Abstract
Age, disease, and exposure to environmental factors can induce tissue remodelling and alterations in protein structure and abundance. In the case of human skin, ultraviolet radiation (UVR)-induced photo-ageing has a profound effect on dermal extracellular matrix (ECM) proteins. We have previously shown that ECM proteins rich in UV-chromophore amino acids are differentially susceptible to UVR. However, this UVR-mediated mechanism alone does not explain the loss of UV-chromophore-poor assemblies such as collagen. Here, we aim to develop novel bioinformatics tools to predict the relative susceptibility of human skin proteins to not only UVR and photodynamically produced ROS but also to endogenous proteases. We test the validity of these protease cleavage site predictions against experimental datasets (both previously published and our own, derived by exposure of either purified ECM proteins or a complex cell-derived proteome, to matrix metalloproteinase [MMP]-9). Our deep Bidirectional Recurrent Neural Network (BRNN) models for cleavage site prediction in nine MMPs, four cathepsins, elastase-2, and granzyme-B perform better than existing models when validated against both simple and complex protein mixtures. We have combined our new BRNN protease cleavage prediction models with predictions of relative UVR/ROS susceptibility (based on amino acid composition) into the Manchester Proteome Susceptibility Calculator (MPSC) webapp. Application of the MPSC to the dermal proteome suggests that fibrillar collagens and elastic fibres will be preferentially degraded by proteases alone and by UVR/ROS and protease in combination, respectively. We also identify novel targets of oxidative damage and protease activity including dermatopontin (DPT), fibulins, and defensins.
Introduction
Human skin is constantly under siege from environmental stressors such as ultraviolet radiation (UVR) and reactive oxygen species (ROS), alongside internal factors like endogenous proteases. This collective degradation process, termed the "degradome," leads to tissue remodeling and photo-aging. While UV susceptibility is linked to amino acid composition, predicting enzymatic cleavage by Matrix Metalloproteinases (MMPs) is complex, as it depends on sequence context, protein folding, and accessibility. Existing bioinformatic tools often struggle to accurately predict these "cleavage sites" in native, non-denatured proteins. This study introduces the Manchester Proteome Susceptibility Calculator (MPSC) to bridge this gap using deep learning.
Methods
The researchers developed the Manchester Proteome Cleave (MPC) algorithm, utilizing a Deep Bidirectional Recurrent Neural Network (BRNN) with four Bi-LSTM layers. This architecture captures the context of amino acid sequences in both directions. The model incorporates predicted structural features, including secondary structure (helices vs. sheets), disordered regions, and solvent accessibility.
To overcome data scarcity for specific proteases, transfer learning was employed: a generalist model was pre-trained on a large protease dataset and then fine-tuned for specific enzymes like MMP-9 or Cathepsin K. The protease predictions were integrated with a mathematical model for UV/ROS susceptibility based on amino acid composition. Validation was performed using public databases (MEROPS) and new experimental benchmarks derived from LC-MS/MS analysis of purified proteins and complex fibroblast-derived ECM exposed to MMP-9.
Results
The MPC model demonstrated superior performance compared to state-of-the-art tools like PROSPER and DeepCleave. On a complex fibroblast proteome, MPC achieved an AUC of 0.71, significantly higher than DeepCleave (0.64) and PROSPER (0.55). The model successfully predicted cleavage sites for 15 different proteases, including nine MMPs and four cathepsins.
Predicted UV/ROS susceptibility scores showed a strong linear correlation ($R^2 = 0.94$) with experimental observations (Figure a). The analysis revealed that degradation is synergistic: elastic fibers are susceptible to both UVR/ROS and proteases, whereas fibrillar collagens are primarily susceptible to proteases. Furthermore, the tool identified novel targets of degradation, such as dermatopontin (DPT) and members of the defensin family (e.g., DEFB1, DEFA3), which may serve as new biomarkers for photo-aging (Figures b and c).
Discussion
Predicting protein cleavage in native environments remains a challenging task. While the MPC model improves upon existing methods by incorporating predicted structural features, it currently relies on 2D structural predictions rather than experimental 3D structures. Future iterations could integrate high-resolution structural data from tools like AlphaFold and larger experimental datasets. The MPSC webapp provides a valuable resource for identifying novel biomarkers and therapeutic targets in age-related and inflammatory diseases.
Resource: The Manchester Proteome Susceptibility Calculator is publicly available at http://www.manchesterproteome.manchester.ac.uk/#/MPSC.
Summary updated