Biochemical and Immunological Predictors of Non-healing in Individuals with Early-stage Diabetic Foot Ulcers

Objective: The goal of this study was to identify biochemical and immunological parameters from the blood as predictors of non-healing in early-stage diabetic foot ulcers. Research Design and Method: We performed a cross-sectional prospective cohort study among individuals with early-stage foot ulcers visiting the Karnataka Institute of Endocrinology Research over a 2.5-year period. Histopathological, biochemical, and immunological data (a total of 31 parameters) from 52 individuals were collected and analyzed to determine if predictors of non-healing may be identified. Data analysis was performed using traditional univariate analyses as well as univariate and multivariable logistic regression. Results: Individual histopathological and biochemical parameters did not show any differences between healed and non-healed individuals. However, conventional univariate analysis and univariate logistic regression analysis showed that the expression of the cell-surface proteins CD63, HLA-DR and CD11b on monocytes (CD14+) was significantly lower in non-healed individuals, but with moderate discriminative ability as assessed by area under the curve (AUC) of Receiver Operating Characteristics (ROC) curve. In comparison, a multivariable logistic regression model identified four of the 31 parameters to be salient predictors and demonstrated high discrimination ability with an AUC of ROC value of 0.87. Among the four identified parameters, LDL cholesterol (OR 18.83, CI 18.83-342) and cell-surface expression of CD63 on monocytes (OR 0.12, CI 0.12-0.45) were significant. Conclusion: Through this study we conclude that LDL cholesterol and cell-surface expression of CD63 on monocytes are strong positive and negative predictors of non-healing, respectively, in individuals with early-stage DFU. Following validation in a larger cohort, these parameters may be used by the clinician for early identification of non-healers.


INTRODUCTION
A few studies also suggest presence of chronic low-grade systemic inflammation in addition to 24 local inflammation in DFU individuals (16,17), quantification of which are possible in most clinics. 25 One such cohort study by Dinh and colleagues showed that non-healed individuals had increased 26 (as compared to individuals whose ulcers healed) serum levels of tumor necrosis factor-α (TNF-27 α), monocyte chemoattractant protein-1 (MCP-1), matrix metallopeptidase-9 (MMP-9), and 28 fibroblast growth factor-2 (FGF-2) about 8 months prior to the development of ulcers (18). We 29 conjectured that as myeloid cells play a vital role in the secretion of these cytokines and the 30 establishment of inflammation, they might contribute to slower or absence of healing. However, 31 their role in the healing of early-stage ulcers remains poorly characterized. The goal of the current 32 study was to characterize the phenotype of myeloid cells in individuals with active early-stage 33 ulcers and determine if their phenotype along with other clinical parameters may be used to predict 34 non-healing of ulcers.

35
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. days of biopsy collection and used for biochemical testing and immunophenotyping. Patient's 59 wounds were dressed regularly. Wound healing was monitored for a period of one month by 60 capturing images using a near infra-red based camera (WoundZoom, USA) at the time of dressing. 61 Image analysis 63 Wound images captured by a near infra-red based camera were analyzed using ImageJ. Percentage 64 reduction in wound area was calculated as a measure of healing. Threshold of 50% reduction in 65 wound area on Day 30 was used to classify patients as healed.   CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021. were prepared using compensation beads (BD Biosciences, USA), which were stained with 102 antibodies following manufacturer's protocol. Additionally, fluorescence minus one (FMO) 103 controls were prepared to correct for group effect of fluorophores on spectral spill. Briefly, live-104 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint dead dye stained cells were stained with antibodies for all colors except one which was replaced 105 by respective isotype to prepare FMO for a color of interest. Similar controls were prepared for all 106 fluorophores mentioned in Supplementary Table 1 and used for intensity correction at the time of 107 data analysis. Voltages were set on system using compensation beads at the time of sample 108 acquisition and a minimum of 100,000 CD45 + live events were acquired using BD FACSDiva TM 109 Version 6 Software system (BD Biosciences, USA). All data analyses were performed on FCS files using FlowJo TM v10.6 (Becton Dickinson, USA).

113
Compensation was performed on FlowJo using compensation beads. CD14 and CD15 were used 114 to identify monocytes and granulocytes, respectively. FMO's were used to draw appropriate gates.   (20) and magrittr (21) with low Akaike's 126 information criterion (AIC) score as selection criterion for identifying salient predictors.

127
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint Unadjusted odd's ratio was reported with 95% confidence interval for multivariable logistic 128 regression model. ROC analysis was performed using pROC package in R (22). ROC plots were 129 created using the same package for univariate and multivariable logistic regression models. Fisher's exact test was performed to analyze categorical histology data in R using stats package.

133
Heatmap was created using ComplexHeatmap package in R (23). Univariate analysis of all 134 predictors was performed using student's t-test with Welch's correction using GraphPad Prism 8.

135
Wald's test for significance of predictors was performed by glmnet package as part of the routine.

136
Model evaluation statistics such as Chi-square goodness of fit (GOF) test and Hosmer-Lemeshow 137 (HL) test was performed using stats and ResourceSelection (24) package in R respectively. AIC 138 score was obtained for all models using stats package. Model fit plots were created using ggplot2

142
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021.

143
Cohort characteristics 145 The final study cohort comprised of 52 diabetic individuals, with either Stage I (30) or Stage II 146 (22) foot ulcers. Their clinical characteristics are summarized in Table 1. We observe an   163 We also collected peripheral venous blood from the recruited individuals for biochemical analysis 164 and immunological characterization of neutrophils and monocytes in circulation. Immunological 165 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021.  Table 2).

175
However, the analysis did show that the expression levels of three proteins on monocytes (CD14 176 expressing cells), CD63, HLA-DR and CD11b, were significantly higher on individuals whose 177 ulcers had healed as compared to those whose ulcers had not healed (Supplementary Table 2 Table 3) to assess the discriminative capability of each parameter, all three were 180 observed to have relatively moderate area under the curve (AUC) values and were deemed to be 181 not effective in discriminating between non-healed ulcers and healed ulcers. Hence, we next used 182 logistic regression, a binary classifier model to assess outcome prediction. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint Goodness of Fit (GOF) test, and only MFI of CD63 among monocytes (CD14 + ) was significant 189 when evaluated using the Wald's test (Supplementary Table 4). Additionally, the odd's ratio (OR) 190 for MFI of CD63 among monocytes (CD14 + ) was found to be 0.26 (CI 0.08-0.68) suggesting   Table 2). The model performance was found to be significant 207 based on a Chi-square goodness of fit (GOF) test (p = 0.011). A Hosmer-Lemeshow (HL) goodness 208 of fit test was also performed to evaluate agreement between model's predicted and expected event 209 rates across deciles of risk groups. HL test p value was found to be 0.411 implying lack of evidence 210 for disagreement between predicted and expected event rates. Additionally, Akaike Information 211 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint Criterion (AIC), an estimator of out-of-sample prediction error to assess model performance was 212 found to be 65.74. Odds ratio for each predictor along with CI is summarized in Table 2, and 213 Figure 3A shows the predicted probabilities of the model as well as the ROC analysis.
Similarly, a stepwise feature selection identified four of 15 immunological phenotypic parameters 216 to build the Immunological model, three of which were found to be significant according to Wald's 217 test (Table 3). Chi-square and HL GOF tests showed p value 0.002 and 0.368, respectively 218 indicating that the model is statistically significant, and the AIC was found to be 61.21. Odd's ratio 219 and corresponding CI are summarized in Table 3. Figure 3B shows the predicted probabilities of 220 the model as well as the ROC analysis. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint healing outcome. Together, these metrics suggest that the combined model performed the best in 235 predicting non-healing among individuals with early-stage diabetic foot ulcers. 236 237 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

238
The goal of our study was to identify clinically measurable parameters that could help predict 239 individuals in whom early-stage DFUs fail to heal. Hence, we focused on collecting and analyzing 240 histopathological, biochemical, and immunological parameters that are relatively easy to collect 241 in many health-centers. Our analysis revealed that the histopathological data did not correlate with 242 healing, and no individual biochemical or immunological parameter was a good predictor of non-  is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint (13,14,27). While these studies clearly establish that both local and systemic inflammation 261 correlate with poor healing, measurement of the identified mediators are often difficult due to the 262 labor-intensive methodologies involved or the requirement for equipment that are not yet available 263 in many clinical centers. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint function (28). Hence, individuals with increased levels of LDL cholesterol may have systemic low-284 grade inflammation, which might impact many tissues, including DFUs. Additionally, LDL 285 cholesterol has been shown to have inhibitory effect on endothelial cell proliferation, and has been 286 shown to cause delays in wound healing in mouse models of research (29). In concurrence with 287 these observations, our analysis reveals that higher levels of LDL cholesterol correlate with non-   In diabetic individuals, hyperglycemia is known to play a role in establishing and maintaining 302 inflammation (34)(35)(36). In that context, a surprising observation in our study is the lack of 303 correlation between hyperglycemia (both HbA1c and fasting blood sugar) and non-healing. This 304 observation is in contrast to a decade old retrospective study by Christman et al., which showed 305 that higher HbA1c strongly correlates to poor healing in DFU individuals (37). Two possible 306 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint reasons for the difference between our observations and this study, is the stage of ulcer under 307 consideration (we focus on early-stage ulcers compared to aggressive late stage or amputated 308 wounds in the other study) and the method used to classify healing (we measure healing at an 309 endpoint compared to rate of healing used in the other study). However, our data and observations  One of the major limitations of this study is the relatively small sample size of the cohort, which 316 may limit the statistical power. Nevertheless, our observations provide new information on 317 measurable predictors for non-healing. Second is the classification system used for assessing 318 healing status, which was set as 50% reduction in wound area by 4 weeks following thresholds 319 reported by Sheehan et al. and Lavery et al. (7,8). Due to challenges associated with following-up 320 patients in our setting, we were unable to physically verify if the individuals classified as healed 321 showed complete wound healing by ~16 weeks. Third, the duration that the individual has had an 322 ulcer for before presentation in the clinic was determined through an oral conversation but could 323 not be verified as in many cases no prior medical records exist. Lastly, many of the individuals 324 who visit this center have uncontrolled diabetes, and while this may be common for many 325 government health centers in India, it may not be the norm elsewhere. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint In summary, our cross-sectional cohort study analyzing clinical, biochemical, and immunological 329 parameters among individuals with early-stage DFU showed that no single parameter correlates  is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted November 9, 2021.   is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

351
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint indicated by grey dots. B -Receiver operator characteristics (ROC) curve of CD63, HLA-DR and CD11b has been plotted. Corresponding area under the curve was observed to be 0.73, 0.65 and 0.63, respectively.

352
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint

353
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted November 9, 2021. ; https://doi.org/10.1101/2021.11.09.21266108 doi: medRxiv preprint