Directed evolution of rRNA improves translation kinetics and recombinant protein yield

Development of a PACE-compatible orthogonal translation system

PACE has facilitated the exploration of sequence-function relationships of biomolecules with diverse cellular activities^{10,11,12,13,14,15,16}. Briefly, PACE exploits the rapid M13 bacteriophage lifecycle and couples the production of plasmid-borne gIII, encoding the minor coat protein pIII necessary for both bacterial infection and membrane extrusion¹⁷, to the activity of the evolving biomolecule encoded on a pIII-deficient phage genome. The genetic diversity of the evolving biomolecule is easily tuned through a small molecule-inducible expression of mutator proteins from the mutagenesis plasmid (MP)¹⁸. Historically PACE has been limited to protein-coding genes. We envisioned that PACE could be extended to the directed evolution of orthogonal rRNAs (o-rRNAs), allowing efficient traversal of mutational landscapes and uncovering variants with altered translational activity (Fig. 1a). To establish an o-rRNA PACE selection in E. coli, we adapted an orthogonal translation genetic circuit^19,20 to integrate the M13 bacteriophage gIII (which encodes pIII), yielding the Accessory Plasmid 1 architecture (AP1; Fig. 1b) and concurrently engineered selection phages (SPs) to encode the complementary o-rRNA operon. Functional o-rRNAs capable of forming active ribosomes and translating the gIII mRNA using the o-RBS would robustly produce pIII, yielding infectious phage progeny.

While the previously reported o-antiRBS_B efficiently directs translation of an sfGFP reporter bearing the cognate o-RBS_B sequence (Fig. 1c)⁸, a direct adaptation of o-RBS_B to AP1 rendered S2060¹³ cells uninfectable by wild-type M13 phage, indicating high background pIII expression. Thus, we developed a two-stage selection to identify PACE-compatible o-RBS/o-antiRBS pairs with reduced background translation by host ribosomes. We first introduced a degenerate library of 4⁷ RBS variants (o-RBS_lib, Fig. 1c) and assessed the infectivity of the resultant cells to identify sequences poorly recognized by host ribosomes (Supplementary Fig. 1a). This analysis revealed 33 putative o-RBS candidates, and we further characterized the most abundant seven variants (Supplementary Fig. 1b). To discover potential cognate o-antiRBSs, we introduced a degenerate library of 4⁶ antiRBS variants in the SP-borne E. coli o-rRNA (o-antiRBS_lib, Fig. 1c) into E. coli host cells bearing each of the seven o-RBSs. Functional o-antiRBS sequences should efficiently translate gIII and give rise to progeny phage (Supplementary Fig. 1c–e). After further optimization of spacer sequences (Supplementary Fig. 1f), we identified o-RBS_H3 (Fig. 1c) as the optimal orthogonal sequence for subsequent experiments. We adapted o-RBS_H3 to AP1 (AP1_H3) yielding 40- to 163-fold enrichment for SPs encoding the cognate o-antiRBS_H3 (SP_H3) relative to SPs bearing the mismatched o-antiRBS_B sequence (SP_B) (Fig. 1d).

While the o-RBS_H3/o-antiRBS_H3 pair enabled phage propagation in standing culture (Fig. 1d), we hypothesized that alternative solutions may exist under continuous culture conditions. We continuously propagated a degenerate SP library encoding 4⁶ antiRBSs (o-antiRBS_lib, Fig. 1c) using AP1_H3 in S2060 cells yielding comparable phage titers to SP_H3, while SP_B was rapidly washed out (Fig. 1e). We analyzed the resulting SP populations at 40 h by Sanger sequencing (24 clones) and found that SP_lib converged on exclusively two variants: o-antiRBS_H3-1 and o-antiRBS_H3-2 (Fig. 1c). Both variants robustly translate a LuxAB luciferase reporter, showing a similar dynamic range to the initial o-antiRBS_H3 variant (Supplementary Fig. 1g). We note that o-antiRBS_H3-1, but not o-antiRBS_H3-2, appeared in our initial antiRBS library (Supplementary Fig. 1e), suggesting differential o-ribosome activities may depend on culturing conditions.

Although we successfully identified functional o-antiRBS sequences from an unbiased SP library, the final phage titers were considerably lower than those in previous protein-based PACE campaigns^{10,11,13,14,15,16}. We noted that host cells in the turbidostat reside at the transition between the exponential and stationary phase, during which o-rRNAs may be inactivated by hibernation factors²¹. Accordingly, we deleted factors known to inhibit ribosome activity to improve the propagation of o-rRNA SPs (Fig. 1f). Deletion of ribosome hibernation promoting factor (HPF) from S2060¹³ yielded host strain S3317, with a 3400-fold improvement in SP propagation (Fig. 1g). Concurrently, we prepared an AP architecture, AP2 (Fig. 1b), which encoded a growth phase-independent constitutive promoter²² to simplify o-rRNA evolution experiments (Supplementary Fig. 2a–d) and integrated the pSC101 origin of replication for stringent copy number control²³. When introduced into S3317 cells, AP2_H3 supported SP_H3 propagation 4831-fold more efficiently than the mismatched SP_B (Fig. 1g and Supplementary Fig. 2e).

We next competed all o-antiRBS SPs (Fig. 1b) using S3317/AP2_H3 under continuous flow at varying lagoon dilution rates. We note that low lagoon flow rates (<1.0 vol h⁻¹) led to poor SP propagation, consistent with ribosome inactivation at saturated cell densities (Supplementary Fig. 2f)²¹. We analyzed individual SPs propagated at 2 vol h⁻¹ using Sanger sequencing and found that most SPs encoded o-antiRBS_H3-1, in agreement with the SP_lib evolution experiment (Fig. 1e and Supplementary Fig. 1e). In overnight enrichment assays in standing culture, SP_H3-1 similarly showed improved titers (up to 186-fold) over SP_H3 (Supplementary Fig. 2g). Following additional strain engineering (∆fhuA²⁴) to produce S3489 (Supplementary Fig. 2g, h) and plasmid modification to yield the AP3 architecture (Fig. 1b) to limit AP/SP recombination (Supplementary Fig. 2b, i–l), we found the S3489/AP3_H3/SP_H3-1 combination to be the optimal orthogonal translation system and used it for all subsequent experiments.

Continuous directed evolution of orthogonal ribosomes

We and others have recently shown that rRNAs derived from heterologous microbes can robustly support E. coli viability upon deletion of all host-derived rRNAs^20,25. As only E. coli-derived o-rRNAs have been successfully evolved to date^1,26, we hypothesized that diverse heterologous o-rRNA sequences may undergo distinct evolutionary trajectories in PACE, yielding variable solutions to identical selection conditions. However, divergent heterologous ribosomes often suffer from reduced starting activity in an E. coli chassis as compared to wild-type E. coli ribosomes²⁰. The 16S rRNA is a highly conserved sequence, yet encoding poorly conserved residues often residing at the 3-dimensional periphery of the ribosome (Fig. 2a). To define a threshold for heterologous ribosome activity, we generated deletions in E. coli-derived 16S o-rRNA and characterized their activity levels using reporter and SP enrichment assays (Fig. 2b–f). These experiments established that SPs bearing o-rRNAs with activity levels ≥32% of WT E. coli o-rRNA robustly propagate under stringent conditions.

Next, we identified P. aeruginosa (Pa) and V. cholerae (Vc) heterologous o-rRNAs as promising candidates for oRibo-PACE, as they showed comparable activity to E. coli-derived o-rRNA (Fig. 3a) and could successfully propagate in standing culture, albeit at lower efficiency than their E. coli counterpart (Fig. 3b). To evolve heterologous rRNAs, we subjected starting rRNA species to multi-stage selection regimes with increasing selective pressure. We performed 218 h (~268 generations¹⁰) of PACE using E. coli (SP_Ec), P. aeruginosa (SP_Pa), and V. cholerae (SP_Vc) o-rRNAs while varying selection stringency over multiple segments (Fig. 3b–d). In all segments, we employed a previously optimized MP, MP6¹⁸, to enhance o-rRNA sequence diversity, and regularly increased lagoon flowrates to enhance selection stringency.

In the first segment (S1 = 0–68 h), we diversified the clonal SP-borne o-rRNAs through genetic drift by employing a constitutive promoter driving gIII expression from AP3_H3 (proB²²; Fig. 3b, d). During the second segment (S2 = 68–143 h), we increased selection stringency by reducing the gIII promoter strength 8-fold ((pro4²²); Fig. 3b, d), resulting in a > 250-fold decrease in SP propagation efficiency (Fig. 3b). During the third segment (S3 = 143–218 h), we incorporated a split-intein pIII¹⁴ strategy where an inserted protein sequence increased the effective length of gIII by 123% (425–947 amino acids) and decreased SP propagation efficiency further by >120-fold (Fig. 3c, d and Supplementary Fig. 3a–g). Finally, to examine the effect of selection schedule on o-ribosome variant activities, we carried out a fourth segment (S4 = 68–143 h) using the split-intein pIII approach and SP populations immediately following genetic drift (S1→S4) to compare a shorter selection regime to the aforementioned longer version (S1→S2→S3) (Fig. 3d). We note that all SP populations robustly propagated across all segments, with the exception of SP_Pa during S2 (Fig. 3d and Supplementary Fig. 3f), which rebounded during subsequent high stringency selection, suggesting accumulation of mutations to enable enhanced o-rRNA activities. Furthermore, all three SP populations underwent cognate 23S rRNA deletion at virtually identical time points during oRibo-PACE (Supplementary Fig. 3h), reflecting complementation with the host E. coli 23S rRNA as previously described²⁰.

Individual clone sequencing at the end of each segment revealed sweeping mutations in all SP-borne o-rRNAs (Supplementary Fig. 3i and Supplementary Tables 1–3). Collectively, V. cholerae o-rRNAs developed the highest average number of mutations per clone throughout all segments, while P. aeruginosa o-rRNA retained the lowest number of mutations (Fig. 3e). This trend is consistent with propagation efficiencies of the corresponding SPs during oRibo-PACE (Fig. 3d). A number of unique mutations became prevalent in each SP population at varying segments: C1098U (E. coli, S3), G1415A (E. coli, S1), U409C (V. cholerae, S3), and A434U (P. aeruginosa, S1) (Fig. 4a–c and Supplementary Fig. 3i, Supplementary Tables 1–3). Interestingly, we note varying levels of natural sequence conservation at the discovered sites (Fig. 4d, e)²⁷, suggesting that mutations at these positions may not necessarily indicate functional relevance.

Notably, an identical mutation in h27 was evolved independently in all o-ribosomes at different segments: A906G in E. coli and in V. cholerae (S1), and A900G in P. aeruginosa (S3) (Fig. 4a–c, f and Supplementary Fig. 3i, Supplementary Tables 1–3). Two identical mutations were also found in the E. coli (U904C, G1487A) and V. cholerae (U904C, G1488A) populations (Fig. 4a–c, f and Supplementary Fig. 3i, Supplementary Tables 1–3). A906 and U904 (helix 27, E. coli numbering) together with G1487 (h44) form an interface with protein uS12 (Fig. 4f) during tRNA selection and ensure translation accuracy^28,29,30. The observation of three converged mutations (U904C, A906G, and G1487A) in the E. coli and V. cholerae populations suggests adaptive evolution towards enhanced translational output (Fig. 3d). The E. coli-only mutation C1098U (h37) interacts with r-protein uS2³¹ (Fig. 4f) during the final S30 subunit assembly³², whereas G1415A is proximal to G1487A (h44, Fig. 4f) and may influence tRNA selection. The V. cholerae-only mutation U409C forms a wobble base pair with G433 (h16) interacts with uS4, where its mutation to a cytosine may yield a stronger C409-G433 Watson–Crick pair (Fig. 4f)³¹. The P. aeruginosa-only mutation A434U (h17) is near the binding site of protein uS4³ (Fig. 4f). Taken together, these results showcase hallmarks of both similar and independent evolutionary trajectories to overcome identical selection regimes.

PACE-derived o-rRNAs show augmented translation efficiencies

To assess the consequences of PACE-derived mutations on o-ribosome function, we subcloned evolved o-rRNAs into inducible expression plasmids (EPs) and evaluated their activities in vivo using a battery of assays: (1) characterizing translation rate using orthogonal cellular reporter proteins²⁰, (2) quantifying host E. coli growth burden³³ during o-ribosome overproduction, (3) investigating possible context-dependence effects on the translation by using the unrelated B o-RBS/o-antiRBS system⁸, (4) analyzing preferential use of E. coli host factors by evolved heterologous o-ribosomes via complementation with cognate ribosomal proteins²⁰, (5) exploring improvements in genetic code expansion through non-canonical amino acid (ncAA) incorporation³⁴, and (6) analyzing context independence of evolved consensus mutations in unrelated, divergent heterologous rRNAs comparing everything to starting E. coli o-rRNA under the same conditions (Fig. 5a).

Continuous monitoring of luminescence activity was used as a real-time proxy of the translation rate for o-ribosomes. Using kinetic luminescence output at fixed optical densities (OD₆₀₀ = 0.15, Supplementary Fig. 4a), we observed minimal activity improvements of evolved E. coli o-rRNA variants from S2 and S4. However, the six variants isolated from the longer S3 trajectory showed higher (146–196%) activity compared to the starting E. coli o-rRNA (Fig. 5b–d and Supplementary Fig. 4b). Only a single P. aeruginosa o-rRNA variant evolved after 218 h (S3) of PACE yielded similar activity to starting E. coli (96%) luminescence output (Fig. 5c and Supplementary Fig. 4c). Remarkably, almost half (11/24) of V. cholerae 16S o-rRNA variants produced higher (109–186%) activities relative to E. coli (Fig. 5b–d and Supplementary Fig. 4d). These observed differences in evolved o-rRNA populations, which do not correlate with 16S sequence identity to E. coli (P. aeruginosa: 85%; V. cholerae: 90%), suggest that heterologous rRNA choice may affect directed evolution campaign success by as yet unclear determinants.

Orthogonal ribosomes are known to negatively affect host cell fitness, likely due to over-commitment of resources to the production of supplementary ribosomes (Supplementary Fig. 4e)²⁰. We observed a previously reported burden for E. coli o-rRNA expression on host cells³³, and in some cases, these effects are moderately amplified in evolved variants (Fig. 5b–d and Supplementary Fig. 4e–h). In general, host doubling time increased for o-rRNA mutants with respect to starting o-rRNAs (61/67 mutants; 91%), and this trend was held for o-rRNA mutants with enhanced o-ribosome activity as compared to starting scaffold (49/53 mutants; 92.5%) (Fig. 5b–d). However, expressing wild-type or evolved P. aeruginosa or V. cholerae o-rRNAs exerted a lighter metabolic burden on the E. coli host than expressing the corresponding E. coli o-rRNAs in many cases (Fig. 5b–d and Supplementary Fig. 4f–h).

Representative variants from each rRNA origin and evolution segments were selected for further evaluation based on kinetic luminescence output. Using an orthogonal superfolder GFP (sfGFP) reporter, we observed the highest o-ribosome activity (sfGFP yield) from V. cholerae rRNA mutants (Fig. 5e). Further, we hypothesized that E. coli r-proteins may show limited ability to catalyze heterologous ribosome assembly with rRNAs sufficiently divergent to that of E. coli, limiting overall functionality of the P. aeruginosa– and V. cholerae-derived o-rRNAs. To explore this, we complemented o-rRNA variants with cognate r-proteins which we have previously shown can improve heterologous activity²⁰. r-Protein complementation of P. aeruginosa (using bS16, bS20) and V. cholerae (using bS1, uS15, bS16, bS20) o-rRNAs showed greatly increased sfGFP production as compared to the starting E. coli o-rRNA, corresponding to 122–147% and 146–629%, respectively (Fig. 5f). These findings show that oRibo-PACE-derived o-rRNAs evolved to overcome the designed selection pressure and did not appreciably adapt to the E. coli host.

Orthogonal translation systems have been employed to improve genetic code expansion efforts¹, yet no reports have extended these capabilities to heterologous ribosomes. We, therefore, evaluated select evolved o-rRNAs for ncAA incorporation by integrating an amber (UAG) stop codon in sfGFP (residue Y151³⁵) and assessed Nε-((tertbutoxy)carbonyl)-l-lysine (BocK) incorporation using an established Methansarcina barkeri-derived tRNA-synthetase pair³⁴. E. coli-derived o-rRNA mutants showed no significant increase in BocK incorporation over starting E. coli o-rRNA (Fig. 5g). In the absence of cognate phylogenetically divergent r-proteins, P. aeruginosa and V. cholerae o-rRNA also resulted in negligible improvements in ncAA incorporation over starting E. coli (Fig. 5g). However, upon supplementation with cognate r-proteins, P. aeruginosa and V. cholerae-derived evolved o-rRNAs improved ncAA incorporation efficiency up to 195% and 908%, respectively (Fig. 5h). Context dependence of translation initiation was also evaluated by expressing sfGFP containing either B or H3 o-RBS, where we observed a nearly uniform correlation and clustering by species (Fig. 5i). Interestingly, only E. coli-derived o-rRNA variants showed improvements in both B and H3 o-RBS contexts, suggesting that they may have been biased by their initial discovery using E. coli o-rRNAs (Fig. 5i).

Finally, we explored the functional relevance of mutations observed with high frequency during the various oRibo-PACE campaigns. Through singular and combinatorial mutations using two unrelated heterologous o-rRNAs (Salmonella enterica and Serratia marcescens o-rRNAs), we uncovered the combined consensus mutations U409C + G1487A as improving the kinetic capabilities of orthogonal ribosomes (Supplementary Fig. 4i, j). Interestingly, this mutational combination was only observed in the V. cholerae campaign, which typically showed greater activities than the E. coli and P. aeruginosa counterparts across all assays. Both consensus mutations were transplanted into o-rRNAs from increasingly divergent microbes, which resulted in general improvements to translation activities (Fig. 5j). This effect was amplified when tested alongside the cognate r-proteins (Fig. 5j). Excitingly, o-rRNAs from Alteromonas macleodii and Marinospirillum minutulum increased activity up to 332% and 299 as compared to the starting E. coli o-rRNA scaffold, respectively (Fig. 5j). Interestingly, a comparison of wild-type and orthogonal ribosome activities showed that starting E. coli o-ribosomes (B o-antiRBS) alongside the cognate reporter gene (B o-RBS) affords a similar protein yield to wild-type E. coli ribosomes alongside a wild-type reporter gene in an E. coli host (Supplementary Fig. 4k). Accordingly, we normalized all reporter assays to the starting E. coli o-rRNA of a given o-RBS context, as enhancements seen within these assays are expected to show improvements over wild-type E. coli rRNA. Cumulatively, these extensive analyses demonstrated that oRibo-PACE-derived o-rRNAs enabled the discovery of context-independent mutations that broadly improved o-ribosome activities.

Kinetically enhanced rRNAs do not enhance population growth

Analyses of evolved o-rRNA activities suggest that oRibo-PACE can robustly influence ribosome translational kinetics in engineered settings. To elucidate the physiological cost of kinetically enhanced rRNA variants, we introduced the wild-type antiRBS sequence into evolved o-rRNAs and assayed their ability to complement the rRNA efficiency of SQ171 E. coli cells and translate all cellular proteins (Fig. 6a and Supplementary Fig. 5a)²⁵.

In all cases, evolved 16S rRNAs robustly complemented the ribosomal deficiency of this strain (used alongside native E. coli 23S, 5S; Fig. 6b–d and Supplementary Fig. 5b–f). We noted, however, that all evolved variants from oRibo-PACE S3 and S4 that exhibited improved luminescence output (>145% of respective wild-type) showed a concomitant proliferation rate reduction in SQ171 cells (Fig. 6c, d): E. coli (6–12% reduction), P. aeruginosa (11% reduction), V. cholerae (7–24% reduction). E. coli ribosome content and therefore translation rate is thought to correlate with cell proliferation³⁶, yet kinetically evolved rRNA variants did not result in faster proliferating strains. To further explore this point, all E. coli and V. cholerae SQ171 strains were assessed for cell vitality in nutrient-rich growth conditions (Davis Rich Medium, DRM)¹⁵. Analysis of cellular respiration through measurement of electron transport chain function (reductase activity) is a reliable marker of vitality³⁷. By assessing the reductase activity and co-staining with propidium iodide, a membrane integrity marker, using all E. coli and V. cholerae mutants, we observed comparable reductase activity between all strains (Fig. 6e) with indications of minor compromises in membrane integrity as compared to wild-type E. coli (Fig. 6f). We did not pursue further analyses using P. aeruginosa derived ribosomes due to poor overall activities across most assays.

We hypothesized that the observed reduction in membrane integrity and cell population growth may derive from protein mistranslation by evolved rRNAs. Whereas perturbation of translation rates through ribosomal protein (rpsD, rpsE) mutations can impact the fidelity of protein synthesis⁵, to our knowledge no such relationship between speed and fidelity has been identified for kinetically enhanced translation. To explore this relationship, we first tested complemented SQ171 strains for sensitivity to aminoglycosides as a marker of amino acid mis-incorporation³⁸. Interestingly, we find that sensitivity to the aminoglycosides kanamycin and gentamicin correlated negatively with E. coli– (Pearson correlation coefficient, or PCC = −0.6086) and V. cholerae-derived variants (PCC = −0.5248) (Fig. 6g, h and Supplementary Fig. 6a, b). SQ171 strains encoding evolved rRNAs also showed an increase in overall cell volume that positively correlated with a population doubling time (Fig. 6i), suggesting that kinetically enhanced ribosomes may impact cell size by accumulating proteins at a non-physiological rate, thereby impacting the balance between cell growth and division³⁹. We note, however, that increased cell volume under nutrient-rich growth conditions is correlated with a higher average cell growth rate⁴⁰, yet this relationship was absent for our kinetically evolved rRNA variants.

Motivated by these observations, we investigated the translational fidelity of evolved rRNAs. Complemented SQ171 strain-derived sfGFP was subjected to trypsinization and label-free LC–MS/MS to quantify amino acid mis-incorporation (substitution) events (Fig. 6j)⁴¹. For strains encoding wild-type E. coli and the starting V. cholerae strain rRNAs, we observed a median amino acid substitution frequency between 1 × 10⁻³ and 10⁻⁴, suggesting these ribosomes translate with natural tolerable error rates (Fig. 6k). All E. coli-derived and 4/6 tested V. cholerae-derived mutants, display median amino acid substitution frequencies above ≥2 × 10⁻³ (Fig. 6k), although we do not observe a strong correlation between o-rRNA activity (Supplementary Fig. 6c). Interestingly, specific regions of the sfGFP transcript were enriched in mistranslation events in an o-rRNA-independent manner (Supplementary Fig. 6d, e), although no clear codon ambiguity or amino acid mistranslation preference emerged from these analyses (Supplementary Fig. 6f).

PACE-evolved o-rRNA variants typically showed an enhanced translation rate over starting E. coli rRNA under various reporter gene and o-RBS contexts (Fig. 5). To investigate whether these observations extended to proteome-wide translation, we quantified the relative translation rates of complemented SQ171 strains through l-azidohomoalanine (AHA) (Fig. 6l) incorporation in defined minimal medium and quantification following click-chemistry labeling⁴². We observed higher average translation rates in V. cholerae mutants (S2.7, S3.7) and E. coli mutants (S3.5, S3.7) as compared to E. coli and V. cholerae starting ribosomes (Fig. 6m). Analysis of viable cells revealed an average AHA incorporation rate increase of >2-fold by Vc mutants S2.7 and S3.7 as compared to wild-type E. coli (Fig. 6m). Under these conditions, we observed comparable degrees of reductase activity between all strains (Fig. 6n) and slightly decreased membrane integrity in the Vc mutant S3.7 and S4.4 strains (Fig. 6o). Overall, these data showcase the ability of the discovered mutations to impart enhanced kinetic properties to ribosomes in native settings, and that faster translation results in a moderate reduction in translational fidelity. These data indicate that ribosome kinetic potential is not maximized but rather refined within a cellular context to balance translation rate and error.