Periodic Reporting for period 4 - HelixMold (Computational design of novel functions in helical proteins by deviating from ideal geometries)
Reporting period: 2023-10-01 to 2025-03-31
Society faces big biomedical and biotechnological problems e.g. how quickly can we come up with a drug to fight disease, or how we can stop herbicidal compounds from accumulating in our soils? Computational design of novel protein structures is a promising tool to address these challenges. It can be used to make superior biological materials with tailor-made properties, new pharmaceuticals or complex fine chemicals. Tremendous progress has been made over the last couple of years in the field of de novo protein design, which was also recognized by protein design being awarded the 2024 Nobel prize in chemistry. In this project – HelixMold –we started out by leveraging on the ability to design alpha-helical proteins with ultra-fine control of backbone geometries and unprecedented thermodynamic stabilities to design novel enzymes. Moreover, the advent of AI and machine learning assisted protein design tools, some of which we developed ourselves, made these types of design problems more robust and led to revolution in the field during the last two years. Combined with high-throughput biochemical testing, we are using these advancements to investigate the functionalization of parametrically and AI designed de novo proteins. In particular, we aimed at installing diverse catalytic centers into the de novo designed proteins, ranging from metal complexation and single residue active sites to more complex systems requiring build in flexibility and cofactors. Within HelixMold, we are also focusing on developing tools to design proteins with multiple sites of action; this can be more than one binding site for small molecules, or more than one active site to catalyze a reaction. For computationally self-consistent designs, we were conducting in-depth biochemical and biophysical analysis, resulting in several new experimentally determined protein structures with no natural counterpart, active de novo biocatalysts for copper catalyzed reactions (nitrite reductase, amino oxidase and lytic polysaccharide monooxygenase) and retro-aldol as well as Morita-Baylis-Hillman enzymes. Overall, work in this project resulted in some of the first ever datasets on de novo enzymes for various reactions and can now be used in follow up studies to deduce design principle and gain fundamental insight into installing catalytic activity into de novo proteins.
Over the course of the project, we focused on the parametric and machine learning based design of helical proteins with backbones custom fit to harbor binding sites for different ligands as well as an active sites for catalyzing enzymatic reactions. For the initial designs, we used computational protocols that are already in place to sample the folding space of alpha-helical proteins computationally. This allowed us to generate hundreds of thousands of potential starting backbones for proteins with a specific function. In subsequent steps, combinatorial amino acid sequences design calculations were used to get low energy amino acid compositions for the respective backbones. From the multitude of computational designs, several were experimentally tested, and some showed the desired activity. Of the ones that showed our envisioned activity, we could experimentally determine their three-dimensional structures using a technique called X-ray crystallography. The determined structures coincide very well with the design models and thus validated our design process. On the way to these goals, we had to achieve and establish a multitude of additional protocols/experiments. Thus, we have now a semi high-throughput pipeline for testing our designed proteins. In addition, we combined recent machine learning based backbone and sequence design approaches with our established parametric design approach, which led to a set of de novo designed biocatalysts for C-C bond cleavage and formation with activities exceeding previously reported designs by several orders of magnitude. Below is a bullet point list of achievements since the project start:
1. Establishing of a novel enzyme design pipeline, which uses diffusion based back bone design approaches to scaffold artificial catalytic motifs. The artificial catalytic motif approach is a direct result of previous parametric design approaches and has been combined with RosettaFold diffusion (RFdiffusion).
2. Design and characterization of more than 200 de novo proteins in total, which either complex copper to achieve nitrate reduction, or quercitin oxidation, amino oxidation and polysaccharide monooxygenation
3. Biochemical and biophysical characterization (CD, DSF, SAXS, MS, X-ray crystallography) of the designs.
4. Design and characterization of de novo proteins that complex ruthenium half-sandwich complexes and their biochemical and biophysical characterization.
5. Expansion of the parametric design code
6. Construction of a GNN which learns atom identities and their distances, angles and torsions to neighboring atoms assess the quality of 1. Loops connecting any two secondary structure elements of variable length; 2. Design models on a per-residue basis, which provides an additional level of design accuracy assessment on top of structure prediction networks like AlphaFold2/3.
7. An antiparallel single walled six helix bundle, a fold that does not occur in nature and catalyzes the as retro-aldol reaction.
8. Design of de novo three helix bundles that harbor a binding site for b-type heme between their symmetric dimer interface were designed and experimentally characterized
1. Establishing of a novel enzyme design pipeline, which uses diffusion based back bone design approaches to scaffold artificial catalytic motifs. The artificial catalytic motif approach is a direct result of previous parametric design approaches and has been combined with RosettaFold diffusion (RFdiffusion).
2. Design and characterization of more than 200 de novo proteins in total, which either complex copper to achieve nitrate reduction, or quercitin oxidation, amino oxidation and polysaccharide monooxygenation
3. Biochemical and biophysical characterization (CD, DSF, SAXS, MS, X-ray crystallography) of the designs.
4. Design and characterization of de novo proteins that complex ruthenium half-sandwich complexes and their biochemical and biophysical characterization.
5. Expansion of the parametric design code
6. Construction of a GNN which learns atom identities and their distances, angles and torsions to neighboring atoms assess the quality of 1. Loops connecting any two secondary structure elements of variable length; 2. Design models on a per-residue basis, which provides an additional level of design accuracy assessment on top of structure prediction networks like AlphaFold2/3.
7. An antiparallel single walled six helix bundle, a fold that does not occur in nature and catalyzes the as retro-aldol reaction.
8. Design of de novo three helix bundles that harbor a binding site for b-type heme between their symmetric dimer interface were designed and experimentally characterized
Since the start of the project, lots of time was invested to build streamlined experimental pipelines for the quick analysis of our designed proteins. HelixMold pushed enzyme design beyond the current state of the art by demonstrating that a pre-organized catalytic motif can be transplanted, essentially “as is,” into entirely new backbones generated by parametric design or/and a diffusion model. The used catalytic model systems delivered million-fold rate enhancements for a high-barrier C–C bond-cleavage and forming reaction—without any laboratory evolution. Previous one-shot designs worked almost exclusively required post-design mutagenesis to reach useful activities. Within HelixMold we could show that a single computational cycle can now capture the geometric necessities needed for catalysis involving multiple side chains and transition states. Methodologically, we managed to unify fragment-based active-site preservation with parametric design and machine learning based backbone generation, offering a generalized and modular approach that can be used for designing biocatalysts for various chemistries, including cofactor dependent reactions. For biotechnology, such plug-and-play creation of bespoke catalysts promises to significantly shorten development timelines from years to months, eventually enabling rapid construction of e.g. one-pot on-demand synthesis of complex fine chemicals, and agile retro-biosynthetic routes that can outpace traditional chemical process development while operating under mild, sustainable conditions.