*4.4. Clustering Analysis*

We use 2D-RMSD hierarchical clustering using a single linkage scheme within cpptraj [53] and report the centroid structure of each cluster and its population as representative of the clusters. We used the last 500ns of each replica, aligning on the protein (*C<sup>α</sup>* and *C<sup>β</sup>* atoms) and clustering on the overlapping peptide residues (*C<sup>α</sup>* and *C<sup>β</sup>* atoms). For Figure 2, the lowest temperature replica and all replicas for the clustering of the funnel plots, both with *ǫ* =1.5. For those in Figures 2 and A1–A4, we increase *ǫ* to 2.0, to depict more diverse clusters.
