Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2018 Feb 20;16(1):32.
doi: 10.1186/s12967-018-1406-x.

Computational Analysis Identifies Putative Prognostic Biomarkers of Pathological Scarring in Skin Wounds

Affiliations
Free PMC article

Computational Analysis Identifies Putative Prognostic Biomarkers of Pathological Scarring in Skin Wounds

Sridevi Nagaraja et al. J Transl Med. .
Free PMC article

Abstract

Background: Pathological scarring in wounds is a prevalent clinical outcome with limited prognostic options. The objective of this study was to investigate whether cellular signaling proteins could be used as prognostic biomarkers of pathological scarring in traumatic skin wounds.

Methods: We used our previously developed and validated computational model of injury-initiated wound healing to simulate the time courses for platelets, 6 cell types, and 21 proteins involved in the inflammatory and proliferative phases of wound healing. Next, we analysed thousands of simulated wound-healing scenarios to identify those that resulted in pathological (i.e., excessive) scarring. Then, we identified candidate proteins that were elevated (or decreased) at the early stages of wound healing in those simulations and could therefore serve as predictive biomarkers of pathological scarring outcomes. Finally, we performed logistic regression analysis and calculated the area under the receiver operating characteristic curve to quantitatively assess the predictive accuracy of the model-identified putative biomarkers.

Results: We identified three proteins (interleukin-10, tissue inhibitor of matrix metalloproteinase-1, and fibronectin) whose levels were elevated in pathological scars as early as 2 weeks post-wounding and could predict a pathological scarring outcome occurring 40 days after wounding with 80% accuracy.

Conclusion: Our method for predicting putative prognostic wound-outcome biomarkers may serve as an effective means to guide the identification of proteins predictive of pathological scarring.

Keywords: Biomarkers; Computational modeling; Pathological scarring; Predictive analysis.

Figures

Fig. 1
Fig. 1
Computational strategy. a First, we used our computational kinetic model to simulate 120,000 distinct wound-healing scenarios. The output of each simulation comprised the time courses for the 28 model variables at 40 time points after wounding (each simulated time point represented the level of a variable on each of the 40 days post-wounding). In addition to the 120,000 simulations, we used the default parameter set to simulate a normal wound healing (i.e., normal healing) scenario. Second, we calculated fold changes (on day 40) of total collagen and fibroblast concentrations in each of the 120,000 simulations with respect to their corresponding values in the normal-healing simulation. Based on these fold changes, we classified the 120,000 simulations as “normal” (fold change ≤ 1), “mild pathological” (5 ≤ fold change ≤ 10), or “severe pathological” (fold change > 10) scarring simulations. Finally, we analysed the concentration distributions (or histograms) of 18 modeled wound proteins (excluding collagen) in the normal-healing and pathological-scarring simulations to determine the diagnostic and prognostic biomarkers of pathological scarring (see the “Methods” section for further details). b The pie chart shows the number of simulations that fell into each of the three categories of wound healing (i.e., normal, mild pathological, or severe pathological) after implementation of the classification criteria
Fig. 2
Fig. 2
Logistic regression analysis. We provided the classification (i.e., “normal” or “pathological”) and the normalized concentrations of the model-identified prognostic biomarkers from the 120,000 simulations as inputs to logistic regression models. The models yielded logistic regression coefficients for each model-identified biomarker and the probability of a given simulation being “pathological” based on one, two, or three biomarkers as predictors. We used the probabilities resulting from the logistic regression models to derive the ROC curves (see Additional file 1 for further details)
Fig. 3
Fig. 3
Diagnostic biomarkers of pathological scarring. Concentration distributions of a IL-10, b fibronectin, c TIMP-1, d CXCL8, e TGF-β, and f IL-6 in normal-healing simulations (solid green lines), mild pathological-scarring simulations (dotted pink lines), and severe pathological-scarring simulations (dashed pink lines) at the final simulated time point (i.e., day 40). Brackets (x-axis) designate concentrations. y-axis represents the percentage of simulations for each curve (described in “Methods” section). g Solid bars represent the fold changes in protein levels in human scar tissue calculated from published experimental data available in the literature. A fold change was calculated as the level of a protein measured in the material derived from pathological scar tissue divided by its corresponding level measured in the material derived from scar tissue under normal-healing conditions. The assay used to measure the level of a particular protein, as well as the time at which the measurement was performed, varied between different experimental studies. The data on TGF-β and TIMP-1 were taken from Refs. [21, 22], respectively. The data on IL-10 were taken from two separate studies: Ref. [20] [IL-10 (1)] and Ref. [23] [IL-10 (2)]. The data on fibronectin were also taken from two separate studies: Ref. [14] [Fibronectin (1)] and Ref. [24] [Fibronectin (2)]. Open bars represent the corresponding model-simulated fold change values. We have not found any published experimental data on the levels of CXCL8 and IL-6 in pathological scars
Fig. 4
Fig. 4
Putative prognostic biomarkers of pathological scarring. Concentration distributions of ac IL-10, df TIMP-1, and gi fibronectin at simulated times representing days 7, 14, and 21 post-wounding. Brackets (x-axis) designate concentrations. y-axis represents the percentage of simulations for each curve (described in “Methods” section). Solid green lines show the concentration distributions for the normal-healing simulations, dotted pink lines show the concentration distributions for the mild pathological-scarring simulations, and dashed pink lines show the concentration distributions for the severe pathological-scarring simulations
Fig. 5
Fig. 5
Receiver operating characteristic (ROC) curves. a ROC curves derived from logistic regression models, using day-14 concentrations of IL-10 alone (pink line, AUC: 0.77, 95% CI [0.767, 0.777]); TIMP-1 alone (cyan line, AUC: 0.79, 95% CI [0.784, 0.795]); fibronectin alone (green line, AUC: 0.80, 95% CI [0.794, 0.806]); TIMP-1 and IL-10 (red line, AUC: 0.81, 95% CI [0.806, 0.816]); fibronectin and TIMP-1 (brown line, AUC: 0.81, 95% CI [0.807, 0.818]), fibronectin and IL-10 (blue line, AUC: 0.82, 95% CI [0.811, 0.821]); and fibronectin, TIMP-1, and IL-10 (black line, AUC: 0.82, 95% CI [0.818, 0.829]). b ROC curve for tenfold cross validation performed by using day-14 concentrations of fibronectin, TIMP-1, and IL-10 as predictors (AUC: 0.80, 95% CI [0.792, 0.807]). c ROC curves derived from logistic regression models, using day-21 concentrations of IL-10 alone (pink line, AUC: 0.83, 95% CI [0.834, 0.842]); TIMP-1 alone (cyan line, AUC: 0.84, 95% CI [0.840, 0.848]); fibronectin alone (green line, AUC: 0.86, 95% CI [0.859, 0.866]); TIMP-1 and IL-10 (red line, AUC: 0.87, 95% CI [0.866, 0.873]); fibronectin and TIMP-1 (brown line, AUC: 0.87, 95% CI [0.875, 0.882]); fibronectin and IL-10 (blue line, AUC: 0.88, 95% CI [0.876, 0.883]); and fibronectin, TIMP-1, and IL-10 (black line, AUC: 0.89, 95% CI [0.884, 0.891]). d ROC for tenfold cross validation performed by using day-21 concentrations of fibronectin, TIMP-1, and IL-10 as predictors (AUC: 0.86, 95% CI [0.855, 0.870])
Fig. 6
Fig. 6
Summary of results. Among the 21 modeled proteins, six were shown to serve as diagnostic biomarkers of pathological scarring. Three modeled proteins were identified as putative prognostic biomarkers of pathological scarring with a reasonably high predictive accuracy (> 80%) on days 14 and 21 post-wounding

Similar articles

See all similar articles

Cited by 1 article

References

    1. Butzelaar L, Soykan EA, Galindo Garre F, Beelen RH, Ulrich MM, Niessen FB, Mink van der Molen AB. Going into surgery: Risk factors for hypertrophic scarring. Wound Repair Regen. 2015;23:531–537. doi: 10.1111/wrr.12302. - DOI - PubMed
    1. Martin P, Nunan R. Cellular and molecular mechanisms of repair in acute and chronic wound healing. Br J Dermatol. 2015;173:370–378. doi: 10.1111/bjd.13954. - DOI - PMC - PubMed
    1. Rabello FB, Souza CD, Farina Junior JA. Update on hypertrophic scar treatment. Clinics. 2014;69:565–573. doi: 10.6061/clinics/2014(08)11. - DOI - PMC - PubMed
    1. Gauglitz GG, Korting HC, Pavicic T, Ruzicka T, Jeschke MG. Hypertrophic scarring and keloids: pathomechanisms and current and emerging treatment strategies. Mol Med. 2011;17:113–125. doi: 10.2119/molmed.2009.00153. - DOI - PMC - PubMed
    1. Lindley LE, Stojadinovic O, Pastar I, Tomic-Canic M. Biology and biomarkers for wound healing. Plast Reconstr Surg. 2016;138:18S–28S. doi: 10.1097/PRS.0000000000002682. - DOI - PMC - PubMed

Publication types

Feedback