Access to unexplored regions of sequence space in directed enzyme evolution via insertion/deletion mutagenesis

Insertions and deletions (InDels) are frequently observed in natural protein evolution, yet their potential remains untapped in laboratory evolution. Here we introduce a transposon mutagenesis approach (TRIAD) to generate libraries of random variants with short in-frame InDels, and screen TRIAD libraries to evolve a promiscuous arylesterase activity in a phosphotriesterase. The evolution exhibits features that are distinct from previous point mutagenesis campaigns: while the average activity of TRIAD variants is more deleterious, a larger proportion has successfully adapted for the new activity, exhibiting different functional profiles: (i) both strong and weak trade-off in original vs promiscuous activity are observed; (ii) trade-off is more severe (10- to 20-fold increased kcat/KM in arylesterase with ∼100-fold decreases in the original phosphotriesterase activity) and (iii) improvements show up in kcat rather than KM, suggesting novel adaptive solution. These distinct features make TRIAD an alternative to widely used point mutagenesis, providing access to functional innovations and traversing unexplored fitness landscape regions.


INTRODUCTION
Directed evolution aims at identifying proteins with new functional traits by mimicking the natural process of genetic variation through mutations, followed by selection of improved variants. A major challenge for the success of this approach remains that only a very small fraction of the theoretically possible sequence space is accessible experimentally during any screening or selection process, so the type of library determines the success of directed evolution and the features of the functional proteins arising from such protein engineering. Expanding the diversity and the quality of gene libraries has been a major research focus to increase the chances of identifying new variants with desired functions. So far, most directed evolution (and, more generally, protein engineering) experiments have been performed using point substitutions for gene diversification, while insertions or deletions (InDels) remain an overlooked source of variation despite their frequent and functionally beneficial occurrence in natural protein evolution 1 . Combinatorial approaches to incorporate InDels at predefined positions, based on phylogenetic and/or structural analyses, have been developed to alter catalytic specificities of enzymes 2 or to improve the binding affinities of engineered antibodies 3 . While several methods for incorporating InDels randomly within a gene of interest have been developed, they show many limitations in terms of library quality and diversity. Most of these approaches generate frame-shifting InDels at high frequency (>66%) (e.g., using error-prone DNA polymerases 4,5 , terminal deoxynucleotidyl transferase 6 , exonucleases 7,8 , tandem duplication insertion 9 or truncation 10 ) and result in libraries that mostly consist of non-functional variants, which must be removed by high-throughput selection or screening. Methods based on the use of engineered transposons are designed to avoid frameshifts but so far have been limited to the generation of deletions 11,12 or insertions of fixed length and defined sequences 13 .
In the present work, a strategy for random introduction of single short in-frame InDels of one, two or three nucleotide triplets (± 3, 6 or 9 bp) into a given DNA sequence (dubbed TRIAD: Transpositionbased Random Insertion And Deletion mutagenesis) was established and validated by generating libraries of InDel variants of Brevundimonas diminuta phosphotriesterase (wtPTE), a highly efficient enzyme hydrolyzing the pesticide paraoxon 14 with promiscuous esterase and lactonase activities 15 .
The resulting TRIAD libraries were used to investigate the fitness effects of InDels on wtPTE and compare it to that of substitutions. Moreover, screening these libraries for improved arylesterase activity revealed several hits that would have been inaccessible using traditional and widely used point substitution mutagenesis approaches, demonstrating that the introduction of InDels can harvest functional diversity in previously unexplored regions of protein sequence space.  11 . This strategy was extended to the generation of longer contiguous deletions (i.e., -6 and -9 bp) with a second stage, involving the insertion and subsequent MlyI-mediated removal of custommade cassettes (dubbed Del2 and Del3; Figure 1A and Figure 2A). For the generation of insertions, a new transposon, TransIns, was designed as -in contrast to TransDel -an asymmetric transposon ( Figure 1B and Figure 2B), bearing different end sequences (NotI on one end and MlyI on the other).

RESULTS
The latter site marks subsequent insertion sites for the ligation of custom-made shuttle cassettes: Ins1, Ins2 and Ins3 carrying one, two and three randomized nucleotide triplets, respectively. Further digestion using a type IIS restriction enzyme (AcuI) removes the shuttle sequence but leaves triplet insertions behind ( Figure 1B and Figure 2B).

Generation of random InDel libraries by TRIAD
To validate TRIAD, we generated InDel libraries from the gene encoding a highly expressed variant of phosphotriesterase (wtPTE) that had been previously used as starting point to generate an efficient arylesterase by laboratory evolution 17,18 . To enable TRIAD, any recognition sequences for MlyI, NotI and AcuI in the target sequence or the plasmid containing the target sequence must be removed. A corresponds to >8,000 colonies, corresponding to >8-fold coverage of possible insertion sites (~1,000) within wtPTE. The transposon is inserted randomly throughout the entire plasmid, so the fragments corresponding to the target sequence carrying the transposon (~2 kbp) were isolated by restriction digestion and subcloned back into intact pID-Tet, thereby generating TransDel and TransIns transposition libraries. At this stage, transformation of these libraries into E. coli typically yielded >10 6 colonies, maintaining oversampling of transposon insertion sites without skewing the distribution due to sampling.
The TransDel and TransIns insertion libraries were then used as starting material to generate six independent libraries of wtPTE InDel variants: three deletion (-3, -6 and -9 bp) and three insertion libraries (+3, +6 and +9 bp). Without taking into account potential redundancy in the target DNA sequence, the maximal theoretical diversity of TRIAD libraries is a product of the number of positions (~1000 bp for wtPTE) and the diversity introduced at each position: one deletion of each length for deletion libraries and the diversity of randomized triplets (64 1 , 64 2 and 64 3 for one, two or three NNN triplets) for insertion libraries. Therefore, the maximal theoretical diversity for wtPTE is ~1000 variants in each deletion library and 6.4´10 4 , ~4.1´10 6 , and ~2.6´10 8 for +3, +6 and +9 bp insertion libraries (see Supplementary Figure 5). However, depending on the sequence context, two or more neighbouring events may result in identical final DNA sequence, which reduces the accessible theoretical diversity. Theoretical diversities at the protein level are further reduced due to codon degeneracy and occurrence of stop codons as a result of certain InDels (Supplementary Figures S5B   and S5C). Practically, the size of our libraries was limited by transformation efficiency, achieving > 10 6 variants upon transformation into E. coli. Therefore, all deletions as well as +3 bp insertions were oversampled such that the library diversity was maintained between transformations, while the diversity of sampled transposition sites was maintained in larger +6 bp and +9 bp insertion libraries, with only a fraction of theoretical library diversity generated from the outset.

Quality assessment of TRIAD libraries
The quality of the TRIAD libraries was assessed with Sanger sequencing to obtain long read accurate information, as well as deep next-generation sequencing to quantify the library sizes, distribution and diversity of InDels over the target sequence, and the transposition bias. All 121 Sanger-sequenced variants displayed only a single modification from the initial transposon insertion, without any incidental mutations, and 90 among them showed anticipated in-frame InDel mutations (see Supplementary Results 1.3 for details). We then obtained a next-generation sequencing dataset containing ~1×10 6 total 75-bp reads per deletion library and >3×10 6 reads per insertion library  Table S4). In all libraries, the targeted in-frame InDels were found in high abundance, reaching more than 10 5 variants detected by deep sequencing in the most diverse +6 bp and +9 bp libraries (> 10 3 unique deletions and > 10 5 unique insertions overall; Table 1 Figure S7). We extracted the weakly preferred transposition sequence to be 5'N-Py-G/C-Pu-N (see insert in Figure 3A; Supplementary Table S5) and we conclude that the sequence bias of Mu transposons is less pronounced than previously thought 20, 21 .  Figure S1).

Comparison of fitness effects between InDels and point substitutions
To compare the distribution of fitness effects of InDels vs. point substitutions, the levels of native phosphotriesterase (PTE; substrate: paraoxon; Figure 4) and promiscuous arylesterase (AE; substrate: 4-nitrophenyl butyrate, 4-NPB; Figure 4) activities were determined for several hundred wtPTE variants from each TRIAD library and from a trinucleotide substitution library ( Figure 5; Supplementary Tables S8-9). Considering wtPTE is an evolutionarily "optimized" enzyme as a phosphotriesterase (based on the observation that it is operating near the diffusion limit for its native activity 14 ), it is to be expected that very few mutations would be beneficial and that InDels are more deleterious than point substitutions overall. This expectation is underlined by the observation that 83% of deletions and 77% of insertions are strongly deleterious (<0.1 PTE activity), compared to only 24% in the substitution library ( Figure 5A). The average fitness change similarly favours substitutions and is an order of magnitude more deleterious for InDels ( Figure 5C). However, of 485 screened deletions and 351 insertions, a total of 12 were beneficial (>1.5-fold PTE activity increase) against a background of already high catalytic efficiency. By contrast, no beneficial substitutions were found amongst the 342 substitutions screened. Similar frequencies were observed with respect to  Table S8). The frequency of InDels beneficial for arylesterase activity was found to be at least 3-fold higher than that of beneficial substitutions (6% and 7.7% for deletions and insertions, respectively versus 1.8% for substitutions; Figure 5B).
Mapping the observed mutations to the 3D structure of wtPTE provided insight into the location of adaptive InDels in comparison with point substitutions. While substitutions selected for ≥ 50% of wtPTE activity are found throughout the protein, the positions of InDels triggering similar functional effect appear more clustered in loops and on the surface ( Figure 5D). Analysis of surface-accessible solvent area (SASA) suggests that mutations affecting the buried residues are more detrimental than surface-exposed ones (Supplementary Figure S10 and Supplementary Table S10). This observation holds for both InDels and substitutions. For substitutions, the correlation between SASA and fitness effects on activity is weak, while only ~20% of neutral or beneficial InDels affect buried residues (cf. ~40% of substitutions), readily explained by the larger impact of InDels on presumably optimised packing in the protein core.

Screening and identification of adaptive InDels in wtPTE
To demonstrate that TRIAD libraries allow access to functional innovation via adaptive InDels, all the libraries generated from the full-length wtPTE gene (six libraries in total: -3, -6, -9, +3, +6 and +9 bp) were subjected to two parallel screening campaigns to identify variants with enhanced arylesterase activity against either 4-nitrophenyl butyrate (4-NPB) or 2-naphthyl hexanoate (2-NH) ( Figure 4). Both screening campaigns consisted of a general two-step assay workflow. Upon transformation of the TRIAD libraries into E. coli, the resulting colonies (around 1 to 3×10 4 per library) were first screened for either 1-naphthyl butyrate (prior to subsequent screening against 4-NPB in crude cell lysates) or 2-NH hydrolysis (using the FAST Red indicator that reacts with the released naphthol product).
Colonies expressing an active variant (300 to 600 per library) were subsequently grown, lysed and tested for enzymatic activity (for either 4-NPB or 2-NH) in 96-well plates. Note that screening assays on colonies and in cell lysates were both performed after expression of wtPTE variants in the presence of overexpressed GroEL/ES chaperonin as described previously 22 .
Overall, 81 hits (55 insertions and 26 deletions) were identified based on improved arylesterase activity against 2-NH or 4-NPB in cell lysates, with increases ranging from 2-to 140-fold in lysate activity compared to wtPTE ( Table 2; Supplementary Table S11). In contrast with the adaptive substitutions previously identified 18 , these adaptive InDels appeared to have a more drastic effect on the native phosphotriesterase activity, indicating a more severe trade-off on average between maintaining original and enhancing promiscuous activity (average specificity ratio ~ 260; Supplementary Figure S11). However, numerous individual mutants that do not show such strong negative trade-off were also identified (e.g., 64 variants out of 81 showed a specificity ratio < 100; Figure S12).
Sequence analysis of the nature and the location of the InDels responsible for the improvement in arylesterase activity (Table 2) showed that all the adaptive InDels (apart from one double triplet insertion, e.g., V99G/Q99aI99b) were clustered in two flexible regions of wtPTE, namely loop 7 (residues L252 to Q278) and loop 8 (residues S299 to P322)( Figure 6C). Activity against 2-NH was improved by single InDels present in either loop while activity against 4-NPB was enhanced by InDels clustered in loop 7 (Table 2; Supplementary Figure S13). Unexpectedly, the best variant (10.5-fold improvement in AE) found in the -9 bp deletion library exhibited a 12 bp deletion (presumably as a result of a rearrangement during the transposition step in the TRIAD process) resulting in a fouramino acid residue deletion (i.e., ΔA270-G273).
To further demonstrate that the identified InDels genuinely improve the arylesterase activity of wtPTE, the four variants exhibiting the strongest improvement against the 2-NH substrate (i.e., ΔA270-G273, P256R/G256aA256b, S256aG256b and G311a) were purified and characterized to give a 10-to 20-fold increased kcat/KM for 2-NH, while decreasing paraoxon hydrolysis by around 100-fold ( Figure 6B; Table 3). The quantitation of the improvements varies between cell lysates and purified protein (e.g., 140-fold in cell lysates versus 14-fold with the purified protein for P256R/G256aA256b), which may be ascribed to variation in expression levels.

DISCUSSION
Point substitutions, small insertions and deletions account for most evolutionary changes among natural proteins 1 . The ratio of InDels to point substitutions covers a wide variety of ratios across different species, ranging from 1:5 in humans and primates 23 to 1:20 in bacteria 24 , which indicates that InDels are typically subject to stronger purifying selection. Additionally, protein sequence alignments have established that the majority of InDels fixed in protein-coding genes are short (i.e., encompassing 1 to 5 residues) and occur almost exclusively in loops linking secondary structure elements at the solvent-exposed surfaces of proteins 25, 26, 27, 28, 29, 30 . While a large body of experimental evidence reports on the effects of substitutions, the impact of InDels on structural stability and functional divergence in protein evolution is still imperfectly understood, no doubt in part because convenient methods to introduce them in library experiments were missing. Substitutions, being merely side chain alterations, tend to have local effects with typically minor consequences for the overall structure of a protein. By contrast, InDels alter the length of the backbone, opening the way to dramatically larger changes in the packing and orientation of domains that may result in more global effect on the protein structure 31, 32, 33 . Examples of InDels that cause significant repositioning of the backbone and nearby side chains to accommodate the extra or lost residues are on record 34, 35, 36 . If such rearrangements occur near the active site of a protein, the resulting structural changes can change specificity and activity 37,38,39 . Additionally, short InDels occurring at oligomerisation interfaces have also been shown to have important effects on the stability and/or specificity of protein complexes 40,41 . A corollary of the comparatively drastic effect of InDels on protein structure is the perception that they are more deleterious. Indeed, this view is now experimentally corroborated by our work on wtPTE ( Figure 5) as well as a recent deep mutational scanning study investigating the fitness effects of single amino acid InDels on TEM-1 β-lactamase 42 . However, InDels have also been shown to be contribute to functional divergence in several enzyme families, such as lactate and malate dehydrogenases 43 , tRNA nucleotidyltransferases 44 , nitroreductases 45 , o-succinylbenzoate synthases 41 , and phosphotriesterase-like lactonases 2, 37 .
An experimental platform that gives straightforward access to InDel libraries makes it possible to analyse the respective contributions of InDels and point substitutions as sources of functional innovation in experiments against the molecular fossil record. The reliability of gene randomization methods is essential for success in directed evolution experiments. Popular and practically useful methods must meet several key requirements: a high-yielding library generation protocol should create a large number of variants, avoid bias in gene composition or type of variant introduced, and be technically straightforward. When it comes to amino acid substitutions, several approaches (e.g. error-prone PCR, site-saturation mutagenesis starting with synthetic oligonucleotides) have been developed that partially or fully meet these criteria and are widely used. By contrast, the use of InDels in directed evolution experiments has been curtailed by practical limitations in existing methodologies to randomly incorporate insertions and/or deletions (see Supplementary Table S13). Consequently, their application in protein engineering has been sparse, with very few directed evolution campaigns on record that originate from such libraries. For example, the RID protocol 46 , the first attempt towards creating InDel libraries, relies on a complex protocol involving random cleavage of single stranded DNA, so that random substitutions are introduced unintentionally alongside the target mutations. Two other early methods, segmental mutagenesis 8 and RAISE 6 , do not control for the length of the InDel and consequently produce libraries that primarily contain frameshifted variants. In contrast, a codonbased protocol dubbed COBARDE 47 gives a pool of multiple codon-based deletions with <5% frameshifts but requires custom reprogramming of an oligonucleotide synthesizer to create mutagenic oligonucleotides. Alternatively, the viability of transposon-based protocols has been established for generating deletions of various sizes, up to gene truncation variants 10,11,12 . However, the only reported such protocol to create insertions, namely pentapeptide scanning mutagenesis 13,48 , merely gains access to insertions of defined size and sequence.
Improving on existing methodology, the TRIAD protocol meets all major requirements outlined above and gives easy access to large, diverse InDel libraries. The random insertion of a transposon gives excellent sampling of the entire target sequence (Figure 3; Supplementary Figure S8).
Extensive sequencing shows that Mu transposon is even less biased than previously thought, so that functional effects upon insertion/deletion in any region of the protein can be taken advantage of.
Library sizes upwards of 10 5 variants were accessible by covering most of the theoretical diversity of up to two randomised amino acid insertions (Supplementary Figure S5). Introduction of randomised larger insertions exceeds the typically screenable library size, but can be constructed 49 . Finally, the procedure is technically straightforward, consisting of transposition and cloning steps, and does not require access to specialized DNA synthesis equipment (as in 47 ). The TRIAD workflow is a versatile process that can be adapted to create libraries focused on a specific region of a protein, applicable in cases where screening throughput is limited. This approach would be analogous to other procedures (although only a few 47,50 have directly exemplified this case). In the case of TRIAD, this was typically achieved by adding an in-frame seamless cloning step using a type IIS restriction enzyme such as  Table   S12). InDel libraries constructed in this way showed good coverage of the target region, albeit with slightly more pronounced bias than whole-gene TRIAD, presumably due to increased sensitivity to preferential transposon insertions on a short target sequence. Alternatively, TRIAD can be further expanded with a recombination protocol (e.g., DNA shuffling or Staggered Extension Process) to generate variants combining multiple InDels, which can be screened in a high-throughput assay 51 .
The potential of InDel mutagenesis strategies in directed protein evolution is underlined by our comparative analysis of the fitness effect of InDels and point substitutions that showed InDels to be more likely to yield wtPTE variants with improved arylesterase activity than substitutions ( Figure 5B).
A second point of comparison are the evolutionary trajectories followed starting with InDel vs point substitution libraries. The promiscuous esterase activity of wtPTE has previously been used as the starting point of a directed evolution effort that generated an arylesterase which hydrolysed 2-NH with high efficiency 18 . Here the mutation H254R, selected after the first round of mutagenesis, appeared to be a mutation on which the rest of the trajectory was highly contingent. InDel mutagenesis and selection puts us in a position to address the question whether alternative initial mutations would enable access to different evolutionary trajectories leading towards the same functional outcome.
Based on the hypothesis that the use of a wider genetic and functional diversification (i.e., by both substitutions and InDels) might lead to a wider diversity of possible evolutionary trajectories, the first objective was to identify new adaptive mutations improving the promiscuous arylesterase activity of wtPTE by screening InDel libraries of wtPTE generated via TRIAD. This resulted in the identification of multiple beneficial deletions and insertions, confirming that introduction of InDels can give rise to functional and improved catalysts.
We further observed that four of these adaptive InDels increase arylesterase activity 10-to 20-fold (in kcat/KM) against 2-NH, which is more than the 2.6-fold difference brought about by the initial H254R mutation from the previous directed evolution 18 . For all four InDel variants, the improvement in 2-NH catalytic efficiency appears to be due to increased kcat (from 13-to 67-fold), which outweighs an  53 . A similar campaign that selected for eGFP variants with increased brightness in a colony screen identified the surprising eGFP-ΔGly4 deletion, which has significantly more cellular fluorescence likely due to increased refolding efficiency 54 . Finally, a recent focused library approach in a PTE-like lactonase with insertions into loop 7 (that is shorter in lactonases) led to variants enhanced in phosphotriesterase activity, with increased kcat and decreased KM for paraoxon (kcat/KM increased up to 600-fold) 37 . Native lactonase activity was strongly affected in those variants with up to 10 4 -fold decreases in catalytic efficiency. These results in an enzyme closely related to PTE are very similar to our observations of the mixed effect of InDels on wtPTE, as explored based on the larger diversity of adaptive variants rendered available by TRIAD.
We conclude that unprecedented evolutionary trajectories become accessible by screening InDel libraries obtained via TRIAD, establishing a new paradigm that complements current strategies following the 'one amino acid at the time' adage 55 which are believed to lead to successful outcomes slowly, yet steadily. The effect of InDels is on average more deleterious than substitutions ( Figure 5A), while the fraction of hits is increased in InDel libraries ( Figure 5B), suggesting that InDel library strategies tend to 'polarize' properties of library members towards extremes. For thermodynamically more difficult reactions than those studied here, this trend to more extreme outcomes may practically imply low hit rates, in which case high throughput screening would become crucial. For example, ultrahigh-throughput screening based on droplet microfluidics 56, 57 could be combined with InDel mutagenesis to powerfully explore sequence space for evolutionary trajectories and individual variants that would not arise from epPCR mutagenesis libraries. It remains to be seen whether this new way of 'jumping' (rather than 'tiptoeing') across sequence space yields functionally better catalysts -or just different ones.

Sequencing and quality analysis
The mutagenesis efficiency of TRIAD was analysed both by Sanger sequencing (Supplementary   Tables S1-3

DATA AVAILABILITY
Illumina raw sequencing reads were deposited with European Nucleotide Archive (https://www.ebi.ac.uk/ena) at accession number PRJEB28011. The source code along with instructions for all scripts involved in data processing are freely available at https://github.com/fhlab/TRIAD.

SUPPLEMENTARY DATA
Supplementary results describing focused InDel libraries generated by TRIAD, supplementary methods, supplementary figures S1-S13 and supplementary tables S1-14 are available online. Table 1

. Mutagenesis efficiency of TRIAD analysed by deep sequencing
Unique in-frame InDels (i.e., InDels of multiple of three nucleotides) were counted both at the DNA and the protein level. Adjacent amino acid substitutions and truncations (resulting from the occurrence of stop codons) may occur depending on the insertion point of the transposon, resulting in a lower value for the number of observed unique protein InDels. The proportion relative to the theoretical diversity accessible from the wtPTE sequence (both at the DNA and the protein level) was calculated as the ratio between the number of unique in-frame InDels observed by deep sequencing and the theoretical diversity for a given TRIAD library (see Supplementary Figure S5).

Table 2. Analysis of InDel wtPTE variants with at least 2-fold improved arylesterase activity.
Values refer to the activity change of all or AE positive variants relative to wtPTE obtained by comparing the initial rates v0 for the hydrolysis of paraoxon, 4-NPB or 2-NH to that of wtPTE at 200 μM substrate concentration, resulting in a dimensionless ratio (recorded in Table 2). The average effect value was determined as the geometric mean of the relative activities of all the variants listed in Supplementary Table S11. The maximum, median and minimum changes correspond to the maximum, median and minimum relative activities for each substrate among the variants (See also Supplementary Figure S12). Table 3. Kinetic properties of wtPTE variants.             Table 1.

Mutagenesis efficiency of TRIAD analysed by deep sequencing
Unique in-frame InDels (i.e., InDels of multiple of three nucleotides) were counted both at the DNA and the protein level. Adjacent amino acid substitutions and truncations (resulting from the occurrence of stop codons) may occur depending on the insertion point of the transposon, resulting in a lower value for the number of observed unique protein InDels. The proportion relative to the theoretical diversity accessible from the wtPTE sequence (both at the DNA and the protein level) was calculated as the ratio between the number of unique in-frame InDels observed by deep sequencing and the theoretical diversity for a given TRIAD library (see Supplementary Figure S5).
[a] The theoretical protein diversity of +9 bp library is estimated as 21× larger (20 amino acids and a stop codon) than the calculated diversity of +6 bp library.   Table 2). The average effect value was determined as the geometric mean of the relative activities of all the variants listed in Supplementary Table S11. The maximum, median and minimum changes correspond to the maximum, median and minimum relative activities for each substrate among the variants (See also Supplementary Figure S12). The values refer to the number of insertions and/or deletions observed in the entire sequence of wtPTE or in specific regions (e.g., loop 7 (residues 252-278) and loop 8 (residues 299-313)).

Table 3. Kinetic properties of PTE variants.
[a] The symbol ∆ before a residue (or a group of residues) signifies that this (or these) residue(s) have been deleted. Inserted residues are labelled using the number of the position after which they are inserted and alphabetical order (e.g., glutamine and tyrosine residues inserted in this order after the residues at position 230 would be labelled Q230aY230b