Selection for cell yield does not reduce overflow metabolism in E. coli

Overflow metabolism is ubiquitous in nature, and it is often considered inefficient because it leads to a relatively low biomass yield per consumed carbon. This metabolic strategy has been described as advantageous because it supports high growth rates during nutrient competition. Here we experimentally evolved bacteria without nutrient competition by repeatedly growing and mixing millions of parallel batch cultures of E. coli. Each culture originated from a water-in-oil emulsion droplet seeded with a single cell. Unexpectedly we found that overflow metabolism (acetate production) did not change. Instead the numerical cell yield during the consumption of the accumulated acetate increased as a consequence of a reduction in cell size. Our experiments and a mathematical model show that fast growth and overflow metabolism followed by the consumption of the overflow metabolite, leads to a higher numerical cell yield and therefore a higher fitness compared to full respiration of the substrate. This provides an evolutionary scenario where overflow metabolism can be favourable even in the absence of nutrient competition.


Introduction 31
When microbes compete, fast-growing strategies typically become dominant. While numerous 32 microorganisms have the ability to respire during fast growth on high-quality carbon sources like 33 glucose, many bacteria, yeasts and mammalian cells are known to not make full use of this 34 capability, even when sufficient oxygen is available 1-4 . Instead of efficiently catabolizing the 35 available carbon source to CO2, they produce metabolic by-products like acetate, lactate or ethanol 36 at the expense of metabolic efficiency, i.e. biomass yield (gram biomass/mole carbon source). This 37 phenomenon is termed overflow metabolism ( Figure 1A) and has been suggested to be 38 evolutionarily favourable at nutrient excess, because it supports the highest growth rates. It allows 39 microorganisms that employ this strategy to outcompete fully respiratory cells 5-7 . When subjected 40 to low substrate concentrations, cells grow at a lower rate and display full respiratory behaviour, 41 e.g. in chemostats, which is associated with an increase in metabolic efficiency 1,2,6,8 . 42

43
In nature many microbial species are in spatially structured environments where resources can be 44 localized and the microbes undergo cycles of feast and famine. Their metabolic strategies have 45 adapted to those conditions. Characteristic of such conditions are periods of nutrient excess 46 followed by starvation, similar to batch cultures that are continued until complete depletion of the 47 carbon sources. The winning strategy generates the highest number of viable offspring after a 48 single cycle of feast and famine. Thus, microorganisms are faced with an "optimization problem": 49 the maximal number of viable offspring has to be produced in the shortest possible time given the 50 available resources. In this context, metabolic inefficiency of overflow metabolism is generally 51 considered to be the price for being fast. 52 53 As (growth) rate selection is associated with the occurrence of overflow metabolism, one would 54 assume that yield selection should result in an increased metabolic efficiency, typically via 55 enhanced respiratory metabolism. One way to ensure that growth rate competition between 56 microorganisms is prevented, and selection of yield becomes possible, is by culturing individual 57 cells in droplets of water-in-oil emulsions ( Figure 1B). In this regime each cell gets its own 58 "privatized" amount of substrate within a medium droplet, and does not compete for it with other 59 genotypes, ruling out rate selection. The cells in a droplet can be allowed to grow until depletion 60 of all carbon sources. When L. lactis was subjected to this protocol, the mutant cells that fixed had 4 shifted their metabolism from inefficient homolactic to the more efficient mixed-acid 62 fermentation. These mutants had a higher biomass yield and numerical cell yield at a reduced 63 growth rate 6 . This was the first experimental illustration of (cell) yield selection. 64 We exploited this emulsion protocol to evolve E. coli MG1655. Unexpectedly we found no 66 reduction of overflow metabolism after experimental evolution, but rather a decrease in cell size, 67 specifically during growth on the overflow metabolite acetate. This led to an increased numerical 68 cell yield (number of cells/mole glucose). Our experiments and a mathematical model show that 69 in a feast-famine regime fast growth and overflow metabolism followed by the consumption of the 70 overflow metabolite can lead to a higher fitness compared to full respiration of the substrate. 71 72

Results 73
Selection for cell yield does not decrease overflow metabolism 74 The serial propagation of individual microbial cells in emulsion droplets resembles millions of 75 parallel batch cultivations, each inoculated by a single cell. In each droplet, cells can grow for a 76 limited number of generations (in our case 5 to 6, set by the average droplet size and the available 77 carbon source) before they reach stationary phase. After such a growth period, all droplets are 78 merged, the cells are mixed and diluted, and used to inoculate new droplets for the next round of 79 growth. This protocol (figure 1B) selects for increased numerical cell yield per mole glucose. We 80 wondered if this selection protocol would lead to E. coli strains with an increased metabolic 81 efficiency (i.e. yield of gram biomass/mole glucose) and reduced overflow metabolism (more 82 respiratory). To investigate this, we propagated E. coli MG1655 for 53 transfers in emulsions and 83 determined the growth characteristics of 90 isolates (Supplementary Figure S1). One of these 84 isolates, designated IR1, was characterized in more detail as it showed an increased maximal 85 optical density compared to the wild type. 86

101
Besides the increase in four biomass measures (dry weight, total cell volume, total protein content, 102 optical density) of 9-13%, strain IR1 also showed an increased growth rate on glucose (39% 103 higher) (Fig. 1C-F). Genome sequencing (Supplementary Table S3) revealed an 82bp deletion in 104 the rph/pyrE region that was previously characterized 2,9 . This mutation alleviates a documented 105 pyrimidine production deficiency of the ancestral strain 10-12 that limits growth on minimal 106 medium, but not on rich medium. Growth experiments confirmed that IR1 also displays this 107 phenotype (Supplementary Figure S3). We also found that during the glucose-growth phase, IR1 108 shows a 98% decrease in the production of pyrimidine intermediates orotate and dihydroorotate 109 (Supplementary Figure S2). Together this data indicated that the increase in yield and rate is 110 partially due to mutations that relieve the pyrimidine deficiency of the ancestral strain. 111 Mutants that showed a shift from fermentation towards respiration were not found. The amount of 112 acetate produced per mole of consumed glucose was not altered significantly in the evolved mutant 113 IR1 ( Figure 1G). In an additional attempt to isolate mutants with altered overflow metabolism 114 activity, we propagated strain IR1 for another 25 transfers in emulsion droplets. The subsequent 115 characterization of 15 evolved strains showed that all strains still produced acetate (see 116   Supplementary Table S2). 117 Another phenotypic change of the isolated mutant IR, was a decrease in cell volume of 11% 118 compared to the ancestral strain. This cell size decrease was especially obvious during the second 119 growth phase on acetate, and it corresponds to an increase in cell number of 23% ( Figure 1C) when 120 the glucose is depleted. Re-sequencing revealed a mutation that leads to a stop codon in the ygeR 121 gene (Supplementary Figure S2). This gene is involved in septum formation and cell division, and 122 deletion of it has been shown to reduce cell length 13 . The fact that we isolated mutant strains with 123 reduced cell size and that we were not able to identify strains with decreased overflow metabolism 124 led to an alternative hypothesis: In a feast-famine environment where the evolutionary pressure is 7 on the overall numerical cell yield, overflow metabolism is actually more efficient than complete 126 respiration when the consumption of the overflow product acetate is taken into the equation. 127

128
Biphasic substrate utilization optimizes fitness in feast-famine environments 129 During growth in batch culture until complete exhaustion of all the carbon sources, microbes are 130 subjected to continuously changing conditions. This resembles a feast and famine cycle with two 131 feast phases (a first phase on glucose and a second phase on acetate), followed by carbon source 132 exhaustion/ famine phase. If the carbon source is a fast fermentable sugar such as glucose then the 133 first phase will be characterized by a high growth rate and a relatively large cell size 14 . As long as 134 the sugar concentration is high, maximizing the growth rate leads to the highest number of 135 offspring produced per unit time. At this stage it is not relevant if the cell is metabolically 136 inefficient, as there is sufficient substrate available. However, when the initial "fast" substrate 137 becomes limiting, cells will switch to growth on the produced overflow metabolite, acetate in the 138 case of E. coli. Growth on an overflow metabolite is always slower than on the initial fast substrate 139 and slow growth is correlated with smaller cell sizes 14-18 . 140 We show that already during the onset of the second growth phase, the cells started to reduce in 141 size to eventually reach a cell volume in stationary phase that is less than half of what it was during 142 exponential growth on glucose ( Figure 1E). During this cell size adaptation period, making new 143 offspring cells costs less nutrients than during steady-state growth on acetate, because a mother 144 cell is now larger than a daughter cell and likely does not need to double in size before division. 145 Immediately after glucose depletion, cells therefore need to produce proportionally less biomass 146 to divide into two daughter cells. This leads to a proportionally higher increase in cell number 147 compared to the increase in biomass during the transition from glucose to acetate ( Figure 1C Therefore, for a fair comparison, the additional biomass yield on the overflow metabolites needs 194 to be considered as well. In this case it is not immediately clear whether fully respiratory growth 195 is a better evolutionary strategy than overflow metabolism and biphasic growth. 196 197 Another important point is that natural selection acts on the number of viable offspring rather than 198 the amount of produced biomass. Interestingly, many bacteria become larger when they grow 199 faster; also E. coli is known to have a positive correlation between growth rate and cell size 15-18 . 200 This occurs when the generation time is shorter than the DNA replication time, necessitating 201 mother cells to start DNA replication for their future progeny 22-24 . During evolution on a finite 202 nutrient supply however, it pays off to make a higher number of cells per unit nutrient, and this 203 can be achieved by making smaller cells. A fast growth rate that supports an evolutionary 204 advantage under nutrient excess, therefore has a trade-off with a lower number of offspring per 205 unit of substrate. This trade-off is driven by cell size and independent from metabolic efficiency. outgrown by faster growing cells during nutrient competition. We circumvented this trade-off by 208 serially propagating cells in emulsion droplets where single cells are allowed to grow until nutrient 209 exhaustion in physically separated environments. In this regime mutants that produce more 210 offspring will seed more droplets in the next round and increase in frequency. 211 Against initial expectations, full respiration was not found to be a competitive strategy when rate 212 selection was excluded. While selection in emulsion led to increased metabolic efficiency, 213 biomass-and cell-yield in L. lactis 6 , we found that in E. coli the utilization of overflow metabolism 214 was unaffected. However, we did select for a mutant with a smaller cell size, which is possibly 215 linked to a single base deletion in ygeR, a gene known to affect the cell size 13 . We identified two 216 reasons for the difference in the outcome between the L. lactis study and our experiments i) E. coli 217 is able to consume the overflow metabolite which L. lactis cannot, and ii) when growing on the 218 overflow metabolite, the cell number of E. coli increases disproportionately compared to the 219 biomass increase. 220 Our results show that overflow metabolite production and its subsequent utilization maximizes the 221 number of produced offspring irrespective of the available substrate concentration. At high 222 substrate concentrations the growth rate will be maximized while at limiting substrate 223 concentrations a minimization of cell size (maximization of cell number) will occur after the initial 224 "fast" substrate is depleted. This allows the optimization of fitness in dynamic environments. 225 Maximizing the production rate of offspring through sequential substrate utilization seems a very 226 elegant solution that might apply to numerous organisms. There are numerous computational approaches that predict the shift from respiration towards 237 overflow metabolism, based on biochemical and biophysical constraints. These studies include for example flux balance analysis (FBA), where the optimal metabolic flux distribution is predicted 239 that supports high growth rates; trade-offs due to membrane crowding; and optimization of 240 proteome allocation to minimize the investment associated with the costly respiratory machinery 32-241 38 . However, besides some studies in yeast where the main argument is that overflow metabolites 242 lead to a fitness advantage due to their toxicity for competing organisms (Make-Accumulate-243 Consume hypothesis) 27,30,39 the role of evolutionary perspectives are less investigated. The 244 discussion of overflow metabolism in literature and textbooks typically neglects the fact that the 245 secreted overflow metabolite is often further metabolized. Our results argue for the consideration 246 of the consumption of the overflow metabolite as it likely played a role during evolution in natural 247 environments. This is corroborated by a recent study on the regulation of overflow metabolism 40 . 248 Full respiration and reaching the maximal biomass yield are only seen at low substrate 249 concentrations. This might be a rather artificial situation which is created for instance in 250 chemostats. In nature cells are often exposed to dynamic feast-famine cycles, and spatial structure 251 that leads to variations in the selection pressure. 252 If rate selection would be the only force to favour overflow metabolism one might expect higher 253 acetate fractions to be produced by the wild type strain. The fact that in most cases only a small 254 fraction of the flux is directed towards acetate argues for other (additional) selective forces, such 255 as maximizing cell number. In studies by LaCroix et al. 2 and Long at al. 38 rate selection was applied 256 to the same E. coli MG1655 wild type strain. They found a positive correlation between the acetate 257 production and growth rates in the wild type and evolved strains. Such an increase in overflow 258 metabolism at increasing growth rates has been reported in a number of other studies as well 3,7,41,42 . 259 This might indicate that the natural environment from which the wild type strain was isolated was 260 not purely selective for growth rate. 261 Another aspect of natural environments is the possibility of multiple species or strains coexisting. 262 A 'cheater' strain that consumes the acetate before the producer strain can benefit from it would 263 potentially argue against a beneficial effect of overflow metabolism beyond growth rate 264 maximization 43 . However, growth on acetate is typically slow, so even if such a strain would be 265 present at high frequency in the population, its consumption rate would still be significantly lower 266 than what is needed to fully deplete this substrate before the fast growing/acetate producing 267 biomass starts consuming it. In order to pose significant competition, the acetate consumer would 268 therefore need to grow at a similar rate as the acetate producer. Almost all growth rates we found 12 for diverse organisms on acetate were substantially lower than this (see Supplementary Table S5  270 for a number of examples of glucose and acetate growth rate ratios), except for an Acinetobacter 271 strain that displays an exceptionally high growth rate on acetate of 0.91 h -1 44 . 272 In conclusion, this study shows an unexplored consequence of overflow metabolism. It is not only 273 a strategy for fast growth at excess substrate conditions, but it also maximizes the number of emulsion approximately 1 in 10 droplets (~50 µm in diameter) will be inoculated with a single cell 294 (following a Poisson distribution). After overnight incubation at 37°C, the emulsions were broken 295 using perfluoroctanol (PFO)(Alfa Aesar) and the cultures were diluted in fresh medium to repeat 296 the process 6 . 297 To ensure aerobic growth was supported, the oil was saturated with pressurized atmospheric air 298 prior to use. Sufficient O2 supply was confirmed by measuring growth on a non-fermentable 299 carbon source (data not shown). 300 301 Growth curve measurements 302 Strains of interest were precultured and propagated to a 96-wells plate. The spaces between wells 303 were filled with 0.9% saline solution, and the plate sealed with parafilm to reduce evaporation 304 from the wells. Growth was measured every 5 minutes, with shaking in between, overnight in a 305 SpectraMax® plate-reader, at 600 nm and 37°C. Using the R software environment, growth rates 306 and maximum ODs were determined per strain as described earlier 6 . 307 Alternatively, growth was measured manually. Strains were grown in 250 ml Erlenmeyer flasks, 308 shaking at 220 RPM, at 37°C. Every 30 minutes samples were taken to measure OD600 in cuvettes. 309 After the experimental evolution 90 single colonies were randomly picked from the three parallel 312 cell cultures and transferred to a 96-wells plate containing 200 µl M9 + 2.5mM glucose per well. 313 Six wells were inoculated with the wild type strain. After overnight growth, 20 µl from each well 314 was transferred to a 96-wells plates containing fresh medium for proper growth curve 315 measurements. Glycerol was added to a final concentration of 12% and the plates were frozen at -316 80°C. Growth curves were analysed for growth rates and maximal ODs. Several strains per 317 replicate culture were selected for further analysis, based on the maximal OD compared to the 318 wildtype.                              Research had no role in the study design, data collection, and analysis, decision to publish, or 470 preparation of the manuscript. 471 Supplementary Information Selection for cell yield does not reduce overflow metabolism in E. coli Rabbers et al. 18 An underlying assumption in the hypothesis above is that the cell size after entering stationary phase when 19 growing on acetate is smaller than when entering stationary phase after growing on glucose. We found that 20 when wildtype cells are growing in a glucose limited batch culture where the final divisions are on the 21 overflow metabolite acetate, the cell size in stationary phase is approx. 55% smaller than the cell size during 22 exponential growth on glucose. As strain MG1655 always produces acetate in a batch culture on glucose 23

Supplementary Information 1. Comparison of cell size under different growth conditions
investigating the entering of stationary phase without acetate exposure in batch culture is not possible. To 24 mimic the effects on cell size when going into stationary phase at a faster growth rate than on acetate we 25 prepared nitrogen limited batch cultures where no biphasic growth is observed. The results showed that the 26 final cell size after going into stationary phase with nitrogen limitation is approx. 24% smaller than during 27 exponential growth (see Supplementary Figure 4), which is significantly bigger than cells entering 28 stationary phase after growth on acetate. 29 A second assumption is that the growth rate reduction that comes with full respiration does not lead to a 30 cell size decrease that combined with its effect on increasing biomass yield would lead to a higher cell yield 31 than growth on acetate. We found 5 studies with E. coli where maximum growth rates and growth rates that 32 still allow full respiration are reported (see Supplementary Table 4). All five show that a growth rate 33 reduction to 77% -59% of the maximum growth rate is sufficient for a strain to change metabolism towards 34 full respiration. In our case the balanced growth rate on acetate is 28% and 31% of the maximum growth 35 rate for MG1655 and IR1 respectively. This is therefore well below the growth rate reduction required to 36 allow full respiration suggesting that the growth rate reduction on acetate as a substrate adds to the cell 37 number through making smaller cells.

Supplementary Information 2. Model description
Introduction 41 The aim of this model is to evaluate the fitness of biphasic metabolism, incl. overflow metabolism, and 42 compare it to the fitness of a pure-respiration strategy; to identify which parameters determine the winning 43 metabolic strategy, i.e. the one favoured by evolution when cells are confronted with a finite amount of 44 sugar that they can consume until its depletion. . 83 The number of cells that we can make from this amount of acetate equals, 84