Minimization of proteome reallocation explains metabolic transition in hierarchical utilization of carbon sources

enzyme constraint, metabolic model, mixed carbon sources, metabolic transition 12 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 2

13 Cells choose between alternative pathways in metabolic networks under diverse environmental 14 conditions, but the principles governing the choice are insufficiently understood, especially in 15 response to dynamically changing conditions. Here we observed that a lactic acid bacterium 16 Bacillus coagulans displayed homolactic fermentation on glucose or trehalose as the sole carbon 17 source, but transitioned from homolactic to heterolactic fermentation during the hierarchical 18 utilization of glucose and trehalose when growing on the mixture. We simulated the observation by 19 dynamic minimization of reallocation of proteome (dMORP) using an enzyme-constrained genome-20 scale metabolic model of B. coagulans, which coincided with our multi -omics data. Moreover, we 21 evolved strains to co -utilize mixed carbon sources, circumventing the hierarchical utilization and 22 inactivating the choice of heterolactic fermentation. Altogether, the findings suggest that upon rapid 23 environmental changes bacteria tend to minimize proteome reallocation and accordingly adjust 24 metabolism, and dMORP would be useful in simulating and understanding cellular dynamics. 25 26 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 3

27 Cells have alternative pathways, i.e., redundancy, in metabolic networks to adapt to diverse 28 environmental conditions. For instance, while many microbes and mammalian cells use respiration 29 for ATP production, some of them also harbor fermentation as an alternative ATP -producing 30 pathway, which enables not only survival under anaerobic conditions but also fast growth in the 31 presence of oxygen, i.e., aerobic fermentation (1). The alternative pathways empower cells to 32 choose between distinct metabolic strategies, but the principles governing the cellular choice are 33 insufficiently understood (2-4). 34 35 Experimental and theoretical studies suggest that the choices of alternative pathways and 36 metabolic strategies reflect tradeoffs, mostly between metabolic efficiency, e.g., yield, of pathways 37 and cellular resources, e.g., proteome, invested for enzymes of the pathways (5 -9). For instance, 38 fermentation has a lower ATP yield but is more proteome efficient than respiration, and therefore 39 favors fast -growing cells subject to the limited proteome resources (2,10). These studies are 40 predominantly focused on steady -state cultures, e.g., chemostats, of model microbes, and 41 therefore the cellular choice might be a result of a relatively long-term adaption to the environment. 42 However, much less is known whether and how cells can transition between alternative pathways 43 under dynamically changing environments, which are ubiquitous in both natural and artificial 44 biological systems (11). 45 46 Here we studied a lactic acid bacterium Bacillus coagulans , which has alternative pathways 47 converting glycolytic carbon sources to lactate, i.e., homolactic and heterolactic fermentation 48 (12,13). The former converts carbon sources to lactate via the Embden –Meyerhof–Parnas (EMP) 49 pathway, while the later the phosphoketolase pathway with a lower lactate yield (14,15). We 50 observed that B. coagulans displayed the homolactic fermentation on glucose or trehalose as the 51 sole carbon source, but transitioned from homolactic to heterolactic fermentation in the hierarchical 52 utilization of glucose and trehalose when growing on the mixture. We hypothesized that the 53 metabolic transition could be interpreted by dynamic minimization of reallocation of proteome, and 54 performed enzyme -constrained metabolic modeling and multi -omics analysis to test the 55 hypothesis. Finally, by adaptive laboratory evolution (ALE) we evolved strains that can co -utilize 56 the mixed carbon sources, inactivating the heterolactic fermentation. 57 58

59 Lactate yield significantly changed in hierarchical utilization of glucose and trehalose 60 B. coagulans is widely used in the industrial lactate fermentation (13,16,17), where hydrolysates of 61 low-cost starch-based materials are commonly used as the carbon sources (18,19) considering 62 cost and benefit. However, the co-existence of multiple carbon sources in the medium may lead to 63 the hierarchical utilization (20 -25) and thus the waste of substrates (26), the extension of 64 fermentation cycle (27) and the difficulty in product separation and purification (28). Therefore, 65 identifying the principles governing the dynamic behavior of B. coagulans during the hierarchical 66 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 4 utilization of mixed carbon sources would be of significance from both basic and applied science 67 perspectives. 68 69 We performed batch fermentation of B. coagulans on the mixture of glucose and trehalose, a 70 disaccharide with a high percentage in the hydrolysates (29). We observed the hierarchical 71 utilization of glucose and trehalose, which divided the entire fermentation process into three phases 72 (Fig. 1A). In the glucose utilization phase (P gluc, 0-12 h), glucose was exclusively consumed and 73 lactate was rapidly produced with the lactate yield on glucose reaching as high as 0.897 g/g, which 74 was similar to the yield (i.e., 0.882 g/g) on glucose as the sole carbon source (Fig. 1B) and close 75 to the yield of homolactic fermentation (14,30). Upon glucose depletion, the lag phase occurred 76 (Plag, 12-16 h), in which the lactate production stopped, the biomass concentration (measured by 77 OD620) declined (Fig. 1A) and a few organic acids were slightly consumed ( SI Appendix Fig. S1). 78 In the trehalose utilization phase (P tre, 16 -84 h), trehalose was consumed, and lactate was 79 produced again (Fig. 1A). Interestingly, the lactate yield on trehalose in P tre was only 0.526 g/g, 80 which was much lower than that (i.e., 0.851 g/g) on trehalose as the sole substrate (Fig. 1B) and 81 close to the theoretical lactate yield of heterolactic fermentation (14). This was in line with the 82 increased yields of byproducts in Ptre (SI Appendix Table S1). The significant change in the lactate 83 yield during the hierarchical utilization of glucose and trehalose indicated distinct strategies to 84 catabolize the carbon sources in Pgluc and Ptre. 85 86 87 Figure 1. Fermentation results on various carbon sources in 5 L bioreactor by B. coagulans. 88 (A) Fermentation processes on the combination of glucose and trehalose as the mixed carbon 89 sources (left), sole glucose (middle) and sole trehalose (right). Data in the plots are shown with 90 mean ± s.d. of three biologically replicates. (B) Lactate yields under various carbon source phases 91 or conditions. Data in the table are shown with mean ± s.d. of three biologically replicates. 92 93 Dynamic minimization of reallocation of proteome captured metabolic changes 94 To gain a systematic understanding, we reconstructed a genome -scale metabolic model of the 95 strain B. coagulans DSM 1 = ATCC 7050 (31) called iBcoa620, which was evaluated by the Memote 96 score (32) and validated by growth on various carbon sources (SI Appendix Fig. S2). Subsequently, 97 we converted iBcoa620 into an enzyme-constrained version eciBcoa620 using the GECKO toolbox 98 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 5 (33), allowing for understanding of metabolic changes by taking into account the enzyme usage 99 and proteome allocation. 100 101 To simulate the changes during the fermentation process, we performed dynamic flux balance 102 analysis (dFBA) (34) with eciBcoa620 (Fig. 2A). The dFBA approach can divide the entire period 103 into multiple time intervals and obtain the flux distribution at each time interval by performing 104 classical FBA (35) which requires a predefined objective function. While growth maximization 105 appeared reasonable to simulate each time intervals in P gluc, it could not predict the cellular 106 behavior accurately after glucose depletion (Fig. 2B), meaning that cells adjusted the objective in 107 response to the changing environment. Therefore, we attempted other commonly used objective 108 functions (36), including to maximize lactate production and to maximize non -growth-associated 109 maintenance (NGAM), and found that they also failed to capture the cellular behavior in P tre (SI 110 Appendix Fig. S3), i.e., the former predicted much higher lactate production while the latter no 111 lactate production (Fig. 2B). 112 113 114 Figure 2. Simulation of batch fermentation under mixed carbon sources by eciBcoa620. (A) 115 Dynamic simulation of the fermentation process. In the dFBA simulation of P gluc the growth rate 116 was maximized. To simulate P tre, various objective functions could be adopted within the dFBA 117 framework. Additionally, we developed an approach to dynamically minimize the proteome 118 reallocation with the enzyme -constrained models. (B) Simulation of lactate production using 119 different objective functions. Minimization of proteome reallocation predicted the lactate production 120 in Ptre better compared to the others. Data in the plot are shown with mean ± s.d. of three biologically 121 replicates. (C) Root mean square errors of the different objective functions in predicting P tre. Root 122 mean square error quantifies the difference between the predicted and experimental fluxes. The 123 predicted fluxes refer to the production fluxes of organic acids in all simulation steps in P tre, while 124 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 6 the experimental fluxes refer to the values obtained by fitting the detected organic acid 125 concentrations in Ptre. (D) Comparison of the relative fluxes of the central carbon metabolism (CCM) 126 between Pgluc (represented by 7 h) and Ptre (represented by 48h). Relative flux is calculated as the 127 percentage of the absolute flux of each reaction to the uptake flux of the substrates. Note that one 128 unit of trehalose uptake flux was converted to two unit of glucose uptake flux. T6P, trehalose 6 ‐129 phosphate; G6P, glucose 6 ‐ phosphate; F6P, fructose 6 ‐ phosphate; FBP, fructose 1,6 ‐130 bisphosphate; DHAP, dihydroxyacetone phosphate; G3P, glyceraldehyde 3 ‐phosphate; 13BPG, 131 glycerate 1,3 ‐ bisphosphate; 3PG, 3 -phosphoglycerate; 2PG, 2-Phospho-D-glycerate; PEP, 132 phosphoenolpyruvate; 6PG, gluconate 6 ‐ phosphate; E4P, erythrose 4 ‐ phosphate; S7P, 133 sedoheptulose 7 -phosphate; R5P, ribose 5 ‐phosphate; RL5P, ribulose 5 ‐phosphate; XL5P, 134 xylulose 5-phosphate; Acetyl-P, Acetyl phosphate; PGI, glucose -6-phosphate isomerase; PKFA, 135 6-phosphofructokinase; FBA, fructose -1,6-bisphosphate aldolase; GAP, glyceraldehyde -3-136 phosphate dehydrogenase; PGK, phosphoglycerate kinase; GPMI, phosphoglycerate mutase; 137 ENO, enolase; PYK, pyruvate kinase; LDH, L -lactate dehydrogenase; RPI, ribose 5 -phosphate 138 isomerase; RPE, ribulose -phosphate 3 -epimerase; XFP, phosphoketolase; FSA, fructose -6-139 phosphate aldolase; TKT, transketolase; BUDA, acetolactate decarboxylase; ACKA, acetate 140 kinase; ALD, aldehyde dehydrogenase; ADHP, alcohol dehydrogenase. 141 142 Considering the fact that proteome reallocation requires frequent protein synthesis and 143 degradation, which is of high cost to cells (37), we hypothesized that cells tend to dynamically 144 minimize the proteome reallocation in response to rapid environmental changes. To test it, we 145 developed the approach of dynamic minimization of reallocation of proteome (dMORP) to simulate 146 the cellular behavior after glucose depletion (Fig. 2A). The approach leverages the enzyme -147 constrained models, which enable explicit calculation of enzyme usage, and minimizes the sum of 148 absolute differences of all enzyme usage values between previous and current time intervals 149 (Materials and methods). By employing dMORP, we found that the lactate production in P tre was 150 well predicted (Fig. 2B). Moreover, dMORP outperformed other tested objective functions as it led 151 to the smallest Root Mean Square Error (RMSE) (Fig. 2C) and it was capable of predicting 152 production of byproducts in P tre such as acetate and acetoin ( SI Appendix Fig. S3). Therefore, we 153 demonstrated that dynamic minimization of proteome reallocation could be a more realistic 154

function for describing the cellular behavior after glucose depletion. 155 156 Inspection of the simulated flux distributions enabled us to explain the observed lower lactate yield 157 in Ptre. We compared the relative fluxes, i.e., normalized to carbon source uptake, of the central 158 carbon metabolism (CCM) in Ptre (simulated at 48 h, i.e., middle in P tre) to those in Pgluc (simulated 159 at 7 h, i.e., mid -exponential phase in P gluc) and identified in P tre a much greater fraction of carbon 160 flux toward the key reaction of heterolactic fermentation phosphoketolase (Fig. 2D) as well as 161 related reactions such as ribulose-phosphate 3-epimerase and acetate kinase, resulting in carbon 162 loss in the form of acetate and ethanol and thereby the decreased lactate yield (13). This was also 163 in line with the comparison between Ptre and the sole trehalose condition (SI Appendix Fig. S4). 164 165 To compare the metabolic changes between different phases during the hierarchical utilization of 166 glucose and trehalose, we selected the absolute metabolic fluxes simulated at 7 h, 14 h and 48 h, 167 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 7 representing Pgluc, Plag and Ptre, respectively (Fig. 3A). We found that the metabolic fluxes in P lag 168 and Ptre were comparable but generally less than those in P gluc (Fig. 3A), exemplified by the lower 169 glycolytic fluxes, which was due to the considerably slower substrate uptake (Fig. 1A). Interestingly, 170 in order to dynamically minimize the proteome reallocation after glucose depletion, the model 171 predicted high fluxes through both forward and backward directions for some reversible reactions 172 in the glycolysis pathway (Fig. 3A), which minimized the reduction in the enzyme usage meanwhile 173 allowed for a sharply decrease in the net flux. Such high fluxes through both reaction directions 174 might suggest the control of reaction thermodynamics on fluxes as reported previously (38 -41). 175 Additionally, we found in the comparison between Pgluc and Ptre that the reactions phosphoketolase, 176 ribulose-phosphate 3 -epimerase and acetate kinase displayed smaller changes in fluxes than 177 glycolytic reactions (SI Appendix Fig. S5), which ensured an enhanced flux fraction in response to 178 a large reduction in the substrate uptake flux, indicating their essential contribution to heterolactic 179 fermentation, i.e., the low lactate yield. 180 181 182 Figure 3. Dynamical changes in metabolic fluxes and enzyme usage in the fermentation 183 process predicted by eciBcoa620. (A) The absolute metabolic flux distributions of the CCM at 7 184 h, 14 h, and 48 h. These three time points represent P gluc, Plag and P tre, respectively. At 7 h the 185

function was to maximize growth while at 14 h and 48 h it was to minimize proteome 186 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 8 reallocation. (B) Enzymes usage and net fluxes of the central carbon metabolism from 7 h to 48 h. 187 The log2FC value represents the change in enzyme usage or net flux relative to the 7 h point. 188 189 The enzyme -constrained model allowed us to investigate the enzyme usage under each time 190 interval. By plotting the change in enzyme usage versus fermentation time we found that the 191 enzyme usage remained almost constant after glucose depletion at 12 h (Fig. 3B), indicating the 192 effectiveness of the minimization of proteome reallocation between adjacent time intervals. We also 193 noticed that while the change in fluxes kept in step with the change in enzyme usage for reactions 194 like phosphoketolase, many glycolytic enzymes displayed tremendous difference between changes 195 in fluxes and changes in enzyme usage in Plag and Ptre (Fig. 3B), which could be explained by their 196 elevated reverse fluxes (Fig. 3A). This may also indicate the control of metabolites, rather than 197 gene expression, over glycolytic fluxes (42). 198 199 To compare with the dMORP approach, we examined the simulations performed using lactate 200 production maximization as the objective function in a dynamic manner (Fig. 2B, SI Appendix Fig. 201 S6). As expected the simulated flux distribution at 48 h displayed homolactic fermentation, i.e., 202 trehalose was mostly converted to lactate without acetate production ( SI Appendix Fig. S6A), 203 inconsistent with the experimental observation (SI Appendix Fig. S1). In addition, we found that the 204 simulated enzyme usage of many reactions, e.g., glycolytic reactions, altered much more 205 considerably compared with the simulations using dMORP (SI Appendix Fig. S6B). 206 207 Multi-omics data coincided with dynamic minimization of proteome reallocation 208 To validate our hypothesis, we collected transcriptomics and proteomics data at 7 h, 14 h and 48 209 h, representing P gluc, Plag and Ptre, respectively. The transcriptomics and proteomics data showed 210 high consistency among biological triplicates while clear variability among different phases ( SI 211 Appendix Fig. S7). Based on the data, we mapped changes in mRNA and protein levels between 212 phases onto the CCM (Fig. 4A). By comparing Ptre versus Pgluc, we found that most reactions in the 213 glycolysis and phosphoketolase pathway were slightly downregulated (Fig. 4A) as the changes in 214 mRNA and protein levels were either subtle or not significant (SI Appendix Table S2). This is in line 215 with the changes in the enzyme usage simulated by the model with dMORP (Fig. 3B), rather than 216 those with maximizing lactate production (SI Appendix Fig. S6B). In addition, the comparison of Ptre 217 versus Plag showed subtle, even smaller, changes in the mRNA and protein levels of the glycolysis 218 and phosphoketolase pathway (Fig. 4A, SI Appendix Table S2), as predicted by the model (Fig. 219 3B). Therefore, the data showed that while the metabolic fluxes altered considerably after glucose 220 depletion (Fig. 3) the gene expression levels did not alter so much as the metabolic fluxes. This 221 indicates that the cellular proteome tends to maintain minimal adjustment upon rapid environmental 222 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 9 changes and that there are other mechanisms than gene expression that might explain the 223 metabolic changes. 224 225 226 Figure 4. Analysis of transcriptomics, proteomics and intracellular metabolite data. (A) 227 Transcriptomics and proteomics comparison for the CCM between phases. The omics samples of 228 three phases were obtained from three biological replicates, respectively. Significance analysis 229 between two phases were performed using the two -sided t-test, and a cut -off adjusted P value 230 (Padjusted) of 0.01 was adopted. "na" indicates that the data is not available. (B) The co mparison 231 between substrate-product concentration ratio and forward-backward flux ratio. “+”and “-” represent 232 the forward and backward flux of the reaction, respectively. 233 234 As suggested by the model, the mismatch between the changes in the metabolic fluxes and gene 235 expression levels could be explained by the increased reverse fluxes (Fig. 3). To validate it, we 236 measured intracellular metabolite concentrations at 7 h, 14 h and 48 h, representing P gluc, Plag and 237 Ptre, respectively (SI Appendix Fig. S8 and S9). Based on the thermodynamics of reactions (38,43), 238 the ratio of backward over forward flux of a reversible reaction should be proportional to the ratio 239 of product over substrate concentration ( SI Appendix Supplementary Note). Due to the limited 240 coverage of the metabolite data, we could only calculate for the reaction glucose -6-phosphate 241 isomerase whose substrate and product were both detected, and found that for this reaction the 242 ratio of measured product over substrate concentration increased after glucose depletion (Fig. 4B), 243 supporting the contribution of the reverse flux. To make full use of the metabolite data, we lumped 244 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 10 a few glycolytic reactions to carry out the calculation and also found a similar trend, i.e., increased 245 ratio of the final product over the substrate concentration (Fig. 4B). It should be noted that the ratio 246 of the product over substrate concentration did not change so significantly as the ratio of the 247 simulated backward over forward flux, which was as expected since the model did not consider the 248 changes in other factors that could regulate metabolic fluxes such as enzyme saturation (43) and 249 enzyme activity (44). 250 251 In conclusion, after glucose depletion cells tended to utilize the proteome expressed on glucose to 252 consume trehalose, in which the mismatch between the relatively higher gene expression levels of 253 glycolytic enzymes and the lower glycolytic flux might be contributed by the enhancement of some 254 reverse fluxes as indicated by the metabolite data. In addition, the flux through the phosphoketolase 255 pathway did not decrease so much as the glycolytic flux, which led to an increased fraction of 256 carbon source toward phosphoketolase pathway and thus lowered lactate yield. 257 258 Evolved strains performed homolactic fermentation by co -utilizing the mixed carbon 259 sources 260 Considering that the heterolactic fermentation was triggered by the dynamic switch from glucose 261 to trehalose utilization, we hypothesized that simultaneous utilization of the carbon sources would 262 prevent B. coagulans from choosing the heterolactic fermentation. To test it, we carried out ALE 263 (45,46) to obtain strains that could co-utilize glucose and trehalose with improved lactate yield (Fig. 264 5A) (Materials and methods). As a result, we obtained four evolved strains that can utilize glucose 265 and trehalose simultaneously ( SI Appendix Fig. S10), which was further confirmed by the 5 L 266 bioreactor cultivation of one of the evolved strains Ev3 (Fig. 5B). Notably, the evolved strain Ev3 267 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 11 outperformed the wildtype strain in terms of the final lactate titer (35.892 v.s. 26.220 g/L) and the 268 total fermentation time (24 v.s. 84 h) (Fig. 1A & Fig. 5B). 269 270 Figure 5. Adaptive laboratory evolution improved fermentation process. (A) Brief diagram of 271 adaptive laboratory evolution experiment. The addition of 2-Deoxy-D-glucose to trehalose aimed 272 at efficiently utilizing trehalose in the absence of glucose, with the intention of eliminating the 273 catabolite repression effect. After 40 generations, growth and lactate production were basically 274 stable. The strains with notably improved lactate production were selected for subsequent shake 275 flasks validation and 5 L bioreactor experiments. (B) Profiles of fermentation process with mixed 276 carbon sources in 5 L bioreactor by the evolved strain Ev3. Data in the plot are shown with 277 mean ± s.d. of three biologically replicates. (C) Relative fluxes through PFKA and XFP over the 278 substrate uptake fluxes of the evolved strain Ev3 and wildtype strain on different carbon sources. 279 (D) Lactate yields on different pairs of carbon sources o f evolved strain and wildtype strain. The 280 cultivation process was carried out in 250 mL shake flasks, with the carbon source concentration 281 of 20 g/L for each. The concentrations of lactate and carbon sources at 36 h were measured to 282 calculate the lactate yields. Data in bar diagram are shown with mean ± s.d. of three biologically 283 replicates. Statistical analysis was performed using two -sided t-test. Statistically significant 284 differences are described as follows: ** P<0.01, * P<0.05, ns (not significant). 285 286 The lactate yield of the strain Ev3 on glucose and trehalose (Fig. 5B) was close to that of the 287 wildtype strain on sole glucose and sole trehalose (Fig. 1B), indicating a homolactic fermentation 288 mode in the strain Ev3. To investigate further, we simulated the metabolic fluxes of the strain Ev3 289 on glucose and trehalose, and found that when utilizing glucose and trehalose simultaneously the 290 strain Ev3 performed in a similar manner as the wildtype strain on sole glucose and sole trehalose 291 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 12 in terms of the relative fluxes of 6 -phosphofructokinase and phosphoketolase (Fig. 5C), meaning 292 that the simultaneous utilization of carbon sources can eliminate the metabolic transition caused 293 by the switch of carbon sources. By comparing transcriptomics data we also found either subtle 294 (e.g. |log2FC| < 0.5) or not significant changes in expression levels of genes in the CCM between 295 the strain Ev3 on glucose and trehalose and the wildtype strain with homolactic fermentation ( SI 296 Appendix Table S3). Altogether, the physiological, metabolic flux and gene expression data cross-297 confirmed that the strain Ev3 employed homolactic fermentation when co -utilizing glucose and 298 trehalose. 299 300 We sequenced the genomes of the four evolved strains and found that the gene crr encoding 301 glucose-specific phosphotransferase enzyme IIA component (EIIA Glc) was mutated in all evolved 302 strains (SI Appendix Table S4), in line with the findings that in various microbes the deletion of the 303 gene crr diminished glucose repression on other carbon sources (47 -49). In the presence of 304 glucose, EIIAGlc is dephosphorylated which can inhibit transporters (50) and the genes involved in 305 the catabolism of other carbon sources (51). By comparing the transcriptomics data between the 306 strain Ev3 and the wildtype strain on the mixture of glucose and trehalose, we found many 307 differentially expressed genes in the phosphotransferase system ( SI Appendix Fig. S11), which 308 might be associated with the dysfunction of EIIAGlc. 309 310 To investigate whether the simultaneous utilization is specific to the pair of glucose and trehalose, 311 we tested other combinations of carbon sources. We found that the strain Ev3 was also capable of 312 co-utilizing glucose with fructose, mannose and maltose, whereas the wildtype strain employed 313 hierarchical utilization of them ( SI Appendix Fig. S12). Moreover, we found that the strain Ev3 314 resulted in significantly higher lactate yields on various pairs of mixed carbon sources than the 315 wildtype strain (Fig. 5D). Therefore, the evolved strain has improved capacity of utilizing diverse 316 types of mixed carbon sources, which could be a promising candidate for industrial fermentation 317 where the media consist of complex carbon sources. 318 319

320 Here, we observed that B. coagulans utilized glucose and trehalose hierarchically and transitioned 321 from homolactic to heterolactic fermentation during the hierarchical utilization (Fig. 1). We 322 hypothesized that the dynamic minimization of proteome reallocation might result in this 323 observation. To test it, we reconstructed the enzyme -constrained genome-scale metabolic model 324 of B. coagulans , and developed the approach of dMORP. The model with dMORP accurately 325 simulated the transition from homolactic to heterolactic fermentation of B. coagulans (Fig. 2), and 326 predicted increased reverse fluxes through glycolytic reactions (Fig. 3), which explained the 327 noticeable reduction in the glycolytic fluxes while subtle reduction in the gene expression levels of 328 glycolytic enzymes. Our omics data validated the model simulations: the transcriptomics and 329 proteomics data confirmed small changes in gene expression levels of glycolytic enzymes (Fig. 330 4A), and the intracellular metabolite data supported the increased reverse fluxes through glycolytic 331 reactions based on reaction thermodynamics (Fig. 4B). Moreover, we evolved strains to circumvent 332 the hierarchical utilization and prevent the metabolic transition. Therefore, we conclude that the 333 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 13 dynamic minimization of proteome reallocation can explain the cellular choice of the metabolic 334 strategies upon the change, i.e., hierarchical utilization of glucose and trehalose. 335 336 The biological basis of the minimization of proteome reallocation is the high cost and slow 337 adjustment of protein synthesis and degradation (37,52,53) that the cells should consider when 338 adapting to rapid perturbations. Additionally, the minimization of proteome reallocation is in line 339 with the findings of proteome reserves (54,55), i.e., the proteome is not fully optimized for the 340 environmental condition where the cell is living, as a suboptimal proteome could reduce the cost 341 and time of reallocation upon the adaption to a new condition. By integrating the minimization of 342 proteome reallocation into the constraint -based modeling framework, we developed MORP and 343 dMORP that introduce the past proteome as an internal constraint, which provide additional 344

functions for simulating cellular behaviors (56). Particularly, dMORP would serve as a 345 promising algorithm in the field of dynamic metabolic modeling to predict cellular kinetics (57,58) 346 and history-dependent behaviors (52,59,60). 347 348 The approach of minimization of metabolic adjustment (MOMA) has been widely used to simulate 349 cellular behaviors in response to genetic or environmental perturbations (61 -63) by minimizing the 350 sum of absolute changes in metabolic fluxes before and after perturbations (64). Compared with 351 minimizing the adjustment of metabolic fluxes, we argue that it is more biologically meaningful to 352 minimize the adjustment of proteome, which now can be achieved by MORP using enzyme -353 constrained models. It should be noted that MORP can predict the mismatch between enzyme and 354 flux levels caused by co -occurrence of both forward and backward fluxes of a reaction, which 355 appears to be an impossible mission for the MOMA approach implemented on conventional GEMs. 356 In addition, it is less computationally expensive to perform the dynamic simulations on enzyme -357 constrained models than on fine -grained models, such as ME -models (65,66) that explicitly 358 formulate gene expression processes (67,25). Therefore, we expect extensive applications of 359 MORP and dMORP with the development of enzyme-constrained models (68). 360 361 In summary, we applied systems biology approaches to decipher the principles of the cellular 362 response of B. coagulans upon the hierarchical utilization of mixed carbon sources, and identified 363 the dynamic minimization of proteome reallocation to be the cellular objective that dominates the 364 metabolic behavior in response to rapid environmental changes. In addition, we implemented the 365 minimization of proteome reallocation as an effective objective function in the metabolic modeling 366 framework, which would be useful for simulating and understanding cellular responses upon 367 perturbations. 368 369 370

371 Microorganism, media and culture conditions 372 B. coagulans DSM 1 = ATCC 7050 was purchased from the American Type Culture Collection. 373 374 Seed medium was MRS medium (69), simplified chemically defined medium (CDM) used for 375 fermentation in this study, i.e., MCDM3++, which was developed in our previous study (31). 20 g/L 376 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 14 glucose and 20 g/L trehalose were used as the mixed carbon source to better reproduce the 377 phenomenon of hierarchical utilization. 378 379 Seed culture and shake flask fermentation were performed in 250 mL shake flasks with 100 mL 380 working volume at 50 ℃ and 100 rpm, pH was buffered with CaCO3. The cells in seed culture were 381 collected and washed twice using 100 mM phosphate buffer at around 12 h. Collected cells were 382 resuspended in ultrapure water with a determined volume and then transferred to the fermentation 383 medium in shake flasks or 5 L bioreactors. Shake flask fermentation was used to investigate the 384 metabolic properties of different combinations of mixed carbon sources. 385 386 Batch fermentation was performed in a 5 L bioreactor with 4 L working volume at 50 °C, 100 rpm 387 and 7.2 L/h aeration, and pH was automatically controlled at 5.5 using 25% w/v Ca(OH) 2. The 388 samples of different phases were collected for determination of biomass, metabolites, 389 transcriptome and proteome. Fermentations were performed in triplicates for all conditions. 390 391 Determination of biomass and metabolites 392 The optical density (OD) was detected using a spectrophotometer at 620 nm to characterize the 393 biomass concentration. 394 395 Lactate, acetate, citate, pyruvate and acetoin in the fermentation broth were determined by high ‐396 performance liquid chromatography (Shimadzu LC‐20AT) (12). The Hi-Plex H (300 mm × 7.7 mm) 397 column was used with 210 nm wavelength and 50 °C column temperature; the mobile phase was 398 0.01 mol/L sulfuric acid solution and the flow rate was set to 0.4 mL/min. 399 400 The carbon sources in the fermentation broth were determined by ion chromatography (Dionex 401 ICS-3000) with Dionex Carbopac PA 20 column, the method was referred to the previous 402 description (29) and modified. The pretreatment process of the samples was performed as follows: 403 1) The samples were diluted to achieve sugar concentrations within the detection range. 2) 404 Na2C2O4 was used to precipitate calcium ions in the culture medium. 3) Proteins were removed 405 from the samples using ethanol. 4) The processed solution was evaporated to obtain the solid form 406 of the sugar, which was then dissolved in water to obtain the sample for testing. The mobile phase 407 was ultrapure water and 200 mmol/L NaOH, respectively. The elution procedure was ultrapure 408 water and 200 mmol/L NaOH eluted at 9:1 ratio for 18 min, then adjusted the ratio to 2:8 to elute 409 for 17 min, and finally adjusted to 9:1 to balance the column for 35 min, and the flow rate for the 410 entire process was set to 0.4 mL/min. 411 412 Reconstruction of the genome-scale metabolic model 413 The model iBcoa620 was reconstructed based on the model iBag597 (13): 1) A combined draft 414 model from KEGG (70) and MetaCyc (71) pathway databases based on the genome sequences of 415 B. coagulans DSM 1 = ATCC 7050 (GCF_000832905.1) was reconstructed by RAVEN 2.4.0 (72). 416 2) Another draft model was reconstructed by ModelSEED (73) database based on the genome 417 annotation of RAST (74). 3) Shared reactions in the draft models were integrated with iBag597. 4) 418 The IDs of genes, metabolites, and reactions were unified into the form of the KEGG database. 5) 419 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 15 Reactions with mass imbalance were corrected. 6) The memote score system was used to assess 420 the model quality. 7) The model was validated by comparing the predicted and measure growth 421 rates under different carbon sources. 422 423 The enzyme -constrained model eciBcoa620 was reconstructed using the GECKO 3 .0 toolbox. 424 Note that kcat values were obtained from BRENDA (75) and GotEnzymes (76) databases and 425 predicted using DLKcat tool (77). Molecular weights of all enzymes were obtained from UniProt 426 (78) database. 427 428 MORP, dFBA and dMORP 429 MORP is implemented on the enzyme -constraint models, which can estimate both metabolic and 430 enzyme usage fluxes. MORP estimates the metabolic and enzyme usage fluxes of the model upon 431 a perturbation based on the minimization of proteome reallocation. Thus, the fluxes of the model 432 from a reference condition should be used as the input of MORP, which can be obtained in advance 433 by other algorithms, e.g., FBA, with the enzyme -constraint models. The core of MORP is to 434 minimize the sum of absolute differences of enzyme usage fluxes between the reference and the 435 perturbed conditions: 436 min ∑|𝑣enzyme,i,perturb − 𝑣enzyme,i,ref| m i=1 437 subject to 𝑺 · 𝒗 = 𝟎 438 𝑙𝑏j ≤ 𝑣j ≤ 𝑢𝑏j 439 440 where venzyme,i,perturb and venzyme,i,ref represent fluxes of enzyme usage reaction i under perturbed and 441

conditions, and m represents the total number of enzyme usage reactions in the model. 442 As MORP estimates fluxes for the perturbed condition, S and v represent the stoichiometric matrix 443 and flux vector of the model for the perturbed condition, and vj, lbj and ubj are the flux, lower and 444 upper bounds of the reaction j, respectively. 445 446 dFBA was used to simulate for the hierarchical utilization of glucose and trehalose, in which an 447

function should be determined for different time intervals. As mentioned before three 448

functions were used, i.e., maximization of growth, lactate production and NGAM. 449 Generally, dFBA solves: 450 max 𝑣o,k k ∈ N 451 subject to 𝑺 · 𝒗 = 𝟎 452 𝑣s,k = 𝑣s,max[𝑆]k−1 𝐾m + [𝑆]k−1 453 𝑙𝑏j ≤ 𝑣j ≤ 𝑢𝑏j 454 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 16 where k represents the k th time interval, N represents the number of time intervals of the phase of 455 interest and thus is a positive integer, vo,k is the flux of the objective reaction in P gluc, i.e., the 456 biomass formation, or the flux of the objective reaction after glucose depletion, i.e., the biomass 457 formation, lactate production, or NGAM production, vs,k is the flux of the uptake reaction of the 458 sugar, i.e., glucose or trehalose, of the kth time interval, [ S]k-1 is the sugar concentration of the (k -459 1)th time interval, vs,max is the maximum sugar uptake rate, and Km is the sugar saturation constant. 460 461 dMORP was proposed to simulate the entire period after glucose depletion in the hierarchical 462 utilization of glucose and trehalose. Here dMORP solves: 463 min ∑|𝑣enzyme,i,r − 𝑣enzyme,i,r−1| 𝑚 𝑖=1 r ∈ N 464 subject to 𝑺 · 𝒗 = 𝟎 465 𝑣s,r = 𝑣s,max[𝑆]r−1 𝐾m + [𝑆]r−1 466 𝑙𝑏j ≤ 𝑣j ≤ 𝑢𝑏j 467 where r represents the r th time interval after glucose depletion, N represents the number of time 468 intervals of the phase of interest and thus is a positive integer, venzyme,i,r and venzyme,i,r-1 represent the 469 flux of enzyme usage reaction i for the rth and (r-1)th time interval, respectively, vs,r is the flux of the 470 trehalose uptake reaction of the r th time interval, and [S]r-1 is the trehalose concentration of the (r -471 1)th time interval. 472 473 All the simulations were performed with the COBRA toolbox 3.0 (80) on Matlab R2019a, and the 474 solver was Gurobi 9.1 (https://www.gurobi.com/). 475 476 Transcriptome analysis 477 Cells in the fermentation broth were collected and washed three times with phosphate buffer saline 478 for subsequent RNA extraction. Total RNA was isolated from cells using TRIzol Reagent following 479 the instructions and genomic DNA was removed using DNase I (TaKara). Subsequently, RNA 480 quality was determined using a Agilent2100 Bioanalyzer and quantified using ND-2000 (NanoDrop 481 Technologies). 482 483 RNA library construction was performed using TruSeqTM RNA sample preparation Kit from Illumina 484 (San Diego, CA). The rRNA was removed using Ribo -Zero Magnetic kit (epicenter), and double -485 stranded cDNA was reverse transcribed using random primers (Illumina) and SuperScript double -486 stranded cDNA synthesis kit (Invitrogen, CA). To synthesize the second strand of cDNA, it uses 487 dUTP instead of dTTP, and then eliminates the second strand of cDNA containing dUTP, so that 488 the library contained only the first strand of cDNA. After PCR amplification by Phusion DNA 489 polymerase (NEB) and TBS380 (Picogreen) quantification, paired-end RNA-seq sequencing library 490 was sequenced with the Illumina Novaseq (2 × 150bp read length). 491 492 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 17 The data generated from Illumina platform were used for bioinformatics analysis, and the reference 493 genome and gene annotation files of B. coagulans DSM 1 = ATCC 7050 were downloaded from 494 the National Center for Biotechnology Information. All of the analyses were performed using the 495 free online platform of Majorbio Cloud Platform (www.majorbio.com). 496 497 Proteome analysis 498 Cells in the fermentation broth were collected and washed three times with phosphate buffer saline 499 for subsequent protein extraction. Sample lysis and protein extraction were performed using SDT 500 (4% SDS, 100 mM Tris -HCl, 1 mM DTT, pH 7.6) buffer, and the protein was quantified using the 501 BCA Protein Assay Kit (Bio-Rad, USA). Protein digestion was performed with trypsin following the 502 filter-aided sample preparation (FASP) procedure. The digested peptides from each sample were 503 subsequently desalted (Empore™ SPE Cartridges C18 (standard density), bed I.D. 7 mm, volume 504 3 mL, Sigma) and concentrated. The protein for each sample were quantified by SDS -PAGE. 100 505 μg peptide mixture for each sample was labeled using TMT reagent according to the instructions 506 (Thermo Scientific). The labeled peptides were separated by High pH Reversed -Phase Peptide 507 Fractionation Kit (Thermo Scientific) to obtain 10 different fractions, the collected fractions were 508 desalted on C18 Cartridges and concentrated. 509 510 LC-MS/MS analysis was performed using a Thermo Scientific Q Exactive mass spectrometer 511 equipped with the Easy nLC system (Proxeon Biosystems, Thermo Fisher Scientific). Peptides 512 were trapped on a reverse phase trap column (Thermo Scientific Acclaim PepMap100, 100 μm × 513 2 cm, nanoViper C18) and separated on the C18 -reversed phase analytical column (Thermo 514 Scientific Easy Column, 10 cm long, 75 μm inner diameter, 3 μm resin). The separation was 515 achieved using buffer A (0.1% formic acid) and buffer B (84% acetonitrile and 0.1% formic acid) at 516 a constant flow rate of 300 nL/min controlled by IntelliFlow technology. The mass spectrometer was 517 operated in positive ion mode. MS/MS analysis was performed using the data -dependent top10 518

for high-energy collision dissociation (HCD) fragmentation by selecting the most abundant 519 precursor ion (300 -1800 m/z) in the measurement scan. The MS raw data was subjected to 520 identification and quantitative analysis using MASCOT engine (Matrix Science, London, UK; 521 version 2.2) and Proteome Discoverer 1.4 software. The proteome detection and analysis process 522 were conducted in shanghai applied protein technology co.ltd. 523 524 Determination of intracellular metabolites 525 The absolute quantification of intracellular phosphate sugars, organic acids, and amino acids was 526 performed using gas chromatography-mass spectrometry (Agilent, GC-7890A, MS-5975C). Before 527 the extraction of intracellular metabolites, a quenching step was performed to terminate cell 528 metabolism. 2 mL of fermentation broth were added to a 50 mL centrifuge tube containing 30 mL 529 of 60% methanol pre -cooled to -27 °C. After quenching for 5 minutes, filter paper with attached 530 cells was obtained by filtration. Simultaneously, 100 μL of U -13C-labeled cell extracts was added 531 to the filter paper as internal standards. The filter paper was then placed in a 75% ethanol solution 532 pre-heated to 80 °C and boiled for 5 min. The filter paper was removed, and the remaining liquid 533 was concentrated to 600 μL using rapid evaporation. A 100 μL portion was taken as the sample 534 and subjected to freeze-drying before derivatization (13,81,82). 535 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 18 536 Amino acids derivatization method: 100 μL of pyridine was added to the freeze -dried sample, and 537 the mixture was dissolved in a 65 °C for 1 hour. After the sample reached room temperature, 100 538 μL of derivatization reagent (N ‐tert‐butyldimethylsilyl ‐N-methyl-trifluoroacetamide with 1% 539 tert‐butyldimethylchlorosilane) was added, and derivatization was carried out in a 65 °C oven for 540 1 hour. The sample was then centrifuged at 12000 rpm for 5 min, and the supernatant was 541 subjected to GC-MS analysis. 542 543 Phosphate sugars and organic acids derivatization method: 50 μL of 20 mg/mL methoxyamine 544 pyridine solution was added to the freeze-dried sample. The mixture was dissolved in a 65 °C oven 545 for 1 hour. After reaching room temperature, 80 μL of N -methyl-N-(trimethylsilyl) 546 trifluoroacetamide/trimethylchlorosilane (MSTFA/TMCS, 1000:50, v/v) was added, and 547 derivatization was carried out in a 65 ℃ oven for 1 hour. The sample was then centrifuged at 12000 548 rpm for 5 minutes, and the supernatant was subjected to GC-MS analysis. 549 550 The GC-MS detection conditions were as follows: the interface temperature of the GC-MS was set 551 at 280 °C, the transfer line temperature was 250 °C, Ionization was done by electron impact with 552 the temperature being 230 °C, the quadrupole temperature was 150 °C, and the injection volume 553 was 1 μL. The chromatographic column used was HP -5MS with dimensions of 30 mm × 0.25 mm 554 × 0.25 μm. For amino acids, the temperature program consisted of an initial column temperature 555 of 100 °C for 1 minute, followed by an increase at a rate of 10 °C/min to reach 320 °C, which was 556 then maintained for 10 minutes. For phosphate sugars and organic acids, the temperature program 557 started with an initial column temperature of 70 °C for 1 minute, followed by an increase at a rate 558 of 10 °C/min to reach 300 °C, which was then maintained for 10 minutes. 559 560 Adaptive laboratory evolution 561 The adaptive laboratory evolution strategy was performed in a 250 mL shake flask with complex 562 medium, 100 rpm, 50 °C. Alternating cultures of glucose and trehalose were utilized, while the 563 trehalose was supplemented with glucose structural analogue (2 -deoxy-D-glucose), which can be 564 absorbed but not be metabolized, so it can be used for screening mutant strains without catabolite 565 repression effect (83,84). The hope was to obtain strains that could utilize glucose and trehalose 566 simultaneously. During the evolution process, the cells were transferred to new medium in the 567 exponential phase and maintained the same concentration (OD620 about 0.5) after inoculation. 568 The initial total sugar concentration during evolution was kept constant at 130 g/L (glucose 130 g/L, 569 2-deoxy-D-glucose 40 g/L + trehalose 90 g/L), and the concentration of 2 -deoxy-D-glucose was 570 determined according to our previous study. Four evolved strains with the expected phenotypes 571 were examined by MRS plate culture, and the strain with faster sugar consumption rate and higher 572 lactate yield was selected for fermentation experiments in 5 L bioreactor. 573 574 Whole-genome sequencing 575 Cells from the exponential phase in seed shake flasks were collected for genome sequencing. 576 Genomic DNA of B. coagulans DSM 1 = ATCC 7050 was extracted using Wizard Genomic DNA 577 Purification Kit (Promega) according to the protocol. Purified genomic DNA was quantified by TBS-578 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 19 380 fluorometer (Turner BioSystems Inc., Sunnyvale, CA). Whole-genome sequencing of wildtype 579 strain and four evolved strains was performed using the de novo sequencing method based on the 580 Illumina NovaSeq6000 platform in this study. The final genome of five strains were annotated from 581 KEGG, GO, COG and other databases using BLASTP, Diamond and HMMER. The same mutated 582 genes of four evolved strains were considered to be genes that could cause phenotypic change. 583 584 Data availability 585 The genome sequences of four evolved strains have been deposited in NCBI database under the 586 BioProject accession number of PRJNA1012373. The transcriptomics data have been deposited 587 in NCBI database under the BioProject accession number of PRJNA1012495. The mass 588 spectrometry proteomics datasets have been deposited in the ProteomeXchange with the dataset 589 identifier PXD045169. 590 591 Code availability 592 The models and code are available at https://github.com/ChenYuGroup/MORP. 593 594 Acknowledgments 595 This study was financially supported by the National Key Research and Development Program of 596 China (2023YFA0913900), the Shenzhen Medical Research Fund (A2303026) and the National 597 Natural Science Foundation of China (21878084). 598 599 Author Contributions: Y.C., Y.W. and Z.L. conceived the study. Z.L., M.C. and J.H. performed 600 the experiments. Z.L. and Y.C. constructed the models and performed the simulations. Z.L. and 601 Y.C. analyzed the data. Y.C., Z.L. and Y.W. wrote the paper. All authors approved the final paper. 602 Competing Interest: The authors declare no competing interests. 603 604

605 1. M. G. Vander Heiden, L. C. Cantley, C. B. Thompson, Understanding the Warburg effect: 606 the metabolic requirements of cell proliferation. Science 2009, 324(5930): 1029-1033. 607 2. M. Basan et al., Overflow metabolism in Escherichia coli results from efficient proteome 608 allocation. Nature 528(7580): 99-104 (2015). 609 3. B. Niebel, S. Leupold, M. Heinemann, An upper limit on Gibbs energy dissipation governs 610 cellular metabolism. Nat. Metab. 1(1): 125-132 (2019). 611 4. Y. Shen et al., Proteome capacity constraints favor respiratory ATP generation. bioRxiv 612 2022.08. 10.503479 (2022). 613 5. T. Pfeiffer, S. Schuster, S. Bonhoeffer, Cooperation and competition in the evolution of 614 ATP-producing pathways. Science 292(5516): 504-507 (2001). 615 6. D. Molenaar, R. Van Berlo, D. De Ridder, B. Teusink, Shifts in growth strategies reflect 616 tradeoffs in cellular economics. Mol. Syst. Biol. 5(1): 323 (2009). 617 7. A. Flamholz, E. Noor, A. Bar-Even, W. Liebermeister, R. Milo, Glycolytic strategy as a 618 tradeoff between energy yield and protein cost. Proc. Natl. Acad. Sci. U.S.A. 110(24): 619 10039-10044 (2013). 620 8. M. Szenk, K. A. Dill, A. M. R. de Graff, Why do fast-growing bacteria enter overflow 621 metabolism? Testing the membrane real estate hypothesis. Cell Syst. 5(2): 95-104 622 (2017). 623 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 20 9. B. Du, D. C. Zielinski, J. M. Monk, B. O. Palsson, Thermodynamic favorability and 624 pathway yield as evolutionary tradeoffs in biosynthetic pathway choice. Proc. Natl. Acad. 625 Sci. U.S.A. 115(44): 11339-11344 (2018). 626 10. Y. Chen, J. Nielsen, Energy metabolism controls phenotypes by protein efficiency and 627 allocation. Proc. Natl. Acad. Sci. U.S.A. 116(35): 17592-17597 (2019). 628 11. R. Schuetz, N. Zamboni, M. Zampieri, U. Sauer, Multidimensional optimality of microbial 629 metabolism. Science 336(6081): 601-604 (2012). 630 12. X. Lv et al., Effect of pH, glucoamylase, pullulanase and invertase addition on the 631 degradation of residual sugar in L-lactic acid fermentation by Bacillus coagulans HL-5 632 with corn flour hydrolysate. J. Taiwan Inst. Chem. Eng. 61, 124-131 (2016). 633 13. Y. Chen et al., Genome-scale modeling for Bacillus coagulans to understand the 634 metabolic characteristics. Biotechnol. Bioeng. 117, 3545-3558 (2020). 635 14. M. G. Gä nzle, Lactic metabolism revisited: metabolism of lactic acid bacteria in food 636 fermentations and food spoilage. Curr. Opin. Food Sci. 2, 106-117 (2015). 637 15. Y. Wang et al., Metabolism characteristics of lactic acid bacteria and the expanding 638 applications in food industry. Front. Bioeng. Biotechnol. 9: 612285 (2021). 639 16. Q. Wang, L. O. Ingram, K. Shanmugam, Evolution of D-lactate dehydrogenase activity 640 from glycerol dehydrogenase and its utility for D-lactate production from lignocellulose. 641 Proc. Natl Acad. Sci. U.S.A. 108, 18920-18925 (2011). 642 17. L. Wang et al., Jerusalem artichoke powder: a useful material in producing high-optical-643 purity L-lactate using an efficient sugar-utilizing thermophilic Bacillus coagulans strain. 644 Bioresour. Technol. 130, 174-180 (2013). 645 18. M. A. Abdel-Rahman, Y. Tashiro, K. Sonomoto, Recent advances in lactic acid 646 production by microbial fermentation processes. Biotechnol. Adv. 31, 877-902 (2013). 647 19. Z. Lu, F. He, Y. Shi, M. Lu, L. Yu, Fermentative production of L (+)-lactic acid using 648 hydrolyzed acorn starch, persimmon juice and wheat bran hydrolysate as nutrients. 649 Bioresour. Technol. 101, 3642-3648 (2010). 650 20. R. Hermsen, H. Okano, C. You, N. Werner, T. Hwa, A growth-rate composition formula 651 for the growth of E. coli on co-utilized carbon substrates. Mol. Syst. Biol. 11, 801 (2015). 652 21. E. Perrin et al., Diauxie and co-utilization of carbon sources can coexist during bacterial 653 growth in nutritionally complex environments. Nat. Commun. 11, 3135 (2020). 654 22. H. Okano, R. Hermsen, T. Hwa, Hierarchical and simultaneous utilization of carbon 655 substrates: mechanistic insights, physiological roles, and ecological consequences. Curr. 656 Opin. Microbiol. 63, 172-178 (2021). 657 23. A. Solopova et al., Bet-hedging during bacterial diauxic shift. Proc. Natl Acad. Sci. U.S.A. 658 111, 7427-7432 (2014). 659 24. H. Okano, R. Hermsen, K. Kochanowski, T. Hwa, Regulation underlying hierarchical and 660 simultaneous utilization of carbon substrates by flux sensors in Escherichia coli. Nat. 661 Microbiol. 5, 206-215 (2020). 662 25. P. Salvy, V. Hatzimanikatis, Emergence of diauxie as an optimal growth strategy under 663 resource allocation constraints in cellular metabolism. Proc. Natl Acad. Sci. U.S.A. 118, 664 e2013836118 (2021). 665 26. C. Sievert et al., Experimental evolution reveals an effective avenue to release catabolite 666 repression via mutations in XylR. Proc. Natl Acad. Sci. U.S.A. 114, 7349-7354 (2017). 667 27. P. Gawand, P. Hyland, A. Ekins, V. J. Martin, R. Mahadevan, Novel approach to engineer 668 strains for simultaneous sugar utilization. Metab. Eng. 20, 63-72 (2013). 669 28. K. Ma et al., Highly efficient production of optically pure L-lactic acid from corn stover 670 hydrolysate by thermophilic Bacillus coagulans. Bioresour. Technol. 219, 114-122 (2016). 671 29. X. Lv, Y. Guo, Y. Zhuang, Y. Wang, Optimization and validation of an extraction method 672 and HPAEC-PAD for determination of residual sugar composition in L-lactic acid 673 industrial fermentation broth with a high salt content. Anal. Methods 7, 9076-9083 (2015). 674 30. Y. J. Wee, J. S. Yun, D. H. Park, H. W. Ryu, Isolation and characterization of a novel 675 lactic acid bacterium for the production of lactic acid. Biotechnol. Bioprocess Eng. 9, 303-676 308 (2004). 677 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 21 31. Y. Chen, F. Dong, Y. Wang, Systematic development and optimization of chemically 678 defined medium supporting high cell density growth of Bacillus coagulans. Appl. 679 Microbiol. Biotechnol. 100, 8121-8134 (2016). 680 32. C. Lieven et al., MEMOTE for standardized genome-scale metabolic model testing. Nat. 681 Biotechnol. 38, 272-276 (2020). 682 33. Y. Chen et al., Reconstruction, simulation and analysis of enzyme-constrained metabolic 683 models using GECKO Toolbox 3.0. Nat. Protoc. 1-39 (2024). 684 34. J. D. Orth, I. Thiele, B. O. Palsson, What is flux balance analysis?. Nat. Biotechnol. 28, 685 245-248 (2010). 686 35. D. Henriques et al., A multiphase multiobjective dynamic genome-scale model shows 687 different redox balancing among yeast species of the Saccharomyces genus in 688 fermentation. mSystems 6, e00260-21 (2021). 689 36. M. Kafri, E. Metzl-Raz, G. Jona, N. Barkai, The cost of protein production. Cell Reports 690 14, 22-31 (2016). 691 37. E. Noor et al., Pathway thermodynamics highlights kinetic obstacles in central 692 metabolism. PLoS Comput. Biol. 10, e1003483 (2014). 693 38. J. Xu et al., Metabolic flux analysis and fluxomics-driven determination of reaction free 694 energy using multiple isotopes. Curr. Opin. Biotechnol. 64: 151-160 (2020). 695 39. J. O. Park et al., Metabolite concentrations, fluxes and free energies imply efficient 696 enzyme usage. Nat. Chem. Biol. 12(7): 482-489 (2016). 697 40. P. Grigaitis, B. Teusink, An excess of glycolytic enzymes under glucose-limited 698 conditions may enable Saccharomyces cerevisiae to adapt to nutrient availability. FEBS 699 Lett. 596(24): 3203-3210 (2022). 700 41. L. Gerosa et al., Pseudo-transition analysis identifies the key regulators of dynamic 701 metabolic adaptations from steady-state data. Cell Syst. 1, 270-282 (2015). 702 42. D. Davidia et al., Global characterization of in vivo enzyme catalytic rates and their 703 correspondence to in vitro kcat measurements. Proc. Natl Acad. Sci. U.S.A. 113, 3401-704 3406 (2016). 705 43. S. Zhao et al., Regulation of cellular metabolism by protein lysine acetylation. Science 706 327, 1000-1004 (2010). 707 44. M. I. Espinosa et al., Adaptive laboratory evolution of native methanol assimilation in 708 Saccharomyces cerevisiae. Nat. Commun. 11, 5564 (2020). 709 45. D. Choe et al., S. C. Kim, B. K. Cho, Adaptive laboratory evolution of a genome-reduced 710 Escherichia coli. Nat. Commun. 10, 935 (2019). 711 46. B. R. Oh et al., The production of 1, 3-propanediol from mixtures of glycerol and glucose 712 by a Klebsiella pneumoniae mutant deficient in carbon catabolite repression. Bioresour. 713 Technol. 130, 719-724 (2013). 714 47. M. Zhao, D. Shi, X. Lu, H. Zong, B. Zhuge, Co-production of 1, 2, 4-butantriol and ethanol 715 from lignocellulose hydrolysates. Bioresour. Technol. 282, 433-438 (2019). 716 48. Q. Liang et al., Comparison of individual component deletions in a glucose-specific 717 phosphotransferase system revealed their different applications. Sci Rep 5, 13200 718 (2015). 719 49. B. Gö rke, J. Stü lke, Carbon catabolite repression in bacteria: many ways to make the 720 most out of nutrients. Nat. Rev. Microbiol. 6, 613-624 (2008). 721 50. A. Nair, S. J. Sarma, The impact of carbon and nitrogen catabolite repression in 722 microorganisms. Microbiol. Res. 251, 126831 (2021). 723 51. D. L. Rothman, P. B. Moore, R. G. Shulman, The impact of metabolism on the adaptation 724 of organisms to environmental change. Front. Cell. Dev. Biol. 11, 1197226 (2023). 725 52. M. Scott, T. Hwa, Shaping bacterial gene expression by physiological and proteome 726 allocation constraints. Nat. Rev. Microbiol. 21, 327-342 (2023). 727 53. C. Wu et al., Enzyme expression kinetics by Escherichia coli during transition from rich to 728 minimal media depends on proteome reserves. Nat. Microbiol. 8(2): 347-359 (2023). 729 54. R. Yu et al., Nitrogen limitation reveals large reserves in metabolic and translational 730 capacities of yeast. Nat. Commun. 11(1): 1881 (2020). 731 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 22 55. D. B. Bernstein, S. Sulheim, E. Almaas, D. Segrè , Addressing uncertainty in genome-732 scale metabolic model reconstruction and analysis. Genome Biol. 22: 1-22 (2021). 733 56. D. W. Erickson et al., A global resource allocation strategy governs growth transition 734 kinetics of Escherichia coli. Nature 551, 119-123 (2017). 735 57. M. Basan et al., A universal trade-off between growth and lag in fluctuating environments. 736 Nature 584, 470-474 (2020). 737 58. B. Cerulus et al., Transition between fermentation and respiration determines history -738 dependent behavior in fluctuating carbon sources. eLife 7: e39234 (2018). 739 59. L. Vermeersch et al., Do microbes have a memory? History-dependent behavior in the 740 adaptation to variable environments. Front. Microbiol. 13: 1004488 (2022). 741 60. J. S. L. Yu et al., Microbial communities form rich extracellular metabolomes that foster 742 metabolic interactions and promote drug tolerance. Nat. Microbiol. 7, 542-555 (2022). 743 61. W. C. Ho, J. Zhang, Evolutionary adaptations to new environments generally reverse 744 plastic phenotypic changes. Nat. Commun. 9, 350 (2018). 745 62. S. Vaud et al., Engineering improved ethylene production: Leveraging systems biology 746 and adaptive laboratory evolution. Metab. Eng. 67, 308-320 (2021). 747 63. D. Segre, D. Vitkup, G. M. Church, Analysis of optimality in natural and perturbed 748 metabolic networks. Proc. Natl Acad. Sci. U.S.A. 99, 15112-15117 (2002). 749 64. C.J. Lloyd et al., COBRAme: A computational framework for genome-scale models of 750 metabolism and gene expression. PLoS Comput. Biol. 14, e1006302 (2018). 751 65. A.M. Reimers, H. Knoop, A. Bockmayr, R. Steuer, Cellular trade-offs and optimal 752 resource allocation during cyanobacterial diurnal growth. Proc. Natl Acad. Sci. U.S.A. 753 114(31): E6457-E6465 (2017). 754 66. L. Yang, A. Ebrahim, C. J. Lloyd, M. A. Saunders, B. O. Palsson, DynamicME: dynamic 755 simulation and refinement of integrated models of metabolism and protein expression. 756 BMC Syst. Biol. 13, 1-15 (2019). 757 67. Y. Chen, J. Nielsen, Mathematical modeling of proteome constraints within metabolism. 758 Curr. Opin. Syst. Biol. 25, 50-56 (2021). 759 68. A. N. David, Y. Sewsynker-Sukai, E. G. Kana, Co-valorization of corn cobs and dairy 760 wastewater for simultaneous saccharification and lactic acid production: Process 761 optimization and kinetic assessment. Bioresour. Technol. 348, 126815 (2022). 762 69. M. Kanehisa, S. Goto, KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids 763 Res. 28, 27-30 (2000). 764 70. P. D. Karp, M. Riley, S. M. Paley, A. Pellegrini-Toole, The metacyc database. Nucleic 765 Acids Res. 30, 59-61 (2002). 766 71. H. Wang et al., RAVEN 2.0: A versatile toolbox for metabolic network reconstruction and 767 a case study on Streptomyces coelicolor. PLoS Comput. Biol. 14, e1006541 (2018). 768 72. C.S. Henry et al., High-throughput generation, optimization and analysis of genome-scale 769 metabolic models. Nat. Biotechnol. 28, 977-982 (2010). 770 73. R. K. Aziz et al., The RAST Server: rapid annotations using subsystems technology. 771 BMC genomics 9, 1-15 (2008). 772 74. L. Jeske, S. Placzek, I. Schomburg, A. Chang, D. Schomburg, BRENDA in 2019: a 773 European ELIXIR core data resource. Nucleic Acids Res. 47, D542-D549 (2019). 774 75. F. Li, Y. Chen, M. Anton, J. Nielsen, GotEnzymes: an extensive database of enzyme 775 parameter predictions. Nucleic Acids Res. 51, D583-D586 (2023). 776 76. F. Li et al., Deep learning-based kcat prediction enables improved enzyme-constrained 777 model reconstruction. Nat. Catal. 5, 662-672 (2022). 778 77. R. Apweiler et al., UniProt: the universal protein knowledgebase. Nucleic Acids Res. 32, 779 D115-D119 (2004). 780 78. N. E. Lewis et al., Omic data from evolved E. coli are consistent with computed optimal 781 growth from genome-scale models. Mol. Syst. Biol. 6, 390 (2010). 782 79. L. Heirendt et al., Creation and analysis of biochemical constraint-based models using 783 the COBRA Toolbox v. 3.0. Nat. Protoc. 14, 639-702 (2019). 784 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint 23 80. G. Wang, J. Chu, Y. Zhuang, W. van Gulik, H. Noorman, A dynamic model-based 785 preparation of uniformly-13C-labeled internal standards facilitates quantitative 786 metabolomics analysis of Penicillium chrysogenum. J. Biotechnol. 299, 21-31 (2019). 787 81. X. Liu et al., Dynamic changes of metabolomics and expression of candicidin 788 biosynthesis gene cluster caused by the presence of a pleiotropic regulator AdpA in 789 Streptomyces ZYJ-6. Bioprocess. Biosyst. Eng. 42, 1353-1365 (2019). 790 82. J. Liang, R. van Kranenburg, A. Bolhuis, D. J. Leak, Removing carbon catabolite 791 repression in Parageobacillus thermoglucosidasius DSM 2542. Front. Microbiol. 13, 792 985465 (2022). 793 83. A. Veyrat, V. Monedero, G. Perez-Martinez, Glucose transport by the 794 phosphoenolpyruvate: mannose phosphotransferase system in Lactobacillus casei ATCC 795 393 and its role in carbon catabolite repression. Microbiology 140, 1141-1149 (1994). 796 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 25, 2024. ; https://doi.org/10.1101/2024.01.23.576957doi: bioRxiv preprint