Diabetic Nephropathy Is Associated With Gene Expression Levels of Oxidative Phosphorylation and Related Pathways

Participants in this study were selected from a cohort of 125 patients with >8 years of diabetes duration. This cohort, described in detail elsewhere (17), had a wide range of kidney involvement (11,12,17). Briefly, skin and kidney biopsies were obtained from each patient, and patients were ranked based on their mesangial expansion score, an estimate of the rate of development of mesangial expansion ascertained by electron microscopy morphometric analysis (11,17). The 25 patients with the highest mesangial expansion score and microalbuminuria or proteinuria were allocated to the “fast-track” and the 25 with lowest mesangial expansion score and normoalbuminuria to the “slow-track” group (11,12). Of these, 20 fast-track (all proteinuric) and 20 slow-track patients were used for the current study. Demographic characteristics were similar between the groups (Table 1). As expected, glycemia and blood pressure were higher and the glomerular filtration rate was lower in the fast-track patients. A total of 18 of 20 fast-track but none of the slow-track patients were receiving antihypertensive drugs. Glomerular structure was markedly abnormal in the fast-track group, whereas it was normal or only mildly abnormal in the slow-track group (Table 1).

The methods for cell culture and RNA isolation have been previously detailed (12). Briefly, skin fibroblasts were cultured in Dulbecco’s modified Eagle’s medium with 25 mmol/l glucose and 10% FCS for 36 h after synchronization for 24 h. Total RNA was isolated and its integrity evaluated, using the RNA 6000 LabChip kit and Agilent 2100 bioanalyzer (Hewlett Packard, Palo Alto, CA).

Biotinylated cRNA was generated from total RNA according to Affymetrix protocols (Santa Clara, CA). A hybridization mixture containing 20 μg biotinylated cRNA from each patient was hybridized to individual Affymetrix HG-U133A GeneChips. The chips were then washed, scanned, and analyzed by GeneChip Microarray Suite V.5.0. Global scaling was used on all probe sets to standardize the average signal of each array to a 500 target value, and the software provided values of expression levels of each gene relative to this average signal (18). Thus, these data are provided in relative value units. The gene expression data set was selected for further analysis as follows. First, genes that meet the criterion of at least 50% present calls in all array experiments were included in the data set. Present, absent, and marginal designations were replaced in an Excel spreadsheet with present = 1, absent = 0, and marginal = 0.2. The values for a given gene in all of the arrays were then summed, and a PAM (present, absent, and marginal) score was obtained. A score of at least 20 was required for inclusion. Second, the previously excluded data with scores <20 and >5 were analyzed, using a χ² test to discover any asymmetric distribution in the present calls between groups for any given gene. The significance level was set at P < 0.2, Fisher’s exact test (two-tailed) was used for an entry <5, and Yates corrected test was used for an entry >5. The selected gene expression levels data were then used for directional pathway analysis.

The Exploratory Visual Analysis system, which has been previously described (19,20), was modified to test directionally consistent group differences in expression levels within biological pathways defined by the KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway database (available online from http://www.genome.jp/kegg/pathway.html). This system uses a 1,000-fold permutation test on each pathway to evaluate whether the occurrence of the proportion of statistically significantly (Student’s t test, P ≤ 0.05) increased/decreased genes in a biological pathway is higher than would be expected by chance (19,20). For the P-plus test in a given biological pathway, the null hypothesis is that the number of genes with increased expression levels is the same or lower than that found in a “pseudo-pathway” with an identical number of genes randomly selected from the entire database. For the P-minus test, the number of significantly decreased genes is tested. A total of 1,000 pseudo-pathway replications were performed, and the final P value was calculated by dividing the total counts by 1,000. Thus, the P value represents the probability of obtaining the observed number of genes with higher (P-plus tests) or lower (P-minus test) expression levels in a given biological pathway than by chance alone.

The methods for RT-PCR have been described previously (11,12). The sequence of the probe and the forward and reverse primers used for NDUFS1 (NADH dehydrogenase [ubiquinone] Fe-S protein 1) were: 5′-FAM-CAAAGCTGTCACAGAGGGTGCCCA-TAMRA-3′, 5′-CACAGACAATGGCCAAATGTG-3′, and 5′-CAGCATATGGATGGTTCCTCTACTG-3′, respect-ively. The sequence of the probe and the forward and reverse primers used for COX 7B (cytochrome c oxidase subunit VIIb) were: 5′-FAM-CGCACTAAATCGTCTCCAAGTTCGAAGC-TAMRA-3′, 5′-CGATGTTTCCCTTGGTCAAAA-3′, and 5′-CCTTGCCATTGTTTGCTGAA-3′, respectively. These probes were labeled with 6-FAM (6-carboxyfluorescein) at the 5′ end and TAMRA (6-carboxy-tetramethyl-rhodamine) at the 3′ end.

Skin fibroblasts from 20 patients with a very rapid development of diabetic nephropathy (fast-track group) and 20 with a very slow rate of diabetic nephropathy development (slow-track group) were cultured in Dulbecco’s modified Eagle’s medium with 25 mmol/l glucose for 36 h, and their gene expression profiles were assessed by Affymetrix HG-U133A GeneChips. At the time of skin biopsy, HbA_1c (A1C) was higher in fast-track (9.4 ± 1.5%) than in slow-track (8.4 ± 1.2%, P = 0.027) patients. A total of 456 of ∼22,000 genes showed significant group differences (P < 0.05) in mRNA expression levels, with 237 higher and 219 lower in the fast- versus slow-track patients.

There was an increased proportion of genes with higher expression levels in the oxidative phosphorylation pathway (P-plus test, P < 0.001) in the fast-track compared with the slow-track group: expression levels of 10 of 127 genes on the U133A GeneChip belonging to the oxidative phosphorylation pathway were higher in the fast-track patients (Tables 2 and 3). No genes in this pathway had lower expression levels in the fast-track group. The expression levels of two of these genes, NDUFS1 and COX 7B, were confirmed by RT-PCR to be higher in the fast-track than in the slow-track patients (Table 4).

Four additional pathways, three closely related to the oxidative phosphorylation pathway, also had increased proportion of genes with higher expression levels in the fast- versus the slow-track group (Tables 2 and 3). These pathways included electron transport system complex III, citrate cycle, and propanoate metabolism. Group directional differences in transcription factors pathway were also seen (Tables 2 and 3). None of the pathways represented on the Affymetrix GeneChip had a greater proportion of genes with lower expression levels in the fast- versus the slow-track group. General cell regulation pathways, such as pyrimidine metabolism (P-plus test: P = 0.106; P-minus test: P = 0.188), lysine degradation (P-plus test: P = 0.62; P-minus test: P = 0.211), and cell cycle (P-plus test: P = 0.61; P-minus test: P = 0.136) were not different between the groups.

Several studies using single cell lines (3,4,5,6) have led to the hypothesis that oxidative phosphorylation pathway alterations may be important in diabetic nephropathy pathogenesis (2,7). The current study found directional differences in skin fibroblast gene expression levels in oxidative phosphorylation and related pathways. This is consistent with relatively greater oxidative stress in type 1 diabetic patients with diabetic nephropathy. Intracellular ROS are produced through the oxidative phosphorylation pathway by the mitochondrial electron transport chain in response to hyperglycemia (2). High glucose may result in overproduction of electron donors in the citrate cycle, which in turn generates a high mitochondrial membrane potential (2,7). It has been proposed that once a certain threshold is exceeded, high membrane potential inhibits electron transfer at mitochondrial respiratory chain complex III and increases the half-life of superoxide-generating intermediates such as ubisemiquinone (2,3,21). Thus, as a consequence of overall higher gene expression in the oxidative phosphorylation pathway, it is possible that mitochondrial membrane potential may exceed this threshold, resulting in decreased electron transfer at respiratory chain complex III and overproduction of superoxide and ROS.

Four of the five pathways with directionally consistent group differences in gene expression levels were highly clustered among >200 known biological pathways in the genome (Fig. 1). This is, to our knowledge, the first application of this new method of analysis. However, statistical methods are not yet fully developed that would allow calculation of the likelihood that these four related pathways could show directional consistent group differences by chance alone, in part because of the presence of genes that overlap in these related pathways. Nonetheless, it is intuitively very unlikely that these findings are coincidental. The propanoate metabolism pathway is connected to the citrate cycle through acetyl-CoA. Increased citrate cycle activity could further augment the oxidative phosphorylation pathway by increasing production of NADH and FADH₂, causing increased production of electron donors and increased mitochondrial membrane potential. These, in turn, could inhibit electron transfer in the oxidative phosphorylation chain, resulting in increased ROS production (3,21). Importantly, the downstream consequence of these perturbations on several pathways independently linked to diabetic nephropathy risk was abrogated in experiments that blocked the oxidative stress consequences of increased mitochondrial ROS production (2,3,21). If the functional consequence of the current findings were increased ROS production, this, coupled with the already demonstrated antioxidant enzyme defects in skin fibroblasts from diabetic nephropathy patients (9,10), could result in marked oxidative stress in these cells. Because this and other skin fibroblast behaviors are related to diabetic nephropathy (9–13,15,16), it is tempting to hypothesize that similar processes may be occurring in the kidney.

Skin fibroblasts from the patients in this study were cultured under identical conditions for several passages before gene expression profiling was performed. However, high-glucose–induced overexpression of fibronectin and collagen IV in human cultured endothelial cells persisted after several subsequent passages in the normal glucose condition (22). The fast-track patients had higher A1C than the slow-track patients. Thus, the current study cannot definitely differentiate whether genetic or environmental (“cell memory”) effects, or both, may be operative. If the current findings represent cell memory phenomena reflecting in vivo environmental differences, there are several possible candidate influences. Glycemic control, as expected, was worse in the fast-track group. Systemic blood pressure was higher in the fast-track group. Although it is difficult to imagine how higher systemic blood pressure could influence skin fibroblast behavior, this possibility cannot be eliminated. Also, antihypertensive drugs were usually used by the fast-track but not by the slow-track patients. Drugs blocking the renin-angiotensin system have been shown to have effects in rats that persisted (23,24) far beyond any reasonable possibility that these agents had remained in these animals’ tissues (25). Although similar studies have not been performed for other antihypertensive agents, drug environmental cell memory phenomena are possible. Other environmental influences are also possible. However, our previously published data showed that there are, at least in part, heritable regulators of in vitro skin fibroblast behaviors that are associated with diabetic nephropathy risk (14). Further studies of cellular behaviors of skin fibroblasts from identical twins discordant for type 1 diabetes could help to dissect the genetic versus cell memory questions raised here. Moreover, the current findings cannot determine whether group differences in gene expression levels represent an increase in the fast-track patients, a decrease in the slow-track patients, or both. Further studies on cells from normal control subjects would be required to answer this question.

Finally, the pathway differences reported here probably do not fully explain the altered cellular phenotype that characterizes diabetic nephropathy risk (26,27). In fact, individual genes in upstream, downstream, or unrelated pathways may be equally or more important. Thus, directional pathway analysis strategies should probably not preclude other approaches that could be applied to the understanding of the cellular basis of diabetic nephropathy risk.

C.H. was supported by a mentor-based research fellowship training grant from the American Diabetes Association. M.L.C. was supported by a research fellowship training grant from Juvenile Diabetes Foundation International. This work was supported by grants from the National Institutes of Health (DK 13083, DK 54638, and DK 070210), the National Center for Research Resources (M01-RR00400), the Viking Children’s Fund, and the National Kidney Foundation of Minnesota.

We thank John Basgen and Tom Groppoli for performing the morphometric studies, Kim Pinkham for the cell culture, Michele Watrin for patient recruitment, and Sue Kupcho-Sandberg for albumin excretion rate measurements.