Diabetic neuropathy is the most common diabetic complication, affecting up to 60% of patients and contributes significantly to pain, injury, poor wound healing and lower extremity amputation (Edwards et al., 2008
). The pathogenesis of diabetic neuropathy is complex and includes hyperglycaemia-induced oxidative stress and deranged polyol metabolism, changes in nerve microvasculature, decreased growth factor support and dysregulated lipid metabolism (Edwards et al., 2008
; Figueroa-Romero et al., 2008
). Addressing these deficits alone or in combination has yet to result in effective diabetic neuropathy treatment, confirming that an increased understanding of the mechanisms underlying the onset and progression of diabetic neuropathy is of prime importance.
The current study takes an important first step towards this goal by identifying specific sets of genes whose expression accurately classifies patient samples with regard to diabetic neuropathy progression and by analysing their interactions within known cellular pathways. Identifying common elements in these complex networks will yield novel insights into disease pathogenesis, provide new therapeutic targets and identify potential diabetic neuropathy biomarkers. The genes identified in the current study confirm data gathered from experimental models of diabetes and provide a comprehensive picture of the expression of multiple targets in a single human tissue sample.
Our initial analyses of this data set classified the patient samples based on myelinated fibre density and found that two large groups emerged; those with a loss of myelinated fibre density ≥500
over 52 weeks (progressors) and those whose myelinated fibre density was relatively stable (myelinated fibre density loss ≤100
over 52 weeks, non-progressors) (Wiggin et al., 2009
). We examined sural nerve biopsies from two groups of diabetic neuropathy patients (progressors and non-progressors) to discover differences in gene expression that could account for the differences in their clinical course. Gene expression profiling in damaged peripheral nerves by diabetic neuropathy or axotomy has been explored in experimental rodent models (Renaud et al., 2005
; Bosse et al., 2006
; Price et al., 2006
; de Preux et al., 2007
; Karamoysoyli et al., 2008
) but has yet to be examined in humans. These studies indicate that changes in gene expression in injured peripheral nerves are similar to the over-represented biological functions (inflammation and energy metabolism) reported in the current study.
Multiple cells types are affected in diabetic complication-prone tissues. Peripheral nerves contain the cellular extensions (axons and dendrites) of both sensory and motor neurons and their ensheathing glia, the Schwann cells. Other components include fibroblasts, capillary endothelial cells and a complex extracellular matrix. The majority of messenger RNA isolated from any peripheral nerve will be that generated by Schwann cells. Schwann cells are a target of hyperglycaemia and diabetes results in Schwann cell damage in part due to altered axon integrity and defective growth factor signalling (Yu et al., 2008
; McGuire et al., 2009
). In addition, inflammatory pathways including advanced glycation end products/receptor for advanced glycation end products (AGE/RAGE) signalling in axons and Schwann cells are reported in experimental animals with diabetic neuropathy and contribute to nerve damage (Lukic et al., 2008
The upregulated differentially expressed genes in progressors were enriched with ‘defence response’ and ‘inflammatory response’. Inflammation-associated molecules such as chemokines and cytokines are implicated in the development and progression of both diabetic nephropathy and diabetic neuropathy (Rivero et al., 2009
). Among these inflammation genes, bradykinin receptor B2
) is of particular interest. BDKRB2 regulates the expression of genes involved in progressive glomerulosclerosis such as transforming growth factor beta 1
) and p53
(Kakoki et al., 2006
). We recently reported that type 1 diabetic mice with dysregulated BDKRB2 developed enhanced nephropathy and diabetic neuropathy (Kakoki et al., 2010
). Membrane-associated adenosine A3 receptor (ADORA3), is also implicated in the pathogenesis of diabetic nephropathy (Pawelczyk et al., 2005
). Thus, the upregulation of cytokines, chemokines and genes such as DBKRB2
in our study () suggests enhanced inflammation and dysregulated defence responses, thus contributing to more substantial nerve damage in patients with progressive diabetic neuropathy. It is not yet clear if these cytokines and chemokines are expressed by the Schwann cells or by infiltrating macrophages (Sommer et al., 2005
), which may interact with each other in injury and demyelinating diseases (Martini et al., 2008
). Regardless, our data raise the intriguing idea that the inflammatory response should be further explored as a new therapeutic target for diabetic neuropathy.
The downregulated differentially expressed genes in the progressors were enriched with biological functions related to energy metabolism including ‘glucose metabolic process’ and ‘PPAR signalling pathway’ (). Among these differentially expressed genes, PPARγ
, encoding a nuclear receptor for glitazone, plays a key role in regulating glucose and lipid metabolism (Duan et al., 2009
). Agonists of PPARγ are effective in ameliorating diabetic neuropathy and nephropathy in animal models (Maeda et al., 2008
; Yamagishi et al., 2008
). Another key gene is APOE
, encoding an apolipoprotein, which regulates the normal catabolism of triglycerides and cholesterol. A polymorphism of this gene is linked to the progression of diabetic nephropathy (Li et al., 2010
). Decreased levels of PPARγ and APOE as well as other lipid metabolism-related differentially expressed genes correlates with the increased levels of serum triglycerides confirming our recent finding that altered lipid metabolism may play a role in the progression of diabetic neuropathy (Wiggin et al., 2009
). Further experimental work is required to determine how altered lipid metabolism influences the progression of diabetic neuropathy.
We have demonstrated that increased glucose metabolism results in increased oxidative stress, mitochondrial dysfunction and cell death in both in vitro
and in vivo
models of diabetic neuropathy (Vincent et al., 2004
; Russell et al., 2008
). In the current study, genes involved in glucose metabolism are downregulated in progressors, which is counter-intuitive to what we and others have reported for sensory neurons. Yet, considering that the majority of sural nerve RNA originates from Schwann cells, these data do support what we and others have reported regarding Schwann cells i.e. that they are resistant to hyperglycaemia-induced cell death and exhibit an enhanced antioxidant capacity (Delaney et al., 2001
; Vincent et al., 2009
). Our data also imply that sensory neurons and Schwann cells likely employ different pathways when presented with metabolic stressors such as hyperglycaemia.
Although functional enrichment analyses identify over-represented biological functions, they do not reveal how these genes interact with each other. To obtain a global view of the network, we examined gene interaction networks based on literature-derived co-citation information ( and ). Although co-citation of two genes in a single sentence does not necessarily indicate there is a direct interaction, this process may reveal novel associations and lend new insights into function (Schmelzer et al., 2008
). In the current study, the BiblioSphere co-citation network demonstrated potential interactions among differentially expressed genes and identified five major subnetworks centred on the following genes: PPARγ, APOE, SERPINE1, JUN and LEP.
The majority of the key genes identified in our network analyses are implicated in the pathogenesis of diabetes and diabetic complications (mainly diabetic nephropathy) (Supplementary Table 5
). As discussed above, PPARγ and APOE are downregulated in progressors. Downregulation of either gene in adipocytes leads to a decrease in serum lipid uptake with subsequent hyperlipidaemia (Duan et al., 2009
) and a predisposition towards developing diabetic neuropathy (Wiggin et al., 2009
). A fibrinolysis regulating gene, SERPINE1
encodes plasminogen activator inhibitor 1 (PAI-1), whose elevated levels are associated with higher incidences of diabetes (Festa et al., 2002
) and knocking out PAI-1 ameliorated diabetic nephropathy in mice (Nicholas et al., 2005
). A recent study suggested leptin's therapeutic effect in a combinatorial treatment with insulin in type 1 diabetic mice (Wang et al., 2010
). The cell cycle controlling JUN might be also involved in progression of diabetic neuropathy through its close interacting partners c-Jun N-terminal kinases (JNKs). JNKs are key signalling molecules linking inflammation and insulin resistance and are significantly activated in multiple tissues including the sural nerve of patients with type 1 and 2 diabetes (Yang and Trevillyan, 2008
). Thus, the enriched biological functions and the networks of the differentially expressed genes reflect current theories with regard to dysregulation in diabetes and its complications, suggesting their expression changes may be related to the development of diabetic neuropathy.
To fully incorporate all of the co-citation connections among the differentially expressed genes, we applied the Fast Greedy algorithm, a community structure identification algorithm, to the entire co-citation network. Fast Greedy grouped LEP and PPARγ together within the context of glucose and lipid metabolism and JUN and SERPINE1 within the context of cell death and inflammation. Three other subnetworks were identified with noteworthy key genes: ‘cell projection and axonogenesis’ with nerve growth factor receptor (NGFR), ‘cellular homeostasis and inflammatory response’ with thioredoxin and ‘cytoskeletal protein binding’ with stathmin 1 (STMN1).
Nerve growth factor receptor exerts protection against nerve damage and the expression of nerve growth factor receptor protein in plasma correlates with diabetic neuropathy progression in diabetic rats (Chilton et al., 2004
). Thioredoxin, which regulates cellular oxidative stress, is also implicated in diabetes. Thioredoxin's antioxidant activity is significantly inhibited by hyperglycaemia, suggesting its important role in vascular oxidative stress and inflammation in diabetes (Schulze et al., 2004
). No direct implication of stathmin 1, a major regulator of microtubule dynamics, in diabetes is currently known. We and others have reported regenerative changes in response to diabetic neuropathy (Dyck et al., 1986
; Sullivan et al., 2003
). The expression of genes involved in axonal extension may reflect these changes and an individual's ability to recover from nerve damage.
Our next goal was to use observed differentially expressed gene expression to classify a separate subset of biopsies using Ridge regression modelling. Regression modelling using gene expression data has proven extremely useful in predicting the progression of cancer and diabetic nephropathy (Shedden et al., 2008
; Ju et al., 2009
). In the current study, gene expression profiles from secondary biopsy samples with known myelinated fibre density (progressors or non-progressors) were compared and used in training the models. The models were then used to classify the expression profiles of a set of primary biopsies for their progression endpoint 12 months later. The best classification models included 14 genes and correctly identified 11 out of 12 patients with respect to their identification as progressors versus non-progressors. This classification accuracy (92%) is much higher than our previous naïve Bayes-based classification model's accuracy (63%) using only physiological and demographic data of these patients (Wiggin et al., 2009
). Our data demonstrate that the gene expression profile from sural nerve biopsies of patients with diabetic neuropathy achieve a higher prediction accuracy (92%) than the clinical parameters alone and are better predictors of diabetic neuropathy progression.
We hypothesize that the genes identified in our classification models () represent products or ‘genetic biomarkers’ of the biological networks involved in diabetic neuropathy onset and progression. This idea is reinforced by the fact that several of the genes have known associations with diabetes or diabetic complications. We are particularly interested in CST1
, whose expression was increased by 10-fold in progressors. CST1
, encoding a cysteine protease inhibitor, was initially implicated in gastric and colorectal tumourigenesis (Choi et al., 2009
; Yoneda et al., 2009
). Another member of this protein family, cystatin C (CST3), has been identified as a prime predictor of diabetic nephropathy progression (Shimizu et al., 2003
; Taglieri et al., 2009
). Although the CST1
gene product has not been investigated in the context of diabetic complications, it is detectable in saliva, tears and urine (Choi et al., 2009
). To date, there are no definitive biomarkers of diabetic neuropathy progression easily accessed from body fluids, and we speculate that CST1 could prove to be an easily measureable biomarker for diabetic neuropathy.
According to the Ingenuity Pathway Analysis Biomarker Filter (http://www.ingenuity.com
), nine genes (six genes from the base set and three genes from the additional set) out of the 21 genes in the four models, are detectable in readily accessible human biofluids such as saliva, blood or urine (). Initial analyses of urine and blood chemistry were used to qualify patients for study inclusion during the clinical trials. No blood or urine was collected at study end; thus, we cannot evaluate the direct correlation of these candidate genes between the sural nerve and biofluids from the same patients. Sural nerve biopsy has been replaced by minimally invasive skin biopsy as a measure of innervation and it is unlikely that sural nerve biopsies and biofluids from new patients will ever be available; however, our novel results strongly suggest these biofluid detectable genes should be prioritized as potential diabetic neuropathy predictive biomarkers in future prospective studies.
No model is currently able to predict potential development of diabetic neuropathy. We do not know if our models would be useful in identifying patients who could potentially develop diabetic neuropathy since our gene expression-based models are optimized for differentiating patients with progressive diabetic neuropathy from those with non-progressive diabetic neuropathy. We do not expect our models to be used in a clinical setting in itself, especially since sural nerve biopsies are rarely performed; however, these models provide a starting point to identify potential biomarkers since many of the genes in our models are detectable in human biofluids. We expect that these potential biomarkers will be examined in a clinical context to assess their usefulness in identifying patients at risk of developing diabetic neuropathy and those whose disease course may be more severe.
In summary, we report for the first time differential gene expression of human sural nerves from patients with progressive and non-progressive diabetic neuropathy. Biological enrichment and network analyses identified several novel areas of biological importance, yielding new insight into disease pathogenesis and opening up new areas of potential investigation for the discovery of mechanism-based therapies. We have also reported for the first time the expression signatures of gene sets that accurately classify patients with regard to diabetic neuropathy progression. While translating gene expression to predictive biomarkers measurable in the clinic remains a challenge, we report several novel potential biomarker candidates for diabetic neuropathy. Evaluation of these biomarker candidates and refinement of patient classification models represents an exciting new direction in the management and treatment of diabetic neuropathy. Collectively, our results represent the first exploration of gene expression arrays from human sural nerves of patients with varying degrees of diabetic neuropathy and provide new insight into disease pathogenesis and biomarker identification.