Accurate top protein variant discovery via low-N pick-and-validate machine learning

A strategy to obtain the greatest number of best-performing variants with least amount of experimental effort over the vast combinatorial mutational landscape would have enormous utility in boosting resource producibility for protein engineering. Toward this goal, we present a simple and effective machine learning-based strategy that outperforms other state-of-the-art methods. Our strategy integrates zero-shot prediction and multi-round sampling to direct active learning via experimenting with only a few predicted top variants. We find that four rounds of low-N pick-and-validate sampling of 12 variants for machine learning yielded the best accuracy of up to 92.6% in selecting the true top 1% variants in combinatorial mutant libraries, whereas two rounds of 24 variants can also be used. We demonstrate our strategy in successfully discovering high-performance protein variants from diverse families including the CRISPR-based genome editors, supporting its generalizable application for solving protein engineering tasks. A record of this paper's transparent peer review process is included in the supplemental information.

Resurfaced ZIKV EDIII nanoparticle immunogens elicit neutralizing and protective responses in vivo

Zika virus (ZIKV) is a flavivirus that can cause severe disease, but there are no approved treatments or vaccines. A complication for flavivirus vaccine development is the potential of immunogens to enhance infection via antibody-dependent enhancement (ADE), a process mediated by poorly neutralizing and cross-reactive antibodies. Thus, there is a great need to develop immunogens that minimize the potential to elicit enhancing antibodies. Here we utilized structure-based protein engineering to develop "resurfaced" (rs) ZIKV immunogens based on E glycoprotein domain III (ZDIIIs), in which epitopes bound by variably neutralizing antibodies were masked by combinatorial mutagenesis. We identified one resurfaced ZDIII immunogen (rsZDIII-2.39) that elicited a protective but immune-focused response. Compared to wild type ZDIII, immunization with resurfaced rsZDIII-2.39 protein nanoparticles produced fewer numbers of ZIKV EDIII antigen-reactive B cells and elicited serum that had a lower magnitude of induced ADE against dengue virus serotype 1 (DENV1) Our findings enhance our understanding of the structural and functional determinants of antibody protection against ZIKV.

Antigenic evolution of human influenza H3N2 neuraminidase is constrained by charge balancing

As one of the main influenza antigens, neuraminidase (NA) in H3N2 virus has evolved extensively for more than 50 years due to continuous immune pressure. While NA has recently emerged as an effective vaccine target, biophysical constraints on the antigenic evolution of NA remain largely elusive. Here, we apply combinatorial mutagenesis and next-generation sequencing to characterize the local fitness landscape in an antigenic region of NA in six different human H3N2 strains that were isolated around 10 years apart. The local fitness landscape correlates well among strains and the pairwise epistasis is highly conserved. Our analysis further demonstrates that local net charge governs the pairwise epistasis in this antigenic region. In addition, we show that residue coevolution in this antigenic region is correlated with the pairwise epistasis between charge states. Overall, this study demonstrates the importance of quantifying epistasis and the underlying biophysical constraint for building a model of influenza evolution.

Facilitating Machine Learning-Guided Protein Engineering with Smart Library Design and Massively Parallel Assays

Protein design plays an important role in recent medical advances from antibody therapy to vaccine design. Typically, exhaustive mutational screens or directed evolution experiments are used for the identification of the best design or for improvements to the wild-type variant. Even with a high-throughput screening on pooled libraries and Next-Generation Sequencing to boost the scale of read-outs, surveying all the variants with combinatorial mutations for their empirical fitness scores is still of magnitudes beyond the capacity of existing experimental settings. To tackle this challenge, in-silico approaches using machine learning to predict the fitness of novel variants based on a subset of empirical measurements are now employed. These machine learning models turn out to be useful in many cases, with the premise that the experimentally determined fitness scores and the amino-acid descriptors of the models are informative. The machine learning models can guide the search for the highest fitness variants, resolve complex epistatic relationships, and highlight bio-physical rules for protein folding. Using machine learning-guided approaches, researchers can build more focused libraries, thus relieving themselves from labor-intensive screens and fast-tracking the optimization process. Here, we describe the current advances in massive-scale variant screens, and how machine learning and mutagenesis strategies can be integrated to accelerate protein engineering. More specifically, we examine strategies to make screens more economical, informative, and effective in discovery of useful variants.

Informed training set design enables efficient machine learning-assisted directed protein evolution

Directed evolution of proteins often involves a greedy optimization in which the mutation in the highest-fitness variant identified in each round of single-site mutagenesis is fixed. The efficiency of such a single-step greedy walk depends on the order in which beneficial mutations are identified-the process is path dependent. Here, we investigate and optimize a path-independent machine learning-assisted directed evolution (MLDE) protocol that allows in silico screening of full combinatorial libraries. In particular, we evaluate the importance of different protein encoding strategies, training procedures, models, and training set design strategies on MLDE outcome, finding the most important consideration to be the implementation of strategies that reduce inclusion of minimally informative "holes" (protein variants with zero or extremely low fitness) in training data. When applied to an epistatic, hole-filled, four-site combinatorial fitness landscape, our optimized protocol achieved the global fitness maximum up to 81-fold more frequently than single-step greedy optimization. A record of this paper's transparent peer review process is included in the supplemental information.

Optimization of multi-site nicking mutagenesis for generation of large, user-defined combinatorial libraries

Generating combinatorial libraries of specific sets of mutations are essential for addressing protein engineering questions involving contingency in molecular evolution, epistatic relationships between mutations, as well as functional antibody and enzyme engineering. Here we present optimization of a combinatorial mutagenesis method involving template-based nicking mutagenesis, which allows for the generation of libraries with >99% coverage for tens of thousands of user-defined variants. The non-optimized method resulted in low library coverage, which could be rationalized by a model of oligonucleotide annealing bias resulting from the nucleotide mismatch free-energy difference between mutagenic oligo and template. The optimized method mitigated this thermodynamic bias using longer primer sets and faster annealing conditions. Our updated method, applied to two antibody fragments, delivered between 99.0% (32451/32768 library members) to >99.9% coverage (32757/32768) for our desired libraries in 2 days and at an approximate 140-fold sequencing depth of coverage.

Identification and iterative combinatorial mutagenesis of a new naringinase-producing strain, Aspergillus tubingensis MN589840

Naringinase was mainly obtained by microbial fermentation, and mutagenesis was a major way for obtaining excellent mutants. The aim of this study was to screen out a high naringinase yielding mutant to enhance the potential application value of its industrialization and compare the effects of different mutagenic methods on the enzyme activity of the strain. A novel producing naringinase strain, Aspergillus tubingensis MN589840, was isolated from mildewed pomelo peel, later subjected to mutagenesis including UV, ARTP and UV-ARTP. After five rounds iterative mutagenesis, the mutants U1, A6 and UA13 were screened out with 1448·49, 1848·71, 2475·16 U mg^-1 enzyme activity, the naringinase productivity raised by 79·08, 123·56 and 206%, respectively. In addition, the naringinase activity of three mutants rose after each round of iterative mutagenesis. These results indicated that the mutagenesis efficiency of UV-ARTP was higher than that of single ARTP, and both are better than UV. In summary, the iterative UV-ARTP mutagenesis is an effective strategy for screening high naringinase-producing strains.

Escapement mechanisms: Efficient free energy transduction by reciprocally-coupled gating

Conversion of the free energy of NTP hydrolysis efficiently into mechanical work and/or information by transducing enzymes sustains living systems far from equilibrium, and so has been of interest for many decades. Detailed molecular mechanisms, however, remain puzzling and incomplete. We previously reported that catalysis of tryptophan activation by tryptophanyl-tRNA synthetase, TrpRS, requires relative domain motion to re-position the catalytic Mg²⁺ ion, noting the analogy between that conditional hydrolysis of ATP and the escapement mechanism of a mechanical clock. The escapement allows the time-keeping mechanism to advance discretely, one gear at a time, if and only if the pendulum swings, thereby converting energy from the weight driving the pendulum into rotation of the hands. Coupling of catalysis to domain motion, however, mimics only half of the escapement mechanism, suggesting that domain motion may also be reciprocally coupled to catalysis, completing the escapement metaphor. Computational studies of the free energy surface restraining the domain motion later confirmed that reciprocal coupling: the catalytic domain motion is thermodynamically unfavorable unless the PPi product is released from the active site. These two conditional phenomena-demonstrated together only for the TrpRS mechanism-function as reciprocally-coupled gates. As we and others have noted, such an escapement mechanism is essential to the efficient transduction of NTP hydrolysis free energy into other useful forms of mechanical or chemical work and/or information. Some implementation of both gating mechanisms-catalysis by domain motion and domain motion by catalysis-will thus likely be found in many other systems.

Pareto Optimization of Combinatorial Mutagenesis Libraries

In order to increase the hit rate of discovering diverse, beneficial protein variants via high-throughput screening, we have developed a computational method to optimize combinatorial mutagenesis libraries for overall enrichment in two distinct properties of interest. Given scoring functions for evaluating individual variants, POCoM (Pareto Optimal Combinatorial Mutagenesis) scores entire libraries in terms of averages over their constituent members, and designs optimal libraries as sets of mutations whose combinations make the best trade-offs between average scores. This represents the first general-purpose method to directly design combinatorial libraries for multiple objectives characterizing their constituent members. Despite being rigorous in mapping out the Pareto frontier, it is also very fast even for very large libraries (e.g., designing 30 mutation, billion-member libraries in only hours). We here instantiate POCoM with scores based on a target's protein structure and its homologs' sequences, enabling the design of libraries containing variants balancing these two important yet quite different types of information. We demonstrate POCoM's generality and power in case study applications to green fluorescent protein, cytochrome P450, and β-lactamase. Analysis of the POCoM library designs provides insights into the trade-offs between structure- and sequence-based scores, as well as the impacts of experimental constraints on library designs. POCoM libraries incorporate mutations that have previously been found favorable experimentally, while diversifying the contexts in which these mutations are situated and maintaining overall variant quality.

Combinatorial Evolution of Enzymes and Synthetic Pathways Using One-Step PCR

DNA engineering is the fundamental motive driving the rapid development of modern biotechnology. Here, we present a versatile evolution method termed "rapidly efficient combinatorial oligonucleotides for directed evolution" (RECODE) for rapidly introducing multiple combinatorial mutations to the target DNA by combined action of a thermostable high-fidelity DNA polymerase and a thermostable DNA Ligase in one reaction system. By applying this method, we rapidly constructed a variant library of the rpoS promoters (with activity of 8-460%), generated a novel heparinase from the highly specific leech hyaluronidase (with more than 30 mutant residues) and optimized the heme biosynthetic pathway by combinatorial evolution of regulatory elements and pathway enzymes (2500 ± 120 mg L(-1) with 20-fold increase). The simple RECODE method enabled researchers the unparalleled ability to efficiently create diverse mutant libraries for rapid evolution and optimization of enzymes and synthetic pathways.

Combinatorial active-site variants confer sustained clavulanate resistance in BlaC β-lactamase from Mycobacterium tuberculosis

Bacterial resistance to β-lactam antibiotics is a global issue threatening the success of infectious disease treatments worldwide. Mycobacterium tuberculosis has been particularly resilient to β-lactam treatment, primarily due to the chromosomally encoded BlaC β-lactamase, a broad-spectrum hydrolase that renders ineffective the vast majority of relevant β-lactam compounds currently in use. Recent laboratory and clinical studies have nevertheless shown that specific β-lactam-BlaC inhibitor combinations can be used to inhibit the growth of extensively drug-resistant strains of M. tuberculosis, effectively offering new tools for combined treatment regimens against resistant strains. In the present work, we performed combinatorial active-site replacements in BlaC to demonstrate that specific inhibitor-resistant (IRT) substitutions at positions 69, 130, 220, and/or 234 can act synergistically to yield active-site variants with several thousand fold greater in vitro resistance to clavulanate, the most common clinical β-lactamase inhibitor. While most single and double variants remain sensitive to clavulanate, double mutants R220S-K234R and S130G-K234R are substantially less affected by time-dependent clavulanate inactivation, showing residual β-lactam hydrolytic activities of 46% and 83% after 24 h incubation with a clinically relevant inhibitor concentration (5 μg/ml, 25 µM). These results demonstrate that active-site alterations in BlaC yield resistant variants that remain active and stable over prolonged bacterial generation times compatible with mycobacterial proliferation. These results also emphasize the formidable adaptive potential of inhibitor-resistant substitutions in β-lactamases, potentially casting a shadow on specific β-lactam-BlaC inhibitor combination treatments against M. tuberculosis.