Evaluation of the collaborative network of highly correlating skin proteins and its change following treatment with glucocorticoids

Abstract

Background: Glucocorticoids (GC) represent the core treatment modality for many inflammatory diseases. Its mode of action is difficult to grasp, not least because it includes direct modulation of many components of the extracellular matrix as well as complex anti-inflammatory effects. Protein expression profile of skin proteins is being changed with topical application of GC, however, the knowledge about singular markers in this regard is only patchy and collaboration is ill defined.

Material/methods: Scar formation was observed under different doses of GC, which were locally applied on the back skin of mice (1 to 3 weeks). After euthanasia we analyzed protein expression of collagen I and III (picrosirius) in scar tissue together with 16 additional protein markers, which are involved in wound healing, with immunhistochemistry. For assessing GC's effect on co-expression we compared our results with a model of random figures to estimate how many significant correlations should be expected by chance.

Results: GC altered collagen and protein expression with distinct results in different areas of investigation. Most often we observed a reduced expression after application of low dose GC. In the scar infiltrate a multivariate analysis confirmed the significant impact of both GC concentrations. Calculation of Spearman's correlation coefficient similarly resulted in a significant impact of GC, and furthermore, offered the possibility to grasp the entire interactive profile in between all variables studied. The biological markers, which were connected by significant correlations could be arranged in a highly cross-linked network that involved most of the markers measured. A marker highly cross-linked with more than 3 significant correlations was indicated by a higher variation of all its correlations to the other variables, resulting in a standard deviation of > 0.2.

Conclusion: In addition to immunohistochemical analysis of single protein markers multivariate analysis of co-expressions by use of correlation coefficients reveals the complexity of biological relationships and identifies complex biological effects of GC on skin scarring. Depiction of collaborative clusters will help to understand functional pathways. The functional importance of highly cross-linked proteins will have to be proven in subsequent studies.

Figure 1

Collagen type I (red pixel) and collagen type III (green pixel) in skin scar (number of pixel) after treatment with ethanol, low dose GC and high dose GC for 1 to 3 weeks (# p < 0.05; + p < 0.01).

Figure 2

Correlation networks and their analysis in a fictive model system. 2A: Depiction of the collaborative network in the cell infiltrate using significant Spearman's correlation coefficients (p < 0.05). Highly connected proteins with more than 3 significant correlations are marked in grey color; solid lines mark positive correlation coefficient, dotted line negative correlation coefficient. The analyzed tissue proteins include TGF-beta, MMP-2, TNF-R2, COX-2, AXL, c-myc, Ki67, beta-Catenin, ESDN, p53, Gas6, Notch 3, CD68, apoptosis -TUNEL, SMA, collagen type I - red pixel and type III - green pixel. Three therapy groups were compared; ethanol-controls, low dose GC and high dose GC. Figure 2B: Depiction of significant Spearman's correlations within a database of 20 variables, each consisting of 30 random data sets. Figure 2C: Depiction of significant Spearman's correlations within a fictive random database of 20 variables after introduction of a cluster, which included the variables 9 to 14 with equal numerical values, and which led to correlation coefficients of r = 1 between these variables.

Figure 3

Number of correlation coefficients with significance in relation to mean and standard deviation of all coefficients per variable. 3A: 20 markers of the cell infiltrate underneath the scar with mean and standard deviation of all 19 correlation coefficients of every variable in relation to the number of significant correlations per variable. A standard deviation of more than 0.2 indicated highly cross-linked marker proteins, whereas the mean of all coefficients did not. 3B: 20 variables in a fictive model of random figures with mean and standard deviation of all 19 correlation coefficients of every variable in relation to the number of significant correlations per variable. With means between +0.1 and -0.1 and standard deviations of less than 0.25 highly cross-linked variables were rare. 3C: 20 variables in a fictive model of random figures after introduction of a cross-linked cluster with mean and standard deviation of all 19 correlation coefficients of every variable in relation to the number of significant correlations per variable. A standard deviation of more than 0.2 indicated highly cross-linked variables.