Extended Data Fig. 10: Analysis of the HsPINK1 model as predicted by AlphaFold2^37,38.

(a, b) Predicting the structure of HsPINK1 via AlphaFold2 (a, left), resulted in a model remarkably similar to PhPINK1 from the PhPINK1–Ub TVLN complex¹⁰ (b, right), and features a kinked αC helix, ordered insertion-3, and an extended N-helix that binds and extends from the CTR directly into the membrane. Consistently, predicting a complex between HsPINK1 with the ubiquitin-like (Ubl) domain of human Parkin (a, right), places the Ubl domain at HsPINK1 analogously to the PhPINK1–Ub TVLN complex (b, right). Both predictions are dissimilar to unphosphorylated PhPINK1 (b, left) and TcPINK1 (Extended Data Fig. 1). Insets show the detail of the N-lobe with a kinked αC helix and an ordered insertion-3. The prediction is somewhat surprising, as it suggests HsPINK1 to be an active ubiquitin kinase even without Ser228 phosphorylation, which contradicts biochemical analysis. Since AlphaFold2 does not yet predict the impact of post-translational modifications, we interpret the prediction such that it is possible and even likely, that HsPINK1 can adopt a ubiquitin-phosphorylation-competent conformation consistent with our previous PhPINK1–Ub TVLN complex structure¹⁰. c, d, We next used AlphaFold2 to predict a dimer of HsPINK1. c, Strikingly, AlphaFold2 predicts a symmetric dimer with a dimer interface identical to the one shown for PhPINK1 (compare with d and Fig. 3). In fact, we have already validated this arrangement of HsPINK1 molecules via our Cys166 cross-linking experiments in Fig. 3g. However, in the predicted dimer of HsPINK1, the αC helix is kinked and Ser228 does not contact the second molecule. This is different from our conclusions in Extended Data Fig. 8, but again may be a result of not incorporating the effect of Ser228 phosphorylation. We therefore anticipate that unphosphorylated HsPINK1 can also adopt a conformation with an extended αC helix that places Ser228 into the active site of the dimeric molecule to facilitate autophosphorylation prior to forming the depicted conformation. AlphaFold2 predictions hence support the notion that the activation model proposed in Fig. 5g applies to HsPINK1.