Surrogate Markers of Small Fiber Damage in Human Diabetic Neuropathy

Patients attending the Manchester Diabetes Centre and nondiabetic volunteers who underwent neurological examination and electrophysiology were invited to participate unless they had a neuropathy of another cause or absent pedal pulses or wore contact lenses.

Symptoms were assessed by using the diabetic neuropathy symptom score (23) and painful symptoms were quantified with a visual analog score ranging between 0 and 10. Neurological deficits were assessed using the neuropathy disability score (NDS) (24), quantitative sensory tests (cooling detection threshold [CDT] and heat-as-pain visual analog score minimal threshold, intermediate threshold, and differential threshold expressing pain tolerance (HP-VAS 0.5, 5.0, and 0.5–5.0, respectively) (1), autonomic function (deep breathing heart rate variability [DB-HRV]) with a CASE IV system (WR Medical Electronics, Stillwater, MN), and electrodiagnostic studies with a Dantec “Keypoint” system (Dantec Dynamics, Bristol, U.K.) equipped with a DISA temperature regulator to keep limb temperature constantly between 32 and 35°C. The NDS included tuning fork vibration perception, pin prick perception, and temperature perception as well as the presence or absence of ankle reflexes and ranged between 0 and 10; neuropathy was diagnosed if the NDS was ≥3 of 10. Patients were deemed to have mild neuropathy when the NDS was between 3 and 5, moderate neuropathy when the NDS was between 6 and 8, and severe neuropathy when the NDS was 9 or 10. Electrophysiological measures included peroneal (PNOL), sural (SNOL), and tibial (TNOL) nerve onset latencies, amplitude potential (PNAP, SNAP, and TNAP, respectively), conduction velocity (PNCV, SNCV, and TNCV, respectively), and peroneal (PNFL) and tibial (TNFL) F-waves. For small fiber evaluation we adopted the CASE IV normality range. For thermal detection thresholds the 95th percentile and for autonomic function (DB-HRV) the 5th percentile was considered abnormal. We used DB-HRV, CDT, HP-VAS 0.5, and HP-VAS 5.0 to stratify the severity of small fiber dysfunction, and we chose the following criteria: normal (all tests normal), mild (one abnormal test), moderate (two abnormal tests), and severe (≥3 abnormal tests).

Patients were examined with a Confoscan corneal confocal microscope model P4 (Tomey, Erlangen, Germany). Each eye being examined was anesthetized with 1 drop of 0.4% benoxinate hydrochloride (oxybuprocaine hydrochloride; Minims). The objective lens of the confocal microscope was disinfected (isopropyl alcohol 70% vol/vol swabs). A large drop of Viscotears liquid gel (carbomer 940; Ciba Vision Ophthalmics) was applied onto the tip of the lens and advanced slowly forward until the gel touched the cornea, allowing optical but not physical contact between the objective lens and corneal epithelium during the examination. One eye of each subject was selected at random for examination. Several scans of the entire depth of the cornea were recorded by turning the fine focus of the objective lens backward and forward for ∼2 min to acquire satisfactory images of all corneal layers, providing en face two-dimensional images with a lateral resolution of ∼1–2 μm and final image size of 768 × 576 pixels. On average, three to five high-quality images of Bowman's layer were used for investigation in all patients and control subjects. This layer is of particular relevance for defining neuropathic changes because it is the location of the main nerve plexus that supplies the overlying corneal epithelium. The investigator (M.T.) who examined the cornea with the confocal microscope and who obtained the morphometric measurements of the images was masked with respect to the severity of neuropathy of the diabetic patients. By using the images collected, the following three parameters were quantified to define CNF damage and repair: 1) corneal nerve fiber density (CNFD), the total number of major nerves per square millimeter of corneal tissue; 2) corneal nerve fiber length (CNFL), the total length of all nerve fibers and branches (millimeters per square millimeter) of corneal tissue; and 3) corneal nerve branch density (CNBD), the number of branches emanating from major nerve trunks per square millimeter of corneal tissue. Measures 1 and 3 were determined using morphometric software incorporated within the Tomey instrument. Measure 2 was determined using third-party image analysis software (Scion Image for Windows; Scion, Frederick, MD).

To estimate the error in measuring CNFD, CNFL, and CNBD, we acquired images and determined each of these parameters in 15 subjects on two occasions separated by at least 48 h. The coefficients of variation for these parameters were 12% for CNFD, 9% for CNFL, and 24% for CNBD.

A 3-mm punch skin biopsy was taken from the dorsum of the foot, ∼2 cm above the second metatarsal head, with the use of 1% lidocaine local anesthesia. The biopsy site was closed using Steri-strips, and the specimen was immediately fixed in PBS-buffered 4% paraformaldehyde. After 18–24 h, it was rinsed in Tris-buffered saline and soaked in 33% sucrose (2–4 h) for cryoprotection. It was then embedded in OCT (optimum cutting temperature embedding compound), rapidly frozen in liquid nitrogen, and cut into 50-μm sections using a cryostat (model OTF; Bright Instruments, Huntington, U.K.). Four floating sections per subject were subjected to melanin bleaching (0.25% KMnO₄ for 15 min followed by 5% oxalic acid for 3 min), a 4-h protein block with a Tris-buffered saline solution of 5% normal swine serum, 0.5% powdered milk, and 1% Triton X-100, and overnight incubation with 1:1,200 Biogenesis polyclonal rabbit anti-human PGP9.5 antibody (Serotec, Oxford, U.K.). Biotinylated swine anti-rabbit secondary antibody (1:300; DakoCytomation, Ely, U.K.) was then applied for 1 h; sections were quenched with 1% H₂O₂ in 30% MeOH-PBS (30 min) before a 1-h incubation with 1:500 horseradish peroxidase–Streptavidin (Vector Laboratories, Peterborough, U.K.). Nerve fibers were demonstrated using 3,3′-diaminobenzidine chromogen (Sigma-Aldrich, Manchester, U.K.). Sections were mildly counterstained with eosin to better localize the basement membrane and hence the nerve fibers passing through the basement membrane. Negative controls consisted of replacing the anti-PGP9.5 antibody with rabbit immunoglobulin (DakoCytomation) at a concentration matching that of the primary antibody and showed no immunostaining. Basement membrane length (micrometers), average nerve fiber length (micrometers), and number of branches were measured using computer image analysis (Nikon digital camera and Leica QWin Standard V2.4 program). Intraepidermal nerve fiber linear density (IENFD), i.e., the number of fibers per millimeter of basement membrane, was expressed as number per millimeter, intraepidermal nerve fiber branch density (IENFBD), the number of branches in a selected area of epidermis, was expressed as number per square millimeter, and intraepidermal nerve fiber length (IENFL), the average length of epidermal nerve fibers in a given specimen, was expressed as micrometers.

Statistical analysis was performed using SPSS 14.0 for Windows (SPSS, Chicago, IL). The data are presented as means ± SE. Skewed data underwent logarithmic transformation before ANOVA and data without homogenous variance underwent nonparametric ANOVA (Kruskal-Wallis). Post hoc analysis for multiple comparisons between groups was performed using the Tukey test (parametric) or Dunnett T3 test (nonparametric). Correlations were performed using the Spearman rank test and expressed as a coefficient (r_s) with level of significance. For comparison between patients with and without neuropathic pain, the unpaired t test or the Mann-Whitney test was used. P < 0.05 was considered statistically significant.

Table 1 shows the clinical and demographic data with measures of clinical neuropathy in control subjects and diabetic patients matched for age, duration of diabetes, and glycemic control. Fifty-four diabetic patients (no [n = 10], mild [n = 18], moderate [n = 15], and severe [n = 11] neuropathy) were compared with 15 control subjects using ANOVA. SNOL increased (P < 0.01) and SNAP (P < 0.001) and SNCV (P < 0.01) decreased significantly with increasing neuropathic severity. Similarly, PNOL and TNOL (respectively, P < 0.01 and P < 0.001) and PNFL and TNFL (respectively, P < 0.001 and P < 0.001) increased, and PNAP and TNAP (respectively, P < 0.001 and P < 0.001) and PNCV and TNCV (respectively, P < 0.001 and P < 0.001) decreased with increasing neuropathic severity. The CDT (P < 0.001) and minimal (P < 0.05) and intermediate (P < 0.05) pain thresholds increased with increasing neuropathic severity. Autonomic function (DB-HRV) demonstrated a progressive decrease with increasing neuropathic severity (P < 0.001).

IENF loss was demonstrated in diabetic patients compared with control subjects (Table 2 and Fig. 1a and b). IENFD (ANOVA P < 0.0001) showed a progressive reduction with increasing neuropathic severity, which was significant in diabetic patients with mild (P < 0.05), moderate (P < 0.05), and severe (P < 0.01) neuropathy compared with control subjects using post hoc analysis (Fig. 2a). IENFBD (ANOVA P < 0.001) also showed a progressive reduction with increasing neuropathic severity, which was significant for mild (post hoc P < 0.05), moderate (P < 0.01), and severe (P < 0.01) neuropathy compared with that in control subjects (Fig. 2b). IENFL (ANOVA P < 0.05) only showed a significant reduction in patients with severe neuropathy (post hoc P < 0.05) compared with control subjects (Fig. 2c).

CNF loss was also demonstrated in diabetic patients compared with control subjects (Table 2 and Fig. 3a and b). CNFD (ANOVA P < 0.001) showed a progressive reduction with increasing neuropathic severity, which was significant in patients with mild (P < 0.01), moderate (P < 0.05), and severe (P < 0.01) neuropathy compared with control subjects using post hoc analysis (Fig. 4a). CNFBD was significantly reduced (ANOVA P < 0.001) with increasing neuropathic severity, but the reduction was significant even in patients without neuropathy (P < 0.001) and in those with mild, moderate, and severe (P < 0.001) neuropathy compared with control subjects using post hoc analysis (Fig. 4b). CNFL did not differ between diabetic patients and control subjects (ANOVA NS) (Fig. 4c).

IENFD correlated with CNFD (r_s = 0.385, P = 0.001) and IENFBD correlated with CNFBD (r_s = 0.416, P = 0.001). IENFL did not correlate with CNFL.

Both IENFD and IENFBD correlated with NDS, electrophysiology of the peroneal, tibial, and sural nerves, and thermal perception thresholds for cold and heat (Table 4). IENFL correlated with NDS, electrophysiology of peroneal and sural nerves, and cold perception threshold only (Table 4). CCM demonstrated fewer correlations but CNFD correlated with NDS, thermal perception, and DB-HRV and CNFBD correlated with SNOL. There was no correlation between CNFL and any measure of neuropathic severity (Table 4).

As both CNF and IENF pathological changes assess small fiber damage, we specifically assessed changes in CNF and IENF pathology with the degree of small fiber dysfunction. Subjects were stratified for the severity of small fiber deficits in accordance with the number of abnormal test results from CDT, HP-VAS 0.5, HP-VAS 5.0, and DB-HRV. Of the patients, 30% had normal small fiber function, 28% had mild dysfunction, 25% had moderate dysfunction, and 17% had severe dysfunction.

Post hoc analysis of IENFD demonstrated a reduction in diabetic patients even without small fiber dysfunction and also in patients with mild, moderate, and severe small fiber dysfunction (P < 0.001 for all) compared with control subjects (Table 3). IENFBD was also significantly reduced in diabetic patients without (P < 0.05) and with mild and severe (P < 0.01) dysfunction compared with control subjects. IENFL did not differ between groups (NS).

CNFD was significantly reduced in diabetic patients without (P < 0.05), and with mild (P < 0.05) and severe (P < 0.01) small fiber dysfunction using post hoc analysis (Table 3). With post hoc analysis, CNFBD was reduced in diabetic patients without (P < 0.01) and with mild (P < 0.01), moderate (P < 0.01), and severe (P < 0.001) small fiber dysfunction compared with control subjects. CNFL did not differ significantly between groups (NS).

As might be expected with regard to the relationship between small fiber function and structure (Table 4), the CDT correlated inversely with IENFD (r_s = −0.466, P < 0.001) and IENFBD (r_s = −0.408, P < 0.01). The HP-VAS 0.5 also correlated with IENFD (r_s = −0.311, P < 0.05). The CDT showed an inverse correlation with CNFD (r_s = −0.399, P < 0.01) but not with CNFBD (NS). The HP-VAS 0.5 correlated inversely with CNFD (r_s = −0.291, P < 0.05). Heart rate variability correlated with CNFD (r_s = 0.348, P < 0.05) but not IENFD (NS).

Of the patients, 38 (72%) had a positive diabetic neuropathy symptom score (≥1 of 4). Corneal and intraepidermal nerve fiber pathological changes were compared between diabetic patients stratified into those with (visual analog score ≥3 of 10) (n = 28) and without (n = 26) painful neuropathy. There was no significant difference in IENFD (painful 4.8 ± 0.7 vs. painless 6.3 ± 0.7, NS) and IENFBD (painful 27.5 ± 5.7 vs. painless 46.6 ± 9.4, NS). IENFL was significantly reduced in those with painful (25.3 ± 2.8) compared with painless (35.9 ± 3.3) neuropathy (P < 0.05). There was no significant difference for CNFD (painful 23.0 ± 2.3 vs. painless 23.1 ± 2.4, NS) and CNBD (painful 6.5 ± 1.1 vs. painless 8.52 ± 1.7, NS), but there was a significant reduction in CNFL in painful (3.3 ± 0.4) compared with painless (4.6 ± 0.5) neuropathy (P < 0.05).

The National Institutes of Health has deemed the development of a surrogate end point for diabetic neuropathy as a priority area for the complications of diabetes. The main aim of establishing a surrogate marker is to define those at risk of developing a complication, anticipate deterioration, and assess the efficacy of new therapies. Although electrophysiology, QST, and assessment of neurological disability are advocated to define neuropathic severity (1), there is debate as to which test, if any, is appropriate in the early stages of neuropathy (25) and also whether the same tests should be used to define therapeutic efficacy in clinical intervention trials (5).

Regulatory authorities favor clinically relevant surrogate end points. Electrophysiology and assessment of vibration perception threshold quantify large fiber deficits, yet the earliest nerve fibers to undergo damage (8,10,11) and subsequent repair (12) are the small unmyelinated fibers, as evidenced by studies in patients with impaired glucose tolerance (8) and diabetic patients with minimal neuropathy (7). Although functional measures of small fiber damage include assessment of thermal perception thresholds, these tests show considerable variability and are not considered to be sufficiently robust to be included as major end points for clinical trials (1). Sural nerve (6,7) and skin biopsies (8,9) directly quantify small nerve fiber damage and repair and have been proposed as surrogate markers for human diabetic neuropathy, but both are invasive procedures. As an alternative noninvasive test, we (20–22) and others (19) have shown that CCM can also quantify small fiber pathological changes and stratify the severity of somatic neuropathy in diabetic patients. The present study compared in detail CCM and skin biopsy to quantify small fiber pathological changes in diabetic patients stratified in accordance with the severity of neuropathy using established clinical, electrophysiological, and QST techniques.

Skin biopsy is a reliable and reproducible way of quantifying small nerve fiber pathological changes and has been advocated for diagnosing small fiber neuropathy from a variety of causes (14,15), including painful neuropathy due to impaired glucose tolerance (8). Furthermore, in a recent study evaluating the therapeutic benefits of improving multiple risk factors on nerve damage in patients with impaired glucose tolerance, IENF pathology, as opposed to electrophysiology and QST, was the most sensitive measure of neuropathy change over 1 year (12), although, it is interesting that a recent study was unable to show a significant difference in IENFD between diabetic patients with and without neuropathy (26). An additional morphological feature of IENF that has been identified recently is terminal axonal swelling, which occurs with greater frequency in those with paresthesias and also predicts accelerated nerve fiber loss in patients with idiopathic painful neuropathy (27). To further define the ability of IENF assessment to diagnose and assess progression of human diabetic neuropathy and to compare it with corneal nerve pathology (20), we have additionally quantified IENFL and IENFBD.

We evaluated IENF pathology in relation to increasing neuropathic severity. As expected IENFD on the dorsum of the foot, a distal site, was slightly lower than that for other studies in the leg (13,28) and thigh (26). However, neither IENFD, IENFBD, nor IENFL was significantly reduced in diabetic patients without neuropathy compared with control subjects. This result is in keeping with recent observations in diabetic patients without neuropathy assessed using conventional measures of neuropathic severity (26) and suggests that IENF pathological changes may not be an early marker of nerve damage in diabetic patients. This finding is, of course, in contrast with several studies that have demonstrated a reduction in IENF density in patients with impaired glucose tolerance, but it must be remembered that these patients also had painful neuropathy (12). Indeed, it may be that pain is related to loss of IENFs as shown in a recent study in which diabetic patients with painful neuropathy had a significant reduction in IENFD compared with diabetic patients with painless neuropathy (29). Several studies have attempted to address whether there is a relationship between the severity of neuropathy and the degree of IENF loss. In one of the earliest studies to define IENF abnormalities in diabetic patients, Kennedy et al. (30) showed a marked reduction in IENFD and related it to the severity of neuropathy. However, it is important to note that all patients had severe neuropathy as they were undergoing pancreas transplantation. Similarly, a recent study has also demonstrated an association between IENFD and vibration perception threshold; however, the majority of patients in that study also had moderate to severe neuropathy (29). Another study has shown a correlation between loss of IENF and the severity of neuropathy assessed using electrophysiology and QST (28). In the present study, we demonstrate a significant reduction in IENFD and IENFBD in patients with mild neuropathy, which progressed with increasing neuropathic severity. IENFL was only reduced in those with severe neuropathy.

With regard to CNF pathology, whereas CNFD was reduced in diabetic patients with mild neuropathy and progressively worsened with increasing neuropathic severity, CNFBD was significantly reduced in patients even without evidence of neuropathy and showed a progressive reduction with increasing neuropathic severity. Although there was a trend for a reduction in CNFL, this was not significant. One may interpret this reduction as a possible compensatory elongation of remaining fibers; however, we believe that the increased variability in this measure also contributed to the lack of significant difference. The finding that corneal nerve pathology is related to neuropathic severity is in keeping with our previously published study (21) and other data from type 1 diabetic patients in whom a reduction in the number of CNF bundles correlated significantly with the severity of somatic neuropathy (19). A recent study has also shown that CNF loss is associated with the severity of retinopathy in diabetic patients (31).

Although CNF pathology showed fewer correlations to electrophysiology than IENF pathology, it did correlate with the clinically relevant measure of neuropathic severity, namely the NDS. Moreover, it also showed a correlation with autonomic dysfunction. In the present study when we stratified the diabetic patients in accordance with the severity of small fiber dysfunction, both CNF and IENF showed abnormal pathology even in patients without apparent small fiber dysfunction and in those with mild, moderate, and severe dysfunction. Furthermore, we also showed an association between CNFD and IENFD and both cold and heat as pain detection thresholds. A previous study has demonstrated an association between IENFD and cold but not warm detection thresholds (29). Shun et al. (28) have also shown an association between IENFD and both cold and warm detection thresholds.

Painful diabetic neuropathy is debilitating, yet both its etiology and treatment are not clear. As pain is mediated via the C and Aδ fibers, it has been assumed that more specific testing of small fiber dysfunction or damage may help to differentiate patients with and without painful neuropathy. As IENFs and CNFs are C fibers, both techniques provide an ideal opportunity to study in detail any changes, which may occur in painful diabetic neuropathy. Several studies assessing small fiber dysfunction (32,33) and sural nerve unmyelinated nerve fiber pathology (6,34) have failed to show any difference between diabetic patients with and without painful diabetic neuropathy. However, in the present study we show that both corneal and intraepidermal nerve fiber length were significantly reduced in patients with painful compared with painless neuropathy. Sorensen et al. (29) have previously demonstrated a reduction in intraepidermal nerve fiber density in diabetic patients with painful neuropathy. Furthermore, in a lifestyle intervention study in patients with impaired glucose tolerance, an improvement in painful neuropathic symptoms was associated with an improvement in IENF density (12). These data support the contention that small nerve fiber damage may be associated with pain, and this novel finding merits further study.

In conclusion, an ideal surrogate marker for diabetic neuropathy should be easy to use, reliable, sensitive, and noninvasive to enable repeated assessment as often or as long as necessary to define progression or response to therapeutic intervention. We have demonstrated that whereas both techniques of CNF and IENF assessment accurately reflect the severity of somatic neuropathy, CCM provides a significant further advantage as it detects damage before detectable nerve dysfunction, but most importantly it is entirely noninvasive.

This work was supported by grants from the National Institutes of Health (R01 NS46259-01), Juvenile Diabetes Research Foundation International (5-2002-185), and Diabetes UK (RD03/0002624).