Use of Serum Protein Measurements as Biomarkers that Can Predict the Outcome of Diabetic Foot Ulceration

Abstract

Objectives:

To identify proteins that are prognostic for diabetic foot ulcer (DFU) healing and may serve as biomarkers for its management, serum samples were analyzed from diabetic mellitus (DM) patients.

Approach:

The serum specimens that were evaluated in this study were obtained from DM patients with DFU who participated in a prospective study and were seen biweekly until they healed their ulcer or the exit visit at 12 weeks. The group was divided into Healers (who healed their DFU during the study) and Non-Healers.

Results:

Interleukin (IL)-10, IL-4, IL-5, IL-6, and IL-13 and interferon-gamma were higher in the Healers while Fractalkine, IL-8, and TNFα were higher in the Non-Healers. The trajectory of IL-10 levels remained stable over time within and across groups, resulting in a strong prognostic ability for the prospective DFU healing course. Classification and Regression Tree analysis created an 11-node decision tree with healing status as the categorical response.

Innovation:

Consecutive measurements of proteins associated with wound healing can identify biomarkers that can predict DFU healing over a 12-week period. IL-10 was the strongest candidate for prediction.

Conclusion:

Measurement of serum proteins can serve as a successful strategy in guiding clinical management of DFU. The data also indicate likely superior performance of building a multiprotein biomarker score instead of relying on single biomarkers.

Aristidis Veves, MD, DSc

INTRODUCTION

Diabetic foot problems represent one of the major diabetes complications that significantly impair the quality of life of people with diabetes and lead to more than 750,000 new diabetic foot ulcers (DFUs) and 70,000 lower extremity amputations per year in the United States alone.¹ Of great concern, after an initial drop, a resurgence of amputations has been noted in the last decade.² The total medical cost for the management of DFU in the United States is ∼11 billion dollars on top of the cost for management of diabetes alone.³ Given the unabated rise of the diabetes pandemic, the number and cost of DFU are bound to intensify in the future.

There are currently four products that have been licensed by the Food and Drug Administration for the treatment of DFU: A recombinant growth factor, rhPDGF-BB, becaplermin,⁴ two bioengineered skin substitutes Apligraf⁵ and Dermagraft,⁶ and Integra, an acellular, bilayer matrix.⁷ Furthermore, a large number of additional wound care products are available, including human placental membrane products⁸ and advanced wound dressings.⁹ However, all the above treatments carry a high cost and should be reserved for patients who fail to respond to regular care that uses much less expensive wound dressings, such as moistened wound gauze.⁹

Early identification of the patients who are going to heal their ulcers with regular care has the potential to free considerable resources for patients who fail to do so and necessitate more intensive care that includes advanced wound care products. Evaluation of the change of ulcer size over a 4-week period was proposed more than 20 years ago as a strong predictor of healing¹⁰ and is currently the recommended method of choice.¹¹ However, techniques that are based on ulcer size alterations require sequential measurements over prolonged periods of time and simpler, more reliable and rapid methods are urgently needed. The National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) has already established the Diabetic Foot Consortium (DFC) that aims to develop and validate biomarkers for DFU healing that can be harnessed in clinical care and research.¹²

The main goal of the present study was to evaluate changes in serum proteins that are associated with DFU healing and identify serum proteins that can predict DFU healing and serve as biomarkers for its management. To this end, we analyzed serum specimens from a large cohort of DFU patients who were seen every 2 weeks and were followed up for a total of a 12-week period. We first evaluated longitudinal differences between Healers and Non-Healers and then investigated trajectories over time within the diabetic wound healing course. Finally, we used the Classification And Regression Tree (CART) predictive model to predict DFU healing based on a small set of the measured serum proteins. In additional comparative analyses, we explored the gene expression of our most striking findings in Healers and Non-Healers within the DFU single-cell RNA-seq atlas we have previously generated.

CLINICAL PROBLEM ADDRESSED

The development of DFUs is a devastating complication of diabetes that necessitates approaches to predict successful healing and hence facilitate with patient stratification and treatment selection.

MATERIALS AND METHODS

Study population

The serum specimens that were evaluated in this study were obtained from patients who participated in a prospective study, and its results have already been published.¹³ More specifically, we recruited 39 patients with diabetes mellitus (DM) and DFU. Subjects who were smokers, presented with severe peripheral arterial occlusive disease requiring intervention, with any illness that affects wound healing, history of myocardial ischemia, angina, severely increased albuminuria, or any other serious illness were excluded. All participants were recruited and followed for 12 weeks at the Joslin-Beth Israel Deaconess Foot Center (Fig. 1). They were seen fortnightly until they healed their ulcer or the exit visit at 12 weeks.

Figure 1.

Summary graphic illustration.

At the end of the study, subjects were divided into two groups: those who healed their DFU (Healers) (n = 15) or failed to heal (Non-Healers) (n = 24). Serum specimens were obtained at every visit for a maximum of six specimens per subject. Thirty-six (92%) subjects (15 Healers and 21 Non-Healers) had complete protein data at ≥3 visits and were included in the analysis (Table 1). The protocol was approved by the Beth Israel Deaconess Medical Center Institutional Review Board, and all subjects provided written informed consent. Electronic laboratory notebook was not used.

Table 1.

Clinical characteristics of the studied subjects

	DFU Patients
	Healers (n = 15)	Non-Healers (n = 21)	P
Age (years)	50 ± 9.0	59 ± 7.5	0.006
Gender (male)	12 (80%)	15 (71%)	NS
Type 1 diabetes	3 (20%)	1 (5%)	NS
DM duration (years)	19 ± 16	21 (5–25)	NS
Weight (kg)	109.9 ± 20.2	101.4 ± 16.8	NS
BMI (kg/m2)	33.6 ± 6.5	29.74 ± 8.3	NS
Blood pressure (mmHg)			NS
Systolic	132 ± 13	131 ± 19	NS
Diastolic	74 ± 9	70 ± 18	NS
Fasting glucose (mg/dL)	174 ± 103	134 ± 50	NS
HbA1c (%)	9.3 ± 2.3	8.2 ± 1.3	NS
Total cholesterol (mg/dL)	167 ± 36	164 ± 35	NS
LDL (mg/dL)	84 ± 25	86 ± 30	NS
HDL (mg/dL)	44 ± 17	45 ± 23	NS
Triglycerides (mg/dL)	195 ± 122	168 ± 83	NS
Ulcer size (mm)	201 ± 382	158 ± 256	NS
ABI	1.04 ± 0.14	0.98 ± 0.11	NS
NDS	16 ± 7	14 ± 7	NS
VPT	37 ± 18	38 ± 14	NS
SW	6.20 ± 0.65	6.20 ± 0.86	NS

Mean ± SD or frequency (%).

ABI, ankle-brachial pressure; BMI, body mass index; DFU, diabetic foot ulcer; HbA1c, glycated hemoglobin; HDL, high-density lipoprotein; LDL, low-density lipoprotein; NDS, neuropathy disability score; NS, not significant; SD, standard deviation; SW, Semmes-Weinstein monofilament; VPT, vibration perception threshold.

Serum cytokine and growth factors measurements

Serum protein measurements were completed using a Luminex Magpix^® (Austin, TX) apparatus and Millipore (Burlington, MA) multiplex panels. To reduce variability, all measurements were performed at the end of the study. Two plates were used for all measurements, and all serum specimens from each participant were in the same plate. Twenty-eight proteins were measured from each visit with the Luminex platform, resulting in nearly 5,500 data points. Missing values were imputed using last observation carried forward or next observation carried backward.

Gene expression of proteins of interest in various cell types

We leveraged our recently published single-cell transcriptomic atlas of DFUs to examine the differential expression of genes related to serum proteins of interest between DFU Healers and Non-Healers. The main findings of this study have been published.¹⁴ More specifically, we interrogated gene expression in discarded DFU tissue, forearm skin biopsies and peripheral blood mononuclear cells (PBMCs) from another group of Healers and Non-Healers DFU patients. Genes expressed in less than five cells were removed.

Statistical analysis

The Student's t-test was used to compare the clinical characteristics between the two groups. The General Linear Model was used to identify differences in the cytokines' and growth factors' repeated measurements during the six follow-up visits. Age, sex, and race were entered as covariates for adjusting the model. The CART algorithm was used to make prediction on a response variable (e.g., healing), based upon several explanatory variables (e.g., age, sex, race, and serum proteins). CART analysis is a very common modeling technique because it works well for a wide variety of data sets, is easy to interpret, and does not require model assumptions like distribution of the data. The term classification tree is used when the response variable is categorical like in our study (healing vs. nonhealing).

Internal nodes have exactly one incoming edge and then the data are split based upon a decision, that is, a mathematical rule to split the predicted response. All other nodes are called leaves or terminal nodes. To assess the accuracy of the CART model, we used a method to count misclassification rate. The misclassification rate is a straightforward measure of the accuracy of our model predictions. We used a cross-validation technique (e.g., training data vs. test data) to evaluate the performance of the CART model. We generated a confusion matrix to evaluate the accuracy of our final pruned model. A confusion matrix is a table that reports the number and percentage of true positives, false positives, true negatives, and false negatives of a classification model. The Minitab software (State College, PA) was used for the statistical analysis, and significance was defined for p < 0.05.

Single-cell data analysis

We reanalyzed our published data, publicly available on Gene Expression Omnibus with accession number GSE165816. Raw count matrices from CellRanger were imported in R using Seurat.¹⁵ Samples were depth normalized using SCTransform v2¹⁶ and annotated as previously described.¹⁴ Counts were recomputed using the minimum median Unique Molecular Identifier per tissue and markers per cell type were identified by subsetting each cell type and then performing a Wilcoxon test between Healers and Non-Healers using the Seurat FindAllMarkers function with a log fold change threshold of 0.25. To establish a signature of the combination of biomarkers, the average expression of each gene of the significant biomarkers was multiplied by the direction of expression (higher/lower in Healers). Significance was assessed between Healers and Non-Healers for signature expression using the t-test. Data were visualized using Seurat, pheatmap, and ggplot2.

RESULTS

The clinical characteristics of the participants are presented in Table 1. The results of serum protein measurements during the six visits are displayed in Supplementary Tables S1–S6. Repeated measurements analysis by using the General Linear Model and using age, sex, and race as covariates identified differences between Healers and Non-Healers in nine proteins while no differences were observed in the remaining nineteen. Six serum proteins tended to be higher in Healers during the six visits: interleukin (IL)-10, IL-4, IL-5, IL-6, and IL-13 and interferon-gamma (INF-γ) (Fig. 2). Three proteins tended to be higher in Non-Healers: Fractalkine, IL-8, and TNFα (Supplementary Fig. S1). Information for the remaining proteins is provided in Supplementary Table S7 and Supplementary Fig. S2. There was no statistically significant interaction between healing status and the visits.

Figure 2.

Scatter plots of six serum proteins that tended to be higher in the Healers during the observational period (median, 95% confidence interval).

Our search for the best predictors of DFU healing showed that over time the IL-10 trajectory remained stable within and across groups, and at baseline was strongly prognostic for the prospective DFU healing (β = 0.2, p = 0.03). Of note, utilizing measurements from more than one visit had a tendency to improve the model accuracy as reflected by an increase in c-statistics. The odds of healing per one tertile change of biomarker distribution on logistic regression analysis are presented in Table 2.

Table 2.

Circulating interleukin-10 and diabetic wound healing course logistic regression

	Odds Ratio (95% Confidence Interval)	P	Contrast to Visit 1 Δ C-Statistics
Visit 1	2.38 (1.01–5.59)	0.047	REF
Visits 1 and 2	2.97 (1.19–7.37)	0.019	0.041
Visits 1 and 3	2.86 (1.16–7.04)	0.022	0.035

We also used CART analysis to create a decision tree with healing status as the binary response and age, race, sex, and all measured 28 serum proteins during all 6 visits as predictors. Thirty predictors were classified as important, and the presented 11-node tree (Fig. 3) had the least misclassification cost and an area under the receiver operating characteristic curve (c-statistic) of 0.93 (95% confidence interval: 0.89, 0.97) (Supplementary Fig. S3A). In fact, in Fig. 2, we selected a tree size with the smallest number of overall errors when using cross validation. Age was chosen by CART as the first predictor to split the root node while Fractalkine, IL-10, IL-5, IL-6, IL-8, IL-2, INF-γ, and sex were selected for splitting subsequent nodes. Relative variable importance is shown in Supplementary Fig. S3B.

Figure 3.

Classification tree with eleven terminal nodes. Healing status versus all measured serum proteins, age, sex, and race.

The most important predictor variable was IL-10, followed by IL-5 (contribution of 94.5%) and age (contribution of 93.7%). Additional important predictors included Fractalkine, IL-7, IL-2, IL-4, and 1L-6. The confusion matrix with true and false positive and negative rates is shown in Supplementary Table S8.

We subsequently probed our recently generated DFU single-cell atlas for the gene expression of the identified serum biomarkers.

Differential expression analysis results for each cell type between Healers and Non-Healers are shown in Supplementary Table S9 and Supplementary Figs. S4–S6. DFU tissue analysis demonstrated that Healers, when compared to Non-Healers, overexpressed the following genes (Supplementary Fig. S4): CXCL8 (IL-8 encoding gene) in M1-Macrophages (p < 0.0001; logFC = 0.76), Healing-Enhancing Fibroblasts (p < 0.0001; logFC = 1.01), M2-Macrophages (p < 0.0001; logFC = 0.58), and Schwann Cells (p < 0.0001; logFC = 0.32); IL-6 in Fibroblasts (p < 0.0001; logFC = 0.31), in Healing-Enhancing Fibroblasts (p < 0.0001; logFC = 0.59), and M1-Macrophages (p < 0.0001; logFC = 0.4); CX3CL1 (Fractalkine encoding gene) in Sebaceous cells (p < 0.0001; logFC = 0.56); and IFNG in T lymphocytes (p < 0.0001; logFC = 0.33), whereas Non-Healers overexpressed IFNG in Natural Killer cells (p < 0.0001; logFC = −0.41).

Similar analysis at the forearm level (Supplementary Fig. S5) indicated that Non-Healers had higher expression of TNF in M1-Macrophages (p = 0.0004; logFC = −0.3), and Natural Killer cells (p = 0.0069; logFC = −0.92), as well as higher expression of IFNG in Natural Killer cells (p = 0.0465; logFC = −1.66) and T Lymphocytes (p = 0.0061; logFC = −0.4). No genes encoding for any of the significant serum proteins were significantly different between the PBMC samples (Supplementary Fig. S6). Interestingly, while gene expression at the foot wound level does not follow the same direction as observed in the serum, many of those markers do remain significantly dysregulated, underlining their importance in diabetic wound repair, as well as suggesting that different mechanisms govern the expression of those genes and proteins at the skin wound and serum level.

To detect the different cell types that are most actively expressing all the serum highly expressed cytokines at the wound site, forearm, and PBMCs, between Healers and Non-Healers, we used gene expression signatures (Supplementary Fig. S7). More specifically, we used the average log-expression of serum proteins that are higher in Healers (IL-10, IL-4, IL-5, IL-6, IL-13, IFNG) and deducted the average expression of those that are higher in Non-Healers (CX3CL1, CXCL8, TNF). Marked differences were observed between Healers and Non-Healers in the signature activity in DFU tissue (Supplementary Fig. S7A) with the Fibroblasts, Healing-Enhancing Fibroblasts, Mast Cells and Sebaceous Cells having significantly different expression of the signature between Healers and Non-Healers. No significant differences were observed at the forearm, but B Lymphocytes also had significantly different expression of the signature at the PBMC level (Supplementary Fig. S7C).

The distinct expression of those genes between cell types in Healers and Non-Healers illustrates how several genes potentially act in tandem to affect the transcriptional states of these cell types, with Fibroblast and Healing-Enhancing Fibroblast differences being especially important in the context of wound healing.

DISCUSSION

Our data provide strong evidence that measurement of circulating serum proteins can serve as successful biomarkers in guiding clinical management of DFU. The main advantages of such an approach are that it can be easily applicable and can provide results upon completion of the test or soon thereafter, can be cost-effective and its interpretation is simple, and can be based on categorical thresholds. Furthermore, our data indicate a likely superior performance of building a multiprotein biomarker score instead of relying on single biomarkers. The innovation of our approach lies in following the same patients for multiple visits to collect serum, as well as categorizing them in those who healed their ulcers and those who did not.

Previous studies in our unit have attempted to identify serum proteins that are associated with wound healing^13,17 using one single-time serum specimen. However, our efforts were hampered by the lack of adequate information regarding the variability of such serum measurements during the healing period. Possible factors for such variability can include changes in the wound environment, serum protein variability related to systemic effects, such as metabolic changes and variability associated with the applied technique. In this study, we used 6 biweekly measurements over a 12-week period, which allowed us to have a comprehensive idea about possible variabilities of many serum proteins that can be used as DFU healing biomarkers.

Of interest, the main differences between Healers and Non-Healers were observed in inflammatory cytokines. This is in agreement with previous studies in our laboratory that have demonstrated that the mounting of an appropriate inflammatory response, similar to the one that is present in acute uncomplicated wounds, is essential for achieving complete wound healing.^13,14 In addition, age, sex, and race were associated with the observed differences in serum protein levels. These results accentuate the necessity for further investigations that will focus on personalized medicine and will attempt to identify appropriate biomarkers and interventions in specific patient subgroups.

IL-10 emerged as the most promising candidate to serve as a single biomarker. Despite minor variations, the protein tended to be stable in both DFU Healers and Non-Healers, something that allowed a clear separation between the two groups during all six visits. IL-10 is a cytokine produced by macrophages and CD4⁺ T-cells and is known to reduce scarring, in part, through its anti-inflammatory action.^18–20 Notably, type 2 diabetic mellitus (T2DM) patients with DFU have decreased IL-10 levels than T2DM patients without DFU, and IL-10 was found to have high sensitivity, specificity, and predictive value in patients with DFU.²¹ Additional proteins that tended to have a stable trajectory included INF-γ, IL-4, IL-6, IL-8, and Fractalkine. However, differences between Healers and Non-Healers during individual visits did not reach statistical significance, mainly due to high variation and small sample size in each group.

Wound healing is a dynamic process and, as a result, it is not surprising that there was considerable variation among the six biweekly measurements in some of the tested proteins. As the timing of wound healing progression cannot be accurately estimated, only proteins that remain stable during the entire healing period, such as IL-10, can be helpful as one time measurement to predict wound healing. This emphasizes the need for studies with more than one-time measurements for the development of DFU healing predictive algorithms. Furthermore, our data suggest that consecutive measurements over predetermined time periods, such as every two or 4 weeks, as opposed to one-time measurements, may potentially increase the predictive probability of the developed algorithms.

CART analysis is very helpful in constructing a predictive model that can be understood easily without the knowledge of advanced statistics and its main principle is that the outcome is based on the values of the selected variables.²² Our analysis yielded a tree with eleven terminal nodes that involved age, sex, and seven serum proteins and resulted in very satisfactory true positive (93.3%) and negative rates (92.9%). Interestingly, the root node split using age as the first predictor while the next two nodes used IL-10 and Fractalkine. In addition, in agreement with our previous analysis, IL-10 was found to be the most important predictor. These results indicate that age and sex are important factors in predicting DFU healing and should always be taken into consideration. Furthermore, a small panel of serum proteins may be a better alternative than a single biomarker and needs to be corroborated in larger trials.

An important question remains concerning the origin of the measured serum proteins. Although various cell types and tissues could be involved, we focused on the ulcer area, forearm skin, and PBMCs from DFU patients who were available to us from a single-cell RNA-seq study already performed by us.¹⁴ Our findings suggest that the main differences in gene expression between Healers and Non-Healers were observed in cells from the DFU area. Of interest, Healing-Enhancing Fibroblasts from the Healers group overexpressed the CXCL8 (IL-8) and IL-6 genes. Furthermore, in Healers, Fibroblasts overexpressed IL-6, M1 Macrophages overexpressed IL-6 and CXCL8, M2 Macrophages overexpressed CXCL8 while Sebaceous Cells overexpressed CX3CL1 (Fractalkine), and T Lymphocytes overexpressed IFNG. As would be expected, analysis of the forearm skin biopsies showed a few minor differences between Healers and Non-Healers, the most prominent of which was overexpression of IFNG and TNF in the Non-Healers.

Furthermore, no major differences were observed in PBMCs. These results imply that localized DFU microenvironment processes may contribute to the observed serum protein changes. However, it should be noted that as there is no information regarding the role of other tissues and organs that may contribute to the production, post-translational modifications and release of the tested proteins, such as the liver or the immune system, additional extensive studies will be required before firm conclusions can be drawn.

The present study has its limitations, the main of which is the lack of an independent validation cohort to confirm the observed results in the primary discovery cohort. In addition, as CART is prone to overfitting, there is a reasonable possibility that generalization to clinical practice may not be achieved easily. However, the primary objective of our investigation was to interrogate the data of the discovery cohort and identify possible biomarkers and feasibility of proposed techniques that will be tested in a larger, confirmatory cohort. The NIDDK DFC is providing serum specimens from a large cohort of DFU patients that will serve as a confirmatory cohort.

In conclusion, our data indicate that consecutive measurements of proteins associated with wound healing can identify biomarkers that can predict DFU healing over a 12-week period. IL-10 was the strongest candidate to be considered as a single biomarker but the selection of a small panel of serum proteins, based on CART^® Classification, can be an alternative.

INOVATION

Herein, we demonstrate that longitudinal serum quantitation of a panel of proteins can provide biomarkers predictive of DFU wound closure over 12 weeks. This could inform clinicians on further treatment of patients and accelerate wound healing.

KEY FINDINGS

• IL-10 could function as a single biomarker to predict DFU closure for consecutive measurements in the course of treatment.

• A panel of proteins also, including IL-4, IL-5, IL-6, IL-13, and IFNγ, could be additionally predictive of DFU healing

• Fractalkine, IL-8, and TNFα are predictors of nonhealing DFUs

DATA AND RESOURCE AVAILABILITY STATEMENT

The datasets generated during and/or analyzed in the current study are available from the corresponding author upon reasonable request. The single-cell RNA-seq data were downloaded from NCBI's Gene Expression Omnibus with accession number GSE165816.

ACKNOWLEDGMENTS AND FUNDING SOURCES

A.V. was responsible for the study concept and design. G.T., B.S., E.W., I.M., J.G., E.A.V., and I.V. obtained the data. G.T., N.K., and A.V. performed data analysis. G.T., N.K., and A.V. wrote the article. M.H. contributed to data analysis. All authors participated in data interpretation and critical review of the article. They all approved the final report. A.V. received funding from the National Rongxiang Xu Foundation. A.V. was also supported by the National Institutes of Health Grant 1R61DK131915.

AUTHOR DISCLOSURE AND GHOSTWRITING

The authors declare no conflict of interest and no ghostwriters. The listed authors cowrote the article as described in author contributions section.

ABOUT THE AUTHORS

Georgios Theocharidis, PhD, is Assistant Professor of Surgery at Beth Israel Deaconess Medical Center and Harvard Medical School and has a bioengineering background with interests in translational research in impaired cutaneous wound repair. Brandon Sumpio, MD, was a Research Fellow, currently a Vascular Surgery Resident. Enya Wang, BS, was a Research Assistant, currently a Medical School Student. Ikram Mezghani, MA, was a Research Assistant, currently a PhD Student. John M. Giurini, DPM, is Chief of Podiatry at Beth Israel Deaconess Medical Center and Associate Professor of Surgery, Harvard Medical School. Nikolaos Kalavros, MSc, works at Spatial Technologies Unit, Beth Israel Deaconess Medical Center. Eleftheria-Angeliki Valsami, PhD, is Postdoc Fellow at Beth Israel Deaconess Medical Center and Harvard Medical School. Ioannis Vlachos, PhD, is Associate Professor of Pathology at Beth Israel Deaconess Medical Center and Harvard Medical School. Mahyar Heydarpour, PhD is Biostatistician at Mass General Brigham and Harvard Medical School. Aristidis Veves, MD, DSc is a Professor of Surgery at Harvard Medical School and the founding Director of the Rongxiang Xu, MD Center for Regenerative Therapeutics. The primary research focus of his lab is diabetes and its complications, with the main emphasis on wound healing and cardiovascular disease.

Abbreviations and Acronyms

ABI

ankle-brachial pressure

BMI

body mass index

CART

Classification And Regression Tree

CI

confidence interval

DFC

Diabetic Foot Consortium

DFU

diabetic foot ulcer

DM

diabetes mellitus

HbA1c

glycated hemoglobin

HDL

high-density lipoprotein

IFN

interferon

IL

interleukin

LDL

low-density lipoprotein

LogFC

log fold change

NDS

neuropathy disability score

NIDDK

National Institute of Diabetes and Digestive and Kidney Diseases

NS

not significant

OR

odds ratio

PBMC

peripheral blood mononuclear cell

SD

standard deviation

SW

Semmes-Weinstein monofilament

T2DM

type 2 diabetes mellitus

TNF

tumor necrosis factor

VPT

vibration perception threshold