Model-driven design and evolution of non-trivial synthetic syntrophic pairs

Synthetic microbial communities are attractive for applied biotechnology and healthcare applications through their ability to efficiently partition complex metabolic functions. By pairing auxotrophic mutants in co-culture, nascent E. coli communities can be established where strain pairs are metabolically coupled. Intuitive synthetic communities have been demonstrated, but the full space of cross-feeding metabolites has yet to be explored. A novel algorithm, OptAux, was constructed to design 66 multi-knockout E. coli auxotrophic strains that require significant metabolite cross-feeding when paired in co-culture. Three OptAux predicted auxotrophic strains were co-cultured with an L-histidine auxotroph and validated via adaptive laboratory evolution (ALE). Time-course sequencing revealed the genetic changes employed by each strain to achieve higher community fitness and provided insights on mechanisms for sharing and adapting to the syntrophic niche. A community model of metabolism and gene expression was utilized to predict the relative community composition and fundamental characteristics of the evolved communities. This work presents a novel computational method to elucidate metabolic changes that empower community formation and thus guide the optimization of co-cultures for a desired application.

To study the characteristics of designed and optimized communities, a community model of 120 metabolism and expression (ME-model) was constructed [32][33][34] (Figure 1C). Such a modeling 121 approach was necessary since previous methods of genome-scale community modeling have 122 focused on studying the metabolic flux throughout community members (using M-models) 123 without consideration of the enzymatic cost of proteins and pathways that drive these metabolic 124 processes. As proteome optimization via niche partitioning and cell specialization is a driving 125 factor of community formation in ecological systems [35][36][37][38], it is essential to consider 126 proteomic constraints when studying bacterial communities. To this end, community ME-models 127 were successfully utilized to interpret the nascent communities and were used to suggest 128 approaches to optimize the evolved co-cultures and potentially modulate metabolic cross-129 feeding. 130 131 132 Figure 1. Study Overview A: An algorithm was developed to de novo predict reaction deletions that will produce E.

133
coli strains auxotrophic for a metabolite of interest. B: From the set of auxotrophic strain designs, pairs were selected 134 to determine whether they were capable of forming a nascent syntrophic cross-feeding community. C: The chosen 7 co-cultures were both evolved via adaptive laboratory evolution and modeled using a genome-scale model of E. coli 136 metabolism and expression (ME-model) [19,20]. The model predictions of optimal strain abundances and metabolite 137 cross-feeding were verified using resequencing data from the co-culture wet-lab experiments. modifications were made to derive OptAux from RobustKnock. First, the inner growth rate 146 optimization was replaced so that OptAux can be run at a predetermined minimum growth rate 147 bound (set_biomass constraint Figure 2B). This ensures that OptAux designs are auxotrophic 148 at all growth rates (Figure 2A). Second, the objective coefficient was reversed in order to allow 149 the algorithm to optimize for metabolite uptake as opposed to secretion. Third, a constraint was 150 added to allow the model to uptake any additional metabolite that can be consumed by the 151 model (trace_metabolite_threshold constraint Figure 2B). For simulations in which this 152 threshold value was set above zero, all possible exchange metabolites included in the model 153 had their lower bound set to the trace_metabolite_threshold value to compete with a target 154 metabolite uptake, allowing the "specificity" of the knockout solution to be adjusted. Specificity,155 in this case, refers to whether the mutant strain will be auxotrophic for a given metabolite in the 156 presence of other metabolites. High specificity solutions are auxotrophic for only one metabolite, possible in the E. coli K-12 MG1655 metabolic network, which could be used to understand the 178 possible niches of E. coli could inhabit in natural or synthetic communities [42]. 179 OptAux Solution Characteristics 180 The OptAux strain designs were broken into two major categories based on the number of 181 metabolites which, when supplemented, restore cell growth: 1) Essential Biomass 182 Component Elimination Designs (EBC, Figure 3B) and 2) Major Subsystem Elimination 183 Designs (MSE, Figure 3A). The EBC designs are characterized as auxotrophic strains with 184 high metabolite specificity. They were broken into two subcategories: specific auxotrophs (only 185 one metabolite can restore growth, Figure S2) which consists of 104 (23 unique) knockout sets, 186 and nonspecific auxotrophs (defined as strains in which less than 5 metabolites can restore 187 growth, Figure S2) which consists of 55 (20 unique) knockout sets. The specific and nonspecific 188 EBC designs were preferred at high trace metabolite threshold values. There is significant 189 overlap between OptAux predicted EBC designs, and known E. coli auxotrophic mutants 190 [14,[43][44][45][46][47][48][49][50][51][52][53][54]. A summary of experimentally characterized OptAux designs are presented in Table  191 S1. Of note, there are five designs that were not found to be previously characterized in the 192 scientific literature, and these present novel E. coli auxotrophs. 193 194 MSE designs were analyzed as novel auxotrophic strain designs. These were defined as strains 195 in which five or more metabolites could restore growth and consisted of the remaining 69 (23 196 unique) sets of knockouts. At low trace metabolite thresholds, MSE designs were the preferred 197 OptAux solution. This knockout strategy was often accomplished through knockouts to block 198 metabolic entry points into anabolic subsystems. One such example of an MSE design is given 199 in Figure 3B. Here a three reaction knockout design of the FUM, PPC, and MALS reactions can 200 be rescued by one of the four compounds in the figure (citrate, L-malate, 2-oxoglutarate, or L-201 asparagine) at an average required uptake flux of 0.4 mmol gDW -1 hr -1 to grow at a rate of 0.1 202 hr -1 . These rates are higher than the fluxes needed to rescue the EBC design in Figure 3A, 203 which requires uptake of 0.024 mmol gDW -1 hr -1 on average to grow at a rate of 0.1 hr -1 . Another 204 design was a glutamate synthase (GLUSy) and glutamate dehydrogenase (GLUDy) double 205 knockout which effectively blocks the entry of nitrogen into amino acid biosynthesis by 206 preventing its incorporation into 2-oxoglutarate to produce L-glutamate. This renders the cell 207 unable to produce all amino acids, nucleotides, and several cofactors. In order to grow at a rate 208 predicted MSE knockouts disrupt significant biological processes, they produce auxotrophies 227 that require much larger amounts of metabolite supplementation in order to grow, compared to 228 EBC designs (e.g., Figure S3). This makes MSE E. coli mutants attractive from a microbial 229 community perspective because they would require a pronounced rewiring of the metabolic flux 230 of their partner stains in co-culture to secrete the high amount of the auxotrophic metabolite 231 needed for community growth. 232 Adaptive Laboratory Evolution of Auxotrophic E. coli Co-cultures Upon inoculation into the first flask of batch growth, each of the co-culture's growth rates were 245 low (<0.05 hr -1 ) suggesting the strains initially showed minimal cooperativity or metabolic cross-246 feeding ( Figure S4). Following approximately 40 days of ALE, all three co-culture combinations 247 had evolved to establish a nascent community, indicated by an increase in the co-culture growth 248 rate. There was diversity in the endpoint batch growth rates among the independently evolved 249 triplicates for each of the ΔhisD & ΔpyrC and the ΔhisD & ΔgdhAΔgltB co-cultures with endpoint 250 growth rates ranging from 0.09-0.15 hr -1 and 0.08-0.15 hr -1 , respectively. The four successfully 251 evolved independent replicates for the ΔhisD & ΔgltAΔprpC co-cultures also showed endpoint 252 growth rate diversity ranging from 0.12-0.19 hr -1 (Table 1, Figure 4A). The relatively large 253 range in endpoint growth rates for all co-cultures suggests that a subset of replicates evolved to 254 a less optimal state and could be further improved if given more time to evolve. 255 To probe the metabolic strategies of the three co-culture pairs, the genomes of the populations 260 were resequenced at several time points over the course of the 40 day evolution (Figure 4A). 261 The resequencing data was used to identify gene region duplications and acquired mutations 262 ( Figure 4B) that provided insight into the specific mechanisms employed by the co-cultures to 263 establish cooperation. 264

265
The relative strain abundance of each mutant was also tracked to understand the dynamics of 266 community composition in the synthetic co-culture. Each starting strain contained unique 267 characteristic mutations (Table S3) which could act as a barcode to track the community 268 composition ( Figure 4B, Table 1). The breseq mutation identification software [55] was used to 269 calculate the frequency of each of these characteristic mutations within a sequenced co-culture. 270 The relative frequency of the characteristic mutations was used to approximate the fraction of 271 each strain within the co-culture population. This analysis showed that 2 of the 3 co-culture 272 combinations maintained similar relative fractions of the two member strains, whereas one co-273 culture, ΔhisD & ΔpyrC, consistently maintained a relative ΔpyrC abundance of near three 274 quarters of the total population (71-81%, Table 1). Alternatively, the relative abundance of each 275 strain in the populations was predicted by comparing the read coverage of the deleted genes 276 relative to the mean, which showed good agreement with the characteristic mutation-based 277 predictions (Figures S4-5). 278  (Table S5). This included a 121 base pair deletion and a SNP in the binding site of the 304 ArgR repressor in the 5` UTR of hisJ ( Figure 5). The mutation in the argR ORF consisted of a 305 frameshift insertion early in the coding sequence and persisted throughout ALE #8, appearing in 306 the ΔhisD endpoint clone (Table S6). ArgR functions to repress L-arginine uptake and 307 biosynthesis as well as the L-histidine ABC uptake complex [58] in response to elevated L-308 arginine concentrations. All of these mutations could improve L-histidine uptake in the ΔhisD 309 strains either by directly increasing the efficacy of the HisJMPQ ABC uptake system or by 310 preventing ArgR mediated repression of this transporter. 311 (GABA) or L-glutamate. Such a mutation could improve community fitness by facilitating the 358 cross-feeding of either these metabolites to the ΔgdhAΔgltB strain since it is predicted to be 359 auxotrophic for both metabolites (Table S4).  the gltIJKL duplication over the course of the evolution, suggesting L-glutamate or L-aspartate is 440 the preferred cross-feeding metabolite over p-aminobenzoyl-glutamate (Figure 7). 441 Community ME-models were created for each of the three evolved co-culture sets (Figure S10). 458 The models were constructed based on the assumption that, in order to form a stable community when growing exponentially, the strains in co-culture must be growing, on average, 460 at an equal rate. Mass balance conversion terms could then be used to relate the metabolic flux 461 that a strain contributes to the shared compartment and its fractional abundance (see 462 Methods). This approach offered a means to understand which factors drive the structure of the 463 newly established communities (i.e., the relative abundance of the community members) and, 464 ultimately, how this relates to metabolite cross-feeding. 465

466
The community ME-models have the capability of assessing how the community composition 467 could vary depending on the identity of the metabolite that is cross-fed or the enzyme efficiency 468 of the community members. The role of the cross-fed metabolites in defining the structure of the 469 community was assessed using the community ME-models by: 1) allowing metabolic cross-470 feeding to remain unrestricted and 2) restricting the cross-feeding to only one metabolite. When 471 the metabolite cross-feeding was left unrestricted (i.e., any metabolite restoring growth in either 472 strain was allowed to cross-feed in the simulation, Supplemental Text, Figure S11) computed 473 cross-feeding profiles were complex and prediction of the identity of the cross-fed metabolite did 474 not strongly point to one potential metabolite ( Figure S12). However, when turning to the 475 sequencing data, there was general agreement between predicted and experimentally inferred 476 optimal community structure which provided confidence in using the proposed modeling 477 approach (Figure S11). 478 479 Alternatively, the second approach to assess the influence of metabolite cross-feeding on 480 community composition involved restricting the simulation to cross-feed only one of the 481 metabolites computationally predicted to restore growth in the MSE strain. In doing so, the 482 identity of the metabolite being cross-fed could be related to the optimal community growth rate 483 and structure. This approach additionally offered a way to narrow the set of optimal or near 484 optimal cross-feeding metabolites that would be predicted to be cross-fed in vivo. The computations predicted that the ΔhisD & ΔpyrC co-culture would have a community composition 486 and growth rate robust to the metabolite being cross-fed with a slightly higher community growth 487 rate if orotate, uracil, uridine monophosphate, or uridine were cross-fed. The optimal 488 composition of the community was predicted to be skewed toward low percentages (~20%) of 489 the ΔhisD strain for all metabolites in this co-culture. The ΔhisD & ΔgltAΔprpC and ΔhisD & 490 ΔgdhAΔgltB co-cultures, on the other hand, were sensitive to the cross-feeding metabolite 491 where the community structure depended on the identity of the cross-feeding metabolite (Figure  492 8A). For these two co-cultures, the ΔhisD & ΔgltAΔprpC and ΔhisD & ΔgdhAΔgltB pairs were 493 computationally predicted to achieve higher community growth rates when cross-feeding L-494 glutamate, 2-oxoglutarate, citrate, or L-glutamine and 4-aminobutanoate, L-aspartate, L-495 glutamine, L-glutamate, L-alanine, or L-asparagine, respectively. 496 was performed on models constrained to only cross-feed the metabolite that was inferred from the resequencing data 503 (2-oxoglutarate, orotate, and L-glutamate, respectively) ( Table 2). C) Box plots of experimentally measured 504 abundances for each sample (bottom two rows, gray, and dark blue) and the computationally-predicted optimal strain 505 abundances following variation in the cross-feeding metabolite (top row, blue) and in strain proteome efficiency 506 (second and third row, red, and yellow). 507 508 Community ME-models further enable an examination of how each strain's proteome 510 "efficiency" may affect co-culture characteristics when growing in its community niche. Such an 511 analysis was performed by altering a ME-model parameter for each strain corresponding to how 512 efficiently it can export the metabolite that is cross-feeding its partner strain (see Methods). This 513 parameter can be used as a proxy for cellular proteome investment in wasteful or inefficient 514 processes when synthesizing and exporting a metabolite, which is likely to occur in substantial 515 amounts until the strains further adapt to grow as a community. That is, the cells will not be able 516 to optimally rearrange their proteome and metabolic fluxes to efficiently grow as a community 517 over this short-term evolution. It is possible, however, that some strains in co-culture will be able 518 to reorganize their proteome to secrete the necessary metabolite more or less efficiently than 519 their partner strain ( Table 2). The proteome efficiency analysis showed that the community 520 compositions of all three co-cultures were moderately sensitive to this parameter ( Figure 8B). 521 Further, the pairs showed a bimodal behavior depending on whether the ΔhisD strain was more 522 or less efficient than its partner ( Figure 8B). The community models predicted that if the export 523 processes of the ΔhisD strain require a greater protein investment relative to the default export 524 efficiency parameter, the abundances of the ΔhisD strain will increase in the community. 525 Conversely, if the partner strain requires greater protein investment, the community composition 526 remains stable and unchanged. The optimal predicted community composition for the two 527 analyses shown in Figure 8A and B are summarized in Figure 8C. The figure shows general 528 agreement between the computed optimal community compositions and the experimentally 529 inferred community composition, even after varying key features of the community simulation. 530 This suggests that community ME-models have the potential to be useful tools for 531 understanding the behavior of simple communities. 532 This study has demonstrated a novel workflow to design, optimize, and computationally interpret 539 non-trivial syntrophic co-cultures to better understand the characteristics of simple microbial 540 community formation. The simple communities consisted of two strains of E. coli K-12 MG1655 541 which required, in order to grow themselves, the growth of their partner strain. To design the 542 communities to possess characteristics more attractive from an engineering perspective, a 543 novel algorithm, termed OptAux, was used. This algorithm was used to design highly 544 auxotrophic strains which, when paired in co-culture, require high levels of metabolic cross-545 feeding in order for the community to grow. Three co-cultures consisting of OptAux designs 546 were tested in vivo and optimized via adaptive laboratory evolution. By analyzing the genetic 547 changes observed throughout the evolution we could infer the cellular changes underlying 548 improvements in the fitness of the highly metabolically-coupled communities. This work thus 549 provided new insight into cellular mechanisms for establishing syntrophic growth. A community 550 ME-model was developed to computationally interpret the communities and their fundamental 551 properties. Such models are the first to offer a means to study, on the genome-scale, how 552 efficient proteome allocation to metabolic functions in the community members can influence the 553 structure of the nascent microbial communities. 554 OptAux Can be Used to Design Novel Communities 555 To facilitate the design of co-culture communities requiring significant metabolic rewiring and 556 cross-feeding, we constructed the OptAux algorithm to find reaction knockouts that will create 557 auxotrophic strains requiring high amounts of metabolites for growth (Figure 2). OptAux 558 returned two kinds of solutions depending on the parameters used, so-called Major Subsystem 559 Elimination (MSE) and Essential Biomass Component Elimination (EBC) designs (Figure 3). 560 EBC designs are specific with regard to which metabolites are required for the strain to grow 561 and correspond to auxotrophs that have been validated in previous studies [14,50-54]. OptAux 562 EBC predictions resulted in eight designs that were previously verified experimentally and five 563 predictions of untested auxotrophs (Table S1). Conversely, the MSE designs are 564 computationally predicted to grow when supplemented with a any of variety of different 565 metabolites and represent largely new designs that have not been characterized experimentally, 566 though some of the single gene knockout MSE designs were grown in co-culture in [16]. MSE 567 auxotrophs in co-culture need high levels of cross-feeding in order to grow (0.05 and 0.2 mmol 568 gDW -1 hr -1 on average for an EBC and MSE strain to grow at a rate of 0.1 hr -1 , respectively), 569 requiring significant metabolic rewiring in its partner strain (Figure S13).

ALE was Successfully Applied to Increase Fitness of Co-culture 571
Four OptAux predicted auxotrophic E. coli mutants were constructed in vivo, confirmed as 572 auxotrophs, and grown in co-culture. A growth rate selection pressure was applied on these 573 nascent, poorly growing communities via ALE. Three co-cultures of an ΔhisD EBC strain paired 574 with an MSE strain showed reproducible improvements in growth rate throughout the course of 575 the ALE ( Table 1). Under these conditions each of the strains had to rewire its metabolic 576 network to both secrete a metabolite required by its partner strain and efficiently import the 577 metabolite needed to grow itself through mutations that were identified, effectively establishing a 578 new microbial community. By selecting for growth rate, a novel indirect selection pressure was 579 applied on each strain to increase the secretion and uptake of the cross-fed metabolites, thus 580 improving the growth of the co-culture community. This evolution design therefore has potential 581 as a system to self-optimize microbial strains as industrial producers of metabolites of interest. 582 583 Throughout the course of adaptive laboratory evolution, the nascent communities improved 584 community fitness by acquiring beneficial mutations (Tables S5-7, Figures S7-9 Beyond enabling an analysis of how the co-cultures were capable of establishing syntrophy, the 596 sequencing data provided a measure of the structure of the community in terms of relative strain 597 abundance. All auxotrophic mutants contained a unique characteristic starting mutation (Table  598 S3), which was used to track the relative abundance of each member of the co-culture 599 community throughout the evolutions. Community structures appeared to remain remarkably 600 consistent both across ALE replicates of the same strain combinations and over time throughout 601 the ALE lineages (Table 1, Figures S4-5). This finding was corroborated by using the coverage 602 of the gene deletion regions in population resequencing (Figures S5-6). instance, all ALE lineages acquired mutations targeting the ABC uptake system for L-histidine 610 ( Figure 5). Given that all of the three evolved co-culture sets included a strain that was an EBC 611 auxotroph for L-histidine, community growth logically would increase if histidine uptake was 612 improved in this strain via these genetic changes. Similarly, the three MSE strains that were 613 paired with the L-histidine auxotroph, ΔpyrC, ΔgdhAΔgltB and ΔgltAΔprpC, displayed evidence 614 in their resequencing data to suggest that the strains were being cross-fed orotate, glutamate 615 and 2-oxoglutarate, respectively ( Table 2). A community ME-model was constructed for each of 616 the three communities and the model simulations predicted a hierarchy, where clusters of metabolites. In each case, the mutation data inferred cross-feeding metabolites were contained 619 in one of the top computationally predicted clusters. 620 Community ME-modelling Allows for Analyzing Co-culture 621 Composition 622 Community ME-models were employed to understand how the proteome efficiency of each 623 strain drives community composition. ME-models are uniquely capable of addressing this 624 question because they directly incorporate the proteomic cost of catalyzing a metabolic process, 625 which is particularly necessary in this system as there is an inherent proteome cost of each 626 strain to cross-feed the necessary metabolite in co-culture [75]. Kinetic parameters, which play a 627 role in dictating proteome cost in these community ME-models, were therefore systematically 628 adjusted to understand how each strain's proteomic "efficiency" affected the simulation 629 characteristics. The simulations predicted that, for all of the three co-cultures, the proteomic 630 efficiency of the ΔhisD would have the largest impact on the relative abundance of each co-631 culture member (Figure 8B). This is an expected finding due to the fact that the ΔhisD strain 632 has the larger cross-feeding burden since it is paired with an MSE strain in each case. Further, 633 when the ΔhisD secretion proteome efficiency was decreased, the community ME-model 634 predicted its optimal abundance in the co-culture would actually increase. Though unintuitive, 635 this prediction is in agreement with a paradox predicted in a previous computational study of 636 community dynamics [76]. In addition to proteome efficiency, the ME-model predicted that the 637 identity of the metabolite being cross-fed has an effect on optimal community composition 638 ( Figure 8A). The distributions of possible community compositions based on varying 639 metabolites and proteome efficiency aligned well with two of the three co-cultures (ΔhisD & potentially offers a means to study how changes in the characteristics of each strain in co-642 culture will affect the optimal community structure and growth behavior. 643 644 From an industrial perspective, shifting the community composition could increase the 645 production of a specific metabolite of interest. Therefore, this modeling method offers a way to 646 predict how, for instance, LacZ or other unused (i.e., non beneficial) proteins could be efficiently 647 overexpressed to lower a strain's proteome efficiency and alter community composition, thus 648 improving the yield of metabolite secretion. Additionally, this modeling method suggests that the 649 identity of the cross-feeding metabolite can bias the optimal community composition to some implementing a "reverse" version of RobustKnock where the algorithm would optimize the 680 uptake of a metabolite at the maximum growth rate. A "reverse" RobustKnock implementation 681 would lead to strain designs that must take up a high amount of the target metabolite when 682 approaching the maximum growth rate (Figure S1A). In order for a strain to be truly auxotrophic 683 for a particular metabolite, however, it must be required at all growth rates (Figure 2A For the simulations ran in this study (S1 Data), the set_biomass value was set as 1/10 the 691 maximum growth rate for the wild-type simulation in in silico glucose minimal media 692 supplemented with the metabolite whose uptake is being maximized. parameters were set to their default. 749 detected in several samples when ΔhisD was in co-culture with ΔpyrC. This is likely due to the 762 low frequency of the ΔhisD strain in that particular population. In those cases, the ΔhisD strain 763 abundance was predicted using only the frequency of the lrhA/alaA intergenic SNP (Figure S5). 764

765
The second method used the contig read alignment to compare the coverage of the deleted 766 genes in each strain to the fit mean coverage of the sample. As an example, for a strain paired 767 with the ΔhisD strain, the average coverage of the base pairs in the hisD ORF divided by the fit 768 mean for that sample, would give an approximation of its relative abundance in the population. 769 As with the characteristic mutation approach, if the two genes are knocked out in the strain, the 770 average coverage of the two genes is used to make the approximation ( Figure S5). 771 772 When reporting the relative abundance predictions, the predicted abundances of each strain 773 was normalized by the sum of the predicted abundances of the two strains in co-culture. This 774 ensured that the abundance predictions summed to one. Predictions made using the two 775 described methods showed general agreement (Figure S6).  Using this community modeling approach, the fractional abundance (Xi) of each strain in the co-798 culture was implemented as a parameter that could be varied from 0 to 1, which in turn had on 799 impact on the optimal growth state of the community. Simulations were ran varying XStrain1 800 (abundance of strain 1) from 0.05 to 0.95 and the community growth rate was optimized. The 801 metabolites that were allowed to be cross-fed in simulation were limited to the set of metabolites 802 that can computationally restore the growth of each auxotroph (Table S4). 803 804 For the community simulations, the iJL1678b [32] model of E. coli K-12 MG1655 was used with 805 the uptake of metabolites in the in silico glucose minimal growth media into the shared 806 compartment left unconstrained, as the ME-model is self limiting [33]. The non-growth 807 associated ATP maintenance and the growth associated ATP maintenance were set to the 808 default parameter values in the model. The RNA degradation constraints were removed to 809 prevent high ATP costs at the low community growth rates. Since, the newly formed 810 communities are highly unoptimized and growing slowly, the unmodeled/unused protein fraction 811 parameter was set to 75%. If a metabolite had a reaction to import the metabolite across the 812 inner membrane, but no export reaction, a reaction to transport the metabolite from the cytosol 813 to the periplasm was added to the model. For more on the model parameters, refer to [32] and 814 [33]. 815

816
To vary the proteomic efficiency (keff) of the export metabolites, first the exchange reaction into 817 the shared compartment for all potential cross-feeding metabolites except the metabolites 818 inferred from the experimental data ( Table 2) was constrained to zero. Then the enzymatic 819 efficiency of the outer membrane transport process of only the inferred metabolite was altered in 820 each strain. The outer membrane transport reactions for each inferred metabolite (i.e.,HIStex, 821 GLUtex, AKGtex, and OROTtex for L-histidine, L-glutamate, 2-oxoglutarate, and orotate, 822 respectively) have multiple outer membrane porins capable of facilitating the transport process. and their P1 phage lysates were generated for the transduction into the receiving single KO 835 strains. For instance, the ΔgltA or ΔgltB knockout strain was a donor strain and the ΔprpC or 836 ΔgdhA knockout strain was a receiving strain (Table S2) Sunrise plate reader, equivalent to an OD600 of ~1 on a traditional spectrophotometer with a 1 846 cm path length), a point at which nutrients were still in excess and exponential growth had not 847 started to taper off. Four OD600 measurements were taken from each flask, and the slope of Resequencing 850 Co-culture populations samples were collected at multiple points throughout the ALE and 851 sequenced. Additionally, the starting mutant strains and both mutants isolated from the ALE 852 endpoint samples were sequenced. The ΔhisD endpoint clone was unable to be isolated via 853 colony selection for ALE #11 . Genomic DNA of the co-culture populations and mutant clones 854 was isolated using the Macherey-Nagel NucleoSpin tissue kit, following the manufacturer's 855 protocol for use with bacterial cells. The quality of isolated genomic DNA was assessed using 856 Nanodrop UV absorbance ratios. DNA was quantified using the Qubit double-stranded DNA 857 (dsDNA) high-sensitivity assay. Paired-end whole genome DNA sequencing libraries were 858 generated using Illumina's Kappa kit and run on an Illumina MiSeq platform with a PE600v3 kit. 859 DNA sequencing data from this study will be made available on the Sequence Read Archive