neuromuscular synaptogenesis	Breast cancer	23054610
2	2	ZDHHC6	Probable palmitoyltransferase ZDHHC6		ENZ: enzyme, enzymatic properties	Urine	Protein palmitoylation
Detected in urine from cancer patients and other tissues, high confidence dataset
9	5	HIST3H3	Histone H3.1t		TF: transcription and translation	Kidney, urine	Transcription regulation, DNA repair, DNA replication
25	4	HIST1H1E	Histone H1.4		TF: transcription and translation	Kidney, urine	Regulator of gene transcription	Endometrial cancer cells	23682076
3	4	BOLA2	BolA-like protein 2		UK: unknown	Kidney, urine	Redox control	Liver cancer	22653869
116	3	NAV2	Neuron navigator 2		ENZ: enzyme, enzymatic properties	Blood, urine	Neuronal development	Colorectal carcinoma	22810696
25	3	MLL3	Histone-lysine N-methyltransferase MLL3		ENZ: enzyme, enzymatic properties	Blood, urine	Transcriptional coactivation	Colorectal cancer	21853109
39	2	FMR1	Fragile X mental retardation 1 protein	Fragile x tremor/ataxia syndrome; fragile x mental retardation syndrome; premature ovarian failure 1	TF: transcription and translation	Kidney, urine	Translation repressor	Hepatocellular carcinoma	17786358
37	2	TJP1	Tight junction protein ZO-1		CS: cell shape	Kidney, urine	Tight junction assembly, cell migration	Non-small cell lung cancer	24294375
35	2	PCDH17	Protocadherin-17		CS: cell shape	Blood, urine	Calcium-dependent cell-adhesion protein	Laryngeal squamous cell carcinoma	21213369
30	2	NSUN5	Putative methyltransferase NSUN5	Williams-Beuren syndrome	ENZ: enzyme, enzymatic properties	Blood, urine	Methyl-transferase, embryonic development
13	2	ST14	Suppressor of tumorigenicity 14 protein	Ichthyosis with hypotrichosis, autosomal recessive	ENZ: enzyme, enzymatic properties	Kidney, urine	Degradation of extracellular matrix	Breast cancer	20716618
13	2	SON	Bax antagonist selected in saccharomyces 1		TF: transcription and translation	Kidney, urine	Splicing cofactor
9	2	CPB1	Carboxypeptidase B		ENZ: enzyme, enzymatic properties	Blood, urine	Carboxypeptidase, protein degradation	Pancreatic cancer	1688389
6	2	HBG2	Hemoglobin subunit γ-2	Cyanosis transient neonatal	TP: transport, storage, endocytosis, exocytosis, vesicles	Blood, kidney, urine	Oxygen transport
6	2	AKAP2	A-kinase anchor protein 2		SCA: scaffolder, docking, adaptor	Kidney, urine	Protein kinase A-anchoring protein	Ovarian cancer	19123201
5	2	DYNLL2	Dynein light chain 2, cytoplasmic		TP: transport, storage, endocytosis, exocytosis, vesicles	Blood, kidney, urine	Microtubule-based transport
5	2	NCOR1	Nuclear receptor corepressor 1		TF: transcription and translation	Blood, urine	Transcriptional repressor	Prostate cancer	20466759
5	2	MKX	Homeobox protein Mohawk		TF: transcription and translation	Kidney, urine	Morphogenetic regulator of cell adhesion
4	2	ACTN2	α-actinin-2	Cardiomyopathy, dilated, 1aa	CS: cell shape	Blood, kidney, urine	Actin-anchoring	Metastatic pancreatic endocrine neoplasm	15448002
4	2	CORO1A	Coronin-1A	Immunodeficiency 8	CS: cell shape	Blood, kidney, urine	Crucial component of cytoskeletal modulation	Breast cancer	21489049
2	2	CELF5	CUGBP Elav-like family member 5		TF: transcription and translation	Kidney, urine	Regulation of pre-mRNA alternative splicing

[i] Molecules found uniquely in our urinary dataset (with peptide and spectral counts of at least two) but not in non-cancer urine samples are listed by gene and protein names, their individual peptide and spectral counts, and whether they are known to be associated with human disease based on the OMIM database. The tissue type in which the molecule was found, based on meta-analysis of external datasets, a classification-tag, and the molecular function are included. Additionally, a PubMed identification number is listed if the protein has been described to be directly associated with cancer, including the cancer type. The dataset is divided based on whether the proteins were only found in our analysis, or whether they were also detected in other proteomic non-urinary screens.

Discussion

Proteomic large-scale analysis of tissues to define a cancer state can be time- and resource-consuming, especially in light of an unknown end-point. Therefore, it could be helpful to compare a novel dataset with known data in order to establish whether potential disease markers are observable, and thereby analyse a simplified dataset for the disease in question. This approach does not address the issue of quantitative comparisons, but it is rather a qualitative approach. However, the resulting list of potential candidate molecules will have a specificity of 100%. Here, we test this hypothesis by applying a subtractive analysis method in conjunction with large-scale meta-analysis of urinary datasets to screen for potential novel cancer markers observable in human urine.

An initial comparison of functional profiles of urine, blood and kidney proteomes showed no major discernible difference between those datasets. This finding, in itself is not surprising, since it is expected that these systems should reflect an overall similar composition through a combination of immediate environment and source. Blood, containing a substantial amount of cells, is also expected to show a reasonably uniform functional composition profile compared with other tissues e.g. kidney. Our novel urinary dataset, having an expected bias towards an aberrant functional profile due to overexpressed molecules associated with disease, contains more molecules involved in cellular contacts, morphology and cytoskeletal aspects, as well as transcriptional/translational components, which may be directly linked to abnormal and uncontrolled cellular growth.

Comparison of our dataset with known non-cancer urinary proteomes yielded a set of only 44 molecules specific for our cancer data, of which 68% are already known to be involved in cancer. The functional profile of those 44 proteins in comparison to the merged urinary proteome profile showed mainly an enrichment of developmental proteins (5%), signaling molecules (7%) and, most strikingly, transcriptional/ translational proteins (20%). The known cancer-associated molecules described have been suggested to be involved in hepatocellular carcinoma [κ actin (POTEKP) (25); BolA-like protein 2 (BOLA2) (26); fragile X mental retardation 1 protein (FMR1) (27)]; mammary carcinogenesis [polypeptide N-acetylgalactosaminyltrans ferase 6 (GALNT6) (28); protein Daple (CCDC88C) (29); G-protein-signaling modulator 2 (GPSM2) (30); phospho-lipase DDHD2 (DDHD2) (31); downstream of tyrosine kinase 7 (DOK7) (32); suppressor of tumorigenicity 14 protein (ST14) (33); coronin-1A (CORO1A) (34)], lung cancer [tight junction protein ZO-1 (TJP1) (35)], prostate cancer [phos-pholipase A1 member A (PLA1A) (36); transcriptional enhancer factor TEF-4 (TEAD2) (37); nuclear receptor corepressor 1 (NCOR1) (38)], ovarian cancer [A-kinase anchor protein 2 (AKAP2) (39)], colorectal cancer [sterile α and TIR motif-containing protein 1 (SARM1) (40); neuron navigator 2 (NAV2) (41); histone-lysine N-methyltransferase MLL3 (MLL3) (42)], pancreatic cancer [pleckstrin homology domain-containing family G member 2 (PLEKHG2) (43); glial cell line-derived neurotrophic factor (GDNF) (44); carboxypeptidase B (CPB1) (45); α-actinin-2 (ACTN2) (46)], gastric cancer [mRNA-decapping enzyme 1A (DCP1A) (47), a co-activator in TGF-β signaling (48)], melanoma [DNA polymerase α subunit B (POLA2) (49)], multiple myeloma [TEL2-interacting protein 1 homolog (TTI1) (50)], endometrial cancer cells [Histone H1.4 (HIST1H1E (51)], laryngeal squamous cell carcinoma [protocadherin-17 (PCDH17) (52)], and adenocarcinoma [microsomal triglyceride transfer protein large subunit (MTTP) (53)]. Additionally, the latter protein was also described to be a pivotal element in the cancer-associated muscle-wasting disease cachexia (54). Some of these proteins may be differentially regulated across a range of different cancer types and may therefore represent key cancer markers. For example, receptor tyrosine-protein kinase erbB-2 (ERBB2) has been described to be a marker for various cancer types, such as gastroesophageal (55), breast (56), lung (57), gallbladder (58) and pancreatic cancer (59), as well as uterine serous adenocarcinoma (60), and others. Another known protein to be involved in cancer progression is the mitochondrial cytochrome c oxidase subunit 4 isoform 2 (COX4I2), which is part of the Warburg effect, where cancer cells show higher propensity to produce lactate independent of oxygen presence or absence (61).

Of the proteins not previously described in association with cancer, transcription factor Bax antagonists selected in Saccharomyces 1 (SON), homeobox protein Mohawk (MKX) and CUGBP Elav-like family member 5 (CELF5) may represent other potential lead candidates in cancer stratification. Other important markers may include developmental molecules, such as guanylate-binding protein 4 GBP4, which is a negative regulator of virus-triggered cellular responses (62) and is involved in GTP hydrolysis, or neuron navigator NAV1, which has been reported to be a neuronal guidance molecule (63). However, its role in cancer or outside the neuronal environment remains to be elucidated.

In conclusion, we have demonstrated that a subtractive analysis of proteomic datasets can yield a number of potential diagnostic cancer targets in human urine. Further specific screening of urine, based on our findings, using, for example, an antibody-based approach, will establish whether our potential markers are associated with a general cancer status, or if they are specific for a defined cancer type such as pancreatic or esophageal cancer. Additionally, since the data in our database can easily be expanded to contain further datasets, there are other, as yet undefined diseases, which can be addressed by establishing and comparing a relatively small disease-specific dataset. This approach also has the advantage of rapid turnover and increased cost-effectiveness relating to large-scale analyses of tissue and cell proteomes for the discovery of novel molecular markers. In this regard, we are encouraging researchers to submit their published datasets to be incorporated in the LSSR database. All data will be freely available through the PADB portal at www.PADB.org.

Acknowledgements

We thank C.A. Greig, N.A. Stephens and H. Wackerhage for patient recruitment. Funding of this study was provided by the University of Edinburgh.

References