Directed evolution
Copy the gene with a polymerase made deliberately inaccurate (manganese in place of some magnesium, unbalanced dNTPs, a low-fidelity enzyme or simply many doublings) so that each product molecule carries a few random substitutions. The rate is set mainly by how many times the template is copied, and is measured by sequencing unselected clones rather than assumed from a table.
Our ruleSimulations prepare you for the bench but don't replace hands-on practice.
Error-prone PCR
Copy the gene with a polymerase made deliberately inaccurate (manganese in place of some magnesium, unbalanced dNTPs, a low-fidelity enzyme or simply many doublings) so that each product molecule carries a few random substitutions. The rate is set mainly by how many times the template is copied, and is measured by sequencing unselected clones rather than assumed from a table.
How it differs from the standard method
- The spectrum is biased: transitions dominate, and A·T to G·C changes are over-represented, so parts of the mutational space are rarely sampled.
- The template is a functional gene, so it must be linear and digested away or the library fills with parent.
- Rolling the rate up to get more diversity is counter-productive past a few mutations per kilobase.
1Breeding a protein
Directed evolution improves a protein without understanding it. You make a large number of variants of its gene, measure or select for the property you want, keep the best, and do it again.
2Why the mutation rate has an optimum
Suppose each nucleotide substitution has some chance of destroying the protein's function and a much smaller chance of improving the property you are after. If substitutions land independently, a gene carries a Poisson number of them, so the fraction of the library that still works falls exponentially with the mutation rate, while the fraction carrying at least one beneficial change rises.
Error-prone PCR, mutator strains, synthesis
Error-prone PCR with Mn²⁺ and biased dNTPs is the standard, with the rate set by the number of doublings. Nucleotide-analogue mutagenesis, mutator strains and site-saturation of chosen codons are alternatives with different spectra.
Shuffling and staggered extension
Once several improved clones exist, recombining them combines useful changes and sheds the passengers that came with them. Fragment-based shuffling and family shuffling across homologues both do this.
One measurement per clone
A plate assay reads hundreds to tens of thousands of clones, with a real number per clone. Low throughput, high information, and the scatter of the assay sets the smallest improvement you can detect.
Survival, growth or sorting
Link the property to survival, growth or a fluorescent signal and you can interrogate 10⁷–10⁹ variants. Enormous throughput, one bit of information per clone, and a constant worry that the cells found a cheat.
3Screening and selection are different experiments
A screen measures every clone; a selection asks every clone a yes-or-no question. The practical difference is throughput against honesty.
4How it goes wrong
Four failures account for most lost campaigns. Parental carry-over: the template of the error-prone PCR is a functional plasmid, so unless it is linear and digested away, a large share of the library is the unmutated parent and your best hits sequence as the starting gene.
Key terms
Practise these as flashcards →The population of variants that actually exists in cells; capped by the transformation, not by the number of different molecules in the tube.
Substitutions per gene, usually quoted per kilobase of coding sequence or as amino-acid changes per protein.
Share of screened clones that are both still functional and measurably improved; the product of two opposing exponentials in mutational load.
Library size divided by the number of clones sampled; at one-fold coverage a sample sees only 63% of the library.
A template switch during reassembly of fragments, which is what lets shuffling combine substitutions from different parents.
A neutral or mildly deleterious substitution that travels with a beneficial one because they are linked on the same clone.
Dependence of a substitution's effect on the rest of the sequence; diminishing-returns epistasis is why later generations deliver less.
How hard a selection is set, best expressed as a fold over what the current population can already do.
How much more often an improved clone survives a round than the background; the number that decides how many rounds are needed.
Recombining a winner against an excess of the wild-type gene, then re-selecting, to strip passenger mutations.
When to use it
- Improving a property nobody can predict from structure: thermostability, activity on a non-natural substrate, solvent or pH tolerance, soluble expression.
- When a measurable or selectable readout exists for the property, at a throughput matched to how rare the improvement is.
- When several improved variants already exist and the question is whether their changes combine, which is a recombination experiment rather than a mutagenesis one.
- Mapping which residues matter, by sequencing the winners of several independent campaigns.
Compared with alternatives
When a specific residue is the hypothesis, change it directly: one clone and a sequence instead of a library and a screen.
The display formats that make selection on binding possible; the usual choice when the property is affinity rather than catalysis.
Overlap-based assembly is how a library is actually built, and keeping the diversity through the cloning step is part of its design.
The transformation is what caps library size, so its efficiency is a design parameter here, not a routine step.
A variant that assays poorly because it expresses poorly is a common false negative; expression has to be controlled or measured.
Sub-techniques
Choosing one here switches the whole page, including the protocol and the lab, to it.
Copy the gene with a polymerase made deliberately inaccurate (manganese in place of some magnesium, unbalanced dNTPs, a low-fidelity enzyme or simply many doublings) so that each product molecule carries a few random substitutions. The rate is set mainly by how many times the template is copied, and is measured by sequencing unselected clones rather than assumed from a table.
- The spectrum is biased: transitions dominate, and A·T to G·C changes are over-represented, so parts of the mutational space are rarely sampled.
- The template is a functional gene, so it must be linear and digested away or the library fills with parent.
- Rolling the rate up to get more diversity is counter-productive past a few mutations per kilobase.