Every tRNA ends in the same three bases, and the ribosome checks for them. Bases G2251 and G2553 in the 23S rRNA of the E. coli ribosome pair with that 3' CCA end [4], and those Watson-Crick contacts in the large subunit are what stop a tRNA carrying any other 3' sequence from being accommodated during translation [3]. Change the subunit and you have changed the lock rather than the genome: pools of odd-ended tRNAs can then specify a code of their own on ribosomes built to accept them [22].
The reason to bother is a housekeeping problem in cell-free translation. Lysate systems, made from crude cell extracts, keep the host's translation and modification enzymes and scale for bioproduction, but they also carry the host's complete tRNA set, which crosstalks with any reassigned sense codon [11]. The tidy alternative, PURE, rebuilds the reaction from individually purified components, which is where most proof-of-concept code expansion has been done, and it is the poorer choice at preparative scale [12]. A ribosome that will not accept CCA ends renders the host's own tRNAs invisible to it, so the lysate can stay in the tube.
The load-bearing surprise sits one step upstream, at charging. The team found the tRNA 3' end far more tolerant of mutation than assumed, with most odd-ended tRNAs still aminoacylated by all of E. coli's natural synthetases [5]. Code expansion has until now leaned on engineered synthetases alongside engineered ribosomes and tRNAs [14]. Here the enzymes arrive as found, which is the part that removes work rather than adding a step.
Where 34 comes from is worth doing on paper. The standard code spends 64 codons on 20 canonical amino acids [10]; compress it to 20 sense codons plus one start and one stop, and 14 codons come free for non-standard building blocks [8]. Twenty plus fourteen is the 34 [9]: the natural alphabet kept, with fourteen open slots beside it. Church puts the old rate at a decade per amino acid added [18], and his lab's own record, nine years to free the first codon and ten more for the second, works out at about nine and a half years each [17].
The point of fourteen slots is not one exotic residue in one protein. More than 400 non-standard amino acids have already been installed in proteins in living cells [13]. What has not worked is several of them in a single chain at high purity [15], and that, rather than a slightly decorated enzyme, is what a compressed code and an automated, multiplexed prototyping loop [2] are pointed at.
