De novo protein design for catalysis of any desired chemical reaction is a long standing goal in protein engineering, due to the broad spectrum of technological, scientific and medical applications. Currently, mapping protein sequence to protein function is, however, neither computationionally nor experimentally tangible 1,2 . Here we developed ProteinGAN, a specialised variant of the generative adversarial network 3 that is able to 'learn' natural protein sequence diversity and enables the generation of functional protein sequences. ProteinGAN learns the evolutionary relationships of protein sequences directly from the complex multidimensional amino acid sequence space and creates new, highly diverse sequence variants with natural-like physical properties. Using malate dehydrogenase as a template enzyme, we show that 24% of the ProteinGAN-generated and experimentally tested sequences are soluble and display wild-type level catalytic activity in the tested conditions in vitro , even in highly mutated (>100 mutations) sequences.ProteinGAN therefore demonstrates the potential of artificial intelligence to rapidly generate highly diverse novel functional proteins within the allowed biological constraints of the sequence space.

The folate and methionine cycles are crucial for biosynthesis of lipids, nucleotides and proteins, and production of the methyl donor S-adenosylmethionine (SAM). 5,10-methylenetetrahydrofolate reductase (MTHFR) represents a key regulatory connection between these cycles, generating 5-methyltetrahydrofolate for initiation of the methionine cycle, and undergoing allosteric inhibition by its end product SAM. Our 2.5 Å resolution crystal structure of human MTHFR reveals a unique architecture, appending the well-conserved catalytic TIM-barrel to a eukaryote-only SAM-binding domain. The latter domain of novel fold provides the predominant interface for MTHFR homo-dimerization, positioning the N-terminal serine-rich phosphorylation region near the C-terminal SAM-binding domain. This explains how MTHFR phosphorylation, identified on 11 N-terminal residues (16 in total), increases sensitivity to SAM binding and inhibition. Finally, we demonstrate that the 25-amino-acid inter-domain linker enables conformational plasticity and propose it to be a key mediator of SAM regulation. Together, these results provide insight into the molecular regulation of MTHFR.

Abstract5′-aminolevulinate synthase (ALAS) catalyzes the first step in heme biosynthesis, generating 5′-aminolevulinate from glycine and succinyl-CoA. Inherited frameshift indel mutations of human erythroid-specific isozyme ALAS2, within a C-terminal (Ct) extension of its catalytic core that is only present in higher eukaryotes, lead to gain-of-function X-linked protoporphyria (XLP). Here, we report the human ALAS2 crystal structure, revealing that its Ct-extension folds onto the catalytic core, sits atop the active site, and precludes binding of substrate succinyl-CoA. The Ct-extension is therefore an autoinhibitory element that must re-orient during catalysis, as supported by molecular dynamics simulations. Our data explain how Ct deletions in XLP alleviate autoinhibition and increase enzyme activity. Crystallography-based fragment screening reveals a binding hotspot around the Ct-extension, where fragments interfere with the Ct conformational dynamics and inhibit ALAS2 activity. These fragments represent a starting point to develop ALAS2 inhibitors as substrate reduction therapy for porphyria disorders that accumulate toxic heme intermediates.