Customer Insight
Jack Howland¹; Amy A. Caudy¹; Baljit K. Ubhi¹; Katherine Raiche¹; Henry Massey²; Markus Englert²; Ozge Buldag²; Faye Rodgers²; Daniel de la Torre²
¹Matterworks Inc., Somerville, MA | ²Constructive Bio, Cambridge, UK
Nature’s genetic code uses 64 codons to encode just 20 amino acids, leaving substantial built-in redundancy. Constructive Bio has rewritten the genetic code of E. coli to take advantage of that redundancy, reassigning the vacated codons to encode non-canonical amino acids (ncAAs) [1]. This recoding enables the recombinant production of therapeutics with novel properties that were previously impossible to manufacture sustainably [1].
The Syn61 chassis replaces 18,214 codons in the E. coli genome with synonymous codons while encoding the same proteome as the wild-type strain [1]. However, genome-wide recoding at this scale can result in idiosyncrasies in gene expression. Standard strain-engineering playbooks, built around protein-level changes, have no established procedure for finding the cause when the proteome itself hasn’t changed.
Constructive Bio needed to produce ncAA-containing peptides at industrial titers, but Syn61 was not yet performing at the level its proteome would predict. The proteome was unchanged from the wild-type strain, so whatever was limiting productivity had to be found elsewhere. Constructive Bio engaged Matterworks to help diagnose these idiosyncrasies.
Media and cell-pellet samples were collected across 24-hour fermentations of Syn61 and MDS42 (wild-type) at the early exponential, late exponential, and stationary phases. Each sample was extracted in a solvent optimized for small polar biomolecules and combined with Matterworks’s universal calibrants. Samples were run on a Thermo Scientific Vanquish Horizon UHPLC system coupled to an Orbitrap Exploris 120. All data was processed with Pyxis, Matterworks’ co-scientist for omic data, which interprets raw spectra directly and returns metabolite identities and untargeted concentrations without per-analyte standards, method development, or calibration curves. This produced untargeted concentrations for 154 metabolites across every timepoint and both strains.
Uniform Manifold Approximation (UMAP) on all untargeted metabolite concentrations showed the wild-type and recoded strains separating cleanly at every timepoint sampled (Figure 1). Even though the two strains’ genomes encode the same proteome, there were large metabolic differences between them.
Figure 1. Uniform Manifold Approximation (UMAP) on all untargeted metabolic concentrations shows separation of the wild-type and recoded strains at all timepoints. This points to a large metabolic difference between the strains, despite their genomes encoding the same protein complement. EE = early exponential; LX = late exponential; SP = stationary phase.
Comparing metabolite abundance between the two strains in late exponential phase sharpened that picture: of 154 metabolites with untargeted concentrations, 64 showed an absolute log2 fold-change greater than 1.5 between strains at this single timepoint (Figure 2), concentrated in central carbon metabolism and redox-associated pools.
Figure 2. Comparison of metabolite abundance between the wild-type and recoded strains in late exponential phase. Untargeted concentrations were obtained for 154 metabolites, of which 64 showed significant absolute log2FC (>1.5) between strains at this timepoint.
Together with that broader metabolic profile, depleted NAD⁺:NADH ratios and accumulation of NADH in Syn61 pointed to a buildup of excess reductive capacity in the pre-optimized recoded strain (Figure 3), a pattern consistent with a redox imbalance rather than a simple loss-of-function change.
Figure 3. Together with the broader metabolic profile described in Figure 2, depleted NAD⁺:NADH ratios and accumulation of NADH in Syn61 indicate accumulation of excess reductive capacity in the pre-optimized recoded strain.
Constructive Bio overlaid that metabolomics signature with comparative ribosome profiling of the wild-type and recoded strains, identifying two genes that showed altered translational dynamics, including ribosomal stalling, in the recoded strain but not in its wild-type comparator (Figure 4). Both genes encode components of central carbon metabolism, exactly where the depleted NAD+:NADH ratio Pyxis flagged would be expected to originate.
Figure 4. Candidate genetic causes behind the observed metabolic differences were identified through comparative ribosome profiling of the wild-type vs. recoded strains. Altered translational dynamics due to recoding were observed in two genes that encode components of central carbon metabolism. The figure illustrates ribosomal stalling observed in recoded gene x (anonymized for IP protection) and absent from its wild-type comparator.
Alternative recoded sequences for these two genes, designed to restore wild-type-like expression, were introduced into Syn61 via ssDNA oligo recombineering. This single round of targeted optimization significantly increased endpoint OD after 18 hours of shake flask growth (Figure 5).
Figure 5. Two genes involved in central carbon metabolism were identified as having translational idiosyncrasies due to recoding. Alternative recoded sequences for these genes, aimed at restoring wild-type-like expression, were introduced via ssDNA oligo recombineering. After this single round of optimization, endpoint ODs were significantly increased after 18h of shake flask growth. Plot shows mean ± standard deviation of 2 replicates.
New synthetic genome designs may disrupt key cellular processes that affect a cell’s growth and productivity and require genetic fine-tuning to resolve. Pyxis converts raw LC-MS data into metabolite identities and untargeted concentrations within minutes of upload, delivering rich, rapid, and actionable insight without per-analyte calibration curves and dataset-specific method development. Here, combined with ribosome profiling, that insight turned a genome-wide metabolic signature into a short list of genetic targets, and a single engineering round into a measurable gain in productivity.
Fredens, J., Wang, K., de la Torre, D. et al. Total synthesis of Escherichia coli with a recoded genome. Nature 569, 514-518 (2019). doi:10.1038/s41586-019-1192-5