Protective Effects of Cyclooxygenase-2 Gene Inactivation Against Peripheral Nerve Dysfunction and Intraepidermal Nerve Fiber Loss in Experimental Diabetes

Unless otherwise stated, all chemicals were of reagent-grade quality and were purchased from Fisher Scientific (Hanover Park, IL) and Sigma (St. Louis, MO). Kits for measurements of malondialdehyde plus 4-hydroxyalkenals (MDA) and glutathione (GSH) were purchased from Oxis International (Portland, OR). PG and cytokine enzyme-linked immunosorbent assay kits were purchased from Cayman Chemical (Ann Arbor, MI).

The experiments were performed in accordance with regulations specified by the National Institutes of Health's Principles of Laboratory Animal Care, 1985 Revised Version and the institutional animal care and use committees at the University of Toledo and the University of Michigan.

Mice heterozygous for disruption of the COX-2 gene (Ptgs2^tm1Jed) on a C57B6:129S7 strain were purchased from The Jackson Laboratories (Bar Harbor, ME) and subsequently bred at our institutions. COX-2^−/− and littermate COX-2^+/+ male mice were rendered diabetic by streptozotocin (STZ) as previously described (16). Genotype and phenotype were confirmed through PCR and Western blot as previously published (16). At least eight animals per experimental group were maintained for 6 months for end point measurements.

Male Wistar rats (200–250 g) were purchased from Charles River (Wilmington, MA). After fasting overnight, rats were rendered diabetic by STZ as previously described (11). Experimental groups comprised of nondiabetic and diabetic rats (eight or more per group) were randomly assigned to either no treatment or treatment with the selective COX-2 inhibitor (IC50 4.8 nmol/l) celecoxib (50 mg · kg⁻¹ · day⁻¹ in drinking water) and maintained for up to 6 months.

After all electrophysiological measurements were obtained, animals were killed by CO₂ inhalation followed by cervical dislocation. Sciatic nerves and dorsal root ganglia (DRG) were removed, snap frozen in liquid nitrogen, and stored at −80°C for end point measurements.

Motor nerve conduction velocity (MNCV) and sensory nerve conduction velocity (SNCV) were assessed as previously described (16,17). Briefly, animals were anesthetized with 3:1 ketamine:xylene (80 mg/kg i.p.). Hindlimb skin temperature was monitored using a thermistor and was maintained at ∼36°C by radiant heat. Sciatic MNCV was recorded by stimulating proximally at the sciatic notch and distally at the ankle. Digital SNCV was measured by stimulating at the second toe and recording at the medial malleolus. All readings are an average of 10 recordings.

These measurements were performed independently using commercially available kits as previously described (16,17). Briefly, mouse and rat sciatic nerve tissue was homogenized in 125 μl of 0.1 mol/l sodium phosphate buffer, pH 6.5. A total of 5 mmol/l butylated-hydroxy toluene was added to each sample to prevent further lipid peroxidation. A total of 20 μl homogenate at ∼15% weight to buffer were used for spectophotometric measurements of MDA performed in accordance with the manufacturer's instructions. GSH was measured by using a ∼15% wt/vol homogenate using 5% metaphosphoric acid, which prevents enzymatic GSH regeneration, as the homogenizing buffer. GSH was assessed spectophotometrically in accordance with manufacturer's instructions.

Thromboxane B₂ (TXB₂), PGE₂, 6-keto-PGF_1a (stable metabolite of prostacycline), and tumor necrosis factor (TNF)-α were measured by enzyme-linked immunosorbent assay in the sciatic nerve using commercially available kits.

Nerve tissue from treated and untreated nondiabetic and diabetic rats were analyzed as previously described (11). Rabbit anti–COX-1 and –COX-2 antibodies (1:100) (Santa Cruz, CA) or mouse anti–glyceraldehyde 3-phosphate dehydrogenase (1:5,000) (Chemicon, Temecula, CA) were used to visualize the selected proteins.

COX-2 immunoreactivity in sciatic nerves and DRG neurons was assessed by immunofluorescent histochemistry. After in vivo fixation with 4% paraformaldehyde, sciatic nerves and DRG were frozen in imbedding compound (OCT). Sections were stained with COX-2 (1:100) (rabbit; Santa Cruz) and S-1001:50 (goat; Santa Cruz) overnight at 4°C. Slides were stained with fluorescent secondary antibodies (donkey anti–goat 488 and donkey anti–rabbit 594; Invitrogen, Carlsbad, CA). Coverslips were mounted with Prolong Gold mounting media, which contains the nuclear stain DAPI.

Intraepidermal nerve fiber (IENF) density was examined in hindpaw footpads using previously reported techniques (18). Briefly, footpads were collected from the plantar surface of the hindpaws before animal perfusion, fixed in Zamboni's fixative (2% paraformaldehyde, 0.2% picric acid, and 0.1 mol/l sodium phosphate buffer) for 14 h, and cryoprotected through a series of sucrose gradients and embedded in OCT. Free-floating 30-μm footpad sections were stained with anti–PGP 9.5 antibody (a panaxonal marker) (19). Analysis of IENF density was performed using an Olympus FluoView 500 laser-scanning confocal microscope. Fluorescein isothiocyanate fluorescence was excited with a 488-nm blue argon laser, and emission was measured through a 505- to 525-nm barrier filter. Samples were scanned on an Olympus IX-71 inverted microscope using a ×20 objective. Individual IENFs were counted once they crossed the basement membrane so that fibers that branched after crossing the membrane were counted as one fiber. Data are shown as number of fibers per linear millimeter.

Hierarchical clustering was performed on a subset of data evaluating biomarkers of oxidative stress and inflammation using Gene Pattern (available at www.broad.mit.edu/cancer/software/genepattern), specifically adapted with the support of the National Center for Integrative Biomedical Informatics (available at https://portal.ncibi.org/portal). Clusters were assessed by pairwise complete linkage analysis, comparing each column to the others to determine the closest matches. The silhouette width index, a measure of the variability within a cluster versus the variability between clusters, was used to evaluate the separation between defined sample groups (20).

Data are expressed as means ± SE. Differences among experimental groups were determined by ANOVA using SPSS and PRISM 3 software, and the significance of between-group differences was assessed by Tukey multiple-range test. Significance was defined as P ≤ 0.05. If the variances for the variables were found to differ significantly, a logarithmic transformation was performed that corrected the unequal variances. All analyses were then performed on the transformed data.

In nondiabetic animals, COX-2 gene inactivation/COX-2 selective inhibition did not affect body weight or blood glucose levels compared with wild-type mice/untreated rats at baseline or thereafter (Table 1).

There was no significant difference in MNCV and SNCV between nondiabetic COX-2^+/+ and COX-2^−/− mice at any time point. Consistent with our previous short-term observations (16), 6 months of STZ-induced diabetes significantly reduced MNCV and SNCV in diabetic COX-2^+/+ mice compared with nondiabetic mice (P < 0.01), whereas diabetic COX-2^−/− mice were completely protected against diabetes-induced MNCV and SNCV slowing (P = 0.9 and P = 0.9 vs. COX-2^−/− nondiabetic group; P < 0.01 vs. COX-2^+/+ diabetic group) (Fig. 1A).

As expected, in untreated rats diabetes induced a significant slowing in both MNCV and SNCV compared with nondiabetic rats, and the deficits in both variables progressed significantly with 3- and 6-month durations of STZ-induced diabetes (P < 0.001 for both vs. nondiabetic rats) (Fig. 1B and C). Celecoxib treatment prevented both MNCV and SNCV slowing in diabetic rats at each time point without affecting either variable in nondiabetic rats (P = 1.0 and 0.977, respectively, vs. nondiabetic rats).

Oxidative stress was assessed by measurements of MDA and GSH. Under nondiabetic conditions, COX-2 gene inactivation/COX-2 selective inhibition had no effect on sciatic nerve MDA or GSH levels. MDA was significantly increased after 6 months of STZ-induced diabetes by 2.2-fold in COX-2^+/+ mice (P < 0.05 vs. nondiabetic mice) and by 3.1-fold in untreated diabetic rats (P < 0.01 vs. nondiabetic rats) (Fig. 2A and C). In contrast, diabetic COX-2^−/− mice and celecoxib-treated diabetic rats exhibited similar MDA levels to nondiabetic COX-2^−/− mice (P = 0.8) and rats (P = 1.0). GSH was significantly reduced in the sciatic nerves of diabetic COX-2^+/+ mice (P < 0.01 vs. nondiabetic COX-2^+/+ mice) and untreated diabetic rats (P < 0.05 vs. nondiabetic rats). COX-2^−/− diabetic mice preserved similar nerve GSH levels with nondiabetic mice (Fig. 2B and D), and in diabetic rats celecoxib prevented nerve GSH depletion.

Measurements of various PGs were performed in peripheral nerves to determine trends of PG changes induced by diabetes and the effects of interventions. Under nondiabetic conditions, the production of TXB₂ in the peripheral nerve was not different between COX-2^−/− and COX-2^+/+ mice at any time point. However, nondiabetic COX-2^−/− mice had significantly higher nerve 6-keto-PGF_1a levels compared with nondiabetic COX-2^+/+ mice (P < 0.01). Sciatic nerve TXB₂ was significantly increased and 6-keto-PGF_1a was significantly decreased after 6 months of diabetes in COX-2^+/+ mice (P < 0.001 for both vs. COX-2^+/+ nondiabetic mice), whereas diabetic COX-2^−/− mice were completely protected against this imbalance in PG synthesis (P = 0.8 and 0.7, respectively, vs. nondiabetic COX-2^−/− mice) (Table 2). Likewise in rats, diabetes induced a significant increase in nerve PGE₂ and TXB₂ and a significant decrease in 6-keto-PGF_1a levels at all time points (P < 0.01 vs. nondiabetic rats). Celecoxib protected diabetic rats against the increase in the vasoconstrictor and the decrease in the vasodilator PG levels (Table 2).

To further examine the link between COX-2 metabolites and inflammation in the peripheral nerve, we measured levels of nuclear factor (NF)-κB–derived cytokines. Sciatic nerve TNF-α levels were not different in nondiabetic COX-2^−/− and COX-2^+/+ mice or any nondiabetic rats. Diabetes induced a significant increase in nerve TNF-α levels in COX-2^+/+ mice (P < 0.05 vs. nondiabetic mice) and rat nerve TNF-α in untreated rats (P < 0.01 vs. nondiabetic rats), whereas diabetic COX-2^−/− mice and celecoxib-treated diabetic rats were protected against the increase in TNF-α (Fig. 3A and B).

To evaluate the clustering pattern in the change of several biomarkers of oxidative stress and inflammation induced by diabetes and/or intervention, we used methods commonly used for high throughput gene expression analysis (21), specifically adapted for our analysis as described in research design and methods. As shown in Fig. 4, cluster analysis of these biomarkers identified specific diabetes signatures and shows that COX-2^−/− diabetic mice and celecoxib-treated diabetic rats cluster with respective nondiabetic animals for markers of oxidative stress, PG production, and inflammation (silhouette width index = 0.85 and 0.83, respectively) (Fig. 4).

Western blot analysis of COX-2 protein expression demonstrates that COX-2 was undetectable in the sciatic nerve of nondiabetic rats, regardless of treatment group. However, COX-2 protein expression was increased in the sciatic nerve of all diabetic rats, and this increase was not affected by treatment with celecoxib. In contrast, diabetes had no effect of COX-1 protein expression (Fig. 5A).

Immunofluorescence of mouse sciatic nerves and DRG demonstrates a diffusely increased COX-2 immunoreactivity within the axons and DRG of COX-2^+/+ diabetic mice compared with COX-2^+/+ nondiabetic mice, as confirmed by labeling serial adjacent sections with antibody to S-100 and DAPI. No significant COX-2 immunofluorescence was found in the axons or DRG of COX-2^−/− diabetic mice (Fig. 5B and C).

COX-2 gene inactivation had no effect on IENF density in nondiabetic mice. After 6 months of diabetes, IENF density was significantly decreased in COX-2^+/+ mice compared with respective nondiabetic mice (P < 0.05). In contrast, diabetic COX-2^−/− mice were completed protected against diabetes-induced nerve fiber loss (P = 1.0 vs. nondiabetic COX-2^−/− mice) (Fig. 6).

The data reported herein confirm the critical role played by COX-2 pathway activation in mediating peripheral nerve dysfunction in experimental diabetes. Furthermore, we demonstrate that the long-term preventive effects of COX-2 gene inactivation on nerve electrophysiology and specific biomarkers of peripheral nerve function parallel the beneficial effects reported in short-term experiments (16). In addition, we also show that COX-2 gene inactivation protects against diabetes-induced IENF loss. Long-term treatment with the selective COX-2 inhibitor, celecoxib, replicates these effects in diabetic rats.

Several mechanisms may explain these beneficial effects of COX-2 inhibition on experimental DPN. As discussed below, our data suggest that modulation of peripheral nerve oxidative stress and inflammation appear to be important mechanisms explaining the neurovascular protection associated with COX-2 pathway inhibition in diabetes. Vascular mechanisms associated with either PG imbalance and/or reduced nitric oxide availability (14) with effects on endoneurial perfusion are also involved. For instance, our laboratory has previously reported that treatment with a selective COX-2 inhibitor prevented diabetes-induced endoneurial nerve blood flow deficits (11). In addition, interactions between COX-2 and the receptor for advanced glycation end products signaling pathway were recently described (22).

Consistent with our prior findings in short-term experiments (16) and with other groups (6,17), we found that GSH levels were decreased in the sciatic nerves of diabetic COX-2^+/+ mice and diabetic rats. However, diabetic COX-2^−/− mice or celecoxib-treated diabetic rats were protected against GSH depletion, which suggests that in diabetes, COX-2 pathway metabolites act independently of, but synergistically with, hyperglycemia in inducing oxidative stress. This contribution is further underscored by the significant accumulation of lipid peroxidation products in the sciatic nerves of diabetic COX-2^+/+ mice and untreated diabetic rats, whereas diabetic COX-2^−/− mice and celecoxib-treated diabetic rats were protected against this additional marker of oxidative injury.

Activation of the COX-2 pathway is an important consequence of oxidant- and inflammation-initiated metabolic response in various tissues. A chain of events linking ROS generation with NF-κB activation and subsequent COX-2 protein production is supported by multiple studies (14,23,24). The increased lipid peroxidation associated with oxidative stress has been also shown to be an independent and potent inducer of COX-2 (25). However, our data indicate that COX-2 activation itself promotes oxidative stress in the diabetic peripheral nerves. This may be explained through vascular and nonvascular mechanisms induced by imbalanced PG generation and subsequent metabolic and inflammatory alterations in the peripheral diabetic neurovascular structures.

COX is the rate-limiting enzyme in PG synthesis and plays an essential role in neuroinflammation. Two isoforms of the enzyme, COX-1 and COX-2, have been isolated in mammalian cells. While COX-1 is ubiquitous, COX-2 operates as an inducible enzyme with low or undetectable levels under normal conditions but is rapidly upregulated in various disease states including diabetes.

Our laboratory (11) and other groups (26,27) have demonstrated that the COX-2 expression is elevated in the peripheral nerves and vascular tissues in various models of experimental diabetes, and this was further confirmed in the present study (Fig. 5A). We have now extended these studies to show that COX-2 upregulation subsequently induces a progressive PG imbalance due to increased levels of vasoconstrictor TXB₂ and proinflammatory PGE₂ and decreased prostacyclin in the peripheral nerves, which was maintained during long-term diabetic conditions in both animal models (Table 2). This imbalance, in turn, may contribute to the observed oxidative and neurovascular deficits. These findings are consistent with several studies (28–30) reporting a similar shift in the PG balance induced by diabetes. A COX-2–dependent imbalanced PG response favoring PGE₂ has been shown to promote oxidative stress and related neuronal damage in a model of Alzheimer's disease via PGE₂ receptor signaling (31). In addition, COX-2 upregulation stimulates a peroxidase-dependent conversion of PGG₂ to PGH₂, resulting in an increased superoxide production and subsequent lipid peroxidation and protein nitrosylation (32). Although we have not directly measured ROS generation in this study, several reports (32,33) demonstrate a direct relationship between COX-2 and ROS production.

Our data indicate that additional markers of inflammation, such as the NF-κB–derived TNF-α (32), are elevated by diabetes in the peripheral nerves and that either COX-2 gene inactivation or pharmacological COX-2 blockade prevents this increase. Several molecular possibilities may explain different cellular effects involving the interaction between PGs, inflammation, and COX-2 in the peripheral nerves. For instance, COX-2 has an NF-κB binding site in its promoter region, and the NF-κB pathway directly modulates cellular inflammatory processes within the nervous system (32). It has been shown that while PGE₂ upregulates NF-κB activity, with subsequent increasing inflammatory response, other PGs such as PGD₂ and PGJ₂ suppress NF-κB activation and decrease inflammation (32). Therefore, in diabetes, COX-2–derived PGE₂, in concert with oxidative stress, generates a positive feedback loop resulting in NF-κB activation and subsequent inflammation. These data suggest that the relationship between oxidative stress, inflammation, and COX-2 activation in our diabetic models appear to be bi- rather than unidirectional, since COX-2 activation not only results from but also leads to increased oxidative stress and inflammation in the diabetic peripheral nerves.

Clustering analysis was applied to show a regulatory pattern linking PG, oxidative stress, and inflammatory markers as well as the presence of a specific diabetic phenotype signature (Fig. 4). The net cluster demarcation between diabetic COX-2^+/+ mice or untreated diabetic rats and their COX-2^−/− and celecoxib-treated counterparts is an additional strong and original indicator of a similar nondiabetic phenotype induced by COX-2 inactivation/inhibition, supporting our hypothesis that increased COX-2 activity in diabetes mediates an unbalanced PG/ROS/inflammation interplay in peripheral nerves.

As reported previously (11) and confirmed in this communication (Fig. 5), the expression of COX-2, but not COX-1, is elevated during diabetes in the peripheral nerves and DRG. It has also been shown that in the nervous system, different PGs have the ability to antagonize one another and a PG can exert opposite effects in different cell types (32). This spectrum of responses, associated with antagonizing the actions of growth factors, has direct effects on neuron survival and nerve function (32). Studies in purified preparations of human COX-1 and COX-2 have also shown that even when both isoforms are present in the same intracellular compartment, downstream PG generation follows disease-specific paradigms and effective PG synthesis may proceed predominantly through one of the COX isoforms (31,34). Our observations in nondiabetic COX-2^−/− mice, which had consistently significantly higher nerve 6-keto-PGF_1a levels compared with nondiabetic wild-type mice, support the hypothesis that formation of the vasodilator PG occurs mainly by a mechanism independent of the COX-2 pathway. Celecoxib treatment had no effect on COX-2 protein expression in nondiabetic or diabetic rats, suggesting that the inhibition of COX-2 activity is responsible for the beneficial effects observed with celecoxib (Fig. 5A).

In this report, we utilized animal models of type 1 diabetes, and, therefore, it is unclear as to whether inhibition of the COX-2 pathway will be as effective in treating DPN in type 2 diabetes. However, recently reported data (35,36) demonstrate that COX-2 is also upregulated in animal models of type 2 diabetes and that this upregulation is associated with similar inflammatory and metabolic changes to those reported herein and ultimately the development of microangiopathic complications. Therefore, COX-2 inhibition may also provide a potential therapeutic benefit for DPN in type 2 diabetes.

Finally, we have observed that our diabetic COX-2^+/+ mice have a marked decrease in IENF density compared with their nondiabetic littermates, whereas the diabetic COX-2^−/− mice were protected against nerve fiber loss for 6 months of experimental diabetes. In animal models, nerve fiber density is often used to verify nerve fiber degeneration (37,38). Small-diameter nerve fibers can be affected early in peripheral neuropathy, and terminal sensory nerve endings might have degenerated despite normal sural morphometry (39,40). Therefore, routine NCV tests lack the sensitivity to detect small fiber impairment. On the contrary, footpad skin biopsy has proved to be a reliable tool to examine unmyelinated nerve fibers, as assessed by the quantification IENF density (38,41–43), and it is now accepted as a sensitive method in assessing the presence of DPN.

The data presented herein support the concept that COX-2 inhibition has beneficial effects in preventing diabetes-induced peripheral nerve dysfunction and could be used as a new potential therapeutic tool. A limitation of our study, however, was its preventive design, and so an interventional (or reversal) study could potentially yield different results.

Also, emerging data from several clinical trials (44–46) have recently questioned the cardiovascular safety of COX-2 selective inhibitors. One of the potential mechanisms involved includes a predominant decrease in PGI₂, but not TXB₂, production (44,47) and increased blood pressure. However, our data demonstrate the presence of specific PG and inflammatory imbalances in diabetes, which could potentially alter the cardiovascular risk association with COX-2 inhibition. Moreover, significant differences in cardiovascular outcomes were reported with various agents within class, which suggest rather a drug- and not a class-specific effect (44,45,48,49). Indeed, a recent trial specifically exploring cardiovascular risk measures in subjects with type 2 diabetes randomized to celecoxib, rofecoxib, or naproxen reported no increase in systolic blood pressure with celecoxib, while rofecoxib induced a significant increase in systolic blood pressure after 12 weeks (50). Nevertheless, an interesting possibility to be explored in future studies is a conditional tissue-specific COX-2 inhibition approach, which would promote the desired effects in target tissues without the unwanted consequences that may result from a general COX-2 inhibition.

Therefore, in summary, our data support an important role for increased flux through the COX-2 pathway in 1) loss of sensory nerve function assessed by electrophysiological measurements, 2) increased oxidative stress and inflammation in the peripheral nerves, and 3) nerve fiber loss measured by IENF density in footpads. Selective COX-2 inhibition may therefore be useful for preventing or delaying DPN.

This research is supported by National Institutes of Health Grant NINDS-NS047653-03 (to R.P.B). M.J.S. is supported by grant R01AT002146 and Eli Lilly. This work utilized the Morphology and Image Analysis Core of the Michigan Diabetes Research and Training Center funded by NIH5P60 DK20572 from the National Institute of Diabetes and Digestive and Kidney Diseases.

We thank Pfizer for the gift of the selective COX-2 inhibitor celecoxib and Dr. Kelli Sullivan of the University of Michigan for immunohistochemical advice.