Fig. 5: Multidimensional epistasis shapes the formation of a Φ–X–Φ motif in the D-domain.

a Heatmaps are shown for subsets of all 29,563 active sequences, which either include an isoleucine/leucine (Φ) at positions 6 and 7a, 7a and 9, 9 and 11, or 11 and 13 (top row), or the inverse, including anything but isoleucine/leucine (Φ) at those positions. Subsets that contain a Φ–X–Φ motif have washed out colours in the other positions indicating that the sequence preference the other positions is relatively weak. When anything but Φ is present in these subsets, strong enrichment for Φ is observed at the other randomised positions. b The heatmap in the top left shows the enrichment for all 29,563 variants. All other heatmaps are epistasis plots, which like (a), are comprised of subsets of the 29,563 active sequences. Here, the amino acid frequencies in the chosen subset of variant (containing/excluding the Φ–X–Φ motif) are divided by the amino acid frequencies in the full set of active variants; consequently, the heatmaps show the change in preference compared to the full dataset. The top row shows that when the Φ-X-Φ residues are included, there is an epistatic preference for non-hydrophobic residues (Ala, Gly, Pro, Tyr, Asp or Lys) in the rest of the D-domain. Conversely, if we examine the variants without the Φ–X–Φ motif on one end, there is an additional preference for the motif elsewhere in the domain, beyond the general enrichment for Leu/Ile residues.