Refined Quantitation of Sweat Gland Innervation

Abstract

Skin biopsies have gained increasing popularity as a tool to evaluate disorders affecting small nerve fibers. While reports on sweat gland nerve fiber density (SGNFD) to quantitate sudomotor innervation have been promising, methodologies vary significantly. Although conventional stereology is commonly used, no standard technique has been established. We sought to develop an accurate and reproducible technique to quantify SGNFD. Skin punch biopsies from healthy individuals were cut and stained. Images of sweat glands (SGs) were acquired using confocal and widefield microscopes, and optimized using deconvolution. Nerve fibers were reconstructed and nerve fiber length (NFL) was quantified using three-dimensional (3D) automated software. SGNFD was obtained by dividing NFL by SG volume. SGNFD was also assessed using stereology for comparison. Ninety-two SGs from 10 healthy subjects were analyzed by independent observers. Using confocal microscopy, the software reliably traced nerve fibers. In contrast, rendering of nerve fibers was inferior using widefield microscopy. Interobserver reliability was suboptimal using widefield images compared to confocal (ICC = 0.82 vs ICC = 0.98). Correlation between 3D-reconstruction and stereology was poor (ICC = 0.38). The newly developed technique of SGNFD quantitation using 3D reconstruction of SG innervation with confocal microscopy reliably traces nerve fibers, shows outstanding reproducibility, is almost completely unbiased, and superior to conventional stereology methods.

INTRODUCTION

Peripheral neuropathy (PN) is a common complication of many medical conditions, including diabetes mellitus, chronic renal failure, monoclonal gammopathies, paraneoplastic syndromes, rheumatologic disorders, and chemotherapy (1–4). Small nerve fibers are often early or selectively involved in PN (5, 6). These fibers mediate somatic sensory (pain, temperature) and autonomic functions via thinly myelinated Aδ and unmyelinated C fibers and are responsible for a considerable portion of the morbidity associated with PN.

The histologic assessment of small nerve fibers in skin has gained increasing popularity in recent years owing to the ease of access via skin punch biopsies on the one hand, and its diagnostic value in the diagnosis of small fiber neuropathies on the other hand. The main focus has been the quantitation of intraepidermal nerve fiber density (IENFD), and a series of reports have demonstrated its value in the diagnosis of painful small fiber sensory neuropathies (7–12). Consensus panels have accepted the assessment of IENFD as an objective and valid indication of small sensory nerve fiber involvement in PN (1, 5, 13).

Impairment of autonomic function may be the earliest manifestation of small fiber involvement in PN; impairment of sudomotor function in particular has been shown to be one of the most sensitive markers identifying autonomic involvement in PN (14–16). Thus, it comes as no surprise that there has been increasing interest in quantifying autonomic innervation of dermal structures, particularly of sweat glands (SGs), structures densely innervated by autonomic nerve fibers. A number of groups have reported on sweat gland nerve fiber density (SGNFD) as a means to quantitate sudomotor innervation, but the complex, three-dimensional (3D) shape of SGs has made accurate and reproducible quantitation of SG innervation challenging. This is reflected in the wide variation of reported methodologies from imaging to analyzing techniques. While some rely on widefield microscopy for image acquisition, others use confocal microscopy. Image acquisition is not standardized and some use 2D, while others use 3D acquisition modalities. Although conventional stereology techniques are most commonly used, to date no standard technique has been established (17–21). In an attempt to bridge this gap, we sought to explore and compare different imaging and quantitation techniques with the goal to develop a gold standard that allows for accurate and reproducible quantification of SGNFD.

MATERIALS AND METHODS

Subjects

Participants were healthy controls recruited from the communities of southeastern Minnesota. Subjects were male and female, ages 18–80 years, with normal autonomic reflex screen, no symptoms suggestive of PN or autonomic dysfunction, and normal neurologic exam. The study was approved by the Mayo Clinic Institutional Review Board, and written informed consent was obtained from all participants.

Immunohistochemistry

Skin biopsies (4 mm) were obtained from the distal leg, 5 cm proximal to the medial malleolus. Tissue was immediately fixed in Zamboni’s fixative (Newcomer Supply, Middleton, WI) for 24 hours at 4°C. Biopsies were then immersed in 20% sucrose in phosphate buffered saline (PBS) for at least 24 hours at 4°C. They were then frozen on dry ice and stored at −80°C. Sections were cut (50 µm) with the epidermis perpendicular to the knife on a freezing microtome (MICROM International GmbH, Walldorf, Germany), and placed in 96-well Elisa plates filled with antifreeze solution. Every other section after the 10th section was taken for immunohistochemistry (n = 10 per case) and placed in PYREX 9-well spot plates filled with 0.1 PBS with 0.1% Triton X-100. Tissue was blocked with 10% normal donkey serum (NDS) in 0.1 PBS with 0.1% Triton X-100 for 30 minutes with gentle rotation. NDS was drawn off with a glass pipette avoiding contact with sections.

Primary antibodies utilized were PGP 9.5 (1:800, rabbit polyclonal, DAKO, Santa Clara, CA) and Collagen IV (1:800, mouse monoclonal, EMD Millipore, Temecula, CA), diluted in PBS with 0.5% NDS/0.3% Triton X-100. Plates were covered with seal plate film and incubated overnight at 4°C with gentle rotation. Sections were washed in 0.1 PBS with 0.1% Triton X-100 3 times for 5 minutes with gentle rotation. Secondaries were diluted in 0.5% NDS/0.3% Triton X-100, Cy3 (1:300, donkey anti rabbit, Jackson ImmunoResearch, West Grove, PA) and Cy2 (1:200, donkey anti mouse, Jackson ImmunoResearch) and passed through a 0.2-µm syringe filter. Sections were incubated in the dark for 60 minutes at room temperature with gentle rotation and washed in 0.1 PBS with 0.1% Triton X-100 3 times for 5 minutes.

Sections were mounted in 1.35% (w/v) noble agar (heated to 60–65°C) on a slide warmer (60–65°C). Tissues were rinsed briefly in an extra pool of agar to remove any salts and to coat in agar. Solidified mount was then placed directly in 95% ethanol for 30 minutes, followed by 30 minutes in 100%. Tissue was then placed in methyl salicylate for at least 30 minutes and cover slipped with DPX mounting medium (Fluka, Roanoke, VA).

Imaging

Tissue sections were exhaustively searched for SGs. SGs were identified by visualization of Collagen IV staining tubular basement membranes in association with PGP 9.5 antibodies signaling nerve fibers (Fig. 1A, B).

FIGURE 1.

3D reconstruction of sweat gland (SG) nerve fibers. Confocal image stacks of SGs were taken from the distal leg of a control subject. Nerve fibers were immunostained with PGP 9.5 (A) and SG tubules with Col IV. (B) Seeds were automatically placed throughout nerve fiber structures. (C) Trees were traced reclusively in x, y, and z planes throughout the stack. (D) Scale bar = 50 μm.

SGs were initially examined under widefield microscopy (Zeiss Axioimager A1; Zeiss, Thornwood, NY) and 3D Z-stacks acquired at ×20 magnification at 1-μm intervals using a CCD color video camera (MicroFire; Optronics, Goleta, CA), and Stereo Investigator software (v2017.03.2; MBF Bioscience, Williston, VT).

In addition, all tissue sections were imaged using an inverted LSM 780 confocal microscope (Zeiss) and 3D Z-stacks acquired at 1-μm intervals using a ×20 magnification (Fig. 1).

Images were deconvolved to reduce nonspecific background staining and increase resolution (Huygens Professional version 18.04; Scientific Volume Imaging, Hilversum, The Netherlands, http://svi.nl).

Volume Estimation

The 3D nature of SGs requires that nerve fiber length (NFL) be normalized by volume. We used the contouring function in Stereo Investigator to trace the region of interest at the topmost Z-section and bottom Z-section. SG volume was defined as the area within the contours.

Morphometric 3D Reconstruction

Nerve fibers were reconstructed using Neuro Lucida 360 (NL360) (MBF Bioscience), a software platform for reconstruction of neuronal morphology. We used the automated algorithm to trace nerve fibers from 3D (volumetric) Z-stacks. This method locates an initial point, places seeds in linear branched structures and then exploits local image properties to trace the structures recursively in x, y, and z planes (22). Seeds were placed at 90% sensitivity and validated using the refine filter setting at levels 1–2. Additional modifications were made in edit mode to remove seeds placed outside SG structures, often related to nonspecific lipofuscin staining. NFL was quantified using NeuroExplorer (MBF Bioscience), and SGNFD was calculated by dividing NFL by SG volume (Fig. 1C, D).

Stereology

All images were additionally analyzed using the stereologic probe, Spaceballs (MBF Bioscience) (23). In this stereological technique a 3D hemi-sphere is superimposed upon the image stack. Sampling grids were selected to optimize counting sites (200–300 fiber intersections). This methodology returns an estimate of length per region. SGNFD was calculated in the same manner, by dividing NFL by volume.

Statistical Analysis

All analysis was completed twice by 2-independent observers using widefield microscopy (Zeiss Axioimager A1; Zeiss) and confocal microscopy (LSM 780; Zeiss).

Intra-class correlation coefficients (ICCs) were used to assess reliability within and between observers. t-Test for paired data was used to analyze group differences in volume measurements (24). Analysis was performed using SAS software version 9.4 (Cary, NC).

RESULTS

Subjects

Skin biopsies of 10 healthy control subjects (6 male, 4 female) were analyzed. Age range was 18–80 years, with a median of 29. A total of 92 SGs were imaged with both confocal and widefield microscopy. The number of SGs per biopsy ranged from 5 to 14, with a median of 9. SGNFD measurements were performed twice by 2-independent observers.

Microscopy Techniques and 3D Morphometric Reconstruction

Confocal images were superior compared to those from widefield microscopy. Widefield images lacked resolution and displayed significant background noise. Punctate-like structures, unrepresentative of nerve fibers were consistently present throughout widefield image stacks. In contrast, confocal images showed higher resolution and had minimal background noise. Thus, nerve fibers were more readily identifiable with confocal microscopy (Figs. 2 and 3).

FIGURE 2.

Comparison of images acquired with widefield microscopy and confocal microscopy. (A) Widefield images display significant background noise and punctate-like structures rendering differentiation between true nerve signal and artifact challenging. (B) Confocal images show readily distinguishable nerve fiber structures and block out background noise and artifact. Scale bar = 50 μm.

FIGURE 3.

Comparison of images acquired with widefield microscopy (A) and confocal microscopy. (B) Nerve fibers are more easily discernible in confocal images compared to widefield. Scale bar = 50 μm.

Using confocal microscopy, the software reliably traced nerve fibers with minimal operator modifications or corrections required (Fig. 4). In contrast, rendering of nerve fibers was notably inferior when using widefield image stacks, often raising uncertainty about true nerve signal versus artifact (Fig. 4). As a result, the automated analysis required significantly more operator input and adjustments to analysis settings. ICC between confocal and widefield microscopy was 0.51 (CI 0.29–0.68) for observer 1 and 0.65 (CI 0.46–0.78) for observer 2. Inter-rater reliability was ICC = 0.98 (CI 0.97–0.99) for confocal images compared to ICC = 0.82 (CI 0.75–0.88) using widefield images.

FIGURE 4.

Comparison of 3D morphometric reconstruction of nerve fibers using widefield and confocal images. (A) In the widefield image, nonspecific background staining results in erroneous identification of fibers. (B) In contrast, in the confocal image, nerve fibers are properly identified without detection of artifacts due to background noise. Arrows illustrate characteristic erroneous fiber detection in the widefield image which is absent in the confocal image. Scale bar = 50 μm.

3D Morphometric Reconstruction Versus Stereology

To compare performance among quantitative techniques, we correlated SGNFDs derived with confocal microscopy using the 3D automated reconstruction software and the commonly used gird-matrix stereology probe. Correlation between techniques was poor, ICC = 0.38 (−0.64 to 0.676).

Volume Assessment

Volume estimates were assessed by measuring the area within contours for every individual SG. Volumes acquired from widefield images were spuriously 30% larger compared to those from confocal images (p < 0.001).

DISCUSSION

We introduce a novel technique of SGNFD quantitation using confocal microscopy, deconvolution algorithms, and 3D reconstruction of SG innervation. This technique allows for almost completely unbiased and reliable tracing of SG nerve fibers and quantitation of innervation with outstanding reproducibility between observers. While comparison between techniques is arbitrary when a gold standard is not available, the use of confocal microscopy and visual verification of automated fiber detection in a 3D matrix deems this methodology superior to conventional stereology techniques and renders those inadequate given the poor correlation with the here described approach.

Objective quantification of autonomic innervation would be a welcome tool in the diagnosis, characterization, and prognostication of PN and may also help characterize and differentiate neurodegenerative disorders with autonomic involvement, including Parkinson disease, dementia with Lewy bodies, multiple system atrophy, and pure autonomic failure. However, reliable and unbiased quantitative techniques are a prerequisite for clinical utility.

Choosing appropriate imaging modalities for quantitative analysis is a crucial first step. The ideal light microscopy technique optimizes both image resolution and imaging speed. Previous approaches have ranged from widefield to laser-scanning confocal microscopy (17, 20, 21, 23). However, to date, no consensus has been reached, and there is a need for optimization and standardization.

It is known that with widefield microscopy, specimens are subject to significant background noise. As the focal plane is moved throughout a sample and an image comes into focus, surrounding structures become blurry. This translates into reduced resolution and inability to distinguish fine structures such as nerve fibers. Widefield images of SG innervation, in our experience, proved to lack resolution and detail, and often raised questions about true nerve signal versus artifact. 3D automated reconstruction using widefield image stacks was often inaccurate and nerve recognition imprecise. Trees were often traced over unspecific staining, requiring more operator input and adjustments to analysis parameters (25, 26).

Laser-scanning confocal microscopy creates higher resolution images and increased contrast by illuminating samples one point at a time and allowing only in-focus fluorescence through a pinhole. This process blocks out-of-focus light, resulting in better defined images (27, 28). In our experience, confocal images rendered the necessary detail to detect linear, more continuous projections corresponding to nerve fibers and enhanced the ability to distinguish nerve fibers from artifact. This translated into accurate detection of nerve fibers by the software with minimal operator input.

SGNFD measurements were close to identical between raters when using confocal image stacks. Although reliability assessment between raters was reasonably good with widefield images, it was clearly inferior when compared to that of confocal image stacks. Furthermore, correlation between widefield and confocal images was poor. These findings highlight the discordance between the 2 imaging techniques and underline the advantages of confocal images.

Another important determinant of 3D quantitative analysis of SGNFD is volume estimation. From our experience and supported by the literature, confocal microscopy has the ability to properly discriminate the Z-dimension, also known as the axial plane in 3D space. While x and y axes correspond to horizontal and vertical planes, the z axis provides information on depth of field and volume of a specimen. Instead, widefield microscopy due to its overt light dispersion makes it difficult to determine where fluorescence from a sample originates in the Z-plane and therefore makes it challenging to estimate volume (28). Volumes derived from widefield images were systematically larger than those from confocal images, suggesting overestimation of SG volumes with widefield microscopy.

Software systems used to improve image quality have also varied among studies in the past. While the unsharp mask filter technique works by removing the blur in the images, deconvolution algorithms reconstruct the true intensity distribution of the specimen from the observed data and represents evolution of imaging techniques (27). This technique has already been widely applied in quantitative image analysis and demonstrated to improve morphologic characterization (25, 28).

The aforementioned steps were crucial in the process of obtaining optimal images for analysis. The next critical step was optimization of the quantitation methodology. Donadio and colleagues described qualitative techniques introducing subjective scoring systems (29). These techniques gave important insight into dermal autonomic innervation, but their subjective nature limited more widespread use. Gibbons and colleagues were among the first to design a manual technique to quantify SGNFD (23). This technique showed good reproducibility but was labor intensive. The same group later developed a faster, automated technique that quantified highlighted pixels. Both techniques were validated against stereology with good correlations, but all these techniques resulted merely in estimates of the actual length of nerve fibers surrounding SGs (17).

The analysis technique described here allows for complete 3D morphometric reconstruction of SG nerve fibers surrounding SGs. This approach is arguably the most complete approach at quantitating SG innervation using morphometric 3D reconstruction (NL360). This methodology detects and quantifies SGNFD using an almost completely automated process requiring minimal operator adjustments, minimizing bias and the need for investigator judgment. Surprisingly, stereology did not correlate well with the 3D-reconstruction method and showed consistently lower densities, likely due to underestimation of the actual fiber length. Stereology furthermore requires manual selection of pixels representing nerve fibers, while the automated method relies on predetermined parameters that consider voxel clustering, spine shapes and dimensions to reconstruct nerve fibers in a 3D stack, making quantitation more precise and reliable (30, 31). Our findings showed close to perfect correlation between observers.

However, in spite of careful exploration and much consideration of widefield microscopy, we conclude that confocal microscopy is a requirement for proper quantitative analysis of SGNFD. Image quality is further improved with the use of deconvolution algorithms, which is another key step in the process.

One of the areas of consistent concern among these methodologies has been that of time investment (17, 23, 32). We found that morphometric analysis ranged from 1 to 7 minutes (smallest to largest gland) per gland compared to 30 to 45 minutes with stereology, making the described morphometric analysis also the faster, more efficient approach.

We acknowledge limitations of this study. First, since there was no established gold standard for SGNFD quantitation, we were guided by visual verification of the nerve fiber detection and delineation by the morphometric reconstruction process. Second, our data were derived from a relatively small number of subjects, though a large number of individual SGs were analyzed. Finally, the equipment and software required to utilize the described technique will likely limit its use to selected centers, at least at the moment. The process is overall, however, not more time-consuming than currently utilized protocols, so could eventually well be applied in clinical practice.

In conclusion, after careful exploration of imaging techniques and quantitation methodologies for SGNFD assessment, we present a novel, refined approach at quantifying SGNFD that optimizes image acquisition and applies unbiased and reliable quantitation methodology. We believe the data stress the importance of reaching consensus on techniques to quantify dermal autonomic innervation, and support the described novel approach as current gold standard of SGNFD quantitation. Ongoing further development steps are the establishment of normative data and exploration of findings in disease.