In a perspective article published in Frontiers in Industrial Microbiology, Laura R. Jarboe of Iowa State University examined the alleleome of E. coli, the collective record of amino acid substitutions accumulated across hundreds of laboratory evolution experiments. Drawing on the ALEdb database and decades of accumulated mutation data, including the celebrated long-term evolution experiments, the analysis distilled roughly 33,000 documented mutations down to 25,286 unique amino acid substitutions drawn from 278 distinct E. coli experiments. The question was straightforward: of all the amino acid swaps that are biochemically possible, how many has laboratory evolution actually sampled?

The answer is sobering. There are 190 possible non-synonymous exchanges between the twenty standard amino acids, yet only 88 of them, less than half, appear anywhere in the dataset. Even more striking is how lopsided the sampling is. The interconversion of alanine and valine, and of alanine and threonine, together account for more than 18 percent of all observed amino acid shifts. The rest of the possible substitution space remains largely untouched, a vast and silent continent of protein chemistry that short-term laboratory evolution has simply never visited.

Part of the explanation lies in the architecture of the genetic code itself. Some amino acid substitutions require only a single nucleotide change within a codon, a single hop from one codon to another, while others demand two or even three simultaneous nucleotide changes. The type of change matters as well. Transition mutations, which swap one purine for another or one pyrimidine for another, occur far more frequently than transversions, which cross between purine and pyrimidine. Of the 192 possible single-transition codon shifts, only 116 produce a non-synonymous change in E. coli, because many are synonymous and leave the encoded amino acid untouched.

The numbers bear this out with remarkable precision. Amino acid shifts achievable through a single transition mutation account for 14,225 of the observed substitutions, or 56.26 percent of the dataset. Single transversions account for another 10,766, or 42.58 percent. Substitutions requiring two nucleotide changes are vanishingly rare, just 295 observations, or 1.17 percent of the total, and not a single substitution requiring three mutations was observed anywhere in the data. The rarity of multi-hop mutations maps directly onto the sparse sampling of certain amino acid exchanges, suggesting that the genetic code’s structure imposes a hard ceiling on what laboratory evolution can reach in a realistic number of generations.

Yet codon accessibility is not the whole story. Some poorly sampled substitutions cannot be explained by the number of hops required. The shift from arginine to threonine, for example, can be accomplished by two different single-transversion mutations, from AGA to ACA or from AGG to ACG, yet neither was observed. The reason appears to be codon usage: those two arginine codons occur at frequencies of only 2.05 and 1.22 per thousand codons in the E. coli genome, making the necessary starting points inherently scarce. More broadly, the observed frequency of single-hop mutations generally tracks the frequency of the source codon, meaning that the genome’s own composition shapes which evolutionary paths are traveled most often.

The analysis also raises a tantalizing possibility: some of the unexplored substitutions might actually be useful. Jarboe points to amino acid exchangeability scores derived from experimental data, where high scores indicate substitutions that tend to be gentle on protein function. A lysine-to-alanine substitution, with an exchangeability score of 600, has been shown to increase the half-life of a glucanase enzyme at 70 degrees Celsius by more than fivefold. A glutamine-to-valine substitution, scoring 603, boosted the NADH-dependent catalytic efficiency of ketol-acid reductoisomerase by more than 80-fold. Neither of these high-exchangeability shifts appears in the E. coli alleleome data, hinting that at least some of the under-sampled amino acid exchanges could improve protein or enzyme performance, even if they might not confer resistance to the specific selective pressures applied in the laboratory.

To test whether short-term laboratory evolution follows the same strategies as evolution on geological timescales, Jarboe compared the E. coli thermotolerance literature with a landmark study by Pinney and colleagues, who analyzed 14,399 site-specific amino acid changes across organisms whose growth temperatures span eons of divergent evolution. That study identified 26 amino acid shifts with significant directionality linked to high growth temperature, such as the strong enrichment of isoleucine over leucine in high-temperature homologs. Six E. coli thermotolerance ALE studies, spanning 2012 to 2021 and up to roughly 5,000 generations, contributed 685 unique amino acid shifts expected to support heat tolerance. If laboratory evolution recapitulates nature’s long-term playbook, the same directional shifts should appear.

They largely do not. Using Pinney’s significance threshold of a p-value below 5.3 times ten to the minus fifth, derived from a Bonferroni correction across all 190 amino acid pairs, no shift in the E. coli thermotolerance dataset qualified. Under the more permissive Benjamini-Hochberg correction, three pairs showed enrichment, but only the aspartate-to-glutamate shift pointed in the same direction as the organism-wide comparison, while the glutamine-to-lysine and phenylalanine-to-isoleucine shifts were enriched in the opposite direction. Thirteen of Pinney’s 26 significant shifts require two or three nucleotide hops and were never observed in the ALE strains, plausibly because such mutations are simply too rare to arise. The mismatch is not merely an artifact of dataset size: five amino acid pairs in the E. coli data had at least 15 total observed exchanges, enough to satisfy the significance threshold, yet none did.

The takeaway is not that adaptive laboratory evolution is useless; it remains a powerful, knowledge-light method for improving biomanufacturing strains, particularly when redox or ATP maintenance is tied to production of a target molecule. But the analysis makes clear that short-term evolution samples a constrained slice of the available amino acid space, for reasons rooted in the genetic code, mutation biases, and codon frequencies. Complementary strategies exist: directed evolution of individual enzyme targets can access rare substitutions, though it sacrifices ALE’s organism-wide scope by requiring researchers to nominate protein targets in advance. Genome refactoring offers another route, with a proposed strategy to recode 60 of E. coli’s 64 codons to expand evolutionary access, although codon recoding has so far been demonstrated only at a more limited scale. Until such approaches mature, Jarboe argues, researchers should resist relying exclusively on evolution’s cleverness and instead combine multiple strategies for organism improvement, treating laboratory evolution as one tool in a larger kit rather than a shortcut to biology’s full potential.

Subject of Research: Sampling of amino acid substitution space in adaptive laboratory evolution of Escherichia coli

Article Title: Adaptive laboratory evolution of Escherichia coli is clever, but its reach is also relatively short

Article References: Adaptive laboratory evolution of Escherichia coli is clever, but its reach is also relatively short. (n.d.). https://doi.org/10.3389/finmi.2026.1840830

Image Credits: AI Generated

DOI: 10.3389/finmi.2026.1840830

Keywords: adaptive laboratory evolution, Escherichia coli, alleleome, amino acid substitutions, genetic code, thermotolerance, mutation bias, transition mutations, transversion mutations, codon usage, metabolic engineering, directed evolution