Rapid and predictive synthesis of membrane proteins ex vivo by Conary Meyer

Conary Meyer

UC Davis

Membrane proteins are critical to all biological life, yet our inability to synthesize them has limited our understanding and use of these proteins. To address this, I am developing an end-to-end pipeline to generate a predictive model and optimization cycle to robustly synthesize membrane proteins, using cell-free protein synthesis, high-throughput microfluidics, and machine learning.


Engineers have been trying to understand and utilize natural membrane proteins to little avail. These proteins catalyze key reactions, selectively transport molecules, and detect chemicals at concentrations orders of magnitude below standard sensors. These natural functions could be exploited for energy harvesting, drug screening, and medical diagnostics. However, a fundamental challenge lies in synthesizing membrane proteins and inserting them into the membrane. Traditionally, membrane proteins are produced using a heterologous expression host, followed by their purification and insertion into pure membranes. However, the new host often lacks the necessary environment to produce and fold the target protein. This in vivo synthesis method also requires challenging detergent solubilization steps that often fail. An emerging alternative is to utilize natural protein-synthesis machinery outside of cells. This cell-free protein synthesis method coupled with synthetic lipid membranes allows for direct manipulation of the protein and chemical components of the reaction mixture and the composition of the membrane. This widely tunable system has enabled the synthesis of several membrane proteins, but this vast parameter space is difficult to search. Essentially, the scientific community lacks a universal model that predicts successful synthesis of membrane proteins in cell-free systems. The fundamental hypothesis is that the features of a membrane protein and its optimal synthesis conditions can be summarized in a statistical representation that enables the predictive synthesis of diverse membrane proteins. The connection of unique protein features (hydrophobic length, topology, charge profile, etc.) and the reactions components (chaperones, lipids, salts, etc.) will be initially derived analytically and later reinforced by empirical data derived connections through machine learning. To train this model, a high throughput reaction generation and analysis pipeline has been built. Combinatorial reaction assembly is conducted using a microfluidic droplet printer which rapidly prints nanoliter scale droplets from a multitude of reagents. These reactions include a folding-dependent fluorescent reporter to assess thousands of unique membrane protein synthesis reactions. An active learning model will choose the next round of conditions to further elucidate unclear parameters and refine the model. This model will allow researchers around the world to synthesize membrane proteins; giving rise to a dramatic increase in our understanding and use of these proteins.


12 + 6 =