Quantitative Physiology and Proteome Adaptations of Bifidobacterium breve NRBB57 at Near-Zero Growth Rates

ABSTRACT In natural environments, nutrients are usually scarce, causing microorganisms to grow slowly while staying metabolically active. These natural conditions can be simulated using retentostat cultivations. The present study describes the physiological and proteome adaptations of the probiotic Bifidobacterium breve NRBB57 from high (0.4 h−1) to near-zero growth rates. Lactose-limited retentostat cultivations were carried out for 21 days in which the bacterial growth rate progressively reduced to 0.00092 h−1, leading to a 3.4-fold reduction of the maintenance energy requirement. Lactose was mainly converted into acetate, formate, and ethanol at high growth rates, while in the retentostat, lactate production increased. Interestingly, the consumption of several amino acids (serine, aspartic acid, and glutamine/arginine) and glycerol increased over time in the retentostat. Morphological changes and viable but nonculturable cells were also observed in the retentostat. Proteomes were compared for all growth rates, revealing a downregulation of ribosomal proteins at near-zero growth rates and an upregulation of proteins involved in the catabolism of alternative energy sources. Finally, we observed induction of the stringent response and stress defense systems. Retentostat cultivations were proven useful to study the physiology of B. breve, mimicking the nutrient scarcity of its complex habitat, the human gut. IMPORTANCE In natural environments, nutrients are usually scarce, causing microorganisms to grow slowly while staying metabolically active. In this study we used retentostat cultivation to investigate how the probiotic Bifidobacterium breve adapts its physiology and proteome under severe nutrient limitation resulting in near-zero growth rates (<0.001 h−1). We showed that the nutrient limitation induced a multifaceted response including stress defense and stringent response, metabolic shifts, and the activation of novel alternative energy-producing pathways.

I n natural environments, microorganisms are not exposed to nutrient-rich conditions (1). Often, nutrients are not readily available for consumption because of low concentrations and competing microorganisms. This causes a scarcity of nutrients and hinders microbial growth by inducing starvation (2,3). Consequently, microorganisms enter a famine phase where growth becomes slow while they stay metabolically active. This forces bacteria to become energy efficient in order to survive (4,5).
Several systems are used to study microbial physiology in the laboratory. The most common method is batch cultivation. Nevertheless, this model does not well reflect microbial life in natural environments (4,6). In batch cultivations, all nutrients are initially abundant, resulting in fast microbial growth. Nonetheless, in batch cultures, the concentrations of some nutrients may become too low to support further growth. To grow microorganisms under more controlled conditions, chemostat cultivation is used, where the dilution rate determines the growth rate. Thus, fresh medium is continuously fed at a fixed rate into the bioreactor while there is a simultaneous removal of spent medium containing biomass, metabolic products, and unused nutrients (7). However, in chemostat cultivations growth rates below 0.025 h 21 are not possible due to pulse feeding and therefore cannot reach the low levels usually observed in nature, which are close to zero (8).
A better system for the controlled cultivation of microorganisms at very low growth rates is the retentostat (9). The retentostat is a modification of the chemostat in which the biomass is completely retained in the bioreactor using a filter in the effluent line (10,11). In this way, biomass accumulates, reducing the available energy sources per individual cell and resulting in a gradual decrease in growth rate that approaches zero (11). Therefore, unlike the chemostat, retentostat cultivation allows us to emulate the natural environmental conditions of microorganisms (8).
Retentostat studies have shown the importance of this system in elucidating the mechanisms involved in the survival of microbial cells in environments with extremely low nutrient availability (4,10,(12)(13)(14)(15)(16)(17). Under these conditions, microbial cells use different strategies to adapt to the nutritional stress in order to survive. Microbial cells make efficient use of the available energy by shutting down certain processes involved in growth while activating others that would be advantageous for survival (12,18). Hence, maintenance requirements are extremely reduced at near-zero growth rates (13,16,19). Concomitantly, shifts in the metabolism are seen where pathways are activated for the use of alternative energy sources (4). Changes in the morphology, the generation of viable but nonculturable (VBNC) cells, and improved robustness have also been reported as survival strategies at near-zero growth rates (4,11).
For this study, we used an industrially important bacterium, Bifidobacterium breve, which has been studied for its probiotic capacity and its health-promoting traits (20,21). This bacterium has been applied in several food products but most importantly in fermented infant formulas due to its capacity to produce metabolites such as galactooligosaccharides that are beneficial for infants (22,23).
The natural environment of B. breve is one of the most complex ecosystems that exists: the human gastrointestinal tract (24)(25)(26). The human gastrointestinal tract harbors a dense population of microorganisms, which generates a low concentration of nutrients despite its constant inflow. This means that this bacterium often encounters extreme nutrient limitation in the human gut, resulting in slow growth (27). B. breve has never been studied in a continuous fermentation nor in retentostat systems.
The aim of this research was to investigate the physiological response of B. breve NRBB57 at near-zero growth rates using retentostat cultivation with a focus on culturability, metabolism, and maintenance requirements. Proteome analysis was used to link metabolic and physiological changes to changes in the protein abundances in corresponding metabolic pathways and other relevant functions contributing to the performance of B. breve NRBB57 cells under near-zero growth conditions. Implications for B. breve ecophysiology and applications are discussed.

RESULTS
In this study, Bifidobacterium breve NRBB57 was anaerobically cultivated for 3 weeks in three independent retentostat cultivations in a medium containing lactose as the growth-limiting substrate. During cultivation, several parameters were measured, aiming to establish the physiological adaptations of this strain to near-zero growth rates. One of the replicates was stopped after 2 weeks due to clogging of the cross-flow filter and is therefore only included in the proteome analysis.
Culturability, viability, and morphology. Culturability was estimated by plating serial dilutions on transoligosaccharide (TOS)-propionate agar plates and comparing the number of culturable cells with the total number of cells estimated using a counting chamber (chains of cells were counted as 1). CFU/mL decreased over time ( Fig. 2A), while the cell counts increased approximately 10-fold (Fig. 2B).
Likewise, we determined the viability of the culture by assessing the membrane integrity of the cells using the LIVE/DEAD BacLight kit. Bacterial cells were classified into two categories: red propidium iodide (PI)-stained cells conceivably had damaged membranes and therefore were considered not viable, while green SYTO 9-stained cells had intact membranes and were considered viable. The viability decreased from 99% in chemostat mode (t = 0) to 80% at the end of the retentostat culture (Fig. 2C). These observations indicate that most of the bacterial population in this culture stayed viable during the 3 weeks of retentostat cultivation but lost the ability to grow on TOSpropionate agar plates (approximately 70%) (Fig. 2D).
The morphology of B. breve NRBB57 was assessed by scanning electron microscopy (SEM). This showed that the morphology changed in the first and second weeks of retentostat cultivation when cells displayed corrugation (see Fig. S1 in the supplemental material). By the third week, bacterial cells became much longer and showed branching and were counted as 1 as mentioned above.
Metabolism. Concentrations of the products of lactose metabolism (acetate, lactate, ethanol, formate) were measured along with the production of succinate and the consumption of glycerol (Fig. 3). At the start of the retentostat cultivation, lactose was mainly converted into acetate, ethanol, and formate, indicating that B. breve NRBB57 metabolized lactose via the bifid shunt (28). This behavior was also observed in chemostat cultivations of B. breve NRBB57 at higher growth rates up to 0.4 h 21 , while in batch cultures this strain produces a mixture of acetate and lactate (Fig. S2). Throughout the retentostat cultivation, acetate production was stable, while the production of ethanol and formate decreased slightly, and lactate production increased, indicating a small but gradual change in pyruvate dissipation while the growth rate decreases. Furthermore, the concentration of glycerol decreased starting after approximately 1 week of retentostat cultivation, indicating that glycerol was consumed. Interestingly, the concentration of succinate increased over time, which could point to catabolism of amino acids. Therefore, we decided to also assess the concentration of amino acids during the fermentations.
Amino acids. Concentrations of amino acids were measured by ultra-high-performance liquid chromatography (UPLC). During the retentostat cultivation, the concentration of several amino acids decreased (aspartic acid, glutamine/arginine, serine) (Fig. 4). While the succinate concentration gradually increased 2 mM from time zero until the end of the retentostat cultivation, the aspartic acid and glutamine/arginine concentration decreased by only 0.1 mM and 0.75 mM, respectively. This suggests that amino acid metabolism could only partially explain the increase in succinate, indicating that the extra succinate originates from another source. Likewise, the ornithine concentration increased 0.15 mM, conceivably as the end product of arginine catabolism. Ammonia (NH 3 ) production also increased during the retentostat cultivation (1 mM), which correlates with the increased utilization of amino acids. Concentrations of other amino acids (alanine, asparagine, cystine, glutamic acid, glycine, histidine, isoleucine,  leucine, lysine, methionine, phenylalanine, proline, threonine, tryptophan, tyrosine, valine) did not decrease, suggesting that these were not utilized (Fig. S3). Maintenance requirement. Metabolite analysis suggests that B. breve NRBB57 used alternative energy sources under the extremely energy-limited conditions, most likely to produce as much energy as possible and thereby improving its chance of survival. Therefore, we investigated if and to what extend B. breve NRBB57 is capable of decreasing the energy required for its maintenance. To estimate the maintenance requirements at near-zero growth rates, biomass accumulation in the retentostat cultures has been modeled using a modified Verseveld equation, in which ATP production was estimated based on the metabolite production (11) as explained in the experimental procedures. The maximum biomass yield on ATP (Y x/ATP max ) and maintenance coefficient (m ATP,che ) were determined in chemostat cultivations at dilution rates between 0.025 and 0.4 h 21 and estimated to be 16.73 6 0.85 gDW/mol ATP (estimate 6 standard error) and 3.32 6 0.77 mmol ATP Á gDW 21 Á h 21 (estimate 6 standard error), respectively (Fig. 5A). Comparing the m ATP,che with Glutamine and arginine could not be separated in the UPLC, and the decreased concentration could be due to either increased glutamine and/or arginine utilization. Only a selection of the amino acids is shown. All quantified amino acids can be found in (Fig. S3).
Physiology of B. breve at Near-Zero Growth Rates Microbiology Spectrum the biomass specific ATP production rate (q ATP ) during the retentostat cultivation showed that the maintenance coefficient reduced at least 3.6-fold. Therefore, we assumed in the modeling that the maintenance coefficient gradually decreases toward near-zero growth rates as found for Lactococcus lactis (11). Biomass accumulated in the cross-flow filter in the first couple of days after connecting the filter resulting in significant underestimation of the biomass production. Therefore, biomass accumulation in the bioreactor was modeled from 5 days onward, resulting in a good model fit (the square root of the mean squared error is 0.066 and 0.084 gDW/kg (Fig. S4)). The model estimated that during 3 weeks of retentostat cultivation, the maintenance coefficient was reduced 3.6-fold to 0.915 6 0.016 mmol ATP Á gDW 21 Á h 21 (Fig. S5). The estimated growth rates after 1, 2, and 3 weeks of retentostat cultivation were 0.0055 6 0.0011, 0.0018 6 0.00005, and 0.00092 6 0.00006 h 21 , respectively ( Fig. S6). Finally, the estimated growth rates in the retentostat cultures were plotted against the biomass-specific ATP production rate and compared with data from the chemostat cultures used to estimate the Y x/ATP max and m ATP (Fig. 5B). This showed that these estimates are only valid at growth rates above 0.025 h 21 , because at near-zero growth rates the ATP required for maintenance purposes is significantly reduced. Proteome analysis. The proteome of B. breve NRBB57 was analyzed to investigate how this bacterium manages to adapt to the nutrient limitation and consequently decreasing growth rates. Additionally, through this analysis, it was possible to link the phenotypes described in the results above to changes in protein abundances. Cluster analysis of all proteome profiles showed that retentostat and chemostat samples clustered separately, but within the group of retentostat samples (m #0.0055 h 21 ) they cluster per replicate and not per growth rate (Fig. S7).
To detect proteins with a significantly different abundance as a function of the growth rate, the protein abundance ratio (log 2 [condition of interest/0.025]) was plotted against the growth rate (on a log 10 scale). Linear regression was applied on either the high growth rates (m $0.025 h 21 ) or near-zero growth rates (m #0.025 h 21 ) to find the slope and its significance (P value). Because all data were expressed as log 2 values of the protein abundance ratio compared to values obtained at 0.025 h 21 , all regression lines were forced to the point (0.025; 0). Subsequently, each protein was allocated to a group based on their slopes and P values. In total, 1,134 proteins were detected, and 655 out of these showed significant changes in abundance in comparison with the cells A functional enrichment analysis was performed for proteins that showed significant differences in their abundance in cells cultivated at lower growth rates (retentostat cells) in comparison to their abundance in cells cultivated at 0.025 h 21 (chemostat cells). The results show the clusters of orthologous groups (COG) categories to which the proteins with increased and reduced abundances belong to (Fig. S8). Interestingly, among the proteins with increased abundance in the retentostat, the most overrepresented categories (highest score assigned by the software) correspond to (i) posttranscriptional modification, protein turnover, and chaperones; (ii) translation, ribosomal structure, and biogenesis; (iii) nucleotide transport and metabolism; (iv) amino acid transport and metabolism; and (v) replication, recombination, and repair. The category related to "translation and replication" includes many proteins that participate in the stress response, such as chaperones and DNA repair proteins. The COG category "nucleotide transport and metabolism" includes proteins that are likely to be involved in (pp)pGpp synthesis, indicating that the severe nutrient limitation in the retentostat induced a stringent response in B. breve NRBB57. The category "amino acid transport and metabolism" consists of proteins involved in the biosynthesis of several amino acids but also contains several proteins involved in the catabolism of serine (serine dehydratase), aspartic acid (aspartate aminotransferase), and arginine (argininosuccinate synthase, ornithine carbamoyltransferase). Finally, other proteins that were more abundant at near-zero growth rates were proteins involved in polyphosphate synthesis (polyphosphate kinase and proteins involved in phosphate transport) as well as several tRNA synthetases.
On the other hand, for the proteins with decreased abundances at near-zero growth rates, the most overrepresented categories included "energy production and conversion" and "amino acid transport and metabolism." Other proteins that were significantly less abundant at near-zero growth rates were ribosomal proteins (15 out of 52), which indicates a reduction in ribosome synthesis.
Metabolic energy production. Our results show that during the retentostat cultivation, B. breve NRBB57 adapts its metabolism to cope with nutrient scarcity. Lactose was mainly converted to acetate, and at decreasing growth rates lactate concentrations gradually increased. Simultaneously, formate and ethanol concentrations slightly decreased with increasing residence time in the retentostat bioreactor. Additionally, the extracellular concentrations of amino acids (serine, aspartic acid, and glutamine/  Table S1. Physiology of B. breve at Near-Zero Growth Rates Microbiology Spectrum arginine) and glycerol were reduced over time in the retentostat. These observations were linked to the proteome data showing that several proteins involved in the catabolism of amino acids, glycerol, and lactate production had significantly higher abundances than those in the chemostat cultivation. This led to the hypothesis that B. breve NRBB57 started using these amino acids as well as glycerol as additional energy sources during declining growth rates caused by nutrient limitation. Based on the metabolic and proteome analysis, we propose an integration of the catabolic pathways for lactose, arginine, serine, aspartic acid, and glycerol and how the utilization of these substrates potentially contributes to metabolic energy generation (Fig. 7). Lactose metabolism by B. breve leads to the production of acetate, ethanol, formate, and lactate. Moreover, we observed utilization of arginine concomitantly with the production of ornithine. Simultaneously, serine was probably converted to pyruvate and ammonia. In our model, glycerol is converted to pyruvate and aspartate is proposed to be converted to succinate.
(p)ppGpp-induced stringent response. Enzymes involved in the production of the alarmones guanosine tetraphosphate (ppGpp) and guanosine pentaphosphate (pppGpp) and guanosine 59-monophosphate 39-diphosphate (pGpp) (collectively referred as ppGpp) were more abundant at near-zero growth rates (Fig. 8). Proteomic data suggested that synthesis of GDP and GTP, which are precursors of ppGpp and (pp)pGpp, respectively, increased at higher growth rates, but the enzymatic complex RelA/SpoT that converts GDP/GTP into (pp)pGpp was less abundant at higher growth rates, which indicates that GDP/GTP was produced solely for growth purposes. In contrast, proteins involved in GTP synthesis also increased in the retentostat cultivation, but now the enzymatic complex RelA/SpoT increased concomitantly, indicating that at near-zero growth rates the GDP and the GTP are converted to (pp)pGpp to activate the stringent response.

DISCUSSION
In the present study, the physiology of Bifidobacterium breve NRBB57 was investigated at near-zero growth rates using retentostat cultivations. In a retentostat cultivation bacterial cells are calorically restricted but not starving. Several parameters were measured in this study, including biomass accumulation, culturability, viability, morphology, metabolite production and consumption, and proteomic adaptations.
Viable but nonculturable cells. During the 3 weeks of retentostat cultivation, the growth rate of B. breve NRBB57 progressively decreased from 0.025 to 0.00092 h 21 , the latter corresponding to a doubling time of 31 days. Throughout the fermentation, the biomass concentration increased approximately 6-fold. At the same time, culturability significantly decreased to approximately 30%, while viability remained higher than 80%, indicating that 20% of the cells were dead and 50% entered a viable but nonculturable state (VBNC). Morphological changes were also progressively observed during the retentostat cultivation, showing cell elongation and more branching. These results correspond to what has been observed in Lactococcus lactis and Pseudomonas putida (10,11,18), and these changes are likely the result of nutritional stress (29).
The loss of culturability or the increase in the number of cells in the VBNC state in retentostat cultures of P. putida has previously been linked to a decrease in ribosome content (18). The latter phenomenon could also be deduced from the proteome data of B. breve NRBB57, which showed a lower abundance of ribosomal proteins during the retentostat cultivation, indicating a decrease in ribosomal synthesis. Changes in cell morphology have also been linked to VBNC cells as an adaptation to stressful environments. It has previously been hypothesized that this is a mechanism of adaptation to reduce energy requirements (30). By reduced cell division, cells tend to become extended instead of forming a septum and dividing (10). This agrees with (i) our observations of a 3.6-fold reduction of the maintenance requirements of B. breve after 3 weeks of retentostat cultivation and (ii) morphological changes, as is seen with L. lactis as well (10,11,13,19).
Stringent response. Proteins involved in the production of the alarmones ppGpp, pppGpp, and pGpp (collectively referred as ppGpp) had higher abundances at the In this pathway, lactose is mainly converted to lactate and acetate (2:3). Glycerol is converted to ethanol and serine is converted to acetate. Moreover, aspartate is transformed to succinate and arginine is converted into ornithine. The metabolism of one molecule of lactose delivers 5 ATPs. From serine to acetate, 1 ATP is produced. From aspartate to succinate, 1 reduced flavin adenine dinucleotide (FADH 2 ) is consumed, which means that 0.5 acetyl-coenzyme A (CoA) can be converted to acetate instead of ethanol, which is equivalent to 0.5 ATP. From arginine to ornithine 2 ATPs are produced on the known steps in B. breve. Finally, the conversion of glycerol to ethanol gives 1 ATP. Glu-6-P, glucose-6-phosphate; Fruc-6-P, fructose-6-phosphate; Gly-Al-3-P, glyceraldehyde-3-phosphate; PRPP, phosphoribosyl pyrophosphate; PEP, 2-phosphoenolpyruvate; DHAP, dihydroxyacetone phosphate; Gly-3-P, glycerol-3-phosphate.
Physiology of B. breve at Near-Zero Growth Rates Microbiology Spectrum extremely low growth rates, which indicates induction of the stringent response. The stringent response is a stress signaling system mediated by (pp)pGpp in response to nutrient deprivation that controls many important processes, such as DNA replication, transcription, ribosome synthesis, and maturation (31). The production of (pp)pGpp also plays a key role in inducing the nongrowing state in the bacterial cells (32,33), which could explain why B. breve NRBB57 progressively entered a viable but nonculturable state during the retentostat cultivation. The alarmone (pp)pGpp inhibits ribosomal synthesis and maturation, while it upregulates genes involved in nutrient acquisition (e.g., nutrient transporters) and stress resistance (34). A possible indication of the activation of the stringent response by (pp) pGpp accumulation in our study was observed when 15 out of 52 ribosomal proteins were significantly less abundant at near-zero growth rates compared to at a growth rate of 0.025 h 21 . In contrast, many amino acid biosynthesis proteins were more abundant at near-zero growth rates as well as several proteins involved in protein and DNA repair. Moreover, 10 different tRNA-synthetases, which load amino acids to tRNAs, were significantly more abundant at near-zero growth rates.
Moreover, proteomics results also showed the upregulation of polyphosphate kinase at lower growth rates; this enzyme is involved in the production and accumulation of inorganic polyphosphates. Thereby, the stringent response inhibits exopolyphosphatase and triggers the activity of the polyphosphate kinase and subsequent accumulation of polyphosphate (35,36). Polyphosphate accumulates in response to nutrient starvation as seen in the retentostat cells and plays an important role in the stress response interacting in several important cellular processes that include inhibiting reinitiation of DNA replication and promoting fitness during starvation (37,38).
Pyruvate dissipation. The above-described observations indicate that in the retentostat B. breve NRBB57 downregulates growth-related processes as a consequence of

Physiology of B. breve at Near-Zero Growth Rates
Microbiology Spectrum the reduction in the concentration of lactose available per cell but started to utilize alternative energy sources, such as amino acids and glycerol. Moreover, in all retentostat and chemostat cultures pyruvate was mainly converted to acetate, ethanol, and formate for extra energy generation. Even at the highest growth of 0.4 h 21 with a specific substrate consumption rate of 10 mmol hexose equivalents Á gDW 21 Á h 21 , lactate production was low in the chemostat cultures, while lactose is almost exclusively converted to lactate and acetate in batch cultures of B. breve NRBB57 (data not shown). This indicates that not only is the metabolic shift controlled by the specific rate of sugar consumption (39), but the sugar concentration might also play an important role. While lactose was still mainly converted into acetate, formate, and ethanol at high growth rates, in the retentostat lactate production increased. Nonetheless, lactate was still the product with the lowest concentration detected in the retentostat. The slightly higher production of lactate in the retentostat could be associated with the significantly higher abundance of two L-lactate dehydrogenases observed in our proteomic results.
Alternative energy sources. In addition to downregulation of growth-related processes to save energy, consumption of alternative nutrient sources could provide the bacteria with valuable energy during the extreme energy limitation imposed by retentostat cultivation. In this study, the concentration of several amino acids gradually decreased during the retentostat cultivation, while ammonia increased accordingly, pointing to a possible role for amino acids as alternative carbon and energy sources. These amino acids include glutamine/arginine, aspartic acid, and serine. Because glutamine/arginine depletion coincided with an increase in ornithine, increased arginine utilization was more likely than glutamine utilization.
Although arginine utilization coincided with ornithine production, no arginine deiminase and/or arginase have been identified in the genome of B. breve. Therefore, we hypothesize that arginine is catabolized by reversing the arginine biosynthetic pathway (Fig. 7). All three enzymes, argininosuccinate lyase (ASL), argininosuccinate synthase (ASS), and ornithine carbamoyltransferase (OTC), have been described to catalyze reversible reactions (40)(41)(42). Proteome analysis indicated that both ASS and OTC were significantly more abundant at near-zero growth rates when arginine was catabolized, and ASL and aspartate ammonia lyase (AAL) were present at near-zero growth rates. In many bacteria, the produced carbamoyl-phosphate is further degraded by carbamate kinase (CK) to CO 2 and NH 3 , yielding ATP, but no gene encoding CK has been identified in B. breve or in any other Bifidobacterium species. Interestingly, both the small and large subunits of carbamoyl-phosphate synthase (CPS) were significantly more abundant at near-zero growth rate. Therefore, it is tempting to speculate that either the CPS catalyzes a reversible reaction in B. breve or there is a carbamate kinase in B. breve that has not been identified yet. Because ASS converts AMP to ATP and another ATP could be produced by CK, in total 3 ATPs could be produced per arginine by this hypothetical pathway (Fig. 7), outlining how arginine could serve as alternative energy source.
The aspartic acid and serine concentrations decreased during the retentostat cultivation, indicating increased utilization at near-zero growth rates. Serine was most likely converted into pyruvate and ammonia by serine dehydratase (SD) (43,44), which was significantly more abundant at near-zero growth rates. Subsequently, pyruvate could be converted into acetate by pyruvate formate lyase (PFL), phosphotransacetylase (PTA), and acetate kinase (ACK) to yield 1 ATP per pyruvate (Fig. 7).
Aspartic acid was most likely converted first to fumarate via either aspartate ammonia lyase (AAL) or via aspartate aminotransferase (AAT), malate dehydrogenase (MDH), and fumarase (Fum) (Fig. 7). AAL was produced and AAT was even significantly more abundant in the retentostat. However, no MDH-or Fum-encoding genes have been identified in the genome of B. breve, although minor MDH activity was found in other Bifidobacterium spp. (45). The resulting fumarate is most likely converted to succinate by succinate dehydrogenase (SDH), and the abundance of both subunits significantly increased at lower growth rates in the chemostat culture, and their abundance remained high in the retentostat cultures. This hypothesis is supported by the observed increase in the succinate concentration over time in the retentostat. Flavin adenine dinucleotide (FAD) is reduced to FADH 2 by conversion of fumarate to succinate, while aspartic acid acts as an electron acceptor. This allows for the production of 0.5 to 1 ATP per aspartic acid because relatively more pyruvate can be converted to acetate instead of ethanol. Succinate production from aspartic acid could also potentially explain the observed succinate production in batch cultures of other Bifidobacterium species (39), indicating that those species might also metabolize aspartic acid as an alternative energy source.
Glycerol. In addition to the consumption of several amino acids, surprisingly, glycerol was also consumed in the retentostat cultures. Since genes encoding known enzymes involved in glycerol metabolism are absent in the genome of B. breve NRBB57, we analyzed the corresponding proteomes and noted an increased abundance of glycerol-3-phosphate dehydrogenase (G3PDH) at near-zero growth rates in the retentostat. Therefore, we speculate that glycerol was phosphorylated by an unidentified enzyme to glycerol-3-phophate (G3P), which was subsequently oxidized to dihydroxyacetone phosphate (DHAP) entering the glycolytic pathway. Further conversion to ethanol would result in a redox-neutral pathway in which 1 ATP per glycerol could be produced. We did not observe consumption of glycerol by B. breve NRBB57 in batch cultures; nor was growth observed under such conditions with glycerol as the sole substrate for growth, suggesting specific activation of the putative pathway under extreme energy restriction in retentostat cultivations. Additional research is required to elucidate the activation and role of glycerol metabolism in B. breve.
Finally, we highlight how this study contributes to our understanding of the ecophysiology of B. breve. First, we showed the production of acetate, formate, and lactate at near-zero growth rates. These metabolites can act as substrates for cross-feeding of other beneficial bacteria and for the formation of other short-chain fatty acids such propionate and butyrate (46). These fatty acids, besides helping to maintain a low pH in the gut and thereby inhibiting pathogenic bacteria, conceivably contribute to other health benefits to the host (47,48). Similarly, succinate is also a substrate for crossfeeding of other microorganisms in the gut and an intermediate in microbial propionate synthesis (49). Lastly, ornithine has been suggested to be involved in the support of homeostasis in the gut mucosa (50). Our findings show how B. breve can adapt its metabolism at lower growth rates while simultaneously producing metabolites that can be of great benefit to the human host.
To conclude, in this study we demonstrated that B. breve NRBB57 can be cultivated in the retentostat system at near-zero growth rates while remaining viable and metabolically active. Metabolite and proteome analysis revealed that the extreme energy restriction induced a multifaceted response, including stress defense and stringent response, metabolic shifts, and activation of alternative energy-producing pathways. It is conceivable that microorganisms encounter near-zero growth rate-inducing conditions in a range of natural environments, but the mechanisms used to adapt to these conditions are poorly understood. The use of retentostat cultivations provided insights into how B. breve adapts to nutrient scarcity and revealed cellular responses uniquely linked to physiology at near-zero growth rates.

MATERIALS AND METHODS
Bacterial isolate and storage. In this study, the strain Bifidobacterium breve NRBB57 obtained from the strain collection of Danone Nutricia research (Utrecht, The Netherlands) was used. This strain was stored in glycerol stocks (30%) at 280°C until use.
Culture conditions. Glycerol stocks were streaked on TOS-propionate agar (Merck, Germany) and incubated at 37°C for 48 h in anaerobic jars (Advanced Instruments, USA) containing anaerobic gas-generating sachets (Oxoid AnaeroGen). For overnight cultures, single colonies were resuspended in 10 mL TOS-propionate broth (Merck, Germany), which was made according to the manufacturer's instructions, while agar was removed by filtration over a 520B qualitative filter paper (Whatman, United Kingdom) prior to autoclaving the broth filtrate. Overnight cultures were incubated for 20 h at 37°C in anaerobic jars containing anaerobic gas-generating sachets before they were used to inoculate the chemostat and retentostat cultures.
Chemostat cultivation. B. breve NRBB57 was grown in chemostats in biological triplicates at dilution rates of 0.12, 0.25, and 0.4 h 21 in 0.5-L bioreactors and 0.025 h 21 in 1-L bioreactors (Multifors, Infors HT, Switzerland). The temperature was kept constant at 37°C during the fermentation, the pH was controlled at 6.5 by automatic addition of 5 M NaOH, and the stirring speed was set at 300 rpm. For anaerobic conditions, the headspace was flushed with gas composed of 5% CO 2 , 5% H 2 , and 90% N 2 at a rate of 0.06 L/min. Bioreactors were inoculated with 1% of an overnight culture made in TOS-propionate broth (Merck, Germany) as previously described. Bacteria were grown until the end of the exponential phase. Subsequently, the supply of fresh medium was turned on to start the chemostat mode at the dilution rates previously mentioned. Samples were taken once the steady state was reached, which was considered to be achieved after a minimum of 5 volume changes.
Retentostat cultivation. Biological triplicate retentostats were carried out in 1-L bioreactors (Multifors, Infors HT, Switzerland). Temperature (37°C), pH (6.5), stirring speed, and the anaerobic environment were kept constant as previously described for the chemostat cultures. The bioreactors were inoculated with an overnight culture (1%) made in TOS-propionate broth (Merck, Germany) as previously described. Bacteria were allowed to grow until the end of the exponential phase, after which a fresh medium supply was turned on to start the chemostat mode at a dilution rate of 0.025 h 21 . After the chemostat reached the steady state, a polyethersulfone crossflow filter (0.2 mm; Spectrum Laboratories, USA) was connected to an outer loop in the effluent line to begin the retentostat mode. Retentostats were run for approximately 3 weeks. Samples were taken every 2 to 3 days.
Culturability. To determine the concentration of culturable cells, samples were serially diluted in peptone physiological salt solution (PPS; Tritium Microbiology, The Netherlands), and appropriate dilutions were spread plated on TOS-propionate agar (Merck, Germany). Plates were incubated anaerobically at 37°C for 48 to 72 h, after which the colonies were counted.
Viability. The culture viability was determined using the LIVE/DEAD BacLight kit (Molecular Probe Europe, Leiden, The Netherlands) according to the manufacturer's instructions. The fluorescence of the cells in the treated sample was visualized with a fluorescence microscope (Olympus) at a magnification of 630 times. Bacterial cells with a compromised membrane are considered dead and will stain red with propidium iodide (PI). Meanwhile cells with an intact membrane were stained green with the DNA probe SYTO 9 and were presumed viable. Cells were counted up to 100 per sample, and the percentage of viable cells was calculated.
Cell counts. To determine the cell concentration, samples taken from the bioreactor were diluted 10 to 100 times in peptone physiological salt solution, and 25 mL was placed in a cell counting chamber (CellVision Technologies, The Netherlands). Subsequently, bacterial cells were counted in a phase-contrast microscope (Olympus, Japan) at a magnification of 1,000 times. Chains of cells were counted as one, assuming that chains would only form one colony.
Cell dry weight. Approximately 3 mL of sample (previously weighed) taken from the bioreactor was passed through a preweighed 0.2-mm filter (Pall Corporation, USA) with the help of a vacuum filtration pump. Filters were then washed with approximately 40 mL of demineralized water and dried at 80°C for 48 h. Afterward, the filters were weighed on an analytical balance again to calculate the cell dry weight (CDW) concentration of the sample. CDWs were measured in duplicate for every sampling point.
Scanning electron microscopy. The morphology of B. breve NRBB57 cells in the retentostat was investigated with scanning electron microscopy (SEM). The samples were collected at time zero (chemostat) and 7, 15, and 20 days after switching to retentostat mode and centrifuged at 17,000 Â g for 1 min, and the pellets were frozen at 220°C until use. Cells were visualized by SEM as previously described (11). Briefly, samples were thawed and resuspended in PPS. For every sample, a drop of the suspension was placed in a poly-L-lysine coated coverslip (Corning BioCoat, USA) and left for 1 h at room temperature. Then, the coverslips were rinsed with phosphate-buffered saline and fixed with 3% glutaraldehyde buffer for 1 h. Afterward, the samples were dehydrated in a graded series of ethanol followed by drying with CO 2 (Leica EM CPD 300, Leica Microsystems, Germany). The coverslips were fitted onto sample stubs with carbon adhesive tabs and sputter coated with 10 nm tungsten (Leica SCD500). Finally, samples were imaged at 2 KV, 6 pico-amperes, at room temperature in a field emission scanning electron microscope (Magellan 400, FEI Company, USA).
Metabolite analysis. Samples were taken directly from the bioreactor and centrifuged at 17,000 Â g for 1 min, and supernatants were stored at 220°C until quantification of extracellular metabolites by high performance liquid chromatography (HPLC) and ultra-high-performance liquid chromatography (UPLC).
For the quantification of lactose, lactate, acetate, ethanol, formate, glycerol, and succinate, samples were deproteinated by the addition of 100 mL of cold Carrez A (0.1 M potassium ferrocyanide trihydrate) to 200 mL sample. After mixing, 100 mL of cold Carrez B (0.2 M zinc sulfate heptahydrate) was added, followed by mixing and centrifugation at 17,000 Â g for 10 min. Then, 10 mL of the deproteinated sample was injected on an Ultimate 3000 device (Dionex, Germany) equipped with an Aminex HPX-87H column (300 by 7.8 mm) with guard-column (Bio-Rad). As the mobile phase, 5 mM H 2 SO 4 was used at a flow rate of 0.6 mL/min. The column temperature was kept at 40°C. Compounds were detected using a refractive index detector (RefractoMax 520). The analysis was performed with technical duplicates.
Amino acids (alanine, asparagine, aspartic acid, cystine, glutamic acid, glutamine, arginine, glycine, histidine, isoleucine, leucine, lysine, methionine, phenylalanine, proline, serine, threonine, tryptophan, tyrosine, valine) and ammonia (NH 3 ) were quantified by UPLC. Then, 40 mL supernatant was deproteinated by addition of 50 mL of 0.1 M HCl, containing 250 mM of norvaline as the internal standard and 10 mL of 30% sulfosalicylic acid (SSA). Subsequently, the solution was mixed and centrifuged at 17,000 Â g for 10 min at 4°C. Amino acids and ammonium were derivatized using the AccQTag Ultra derivatization kit (Waters, USA). Next, 20 mL of the deproteinated supernatant or standard amino acids mixture was mixed with 60 mL of AccQTag Ultra borate buffer in glass vials. For deproteinated samples, 75 mL of 4 M NaOH was added to 5 mL of borate buffer to neutralize the addition of SSA. Subsequently, 20 mL of a AccQTag reagent dissolved in 2.0 mL of AccQTag Ultra reagent diluent was added and immediately vortexed for 10 s. Then, the sample solution was capped and heated at 55°C in a heatblock for 10 min. Amino acids and ammonium were quantified by UPLC by injection of 1 mL of sample on an Ultimate 3000 device (Dionex, Germany) equipped with a AccQTag Ultra BEH C 18 column (150 mm by 2.1 mm, 1.7 mm) (Waters) with a BEH C 18 guard column (5 mm by 2.1 mm, 1.7 mm) (Waters). The column temperature was set at 55°C, and the mobile-phase flow rate was maintained at 0.7 mL/min. Eluent A was 5% AccQTag Ultra concentrate solvent A, and eluent B was 100% AccQTag Ultra solvent B. The separation gradient was 0 to 0.04 min 99.9% A, 5.24 min 90.9% A, 7.24 min 78.8% A, 8.54 min 57.8% A, 8.55 to 10.14 min 10% A, and 10.23 to 17 min 99.9% A. Compounds were detected by UV measurement at 260 nm. Glutamine and arginine could not be separated in the UPLC analysis.
Modeling biomass accumulation and estimation maintenance. The accumulation of biomass in the retentostat cultivations was modeled using a modified Verseveld equation in which metabolic changes are taken into account (11) (equation 1). Moreover, we assumed that the maintenance coefficient gradually decreases toward near-zero growth rates as found for Lactococcus lactis (equation 2), (11), in which the growth rate was calculated by taking the first derivative of equation 1 giving equation 3.
in which t is the time (in hours), C x is the biomass concentration (gDW/kg), C x,0 is the biomass concentration at t = t -Dt (gDW/kg), D is the dilution rate (h 21 ), C s,in is the substrate concentration in the medium (15 mmol/kg lactose), C S is the substrate concentration in the effluent, Y ATP/s is the ATP yield on substrate (mol ATP/CmolS), m ATP is the maintenance coefficient (mol ATP Á gDW 21 Á h 21 ), Y x/ATP max is the maximum biomass yield on ATP measured in chemostat cultivations (gDW/mol ATP), m ATP,che is the maintenance coefficient measured in chemostat cultivations (mol ATP Á gDW 21 Á h 21 ), m is the growth rate (h 21 ), a and b are parameters describing the relation between the maintenance coefficient and the growth rate.
The Y ATP/s was calculated based on the measured metabolite production (equation 4) assuming lactose was the only energy source; 1 mol ATP is produced per mol acetate produced in the bifid shunt, while 2 mol ATP is produced per mol acetate produced via pyruvate formate lyase, and 1 mol ATP is produced per mol lactate and ethanol: (4) in which R i is the production rate (mol/h) of compound i. Because in the bifid shunt 1 mol lactose is converted to 3 mol acetate, equation 4 can be rewritten as: Input data for the modeling were online optical density measurements, which were converted to biomass dry weight concentrations using a second-order polynomial relation. To fit the model, the variable parameters a and b were optimized by minimizing the sum of squared errors between the model and the data in 10-min time intervals. This was done using the solver add-in of Excel.
Proteome analysis. For the proteome analysis, samples were taken from the steady states of 3 biological replicates of the chemostats (m, 0.4 h 21 ; m, 0.12 h 21 ; m, 0.05 h 21 ; m, 0.025 h 21 ) and from 3 biological replicates of the retentostats after 1, 2, and 3 weeks after connecting the filter. Samples were centrifuged at 13,000 Â g for 5 min, and pellets were frozen at 280°C until use.
Cell lysis was achieved through bead beating in a FastPrep-24 5G instrument (MP Biomedicals) 6 times for 30 s at 6.5 m/s with cooling after every bead step. Protein concentration was assessed using the Pierce bicinchoninic acid (BCA) protein assay (Thermo Fisher Scientific, Waltham, MA, USA). Subsequent protein digestion was performed overnight using dithiothreitol (DTT, 2 mM), iodoacetamide (IAA, 4 mM), and trypsin (1:50 of a 1 mg/mL solution) at 37°C. Cleanup was performed with solid phase extraction (SPE) columns (Thermo Fisher Scientific) with acetic acid (100 mM in 95% acetonitrile) to be Physiology of B. breve at Near-Zero Growth Rates Microbiology Spectrum finally dissolved in the eluent acetic acid (100 mM). Samples were analyzed by nano-liquid chromatography high-resolution mass spectrometry (nano-LC-HRMS/MS) as previously described (51). An Agilent 1200 HPLC system (Agilent Technologies, Santa Clara, USA) was used, connected to a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific). Peptides were trapped on a 100-mm inner diameter trap column packed using ReproSil-Pur C 18 -AQ, 3 mm resin (Dr. Maisch, Ammerbuch, Germany) at 5 mL/min in 100 mM acetic acid. Afterward, the peptides were eluted at 100 nL/min in a 90-min extended gradient from 10 to 40% acetic acid solvent (in 95% acetonitrile) to a 20-cm IntegraFrit column (50 mm inner diameter; Reprosil-Pur C 18 -AQ 3 mm, New Objective, Woburn, MA, USA). The acquired spectra were analyzed using Thermo Proteome Discoverer (v2.4) in combination with Mascot (v2.5) (Thermo Fisher Scientific). The reference database comprised protein sequences from B. breve NRBB57 from UniProt and typical contaminants. Built-in Percolator was used with default settings for postprocessing of Mascot peptide spectrum matches (PSMs) from Proteome Discoverer. In all experiments, PSMs were filtered to a peptide false-discovery rate of 1% using q values that were calculated based on PSM score distributions for decoy database searches as well as considering a minimum peptide length of 6 amino acids. Proteins were filtered to a protein false-discovery rate of 1% and a minimum requirement of 2 unique peptides. Relative protein quantification was performed with Proteome Discoverer based on peptide intensity signals using default settings. The obtained abundances of all detected proteins are listed in Table S1. Proteome data analysis. Samples were taken in biological triplicates from a wide range of growth rates in chemostat cultures (0.4, 0.25, 0.12, and 0.025 h 21 ) and retentostat cultures (0.0055, 0.0018, and 0.00092 h 21 ). Protein abundances in all conditions were compared with the chemostat at 0.025 h 21 , because this condition also served as the starting point for the retentostat cultivations. To determine significant differences in the proteome between growth rates, the abundances of the proteins were plotted against the growth rate. Afterward, linear regression was calculated to find the slopes of the line and check how significantly this slope differed from 0 (P value) in R (v3.63). The significantly different abundances were considered when the P value of the slope was ,0.05. Finally, proteins that changed significantly were clustered with GSEA-pro (v3.0) software (http://gseapro.molgenrug.nl/). The proteins were classified into clusters of orthologous groups (COGs) to predict the functions of the different sets of proteins.
Data availability. The raw proteomic data supporting the conclusions of this article have been deposited on 4TU.ResearchData (https://data.4tu.nl/) with the digital object identifier 10.4121/21960575.