Advertisement
Research Article

A Molecular Signature of Proteinuria in Glomerulonephritis

  • Heather N. Reich mail,

    hreich@uhnresearch.on.ca

    Affiliations: Division of Nephrology, University Health Network, University of Toronto, Toronto, Ontario, Canada, The Toronto Glomerulonephritis Registry, Toronto, Ontario, Canada

    X
  • David Tritchler,

    Affiliations: Division of Epidemiology and Statistics, Ontario Cancer Institute, and University of Toronto, Toronto, Ontario, Canada, Department of Laboratory Medicine and Pathology, University Health Network, University of Toronto, Toronto, Ontario, Canada

    X
  • Daniel C. Cattran,

    Affiliations: Division of Nephrology, University Health Network, University of Toronto, Toronto, Ontario, Canada, The Toronto Glomerulonephritis Registry, Toronto, Ontario, Canada

    X
  • Andrew M. Herzenberg †,

    † Deceased.

    X
  • Felix Eichinger,

    Affiliation: Department of Internal Medicine, University of Michigan, Ann Arbor, Michigan, United States of America

    X
  • Anissa Boucherot,

    Affiliation: Department of Internal Medicine, University of Michigan, Ann Arbor, Michigan, United States of America

    X
  • Anna Henger,

    Affiliation: Department of Internal Medicine, University of Michigan, Ann Arbor, Michigan, United States of America

    X
  • Celine C. Berthier,

    Affiliation: Department of Internal Medicine, University of Michigan, Ann Arbor, Michigan, United States of America

    X
  • Viji Nair,

    Affiliation: Department of Internal Medicine, University of Michigan, Ann Arbor, Michigan, United States of America

    X
  • Clemens D. Cohen,

    Affiliation: Division of Nephrology and Institute of Physiology, University Hospital Zurich, Zurich, Switzerland

    X
  • James W. Scholey equal contributor,

    equal contributor Contributed equally to this work with: James W. Scholey, Matthias Kretzler

    Affiliations: Division of Nephrology, University Health Network, University of Toronto, Toronto, Ontario, Canada, The Toronto Glomerulonephritis Registry, Toronto, Ontario, Canada

    X
  • Matthias Kretzler equal contributor

    equal contributor Contributed equally to this work with: James W. Scholey, Matthias Kretzler

    Affiliation: Department of Internal Medicine, University of Michigan, Ann Arbor, Michigan, United States of America

    X
  • Published: October 18, 2010
  • DOI: 10.1371/journal.pone.0013451

Abstract

Proteinuria is the most important predictor of outcome in glomerulonephritis and experimental data suggest that the tubular cell response to proteinuria is an important determinant of progressive fibrosis in the kidney. However, it is unclear whether proteinuria is a marker of disease severity or has a direct effect on tubular cells in the kidneys of patients with glomerulonephritis. Accordingly we studied an in vitro model of proteinuria, and identified 231 “albumin-regulated genes” differentially expressed by primary human kidney tubular epithelial cells exposed to albumin. We translated these findings to human disease by studying mRNA levels of these genes in the tubulo-interstitial compartment of kidney biopsies from patients with IgA nephropathy using microarrays. Biopsies from patients with IgAN (n = 25) could be distinguished from those of control subjects (n = 6) based solely upon the expression of these 231 “albumin-regulated genes.” The expression of an 11-transcript subset related to the degree of proteinuria, and this 11-mRNA subset was also sufficient to distinguish biopsies of subjects with IgAN from control biopsies. We tested if these findings could be extrapolated to other proteinuric diseases beyond IgAN and found that all forms of primary glomerulonephritis (n = 33) can be distinguished from controls (n = 21) based solely on the expression levels of these 11 genes derived from our in vitro proteinuria model. Pathway analysis suggests common regulatory elements shared by these 11 transcripts. In conclusion, we have identified an albumin-regulated 11-gene signature shared between all forms of primary glomerulonephritis. Our findings support the hypothesis that albuminuria may directly promote injury in the tubulo-interstitial compartment of the kidney in patients with glomerulonephritis.

Introduction

Proteinuria is the clinical hallmark of glomerulonephritis, and the most important predictor of outcome in both diabetes-related and idiopathic glomerular-based kidney disease [1][8]. IgA nephropathy (IgAN) is the most common form of primary kidney disease world-wide [9], [10]; up to 40% of patients with IgAN progress to renal failure within 10 years of diagnosis [11]. Studies have consistently shown that proteinuria is the most powerful predictor of the rate of kidney function decline and kidney survival in IgAN [12][17], and that in patients with IgAN, this relationship is particularly strong even at low levels of proteinuria [18].

One of the pathologic features common to all forms of progressive glomerular-based kidney disease is tubulo-interstitial fibrosis, which shows consistent correlation with renal functional impairment [19][21]. Tubulo-interstitial fibrosis may be triggered by a variety of processes [22]; one proposed mechanism includes exposure of tubular cells to protein. Experimental evidence suggests that proteinuria is not only a marker of disease progression, but is directly involved in the pathogenesis of tubulo-interstitial fibrosis, and the progression of kidney injury [23], [24]. In patients with glomerular disease, proximal tubular epithelial cells are exposed to pathologically high concentrations of urinary proteins, including albumin. This induces a number of potentially injurious biologic responses in tubular epithelial cells, including inflammation, apoptosis, production of reactive oxygen species, and transition to a myofibroblast phenotype, ultimately contributing to tubulo-interstitial fibrosis [25][30]. These cellular responses may be dependent upon direct receptor-mediated uptake of albumin by tubular epithelial cells and subsequent stimulation of down-stream responses (such as NFκB-dependent gene transcription) or endocytosis-independent activation of signaling cascades by albumin [31]. While tubular cell exposure to protein is not the only proposed mechanism by which glomerular diseases result in tubulo-interstitial injury and progressive loss of renal function [22], [32], these observations may explain, in part, the important relationship between proteinuria, tubulo-interstitial injury, and long term outcome in glomerular-based kidney disease [19][21], [33].

Genome wide mRNA expression profiling tools, combined with robust statistical approaches, provide an unbiased approach to study the tubulo-interstitial transcriptional response initiated by proteinuria [34]. Using this strategy, we have demonstrated in an in vitro model of proteinuria that exposure of primary human renal proximal tubular epithelial cells to albumin induces the differential mRNA expression of a number of “albumin-regulated” genes, including interleukin-8 (IL-8) and the epidermal growth factor receptor (EGFR) [29]. Using this model system we demonstrated that albumin exposure in vitro results in the enhanced expression of IL-8 via activation of the mitogen-associated protein kinase ERK, an effect that was dependent upon transactivation of the EGF receptor and the generation of reactive oxygen species. While this in vitro model is a highly simplified representation of the in vivo disease process, in vitro findings using similar systems have been confirmed in studies of human kidney disease [30], [35].

To better understand the relationship between proteinuria and tubular epithelial cell responses, we studied the expression of “albumin-regulated genes”, defined in vitro, in the tubulo-interstitium of human kidney biopsies from patients with glomerulonephritis. First, primary human renal tubular epithelial cells were exposed to albumin in vitro, and differential gene expression was assessed using mRNA microarray analysis. A set of 231 differentially expressed “albumin-regulated” genes was derived, and the expression of these genes was then measured in the tubulo-interstitial compartment of kidney biopsy tissue from patients with primary glomerulonephritis and healthy live kidney donors. We first studied the expression of these transcripts in IgAN biopsies, given that this is the most common type of primary glomerulonephritis, and given the particularly close relationship between proteinuria and kidney function in this disease [18]. We then studied mRNA expression in the tubulo-interstitial compartment of patients with other forms of idiopathic glomerulonephritis to determine if there are shared mechanisms of tubulo-interstitial injury.

Results

Differential gene expression in an in vitro model of proteinuria

To determine the effect of albumin on gene expression in cultured primary human renal tubular epithelial cells, mRNA expression was measured using data from 8 microarrays (4 with control conditions representing 8 experimental in vitro replicates, and 4 BSA-treated conditions, representing 8 experimental replicates). Using conservative thresholds for differential gene expression, we identified 231 transcripts differentially expressed in cells treated with 1% bovine serum albumin (BSA) versus control conditions for 6 hours. A selection of the mRNA transcripts found to be differentially expressed in this model, using stringent statistical selection criteria, is provided in Table 1 (full list Supplementary Data S1). Albumin-dependent mRNA regulation was seen in genes involved in apoptosis, cell growth and metabolism, cell-signaling, lipid metabolism, matrix turnover, cell cycle, cell movement, lipid metabolism, and reactive oxygen species scavenging.

thumbnail

Table 1. Genes differentially expressed in cells exposed to albumin.

doi:10.1371/journal.pone.0013451.t001

Expression of “albumin-regulated genes” in human kidney biopsy sample

Affymetrix microarray mRNA expression data were generated from the tubulo-interstitial compartment of biopsies from 25 patients with IgAN and proteinuria and 6 control subjects enrolled in the European Renal cDNA Bank Bank - Kröner-Fresenius biopsy bank (ERCB-KFB). The clinical characteristics of subjects are shown in Table 2; subjects with IgAN had a wide range in proteinuria (trace to 10g/day) and renal function (normal to stage 4/5 CKD). Expression data for 231 unique “albumin-regulated” genes (derived above) were extracted from the human renal biopsy microarray data.

thumbnail

Table 2. Patient characteristics – subjects with IgA nephropathy.

doi:10.1371/journal.pone.0013451.t002

To determine if the expression levels of these 231 genes were associated with the kidney disease process, hierarchical cluster analysis of the expression data was performed (Figure 1). Cluster analysis distinguished between renal biopsies from control subjects and the renal biopsies from patients with IgAN based upon the expression of the 231 “albumin-regulated genes”. The cluster analysis findings were not reproduced using the expression data of randomly selected gene sets of similar size. Of the 231 genes, 49 (21%) were differentially expressed in the IgAN samples compared to the healthy control samples (FDR 5%, see Supplementary Data S1) compared to 4% of the genes in the complete Affymetrix data (χ2 = 147, p<0.01), confirming the enrichment of regulated genes in the in vitro defined gene set.

thumbnail

Figure 1. Cluster analysis.

Clustering of microarrays according to disease status (horizontal axis) based upon expression of the 231 albumin-regulated genes (vertical axis). Control = control group, IgAN = IgA nephropathy.

doi:10.1371/journal.pone.0013451.g001

A gene expression signature of proteinuria in IgAN

Given the critical relationship between proteinuria and outcome in IgAN, even at low levels of proteinuria [18], we specifically examined genes differentially expressed in IgAN biopsies compared with control biopsies. The 231 gene set was derived from an in vitro model of proteinuria designed to be representative of the biologic response of kidney tubular cells in to proteinuria, so we studied the relationship between the expression levels of the “albumin-regulated” genes in the tubulo-interstitial compartment from the renal biopsies of the patients with IgAN with the level of proteinuria at the time of kidney biopsy, and selected the mRNAs with expression levels significantly correlated with proteinuria.

We then utilized the following criteria to select a list of genes for the validation studies: 1) genes differentially expressed by human renal tubular cells after exposure to albumin, in vitro; 2) genes differentially expressed in the tubulo-interstitial compartment of patients with IgAN compared to control subjects (this compartment includes tubular epithelial cells, interstitial tissue and cells such as fibroblasts, and endothelial cells in vascular structures) based upon the biopsy Affymetrix microarray data set; 3) genes with expression levels related to level of proteinuria within IgAN. This selection process yielded a set of 11 genes (see Table 3).

thumbnail

Table 3. Validation gene set.

doi:10.1371/journal.pone.0013451.t003

Albumin-regulated genes in primary GN

In order to determine if the relationship between the expression of the albumin-regulated genes and the kidney disease process is unique to IgAN, the Affymetrix microarray mRNA expression data for these genes were derived from the tubulo-interstitial compartment of biopsies from patients with primary focal segmental glomerulosclerosis (FSGS, n = 10), membranous GN (MGN, n = 18) and minimal change disease (MCD, n = 5) and proteinuria and 21 control subjects (healthy kidney donors with normal renal biopsies) enrolled in the ERCB-KFB. The clinical characteristics of subjects are shown in Table 4. As reference samples healthy living kidney donors known to have normal renal function and no proteinuria were used.

thumbnail

Table 4. Patient characteristics – European Renal cDNA Bank glomerular disease subjects.

doi:10.1371/journal.pone.0013451.t004

We found that the expression of these 11 mRNAs differed in the tubulo-interstitial compartment of all biopsies of subjects with primary glomerular disease compared to controls. Hierarchical cluster analysis confirmed that renal biopsies from healthy donors can be distinguished from the renal biopsies of patients with all forms of primary GN based solely upon the expression of the 11 mRNA gene signature (Figure 2). Cluster analysis using expression data of randomly selected gene sets of similar size did not distinguish kidney biopsies of control subjects from the biopsies of subjects with GN.

thumbnail

Figure 2. Cluster analysis.

Clustering of microarrays according to disease status (columns) based upon expression of the 11 albumin-regulated genes (rows). Control = control group, FSGS = focal segmental glomerulosclerosis, MGN = membranous GN, MCD = minimal change disease.

doi:10.1371/journal.pone.0013451.g002

Potential common regulatory pathways

To define the functional context of the proteinuria associated genes, a transcriptional network was constructed using a co-citation natural language processing (NLP) tools (Genomatix Bibliosphere), considering all of the transcripts of the 11-gene mRNA signature (Supplementary Data S1). EGR1 emerged as a central node linking EGR1 to the remaining 10 mRNA transcripts. To determine if common transcription factor promoter elements could explain the functional relationship between these 11 albumin-regulated mRNA transcripts, promoter regions (500bp up and 100bp downstream of the transcription start sites) was performed (Genomatix Bibliosphere). In concordance with the central role of the EGR node in the NLP analysis, most of the genes encoding the 11-gene mRNA signature contain a proximal EGR1 promoter regions, consistent with a putative common transcriptional regulation (Figure 3).

thumbnail

Figure 3. Transcription factor analysis.

Green shade indicates the presence of a transcription factor binding site sequence, upstream from the gene of interest. Each row represents a gene of interest, and each column represents a putative transcription factor.

doi:10.1371/journal.pone.0013451.g003

Discussion

Proteinuria is an important determinant of outcome in primary GN but the mechanisms responsible for this association have not been fully elucidated. Although in vitro and experimental studies suggest that proteinuria, and in particular albumin, elicits a biological response in kidney tubule epithelial cells that contributes to progressive tubulointerstitial injury [35], it is uncertain if proteinuria has a direct effect on gene expression in human kidney disease. Development of new unbiased molecular and statistical tools for studying mRNA expression in renal tissue has greatly advanced our ability to study renal disease, and to translate findings from basic molecular and cell biology research to human disease [34], [36][44]. Accordingly, the aim of this study was to test the hypothesis that there is a steady-state change in gene expression in the renal tubulointerstitium of subjects with primary GN that reflects a biological response of the tubule cells to proteinuria.

In order to address this hypothesis we exposed primary human renal proximal tubular epithelial cells to albumin, in vitro, simulating the exposure of tubular cells to proteinuria in human proteinuric glomerulonephritis. Our first major finding was the identification of a distinct set of 231 mRNAs differentially regulated in human renal tubular cells by albumin exposure in vitro (Supplementary Data S1). Gene ontology (GO) analysis identified several pathways that were statistically over-represented in the in vitro expression data, and the proteins encoded by these genes are involved in diverse biological processes including apoptosis, cell cycling, pro-inflammatory cell signaling cytokines, connective tissue development and fibrosis, and free radical scavenging, and lipid metabolism (Table 1).

Although the genes regulated by albumin in vitro can be related to injury pathways and outcomes like cell loss, inflammation, and fibrosis, we sought to translate these findings to human kidney disease by comparing the expression levels of these genes in the kidneys of normal subjects and subject with primary GN. mRNA levels were measured by microarray analysis in the microdissected tubulointerstitial compartment of renal biopsy samples in order to capture the kidney tubular cell transcriptome in vivo. We chose to focus first on subjects with IgAN because new evidence has shown that incremental increases in proteinuria are associated with dramatic reductions in renal survival in IgAN [18], and that in patients with IgAN, these changes occur at far lower levels of proteinuria compared to patients with other forms of primary GN [45]. Our second major finding was that we could distinguish between the kidney biopsies of subjects with IgAN and healthy living donors based solely upon the tubulo-interstitial expression levels of the 231 “albumin-regulated genes” using hierarchical cluster analysis.

In order to determine if this clustering phenomenon was a chance event, we tested random sets of 231 genes generated from the full Affymetrix expression dataset but cluster analysis of these random gene sets failed to segregate IgAN from control biopsies. In addition, we found significantly enriched differential expression of the 231 “albumin-regulated genes” in the tubulo-interstitial tissue of IgAN biopsies compared to control biopsies. Taken together, these analyses support the hypothesis that the biological response to albumin that we observed in vitro may also be present, at least in part, in the kidneys of subjects with IgAN and proteinuria.

In order to further explore the link between proteinuria and gene expression in vivo, we then studied the relationship between the expression levels of the 231 genes and the levels of proteinuria at the time of biopsy within the group of subjects with IgAN. The rationale for this analysis was twofold: first, proteinuria is known to be the most powerful predictor of outcome in glomerular-based diseases, including IgAN, and, second, the tubular response to proteinuria (modeled by our in vitro experiment) may be a key factor determining progressive tubulo-interstitial fibrosis and nephron loss, important pathologic indicators of prognosis in IgAN [11], [18]. Our third major finding was that the tubulo-interstitial expression levels of 11 mRNA transcripts (of the 231 albumin-regulated genes) correlated significantly with the level of proteinuria at the time of biopsy, suggesting that there may be a biological relationship between gene expression in the tubulointerstitial compartment of the kidney and proteinuria. We labeled this set of 11 genes the “proteinuria signature”.

The 11 genes in the “proteinuria signature” encode some proteins previously implicated in the tubular response to proteinuria, as well as some proteins that have not been studied in the context of kidney disease, and thus may be potential new mediators of progressive kidney injury. For example, protein members of the coagulation cascade have been implicated in extracellular matrix protein turnover in the kidney, and we found that Serpine1 (PAI-1) was a transcipt identified in the proteinuria signature. Serpine1 is not normally produced in the kidneys, however experimental evidence suggests that it is an important promoter of renal fibrosis [46]. The mechanisms by which it promotes fibrosis are not entirely elucidated, however in addition to inhibiting protease activity in the extracellular compartment, it may modulate inflammatory cell recruitment, and fibroblast activation [47]. Work in experimental models of kidney disease including protein overload injury, proliferative glomerulonephritis, and obstructive kidney disease suggests that increased PAI-1 expression is associated with interstitial fibrosis, and reduction in PAI-1 expression (via drug therapy or recombinant techniques) is associated with attenuation of renal fibrotic injury [46][50]. Furthermore, de novo PAI-1 protein expression is documented in kidney biopsies of patients with glomerular based kidney disease [51][53]. Our data support the hypothesis that proteinuria may contribute to fibrosis by increasing PAI-1 expression by tubular cells. In addition, the early growth response gene 1 (EGR-1) has been implicated in the TGFβ-mediated fibrosis [54] and in the regulation of interstitial fibrosis in the experimental unilateral ureteric obstruction model [55]. Finally, our discovery that the gene for collagen I, alpha 1, was also in the “proteinuria signature” also supports a link between proteinuria, gene expression in the tubulointerstitium, and fibrosis.

In addition to interstitial fibrosis, apoptosis of tubular cells contributes to progressive kidney injury in primary GN by promoting cell loss [56], [57]. In this regard, EGR-1 has also been reported to regulate apoptosis in response to oxidative stress [58]. Two other apoptosis-related genes, immediate early response 3 (IER3) and myeloid cell leukemia sequence 1 (MCL1), were also in the 11 gene “proteinuria signature”. The MCL1 gene encodes a member of the bcl2 family that may promote or inhibit apoptosis depending on the tissue [59][61]. For example, MCL1 has been found to support cell survival in Wilms' Tumour cell lines [62]. Neither MCL1 nor IER3 have been studied in the context of primary GN, and further studies will be necessary to better define their role in kidney disease. Similarly, the role of the proteins encoded by genes MAFF, TYMS, and SAMD4A in the progression of GN is unknown.

The next goal of this study was to determine if these 11 genes represented a generic response to proteinuria that was present in other forms of primary GN because our in vitro experiments identified a response of tubular cells to albumin that was independent of glomerular injury. In order to address this question, we studied the expression levels of the 11 genes in kidney biopsy samples from subjects with three other common forms of primary GN: focal segmental glomerulosclerosis (FSGS), membranous nephropathy (MGN), and minimal change disease (MCD). The biopsies of subjects with FSGS, MGN, and MCD could be distinguished from control biopsies based solely upon the expression of the 11-gene signature, supporting the conclusion that this 11-gene set is part of a common pathway linking proteinuria to gene expression in the kidney.

Finally, we subjected the 11 genes to a bioinformatics analysis in order to explore relationships between the component genes in the “proteinuria signature”. We first constructed a transcriptional network (Supplemental Figure S1 in Supplementary Data S1) utilizing Genomatix Bibliosphere. The network derived from this analysis placed EGR1 in the central node. Based on this finding, we then went on to perform a transcription factor analysis utilizing gene promoter sequence data for the 11 genes, and we found that the consensus sequence for EGR1 was present in 6 of the genes in the “proteinuria signature”. This analysis suggests that EGR1 may play a key role in a common pathway orchestrating the transcriptional response kidney tubule cells to proteinuria in vivo. The transcription factor analysis further suggests that the transcription factor ELF3 may also mediate this response, at least in part.

There are some important limitations in the current study. First, stringent and conservative statistical thresholds were used in the initial selection of differentially expressed genes in vitro at a single time point of 6-hours of albumin exposure. This design may have precluded identification of other important genes involved in the tubular cell response to albumin exposure (Type II Error). In addition, we used the in vitro data set for a targeted analysis of the human subjects with GN, capturing a different gene set than that identified by Rudnicki et al. [63] in studies of laser-microdissected proximal tubular cells in GN.

Finally, while we derived our list of candidate genes from an in vitro model of proteinuria, it is likely that the transcriptional response of the genes measured in the tubulo-interstitium of the kidney biopsies is not due entirely to exposure to albumin the ultrafiltrate. We may also be capturing the tubular response to other factors that play a role in progression. For example, it is possible that the transcriptional response is related to exposure of tubular cells to filtered growth factors [64], other proteins in the ultrafiltrate, or modified albumin moieties [42], [65][69]}. In this regard, Kritz and coworkers have suggested that albumin uptake by tubular cells is not an absolute prerequisite for tubulo-intertitial injury in mice with glomerulonephritis [32]. Other contributors to nephron loss in glomerular-based kidney disease include misdirected filtrate and obstruction of the tubulo-glomerular junction with subsequent tubular cell injury and ischemia [22], [32]. The importance of multifactorial contributions to tubulo-interstitial fibrosis beyond tubular cell albumin exposure is highlighted by the clinical observation that in IgAN, tubulo-interstitial injury and functional decline occurs at far lower levels of proteinuria [45]. Of interest, our mRNA signature was able to discriminate minimal change nephropathy biopsies from control samples, despite the fact that this entity is only more rarely associated with tubulo-interstitial fibrosis and progressive functional loss [70]. However, when proteinuria reduction is not achieved in patients with this disease, sclerosis, fibrosis, and functional loss does occur [70].

It is also possible that the tissue mRNA signature is not entirely derived from tubular cells. There are little morphometric data that quantify the relative abundance of cells that comprise the cortical interstitium in the disease state. In “normal” control biopsies, 87% of the tubulo-interstitial compartment is comprised of the tubular cell component; in biopsies of patients with type 1 diabetes with early nephropathy, this reaches 91% [71]. The non-tubular cell component comprises only 13% of the healthy cortex or less [72]. Of the non-tubular cell portion, only 14% (ie. <2% total) is composed of peritubular capillaries, in biopsies with moderate interstitial expansion due to diabetic nephropathy [71], [72]. Taken together, these data suggest that the predominant resident cell transcriptome signal is derived from tubular cell mRNA expression. Emerging evidence also suggests that number of dendritic cells in the cortical tubulo-interstitium of kidney biopsies from patients with glomerulonephritis is increased in comparison to control biopsies [73], and appear to contribute to progressive kidney disease in animal models [74]. Interstitial macrophage infiltration occurs in many forms of primary glomerulonephritis; while data suggest that cellular infiltration correlates with renal function at the time of biopsy, the relationship to proteinuria is not as clear (reviewed in [75]). We cannot discount the possibility that these cells are also contributing to the mRNA expression profile.

In summary, we have used an in vitro model of proteinuria to identify a set of “albumin-regulated genes” in primary human renal tubular cells. We have translated these findings to human primary GN, and identified a subset of mRNA transcripts with expression levels that correlate with the level of proteinuria, and that distinguish biopsies of subjects with GN from biopsies of control subjects. Further studies will be necessary to define the biological role of these genes in proteinuric kidney disease and to determine if measures of expression of these genes are predictive of long-term clinical outcome.

Methods

Primary cell culture system

The cell system used was previously described [29]. Eight flasks of primary human renal tubular epithelial cells (Cambrex, Walkersville, MD) were exposed to medium alone and 8 flasks of cells were exposed to medium containing 1% bovine serum albumin for 6 h. The RNA extracted (Qiagen RNeasy kit, Valencia, CA); RNA from cells grown in two flasks was pooled to form one experimental sample, and each experiment was performed in quadruplicate (total 16 flasks for eight microarrays—four arrays from control cell RNA and four arrays from albumin-treated cell RNA). RNA quality was verified using the Agilent bioanalyzer (Agilent Technologies, Palo Alto, CA).

Synthesis of cDNA and array hybridization, washing, and scanning were performed by the Affymetrix Gene Chip core facility at The Centre for Applied Genomics at Toronto's Hospital for Sick Children (Ontario, Canada) according to Affymetrix-recommended protocols (Santa Clara, CA) using the hgu 133A Affymetrix Gene Chip and an Affymetrix Fluidics station.

RNA extraction and mRNA expression profiling of human renal biopsy tissue

The study was performed as outlined previously in detail [36], [40]. Human renal biopsy specimens were procured in an international multicenter study, the European Renal cDNA Bank- Kröner-Fresenius biopsy bank (ERCB-KFB, see acknowledgements for participating centers) [36]. Renal biopsies were obtained after written consent and approval of the ethics committee and in the frame of the ERCB-KFB approved by the specialized subcommittee for internal medicine of the cantonal ethics committee of Zurich. The characteristics of patients are shown in Table 2. Control biopsy samples were obtained during the cold ischemia period of living-related donor transplantation.

Total RNA was extracted from manually microdissected tubulo-interstitial compartments obtained from living donors (n = 6) and patients with IgA nephropathy (n = 25). After one round of amplification of 300–800 ng of total RNA, RNA quality and quantity was verified (Agilent Technologies, Waldbronn, Germany). The fragmentation, hybridization, staining and imaging was performed according the manufacturer's guidelines (Affymetrix). For a detailed description of the protocol see reference [44]. All microarray data are MIAME compliant as detailed on the MGED Society website http://www.mged.org/Workgroups/MIAME/mia​me.html. The raw data will be GEO accessible through GEO Series accession.

II. Data filtering strategy to determine renal response to proteinuria

In order to rationally filter the large volume of data derived from the microarray experiments, the following strategy was employed to select the genes that are characteristic of the renal response to proteinuria:

  • 1 – Identification of genes differentially expressed in the in vitro model of proteinuria by SAM and Limma analysis (described below).
  • 2 – Identification of genes differentially expressed in the mRNA expression profiling data from tubulo-interstitial tissue of patients with IgA nephropathy vs. control samples.
  • 3 – Identification of genes correlating with and predictive of proteinuria in vivo by linear models using Limma and lasso regression procedure, respectively.

Statistical tools employed for data analysis

The microarray data obtained from the in vitro model were examined and visualized using Affymetrix Microarray Suite 5.0 software and Bioconductor [76], [77]. The calculation of expression values from probe intensities and normalization of arrays was performed using the RMA method [78] using Bioconductor and RMAexpress [79] (accessed 2006). Differential gene expression was determined using Limma (Linear models for microarray data) and SAM (Significance Analysis of Micoarrays) through Bioconductor [80], [81], with a highly conservative false-discovery rate set at 0.01, and genes were not filtered based upon an arbitrarily-selected fold-change in expression. Differential expression was assessed in the in vivo tubulo-interstitial samples using SAM and dChip [82].

Cluster analysis was performed using Sammon mapping/multi-dimensional scaling, as well as spectral clustering [83] for experimental cell data, and hierarchical cluster analysis was performed using dChip [82], [84] in the renal biopsy dataset (centroid-based, distance metric: 1-correlation).

In order to explore the ontology of genes differentially expressed in vitro, genes were ranked by limma topTable function (by adjusted p-value), and 600 up and down-regulated genes were selected to study possible common ontology patterns. Enriched expression of gene ontology (GO) terms was assessed with Ingenuity Pathway Analysis Software 4.2 (Redwood City, CA) and confirmed using the Bioconductor package GOstats. These programs determine which gene ontology terms found in gene lists are statistically over or under represented, compared with the GO terms represented in the microarray as a whole [85], [86]. A list of enriched GO terms is produced, including the test statistic and associated p value, suggesting functional mechanisms that may underlie the biological response captured in the data set.

Clinical data were extracted for the patients who underwent renal biopsy and inspection revealed that proteinuria values and residuals were skewed, and should be normalized by log transformation for regression analysis. To select transcripts with mRNA expression most closely related to proteinuria in IgAN, mRNA expression was correlated with proteinuria in vivo using advanced regression analysis with linear models (with limma and topTable function in Bioconductor) [81]. Partitioning methods were also employed to use the biopsy gene expression data to predict proteinuria. Lasso regression procedure was also used to confirm genes that were most predictive of log proteinuria tuned by a 10-fold cross-validation procedure [87].

Once this filtration strategy was applied, and the 11-mRNA signature identified, the normalized mRNA expression data were then extracted from the full datasets from MGN, FSGS, and MCD biopsies. Hierarchical cluster analysis was performed on the human renal biopsy data set using dChip [82] (centroid-based, distance metric: 1-correlation). Tests of the correlation between proteinuria and mRNA expression were performed by relating the normalized mRNA expression values to proteinuria using Pearson correlation.

Supporting Information

Dataset S1.

A molecular signature of proteinuria in glomerulonephritis.

doi:10.1371/journal.pone.0013451.s001

(0.86 MB DOC)

Acknowledgments

Array hybridization was performed with assistance from Bodo Brunner and Almut Nitsche, Sanofi-Aventis Deutschland GmbH.

The members of the ERCB-KFB at the time of the study included: C. D. Cohen, M. Fischereder, H. Schmid, M. Kretzler, D. Schlöndorff, Munich; J. D. Sraer, P. Ronco, Paris; M. P. Rastaldi, G. D'Amico, Milano; F. Mampaso, Madrid; P. Doran, H. R. Brady, Dublin; D. Mönks, C. Wanner, Würzburg; A. J. Rees, Aberdeen; P. Brown, Aberdeen; F. Strutz, G. Müller, Göttingen; P. Mertens, J. Floege, Aachen; N. Braun, T. Risler, Tübingen; L. Gesualdo, F. P. Schena, Bari; J. Gerth, G. Wolf, Jena; R. Oberbauer, D. Kerjaschki, Vienna; B. Banas, B. K. Krämer, Regensburg; W. Samtleben, Munich; H. Peters, H. H. Neumayer, Berlin; K Ivens, B. Grabensee, Düsseldorf; R. P. Wüthrich, Zürich; V. Tesar, Prague.

Author Contributions

Conceived and designed the experiments: HNR DT AMH FE AH CB CDC JWS MK. Performed the experiments: HNR AB AH CDC. Analyzed the data: HNR DT DC FE AB AH CB VN JWS MK. Contributed reagents/materials/analysis tools: DC AMH CDC JWS MK. Wrote the paper: HNR JWS MK. Reviewed the manuscript: AMH.

References

  1. 1. Mogensen CE (1976) Progression of nephropathy in long-term diabetics with proteinuria and effect of initial anti-hypertensive treatment. Scand J Clin Lab Invest 36: 383–388.
  2. 2. Row PG, Cameron JS, Turner DR, Evans DJ, White RH, et al. (1975) Membranous nephropathy long-term follow-up and association with neoplasia. Ateneo Parmense Acta Biomed 46: 433–454.
  3. 3. Cameron JS (1990) Proteinuria and progression in human glomerular diseases. Am J Nephrol 10: Suppl 181–87.
  4. 4. Gall MA, Nielsen FS, Smidt UM, Parving HH (1993) The course of kidney function in type 2 (non-insulin-dependent) diabetic patients with diabetic nephropathy. Diabetologia 36: 1071–1078.
  5. 5. Lewis EJ, Hunsicker LG, Bain RP, Rohde RD (1993) The effect of angiotensin-converting-enzyme inhibition on diabetic nephropathy. The Collaborative Study Group. N Engl J Med 329: 1456–1462.
  6. 6. Myers BD, Nelson RG, Tan M, Beck GJ, Bennett PH, et al. (1995) Progression of overt nephropathy in non-insulin-dependent diabetes. Kidney Int 47: 1781–1789.
  7. 7. Christensen PK, Rossing P, Nielsen FS, Parving HH (1999) Natural course of kidney function in Type 2 diabetic patients with diabetic nephropathy. Diabet Med 16: 388–394.
  8. 8. Philibert D, Cattran D (2008) Remission of proteinuria in primary glomerulonephritis: we know the goal but do we know the price? Nat Clin Pract Nephrol 4: 550–559.
  9. 9. D'Amico G (1987) The commonest glomerulonephritis in the world: IgA nephropathy. Q J Med 64: 709–727.
  10. 10. D'Amico G (2004) Natural history of idiopathic IgA nephropathy and factors predictive of disease outcome. Seminars in Nephrology 24: 179–196.
  11. 11. Bartosik LP, Lajoie G, Sugar L, Cattran DC (2001) Predicting progression in IgA nephropathy. Am J Kidney Dis 38: 728–735.
  12. 12. Kobayashi Y, Tateno S, Hiki Y, Shigematsu H (1983) IgA nephropathy: prognostic significance of proteinuria and histological alterations. Nephron 34: 146–153.
  13. 13. Ibels LS, Gyory AZ (1994) IgA nephropathy: analysis of the natural history, important factors in the progression of renal disease, and a review of the literature. Medicine (Baltimore) 73: 79–102.
  14. 14. Radford MG Jr, Donadio JV Jr, Bergstralh EJ, Grande JP (1997) Predicting renal outcome in IgA nephropathy. J Am Soc Nephrol 8: 199–207.
  15. 15. Ruggenenti P, Perna A, Mosconi L, Pisoni R, Remuzzi G (1998) Urinary protein excretion rate is the best independent predictor of ESRF in non-diabetic proteinuric chronic nephropathies. “Gruppo Italiano di Studi Epidemiologici in Nefrologia” (GISEN). Kidney Int 53: 1209–1216.
  16. 16. Li PK, Ho KK, Szeto CC, Yu L, Lai FM (2002) Prognostic indicators of IgA nephropathy in the Chinese–clinical and pathological perspectives. Nephrol Dial Transplant 17: 64–69.
  17. 17. Donadio JV, Bergstralh EJ, Grande JP, Rademcher DM (2002) Proteinuria patterns and their association with subsequent end-stage renal disease in IgA nephropathy. Nephrol Dial Transplant 17: 1197–1203.
  18. 18. Reich HN, Troyanov S, Scholey JW, Cattran DC (2007) Remission of proteinuria improves prognosis in IgA nephropathy. J Am Soc Nephrol 18: 3177–3183.
  19. 19. Bohle A, Mackensen-Haen S, von Gise H (1987) Significance of tubulointerstitial changes in the renal cortex for the excretory function and concentration ability of the kidney: a morphometric contribution. Am J Nephrol 7: 421–433.
  20. 20. Nath KA (1992) Tubulointerstitial changes as a major determinant in the progression of renal damage. Am J Kidney Dis 20: 1–17.
  21. 21. Gilbert RE, Cooper ME (1999) The tubulointerstitium in progressive diabetic kidney disease: more than an aftermath of glomerular injury? Kidney Int 56: 1627–1637.
  22. 22. Kriz W, Hartmann I, Hosser H, Hahnel B, Kranzlin B, et al. (2001) Tracer studies in the rat demonstrate misdirected filtration and peritubular filtrate spreading in nephrons with segmental glomerulosclerosis. J Am Soc Nephrol 12: 496–506.
  23. 23. Remuzzi G, Bertani T (1998) Pathophysiology of progressive nephropathies. N Engl J Med 339: 1448–1456.
  24. 24. Eddy AA (2004) Proteinuria and interstitial injury. Nephrol Dial Transplant 19: 277–281.
  25. 25. Abbate M, Zoja C, Rottoli D, Corna D, Tomasoni S, et al. (2002) Proximal tubular cells promote fibrogenesis by TGF-beta1-mediated induction of peritubular myofibroblasts. Kidney Int 61: 2066–2077.
  26. 26. Donadelli R, Abbate M, Zanchi C, Corna D, Tomasoni S, et al. (2000) Protein traffic activates NF-kB gene signaling and promotes MCP-1-dependent interstitial inflammation. Am J Kidney Dis 36: 1226–1241.
  27. 27. Zoja C, Morigi M, Figliuzzi M, Bruzzi I, Oldroyd S, et al. (1995) Proximal tubular cell synthesis and secretion of endothelin-1 on challenge with albumin and other proteins. Am J Kidney Dis 26: 934–941.
  28. 28. Zoja C, Donadelli R, Colleoni S, Figliuzzi M, Bonazzola S, et al. (1998) Protein overload stimulates RANTES production by proximal tubular cells depending on NF-kappa B activation. Kidney Int 53: 1608–1615.
  29. 29. Reich H, Tritchler D, Herzenberg AM, Kassiri Z, Zhou X, et al. (2005) Albumin activates ERK via EGF receptor in human renal epithelial cells. J Am Soc Nephrol 16: 1266–1278.
  30. 30. Tang S, Leung JC, Abe K, Chan KW, Chan LY, et al. (2003) Albumin stimulates interleukin-8 expression in proximal tubular epithelial cells in vitro and in vivo. J Clin Invest 111: 515–527.
  31. 31. Diwakar R, Pearson AL, Colville-Nash P, Brunskill NJ, Dockrell ME (2007) The role played by endocytosis in albumin-induced secretion of TGF-beta1 by proximal tubular epithelial cells. Am J Physiol Renal Physiol 292: F1464–F1470.
  32. 32. Theilig F, Kriz W, Jerichow T, Schrade P, Hahnel B, et al. (2007) Abrogation of Protein Uptake through Megalin-Deficient Proximal Tubules Does Not Safeguard against Tubulointerstitial Injury. J Am Soc Nephrol 18: 1824–1834.
  33. 33. Meyer TW (2003) Tubular injury in glomerular disease. Kidney Int 63: 774–787.
  34. 34. Ju W, Eichinger F, Bitzer M, Oh J, McWeeney S, et al. (2009) Renal gene and protein expression signatures for prediction of kidney disease progression. Am J Pathol 174: 2073–2085.
  35. 35. Zoja C, Benigni A, Remuzzi G (2004) Cellular responses to protein overload: key event in renal disease progression. Curr Opin Nephrol Hypertens 13: 31–37.
  36. 36. Cohen CD, Frach K, Schlondorff D, Kretzler M (2002) Quantitative gene expression analysis in renal biopsies: a novel protocol for a high-throughput multicenter application. Kidney Int 61: 133–140.
  37. 37. Kretzler M, Cohen CD, Doran P, Henger A, Madden S, et al. (2002) Repuncturing the renal biopsy: strategies for molecular diagnosis in nephrology. J Am Soc Nephrol 13: 1961–1972.
  38. 38. Striker LJ, Striker GE (2003) Windows on renal biopsy interpretation: does mRNA analysis represent a new gold standard? J Am Soc Nephrol 14: 1096–1098.
  39. 39. Waga I, Yamamoto J, Sasai H, Munger WE, Hogan SL, et al. (2003) Altered mRNA expression in renal biopsy tissue from patients with IgA nephropathy. Kidney Int 64: 1253–1264.
  40. 40. Schmid H, Cohen CD, Henger A, Schlondorff D, Kretzler M (2004) Gene expression analysis in renal biopsies. Nephrol Dial Transplant 19: 1347–1351.
  41. 41. Peterson KS, Huang JF, Zhu J, D'Agati V, Liu X, et al. (2004) Characterization of heterogeneity in the molecular pathogenesis of lupus nephritis from transcriptional profiles of laser-captured glomeruli. J Clin Invest 113: 1722–1733.
  42. 42. Porubsky S, Schmid H, Bonrouhi M, Kretzler M, Malle E, et al. (2004) Influence of native and hypochlorite-modified low-density lipoprotein on gene expression in human proximal tubular epithelium. Am J Pathol 164: 2175–2187.
  43. 43. Henger A, Kretzler M, Doran P, Bonrouhi M, Schmid H, et al. (2004) Gene expression fingerprints in human tubulointerstitial inflammation and fibrosis as prognostic markers of disease progression. Kidney Int 65: 904–917.
  44. 44. Schmid H, Boucherot A, Yasuda Y, Henger A, Brunner B, et al. (2006) Modular activation of nuclear factor-kappaB transcriptional programs in human diabetic nephropathy. Diabetes 55: 2993–3003.
  45. 45. Cattran DC, Reich HN, Beanlands HJ, Miller JA, Scholey JW, et al. (2008) The impact of sex in primary glomerulonephritis. Nephrol Dial Transplant 23: 2247–2253.
  46. 46. Eddy AA, Giachelli CM (1995) Renal expression of genes that promote interstitial inflammation and fibrosis in rats with protein-overload proteinuria. Kidney Int 47: 1546–1557.
  47. 47. Eddy AA, Fogo AB (2006) Plasminogen activator inhibitor-1 in chronic kidney disease: evidence and mechanisms of action. J Am Soc Nephrol 17: 2999–3012.
  48. 48. Brown NJ, Nakamura S, Ma L, Nakamura I, Donnert E, et al. (2000) Aldosterone modulates plasminogen activator inhibitor-1 and glomerulosclerosis in vivo. Kidney Int 58: 1219–1227.
  49. 49. Oda T, Jung YO, Kim HS, Cai X, Lopez-Guisa JM, et al. (2001) PAI-1 deficiency attenuates the fibrogenic response to ureteral obstruction. Kidney Int 60: 587–596.
  50. 50. Zhang G, Kim H, Cai X, Lopez-Guisa JM, Alpers CE, et al. (2003) Urokinase receptor deficiency accelerates renal fibrosis in obstructive nephropathy. J Am Soc Nephrol 14: 1254–1271.
  51. 51. Yamamoto T, Noble NA, Cohen AH, Nast CC, Hishida A, et al. (1996) Expression of transforming growth factor-beta isoforms in human glomerular diseases. Kidney Int 49: 461–469.
  52. 52. Grandaliano G, Gesualdo L, Ranieri E, Monno R, Schena FP (2000) Tissue factor, plasminogen activator inhibitor-1, and thrombin receptor expression in human crescentic glomerulonephritis. Am J Kidney Dis 35: 726–738.
  53. 53. Hamano K, Iwano M, Akai Y, Sato H, Kubo A, et al. (2002) Expression of glomerular plasminogen activator inhibitor type 1 in glomerulonephritis. Am J Kidney Dis 39: 695–705.
  54. 54. Chen SJ, Ning H, Ishida W, Sodin-Semrl S, Takagawa S, et al. (2006) The early-immediate gene EGR-1 is induced by transforming growth factor-beta and mediates stimulation of collagen gene expression. J Biol Chem 281: 21183–21197.
  55. 55. Nakamura H, Isaka Y, Tsujie M, Rupprecht HD, Akagi Y, et al. (2002) Introduction of DNA enzyme for Egr-1 into tubulointerstitial fibroblasts by electroporation reduced interstitial alpha-smooth muscle actin expression and fibrosis in unilateral ureteral obstruction (UUO) rats. Gene Ther 9: 495–502.
  56. 56. Erkan E, De Leon M, Devarajan P (2001) Albumin overload induces apoptosis in LLC-PK(1) cells. Am J Physiol Renal Physiol 280: F1107–F1114.
  57. 57. Erkan E, Garcia CD, Patterson LT, Mishra J, Mitsnefes MM, et al. (2005) Induction of renal tubular cell apoptosis in focal segmental glomerulosclerosis: roles of proteinuria and Fas-dependent pathways. J Am Soc Nephrol 16: 398–407.
  58. 58. Stuart JR, Kawai H, Tsai KK, Chuang EY, Yuan ZM (2005) c-Abl regulates early growth response protein (EGR1) in response to oxidative stress. Oncogene 24: 8085–8092.
  59. 59. Yang-Yen HF (2006) Mcl-1: a highly regulated cell death and survival controller. J Biomed Sci 13: 201–204.
  60. 60. Sitailo LA, Tibudan SS, Denning MF (2006) The protein kinase C delta catalytic fragment targets Mcl-1 for degradation to trigger apoptosis. J Biol Chem 281: 29703–29710.
  61. 61. Liu H, Eksarko P, Temkin V, Haines GK III, Perlman H, et al. (2005) Mcl-1 is essential for the survival of synovial fibroblasts in rheumatoid arthritis. J Immunol 175: 8337–8345.
  62. 62. Timofeeva OA, Plisov S, Evseev AA, Peng S, Jose-Kampfner M, Lovvorn HN, Dome JS, Perantoni AO (2006) Serine-phosphorylated STAT1 is a prosurvival factor in Wilms' tumor pathogenesis. Oncogene 25: 7555–7564.
  63. 63. Rudnicki M, Eder S, Perco P, Enrich J, Scheiber K, et al. (2007) Gene expression profiles of human proximal tubular epithelial cells in proteinuric nephropathies. Kidney Int 71: 325–335.
  64. 64. Wang SN, LaPage J, Hirschberg R (2000) Role of glomerular ultrafiltration of growth factors in progressive interstitial fibrosis in diabetic nephropathy. Kidney Int 57: 1002–1014.
  65. 65. Iglesias J, Levine JS (2001) Albuminuria and renal injury–beware of proteins bearing gifts. Nephrol Dial Transplant 16: 215–218.
  66. 66. Kees-Folts D, Sadow JL, Schreiner GF (1994) Tubular catabolism of albumin is associated with the release of an inflammatory lipid. Kidney Int 45: 1697–1709.
  67. 67. Kamijo A, Kimura K, Sugaya T, Yamanouchi M, Hase H, et al. (2002) Urinary free fatty acids bound to albumin aggravate tubulointerstitial damage. Kidney Int 62: 1628–1637.
  68. 68. Thomas ME, Harris KP, Walls J, Furness PN, Brunskill NJ (2002) Fatty acids exacerbate tubulointerstitial injury in protein-overload proteinuria. Am J Physiol Renal Physiol 283: F640–F647.
  69. 69. van Timmeren MM, Bakker SJ, Stegeman CA, Gans RO, van Goor H (2005) Addition of oleic acid to delipidated bovine serum albumin aggravates renal damage in experimental protein-overload nephrosis. Nephrol Dial Transplant 20: 2349–2357.
  70. 70. Tarshish P, Tobin JN, Bernstein J, Edelmann CM Jr (1997) Prognostic significance of the early course of minimal change nephrotic syndrome: report of the International Study of Kidney Disease in Children. J Am Soc Nephrol 8: 769–776.
  71. 71. Drummond K, Mauer M (2002) The early natural history of nephropathy in type 1 diabetes: II. Early renal structural changes in type 1 diabetes. Diabetes 51: 1580–1587.
  72. 72. Katz A, Caramori ML, Sisson-Ross S, Groppoli T, Basgen JM, et al. (2002) An increase in the cell component of the cortical interstitium antedates interstitial fibrosis in type 1 diabetic patients. Kidney Int 61: 2058–2066.
  73. 73. Woltman AM, de Fijter JW, Zuidwijk K, Vlug AG, Bajema IM, et al. (2007) Quantification of dendritic cell subsets in human renal tissue under normal and pathological conditions. Kidney Int 71: 1001–1008.
  74. 74. Heymann F, Meyer-Schwesinger C, Hamilton-Williams EE, Hammerich L, Panzer U, et al. (2009) Kidney dendritic cell activation is required for progression of renal disease in a mouse model of glomerular injury. J Clin Invest 119: 1286–1297.
  75. 75. Nikolic-Paterson DJ, Atkins RC (2001) The role of macrophages in glomerulonephritis. Nephrol Dial Transplant 16: Suppl 53–7.
  76. 76. Gautier L, Cope L, Bolstad BM, Irizarry RA (2004) affy–analysis of Affymetrix GeneChip data at the probe level. Bioinformatics 20: 307–315.
  77. 77. Gentleman RDaH (2003) The Bioconductor FAQ. http://www.bioconductor.org/ (accessed 2008).
  78. 78. Irizarry R, Gautier L, Cope L (2003) An r package for analyses of affymetrix oligonucleotide arrays. In: Parmigiani G, Garrett ES, Irizarry RA, Zeger SL, editors. The Analysis of Gene Expression Data: Methods and Software. New York: Springer. pp. 102–119.
  79. 79. Bolstad BM http://rmaexpress.bmbolstad.com/ (Accessed 2007).
  80. 80. Tusher VG, Tibshirani R, Chu G (2001) Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci U S A 98: 5116–5121.
  81. 81. Smyth GK (2004) Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat Appl Genet Mol Biol 3: Article3.
  82. 82. Eisen MB, Spellman PT, Brown PO, Botstein D (1998) Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95: 14863–14868.
  83. 83. Tritchler D, Fallah S, Beyene J (2005) A spectral clustering method for microarray data. Computational statistics & data analysis 49: 63–76.
  84. 84. Li C, Wong WH (2001) Model-based analysis of oligonucleotide arrays: expression index computation and outlier detection. Proc Natl Acad Sci U S A 98: 31–36.
  85. 85. Gentleman R (2007) Using GO for Statistical Analysis. www.bioconductor.org (accessed 2008).
  86. 86. Wettenhall JLab 7: Gene ontologies and gene sets in analysis of differential expression. http://www.amstat.org/meetings/jsm/2005/ (accessed 2007).
  87. 87. Tibshirani R (1996) Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society Series B 58: 267–288.