Skin barrier homeostasis in atopic dermatitis: feedback regulation of kallikrein activity

Abstract

Atopic dermatitis (AD) is a widely spread cutaneous chronic disease characterised by sensitive reactions (eg. eczema) to normally innocuous elements. Although relatively little is understood about its underlying mechanisms due to its complexity, skin barrier dysfunction has been recognised as a key factor in the development of AD. Skin barrier homeostasis requires tight control of the activity of proteases, called kallikreins (KLKs), whose activity is regulated by a complex network of protein interactions that remains poorly understood despite its pathological importance. Characteristic symptoms of AD include the outbreak of inflammation triggered by external (eg. mechanical and chemical) stimulus and the persistence and aggravation of inflammation even if the initial stimulus disappears. These characteristic symptoms, together with some experimental data, suggest the presence of positive feedback regulation for KLK activity by inflammatory signals. We developed simple mathematical models for the KLK activation system to study the effects of feedback loops and carried out bifurcation analysis to investigate the model behaviours corresponding to inflammation caused by external stimulus. The model analysis confirmed that the hypothesised core model mechanisms capture the essence of inflammation outbreak by a defective skin barrier. Our models predicted the outbreaks of inflammation at weaker stimulus and its longer persistence in AD patients compared to healthy control. We also proposed a novel quantitative indicator for inflammation level by applying principal component analysis to microarray data. The model analysis reproduced qualitative AD characteristics revealed by this indicator. Our results strongly implicate the presence and importance of feedback mechanisms in KLK activity regulation. We further proposed future experiments that may provide informative data to enhance the system-level understanding on the regulatory mechanisms of skin barrier in AD and healthy individuals.

Figure 1. Models of skin desquamation.

A: Cartoon model of skin desquamation. Skin barrier is physically composed of the cornified layer, where keratin-filled and anucleated keratinocytes (corneocytes) are densely packed with skin lipids. Corneocytes are interconnected by corneodesmosomes. Skin desquamation occurs by elimination of corneocytes at the skin surface. B: Cartoon model of protein interactions involved in KLK5 activation regulation. (a) KLK5 and their inhibitor LEKTI are secreted from granular cells into the intercellular space at the interface of cornified and granular layers; (b) KLK5 self-activates by proteolysis; (c) Direct binding of LEKTI inhibits the activity of KLK5; (d) Active KLK5 physically cleaves corneodesmosomes, which bind corneocytes together, resulting in elimination of corneocytes; (e) PAR2 is cleaved by active KLK5 to be activated and internalized. Figure was modified from . C: Simplified model for KLK5 activation regulation system proposed in this paper. KLK5* and PAR2* represent the activated forms of KLK5 and PAR2, respectively. (a) KLK5 self-activation by proteolysis; (b) Association and dissociation of LEKTI and KLK5*; (c) PAR2 activation by KLK5*; (d) Feedback from PAR2* to production of LEKTI (); (e) Feedback from PAR2* to production of KLK5 (); Inflammation level is denoted by the level of PAR2*.

Figure 2. PAR2 signalling downstreams in granular cells.

A: For the non-inflammatory states, a small amount of active PAR2, , is constitutively produced, maintaining the basal activity of KLK5 production for the normal desquamation process. Various external stimuli can influence to be fluctuated around its nominal value. B: For the inflammatory states, a large amount of activated PAR2, [PAR2*], is induced and internalized, which then transduce stronger canonical signalling cascades and increase the expression of inflammatory genes including IL1 , IL1 , IL8 and TNF- .

Figure 3. Bifurcation diagrams of Model 1 showing the inflammation outbreak and its persistence.

Inflammation outbreak and its persistence appear as bistability of inflammation level as the external stimulus level changes. The solid and dotted lines show the stable and unstable steady states, respectively. A: The thickness of each bifurcation curve corresponds to positive feedback strength . Stronger positive feedback leads to more persistent inflammation, as is shown by the larger range of the bistability. B: The behaviours are compared for HC (black), AD-LEKTI (blue), and AD-pH (red) with and . The inflammation threshold is lower for AD conditions than that for HC.

Figure 4. (A) Four bifurcation patterns exhibited by our models and (B) quantitative indices of bistable behaviours.

A: Reversible bistability (red), irreversible bistability (blue), continuous monostability (green), and discontinuous monostability (cyan). Bistability patterns match the characteristic switching of inflammation. Irreversible bistability patterns correspond to more severe symptoms than reversible bistability patterns. B: Inflammation outbreaks at the inflammation threshold and persists until decreases to reach the deactivation threshold , where the inflammation level returns to zero. The range of bistability represents the required level of decrease in the external stimulus for the inflammation to cease. Smaller values of indicate an increased sensitivity of the skin to external stimulus; Larger values of indicate that the inflammation is persistent.

Figure 5. Bifurcations for Model 1 with different parameters.

Calculated for pair of feedback strength and . KLK production rate is higher at the bottom right corner. (Top) Bifurcation patterns with colours corresponding to those in Fig. 4A. Stronger KLK activation results in more severe symptoms of irreversible bistability. (Middle) Inflammation threshold for bistability patterns; for monostability patterns. (Bottom) Range of bistability for reversible bistability; for other patterns. The inflammation is more persistent ( is larger) with stronger KLK activation. A: Comparison for HC, AD-LEKTI, and AD-pH. AD conditions exhibit smaller and more severe symptoms than HC. B: Comparison for different degradation rates for KLK5 and KLK5* in HC with (nominal), and . Slower KLK5 degradation (smaller ) results in the stronger KLK activity and shows similar effects as in AD-pH condition leading to more irreversible bistability patterns and lower threshold values. C: Comparison for different degradation rates for PAR2 and PAR2* in HC with (nominal) and . Slower PAR2 degradation (smaller ) results in the stronger inflammation and shows similar effects as in AD-LEKTI condition leading to more irreversible bistability patterns and little changes in threshold values.

Figure 6. Sensitivity indicator for Models 1 and 2 calculated by eFAST.

Global sensitivity analysis of Models 1 and 2 with respect to the steady state level of inflammation [PAR2*]. Baseline parameter values are given in Table 1. Parameters were perturbed over one order of magnitude ( = 2000 simulations for eFAST).

Figure 7. Microarray data for AD and HC samples.

PAR2 score was derived using the data of seven PAR2 downstream genes (see Methods). PAR2 score is plotted against expression data of (A) KLK5, (B) SPINK5, and (C) KLK7 for lesional AD (red squares), non-lesional AD (blue squares), and HC (black circles). The dotted lines indicate the median values of HC samples.

Figure 8. Model results of inflammation level against production level of KLK5 and LEKTI.

A: Model 1 with and , B: Model 2 with and . Lines with different colours correspond to different conditions: HC (black), AD-LEKTI (blue), AD-pH (red), and AD-LEKTI/pH (green). Microarray data in Fig. 7 is plotted for comparison after scaling: PAR2 score () is scaled by , where is the median of for HC, to compare with the inflammation level; KLK5 () and SPINK5 expression () data are scaled by and for comparison with Model 1, and and for comparison with Model 2, where is the median of for HC. Data with positive values are only shown here.