I1: RiskScore

Project I1 associates omics data with the clinical phenotype of aged human skin. Aging processes are determined by genetic factors (intrinsic aging) as well as by environmental factors (extrinsic aging). Whether at all and, if so, how intrinsic and extrinsic aging mutually influence each other is still not yet fully understood. In preliminary work, we have determined the characteristics of the secretome of human fibroblasts of intrinsically versus extrinsically aged skin for individuals of three different age groups (18-25 a, 35-49 a, 60-67 a). The secretome is a set of proteins, which are secreted into the extracellular space. In Waldera-Lupa et al. (2015) we found that intrinsically aged fibroblasts acquire an age-dependent secretory phenotype. In addition to the secretome data, proteome data is already available and RNAseq data will be obtained in 2020.

In this project, we will investigate whether and how the omics data sets are associated with the clinical phenotype of aged human skin. The project profits from big, systematically collected clinical data sets that allow a comprehensive description of the skin aging phenotype, e.g., via validated skin aging scores, skin physiological parameters and a questionnaire-based assessment of lifestyle factors.

For the separate analysis of each of the omics data, we will develop weighted risk scores (RS) for the association with the outcome Y, following Hüls et al. (2017a) and Hüls et al. (2017b). In the absence of external weights, we will divide the data set into a training data set, on which the weights are estimated, and a test data set, on which the association analysis is performed. The weighted risk scores will then be used jointly as predictors in an integrated regression model for the skin aging phenotype (Y), given by

Y = β₀ + β₁ RS_RNAseq + β₂ RS_proteome + β₃ RS_secretome .

The risk scores are weighted using regularised regression procedures. Besides the lasso or the elastic net, we will also explore Random Forests (in collaboration with project R3). For SNP data, we have been able to show that this method is well suited for handling many highly correlated data sets with simultaneous variable selection (Hüls et al., 2017a). This will be investigated for the other omics data types accordingly.

The model will be extended in various ways. Due to the aging process, risk scores might differ between different age groups and will be considered age-dependent. This can be reflected by formulating a hierarchical mixed model that generalises the regression model above. Moreover, interactions between intrinsic and extrinsic factors might play an important role and will be included for each of the risk scores. Furthermore, interactions between the different risk scores for the association with the clinical phenotype aged human skin will be investigated. We will extend the approach to additionally include existing pathway or network information from external databases in the weighting of the omics risk scores as part of a Bayesian regression analysis.

• Hüls A, Ickstadt K, Schikowski T, Krämer U (2017a). Detection of gene-environment interactions in the presence of linkage disequilibrium and noise by using genetic risk scores with internal weights from elastic net regression. BMC Genet 18, 55, doi: 10.1186/s12863-017-0519-1.
• Hüls A, Krämer U, Carlsten C, Schikowski T, Ickstadt K, Schwender H (2017b). Comparison of weighting approaches for genetic risk scores in gene-environment interaction studies. BMC Genet 18, 115, doi: 10.1186/s12863-017-0586-3.
• Waldera-Lupa DM, Kalfalah F, Safferling K, Boukamp P, Poschmann G, Volpi E, Götz-Rösch C, Bernerd F, Haag L, Huebenthal U, Fritsche E, Boege F, Grabe N, Tigges J, Stühler K, Krutmann J (2015). Characterization of skin aging-associated secreted proteins (SAASP) produced by dermal fibroblasts isolated from intrinsically aged human skin. J Invest Dermatol 135(8), 1954-1968, doi: 10.1038/jid.2015.120.