Proteomics is widely envisioned as playing a significant role in the translation of genomics to clinically useful applications, especially in the areas of diagnostics and prognostics. In the diagnosis and treatment of kidney disease, a major priority is the identification of disease-associated biomarkers. Proteomics, with its high-throughput and unbiased approach to the analysis of variations in protein expression patterns (actual phenotypic expression of genetic variation), promises to be the most suitable platform for biomarker discovery. Combining such classic analytical techniques as two-dimensional gel electrophoresis with more sophisticated techniques, such as MS, has enabled considerable progress to be made in cataloguing and quantifying proteins present in urine and various kidney tissue compartments in both normal and diseased physiological states. Despite these accomplishments, there remain a number of important challenges that will need to be addressed in order to pave the way for the universal acceptance of proteomics as a clinically relevant diagnostic tool. We discuss issues related to three such critical developmental tasks as follows: (i) completely defining the proteome in the various biological compartments (e.g. tissues, serum and urine) in both health and disease, which presents a major challenge given the dynamic range and complexity of such proteomes; (ii) achieving the routine ability to accurately and reproducibly quantify proteomic expression profiles; and (iii) developing diagnostic platforms that are readily applicable and technically feasible for use in the clinical setting that depend on the fruits of the preceding two tasks to profile multiple disease biomarkers.
- protein expression
- renal disease
- translational research
In a blueprint for the genomic era, the US NHGRI (National Human Genome Research Institute) laid out the major challenges ahead in harnessing the potential of the Human Genome Project to improve humanity's health and well being . In their genomics-to-biology vision, they have identified as one of the grand challenges the ability ‘to take an accurate census of the proteins present in particular cell types under different physiological conditions’. Indeed, this is the primary goal of a collection of efforts under the rubric proteomics. Further to their vision of genomics-to-health, the NHGRI also cited proteomic analysis of body fluids as one of the most promising approaches for the development of non-invasive diagnostic tools for the early detection of human disease. Proteomics is widely envisioned as playing a significant role in the translation of genomics to real health benefits, especially in the areas of diagnostics and prognostics.
In the US alone 20 million people suffer from kidney disease, among which approx. 60000 patients with ESRD (end-stage renal disease) die yearly according to the American Society of Nephrology (http://www.asn-online.org). In the clinical arena of the diagnosis and treatment of kidney disease, a major priority is the identification of disease-associated biomarkers that may find application in, for example, population-based preventive screening of early kidney disease, in the early detection of acute renal failure, in the non-invasive diagnosis of acute renal allograft rejection and in the specific non-invasive diagnosis and prognosis of primary and secondary renal diseases. For the latter two examples, the availability of appropriately sensitive and specific biomarkers may largely obviate the need for diagnostic renal biopsy. Biomarker discovery for use in the early detection of specific renal disease is a nascent field that is quickly gaining impetus. To appreciate the challenges involved in such an endeavour, see Table 1 in which we propose a list of the major characteristics that a biomarker for renal tubular injury should possess (many of these criteria apply to diagnostic biomarkers in general). Assays for such biomarkers should be sensitive enough to detect expression at an early phase of the disease. The presence of the biomarker should be disease-specific, especially if the nature of the injury dictates the therapeutic approach. As a general correlate with the latter criterion, a specific molecular signature may also serve to more precisely reclassify diseases, improving on existing classifications that rely heavily on empirical data. To achieve this, there will probably be a need to make a paradigm shift from the one marker/one disease concept for the case of disease-specific markers that is the basis for most existing clinical markers, to the paradigm of a diagnostic signature or ‘bar code’ that relies on a combination of multiple qualitative and quantitative biomarkers to discriminate between normal and disease state and/or distinguish between related but distinct diseases.
The new era of biomarker discovery has been made possible and is largely facilitated by the recent advances in genomics and related technologies. Through the efforts of functional genomic approaches to ‘make sense’ of the genetic information made available in the form of completed human genome sequence databases, molecular genetic approaches have already been successfully applied to the development of novel molecular diagnostic and prognostic tests in the form of the detection of genetic variations associated with disease and response to therapy (the purview of pharmacogenomics). But it is proteomics, with its high-throughput unbiased approach to the analysis of variations in protein expression patterns (actual phenotypic expression of genetic variation), which promises to be the most suitable platform for biomarker discovery. In this review, we look at the progress of the application of proteomics to the diagnosis of renal diseases in the last 5 years. We begin by surveying the most commonly employed proteomic technologies, briefly describing their strengths and limitations. Subsequently, we consider the application of proteomics in the identification of protein markers in various renal diseases, emphasizing the potential of this approach and the challenges that it presents, and then provide a short overview of representative studies covering proteomics applied to the important study of nephrotoxicity. Finally, we conclude by identifying several critical pathways for development that we propose are important to undertake if we are to realize fully the potential of proteomics as applied to the diagnosis of renal disease.
PROTEOMIC TECHNOLOGICAL APPROACHES
The two most commonly used strategies in proteomics are illustrated in Figure 1 in the form of a simplified analogy with the identification of baggage for an airplane passenger. The first makes use of 2-DE (two-dimensional gel electrophoresis) to resolve proteins according to their molecular mass and pI (iso-electric point), and then picks out protein spots from the 2-DE gel for MS analysis and identification (Figure 1A). The usual choice for identifying the protein is PMF (peptide mass fingerprinting) in which the protein is enzymatically digested (usually with trypsin) and the resulting mass spectrum of the fragmentation peptides constitutes a kind of ‘barcode’ or combination tag to unambiguously identify the protein. The strength of this strategy lies in the user's control in selecting which proteins to submit for MS analysis. Since the resolving step, 2-DE, is itself a technique for comparing protein expression levels, spots corresponding to differentially expressed proteins can conveniently be picked out for further analysis. This is one reason why this strategy has found significant use in comparative proteomics of different tissue compartments and subcompartments in the kidney, such as the proteomic comparison between the cortex and medulla by Witzmann et al.  and Arthur et al. , and between IMCD (inner medullary collecting duct) and IMCD-subtracted kidney lysates by Hoffert et al. .
The second strategy, dubbed shotgun proteomics, couples an inline protein separation technique, usually LC (liquid chromatography), with MS identification (Figure 1B). This strategy is best applied to complex protein mixtures that have undergone trypsin digestion. MS/MS (tandem MS), capable of peptide sequencing, is used to identify the proteins from their fragmentation peptides in the ‘random’ order in which they are presented for MS/MS by their differential elution from the chromatograph. Aside from its speed and scalability, the use of LC overcomes many of the limitations inherent in 2-DE, such as exclusion of very hydrophobic proteins and proteins with extreme pI values. LC can be extended to exploit a number of properties of proteins to achieve resolutions unattainable by 2-DE. One such approach is called MudPIT (multi-dimensional protein identification technology), which employs a strong cation exchanger, followed by reverse-phase chromatography prior to MS [5,6]. Because of the high-resolving capability of LC–MS/MS, it has found promising application in highly complex biological fluids, such as serum and urine, in which the dynamic range of protein abundance (approx. 109) poses an enormous technical challenge . Using this strategy, Spahr and co-workers [8,9] were able to identify a total of 124 unique proteins in urine, despite the complexity of a trypsin-digested unfractionated sample.
Another subset of proteomic technologies constitutes what is known as protein chips. To these belong techniques for profiling proteins based on their specific interaction with functionalized surfaces. A common example is the SELDI (surface-enhanced laser-desorption–ionization) platform, which combines surface affinity capture of proteins with MS analysis . This method has the advantage of speed and ease of sample preparation. Despite issues raised about its robustness and lack of intrinsic capability for de novo protein identification, its potential in biomarker discovery continues to be exploited, as witnessed by an impressive and increasing body of published reports (see below) demonstrating the successful application of SELDI technology in various clinical diagnostic settings.
PROTEOMICS AS A DIAGNOSTIC TOOL
Whole-expression proteomics takes advantage of the high-throughput nature of current proteomic technology platforms to determine the profile of differential protein expression between normal and disease states. This unbiased approach has the potential to identify protein components of novel pathways that may not only add to our understanding of disease pathogenesis, but also suggest disease-associated diagnostic biomarkers as well as therapeutic targets. The two strategies described in Figure 1 are commonly used for this approach to biomarker discovery.
Identification of diagnostic biomarkers for RCC (renal cell carcinoma) was the focus of a significant number of early applications of 2-DE, prompted by the characteristically late diagnosis of this malignancy resulting in high mortality rates. Studies by both Balabanov et al.  and Unwin et al.  showed a decreased expression in tumour tissues of various mitochondrial enzymes involved in carbohydrate metabolism. The latter study, in particular, showed a clear pattern of increased and decreased expression (32 proteins and protein variants increased, and 41 decreased in tumour tissues) consistent with the Warburg effect: an increased glycolytic flux at the expense of the gluconeogenic reactions.
From a prior knowledge of the immunological response to RCC, Klade et al.  took a serological approach to identifying tumour antigens within the RCC proteome, which they call SERPA (serological proteome analysis). Using autologous and allogeneic (patient and normal) sera as sources for reactive antibodies, they performed 2-DE Western blotting of tumour and matched normal tissue samples, and identified immunopositively stained proteins by referring back to the 2-DE spots, which were excised and sequenced by Edman degradation. This method yielded smooth muscle 22-α and carbonic anhydrase I as two antigenic proteins expressed in RCC, but not in normal kidneys. In a similar approach that they called PROTEOMEX, Kellner et al.  showed that protein expression levels of cytokeratin 8, stathmin and vimentin were up-regulated in RCC. An additional nine metabolic proteins were reported to be differentially expressed, with thioredoxin being the most highly up-regulated in RCC . Interestingly, with the exception of vimentin, none of the other six above-named proteins were represented among the 32 proteins reported by Unwin et al.  as having an increased expression in RCC. As sensitivity is not an issue (both employed silver staining), the most likely basis for this apparent discrepancy is differences in sample preparation and sample heterogeneity. Sarto et al.  demonstrated that, by using laser-capture microdissection and magnetic microbeads conjugated with epithelial-cell-specific antibodies, the homogeneity and consistency of kidney cortex samples could be improved.
The use of animal models of kidney disease subjected to 2-DE analyses has also been fruitful in identifying potential disease biomarkers. Pinet et al.  used the two-kidney, one-clip model of renovascular hypertension in the Lewis rat to show the under-expression (greater than 3-fold) of troponin T in the afferent arterioles of the underperfused kidney relative to the contralateral control kidney. Confocal microscopy revealed a pattern of intense renin staining and dramatically reduced troponin T staining in the clipped kidney that correlated well with the high renin levels observed in renovascular hypertension. Using a mouse model of diabetic nephropathy, Thongboonkerd et al.  showed an increase in monocyte/neutrophil elastase inhibitor expression and a decrease in elastase IIIB expression in diabetic kidneys compared with normal kidneys. They correctly predicted from this an increased level of elastin in diabetic kidneys, which they confirmed in three human renal biopsies from patients with Type I diabetes.
The difficulty in applying 2-DE to urinary samples for diagnostic biomarker discovery is largely attributable to the wide dynamic range of protein abundance, as mentioned above, and to potentially high variability between individual urine samples due to the method of collection (random samples compared with 24 h urine collection) and preparation. This may explain, at least in part, the relative paucity of published 2-DE studies aimed at urinary biomarker discovery. As a recent study  has emerged demonstrating that immunosubtraction of highly abundant proteins (i.e. albumin and immunoglobulin) and prefractionation can largely overcome these limitations, we can anticipate more published 2-DE studies of the urinary proteome in health and disease in the near future. Moreover, Lafitte et al.  have shown that it is possible to distinguish 2-DE patterns of four representative samples of kidney diseases from the normal pattern, even when taking into account the inter- and intra-individual variations in the normal urine, although no statistical significance could be inferred from their study. Ward and Brinkley  have reported that uraemic ultrafiltrates from haemodialysis patients exhibit a significantly different proteomic expression profile compared with normal plasma ultrafiltrates by 2-DE analysis. Such studies aim to catalogue a more comprehensive list of candidate uraemic toxins in ESRD.
Shotgun proteomics, on the other hand, has been used by Cutillas and co-workers [22,23] to identify specifically expressed urinary proteins in Dent's disease, a renal Fanconi syndrome characterized by a loss of tubular reabsorption function. By using solid-phase extraction and SCX (strong cation exchange) column fractionation of whole proteins, and subsequent reverse-phase nano LC–MS/MS analysis of products resulting from digestion with trypsin, they were able to identify 100 polypeptides, mostly bioactive proteins (chemokines and cytokines) not detected in normal control samples. A limitation inherent in MS/MS is the poor reproducibility in quantifying signals. By using ICAT (isotope-coded affinity tags) , the same group was able to take advantage of the multi-dimensional LC–MS/MS approach to compare protein quantities between Dent's disease and normal urine . Supplemented by 2-DE, they have shown  that, in addition to an enrichment of cytokines and complement components in Dent's samples, there is also a diminution of vitamin and prosthetic group carriers, pointing to the multiple normal physiological roles of the proximal tubule in reabsorptive transport function as well as regulation of the inflammatory response. Pang et al.  also employed an array of techniques, 2-DE, one-dimensional LC–MS/MS and two-dimensional LC–MS/MS, to search for inflammatory signatures in urine, and verified a number of proteins, one of which is orosmucoid, which has been linked previously to inflammation. A different approach was taken by Kaiser and co-workers [27–29], who used capillary electrophoresis coupled with MS in order to establish migration time compared with mass patterns in normal and disease urine, and in dialysate. The method is fast and mimics the 2-DE map display in the ease with which interpretation and comparison across samples can be performed. A recent application of this technique was able to differentiate diabetic nephropathy from normal control urine, and to further stratify diabetic nephropathy cases into those manifesting albuminuria and those with normal urine albumin levels [29a].
Even with the availability of high-throughput analytical methods, it is often useful to employ a ‘candidate’ approach when possible. This is the case with targeted proteomics, which focuses measurement and analysis to a subset of proteins implicated previously or that are known to play a role in the molecular pathogenesis of a disease, or whose putative role in disease pathogenesis is biologically plausible. In practice, this is done with antibodies, because of the unparalleled specificity that antibody-based analytical methods provide. Using standard SDS/PAGE and immunoblotting, Brooks et al.  studied the differential expression of Na+ and water transporters in the renal tubules of knockout mouse models of the AT1a (angiotensin II type 1a) receptor, the Na+/H+ exchanger type 3 co-transporter and the thiazide-sensitive Na+/Cl− co-transporter. The latter two knockout models were studied specifically to identify a role for compensatory mechanisms when various major reabsorption pathways in the proximal tubule were ablated . Looking at the same set of proteins by a similar approach, Bickel et al.  found a decreased abundance of pre-macula densa Na+ transporters in obese Zucker rat models of Type II diabetes. In a different approach combining flow cytometry with immunostaining of malignant and normal renal cells, and different renal carcinoma cell lines, Li et al.  were able to quantitatively map various known cancer antigens to these cells/cell lines.
MS immunoassay, a technique employed by Kiernan et al. , combines antibody capture and MS to assign variant structures to a number of proteins found in the urine of healthy controls and patients. For instance, they found  that retinol binding protein has a number of catabolic variants distributed differently between plasma and urine, which may be characteristic of healthy functional kidneys. However, their use of intact proteins for MS analysis, although rapid and easy to interpret, imposes limits in resolution that becomes important in proportion to the size of the protein, especially as molecular-mass variants are usually not more than hundreds of Daltons in magnitude. A more focused application of the PROTEOMEX platform by Lichtenfels et al. [36,37] also exemplifies this targeted proteomics strategy. By immunoblotting 2-DE gels with specific sets of antibodies, they were able to look at the expression of the MHC class I antigen-processing and -presentation pathway proteins  and heat-shock proteins  in both IFN-γ (interferon-γ)-treated and untreated RCC lines.
The SELDI platform has attracted a lot of attention in the field of biomarker discovery because it is a rapid technique, is easy to scale up and does not involve specialized sample preparations. For instance, crude lysates of archival cytology specimens, including RCC, have been spotted on to hydrophobic chips to establish ‘protein fingerprints’ that distinguish between related, but clinically distinct, tumour cells, leading to an 87% success rate in correctly classifying tumours from a blinded sample set of histologically diagnosed specimens . Won et al. , employing a decision tree classification algorithm, identified five polypeptide mass values (3–5 kDa range) discriminatory for RCC in serum from RCC patients, healthy controls and patients with other urological disorders spotted on to weak cation exchange chips. Following a similar approach, Tolson et al.  identified potential biomarkers for RCC in the 9–12 kDa mass range. Four of these proteins have been identified by SDS/PAGE and PMF as haptoglobin (9192 Da) and three variants of serum amyloid 1 (intact, 11682 Da; -R, 11526 Da; and -RS, 11439 Da; where R and S stands for truncated amino acids), whereas a protein of 10.85 kDa has remained unassigned to any known protein.
Several groups have employed SELDI profiles from urine to identify peaks or patterns of peaks that might allow for early and specific detection of renal allograft rejection. The earliest study by Clarke et al.  correctly classified 91% of their 34 renal transplant patients (consisting of 17 with acute rejections) by using the CART (classification and regression tree) algorithm. Using an independent statistical analysis, they identified five peaks with the highest ability to discriminate between transplant patients with acute rejection compared with stable allograft function, all of whom had undergone allograft biopsy by protocol for histological diagnosis (‘gold standard’). Working with a slightly larger cohort of 18 patients with acute rejection, compared with 22 patients with stable allograft function, and including a number of controls (ten transplant patients with potential confounding renal diseases, 32 normal patients and five subjects with urinary tract infection), Schaub et al.  defined two general profiles: one associated predominantly with the normal and stable transplant groups, the other with the acute rejection group. Patient time-course studies were also conducted to examine changes in the urinary proteomic profile in response to long-term changes in allograft status. Although rigid criteria for patient classification were applied (all transplant patients, both stable and with acute rejection, were assigned based on allograft biopsy), a number of false positives and false negatives, and unexplained reversal of profiles (i.e. from normal to acute rejection and vice versa), were observed. O'Riordan et al.  more recently studied a patient population of comparable size and achieved a precision of >90% classified correctly as either having acute rejection or stable allograft function by employing two classification algorithms. Notably, although a majority of the diagnostic peaks reported above cluster around the 5–7 kDa range, the three studies did not identify the same peaks (e.g. not a single unique peak was reported in more than one study).
The issue of the robustness of SELDI as a diagnostic tool, especially when using urine samples, has been a subject of recent studies. Rogers et al.  used urine samples obtained from RCC patients, healthy controls and outpatients with benign urogenital disease, which they spotted on to WCX (weak cation exchange) chips to train a NN (neural network) to distinguish RCC from non-RCC cases. A number of models generated were able to classify a ‘blinded’ test set with sensitivities and specificities of 82–83%. However, when subsequently tested on a new set of samples 10 months later, the same NN achieved only sensitivities of 40–60%. Further analysis showed that stability and consistency of samples and their processing contributed minimally to the dramatic decline in predictive performance, whereas the influence of laser and detector performance, chip-to-chip consistency (especially among batches) and possibly robustness of the learning model were more pronounced. Their findings are supported by Schaub and co-workers , who reported that spectra reproducibility is practically unaffected by long-term storage at −70 °C or by freeze/thaw cycles of urine samples. This group also showed close to the same coefficient of variance for peak intensity in reproducibility studies (8–30% compared with 14–57% reported by Rogers et al. ), confirming the poor dependability of SELDI MS in giving quantitative information. Their serial dilution experiment also seemed to suggest a departure from linearity in the peak intensity response to the concentration of a protein calibrant spiked into the urine sample. Finally, the limited usefulness of peak intensities can be inferred from a lack of significant improvement, especially in the subsequent time-delayed trial, to the NN model's performance by incorporating peak intensities as an additional variable .
PROTEOMICS IN NEPHROTOXICOLOGY
Proteomics has also found significant application in studying the effects of chemical insults on the kidney, particularly as a result of environmental toxins, drugs and other bioactive agents. The ability to profile changes in protein expression resulting from these external stimuli complements the toxicologist's traditional arsenal, and has the potential to overcome limitations in the use of animal models of toxicity. Table 2 summarizes a number of representative studies on this important subject, which has recently been reviewed elsewhere .
CONCLUSIONS AND FUTURE DEVELOPMENTS
The potential for proteomics to enhance our capability to diagnose renal diseases is evident in the numerous studies that have been cited above. In so far as proteins are the direct effectors and mediators that determine the disease phenotype, a proteomic approach to disease diagnosis would appear to be ideal. It may even be argued that, for diagnostics to move forward in terms of predictive performance, proteomics is a necessary approach. Because clinically significant perturbations to the normal state of a physiological system (i.e. from the healthy state to the diseased state) will invariably invoke a complex response that alters proteomic expression profiles, it is desirable to have a diagnostic approach that can capture information arising from this complexity by taking a ‘high-resolution global snapshot’. Continued technological advances in proteomic analysis promise to allow the capture of just such types of marvellous complexity and information content. For the time being, proteomics in the post-genome era must still be considered a very young science that requires further development and refinement if it is to become universally accepted as a clinically relevant diagnostic tool. What follows in lieu of a summary is our attempt to identify several critical developments that must take place to fully realize the more focused goal of proteomics as a clinical diagnostic tool (Figure 2).
First is the challenge of completely defining the various proteomes in different compartments (e.g. tissue, serum and urine) and in the contexts of both health and disease. There is a current debate regarding the issue of what proportion of the proteome is currently visible . There is a pervading sense that what remains invisible may be as, if not more, important than what is already visible in holding the secrets to disease pathogenesis. Considering the size of the kidney transcriptome, with genes numbering tens of thousands , the successful identification of several hundreds of proteins in the kidney proteome to date by current proteomic strategies may be considered only a very modest achievement at best. If we add splice variants and post-translational modifications, we are indeed only seeing the tip of the iceberg. It is clear that significant technological advances will be needed to overcome the sheer dynamic range and sensitivity requirements for probing deeper into the proteome. 2-DE may have reached its limits and may only attain a qualitative difference in overall performance through its scaling down, such as lab-on-a-chip technology . MS is the technology with the potential to carry forward the next generation of proteomic analysis. As higher mass accuracy and resolution are being achieved, ushered in by such technologies as FTICR (Fourier-transform ion cyclotron resonance) MS , it may be possible to analyse whole-protein mixtures without the need for enzymatic digest, thereby reducing the complexity of the sample and increasing the chance for detecting low-abundance components. On the other hand, using ingenious sample manipulation, a large impact can already be made with current technologies by deeply interrogating subproteomes, as illustrated in the exploration of vesicular membranes in urine .
The second developmental task is to achieve the routine ability to quantify proteomic expression. If indeed MS will be the technology of choice as proteomics matures, it will have to overcome its limitation in quantitative analysis. As mentioned above in an earlier discussion on SELDI, ion abundance may correlate with protein quantity, but not in a predictable manner useful for the purpose of quantifying proteins. The present approach is to use chemical or isotopic labelling, which shifts the peptide mass values of one sample mixture and thus enables them to be run alongside a standard or comparison mixture. This is indeed the case in ICAT , which specifically targets cysteine residues and has been successfully applied in at least one study of renal disease, the Fanconi syndrome . There are also those approaches that more universally target the peptide population, such as quantification using enhanced signal tags developed by Beardsly and Reilly . This approach, at best semi-quantitative, is not without its inherent difficulties. For example, labelling introduces chemical moieties that may contribute to further ionization disparity among components of a sample, whereas incorporating chemical tags to saturation presents another challenge, to name but a few.
The third critical developmental task is to make use of the fruits of the preceding two tasks to create diagnostic platforms that are readily applicable and technically feasible for use in the clinical setting. Conceptually, the simplest approach is to make use of proteomics as a screening technology to identify candidate biomarkers that can then be assessed in terms of selectivity and specificity by more traditional, albeit very sensitive, methods such as ELISA (e.g. an ELISA-based colorimetric urine dipstick or urine spot test to diagnose acute renal allograft rejection). Skates and Iliopoulos  have illustrated this approach in a proposed algorithm for the diagnosis of RCC. An intermediate approach is achieved by combining immunoassay formats with MS and other parallel methods, as exemplified by SERPA/PROTEOMEX [13,14] and MS immunoassay . Mass-resolving techniques have the ability to detect minute changes in native proteins due to post-translational modifications, point mutations and splice variants, where traditional immunoassays are ‘blind’ to them. Thus combining these techniques has a number of advantages. Although worth exploring further, this method is again dependent on the availability of a host of high-quality antibodies, which can only exist downstream of any proteomics effort and whose availability may thus prove rate limiting. At the other end are issues related to the application of bioinformatics on a large data set, for example an MS output, to extract diagnostic information on a routine basis. As evident from the preceding discussion on SELDI, a number of issues concerning robustness will need to be addressed in order for this approach to come close to being acceptable as a clinical tool. One basic problem is that the models generated from such an analysis are usually over-fitted. Sample processing will also need to be rationalized not merely for reproducibility, but also to account for artefacts such as proteolysis and adduct formation that can confound analysis. Lastly, a large enough sample population is necessary to identify statistically significant variations detected in proteomic case-control association studies. When we consider that the proteome is vastly more complex than the genome, the issue of statistical power to detect a real difference between comparison groups will probably require sizeable numbers that exceed those required for molecular genetic association studies.
We thank Newman Sze's and Eastwood Leung's Proteomics Groups (Genome Institute of Singapore) for so generously sharing their knowledge and expertise. The authors were supported by the Genome Institute of Singapore.
Abbreviations: 2-DE, two-dimensional gel electrophoresis; ESRD, end-stage renal disease; IMCD, inner medullary collecting duct; MS/MS, tandem MS; NHGRI, National Human Genome Research Institute; NN, neural network; pI, iso-electric point; PMF, peptide mass fingerprinting; RCC, renal cell carcinoma; SELDI, surface-enhanced laser-desorption–ionization
- The Biochemical Society