An improved Escherichia coli screen for Rubisco identifies a protein–protein interface that can enhance CO2-fixation kinetics

An overarching goal of photosynthesis research is to identify how components of the process can be improved to benefit crop productivity, global food security, and renewable energy storage. Improving carbon fixation has mostly focused on enhancing the CO2 fixing enzyme ribulose-1,5-bisphosphate carboxylase/oxygenase (Rubisco). This grand challenge has mostly proved ineffective because of catalytic mechanism constraints and required chaperone complementarity that hinder Rubisco biogenesis in alternative hosts. Here we refashion Escherichia coli metabolism by expressing a phosphoribulokinase-neomycin phosphotransferase fusion protein to produce a high-fidelity, high-throughput Rubisco-directed evolution (RDE2) screen that negates false-positive selection. Successive evolution rounds using the plant-like Te-Rubisco from the cyanobacterium Thermosynechococcus elongatus BP1 identified two large subunit and six small subunit mutations that improved carboxylation rate, efficiency, and specificity. Structural analysis revealed the amino acids clustered in an unexplored subunit interface of the holoenzyme. To study its effect on plant growth, the Te-Rubisco was transformed into tobacco by chloroplast transformation. As previously seen for Synechocccus PCC6301 Rubisco, the specialized folding and assembly requirements of Te-Rubisco hinder its heterologous expression in leaf chloroplasts. Our findings suggest that the ongoing efforts to improve crop photosynthesis by integrating components of a cyanobacteria CO2-concentrating mechanism will necessitate co-introduction of the ancillary molecular components required for Rubisco biogenesis.

Improving carbon fixation in agriculture has mostly focused on enhancing the activity of the CO 2 -fixing enzyme ribulose-1,5-bisphosphate carboxylase/oxygenase (Rubisco) 3 by modify-ing the enzyme itself or increasing the CO 2 levels around it (1). Improving the kinetics of Rubisco has proved challenging because of its complex catalytic mechanism of fixing CO 2 to ribulose-1,5-bisphosphate (RuBP) and cleaving it into two 3-phosphoglycerate (3-PGA) molecules (2)(3)(4). The carboxylation rate is slow (k cat c ϭ ϳ1 to 5 reactions/s in plants) and often confines flux through the Calvin-Benson-Bassham (CBB) cycle, hence limiting the rate of photosynthesis and plant growth. Because of this limitation, large amounts of Rubisco are needed to support adequate CO 2 -assimilation rates (5)(6)(7). This results in Rubisco being the Earth's most abundant protein (8). Further encumbering its performance, Rubisco also catalyzes RuBP oxygenation to produce 3-PGA and 2-phosphoglycolate. The recycling of 2-phosphoglycolate back into 3-PGA via photorespiration is considered wasteful because it consumes energy and releases fixed CO 2 (9).
The potential for successfully improving the carboxylation properties of Rubisco is spurred on by findings that crop Rubisco is not the pinnacle of evolution, with kinetically more efficient Rubisco forms found in some red algae (1,3,6,10). Evolutionary adaptation of Rubisco kinetics appears to have been constrained by the complexity of its catalytic chemistry and multiplexed subunit folding and assembly requirements (3,(11)(12)(13). The need to retain complementarity with ancillary proteins involved in the biogenesis and metabolic repair of Rubisco appears particularly pertinent for form I Rubisco, which comprises eight RbcL and eight RbcS subunits to form a L 8 S 8 complex. Examples of Rubisco-dedicated components include the assembly chaperones BSD2, RbcX, Rubisco accumulation factors 1 and 2 (Raf1, Raf2), and the metabolic repair protein Rubisco activase (14). The protein-folding chaperonins of bacteria (GroEL), chloroplasts (Cpn60), and their protein co-factors GroES (bacteria) and Cpn10/Cpn20 (chloroplasts) are also essential components in Rubisco biogenesis (15). The need for Rubisco to maintain complementarity with these varied chaperone and chaperonin components appears to limit the span of Rubisco isoforms that can be produced in Escherichia coli, as well as those that can be bioengineered in the chloroplasts of vascular plants. For example, the chloroplasts of the model C 3 plant tobacco can produce high levels of the RbcS-lacking bacterial form II Rhodospirillum rubrum L 2 Rubisco and Methanococcoides bur-tonii archaeal L 10 Rubisco (ϳ10 -25 mol active sites⅐m 2 ) because of their simple biogenesis requirements (16,17). Differences in the biogenesis requirements of L 8 S 8 Rubisco from red algae and monocot plants, however, preclude their assembly in tobacco and E. coli (10,18). By contrast the assembly requirements of Synechococcus elongatus PCC6301 L 8 S 8 Rubisco (Se-Rubisco) are partially met in chloroplasts (ϳ5 mol active sites sites⅐m 2 ) (19) and in E. coli, with ϳ98% of the Se-RbcL produced forming insoluble, misfolded aggregates in the bacterium (20).
To avoid the constraints a photosynthetic environment may pose on the catalytic evolution of Rubisco, modern laboratory evolution applications have made particular use of Rubisco-dependent E. coli (RDE) screens (12,21,22). An elegant advance has been the extensive rewiring of E. coli metabolism to incorporate a non-native CBB cycle where cell survival can be made dependent on CO 2 -fixation by Rubisco (23). This contrasts with conventional RDE screens that simply express recombinant phosphoribulokinase (PRK) to produce RuBP, the 5-carbon substrate of Rubisco. As summarized in Fig. 1A, RuBP is unexplainably toxic to E. coli such that the rate of colony growth in RDE screens is dependent on the level of Rubisco activity. This type of selection has proved useful for evolving M. burtonii L 10 Rubisco mutants with desired improvements in carboxylation rate (k cat c ), CO 2 affinity (lower K m for CO 2 , K c ) and specificity for CO 2 over O 2 (S c/o ) (17). Other attempts to improve the catalysis of L 8 S 8 Rubisco using RDE screens have only identified RbcL mutations that enhance the folding and assembly of the subunit with RbcS into a functional L 8 S 8 complex (i.e. improving Rubisco solubility) (22).
A feature of RDE screens that limit their throughput and reliability (i.e. selection fidelity) is the high proportion of false positives selected (24, 25). These arise from the inactivation of PRK function by transposon integration (Fig. 1A and Ref. 22). Here we present the development and implementation of a simple, faster, high-throughput screen called RDE2 that specifically selects for improvements in Rubisco activity, not PRK inactivation mutants. By eliminating false-positive selection, we demonstrate the versatility of the RDE2 screen in selecting for mutants of Thermosynechococcus elongatus BP-1 (Te-) Rubisco that improve RbcL folding and assembly in E. coli (i.e. Te-Rubisco solubility), the enzyme carboxylation kinetics, or both. We subsequently reveal, using chloroplast genome transformation, how the specialized folding and assembly requirements of Te-Rubisco hinder its translational testing in tobacco (Nicotiana tabacum). The outcomes demonstrate that Rubisco biogenesis in E. coli is not a reliable proxy for expression in chloroplasts. As a consequence, deciphering the crucial molecular partnerships required for Rubisco biogenesis is needed to optimize heterologous Rubisco expression in crop chloroplasts, as well as adapting the RDE2 for future evolution of eukaryotic form I Rubisco.  Figure 1. An improved RDE screen. A, RDE screens rely on the toxicity of ribulose-1,5-P2 (RuBP), a foreign metabolite produced from the pentose phosphate pathway (PPP) intermediate ribulose-5-phosphate (ribulose-5-P) via recombinant PRK expression (regulated by arabinose addition). Cell viability is restored either through Rubisco expression (IPTG-induced regulation), where growth rate is proportional to Rubisco activity level, or through transposon silencing of PRK expression that produces false positives at a high frequency (22). The new RDE2 screen utilizes a PRK-NPTII fusion where NPTII is tethered in frame to the C terminus of PRK rendering kanamycin resistance to E. coli unless PRK-NPTII expression is transposon silenced. Toxic pathways to E. coli are highlighted in red. RDE2 screens are performed under high CO 2 (air ϩ 1 to 2% (v/v) CO 2 ) with the ribulose-1,5-P 2 carboxylation reaction of Rubisco (positive sign) producing 3-PGA, an intermediate of the glycolysis/gluconeogenesis pathways (shaded yellow) (reviewed in Refs. 12 and 21). B, summary of the plasmids and E. coli growth conditions used in the RDE2 screen of an rbcL (ϮrbcS) mutant library. P trc and P BAD indicate IPTG-and arabinose-inducible promoters, respectively.

Novel solutions for improving Rubisco catalysis
selected for one or more amino acid substitutions in RbcL that improve Rubisco biogenesis (solubility), not catalysis (12,22). Impeding success are the poor transformation efficiencies of photosynthetic hosts and the high proportion of false positives selected in RDE screens. As summarized in Fig. 1A, false-positive selection in RDE systems arises from terminating RuBP production via transposon silencing of PRK expression (22). To circumvent this impediment, a synthetic gene coding a fusion protein comprising S. elongatus PCC6301 PRK (Se-PRK) and a C-terminal neomycin phosphotransferase (NPTII) was made (Fig. 1B). The PRK-NPTII fusion protein was found to preserve the kanamycin resistance conferring properties of NPTII to E. coli in addition to retaining 70% the PRK activity (k cat ϭ 95 Ϯ 3 s Ϫ1 ) of unmodified Se-PRK (134 Ϯ 6 s Ϫ1 ).
The fidelity of the RDE and RDE2 screens were compared by plating XL-1Blue cells producing Se-PRK or the PRK-NPTII fusion, along with wild-type Te-Rubisco, on medium containing 0.2% (w/v) arabinose (high PRK inducing) and no kanamycin. Under these conditions, Te-Rubisco cannot support cell growth, meaning any colonies formed are false positives caused by transposon silencing of PRK activity ( Fig. 1A and Ref. 22). More than 500 false-positive colonies/10 6 cells plated grew in the RDE screen, whereas only 28 Ϯ 17 colonies/10 6 cells plated grew using the RDE2 screen. Inclusion of 100 g/ml kanamycin in the growth medium of replica RDE2 cell platings resulted in no colony growth. This confirmed that the RDE2 screen is immune from generating false positives because transposon silencing of PRK-NPTII activity also confers the cells sensitive to kanamycin (Fig. 1A).

Identifying a suitable Rubisco to evolve
The inability of E. coli to meet the folding and assembly requirements of plant and algae Rubisco currently prevents their directed evolution using RDE screens (10,12). By contrast the biogenesis requirements of cyanobacteria L 8 S 8 Rubisco can be partially met by E. coli. For example, only a small proportion (Ͻ2%) of the 52-kDa Se-RbcL produced in E. coli can assemble with the more soluble 14-kDa Se-RbcS into functional Se-Rubisco (20). As a consequence Se-Rubisco produced in E. coli only accounts for 1-2% (w/w) of the cell-soluble protein (CSP). RDE screens have therefore primarily only identified substitutions in RbcL that enhance the biogenesis of Se-Rubisco (i.e. improved its solubility) (22). Among these are the F140I, V189A, and F345I substitutions in RbcL (numbering relative to spinach and Te-Rubisco) that increase Se-Rubisco solubility between 3-and 14-fold in E. coli (Fig. 2, A and B, and Table 1). Notably these mutations all impair k cat c , contrary to that proposed previously (11). By comparison the biogenesis requirements of Te-Rubisco, that already codes Ile-140, are more readily met in E. coli and expressed at ϳ6% (w/w) CSP ( Fig. 2A). Moreover, the apparent carboxylation efficiency under ambient O 2 (k cat c /K c 21%O 2 ) and specificity for CO 2 over O 2 (S c/o ) of Te-Rubisco are ϳ40 and 18% higher, respectively, than Se-Rubisco (Table 1). Notably these favorable Te-Rubisco kinetics come at the expense of a slower carboxylation rate (k cat c ) that is ϳ40% less than Se-Rubisco (Fig. 2, A and B). The high solubility of Te-Rubisco in E. coli and distinctive kinetics made it a superior target for evolution testing using RDE2.

First-generation Te-Rubisco solubility mutant selection in RDE2
The efficacy of RDE2 to select for increased Te-Rubisco activity was initially tested using a library of 2.7 ϫ 10 5 Te-rbcL mutants with an average mutation rate of 2.5 nucleotides (1.6 amino acid substitutions) per variant. After 5 days at 25°C, 18 faster-growing colonies were identified on plates containing 0.1% (w/v) arabinose (defined as moderate PRK-NPTII induction) and found to code either V300A, F345I, F345L, or P415A substitutions in RbcL. The kinetics of the slower growing V300A mutant resembled Te-Rubisco, suggesting that it improved RDE2 fitness possibly from the small (Ͻ10%) increase in the amount of soluble L 8 S 8 made ( Table 1). The faster-growing F345I, F345L, and P415A mutants produced significantly more Te-Rubisco, especially the F345I mutation that stimulated Te-Rubisco biogenesis to ϳ14% (w/w) of the CSP (Fig. 2, A and B). When coupled with the P415A mutation, the solubility of the Te-F345I/P415A mutant increased to ϳ17% (w/w) CSP (Fig. 2, A and B). This improvement came at a cost to the kinetics of each mutant that showed reductions in k cat c , k cat c /K c 21%O 2 , and S c/o (Table 1). Sequence comparisons found that both Phe 345 and Pro 415 are highly conserved among cyanobacteria and higher plant form IB RbcL but are located in separate, unconnected ␣-helical regions (Fig. 2C). Interestingly, these loci are modified in the form IA Rubisco lineage that has evolved independent of the assembly chaperones Raf1 and RbcX (11,14).

Second-generation Te-Rubisco catalytic mutant selection
Improving the kinetics of Te-Rubisco required a second round of RDE2 screening that targeted mutagenesis of both RbcL and RbcS. A 3 ϫ 10 5 member mutagenic library of the full-length Te-rbcLS operon coding the first-generation P415A RbcL mutant (abbreviated Te-P415A) was used. Unlike the F345I and F345I/P415A mutants, the catalytic properties of Te-P415A more closely matched wild-type Te-Rubisco ( Fig.  2A). Under strong PRK-NPTII selection (i.e. on medium containing 0.2% (w/v) arabinose), the growth of cells producing wild-type (abbreviated Te-LS) or Te-P415A Rubisco were impeded (Fig. 3A). After 6 days growth at 25°C in air supplemented with 2% (v/v) CO 2 , 15 colonies were isolated. The sequence and biochemistry of these second-generation mutants (abbreviated Te-2G Rubisco variants) were found to enhance RDE2 growth either through further improving Te-P415A Rubisco solubility and/or significantly enhancing its carboxylation properties ( Fig. 3A and Table 1). The fastest growing Te-2Ga mutant was independently selected four times and coded a V98M RbcS point mutation that improved Rubisco biogenesis, k cat c , k cat c /K c 21%O 2 , and S c/o by ϳ310, 28, 43, and 6% relative to Te-Rubisco (Fig. 3, B-E). Four additional Te-P415A Rubisco mutants with improvements in k cat c and k cat c /K c 21%O 2 and unchanged solubility were also identified. They were found to code for point mutations in either RbcS (A48V, Te-2Gb; H37L, Te-2Gc; or Y36N/G112D, Te-2Gd) or RbcL (L74M/ D397N, Te-2Gg). The Te-2Gg mutant also showed a small but significant increase of ϳ4% in its S c/o relative to native Te-Rubisco (Table 1 and Fig. 3E).

Clustering of the catalytic mutations at a RbcL-RbcS interface in Te-Rubisco
Structural analysis revealed that the H37L, Y36N, V98M, L74M, and D397N substitutions all occur at residues that cluster on a surface exposed region of an RbcS-RbcL interface (Fig.  4, A and B). The A48V mutation occurs on the same interface but is oriented toward the inner enzyme surface (Fig. 4, A and C). The common locality of these mutations suggest this region to be a hot spot for Te-Rubisco catalytic modification, despite being located quite distant from the catalytic pockets in adjoining regions of paired RbcL subunits (Fig. 4A). In support of this assertion, the RbcS residue Tyr 36 is located directly adjacent to Val 98 in the Te-Rubisco holoenzyme (Fig. 4B) and can convey comparable improvements in k cat c and k cat c /K c 21%O 2 when mutated to Asn (Y36N mutant Te-2Gd; Fig. 3, C and D). Curiously the selected RbcS mutants comprise residue changes adjacent to highly conserved regions of cyanobacterial and higher plant form IB Rubisco (Fig. S2). This suggests that this RbcL-RbcS interface may pose a hot spot for manipulating the kinetics in related plant Rubisco lineages.

Second-generation Te-Rubisco solubility mutant selection
Of the 10 different Te-P415A Rubisco-derived mutants selected, five arose from improvements in the biogenesis of soluble Te-Rubisco, not catalysis (Fig. 3A). The improved solubility derived either from a R51H substitution in RbcS (or A48V, see above) or from I393M, A398T, or A414T mutations in RbcL (    (11,20) relative to T. elongatus BP1 (Te)-Rubisco (in blue) and the RbcL solubility F345I and P415A mutants identified using the RDE2 screen. The data are the means Ϯ S.E. of more than three biological samples with significance relative to each WT Rubisco analyzed by one-way ANOVA with Turkey's post hoc analysis. **, p Ͻ 0.01. See Table 1

Novel solutions for improving Rubisco catalysis
selected in the first-generation RDE2 screen, suggesting that they complement the P415A mutation. Structural analyses revealed, however, that the residues associated with increasing Te-Rubisco solubility show no apparent relation to one another and are located quite distant from Pro 415 within the quaternary Rubisco structure (Fig. S3). Consistent with that postulated previously (11), this suggests these residues pose important determinants of RbcL subunit folding by GroEL or/and provide structural stability to avoid their misfolding and aggregation in the absence of ancillary molecular assembly components.

Comparative analysis of Te-Rubisco biogenesis in chloroplasts
The varied biogenesis requirements within the Rubisco superfamily limit not only which isoforms that can be produced in E. coli but also those that can be assembled within leaf chloroplasts (10,14,18). The availability of efficient plastome transformation capabilities in tobacco has made it the model plant for Rubisco bioengineering (13,16). Consistent with leaf photosynthesis modeling (Fig. 5A), the growth of tobacco producing Se-Rubisco necessitate high levels of CO 2 for growth in soil to compensate for the enzyme's poor kinetics and limited biogenesis potential in chloroplasts (19). Comparable modeling using Te-Rubisco kinetics show it would support higher rates of CO 2 assimilation in a C 3 plant over all intracellular CO 2 concentrations, more so in plants producing the Te-2Ga Rubisco mutant selected in this study using the RDE2 screen (Fig. 5A).
To test the correlative potential between Te-Rubisco biogenesis in E. coli and leaf chloroplasts, a synthetic Te-rbcLS operon coding for either Te-Rubisco or the high-solubility F345I/ P415A mutant was transformed in the tobacco chloroplast genome (plastome) in place of the native tobacco rbcL gene ( Fig. 5B and Fig. S4). To optimize translation, the codon use of the transformed genes matched the native tobacco rbcL, and 12 amino acids of the native tobacco RbcL N terminus were retained (Fig. S4a). Independently transformed lines for each tobacco genotype (called tob TeLS and tob TeLS-FIPA ; Fig. 5B) were obtained and grown in tissue culture until homoplasmic (Fig.  5C). Plastome sequencing confirmed correct integration of the transgenes, and RNA blots confirmed high Te-rbcL-rbcS and Te-rbcL-rbcS-aadA mRNA levels (Fig. S4b). Nevertheless the tob TeLS and tob TeLS-FIPA plants shared a pale green phenotype and could only survive in tissue culture (Fig. 5C). CABP binding (Fig. 5D) and native PAGE analysis (Fig. 5E) confirmed the assembly of functional L 8 S 8 Te-Rubisco in tob TeLS and slightly more of the F345I/P415A mutated Te-Rubisco in tob TeLS-FIPA . However, the amount of Te-Rubisco produced were more than 1000-fold lower than Rubisco production in the wild-type tobacco controls (Fig. 5D). Reciprocal mutagenic tests showed introducing the F345I, P415A, or dual F345I/P415A mutations into the N. tabacum RbcL did not permit the assembly of tobacco Rubisco in E. coli (Fig. S5).

Discussion
Following successive evolution rounds, this study reveals a novel region in the subunit interface of the plant-like Rubisco from the T. elongatus BP1 that offers multiple solutions for improving carboxylation rate, efficiency, and specificity. Key to this success is the unwavering fidelity, faster screening, and higher throughput of the new RDE2 screening platform. To date, the RDE screen designed around a ⌬gapA E. coli mutant (a

S 8 Rubisco content and catalytic parameters at 25°C
The values are the means Ϯ S.E. of typically N Ն 3 biological samples assayed in duplicate or triplicate (details shown in top row). One-way ANOVA was undertaken with reference to either WT Synechococcus PCC 6301 Rubisco (Se-LS) or WT Te-Rubisco (Te-LS). Symbols show the statistical significance levels relative to WT enzymes: *, p Ͻ 0.05; **, p Ͻ 0.01. 1G, first generation RDE2 Te-LS Rubisco mutants selected on media containing 0.1% (w/v) arabinose; 2G, second generation mutants derived from Te-P415A under higher PRK selection on 0.25% (w/v) arabinose media (Fig. 3); NA, not applicable; NM, not measured. and 36 Ϯ 2 M (this study), respectively, ϳ10-fold less than the 0.4 mM RuBP concentration used in the 14 CO 2 -fixation assays. Rubisco expression levels were confirmed by native PAGE (Fig. S1).

Novel solutions for improving Rubisco catalysis
strain called MM1) (26) has proven the most effective in Rubisco-directed applications because of the low number of false positives selected (12). Offsetting this improved fidelity is the slow growth rate of the MM1 strain and its need for specialized growth medium that prolongs the screening time frame to 9 -12 days at the optimal growth temperature of 25°C (11,17,20,26). Comparatively, the RDE2 screen is immune to falsepositive production and uses LB growth medium, allowing the screens to be completed in 6 -7 days at 25°C. The feasibility of adapting the new CO 2 -dependent sugar synthesizing CBB-E. coli strains (23) as an alternative RDE screen for selecting Rubisco catalytic mutants remains to be demonstrated (21). There has been long-standing interest in modifying Rubisco catalysis via mutagenesis of RbcL (3, 5-7, 10). More recent RbcS mutagenic studies have demonstrated how it can pervasively influence plant Rubisco kinetics (27), a finding supported by structure-function surveys that indicate a role by the RbcS in the kinetic adaptation of Rubisco to environmental cues (18). X-ray crystallography comparisons show that the quaternary RbcS structure remains highly conserved among the differing L 8 S 8 lineages, especially among the plant and cyanobacteria Form 1B isoforms (Fig. 6A). This suggests that RbcS-mediated changes to catalysis can be strongly influenced by subtle changes in electron density within key areas. One particular area, as demonstrated in this study, is located between the antiparallel ␤-sheets around Val 98 , Tyr 36 , and His 37 in Te-RbcS that share structural similarity to Cys 122 , Trp 38 , and Val 39 in tobacco RbcS (Fig. 6B). Generally, the Met 98 , Asn 36 , or Leu 37 substitutions differentially improved Te-Rubisco catalysis by modification toward more plant-like structures (Fig. 6C). Notably these substitutions did not arise from limitations in codon redundancy at these residues in Te-RbcS. This suggests that further improvements may be possible by additional rounds of laboratory evolution, or possibly even by rational design guided by the vast array of crystal structure information already available for plant and algae Rubisco (4).
An obvious extension of our study is to test how amino acid changes around this region in RbcS influence plant Rubisco catalysis. Such efforts are hindered by the inadequate transgenic methods for replacing or modifying all the multi-RbcS gene copies in plants. The low throughput of modern nucleus gene editing approaches limit their usefulness in mutagenic screening, despite their capacity to introduce multiplexed nucleotide changes with relatively high accuracy (28). Although

Novel solutions for improving Rubisco catalysis
our RDE2 screen poses a potential remedy in terms of screening throughput, our data highlight its versatility for evolving vascular plant Rubisco will require incorporation of requisite Rubisco assembly components from leaf chloroplasts. This observation accords with the underpinning need for ancillary protein complementarity for folding and assembly of form I Rubisco in leaf chloroplasts (13). Understanding these requirements is critical to enabling heterologous Rubisco expression potential in cyanobacteria, E. coli, and chloroplasts. Despite the common prokaryotic connections of these hosts, our work demonstrates that E. coli cannot be used as a proxy for chloroplast Rubisco assembly. This underpins the importance of understanding the specific chaperonin (Cpn60␣, 60␤), co-chaperonin (Cpn10, 20, and 21), and assembly chaperone requirements of plant and algal Rubisco or strategies to circumvent their necessity for successful production in a nonnative host. The last decade has seen significant advances into understanding the mechanisms for many of these components (14), but they have not yet been incorporated into a screen such as RDE2. Conceivably such an approach would provide a highthroughput screen of plant (or algae) Rubisco activity that would be envisaged to function equivalently chloroplasts (29).
Efforts to introduce a cyanobacterial CO 2 -concentration mechanism (CCM) into tobacco plastids are also likely to require the inclusion of Rubisco biogenesis components. In this study we show that Te-Rubisco has the highest CO 2 affinity and CO 2 /O 2 specificity of known cyanobacteria Rubisco. These kinetic features make Te-Rubisco better able to support C 3 photosynthesis, relative to the more commonly studied Se-Rubisco. The reverse would be the case if the extensive requirements for building a functional CCM could be met in plastids to support the higher k cat c of Se-Rubisco (30,31). The limited biogenesis capacity of Te-Rubisco (Fig. 5) and Se-Rubisco (19,30,32) in tobacco suggest that building a CCM in chloroplasts will require not only introducing the multiple carboxysome components, inorganic carbon membrane transporters, and modulating CA levels, but also incorporating the necessary chaperonin/ chaperone components for assembling cyanobacteria Rubisco.

Conclusion
Much of the effort to identify superior Rubiscos to date has focused on measuring the natural variation in Rubisco kinetics. This approach is slowed by the complexity of the catalytic assay methods and the vast spectrum of natural diversity available. Here we show how directed evolution using RDE2 poses a potentially faster alternative. The challenge is to extend past the prior success of RDE screens in investigating prokaryotic and Archaea Rubisco sources and adapt RDE2 for the directed evolution of plant and algae Rubisco. As with Rubisco engineering in plant chloroplasts, our data advocate that this will necessitate the co-expression of the complimentary ancillary proteins needed for the biogenesis and metabolic repair of the target Rubisco.

Expression, purification, and assay of PRK and PRK:NPTII
The prkA gene from Synechococcus PCC6301 was synthesized (Genscript) and cloned downstream of the arabinose inducible BAD promoter in pAC BAD (26) to generate pACYC PRK . Sequence coding a synthetic nptII gene was cloned in frame to the 3Ј of prkA to produce the prk-nptII gene coding a PRK-NPTII fusion protein (GenBank TM accession no. MG000147). The prkA and prk-nptII genes were cloned in frame to a N-terminal His 6 -ubiquitin fusion protein using the expression , and the catalytically improved Te2Ga-Rubisco (Te-2Ga) on CO 2 assimilation rate (A) in tobacco (a model C 3 plant) at 25°C in response to varying chloroplast CO 2 partial pressures (C C ). A was calculated as the minimum of the Rubisco carboxylase-limited (A c ) and lightlimited (A j ) rates modeled as described under "Experimental methods" and assuming a leaf Rubisco content of 30 mol active sites m Ϫ2 (or 5 mol active sites m Ϫ2 as observed in tob-SeLS (19)). B, the tobacco plastome was transformed with plasmids pLEV-TeLS and pLEV-TeLS-FIPA containing homologous plastome flanking sequences (indicated by dashed lines; numbering relative to N. tabacum (WT) plastome; GenBank TM accession no. Z00044) that directed integration of codon optimized rbcLS operon (GenBank TM accession no. MG000149) coding WT or FIPA Te-Rubisco to produce the tobacco genotypes tob TeLS and tob TeLS-FIPA (see Fig. S4 for further detail). C-E, the tob TeLS and tob TeLS-FIPA plants could be grown in tissue culture (C) but not in soil because both produced finite levels of L 8 S 8 Te-Rubisco as quantified by [ 14 C]-CABP binding (D) and confirmed by native PAGE (E). Loading controls included tobacco Rubisco (tob), known amounts of Te-Rubisco (pTrc TeLS ) expressed in E. coli, and an empty vector (Trc) negative control. No L 8 S 8 Te-Rubisco band was evident by Coomassie staining in the tob TeLS and tob TeLS-FIPA leaf protein samples but detected by immunoblotting using a Se-Rubisco antibody. *, non-Rubisco tobacco protein that separates at the same location as Te-Rubisco by native PAGE.

Novel solutions for improving Rubisco catalysis
plasmid pHue, expressed in BL21(DE3) E. coli, purified by immobilized metal affinity chromatography and the 10-kDa His 6 -ubiquitin sequence removed as previously described (33). PRK activity (k cat ) was measured using a NADH enzymelinked assay as described (34).

Library construction and Rubisco selection using the RDE2 screen
The rbcLS operon in pTrc TeLS (first-generation library) or pTrc TePA (second-generation library) were amplified from 10 ng of plasmid DNA by error-prone PCR using the primers Trc55 (5Ј-GAGGTATATATTAATGTATCG-3Ј) and Trc33 (5Ј-ATCTTCTCTCATCCGCCA-3Ј) and the Genemorph II random mutagenesis kit (Agilent), as per the manufacturer's recommendations. Libraries were transformed into CaCl 2competent RDE2 cells and plated onto LB medium containing 32 g/ml chloramphenicol, 200 g/ml ampicillin, 100 g/ml kanamycin, 0.5 mM IPTG, and 0.05 to 0.25% (w/v) arabinose. After growing at 25°C for 3-7 days in air containing 2% (v/v) CO 2 , the faster-growing colonies were replated, and then their pTrc TeLS plasmids were purified, sequenced, and retransformed into RDE2, and the screen was repeated to confirm their selective advantage.

Rubisco content and catalysis
Cyanobacteria Rubisco expression was induced in XL-1Blue E. coli transformed with pTrc TeLS , pTrcSynLS (20), or their mutant plasmid derivatives with 1 mM IPTG at 28°C. After 6 h the cells were harvested by centrifugation (5 min at 4°C, 6200 ϫ g), and the cell pellets were N 2 -frozen and stored at Ϫ80°C. Their soluble protein was isolated following lysis using a French pressure cell (140 MPa) in ice-cold extraction buffer (100 mM EPPES-NaOH, pH 8.05, 15 mM MgCl 2 , 0.5 mM EDTA, 1 mM PMSF, 2.5 mM DTT). After centrifugation (3 min, 2°C, 15,000 ϫ g), the supernatant was mixed with an equal volume of reaction buffer (100 mM EPPES-NaOH, pH 8.05, 15 mM MgCl 2 , 0.5 mM EDTA) containing 40 mM NaH 14 CO 3 and incubated at 25°C for 8 or 12 min (technical repeats) to activate Rubisco. The 14 CO 2 fixation assays (0.5 ml of total volume) were performed in 7.7 ml of septum-capped scintillation vials in reaction buffer containing 10 g ml Ϫ1 carbonic anhydrase and saturating (0.4 mM) RuBP synthesized and purified according to Kane et al. (36). All assay components were equilibrated in CO 2 -free air (i.e. 20.9% (v/v) O 2 in N 2 ) prior to adding a series of five differing amounts of NaH 14 CO 3. The final 14 CO 2 concentrations were 0 -240 M or 0 -350 M when assaying the wild-type and mutant variants of Te-Rubisco or Se-Rubisco, respectively. The assays were initiated by addition of 20 l of 14 CO 2 -activated E. coli soluble protein. RuBP-independent 14 CO 2 fixation control assays were run for each protein sample and contained H 2 O in place of RuBP. The assays were terminated after 1 min by rapid mixing with 0.2 ml of 20% (v/v) formic acid and then dried at 80°C before adding 0.5 ml of H 2 O and mixing with 1 ml of Ultima-Gold scintillant (PerkinElmer). The fixed 14 C was measured in a Tri-carb 4910TR scintillation counter with the carboxylase activity between the technical repeats varying by Ͻ2%, confirming the extracted Rubisco was fully activated and stable.
The specific activity of each NaH 14 CO 3 stock was determined in assays (n ϭ 4) containing the highest 14 CO 2 concentration and 5 nmol of pure RuBP. These assays were allowed to react for 30 or 90 min to ensure full RuBP fixation. After acid treatment and drying, the scintillation counter 14 C values were divided by 5 to derive the specific activity value, which varied between 1500 and 1800 cpm/nmol CO 2 fixed.
The CO 2 levels in the assays were calculated using the Henderson-Hasselbalch-derived equation,

Novel solutions for improving Rubisco catalysis
where V/v is the ratio of reaction vial headspace (V) to assay volume (v); C t is the total NaHCO 3 concentration (including the 0.4 mol of NaH 14 CO 3 in the E. coli soluble protein); q is the CO 2 solubility at 1 atm (0.0329 mol/liter/atm at 25°C); R is the universal gas constant (0.082057 liter/atm/K/mol); T is the assay temperature (298 K); and dissociation constants are 6.251 (pK 1 ) and 10.329 (pK 2 ). The 14 (20,35). The soluble protein concentration in each E. coli extract was quantified using a Coomassie dye binding assay against BSA. Assays to measure the K m for RuBP of Te-Rubisco (see legend to Table 1) were undertaken in reactions containing 20 mM NaH 14 CO 3 and a series of six RuBP concentrations (0 -150 M).
Rubisco S C/O measurements were made as described by Kane et al. (36) using recombinant Rubisco purified from E. coli by anion exchange chromatography and then Superdex 200 (GE Life Sciences) size exclusion column chromatography (35). Each purified Rubisco was equilibrated with a gas mixture of 0.05% (v/v) CO 2 and 99.95% (v/v) O 2 (accurately mixed using Wostoff gas mixing pumps) at 25°C in a replica septum seal 20-ml glass vial assays comprising 1 ml of specificity buffer (30 mM triethanolamine, pH 8.1, 10 mM MgSO 4 , 10 g/ml carbonic anhydrase). After 1 h the reactions were initiated by the addition of 1 nmol [1-3 H]-RuBP (10 MBq/nmol), and then 10 units of alkaline phosphatase (Sigma) was added 30 -60 min later. The resulting [ 3 H]glycerate and [ 3 H]glycolate products were then separated by HPLC and measured by scintillation counting as described (36). The S C/O factor was calculated using the following equation, where from R is the ratio of [ 3 H]glycerate to [ 3 H]glycolate; M O 2 and M CO 2 are the mole fractions of O 2 and CO 2 , respectively, in the assay; and 0.037 is the ratio between the aqueous solubility of O 2 and CO 2 at 25°C.

PAGE analyses
The leaf and E. coli soluble protein extracts were prepared and analyzed by SDS-PAGE, native PAGE, and immunoblot analysis as described (35).

Tobacco chloroplast transformation
The plasmids pLEVTeLS (GenBank TM accession no. MG000149) and pLEVTeLS FIPA were biolistically transformed into the plastome of the cm trL tobacco genotype as previously described (16). The resulting transplastomic genotypes tob TeLS and tob TeLS-FIPA coded a synthetic rbcLS operon and the aadA marker gene (coding spectinomycin resistance) in place of the tobacco rbcL gene (Fig. 5A; see also Fig. S4 for additional trans-formation and RNA blotting detail). Correct transgene insertion was confirmed by fully sequencing the PCR product amplified from total leaf genomic DNA isolated using the DNeasy plant mini kit and primers LSD (5Ј-CACGGAATTCGTGTC-GAGTAG-3Ј) and LSZ (5Ј-ATCCTTCTTTATTTCCTGC-3Ј).

Leaf photosynthesis simulations
Photosynthetic CO 2 assimilation rates (A) in tobacco at 25°C under varying chloroplast CO 2 partial pressures (C C ) were simulated as the minimum of the Rubisco carboxylase-limited (A c ) and light-limited (A j ) CO 2 -assimilation rates modeled according to Ref. 37, using the Rubisco kinetic values listed in Table S1 and a leaf Rubisco content (m) of 30 mol active sites m Ϫ2 (or 5 mol active sites m Ϫ2 for the assembly impaired Se-Rubisco producing tobacco line tob-SeLS, (19)); a maximal RuBP regeneration rate (J max ) of 160 mol m Ϫ2 s Ϫ1 ; a mitochondrial respiration rate (R d ) of 1 mol m Ϫ2 s Ϫ1 and using the solubility constants 0.0334 M bar Ϫ1 (s c ) and 0.00126 M bar Ϫ1 (s o ) to calculate the CO 2 (C c ) and O 2 (O c ) concentrations in the chloroplast.