Metabolic Dysfunction Is Restricted to the Sciatic Nerve in Experimental Diabetic Neuropathy

Diabetic neuropathy (DN) will develop in ∼30–50% of patients with diabetes. DN typically presents with sensory symptoms in a distal glove-and-stocking distribution (1,2). It is a poorly understood complication of diabetes, and currently, few treatments are available (3). Raised glucose has long been believed to instigate pathology in DN either through direct neurotoxicity or from the activation of secondary pathways (4,5). However, exactly how these pathways cause nerve conduction velocity deficits, neuropathic pain, distal axonopathy, and numbness in the extremities continues to elude us, and many clinical trials aimed at specific targets have failed due to lack of efficacy (3).

Crossover exists between proposed pathogenic pathways in DN, but how these interact is unclear. This question can be approached by implementing extensive -omic technologies to measure many transcripts, proteins, or metabolites in parallel. Gene microarrays were among the first of these technologies to be used, and transcriptomic analyses have been performed on tissues such as the dorsal root ganglia (DRG) from streptozotocin (STZ)-diabetic rats compared with healthy controls (6), the sciatic nerve (SN) of db/db mice compared with those from db/+ mice (7), and sural nerve biopsy specimens from human patients whose neuropathy progressed over 1 year compared with those whose neuropathy did not (based on myelinated fiber density loss) (8). Common changes across these gene array studies highlight altered carbohydrate and lipid metabolism.

Because gene transcript levels do not always correlate with protein expression due to varying transcriptional/translational control, these studies have been expanded by the measurement of proteins through proteomics using a cell culture–based method known as stable isotope labeling by amino acids in cell culture, or SILAC. In neurons derived from the DRG of diabetic rats (22 weeks after STZ) compared with healthy controls and maintained in culture with high and normal glucose media, respectively, there was significant downregulation of proteins related to oxidative phosphorylation and the tricarboxylic (TCA) cycle in diabetes/high glucose (9). In contrast, Schwann cells derived from the SN of neonatal rats and cultured in high glucose for 2, 6, or 16 days compared with normal glucose showed upregulation of oxidative phosphorylation and the TCA cycle at all time points (10). Both findings strongly point to metabolic dysfunction and mitochondrial impairment occurring in DN.

These studies describe the effects of high glucose on a single cell type in culture but do not replicate the multifaceted in vivo environment. The nervous system exists as a complex arrangement of cell types (encompassing neurons, glia, endothelial cells, and more), and therefore, primary single-cell culture models are not able to address tissue-wide changes in diabetes and the distal-proximal presentation of DN. Tissue measurements address such limitations, and targeted metabolomic analysis of the sural nerve, SN, and DRG of db/db mice compared with db/+ mice found that the metabolic intermediates of glycolysis and the TCA cycle are significantly downregulated in the sural nerve and SN but not in the DRG in diabetes (11).

Targeted analyses, however, are based on a priori assumptions; thus, there is sufficient scope for a reassessment of established theories and the generation of new targets by analyzing proteins and metabolites in multiple nervous tissues alongside one another in an unbiased and holistic analysis. Because DN is a complex, multifactorial process with not all sites of the peripheral nervous system (PNS) affected equally, we combined proteomic and metabolomic analyses of three different regions of the PNS from control and diabetic rats (12 weeks after STZ). The integration of both technologies presented here provides novel insights into the distal-proximal pathogenesis of DN, and these data enable the generation of new therapeutic targets.

All reagents were purchased from Sigma-Aldrich unless otherwise stated.

All animal experiments were carried out in accordance with the U.K. Animal (Scientific Procedures) Act 1986 and institutional ethics policies. Adult male Sprague-Dawley rats (mean ± SD initial weight 370 ± 19 g; Charles River) were randomly assigned to treatment groups and injected intraperitoneally with 55 mg/kg STZ in 0.9% sodium chloride (n = 16) or an equivalent volume of saline (n = 14) after an overnight fast. Hyperglycemia was confirmed 3 days after STZ, and animals were housed in groups of two to three under a 12-h light-dark cycle with ad libitum access to food and water. After 12 weeks, nerve conduction velocity was measured in the SN of terminally anesthetized rats (2% volume for volume [v/v] isoflurane in oxygen) as previously described (12).

Terminally anesthetized rats were culled by decapitation, and blood glucose was measured from core blood with a strip-operated reflectance photometer (12). SN, lumbar 4/5 DRG, trigeminal ganglia (TG), and plantar skin were dissected. All tissue collection was limited to a 90-min period each day to minimize circadian influences on metabolism and limit analyte degradation. Tissues destined for metabolomics were rapidly washed in ice-cold 0.9% NaCl in water. Tissues for proteomics were rapidly washed in ice-cold PBS followed by ice-cold 0.25 mol/L sucrose. Tissue was snap frozen in liquid nitrogen to halt metabolic and proteolytic processes. The mass of each tissue was recorded and samples stored at −80°C. Intraepidermal nerve fiber density was measured by quantification of protein gene product 9.5–immunoreactive nerve fibers in the glabrous surface of the hind paw as previously described (12).

Analysis of tissue metabolites was performed by lysing tissue in a mixture of choloroform, methanol, and water to separate polar and nonpolar metabolites before analysis by gas chromatography (GC) or liquid chromatography (LC) mass spectrometry (MS). The weight of each sample was equalized and tissue lysed in 800 µL 50:50 MeOH:CHCl₃ containing isotopically labeled internal standards (0.016 mg/mL each citric acid-d₄, ¹³C₆-d-fructose, l-tryptophan-d₅, l-alanine-d₇, stearic acid-d₃₅, benzoic acid-d₅, and leucine-d₁₀; Cambridge Isotopes) in a TissueLyser II (QIAGEN). Four hundred microliters of water were added and samples centrifuged at 2,400g for 15 min to separate phases. After separation, tissue debris lay at the interface between the lower (nonpolar, CHCl₃) phase and the polar (MeOH:H₂O) phase. Metabolites in the polar fraction were derivatized to methoxime/trimethylsilyl derivatives as previously described (13) and analyzed by GC-MS (Agilent 7890A GC and LECO Pegasus high-throughput GC with time-of-flight MS). Metabolites in the nonpolar fraction were resuspended in methanol and analyzed by LC-MS (Accela ultra-high-pressure LC system coupled to an Orbitrap Velos MS). Instrumental analysis and data reduction and interpretation followed a previously described protocol (14). Univariate statistical tests (mean ratio with CIs, Mann-Whitney P value, false discovery rate [FDR]–corrected q value) were applied in R. All metabolites identified/quantified are shown in Supplementary Table 1.

Proteomics was performed using 8-plex iTRAQ (AB Sciex) (isobaric tags for relative and absolute quantitation), a labeled method that allows simultaneous analysis of eight independent tissue samples as previously described (15) but with the following modifications. Tissue was lysed in 1 mol/L triethylammonium bicarbonate (TEAB) with 0.1% weight for volume SDS (SN 600 μL, DRG 150 μL, TG 250 μL in TissueLyser II). Following determination of protein concentration using a modified Bradford assay (Bio-Rad), 100 μg protein were aliquoted and volumes equalized to 30 μL with 1 mol/L TEAB before cysteine reduction, alkylation, and digestion with 7.5 μg trypsin at 37°C overnight. Tryptic digests were dried in a SpeedVac concentrator (Eppendorf), resuspended in 30 μL 1 mol/L TEAB, and labeled with an 8-plex iTRAQ reagent-labeling kit. Labeled peptide samples were pooled within one 8-plex and dried in a SpeedVac concentrator.

Peptides were fractionated off-line using high-pH reversed-phase chromatography on a 3-μm Extend-C18 column (4.6 × 100 mm; Agilent) at 45°C using a 30-min gradient from 3% to 40% acetonitrile in 0.1% ammonium hydroxide at 0.75 mL/min, with 30-s fractions collected, dried in a SpeedVac concentrator, and stored at −20°C until analysis. For subsequent analysis by low-pH reversed-phase LC-tandem MS analysis, dried fractions were resuspended in 15 μL 3% v/v acetonitrile and 0.1% v/v trifluoroacetic acid, with 5 μL analyzed by low-pH reversed-phase chromatography using a nanoACQUITY LC system (Waters) online to a QSTAR Elite mass spectrometer (AB Sciex) as previously described (15).

To identify proteins from their peptide spectra, raw data files were analyzed with ProteinPilot 4.0 using default search settings against a rat-specific Uniprot database (15,190 proteins; release 2011_04), which was concatenated with a reversed-sequence decoy version of the same database to enable the FDR for identifications to be determined.

To identify differentially expressed proteins, protein-level quantification was performed on the ProteinPilot spectrum-level iTRAQ measurements by Bayesian mixed-effects modeling in R (16). For each protein identified, statistically significant FDR-controlled differential protein expression between diabetic and control animals was inferred using the measurements unique to that protein. We based our one-sided significance test on the posterior probability that the mean fold change is at least 5% above or below control expression (17). The reciprocal of this posterior probability represents the local FDR (i.e., the probability that this specific test is a false discovery). We defined a significant difference in protein expression using a global FDR threshold of 5% (17).

We framed the problem as a generalized linear mixed model with Poisson likelihood distribution and log link. In the generalized linear mixed model design, condition (diabetes/control) was treated as a fixed effect and subject as a random effect. Additional random effects were fitted for samples within each peptide so that both technical peptide–level variation and biological protein–level variation were captured. The sample normalization scaling factors represent the mass spectrometer’s exposure to each sample and hence, were included as a fixed offset within the model.

Finally, the set of measurements within each iTRAQ spectrum were assigned as follows: 1) their own baseline fixed effect to account for differences in the location of the tandem MS selection window across the chromatographic profile as well as varying ionization/fragmentation efficiencies across peptides and charge states and 2) their own independent and identically distributed log-normal residual variance to account for overdispersion due to background contamination and incorrectly identified spectra. Residual variances were assigned inverse-γ prior distributions, whereas random effects were assigned parameter-expanded Cauchy prior distributions. The model was then tested with various prior scale factors to establish that the prior distributions were not informative to the outcome. For each protein, model inference was run initially with eight chains of 20,000 samples each and 50% burn-in. Mixing was deemed successful if for each significance test the variance of the posterior probabilities for the eight chains did not exceed 0.02 (tested using a Bayesian generalized linear model with binomial likelihood distribution, logit link, 200,000 samples and 50% burn-in). If this test failed, the model was continually rerun with double the number of samples (thinned to 10,000) until the test passed. All proteins identified/quantified are shown in Supplementary Table 2.

The results of the Bayesian analysis were analyzed with Ingenuity Pathway Analysis (IPA; QIAGEN [www.qiagen.com/ingenuity]). Protein lists were inputted into IPA using Uniprot accession numbers, mean log₂ ratios, and global FDRs for each of the three tissues separately. Significant changes (global FDR <0.05) were compared with the user input data set as the reference set. To assess solely the mitochondrial proteome, proteins localized to the mitochondria were extracted through an upload of the SN data set to ConsensusPathDB (Max Planck Institute for Molecular Genetics [http://cpdb.molgen.mpg.de]). One hundred ninety-seven proteins were reported as being present in any part of the mitochondria. All mitochondrial proteins were then divided into two groups based on whether they were annotated as involved in metabolism (127 proteins) or as nonmetabolic (70 proteins).

After 12 weeks of diabetes, rats displayed a phenotype of peripheral neuropathy (18), including decreased sensory and motor nerve conduction velocities in the SN and loss of intraepidermal nerve fibers in the hind paw (Table 1). To investigate the PNS in diabetes, we first performed metabolomics on the distal SN, the corresponding lumbar 4/5 DRG, and the cranial TG of diabetic and control rats (Fig. 1A). Analysis of polar metabolites by GC-MS identified and quantified 47 metabolites in the SN, 46 in the DRG, and 47 in the TG (n = 4 per tissue). All tissues showed a similar extent of alteration, with 18 metabolites (38.3%) in the SN, 16 (34.8%) in the DRG, and 19 (40.4%) in the TG significantly changed in diabetes (Fig. 1B–D).

All tissues showed increases in glucose (SN 4.8-fold, DRG 14.6-fold, TG 19.6-fold), sorbitol, and fructose and decreases in myo- and scyllo-inositol in diabetes (Fig. 1E) characteristic of polyol pathway activation (19). Because free glucose should not normally accumulate, impaired glucose utilization appears to be present in all three tissues in diabetes. All metabolites identified/quantified are shown in Supplementary Table 1.

The GC-MS analysis also showed decreases in palmitic (16:0), stearic (18:0), and eicosanoic (20:0) fatty acids in the SN, but these were unchanged in the DRG or TG (Fig. 1E). To explore lipid species more widely, we performed nonpolar metabolomics by LC-MS. Initial analysis produced 9,166 putative metabolite features across the three tissues (n = 6 per tissue). We extracted features that showed potential changes (q <0.1) in any tissue and filtered these to 397 unique metabolite features (Supplementary Table 1). Of these, 257 of 397 (64.7%) in the SN (Fig. 2A), 15 of 321 (4.7%) in the DRG (Fig. 2B), and 1 of 365 (0.3%) in the TG (Fig. 2C) were significantly altered in diabetes (q <0.05).

The most comprehensive lipid changes observed were in the triacylglycerols (TAGs). In the SN, 73 of 110 identified TAG species (66.4%) were significantly changed in diabetes (Fig. 2E, inset). These exhibited a length-dependent phenomenon whereby relatively short-chain TAGs (<54:2) were reduced in diabetes, whereas longer-chain species increased (Fig. 2E). There was no change in any TAG species in either the DRG (Fig. 2F) or the TG (Fig. 2G). Moreover, two abundant acylcarnitine species, palmitoylcarnitine (16:0) and linoleylcarnitine (18:2), showed increases in the SN (Fig. 2H) but were unchanged in the DRG (Fig. 2I) or TG (Fig. 2J). Both families are functionally important in metabolism: TAGs for storage and acylcarnitines for fatty acid transport into the mitochondria for energy generation (Fig. 2D). Beside alterations to metabolic lipids, we observed changes in major structural/membrane lipids, including phospholipids and sphingolipids/ceramides specifically in the SN (Supplementary Table 1). These lipid changes appear severe in the distal SN, moderate in the proximal DRG, and not evident in the cranial TG, suggesting that lipid dysfunction presents distally in the PNS in diabetes.

To investigate the putative underlying mechanism of this spatial grading of metabolic dysfunction, we analyzed all three tissues by iTRAQ proteomics (control n = 4, diabetic n = 6 per tissue). We inferred differential protein expression between diabetic and control tissues by Bayesian mixed-effects modeling (see 2research design and methods). In the SN, 683 of the 2,356 proteins (28.9%) identified and quantified showed significant changes of expression in diabetes (Fig. 3A and Supplementary Table 2). IPA highlighted coordinated dysregulation to oxidative phosphorylation, liver/retinoid X receptor (LXR/RXR) activation, and glycolysis (Fig. 3B). In contrast, only 85 of 1,649 proteins (5.2%) in the DRG (Fig. 3C) and 60 of 1,734 (3.4%) in the TG (Fig. 3E) significantly changed. Pathway analysis showed changes in the acute phase response and LXR/RXR activation in both the DRG and the TG (Fig. 3D and F) but overall protein expression in the DRG and TG was relatively unaffected (Supplementary Table 2) despite the higher levels of glucose.

Mitochondrial oxidative phosphorylation was particularly conspicuous among tissues, with 32 of 37 identified proteins in the SN (86%) increased in diabetes, whereas not one of the 29 oxidative phosphorylation proteins quantified in the DRG or TG was significantly altered (Fig. 4). Dysregulation in the SN comprised increased expression in multiple subunits of complexes I, III, IV, and V, with no observed change in complex II. Because this upregulation could be, in principle, an artifact of increased mitochondrial numbers in the SN, we analyzed the composition of the mitochondrial proteome. We found mitochondrial ρ-GTPase 1 (Miro1), a protein mediating axonal transport of mitochondria (20), to be upregulated by diabetes in the SN (Supplementary Table 1), which might reflect aberrant mitochondrial transport and accumulation of mitochondria in axons. However, we found only 14 of 70 nonmetabolic mitochondrial proteins (20%) to be upregulated by diabetes compared with 86% of oxidative phosphorylation proteins. This disproportionate difference suggests targeted effects on the proteins of oxidative phosphorylation rather than increased mitochondrial numbers.

In glycolysis, 8 of 13 proteins (62%) significantly increased in the SN, with the DRG (0 of 15) and TG (1 of 15) again unaffected (Fig. 5B). Because we now have data on both the metabolites and the proteins of glycolysis, we can integrate these to assess pathways in detail. We found that although multiple glycolytic proteins increased in the SN (Fig. 5B), glycolytic intermediates glucose-6-phosphate, fructose 1,6-bisphosphate, and glyceraldehyde-3-phosphate did not significantly change (Fig. 5C). Although pathway analysis did not highlight the TCA cycle, we found that 6 of 13 TCA cycle enzymes (46%) were significantly increased in the SN, whereas none changed in either the DRG or the TG (Fig. 5E). Once again, we saw a similar phenomenon whereby proteins of the TCA cycle showed increases in the SN (Fig. 5E) without coincident increases in the metabolic intermediates citrate, succinate, or malate (Fig. 5F). Overall, we conclude that all tissues measured in the PNS of STZ-diabetic rats exhibit impaired glucose utilization but that metabolic dysfunction is restricted to the SN.

Our integrated metabolomic and proteomic analysis in experimental DN has revealed coordinated dysregulation of sugar, lipid, and mitochondrial metabolism in the distal axonal/glial compartment of the SN that is not present in the corresponding cell bodies of the lumbar 4/5 DRG or the cranial TG. Integrating both analyses allows construction of a comprehensive model of defective metabolism in the SN, where upregulation of protein components of glycolysis, the TCA cycle, and oxidative phosphorylation occur alongside complex changes in lipid metabolism (Fig. 6). Because DN predominately presents with distal symptoms, these site-specific molecular changes may directly contribute to disease pathogenesis.

We measured accumulation of glucose, fructose, and sorbitol in all tissues, supporting the view that the polyol pathway is prominently altered in DN (4). Because the polyol pathway is usually a minor route for glucose metabolism, these alterations point to changes in glucose utilization in all tissues studied. Thus, raised glucose levels cannot alone explain the pathogenesis of DN. The differential impact of impaired glucose utilization on the PNS implies that metabolic regulation differs throughout the nervous system, with some regions more susceptible to dysfunctional metabolism than others.

The synchrony of diabetes-induced dysregulation to glycolysis, the TCA cycle, and oxidative phosphorylation in the SN indicates substantial bioenergetic dysfunction. We found that although protein components of glycolysis/TCA cycle increased, their metabolic intermediates showed no evidence of alteration. Because we did not perform flux experiments, we cannot specifically comment on pathway activity, but we suggest that the discrepancy between protein and metabolite measures indicates that upregulation of the enzymes of glycolysis, the TCA cycle, and oxidative phosphorylation may reflect a compensatory response to metabolite overload in the SN. Such changes might indicate reduced ATP production because increased expression of respiratory chain components in cultured Schwann cells coincides with decreased respiration efficiency (10) and/or oxidative stress (21). Glycolytic/TCA cycle intermediates notably decrease in the sural nerve and SN but not in the DRG of db/db mice (11), supporting our interpretation that pathway activity may not be increased. However, because both studies relied on steady-state measures, definitive pathway flux is currently unknown.

Even though the normalization of all protein ratios was performed, the majority of the highlighted protein changes showed an upregulation in the SN in diabetes. In principle, spatial differences could result from differential metabolism or an artifact, such as increased numbers of mitochondria in the SN. We observed evidence of altered mitochondrial transport through increased expression of Miro1 in the SN in diabetes (20); however, increases in nonmetabolic mitochondrial proteins were infrequent compared with the 86% of oxidative phosphorylation proteins. We believe that this indicates that mitochondria are dysfunctional in distal axons but do not change in number. To support this, analysis of human skin biopsy specimens revealed no difference in mitochondrial numbers within intraepidermal nerve fibers between patients with and without DN, but mitochondrial volume increased in DN (22).

The main question arising from this work is why does the SN show such a disrupted proteomic/metabolomic signature, whereas the DRG and TG appear relatively unaffected? We believe that at least two nonmutually exclusive explanations could account for these observations. The first is due to a difference in tissue composition. Both the DRG and TG are largely neuronal, with smaller contributions from satellite glial cells. In the SN, however, there is a high proportion of Schwann cells that could be responsible for the disrupted tissue metabolism observed. This could be either a direct influence (the dysregulated proteins we measured derive predominately from Schwann cells) or an indirect influence (metabolic dysfunction in Schwann cells affects axonal health and protein expression).

The current data harmonize with elegant molecular studies of peripheral neuropathy. Schwann cell–specific mitochondrial dysfunction and metabolic stress in mice lead to symptoms of peripheral neuropathy with decreased conduction velocity, loss of small unmyelinated fibers and degeneration of large myelinated fibers, and thermal hypoalgesia (23,24). This work demonstrates that Schwann cell metabolism plays a critical role in supporting neuronal function along long peripheral nerves, and its disruption can cause axonal degeneration/neuropathy. Because glucose is preferentially taken up into Schwann cells in peripheral nerves (25) and aldose reductase predominately localizes to Schwann cells (26,27), it is possible that high glucose levels principally instigate metabolic dysfunction in Schwann cells, which affects neuronal health and results in neuropathy. Further investigation is required, but this represents an attractive target for future mechanistic and therapeutic studies in DN.

Glycolytic glia are believed to support neuronal oxidative metabolism through the transfer of lactate, which can support neuronal function during metabolic stress in Schwann cells (24,28). We did not observe any change in lactate levels in the SN, but this does not preclude alterations in the flux of lactate between neurons and glia. Lactate transfer is mediated by the monocarboxylate transporters (MCTs), of which MCT-1 is the main isoform expressed in peripheral nerves (29). We identified/quantified MCT-1 in both the SN and the DRG but did not observe significantly altered expression in either tissue (Supplementary Table 2). Changes in localization of the MCTs in the PNS in diabetes could alter lactate transfer, but few available data address this point (5).

Besides impaired metabolic support from Schwann cells, another possibility is that metabolic stress in Schwann cells themselves could result in accumulation of toxic intermediates, such as acylcarnitines, which have been shown to be neurotoxic and coincident to the development of peripheral neuropathy (30). The current data support the presence of this phenomenon in DN because we observed concomitant mitochondrial dysfunction and deficits in lipid utilization in the SN, including alterations to acylcarnitines, which were not present without metabolic stress in the DRG and TG.

Dysfunctional lipid metabolism has previously been linked to the pathogenesis of neuropathy (23,24), but its mechanism is poorly understood. The current data add that there are profound changes to lipid intermediates in the SN but not in the DRG or TG in STZ-diabetes. Total lipid content was ∼20% higher in control SN than in control DRG (data not shown). Theoretically, this difference could make changes in the SN more detectable, but this relatively small difference is unlikely to solely explain the observed extensive difference between control and diabetic tissues. These disease-associated lipid alterations occurred in the absence of protein changes to β-oxidation (Supplementary Table 2), highlighting that protein expression levels alone do not capture possible posttranslational modification effects on protein activity.

A second hypothesis approaches the relationship between the cell bodies in the DRG and the axons in the SN. The neuronal cell bodies in the DRG possibly are more capable of responding to metabolic insult through constant synthesizing and refolding of proteins than the axonal compartment, which would explain why we did not observe abundant alterations in the DRG proteome/metabolome. Perikaryal preservation is a key feature of DN because no neuronal loss was detected in the DRG after 12 months of hyperglycemia in STZ-diabetes (31,32). The DRG plays a crucial role in axonal support, as evidenced by findings in STZ-diabetes when direct support of DRG neurons through intrathecal administration of insulin (which did not reduce hyperglycemia) improved SN conduction velocity and protected against distal axonal atrophy and intraepidermal nerve fiber loss (33,34). Therefore, even apparently small phenotypic changes in the DRG may affect distal axon health and play a role in the pathogenesis of DN.

Although the current findings highlight that metabolic dysfunction is most evident in the SN, this does not negate the role played by the DRG in the pathogenesis of the disease. We believe that distal axon degeneration is most likely to result from a complex combination of both local axonal/Schwann cell dysfunction and failed axonal support from the cell body. In future work, study of the fundamental mechanisms of metabolism within the PNS will be important to allow dissection of the responses of Schwann cells, neurons, and axons.

During the present analysis, we focused on mechanisms of dysfunction in carbohydrate/lipid metabolism because these signals were the most enriched and concordant in the data sets. However, these data contain many more dysregulated proteins and metabolites that could yield interesting targets for future study. For example, endoplasmic reticulum stress and eIF2 signaling, key stress pathways linked to the pathogenesis of peripheral neuropathy (30), appear prominently altered, and LXR/RXR activation shows complex alteration that differs between tissues. Data on all of these are provided in the Supplementary Data as a resource. Because similar changes in carbohydrate metabolism have been reported in the type 2 db/db mouse model of diabetes (11), we believe that these alterations may well be applicable to further models and patients with DN. Investigation of shorter and longer durations of diabetes would be beneficial to describe the dynamics of some of the key changes highlighted here and their relationship to disease progression. Likewise, the reversion of some of these key changes with insulin or experimental therapeutics would be of interest.

Although it remains unclear whether neuron-glial metabolic coupling is disrupted in diabetes, the current measurements support the notion of a compartmentation of dysfunctional energy metabolism in peripheral neuropathy. The observation that metabolic regulation differs between the proximal cell bodies of the lumbar 4/5 DRG to the axonal/Schwann cell–rich SN may help to explain the underlying molecular basis for the glove-and-stocking distribution of peripheral neuropathies. These findings would not have been possible had we not integrated both protein and metabolite measures, and the performance of both technologies in parallel has proven vital to help to interpret pathway interactions. This comprehensive network view of dysfunctional metabolism identifies new therapeutic targets to rescue a healthy state of metabolism and to treat the progression of DN.

Funding. This research was funded by Medical Research Council Studentship and Strategic Skills Award G1001609/MR/J500410/1 (to O.J.F., R.S.P., G.J.S.C., and N.J.G.) and grant MR/L011093/1 (to A.W.D.) and facilitated by the Manchester Biomedical Research Centre, the NIHR Greater Manchester Comprehensive Local Research Network, and the Wellcome Trust (097820/Z/11/B). N.J.G. was supported by a JDRF Career Development Award (2-2009-226).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. O.J.F. contributed to the study design, data collection and analysis, discussion, and writing and review of the manuscript. R.D.U. and W.B.D. contributed to the study design, data analysis, discussion, and writing and review of the manuscript. A.W.D. developed and performed the Bayesian analyses and contributed to the discussion and review of the manuscript. P.B. performed the GC-MS analysis and contributed to the data analysis, discussion, and review of the manuscript. S.A. and R.S.P. contributed to the animal study, discussion, and review of the manuscript. K.A.H. and N.R. performed the ultra-high-performance LC-MS analysis and contributed to the discussion and review of the manuscript. G.J.S.C. and N.J.G. contributed to the study design and supervision, discussion, and writing and review of the manuscript. G.J.S.C. and N.J.G. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 23rd NEURODIAB Meeting (Diabetic Neuropathy Study Group), Barcelona, Spain, 19–22 September 2013; BNA2015: Festival of Neuroscience, Edinburgh, U.K., 12–15 April 2015; and Peripheral Nerve Society Meeting, Quebec City, QC, Canada, 28 June–2 July 2015.