Supplementary Figure 1: Data analysis workflow.

First step: Identification of valid datasets. Amplitudes and number of valid maxima are investigated. If more than 6 local maxima are identified the curve is neglected. Second step: Background subtraction. The signal from both buffer references are averaged and subtracted from all individual curves in the dataset. Third step: Normalization. All datasets are normalized to a range of 0 to 1000, by setting the lowest local minimum to 0 and the highest relevant local maximum to 1000. Fourth step: Identification and removal of datasets containing air bubbles. The slope from the first data-point to the first relevant local minimum is evaluated. Curves with large slopes are omitted. This also removes transitions of proteins, which were already aggregated in solution prior to measurement. Fifth step: Data approximation according to the thermodynamic framework presented here for two- to five-state models. Sixth step: Parameter extraction for two-state and best fitting models. The values for T_m, ∆H_m, R² are extracted for the two-state model and the best fitting model. These extracted parameters are then hierarchically sorted to evaluate the most stabilizing conditions.