There are at least 32 peer-reviewed publications from 12 separate laboratories supporting the in vitro study of SFs as a model to study diabetes and its complications, especially diabetic nephropathy. Skin biopsies are well accepted by research subjects and SFs grow well under appropriate culture conditions, retaining many of their original phenotypic characteristics. Using a culture system whereby cells from different individuals are exposed to identical culture conditions for several weeks and multiple passages likely reduces influences of acute metabolic disturbances may thus better reflect longer-term epigenetic influences of the diabetic state in these genetically identical T1D discordant twins.
We incubated the SFs in HG in order to seek proof for the concept that SF in vitro behaviors can reflect memory for the in vivo metabolic differences between the diabetic and nondiabetic identical twins. Thus, this study was designed to test whether long-term hyperglycemia changes cellular behavior in a persistent and systematic way, above and beyond the effects of several weeks of in vitro exposure of the nondiabetic twins’ cells to these HG levels. In this in vitro condition, systematic differences between the twin pairs would more likely reflect the long-term consequences of in vivo hyperglycemia rather than acute effects of HG on SF behaviors. Studies of these cells under NG conditions would also be of interest, perhaps revealing time lines and mechanisms for reversal of the in vivo effects.
Statistical hypothesis testing seeks to find a balance between maximizing the power (1-type II error) to detect a valid signal and minimizing the probability of finding a false-positive (type I error). This balance between type I and type II errors is often a source of philosophical debate between experimental biologists and statisticians. Statisticians tend to be more concerned about type I errors and thus establish very conservative significance criteria that control for false-positives in the context of multiple testing. Statistical methods such as Bonferroni that correct for all tests performed assume that the universal null hypothesis of no valid relationships is true. Experimental biologists, on the other hand, are often more concerned with type II errors and are unwilling to assume that the universal null hypothesis is true. This is because biological experiments almost always build on the prior work of others. Given a set of prior work, biologists can use their knowledge of the end point being studied and its related biological processes to help determine whether a set of statistical results are likely to provide evidence in support of a given hypothesis. Here we carried out statistical analyses to determine whether differentially expressed genes are enriched in biological pathways defined by the KEGG database. We established a liberal statistical significance criteria of α = 0.05 that was uncorrected for the number of pathways that we examined. We chose not to correct for the number of pathways considered because the existing biological knowledge about diabetes and its complications suggests that the universal null hypothesis is unlikely to be true. Further, we chose to maximize our power to detect pathways by using a more liberal significance threshold. Using this approach, we found a total of eight pathways that were enriched for differentially expressed genes. The statistical concern is that a liberal threshold would yield a number of false-positives. We chose to address this concern through biological interpretation using prior knowledge from published studies. We review these results below. Suffice it to say that with more conservative statistical approaches, the present studies could be entirely unrevealing, whereas potentially important observations emerged with the more liberal approaches used here. Particularly in very slowly progressing chronic conditions such as diabetes complications, subtle cellular perturbations sustained over decades are more likely to occur than more dramatic pathway disturbances. If so, new analytic strategies may be needed to better understand these processes.
Identical twins discordant for diabetes could represent an ideal human model for the study of epigenetic processes related to the diabetic state. Although systematic differences between these twin pairs could also represent differences relating to the triggering factors leading to diabetes, this is unlikely given that such factors probably vary among T1D patients and may be unlikely to be influencing cell behavior so many years after T1D onset. Moreover, one might predict such long-lasting influences of T1D triggers to be in inflammatory pathways, but, in one study, monocyte gene expression profiles did not differ among T1D-discordant identical twins, whereas both the diabetic and nondiabetic twins differed from controls in such pathways (18
). In a study of Vα24JαT-cell clones activated by anti-CD3, differences between discordant identical twin pairs were found in a large number of genes in multiple pathways, not restricted to inflammatory processes (19
). Importantly, the gene expression differences found in the current study were in pathways that are known to be affected when normal cells are placed in HG, and in pathways that are, in most instances, highly relevant to diabetes complications.
For example, the cyclooxygenase direction of the arachidonic acid pathway can initiate the inflammatory response through conversion of arachidonic acid into an unstable endoperoxide intermediate that is converted to prostacyclin and thromboxane A2
, resulting in activation of NF-κB and inducible nitric oxide synthase, culminating in mitochondrial stress, which has been linked to diabetes complications (20
). The 12-lipoxygenase direction of the arachidonic acid pathway mediates some actions of angiotensin II, including increasing extracellular matrix synthesis in mesangial cells (21
). 12-Lipoxygenase also has a role in p38-mitogen activated protein (MAP) kinase and α5 (IV) collagen synthesis in diabetic rats and cultured mouse podocytes (22
), linking this pathway to diabetic nephropathy. This link was strengthened by the demonstration that the 12-lipoxygenase genotype is associated with albuminuria in T2D patients with poor glycemic control (23
The TGF-β pathway has long been considered to play a pivotal role in diabetes complications (24
). Thus, as reviewed elsewhere (24
), studies in cell culture systems, animal models, and humans support the hypothesis that increased TGF-β production mediates hypertrophic and fibrotic manifestations of diabetic nephropathy. This appears to result from the convergence of oxidative stress, nonenzymatic glycation, and hemodynamic processes on the upregulation of TGF-β signaling and, consequently, increases in extracellular matrix molecule production by glomerular, tubular, and interstitial cells (24
Glutathione metabolism, representing one of the most important antioxidant pathways, has been linked to diabetes complications (26
). Ceriello et al. (28
) reported inadequate antioxidant enzyme responses to HG, including that of glutathione peroxidase in SFs of T1D patients with diabetic nephropathy, whereas these responses were normal in T1D patients without nephropathy. We found no statistically significant evidence of heritability in SF glutathione peroxidase gene expression in T1D sibling pairs who were concordant for diabetic nephropathy lesions (14
), which is consistent with epigenetic explanations for the SF behavior differences noted by Ceriello et al. (28
The complement pathway has also been linked to diabetes complications. The eyes of T2D patients showed deposition of the C5b-9 terminal product of complement activation (the membrane attack complex) in luminal endothelium (30
) and choriocapillaris (31
). These findings were confirmed in streptozotocin-diabetic rats and in T2D patients’ eyes, and were associated with decreased levels of glycosylphosphatidylinositol-anchored complement inhibitors (30
), thus linking with yet another differentially expressed pathway among the discordant identical twins. Membrane attack complex deposition was also found in endoneurial microvessels of patients with diabetic neuropathy (32
), and in glomeruli, tubular basement membranes, and vessel walls in advanced diabetic nephropathy (33
). Mannose binding lectin (MBL) can directly activate the complement system. High levels of MBL early in the course of T1D independently predicted the later development of overt nephropathy. Moreover, high MBL expression genotypes were more common in T1D patients with overt nephropathy (34
Increased endothelial permeability has been associated with both diabetic nephropathy and retinopathy (37
). Endothelial permeability can be modulated by alterations in the adherens junction pathway proteins. Vascular endothelial cadherin (VE-cadherin) associates with vascular endothelial growth factor (VEGF) receptor 2, markedly reducing VEGF's induction of phospholipase Cγ and MAP kinase (39
). HG induces F-actin reorganization and associated adherens junction proteins in cultured rat heart endothelial cells (40
). Increased retinal vascular permeability in diabetic rats was associated with decreased VE-cadherin expression (41
). Interestingly, VE-cadherin is also a critical endothelial regulator of TGF-β signaling through recruitment of components of the TGF-β receptor complex, including TGF-β receptor and endoglin (42
). We found that both latent TGF binding protein-1, necessary for local TGF-β activity (12
), and endoglin (43
) gene expression were decreased in SFs of T1D patients with very slow versus rapid development of diabetic nephropathy and versus normal controls, which is consistent with protective mechanisms.
As already mentioned, oxidative stress mechanisms are likely very important in the pathogenesis of diabetic nephropathy (14
). Based on an earlier report suggesting that inhibition of the proteasome protects oxidative stress–induced endothelial dysfunction (44
), Luo et al. (45
) tested this idea in a diabetic rat model. Nontoxic proteasome inhibition reduced proteinuria and diabetic nephropathy lesions, upregulated Nrf2, and improved the diminished renal antioxidant enzyme gene expression in diabetic rats (45
Interestingly, there were also systematic differences in expression in genes important in epigenetic processes. For example, the histone methyltransferase Set7 was increased in expression in the nondiabetic twin SFs exposed to HG in vitro. As mentioned above, transient exposure of aortic endothelial cells and normal mice to HG led to persistent aortic endothelial cell epigenetic changes in the promoter of the NF-κB p65 subunit. Apparently, transient HG increased H3K4me1 in the promoter region of p65 by Set7, causing a sustained increase in p65 gene expression and, consequently, in potentially damaging NF-κB-responsive genes (8
). We also found increased expression of the H3K4 methyltransferase (MLL3) gene in SFs of nondiabetic twins exposed to HG. Studies showed that the TGF-β1–induced increases in the expression of α1 (I) collagen, connective tissue growth factor, and plasminogen activator inhibitor 1 in cultured rat mesangial cells correlated positively with increased H3K4me1, H3K4me2, and H3K4me3 at their promoters (7
). The histone deacetylase 8 gene, also upregulated in nondiabetic twins exposed to HG, has been shown to be activated in the retina and its capillary cells in diabetic rats, and this persists despite the subsequent induction of good glycemic control (46
). Bovine retinal endothelial cells exposed to HG for 4 days showed similar responses (46
). Finally, we also found downregulation of histone deacetylase 4 in the nondiabetic twins. Exposure of rat cardiac myocytes to HG resulted in a reduction in the proportion of acetylated histone-4 associating with the insulin-like growth factor 1 receptor (IGF-1R) promoter. HG also resulted in decreased IGF-1R mRNA and IGF-1R protein levels, and this was linked to the HG-induced increase in apoptosis in these cells (47
In summary, these studies show that, despite 6 weeks and multiple passages in identical in vitro HG conditions, SFs of the T1D twins have gene expression differences from nondiabetic identical twins in pathways that have, in most instances, been previously implicated in diabetes complications. Although there are currently no statistical procedures that allow us to calculate the probability of identifying such pathway clustering, as opposed to random unassociated pathways, intuitively, the likelihood of this happening by chance would seem to be extremely low. Similarly, the probability that the differences in expression levels of genes related to epigenetic processes would all be for molecules previously reported to be persistently altered by HG in different cell lines and in some animal models would also appear to be very small. Moreover, given that the nondiabetic twin cells were in HG for weeks and that persistent epigenetic alterations can be induced in cells within 16 h (8
), differences between these discordant twins may have been blunted. It is, therefore, possible that a more complete picture of the prolonged impact of in vivo hyperglycemia would emerge from in vitro discordant identical twin studies that are also carried out in an NG environment. It should also be noted that many studies from different laboratories have demonstrated multiple in vitro behavior differences between SFs of T1D patients with and without diabetic nephropathy. Since, regularly, T1D cohorts with diabetic nephropathy have, on average, worse glycemia, and given the strong evidence for familial concordance for diabetic nephropathy risk (48
), it will be important to design studies that can dissect the genetic versus the epigenetic processes in the pathogenesis of diabetes complications.