Engineering indel and substitution variants of diverse and ancient enzymes using Graphical Representation of Ancestral Sequence Predictions (GRASP)

Abstract

Ancestral sequence reconstruction is a technique that is gaining widespread use in molecular evolution studies and protein engineering. Accurate reconstruction requires the ability to handle appropriately large numbers of sequences, as well as insertion and deletion (indel) events, but available approaches exhibit limitations. To address these limitations, we developed Graphical Representation of Ancestral Sequence Predictions (GRASP), which efficiently implements maximum likelihood methods to enable the inference of ancestors of families with more than 10,000 members. GRASP implements partial order graphs (POGs) to represent and infer insertion and deletion events across ancestors, enabling the identification of building blocks for protein engineering. To validate the capacity to engineer novel proteins from realistic data, we predicted ancestor sequences across three distinct enzyme families: glucose-methanol-choline (GMC) oxidoreductases, cytochromes P450, and dihydroxy/sugar acid dehydratases (DHAD). All tested ancestors demonstrated enzymatic activity. Our study demonstrates the ability of GRASP (1) to support large data sets over 10,000 sequences and (2) to employ insertions and deletions to identify building blocks for engineering biologically active ancestors, by exploring variation over evolutionary time.

Evidence linking APOBEC3B genesis and evolution of innate immune antagonism by gamma-herpesvirus ribonucleotide reductases

“…To gain deeper insights into the interaction between A3B and gamma-herpesvirus RNRs, we generated a phylogenetic tree of all publicly available primate A3B sequences and used the GRASP-suite ( Foley et al, 2022 ) to reconstruct the amino acid sequence of ancestral A3B ( Figure 8A and Figure 8—figure supplement 1 ). We then examined an expanded panel of primate gamma-herpesvirus RNRs (i.e., 11 different viral proteins) for a capacity to bind human and ancestral A3Bctd in co-IP experiments ( Figure 8B ).…”

“…To generate the phylogenetic tree in Figure 8A , primate A3B amino acid sequences were aligned using MUSCLE ( Edgar, 2004 ; Figure 8—figure supplement 1 ) and a maximum likelihood phylogeny was inferred using PhyML 3.0 ( Guindon et al, 2010 ) using the SMS tool ( Lefort et al, 2017 ) and 100 bootstraps. Reconstruction of the ancestral A3B protein sequence was performed with GRASP ( Foley et al, 2022 ) using the sequence alignment and phylogenetic tree above.…”

Computational Scoring and Experimental Evaluation of Enzymes Generated by Neural Networks

“…In this work, we consider a successful experimental validation as a protein which can be expressed and folded in E. coli and, after purification, have detectable activity significantly above the background in an in vitro assay (Methods). We focused on testing three distinct protein sequence generative models, namely, a transformer-based protein multiple sequence alignment language model ESM-MSA 46 , a convolutional neural network with attention trained as a generative adversarial network (GAN/ProteinGAN) 15 , and a phylogeny-based statistical model for the reconstruction of ancestral sequences (ASR) 26 . The latter is not a truly generative sequence model as it is constrained within a phylogeny of trees by traversing backwards in evolution without the ability to navigate sequence space in a novel direction, but it has been successfully applied for resurrecting ancient sequences 47 and increasing enzyme thermotolerance 48 .…”

“…A total of 144 selected passing and control sequences were expressed in E. coli, purified, and assayed for activity in the same way as in previous rounds (Supplementary Figs. [25][26][27][28][29][30]. Most of the sequences the filter selected showed in vitro activity, including 94% (17 out of 18) and 100% of ESM-MSA CuSOD and MDH enzymes, respectively (Fig.…”

“…The underlying assumption of generative protein models is that natural proteins are under evolutionary pressure to be functional, so novel sequences drawn from the same distribution will also be functional 17 . Multiple different generative protein models have been proposed, including methods based on deep neural networks such as generative adversarial networks (GANs) 15 , variational auto-encoders (VAEs) 16,18 , and language models 13,14,[19][20][21][22] , and other neural networks 23,24 , as well as statistical methods such as ancestral sequence reconstruction (ASR) 25,26 and direct coupling analysis (DCA) [27][28][29] . However, comparing the performance of these methods, i.e.…”