Diabetic Nephropathy: A Frontier for Personalized Medicine

Proteomics of Kidney Tissue in DNP

Proteomics measures protein level at a global scale. Proteomics studies are becoming increasingly popular because they provide direct information about protein levels, which are essential to execute gene functions (16). Proteomics has been applied successfully to rodent models of diabetic nephropathy (17), but it requires large amounts of kidney material; therefore, it is less suitable for the analysis of human kidney biopsy materials. Recently, Thongboonkerd et al. (17) performed proteomic analysis on kidneys of 120-d-old OVE26 transgenic mice, which display many characteristics of early-onset human type 1 diabetes (18). They used two-dimensional gel electrophoresis with SYPRO Ruby staining, quantitative intensity analysis, and matrix assisted laser desorption ionization–time of flight mass spectrometry. Thirty proteins were identified as differentially expressed in the diabetic kidney, including proteases, protease inhibitors, apoptosis-associated proteins, regulators of oxidative tolerance, calcium-binding proteins, transport regulators, cell signaling proteins, and smooth muscle contractile elements. Nineteen of the altered proteins had previously been shown to be regulated during diabetes, whereas roles for the other 11 altered proteins had not previously been established, suggesting that they may function by novel mechanisms in diabetic nephropathy.

Proteomics to Analyze Urinary Protein Profiles

Application of proteomics tools to analyze urinary protein profiles is rapidly emerging as a promising research area for marker discovery (19). The most important advantage of urine proteomics is the prospect of a noninvasive and easy sampling method. Urinary proteins are a mixture of proteins filtered from plasma or secreted by kidney cells. Normal urinary proteins generally reflect normal tubular physiology. Information on changes in urinary protein excretion by various interventions is essential for a better understanding of tubular and glomerular responses to physiologic stimuli. Rossing et al. (20) performed urinary proteomics analysis in four groups of patients with type 2 diabetes, matched for age gender and diabetes duration, including 20 normoalbuminuric patients, 20 microalbuminuric patients with retinopathy, and 18 macroalbuminuric patients with retinopathy. Furthermore, changes in urinary polypeptide patterns during treatment with the angiotensin receptor blocker candesartan were evaluated in the macroalbuminuric patients in a randomized, double-blinded, cross-over trial in which each patient received treatment with placebo and with candesartan 8, 16, and 32 mg/d each for 2 mo. They used a combination of capillary electrophoresis and mass spectrometry to identify differentially regulated proteins and were able to identify 4551 different polypeptides in the urine samples. Urinary polypeptide patterns were not significantly different in normo- and microalbuminuric patients, whereas distinct differences were found in macroalbuminuric patients. Differences in urinary polypeptide patterns between normo- and macroalbuminuric patients permitted the establishment of a “diabetic renal damage” pattern that consisted of 113 polypeptides. Eleven of these polypeptides had been sequenced and identified. Candesartan treatment in macroalbuminuric patients significantly changed 15 of the 113 polypeptides in the diabetic renal damage pattern toward levels in normoalbuminuric patients. Change in the diabetic renal damage pattern was not candesartan dose dependent, but individual changes correlated with changes in urinary albumin excretion at each dose level.

Similar studies have been conducted by Mischak et al. (21) on patients with type 2 diabetes and DNP. These studies indicate that urine proteomics could be adapted for pharmacoproteomics approaches, such as identification of responders and nonresponders of RAS blockade, etc. Sharma et al. (22) used two-dimensional differential in-gel electrophoretic analysis to examine protein patterns in urine samples in patients with DNP. All patients had longstanding diabetes, impaired renal function, and overt proteinuria. They identified 99 differentially regulated spots in the urine proteome of the diabetic samples, with 63 up- and 36 downregulated. One spot was consistently upregulated by 19-fold across individuals in the diabetic group, and it was identified as α 1 antitrypsin. ELISA of urine samples from a separate group of patients and control subjects confirmed a marked increase of α 1 antitrypsin in patients with diabetes.

Barriers for Proteomics Applications in Marker Discovery

Although proteomics strategies are rapidly emerging in renal research, considerable technical limitations remain to be overcome. First, current technologies lack sensitivity to allow detection of moderate- and low-abundance proteins. Second, reliable algorithms to facilitate standardization of proteomics data sets to enable reliable and large-scale comparisons between many samples or even between laboratories are not readily available and need to be developed. Third, proteolytic cleavage of proteins into peptides occurs rapidly and uncontrollably and presents a major problem for the analysis of proteomics data sets. New approaches are under development using gel-free technology and sample fractionation before analysis to improve sensitivity for detection of membrane and low-abundance proteins.

Technology and Applications

The development of microarray technology has revolutionized functional genomics by providing tools that allow the parallel measurement of gene expression in whole genomes with thousands of genes (23–25). Microarrays are artificially constructed grids of probes, such that each element of the grid probes for a specific RNA sequence; that is, each holds a DNA sequence that is a reverse complement to the target RNA sequence. In gene expression microarrays, either synthetic oligonucleotides or cDNA fragments are used as probes (26,27). Although, there are many protocols and platforms available, the basic technique involves extraction of RNA from biologic samples, followed by complementary RNA copying with incorporation of either fluorescence nucleotides or tags that can be labeled subsequently. Thus, samples contain a complete complement of labeled RNA or cDNA species that can be hybridized to microarrayed specific complementary sequence probes. This procedure typically generates thousands of measurements of gene expression per biologic sample. The cost of microarray experimentation continues to decrease rapidly, allowing for much broader use of this powerful technology in many translational research disciplines, including biomarker discovery, drug discovery, pharmacogenomics, toxicogenomics, systems biology, and molecular pathology (28,29).

Technical Barriers for Application of Gene Expression Profiling in Renal Biopsy Samples

Although clearly very promising, the application of microarray technology to analyze gene expression profiles in renal biopsy tissue is also extremely challenging (30–32). Renal biopsy tissue is typically very heterogeneous, with proportions of glomerular and tubular segments often varying dramatically, depending on the path of the biopsy needle in kidney tissue during the procedure. The considerable sample heterogeneity of undissected kidney biopsy cores precludes reliable sample-to-sample comparisons. Two different solutions to this problem are being pursued by various investigators. Pioneering studies by Kretzler et al. (32,33) have established that manual microdissection of fresh renal biopsy cores is a productive approach that allows efficient separation of glomeruli and tubuli. It is critical that biopsy cores be placed immediately in storage solutions that contain inhibitors of RNase to prevent RNA degradation and to conserve an in vivo transcriptome profile. Similarly, our group has developed a microdissection-based method to prepare glomerular and tubular segments of fresh kidney biopsy cores for microarray analysis (K.S. and E.P.B., unpublished observations). Although this approach provides consistently high-quality RNA samples, it is very labor-intensive and requires considerable skills in microdissection, limiting its current application to clinical and translational research projects.

Laser capture microdissection is an alternative technology that can be applied to retrieve glomeruli from fixed kidney tissue sections under microscopic guidance (34). The procedure allows accurate identification and procurement of glomerular or tubular cells from tissue samples under direct microscopic visualization. Disadvantages of laser capture microdissection approaches are the considerable inconsistency of RNA quality caused by degradation induced during tissue fixation, sectioning, and laser-induced degradation. Both manual and laser capture microdissection methods typically provide very small amounts of tissue. Until recently, it was not possible to isolate and quantify very small amounts of RNA (typically picogram quantities) reliably. In addition, quality control of such small samples was not possible. Several research tool companies recently introduced new technologies to overcome these technical limitations, including resin-based RNA extraction and lab-on-chip–based microfluidic electrophoresis methods. Thus, these technical challenges now have been solved through the development of refined protocols and improved reagents and tools.

Paucity of Kidney Biopsies in Individuals Who Have Diabetes without and with Clinical DNP

In general, the diagnosis of DNP is established on the basis of clinical findings, including patient history, signs and symptoms, and urinary protein excretion. Because analysis of kidney tissue by renal pathologists is not required to diagnose DNP, kidney biopsies are only rarely performed in individuals with diabetes. As a consequence, kidney tissue is typically not available for research studies. The paucity of new kidney biopsy tissue is severely restricting translational molecular research in DNP and prevents the application of microarray analysis for gene expression profiling in human DNP. Thus, gene expression profiling in DNP research is currently largely restricted to experimental animal models. In addition, it has proved very difficult to obtain approval for protocol research kidney biopsies from institutional review boards at US medical centers, with few exceptions, largely because of the risk associated with this invasive procedure. Although considerable progress has been achieved in our understanding of the pathomechanisms of established DNP with clinical manifestations in both type 1 and type 2 diabetes, the early molecular and cellular changes that are induced in kidney with the onset of diabetes in individuals who are at risk for developing DNP remain unclear. Therefore, the paucity of kidney biopsies in individuals who have diabetes without and with clinical DNP is to a large extent responsible for the relative standstill in translational research of early DNP.

Challenges for Biologically or Clinically Meaningful Analysis of Gene Expression Data

Microarrays deliver large amounts of data on tens of thousands of genes. The result is an immense quantity of biologic information that needs to be analyzed, presented, and archived in a meaningful way. Therefore, functional genomic studies should be combined with advanced computational and biostatistical approaches (35). For potential identification of biomarkers, the gene expression data have to be analyzed in conjunction with patient and sample variables. The most basic question that one can ask in a transcriptional profiling experiment is which genes’ expression levels changed significantly when gene expression levels in two different groups are compared. Although the statistical methods to identify lists of differentially expressed genes are very powerful, it is considerably more challenging to determine meaningful correlations between gene expression patterns and clinical parameters.

In general, pattern discovery methods such as clustering provide a high-level overview of a data set and may be the first analysis step in a study that ultimately involves other analytical methods (36). Such techniques include dimension-reduction methods, as well as various “clustering” techniques designed for finding groups within the data. What these methods have in common is that they simplify the data set, ideally in ways that impart additional information about its structure, and that they are considered “unsupervised,” meaning that the reduction is derived solely from the data rather than reflecting any previous knowledge or classification scheme (37).

In contrast to pattern discovery, class prediction methods are techniques specifically designed to classify objects into known groups (36). Numerous reports describe machine-learning algorithms as computational techniques for classifying multidimensional data. Most methods include a training phase, run-on samples whose classes are already known (“training set”), and a testing phase in which the algorithm generalizes from the training data to predict classifications of previously unseen samples (“test or validation set”). Because direction is provided in the training phase, prediction methods are referred to as “supervised” classification methods. For microarray data derived from clinical studies, prediction generally refers to the classification of patients’ samples by characteristics such as disease subtype or response to treatment (38–40). These reports demonstrate convincingly that microarray data and class prediction methods provide a powerful approach to refine disease classifications and to predict outcome or treatment response. Once highly predictive classifier gene sets are identified, expression patterns of these genes must be verified by methods other than microarray. Typically, quantitative real-time PCR analysis is performed on the genes of interest in the same RNA samples that were subjected to microarray analysis. Finally, the performance of prediction and classification algorithms should be verified in new, unknown groups of patients/samples (41).