Respiratory Physiology of Lactococcus lactis in Chemostat Cultures and Its Effect on Cellular Robustness in Frozen and Freeze-Dried Starter Cultures

Lactococcus lactis is used in large quantities by the food and biotechnology industries. L. lactis can use oxygen for respiration if heme is supplied in the growth medium. This has been extensively studied in batch cultures using various mutants, but quantitative studies of how the cell growth affects respiratory metabolism, energetics, and cell quality are surprisingly scarce. Our results demonstrate that the respiratory metabolism of L. lactis is remarkably flexible and can be modulated by controlling the specific growth rate. We also link the physiological state of cells during cultivation to the quality of frozen or freeze-dried cells, which is relevant to the industry that may lack understanding of such relationships. This study extends our knowledge of respiratory metabolism in L. lactis and its impact on frozen and freeze-dried starter culture products, and it illustrates the influence of cultivation conditions and microbial physiology on the quality of starter cultures.

lactis (16). However, in the presence of heme, the ETC functions as a proton transport system causing respiration-driven efflux of protons in addition to, or instead of, H ϩ -ATPase (17). L. lactis thus saves ATP through heme-activated respiration, as NADH dehydrogenase regenerates NAD ϩ while protons are extruded at no additional ATP cost. Some studies suggest that additional ATP generation may occur; however, this has not yet been fully verified (8,15).
Furthermore, respiration-permissive conditions direct metabolism toward flavorforming pathways. The flux from pyruvate to lactate decreases significantly, and acetoin, diacetyl, and CO 2 are formed instead. In the production of acetoin and diacetyl, pyruvate undergoes decarboxylation by acetolactate synthase (ALS) into acetolactate (Fig. 1). Acetolactate is unstable and can either undergo spontaneous conversion to diacetyl by a chemical reaction with O 2 , or it can be enzymatically converted to acetoin by acetolactate decarboxylase (18).
The advantages of respiration-permissive conditions are well established in both the literature and in industrial practice; the intracellular oxygen level is decreased, acid production is reduced, and thus, the stress on the cells is lower. This leads to an increase in biomass yield, resistance to oxygen, and long-term cellular survival in liquid-form cultures stored at 4°C (4,5,19). Nevertheless, respiration in L. lactis has exclusively been conducted in batch cultivations and at low sugar concentrations. Quantitative analyses of metabolic fluxes and energetics are scarce and have considered neither the potential relationship between the specific growth rate and the beneficial effects of respiration nor the potential effects on the robustness in terms of tolerance to downstream processing.
The aim of the present study was to understand the relationship between respiration-permissive conditions during starter culture production and the microbial robustness during downstream treatment of the culture. We mimicked conditions that prevail during different phases of industrial batch production by culturing L. lactis subsp. lactis to steady states at three dilution rates in respiration-permissive and anaerobic chemostat cultures. We performed quantitative metabolic analysis to investigate how cellular metabolism and energetics depend on respiration and the specific growth rate. Furthermore, we explored how the specific growth rate and respiration influence the robustness of frozen and freeze-dried starter cultures by assessing the viability and acidification activity after industry-like downstream processing.

RESULTS
Metabolic fluxes and yields. Lactococcus lactis CHCC2862 was grown in continuous cultures on rich medium with 40 g liter Ϫ1 lactose under respiration-permissive conditions, i.e., aerobically and with hemin supplementation, and under anaerobic conditions in order to study the effects of respiration and specific growth rate on metabolite yields, energetics of growth, and robustness during and after downstream processing. The experimental procedures are schematically illustrated in Fig. 2.
The residual lactose concentration increased with increasing dilution rates under both anaerobic and respiration-permissive conditions ( Fig. 3A and B). The biomass concentration was higher under respiration-permissive than anaerobic conditions. Lactate was the main product under anaerobic conditions ( Fig. 3A and C), whereas acetoin was the main product under respiration-permissive conditions, together with CO 2 and lactate, depending on the dilution rate ( Fig. 3B and D, Table 1). At D ϭ 0.1 h Ϫ1 , no detectable lactate was formed in the respiration-permissive chemostat.
Under anaerobic conditions, lactose was almost exclusively converted to lactate, and only minor amounts of formate, acetate, and ethanol were formed (Fig. 3C, Table 1). The metabolism remained homofermentative, with mainly lactate being produced at all dilution rates. Under respiration-permissive conditions, the metabolite profile instead consisted of a mixture of acetoin, CO 2 , acetate, and lactate, with CO 2 exhibiting the highest specific flux (Fig. 3D, Table 1). At D ϭ 0.1 h Ϫ1 the lactate flux was close to zero, but it increased linearly with increasing dilution rates. The specific O 2 uptake rate also increased with increasing dilution rates ( Fig. 3D; P Ͻ 0.05). However, the ratio between CO 2 production and O 2 uptake (the respiratory quotient, RQ) was higher at D ϭ 0.1 h Ϫ1 and RQ ϭ 1.97 than at D ϭ 0.5 h Ϫ1 and 0.8 h Ϫ1 where RQ ϭ 1.88 (P Ͻ 0.05). This was in line with the higher activity of CO 2 -producing metabolic pathways at D ϭ 0.1 h Ϫ1 , where acetoin was the main product (Table 1). Diacetyl was detected in all respiratory cultures but at low concentrations.
Under anaerobic conditions, the biomass yield on consumed lactose was significantly lower (P Ͻ 0.05) at D ϭ 0.1 h Ϫ1 (0.11 mol carbon [C-mol] C-mol Ϫ1 ) than at D ϭ 0.5 h Ϫ1 and D ϭ 0.8 h Ϫ1 (0.24 and 0.26 C-mol C-mol Ϫ1 , respectively). Lactate was the main product, and lactate yields were all similar at 0.96 to 0.99 C-mol C-mol Ϫ1 (Table 1). Acetate yields were low and decreased significantly with increasing dilution rate (P Ͻ 0.05).
Under respiration-permissive conditions, the biomass yield at D ϭ 0.1 h Ϫ1 tended to be lower than that at D ϭ 0.5 h Ϫ1 and D ϭ 0.8 h Ϫ1 ; however, the differences were much smaller. At D ϭ 0.1 h Ϫ1 , no lactate was produced, and carbon was instead directed to the flavor-forming and mixed acid metabolites acetoin, acetate, and CO 2 . The lactate yield on consumed lactose increased with increasing dilution rates, while acetoin and CO 2 yields decreased (Table 1; P Ͻ 0.05).
Energy and redox balances. The substrate-related maintenance coefficient (m S ) and maximum biomass yield on lactose (Y SX max ) of L. lactis CHCC2862 were estimated from the specific lactose consumption by linear regression. The relationships between the specific lactose consumption rate and the dilution rate were linear and clearly significant under both anaerobic and respiration-permissive conditions (observed F-test statistic F obs, AN ϭ 656 Ͼ F 0.001,1,10 ϭ 21 and F obs, RES ϭ 1,072 Ͼ F 0.001,1,9 ϭ 23, respectively), indicating that the maintenance coefficient was independent of the specific growth rate in both cases (Fig. 4A). The maintenance coefficient was significantly lower under respiration-permissive conditions, 2.1 Ϯ 4.7 mmol of carbon per gram dry cell weight per hour (C-mmol gDW Ϫ1 h Ϫ1 , 95% confidence interval) than under anaerobic conditions (26.0 Ϯ 5.6 C-mmol gDW Ϫ1 h Ϫ1 ). The maximum biomass yields on lactose were very similar under respiration-permissive and anaerobic conditions, 1/(120 Ϯ 8) and 1/(122 Ϯ 10) gDW C-mmol Ϫ1 , respectively, corresponding to 0.33 Ϯ 0.03 C-mol C-mol Ϫ1 .
As a consequence, the fraction of carbon (C-mol C-mol Ϫ1 ) used for maintenance under respiration-permissive conditions was 14%, 3%, and 2% of the consumed lactose at D ϭ 0.1 h Ϫ1 , 0.5 h Ϫ1 , and 0.8 h Ϫ1 , respectively (Fig. 4B). The maintenance energy demand was higher under anaerobic conditions; 67%, 29%, and 22% of the consumed lactose was used for maintenance at D ϭ 0.1 h Ϫ1 , 0.5 h Ϫ1 , and 0.8 h Ϫ1 , respectively.
The NADH-NAD ϩ balance in the central carbon metabolism and the ATP produced via substrate-level phosphorylation, ATP SLP , were estimated based on the lactose catabolism and yields of all by-products (Table 2). Under anaerobic conditions, the balance of NADH formed and NAD ϩ regenerated in catabolism (see equation 2) was almost complete, indicating small experimental errors. An excess of NADH was found in lactose catabolism under respiration-permissive conditions (see equation 3). However, under respiration-permissive conditions, NADH can donate electrons to the ETC, with O 2 being the final acceptor. The estimated NADH reoxidized in the ETC due to O 2 uptake (see equation 4) showed that the overall NADH-NAD ϩ balance was close to zero also under respiration-permissive conditions (see equation 5; Table 2). Acetoin production was associated with 71% Ϯ 13% of the NADH consumed in respiration.
The estimated ATP formed by substrate-level phosphorylation (ATP SLP ; see equation 6) was lower under respiration-permissive conditions at all dilution rates (P Ͻ 0.05), despite the fact that the biomass yields were significantly higher ( Table 2).
Degree of cell damage at each process step. Cell robustness is essential throughout the production process to ensure starter cultures of high quality. To investigate the effect of fermentation conditions on cellular robustness, cells were harvested from each fermen-  tation experiment after steady-state samples had been collected. The complete fermentation broth (FB) was divided into two equal parts that were used to produce frozen product (FP) and freeze-dried product (FDP), respectively (Fig. 2). The robustness was assessed by measuring the number of cells and their membrane potential using flow cytometry to estimate the percentage of damaged cells after each process step (Fig. 5). The total number of cells per gram increased between the downstream process steps as cells were concentrated by centrifugation and freeze-drying (Fig. 5A). According to cell counts measured in each of the process steps (FB, FP, FDP), cell recoveries after downstream treatments were complete within 3% Ϯ 2% and 7% Ϯ 1% for FP and FDP, respectively. In general, cells in the frozen and freeze-dried products had a higher degree of cell damage than those in the fermentation broth (Fig. 5B). There was a tendency toward an increasing percentage of damaged cells with increasing dilution rate under anaerobic conditions, and the highest percentage of damaged cells (14%) was determined in the frozen product harvested at D ϭ 0.8 h Ϫ1 . Cells harvested under respiration-permissive culture conditions showed the opposite trend; the percentage of damaged cells decreased with increasing dilution rate and was only about 3% in the frozen and freeze-dried products harvested at 0.5 h Ϫ1 and 0.8 h Ϫ1 , respectively (Fig. 5B). Acidification performance. The effect of the cultivation conditions on the acidification performance after each process step was assessed by a standardized milk acidification assay (Fig. 6). For anaerobically cultured cells, the downstream processing alone did not seem to influence the maximum acidification rate (r pH,max ) between cells in the fermentation broth (FB), frozen product (FP), and freeze-dried product (FDP). However, it was significantly affected by low dilution rates in all process steps. At D ϭ 0.1 h Ϫ1 , the maximum acidification rates of FB, FP, and FDP were about 13% lower  than at the higher dilution rates ( Fig. 6A; P Ͻ 0.05). No significant differences were found in the specific acidification time (t spe ) across product types or the dilution rates, although the freeze-dried products tended to perform slightly better at the higher dilution rates (Fig. 6B).
In respiration-permissive conditions, the maximum acidification rate (r pH,max ) appeared to be independent of the dilution rate, and no significant differences were observed among the fermentation broth and the frozen and freeze-dried products prepared (Fig. 6A). On the other hand, the specific acidification time (t spe ) depended on both the dilution rate and the downstream processing. After cultivation at the two higher dilution rates, the t spe for the FB, FP, and FDP products was shorter (i.e., better) than at D ϭ 0.1 h Ϫ1 (Fig. 6B). At low dilution rates, the cells were highly sensitive to freeze-drying. The specific acidification time of the freeze-dried product was more than 60% longer than that of the frozen product at D ϭ 0.1 h Ϫ1 . The sensitivity to freezedrying was less severe at the higher dilution rates, although a difference in t spe could still be observed between the freeze-dried and the frozen products at D ϭ 0.5 h Ϫ1 . Moreover, compared to the anaerobic conditions, the frozen products of D ϭ 0.5 h Ϫ1 and 0.8 h Ϫ1 and the freeze-dried product of D ϭ 0.8 h Ϫ1 resulted in similar r pH,max and shorter t spe than their anaerobic counterparts, making the respiro-fermentative cells at D ϭ 0.5 h Ϫ1 and 0.8 h Ϫ1 the best-performing products among all conditions tested.

DISCUSSION
Metabolic shifts and respiration in Lactococcus lactis. The metabolic shift between homofermentative and mixed-acid fermentation under anaerobic conditions can be strain dependent, although factors such as carbon excess, high glycolytic flux, and rapid growth are often associated with homofermentative metabolism. Mixed-acid  fermentation, on the other hand, is associated with low specific growth rate, carbon limitation, and low glycolytic flux (11,14). Aerobic conditions appear to lead to improved growth yields in most lactic acid bacteria (20). However, while six out of six strains of L. lactis displayed a respiration-associated metabolic shift upon aeration and addition of heme (21), only about 20% of 76 screened strains of Lactobacillus, Leuconostoc, and Weissella species showed respiratory phenotypes (20).
We used chemostat cultures under anaerobic and respiration-permissive conditions to enable quantitative analysis of the metabolic shifts between homofermentative fermentation, mixed-acid fermentation, flavor formation, and respiration in L. lactis CHCC2862 using an industry-like rich medium. We found that L. lactis CHCC2862 remained homofermentative in anaerobic chemostats, regardless of the dilution rate. The acetate yield on lactose increased with decreasing dilution rates, but the overall yield of lactate on consumed lactose was above 96% in all anaerobic chemostats.
In contrast, under respiration-permissive conditions the metabolite production profile varied with the dilution rate. In conjunction with respiration, the flavor-forming and mixed-acid pathways were active in L. lactis CHCC2862 at all growth rates, but clear differences were observed in lactate formation. The lactate concentration was below the HPLC detection limit at the lowest dilution rate, while a mixture of lactate and acetoin, acetate, and CO 2 was produced at the two higher dilution rates. Several enzymes, primarily LDH, ALS, PDH, and PFL, compete for pyruvate as the substrate, and LDH also competes with, e.g., NADH oxidase for NADH. It was clear from the overall yields that, at the highest dilution rates, the most active enzymes were ALS and LDH, as the pyruvate flux was directed toward acetoin and lactate. Acetate was also formed, but the specific rate of acetate formation plateaued, along with decreasing yield of acetate on lactose. This may be attributed to allosteric regulation of several enzymes by fructose-1,6-bisphosphate (FBP). The FBP level can be expected to increase at high specific growth rates (13,14,22). FBP increases the affinity of LDH toward both pyruvate and NADH (23), and it is known to inhibit PTA and ACKA, which are involved in acetate formation (22,24). Together, this would mean that carbon flux is redirected away from acetate toward lactate at high dilution rates. It is, however, interesting that such a shift was not clearly seen under anaerobic conditions in this strain, which showed similar yields at all anaerobic steady states. Tentatively, this could be due to regulation via the NADH/NAD ϩ ratio, which should generally be higher under anaerobic than under aerobic conditions. Moreover, at D ϭ 0.1 h Ϫ1 , the specific lactose uptake rate, which translates into the glycolytic rate, was almost three times higher under anaerobic conditions (q S ϭ 3.25 Ϯ 0.41 mmol gDW Ϫ1 h Ϫ1 ) than under respiration-permissive conditions at the same dilution rate (q S ϭ 1.22 Ϯ 0.09 mmol gDW Ϫ1 h Ϫ1 ).
Previous studies have shown that few changes in transcription and protein synthesis are direct results of respiration-permissive conditions (7,25). It has been suggested that NADH is a key regulator in carbon metabolism, as a low NADH/NAD ϩ ratio could allosterically, or via substrate concentration-related effects, redirect metabolism in favor of enzymes using NAD ϩ or pyruvate as the substrate, without necessarily changing the gene expression levels or enzyme concentrations (12,26). Recently, Cesselin and colleagues analyzed the transcriptional response in early stationary phase under respiration-permissive conditions and found that late changes in the transcriptome included significant upregulation of several genes in the acetate (pdhABCD, poxL, and pflB) and acetoin (als, aldB, butA, and butB) pathways (21). Using mutants with inactivated pdhA, pfl, pta, and als, they concluded that acetate production is necessary for the increased biomass yield, while the acetoin production, being pH neutral, alleviates acid stress and prolongs survival. Moreover, pyruvate accumulation was identified as the main driver for als activity (21). The kinetic properties of ALS, PDH, and LDH also favor acetoin production at high intracellular pyruvate concentrations (27).
Our results suggest that, in addition, acetoin production is required to compensate for the NADH consumed in respiration. We calculated that there was a net production of NADH from catabolism of lactose in respiratory cultures. This is related to the higher levels of flavor-forming and mixed-acid products. Under respiration-permissive conditions at the lowest dilution rate, almost no lactate was produced, and acetoin and CO 2 were produced instead. If no lactate is formed, NADH must be oxidized either via ethanol formation or via the ETC. Only traces of ethanol were observed, regardless of the dilution rate, which leads us to the conclusion that NADH is oxidized via the ETC. We measured the oxygen uptake rate and found that the respiratory NADH consumption aligned with the net catabolic NADH production to within Ϯ8%, indicating that the net NADH production in catabolism is well balanced with the NADH consumption due to respiration.
The increase in biomass yield is a clear result of heme-activated respiration, whether additional energy is released or whether the energy demand is reduced. Under anaerobic conditions, the H ϩ -ATPase transfers protons across the cytoplasmic membrane at the expense of ATP to maintain the intracellular pH (8,16). Under respirationpermissive conditions, the flow of electrons through the ETC extrudes protons without any additional need for ATP, thereby reducing or eliminating the cost of cellular ATP proton transfer (8). However, respiration in L. lactis is different from that in other respiring organisms such as E. coli and B. subtilis, which use the Krebs cycle to produce NADH for full respiration. L. lactis does not possess all the enzymes required for a complete Krebs cycle and thus generates NADH via fermentation. During respiration, the reoxidation of NADH takes place via the ETC, and the accumulated pyruvate must therefore be consumed via NADH-producing or redox-neutral reactions, such as acetoin formation, to support respiration, which is completely in line with our results and previously reported data (21). Cellular economy. Respiration capacity is often related to both the NADH/NAD ϩ and the ATP/ADP ratios and affects biomass production. However, NADH oxidation and ATP production are only weakly connected in L. lactis, as most of the ATP is generated by substrate-level phosphorylation (15). The function of H ϩ -ATPase in relation to the ETC has not yet been fully elucidated, and the question is whether H ϩ -ATPase contributes with additional ATP besides that produced in glycolysis and mixed-acid fermentation (15). We estimated the substrate-level phosphorylated ATP (ATP SLP ) from the metabolite yields (Table 2). Anaerobic conditions led to higher ATP SLP values than respiration-permissive conditions did. Nevertheless, the biomass yield was significantly increased under all respiration-permissive conditions studied. At the lowest dilution rate, the biomass yield increased 165% compared to anaerobic conditions, whereas it increased 36% and 23% at D ϭ 0.5 h Ϫ1 and D ϭ 0.8 h Ϫ1 , respectively (Table 1).
Heme-activated respiration indisputably increases biomass yield. However, it is not clear if this is because additional ATP is formed or because the cellular ATP requirement is reduced. When environmental conditions are favorable and nutrient supplies are abundant, microorganisms direct nutrient and energy resources primarily toward growth-related cellular processes (28,29). In contrast, when nutrient and energy supplies are limited, microorganisms predominantly employ metabolic energy for maintenance-related processes, which can lead to shifts in metabolic pathways, including changes in energy storage and utilization pathways (30). The maintenance coefficient obtained under anaerobic conditions (Fig. 4A) corresponds to previously reported values of comparable L. lactis strains, such as L. lactis subsp. lactis IL-1403 (31,32). The maintenance energy demand is the likely reason for the lower biomass yield at the lowest dilution rate under both anaerobic and respiration-permissive conditions ( Table 1).
The m S obtained from respiration-permissive conditions was only 10% of the anaerobic value. This explains the large difference between anaerobic and aerobic biomass yields at D ϭ 0.1 h Ϫ1 , as a larger fraction of the lactose consumption would be used for maintenance energy requirements under anaerobic conditions, instead of providing energy for biomass formation. In addition, when excluding the biomass formation, the carbon recovery was below 90% for all the respiratory cultures and around 100% for the anaerobic cultures. Since the calculated carbon recoveries did not include carbon potentially incorporated into biomass, the low carbon recoveries suggest that lactose was assimilated into biomass under respiration-permissive conditions.
These observations indicate either that additional ATP formation could take place from the available lactose due to respiration or that respiration-permissive conditions cause a lower energy requirement for maintenance. Our data do not allow us to clearly distinguish between these mechanisms, but both would lead to fulfilment of the maintenance energy demand at a lower substrate consumption rate. Cellular robustness. In industry, cells are usually cultivated in batch fermentation and harvested in the stationary phase. Batch cultivations are characterized by dynamic changes throughout the process. Early in the batch, the substrate concentration and the specific growth rate are high. Toward the end, the substrate is depleted, product concentrations accumulate, and the specific growth rate decreases. This activates various stress resistance mechanisms due to nutrient limitations, leading to an increased tolerance to other stress conditions, such as freezing and freeze-drying (1). Previous studies have shown that fermentation conditions and harvest time influence the cellular robustness of starter cultures (3). In the present study, cells were instead harvested at steady state in chemostat cultures to investigate how the physiological state of cells affects frozen and freeze-dried products. At steady state, growth occurs at a constant rate lower than the maximum specific growth rate and all culture variables remain constant. This offers the possibility to examine the cell biology in a quantitative and controlled manner. Our results showed that at high dilution rates, a higher degree of cell damage was observed under anaerobic conditions than under respiration-permissive conditions, indicating that respiration leads to more robust cells as long as it occurs in parallel with lactate production at an intermediate or high specific growth rate. The reasons for this are not clear, but the lower acid stress and modulation of the membrane composition in the presence of oxygen may affect the robustness of the final product (21,33).
The acidification activity is an important technological property of industrial dairy starter cultures. The adaptation of cells to a milk medium after their production is crucial for successful acidification in standard dairy production processes. Under anaerobic conditions, the initial acidification activity, here expressed as the specific acidification time (t spe ), did not differ significantly between the three types of product formulations (fermentation broth, frozen, and freeze-dried). However, in respirationpermissive conditions, freeze-dried products had a slightly longer specific acidification time at D ϭ 0.5 h Ϫ1 , and much longer at D ϭ 0.1 h Ϫ1 , than at D ϭ 0.8 h Ϫ1 . These observations suggest that the metabolic shift toward flavor-forming pathways affects the initial ability of freeze-dried cells to acidify milk. They also highlight that the freeze-drying process may have a huge impact and should be included in investigations of the robustness of starter cultures.
The maximum acidification rate (r pH,max ) was not significantly affected by any of the cultivation conditions, except by anaerobic growth at D ϭ 0.1 h Ϫ1 , after which it was on average about 13% lower than in the other cases (P Ͻ 0.05). The reason for the low value of r pH,max at low specific growth rates under anaerobic conditions may potentially be the high-maintenance energy demand, hypothetically leading to loss of function resulting from energy limitation. The relatively high values of both r pH,max and t spe in the freeze-dried product prepared after respiration-permissive cultivation at D ϭ 0.1 h Ϫ1 indicate that the shift of metabolism does not change the potential metabolic acidification rate but does affect the ability of the cells to adapt and initiate growth and milk acidification. The physiological state of cells under respiration-permissive conditions at D ϭ 0.1 h Ϫ1 suppresses lactate formation, which is the principal acidification activity. Furthermore, in respiration-permissive conditions at D ϭ 0.1 h Ϫ1 , high levels of acetoin were present which may have contributed to the poor adaptation, since acetoin has been shown to inhibit cell growth (34).
L. lactis starter cultures are produced in batch cultivations striving for high biomass yield of robust high-performance cells. Chemostat cultivations represent snapshots of what takes place at different points in a batch cultivation. Clearly, the respiratory metabolism was quite flexible. Low specific growth rate resulted in undetectable lactate formation and high acetoin formation. The frozen products of D ϭ 0.5 h Ϫ1 and 0.8 h Ϫ1 and the freeze-dried product of D ϭ 0.8 h Ϫ1 exhibited the highest acidification activity among all tested conditions, whereas cells produced under respiration-permissive conditions at the lowest dilution rate performed very poorly after freeze-drying. This raises the question of whether the activity of the LDH pathway is beneficial or if the high level of acetoin is detrimental for cellular robustness during storage of starter cultures. The harvest point in current starter culture production setup takes place in the stationary phase, where the specific growth rate is very low. To achieve high L. lactis biomass yield and high-performance freeze-dried starter cultures, one may want to reconsider this harvesting procedure.
In conclusion, this study provides a quantitative analysis of chemostat cultures of L. lactis under respiration-permissive and anaerobic conditions. Respiration-permissive conditions gave clearly higher biomass yield, ranging from 0.29 to 0.33 C-mol C-mol Ϫ1 , than anaerobic conditions, 0.11 to 0.26 C-mol C-mol Ϫ1 . Lactate formation was eliminated under respiration-permissive conditions at a dilution rate of D ϭ 0.1 h Ϫ1 . Respiratory cultures have very low maintenance requirements compared to anaerobic cultures, which is probably the main reason for the increased biomass yield. Respiration-permissive conditions have been reported to have several advantages in previous studies, and our findings confirm their superior performance at high dilution rates but not necessarily in freeze-dried products obtained at lower dilution rates. This work thus underlines the importance of systematically studying the upstream as well as the downstream aspects of production processes.

MATERIALS AND METHODS
Upstream procedure. (i) Organism and preculture preparation. The strain used in this study was Lactococcus lactis subsp. lactis CHCC2862, an industrial strain used in cheese-making (Chr. Hansen Culture Collection, Hørsholm, Denmark). Strain material was transformed into glycerol stocks using overnight static cultivation in M17 medium (Oxoid) and 5 g liter Ϫ1 lactose (Merck). The working volume was 100 ml, the temperature was 30°C, and the starting pH was 6.5. At harvest, the broth was mixed with glycerol solution to a final volume ratio of 15% (vol/vol), divided into vials of 1.0 ml, and stored at -50°C. For each cultivation condition, a new glycerol stock vial was used to inoculate static precultures containing M17 medium and 5 g liter Ϫ1 lactose in a working volume of 100 ml, at a temperature of 30°C and starting pH of 6.5. Cells were harvested in the late growth phase before lactose depletion and added to the bioreactor to achieve an initial optical density at 600 nm (OD 600 ) of 0.06.
All fermentations were carried out in a 1.5-liter Sartorius Biostat B twin tower with a 1-liter working volume. Initially, 0.5 ml liter Ϫ1 antifoam (suppressor 3130; Hydrite) was added. The temperature was 30°C and the pH set point was 6.0, adjusted with 24% ammonium hydroxide. Anaerobic conditions were established by continuously flushing the headspace with nitrogen gas at 0.1 liter min Ϫ1 at an agitation rate of 300 rpm. Under respiration-permissive conditions, the bioreactors were sparged with air. A minimum threshold of 60% dissolved oxygen was ensured by using cascade control, i.e., gradually increasing the stirrer speed and the aeration rate using Sartorius control software. The initial airflow was set to 0.5 liter min Ϫ1 , and the agitation rate was 400 rpm. The outlet gas was monitored with a Prima BT benchtop mass spectrometer (Thermo Fischer Scientific).
All chemostat fermentations were initially run in batch mode. Batch operation was switched to chemostat mode at the late exponential growth phase before lactose depletion, using the same cultivation conditions. We aimed to achieve the three dilution rates, 0.1 h Ϫ1 , 0.5 h Ϫ1 , and 0.8 h Ϫ1 , for both anaerobic and respiration-permissive conditions. However, the actual dilution rates were 0.11 h Ϫ1 , 0.52 h Ϫ1 , 0.74 h Ϫ1 , and 0.78 h Ϫ1 for anaerobic conditions. For respiration-permissive conditions, the actual dilution rates were 0.11 h Ϫ1 , 0.52 h Ϫ1 , and 0.8 h Ϫ1 . Steady states were verified after a minimum of eight residence times by repeated measurements of CO 2 , OD 600 , and metabolites. All fermentations were performed in biological duplicates, and each steady state originated from an independent preculture and chemostat startup. The results given are based on two measurements at each steady state and each biological replicate, separated by at least one residence time.
(iii) Sampling. Steady-state sampling was performed by withdrawing approximately 15 ml broth and transferring it directly to a 50-ml Nunc tube containing stainless steel beads (40 g, 4 mm diameter) precooled to -30°C. The sample did not freeze and was quickly distributed to more analysis tubes (37).