Dynamic substrate preferences and predicted metabolic properties of a simple microbial consortium

Microorganisms are typically found as complex microbial communities that altogether govern global biogeochemical cycles. Microbes have developed highly regulated metabolic capabilities to efficiently use available substrates including preferential substrate usage that can result in diauxic shifts. This and other metabolic behaviors have been discovered in studies of microbes in monoculture when grown on low-complexity (e.g. two-component) mixtures of substrates, however, little is known about how species partition environmental substrates through substrate competition in more complex substrate mixtures. Here we use exometabolomic profiling to examine the time-varying substrate depletion from a mixture of 19 amino acids and glucose by two Pseudomonads and one Bacillus species isolated from ground water. We examine if the first substrates depleted result in maximal growth rate, or relate to growth medium or biomass composition and find surprisingly few correlations. Patterns of substrate depletion are modeled, and these models are used to examine if substrate usage preferences and substrate depletion kinetics of three microbial isolates can be used to predict the metabolism of the pooled isolates in co-culture. We find that most of the substrates fit the model predictions, indicating that the microbes are not altering their behaviors for these substrates in the presence of competitors. Glucose and histidine were depleted more slowly than predicted, while proline, glycine, glutamate, lysine, and arginine were all consumed significantly faster; these compounds highlight substrates that could be involved in species-species interactions within the consortium. Author Contributions OE, TRN conceived and designed the experiments OE, BPB, SMK, SJ, RL performed the experiments OE, BPB, SJ, TRN analyzed the data OE, TRN wrote the manuscript TRN contributed materials and analysis tools

individuals within a mixed community [13][14][15] . We have recently found exometabolite 78 niche partitioning in two soil environments where sympatric microbes were found 79 to target largely non-overlapping portions of the available substrates, thus 80 minimizing substrate competition 14 . These experiments were focused on the 81 endpoint depletion of substrates by isolates, not the temporal sequence of 82 utilization. However, the order of substrate utilization (i.e. substrate preferences) 83 may further discriminate the adaptive strategies of individual organisms for 84 common substrates.

86
While some work on mixed-substrate growth has been performed in continuous 87 culture at steady state 16 , understanding substrate usage and competition in 88 batch cultures may have both ecological and practical applications. Many 89 environmental processes happen with pulsed inputs: for example the release of 90 substrates into the soil following rainfall, light-dark cycles, digestion in animals, 91 etc. Additionally, some biotechnologies that use microorganisms are also batch 92 processes, such as the large-scale fermentations of microbe-processed foods 93 (e.g. cheese, wine, etc.). Most of these processes use mixed microbial cultures, 94 including one-pot processes of biomass conversion to biofuels and other 95 biosynthetic products [17][18][19] . Studying the temporal substrate utilization by 96 individuals is an important first step in developing approaches to better model 97 these biochemical processes.

99
As recently shown in the pioneering work by Behrends et al., the kinetics of 100 substrate depletion from a mixture of substrates can be effectively fit using a few 101 parameters 20 : see Equation (1) in Materials and Methods. When compared 102 across all substrates in an environment, these parameters have great potential in 103 providing a direct measure of an organism's substrate preferences within that 104 environment, effectively creating a metabolic model for the organism. Such 105 models may be useful in classifying microorganisms for in-depth characterization 106 of their metabolism and regulatory networks to understand the biochemical or 107 evolutionary basis for these preferences. Furthermore, when taken into 108 consideration with other species' models, they may also enable the prediction of 109 the overall net metabolism of microbial consortia by aggregating individual 110 contributions to environmental substrate usage. Observed deviations from these 111 predictions could help identify interspecies interactions that modulate an 112 organism's metabolism, e.g. communication and antagonism between microbes 113 within communities.

115
Here we compare the temporal depletion of 20 substrates by 3 isolates and fit 116 these data to the Behrends model (Equation The absolute concentrations of the 20 growth substrates were quantified at each 148 time point, and the data were fit to a previously described model for compound 149 depletion during microbial batch culture 20 (Figure 1, Algorithm 1). We observed 150 that all compounds followed the Behrends model over the course of growth for 151 each species, with the exception of two compounds: glycine increased over the 152 first 5 hours of culture from all three species and then decreased logarithmically, 153 and the methionine depletion profile for Bc was indeterminable due to both 154 variance in the data and a lack of time points from 12 to 24 hours (Supplemental 155 File 1). These observations corroborate previous assertions that substrate 156 utilization by microbes in batch culture follow the shape of a logistic growth type 157 curve 20-22 .

159
To examine the sequence of substrate deletion in finer detail, we used the model 160 to calculate the time at which each species depleted half of the total amount of 161 each compound (T h ), and when the compound was depleted from 90% to 10% of 162 the total amount available to the species (usage window) (Figure 1), and 163 mapped them onto the growth curve of each species (Figure 2A-C). For Bc, we 164 observed that compounds were half-depleted in three distinct groups ( Figures  165  2A and D growth, and finally half-depleted remaining 6 substrates in late exponential and 168 stationary phases. Neither of the pseudomonads appeared to utilize substrates in 169 these types of groups, but instead had a more even distribution throughout their 170 growth curve (Figures 2B-D). However, the growth curve of Pb did show multiple 171 growth phases (Figure 2C), and so compounds can be mapped to the growth 172 phase in which they are half-depleted ( Figure 2D). This observation is more in 173 line with the traditional view of catabolite repression and multi-auxic growth, 174 where a lag phase will be observed each time the organism reorganizes its 175 metabolism to utilize different substrates 23 .

177
It is surprising that for these three species we observed three different 178 combinations of growth curve and substrate utilization profile: a temporally 179 distinct grouping of compound utilization with only one observed growth phase 180 (Figure 2A), an even distribution of substrate utilization with only one growth 181 phase ( Figure 2B) and an even distribution over multiple growth phases ( Figure  182 2C). This is quite significant given that two of the species belong to the same 183 genus (Pl and Pb). This suggests that the metabolic regulatory systems between 184 the two species are different: while Pb slows down its growth, presumably 185 because it is undergoing a large-scale "switch" of metabolic systems, Pl does 186 not, which may indicate that either all its metabolic systems are constitutively 187 active, or the regulation of the systems is so perfectly timed that the organism 188 can seamlessly switch from one metabolic regime to another. Bc may also have 189 an efficient metabolic regulatory system, as even though we observe distinct 190 temporal gaps between groups of compounds, we did not observe multiple 191 growth phases.

193
To compare the differences in substrate depletion between species, we 194 compared T h across the three species ( Figure 2D and Supplemental Table 1).

195
Across all three species, glutamine, glutamate, alanine, arginine, proline, and 196 asparagine, were half-depleted within one hour of each other. Additionally, the T h 197 values across all substrates for the two Pseudomonas species were close, but 198 not identical, consistent with their short phylogenetic distance but different 199 species identity ( Figure 2D); a similar observation has been described previously 200 22 . Considering the differences in growth curves between the two species, this is 201 quite intriguing, as the general order in which the species consume the 202 metabolites is not different, but there is this difference in growth profiles, 203 supporting the hypothesis that there could be significant physiological differences 204 between such closely related species.

206
Bc was markedly different from the two pseudomonads, differing greatly in the 207 amount of time it depleted 8 of the compounds ( Figure 2D and Supplemental 208 Table 1). Of these, the utilization of glucose was particularly interesting, as it was 209 predominantly depleted before there was any appreciable increase in biomass 210 (Figure 2A). This may indicate that there is a significant delay in substrate 211 conversion to biomass in this species, or that Bc rapidly transforms glucose into 212 some other compound, for example glycogen.

214
We next wondered if the preferred substrates offer some physiological benefit 215 over less preferable substrates. It is a general assumption in microbiology that 216 substrates consumed first may be more advantageous than those consumed 217 later 24 , and that this would depend on the competitive 'strategy' of the organism. 218 Major strategies suggested include maximal biomass production rate, maximal 219 growth rate and maximal biomass yield. Generally, copiotrophs are thought of as 220 r-strategists (maximal growth rate) and oligotrophs as K-strategists (maximum 221 yield) 25,26 . Given the relatively fast growth rates and high substrate 222 concentrations in this study we would expect that the order of substrate 223 consumption would be related to maximal growth rate or biomass production rate 224 27 .

226
We tested some of these general assumptions by comparing the calculated T h 227 values and maximum usage rate of each compound to the specific growth rate, 228 starting molarity of the compound, and predicted total protein composition of 229 each species, in order to determine what the substrate preference order might be 230 correlated with (Figure 3 and Supplemental Figure 3). The specific growth rate 231 of a species on a compound was determined by growing the species on that 232 compound as a sole carbon source (see Materials and Methods). Surprisingly, 233 the only significant (p < 0.05) correlations between all of these tests were that the 234 specific growth rate of Pl on a given compound was weakly correlated with the T h 235 of that compound (r = -0.652, p = 0.030), and moderately correlated with the 236 maximum depletion rate of that compound by Pl (r = 0.791, p = 0.004) ( Figures  237  3C,D). These correlations support the common assumptions listed above, 238 especially for flux balance analysis, as the compound that provides the higher 239 rate of growth is depleted earlier and more rapidly than others. It is interesting 240 that glucose did not confer the fastest specific growth rate for any of the strains, 241 despite glucose generally being considered a superior source of energy. This is 242 not surprising, however, as it is known that pseudomonads preferentially use 243 amino acids over glucose 28 . The rationalization of this phenotype is that in the 244 soil environments where many pseudomonads (and B. cereus) live, 245 decomposition products such as amino acids and organic acids are more readily 246 available than sugars 28 . However, the lack of any strong or significant 247 correlations in the bacillus and the other pseudomonad indicates that there are 248 other factors at play that determine an organism's preferred substrate usage. It is 249 apparent that not all microbes prefer to use substrates sequentially at all; the 250 grouping of substrate utilization by Bc is a striking example of this. The resources 251 within the second utilization group (Figure 2A) conferred a wide range of specific 252 growth rates, from zero to the highest observed for all substrates, and all were 253 utilized within two hours of each other ( Figure 3A). It is likely the case that the 254 simultaneous usage of these substrates confers the greatest physiological 255 advantage. Bc could possess a metabolic strategy that does not perfectly follow 256 the well-established paradigm of catabolite repression. Ultimately, it is clear that 257 bacteria dramatically differ in regulation of catabolite uptake, and it is not prudent 258 to make general assumptions on microbial metabolism based solely on 259 observations from a few model organisms and/or the energetic potential of 260 substrates.

262
Our experiments to test these correlations yielded a number of interesting results 263 in addition to those described above. First, all three species grew on glucose as 264 the sole carbon source without added amino acids. This was not predicted based 265 on genomic predictions of the species in the Integrated Microbial Genomes (IMG) 266 database (img.jgi.doe.gov), which indicated auxotrophy for lysine, phenylalanine, 267 tyrosine, histidine and serine in the case of Bc, and for lysine, histidine, leucine, 268 and coenzyme A for Pl and Pb. This observation highlights that all computational 269 predictions should be treated as only suggestions, and should always be tested 270 experimentally before making any assertions. Additionally, there were a number 271 of compounds that did not support growth as sole carbon sources, but were 272 depleted throughout the growth of the species in our complete defined medium 273 (Figure 3, lightly shaded compounds). This finding indicates that caution should 274 be employed when making physiological assertions based on single-substrate 275 studies, for example those that have individual substrates arrayed in multi-well 276 plates; many microbes can only utilize certain compounds when other substrates 277 are present, the phenomenon of co-metabolism 29 . We should note, however, 278 that we do not know the details of how these compounds are depleted in the rich 279 defined medium, only that they are depleted from the medium; they may simply 280 be exogenously transformed. Finally, we observed the maximum depletion rate 281 of all the substrates by the three species to be less than 130 µg/mL/hour except 282 for glutamate depletion by Bc, which we calculated to be about 640 µg/mL/hour 283 (Supplemental Table 1). This rate corresponds to a near instantaneous 284 depletion of glutamate by Bc at about 5 hours into the growth curve (see 285 Supplemental File 1), which is towards the end of the second group of 286 compounds utilized by this species (Figure 2A). Why glutamate would be 287 depleted so much faster than the other compounds for Bc is a mystery, but it 288 does suggest that there is something unique about the compound that requires 289 or allows for the flux to be so rapid. Interestingly, in a previous study of 290 metabolite depletion of a mixture of 470 compounds glutamate was one of two 291 metabolites depleted by all of the isolates 14 , so it is clearly an important or high-292 value compound that Bc may have evolved to deplete quickly in order to gain a 293 competitive advantage. 294 295 296 Predicting consortium metabolism based on models of individual isolates 297 298 Having modeled the substrate usage of each species for each compound, we 299 hypothesized that these models could be combined to predict how a consortium 300 composed of the three species might utilize the substrates. We simulated the 301 time-dependent depletion of each compound by a consortium composed of the 302 bacillus and two pseudomonads (see Materials and Methods, Equation 2, and 303 Algorithm 2). Briefly, the functions describing the compound usage by each 304 species were summed (Supplemental Figure 2A), and the time at which this 305 summed use curve reached the total available compound was determined. This 306 time of depletion was then used to predict how much of a given metabolite each 307 species would have utilized when grown in co-culture, and the compound usage 308 by each species was re-modeled (Supplemental Figure 2B colored dashed 309 lines) and added together to form the co-culture prediction (Supplemental 310 Figure 2B solid black line). These predictive models allowed us to make several 311 hypotheses that are relatively simple to test. First is the usage curve of each 312 metabolite by the co-culture. Related to this, we can predict the time at which all 313 of a given metabolite will be depleted, and when all metabolites will be depleted. 314 From this we predict that 14 compounds will be nearly depleted (less than 10% of 315 starting concentration) by six hours, and all but methionine will be completely 316 consumed by 9 hours (Figure 4). Based on this, one could reasonably argue that 317 a consortium composed of these three species would reach stationary phase 318 sometime between 6 and 9 hours, in contrast to the individual species, which all 319 reached stationary phase after 9 hours. 320 321 To test our predictions, we inoculated a 3-member co-culture at equal optical 322 density in the defined medium (see Materials and Methods), collected 323 supernatant time points every hour, and measured the concentrations of all 20 324 substrates as described for monocultures. We found that many of our predictions 325 were valid: nearly all compounds (17) were depleted to below 10% of starting 326 concentration by 6 hours (Figure 4, gold), and the co-culture accordingly 327 reached stationary phase at this time as well (Supplemental Figure 4), 328 presumably because all available substrates were consumed.

330
Compounds that follow the model are evenly shared 331 332 When analyzing the kinetics of depletion of the compounds, we observed that 333 many (13) compounds agreed very well with the prediction, having R 2 values 334 greater than 0.9 (Figure 4). Most of the compounds with high R 2 values began 335 to decrease slightly earlier or at a slightly faster rate than predicted, which could 336 be attributed to experimental error in initial culture density. However, the 337 depletion of most compounds were still very close to the predicted model, 338 indicating that the shared usage between the species could be very close to proline, and glycine) deviated significantly from our predictions (R 2 < 0.9) ( Figure  355 4, red text), suggesting some additional species-species interaction(s) is/are 356 present that affect the depletion of those compounds. It is intriguing that we 357 detected metabolites that showed both positive and negative deviations.

359
Glucose and histidine were both depleted more slowly than predicted. The 360 simplest explanation for this is that the metabolic systems that deplete these 361 compounds are indeed concentration dependent. Another possibility for this 362 would be that there is a buildup of product in the co-culture that exerts feedback 363 inhibition on the metabolism of these two compounds. This is easily rationalized 364 for histidine utilization, which is an expensive process for bacteria 33 ; they may be 365 exposed to better carbon sources in a mixed culture as a byproduct of another 366 microbe. However, glucose being utilized slower is curious. In the monoculture 367 experiments, we observed Bc to deplete glucose before it or either pseudomonad 368 even started producing appreciable biomass (Figure 2). Perhaps this behavior is 369 inhibited in the presence of the pseudomonads or is a result of changes in the 370 community structure over the experiment, the assessment of which are 371 unfortunately beyond the scope of the current study.

373
In contrast, glycine, proline, lysine, arginine, and glutamate were all depleted 374 faster than predicted. This is more difficult to explain and suggests at least one 375 microbe has altered its phenotype due to the presence of other microbes, or that 376 other exometabolites are influencing consortial behavior. For example, one 377 species may have up-regulated metabolic pathways involving these compounds 378 in an effort to outcompete others, either for the purpose of direct competition for 379 the substrate, or in order to synthesize antibiotic compounds 34 . Alternatively, 380 another member may have otherwise sequestered those compounds, effectively 381 taking them out of a common pool, for example by converting the compound into 382 some storage molecule, or sequestering it in a way similar to how siderophores 383 sequester iron. Testing these hypotheses would require an extensive untargeted 384 metabolomics study, an extremely interesting direction for future studies. Another 385 potential reason for this early depletion is that the co-culturing of these microbes 386 has resulted in an emergent function of increased flux of the substrate(s) through 387 the system. This could be due to a cross-feeding effect where one microbe 388 depletes an inhibitory compound of another microbe or one microbe's products 389 induce the co-metabolism of that product and one of these substrates.  all growth experiments, the water used to prepare the medium and uninoculated 450 medium were incubated alongside the experimental flasks, as controls.

452
Growth assays of species on individual carbon sources were performed in 96-453 well Falcon tissue culture plates with flat bottom and low evaporation lid, in a total 454 volume of 200 µL. The medium consisted of the same concentrations of Wolfe's 455 vitamins and minerals, ammonium chloride and potassium phosphate. Individual 456 carbon sources were added at a concentration of 0.5 mg/mL. Species were pre-457 cultured and washed as before, and wells were inoculated at an OD 600 of 0.05.

458
The plates were incubated at 30 °C, shaking at "medium" speed in BioTek 459 Synergy HT and Tecan Infinite F200 Pro plate readers, for 48 h.

461
Metabolomics sample extraction 462 463 Hourly time points of 1 mL of cell culture and controls (see above) were aspirated 464 and centrifuged at 5,000 xg to pellet the cells. 800 µL was aspirated from the top, 465 taking care not to disturb the cell pellet, and split into two 400 µL aliquots, which 466 were immediately frozen at -80 °C. A calibration curve was created with the 467 medium used for culturing: 1x culture medium, 1/2x, 1/10x, 1/100x, 1/1000x, and 468 1/10000x dilutions were prepared using culture medium without any carbon 469 sources as the diluent. All experimental, control, and calibration curve samples 470 were lyophilized overnight, and metabolites were extracted in 300 µL methanol 471 with 25µM 13 C-phenylalanine for use as an internal standard. The Anaconda package and IPython notebooks were used for all computational 494 tasks 42 , which will be made publicly available at https://github.com/biorack in the 495 "Predicting metabolic properties of a microbial co-culture" repository upon 496 manuscript publication by a peer-reviewed journal. Data were stored and 497 organized using Pandas 43 and NumPy 44 , and graphs created using Matplotlib 45 . 498 Metabolite depletion was modeled using leastsq from scipy.optimize 46 , fitting 499 the data to the Behrends model (eq 1): 500 501 Where a is amplitude and o is offset (see Figure 1). These two parameters were 503 defined from the data: amplitude was defined to be the average of the t=0 data 504 point and the maximum value data point in the data set of each compound, and 505 offset was defined as the lowest value in the data set. All other parameters were 506 solved using leastsq, with the criteria that they had to be positive values. The leastsq parameter fitting of !" !" and w ij to data T h and usage window values were calculated from the Behrends model. All 510 correlation coefficients and p-values were calculated using the pearsonr function 511 in the stats package of scipy. 512 513 Co-culture predictions 514 515 The equations representing the depletion of a compound by a species were 516 subtracted from the initial starting concentration of the compound, creating an 517 expression that represented the amount of compound used by each species over 518 time; these are the curves shown in Supplemental Figure 2A. These 519 expressions were summed to generate an approximate total usage curve, and 520 the time at which this curve crossed the total amount of available compound was 521 determined. The amount of available compound was defined to be the starting 522 concentration of a compound minus the lowest offset parameter between the 523 three species, as the species with the lowest offset parameter for a substrate will 524 presumably deplete the substrate to that level, but not more, even in a co-culture. 525 The time of total depletion was used to approximate the amount of compound 526 that each species would have consumed by that time. The individual usage 527 curves were capped at this compound level at this time, and transformed back to 528 compound depletion curves, which were then used to re-fit to the Behrends 529 equation, generating new models of compound depletion in mixed conditions. 530 These new models were then summed, producing the predicted total co-culture 531 usage of each compound. This can be summarized by the general Equation 2: 532 533 Where C is the total amount of substrate j that is available to the mixed culture of 535 set species. This is defined as the starting concentration of j minus the smallest 536 o j in species. !" ′, !" ! , !" !" , and !" ′, are parameters that describe the depletion 537 of j by species i in the co-culture of the individual in the set species, shown in 538 Algorithm 2: 539 540 Algorithm 2. Predicting co-culture substrate usage 1 for j in substrates: for i in species: or when half of the total amount of compound has been depleted, and the red bar 691 depicts the calculated usage window, or time when the compound is depleted from 692 90% to 10% of the total amount used by the species. 693