The Inuence of Long and Short Cycle Schemes of Self-Cycling Fermentation on the Growth of E. Coli and S. Cerevisiae

Self-cycling fermentation (SCF), a cyclic process in which cells divide once per cycle, has been shown to lead to improvements in productivity during bioconversion and, often, whole-culture synchronization. Previous studies have found that in some cases, the completion of synchronized cell replication occurred simultaneously with depletion of a limiting nutrient. However, exceptions were also observed when the end of cell doubling occurred before the exhaustion of the limiting nutrient. In order to better understand the underlying mechanisms and impacts of these growth patterns on bioprocessing, we investigated the growth of Escherichia coli and Saccharomyces cerevisiae in long- and short-cycle SCF strategies. Three characteristic events linked to SCF cycles were identied: (1) the completion of synchronized cell replication, (2) the depletion or a plateau of the limiting nutrient, and (3) characteristic points of control parameters (e.g., the minimum of dissolved oxygen and the maximum of carbon dioxide evolution rate). Three major trends stemming from this study and SCF literature were observed: (A) co-occurrence of the three key events in SCF cycles, (B) cycles for which cell replication ended prior to the co-occurrence of the other two events, and (C) cycles for which the time of depletion or a plateau of the limiting nutrient occurred later than the concurrence of the other two events. Based on these observations, a novel denition for SCF is proposed.

Introduction SCF and continuous phasing (Dawson 1965(Dawson , 1972Sheppard andCooper 1990, 1991), its forebearer, share many similarities. One is the entrainment mechanism that describes the periodic availability of essential nutrients inducing synchronization (Sheppard and Dawson 1999). A sharp increase in cell count within a narrow time window during a given cycle was observed for both continuous phasing and SCF (Sheppard and Dawson 1999). Flow cytometry validated that synchrony had been achieved during continuous phasing of bacteria by the determination of DNA content and cell sizes (Fritsch et al. 2005).
Transcriptomic patterns were used to con rm that synchrony was obtained during SCF of yeast (Tan et al. 2022). Many relevant trends could be observed in the prior study, including that most genes related to DNA replication and half of the genes associated with the yeast cell cycle were signi cantly up-regulated at the same point during the early SCF cycles (Tan et al. 2022).
The incorporation of a feedback control system in SCF is a major difference and improvement compared to continuous phasing. It prevents either starvation or washout, and triggers medium replacement precisely at the exhaustion of the limiting nutrient (Brown 2001). The decisive in uence of cell growth on cycle time is paramount to SCF operation. In a large number of studies implementing the SCF technique, the cycle time was found to be equal to the doubling time of the same microorganism growing under the same nutrient conditions (Sheppard and Cooper 1990;Brown and Cooper 1991;Sarkis and Cooper 1994; McCaffrey and Cooper 1995; Wentworth and Cooper 1996;Brown et al. 1999). Subsequently, SCF cycle time was used to re ect the nutrient quality of the environment in a number of physiological studies; wherein a shorter cycle time suggested more e cient cell replication and thus a more bene cial nutrient condition (Brown 2001). was also found for other microorganisms (Sheppard 1993;Marchessault and Sheppard 1997;Crosman et al. 2002). However, the trends in the control parameters could be different (Sheppard 1993; Marchessault and Sheppard 1997). With these discrepancies in mind, we show that an SCF short cycle strategy, cycling when the maximum in CER occurred, could lead to stable cyclic operation of E. coli and S. cerevisiae with enhanced volumetric biomass productivity. Also, transcriptional shifts of selected cyclin genes were explored during S. cerevisiae short cycles. These results con rmed the time at which cell replication of these microbes was completed and led to the identi cation of three typical trends during all SCF scenarios. In trend A, the end of synchronized cell replication, the exhaustion of the limiting nutrient, and the characteristic points of control parameters (three characteristic events) occur concomitantly. In Trend B, cell doubling completes before the co-occurrence of the other two events. Finally, in Trend C, the limiting nutrient is depleted or reaches a plateau after the joint occurrence of the other two events. The results from this study enable a better understanding of the cellular processes during SCF and can guide the development of more e cient bioconversion processes.

Results
E. coli grown in BR and SCF operation E. coli MG1655 was grown in a batch set-up for over 24 h. Two CER local minima were observed, one at 2 h and another at 7.5 h (Fig. 1A). These were likely due to either metabolic stalling or transitions between metabolic regimes. OD 600 reached its maximum at 7.5 h, as CER reached a local minimum, and remained at approximately 2.5 for the remainder of the experiment (Fig. 1B). Glucose was consumed rapidly until 9 h, corresponding to the second CER maximum, after which point it remained relatively constant.
Glucose, which was expected to be the limiting nutrient, was not exhausted even after an extension of the experiment well into stationary phase. Therefore, the abundance of nitrogen, in the form of ammonium, nitrate and nitrite, was examined; nitrogen compounds were found to remain abundant during the entire batch operation (Fig. 1C). Additional calcium and iron, and yeast extract were added to fresh media in independent experiments; these approaches did not result in glucose exhaustion after batch operation (data not shown). Hence, the cessation of glucose consumption was likely not due to a deprivation of a limiting nutrient but more likely due to an inhibitory effect of excessive intermediate metabolites. For example, organic acid overproduction could impact growth. However, pH only decreased from 6.76 to below 6 during operation, which is unlikely to cause cessation of growth by itself.
E. coli MG1655 was then grown under SCF operation. The rst 26 cycles were run in an SCF long cycle scheme, while the succeeding 10 cycles were operated under a short cycle scheme. The transition from long to short cycles occurred within cycle 27, which could be regarded as a disrupted long cycle and was not considered a cycle in either mode of operation. In the long cycle scheme, cycles were triggered when the decreasing trend in CER attened, while in the short cycle operation, cycling was triggered when CER reached its maximum. For SCF long cycles, a repeatable pattern of CER was established directly after the second cycle -an increase in the early stages of a cycle, a decrease during the late campaign, and a plunge upon cycling ( Fig. 2A). The cycle time during long cycle operation averaged 4.54 ± 0.32 h for cycles 3 to 26 ( Fig. 2A). When the operation was tuned to the short cycle scheme, readaptation occurred within the rst two cycles after the transition cycle, and a new stable pattern of CER was obtained after it (Fig. 2C). The new CER pattern only displayed a decrease in CER upon cycling, and the cycle time was reduced to 1.43 ± 0.06 h for short cycles 3 to 10 (Fig. 2C). Meanwhile, the CER maximum increased from 0.0026 mol/L/h to 0.0036 mol/L/h and the average CER (integrated CER per cycle time) signi cantly increased once long cycles were switched to short cycles. Moreover, the stability of SCF long cycle and short cycle operation demonstrated that cells completed one round of cell replication per cycle, regardless of the cycling regime.
The increase in OD 600 and the decrease in glucose concentration were essentially linear towards the end of the cycling regimes represented by long cycle 24 and short cycle 10 ( Fig. 2B and 2D). Long cycle 24 accumulated more biomass, consistent with a greater amount of glucose consumed, over its signi cantly greater cycle time, as compared to short cycle 10 ( Fig. 2B and 2D). The yield of E. coli biomass from glucose was 0.34 L•g glucose -1 during long cycle 24 and was 0.63 L•g glucose -1 for short cycle 10. The productivity of E. coli biomass was 0.15 h -1 and 0.42 h -1 for the long and short cycles, respectively. The glucose consumption rate was also improved in short cycle 10 ( Fig. 2B and 2D). By adjusting to a short cycle scheme, the yield of E. coli cells from glucose increased 1.8-fold, and volumetric biomass productivity was improved 2.7-fold (both relative to the long cycle scheme).

S. cerevisiae grown in SCF operation
Similarly to E. coli, S. cerevisiae was cultivated under SCF long cycle and short cycle schemes; the long cycle operation cycled when the decreasing CER attened, whereas the short cycle scheme triggered cycling upon reaching a maximum in CER ( Fig. 3A and 3D). Both modes of operation were found to be highly stable and reproducible directly after their rst cycles considering their CER pro les ( Fig. 3A and 3D), suggesting that S. cerevisiae cells completed one generation of cell proliferation within a long or short cycle time. However, similarly to E. coli experiments, there was a signi cant difference regarding the shapes of CER curves -no CER decrease was observed during short cycles. CER maxima increased from 0.009 mol/L/h to 0.011 mol/L/h between long and short cycle operation; the average CER (integrated CER per cycle time) also signi cantly increased ( Fig. 3A and 3D). Cycle times were signi cantly different, with an average cycle time of 12.11 ± 0.73 h for long cycles 2 to 16 and 3.80 ± 0.27 h for short cycles 2 to 20 ( Fig. 3A and 3D).  Table 1). The glucose consumption rate was also found to be greater in short cycle 20 ( Fig. 3C and 3F).
However, biomass yield was comparable between the long and short cycles -0.22 and 0.21 L•g glucose -1 , respectively ( Fig. 3C and 3F; Table 1). Additionally, the nitrogen source, ammonium, was always in excess and was not the limiting nutrient (data not shown).
Relative expression levels of selected cyclin genes in S. cerevisiae Relative expression levels (fold changes) of CLN1, CLN2, CLB3, CLB1, and CLB2, were determined for S. cerevisiae growing in SCF short cycles 1 and 21 using qPCR ( Fig. 4A and 4B, respectively). Generally, fold changes of the cyclin genes during BR (short cycle 1) late-log phase were not signi cant: slight decreases in expression were found for CLB1 and CLB2, and slight increases in expression were observed for CLN1 and CLN2 over the cycle (Fig. 4A). In contrast, during short cycle 21 (with a cycle time of 3.7 h), the following observations were made: (1) CLB1 and CLB2 (paralog genes) were signi cantly up-regulated during the early stages of the short cycle up to a cycle time of 2.8 h, with peaks in expression at approximately 1.4 h; (2) the expression of CLN1 and CLN2 (paralog genes) was generally stable but showed signi cant up-regulation at 2.8 h; and (3) in comparison, CLB3 transcription remained relatively steady throughout the short cycle (Fig. 4B). Identical trends in relative expression levels of the cyclin genes were observed in replicated short cycle experiments ( Fig. S1A and S1B for SCF short cycles 1 and 21, respectively).
Considering the signi cant sequential up-regulation of these cyclin genes, it can be established that some extent of synchrony was achieved during the short cycle operation, even though this could not be directly corroborated using cell counts due to occulation of the yeast cells (data not shown). During a standard yeast cell cycle, CLN1 and CLN2 are up-regulated prior to CLB1 and CLB2; CLN1 and CLN2 are expressed in G1 and S phases while CLB1 and CLB2 during M phase (Fitch et al. 1992;Futcher 1996;Cho et al. 1998;Feldmann 2012). However, here in the short cycle 21, the up-regulation of CLB1 and CLB2 was shown to be earlier than that of CLN1 and CLN2. This indicated that partially synchronized (if not completely synchronized) cell replication started and ended in the middle of the short cycles, rather than being aligned with the SCF cycle progression.

E. coli BR Operation
During E. coli BR operation, two major CER peaks could be observed (Fig. 1A), which is consistent with a prior E. coli study ). During the current study, nitrogen, calcium, iron, and yeast extract were not shown to limit the growth. Little glucose was consumed during the late stages of BR and it was not depleted by the end of BR (Fig. 1B). The restricted growth was most likely due to the accumulation of organic acids (e.g., acetic acid), resulting from bacterial Crabtree effect (Mustea and Muresian 1967), rather than the depletion of a limiting nutrient. Excessive organic acids might be produced from rapid growth in the early stages of BR, partly inhibiting the subsequent stages of growth and preventing the full consumption of glucose. Some future investigations can be carried out to identify the mechanisms of this inhibition and eliminate barriers towards complete consumption of glucose.

SCF Long Cycle and Short Cycle Operation
Much like the case for E. coli BR operation, E. coli SCF long cycles did not exhaust glucose when CER attened (cycling condition) (Fig. 2B). This was not the case for S. cerevisiae undergoing SCF operation, wherein glucose depletion was observed at the end of BR and long cycles ( Fig. 3B and 3C). As discussed earlier, the accumulation of organic acids might have an inhibitory effect on E. coli growth. Previous work investigating E. coli undergoing SCF long cycles  found that glucose was totally consumed by the time CER stopped decreasing; however, while that study was performed with similar nutrient conditions, it used a different strain of E. coli (ATCC 11303) ).
The improvement in production of E. coli biomass was signi cant once SCF operation was adjusted to a short cycle scheme; 1.8-fold and 2.7-fold increases in yield and volumetric productivity, respectively (Table 1). In the previous study investigating SCF long cycles with E. coli ATCC 11303 ) the yield was found to be 0.23 L•g glucose -1 , and the productivity of E. coli cells was 0.28 h -1 (also shown in Table 1; calculated based on an original gure (Sauvageau et al. 2010) using Eq. 1 and 2). In comparison, the yield and volumetric productivity during the current short cycle operation also prevailed. As for S. cerevisiae cells grown in SCF operation, short cycles led to a 1.6-fold increase in volumetric productivity while providing a similar yield to long cycles (Table 1). It was also noted that the average glucose consumption rate and average CER were enhanced during short cycles of both E. coli and S. cerevisiae, despite lower cell density ( Fig. 2 and 3) -meaning more glucose was consumed per cell and more CO 2 was released per cell. That is to say, cellular activity was generally more intense during SCF short cycles. Signi cant improvements in yield, productivity, and metabolic activity highlight the advantages of the SCF short cycle scheme as compared to the long cycle counterpart. Yield and productivity would likely be further improved if the SCF short cycle strategy were used in the context of recent E. coli and S. cerevisiae SCF studies, such as E. coli biomass production ( Instantaneous speci c productivity of bacteriophage, recombinant protein, and shikimic acid was found to be optimal near the completion of synchronized cell replication (corresponding to a maximum in CER) during SCF long cycle operation (Storms et al. 2012(Storms et al. , 2014Agustin 2015).
In speci c, the results observed in this study for S. cerevisiae SCF long cycle operation generally agree with a preceding study (Tan et al. 2022). Growing the engineered yeast strain in SCF long cycles had substantially improved shikimic acid yield and productivity compared to BR (Tan et al. 2022). Hence, it may be possible to further enhance shikimic acid production by implementing the SCF short cycle strategy. Based on the improved productivity of cellular biomass (Table 1), the production rate of the primary metabolite will highly likely be further reinforced in SCF short cycles.
The incomplete depletion of glucose during short cycles might be the only concern for the implementation of this SCF scheme. To address this, recycling of the carbon source might be considered in future studies.
A close link between CER maximum and the completion of synchronized cell division can be established.  2015)), step-wise doublings in cell count were observed at the CER maxima. This suggested that cell replication was in unison, and synchronized cell doubling was completed when CER reached its maximum. Thirdly, signi cant upregulation of DNA replication-related genes and selected cyclin genes (CLN1, CLN2, CLB3, CLB1, and CLB2) were observed only during the rst-half of S. cerevisiae SCF long cycles (Tan et al. 2022). It is very likely that there was little replication activity after the maximum in CER. Fourthly, the cycle time of SCF long cycles was signi cantly longer than the expected doubling time in the same nutrient conditions. In contrast, the duration of short cycles was very close to those doubling times.
As to S. cerevisiae SCF short cycles, the expression pro les of the investigated cyclin genes provided valuable information (Fig. 4). Firstly, the amplitudes and sequential changes in up-regulation suggested that a certain level of cell synchrony was established. Cyclical changes in glucose concentration established by the mode of operation provided a forcing function to induce the entrainment effect required for cell synchrony; greater glucose availability early in the SCF cycles preferentially favored some stages of the cell cycle. It was also noted that, compared to SCF long cycles, a decrease in up-regulation amplitudes was observed during short cycles ( Fig. 4B and Tan et al. 2022). This was likely due to the incomplete utilization of glucose, resulting in a tamer entrainment effect during short cycles. Secondly, the sequence of cyclin genes expression suggested that the replication of partially synchronized cells started from the middle of the short cycles and was completed at the same point in the subsequent cycle. CLN1 and CLN2 were expressed later than CLB1 and CLB2 during SCF short cycles (Fig. 4B) -an inverse sequence compared to the standard yeast cell cycle (Fitch et al. 1992;Cho et al. 1998). This unexpected, distinct, cycle-spanning cell replication pattern in short cycles was likely caused by other driving forces apart from the oscillation of glucose concentration, as the nutrient cycle itself is expected to lead to an alignment between the starts of SCF and cell cycles.
This leads to another question -might synchronized yeast cell replication also present a cycle-spanning pattern during SCF long cycles, starting from the middle of a long cycle and ending at the same point in the succeeding long cycle? The answer should be no. One reason is that there was no signi cant expression of selected cyclin genes during the second-half of long cycles (Tan et al. 2022). Hence, there was hardly any replication activity during the late stages of long cycles. Moreover, the onsets of SCF long cycles and the yeast cell cycle were aligned, as suggested by the congruent expression sequence of the cyclin genes during the rst half of long cycles and the standard yeast cell cycle. Furthermore, the cycle time of long cycles was more than twice the doubling time of S. cerevisiae in the same nutrient conditions. It is unlikely that one round of cell replication occurred throughout the whole cycle time of long cycles.
Identical trends and amplitudes of cyclin gene expression during BR late-log phase were identi ed not only between replicate experiments in the present study ( Fig. 4A and S2A) but also amongst current qPCR results, previous qPCR results, and previous RNA-Seq results (Tan et al. 2022). This great alignment of the gene expression pro les across different studies and analytical techniques signi cantly increases con dence in the trends observed. Consequently, relative quanti cation results were considered to truly re ect transcriptional changes during SCF short cycles (Fig. 4B).

An Overview of Characteristic Events in SCF
Signi cant differences in the occurrence of some SCF key events can be observed among SCF studies with different microorganisms implemented. These characteristic events include: (1) the time at which the limiting nutrient was depleted or reached a plateau, (2) the characteristic values of control parameters, and (3) the completion of one generation of synchronized cell division.
Nitrogen or carbon sources are frequently set as the limiting nutrients dictating the cycling of SCF operation. Control parameters used to establish cycling conditions have included dissolved oxygen (DO), carbon dioxide evolution rate (CER), or oxidation-reduction potential (ORP) (Brown 2001 ORP patterns during SCF operation are generally more complex than other parameters. An increasing trend in ORP was observed in the toluene removal study using P. putida ATCC 12633 (Brown et al. 2000), while a decreasing trend was shown in the removal of oxidized nitrogen using Pseudomonas denitri cans ATCC 13867 (Brown et al. 1999). The presence and absence of oxygen in these two studies were likely responsible for these diverging patterns. In contrast, DO and CER generally present similar patterns amongst various studies and hence have been more often applied as the control parameter.
Overall, it is illustrated that DO minimum would coincide with CER maximum during SCF operation, and the in ection point of ORP is likely close to this point. In the present study, the time at which this event occurs is referred to as the control parameters' "characteristic values" or "characteristic points".
In many SCF studies published before 2010, co-occurrence was always identi ed for the depletion of a limiting nutrient and the characteristic values of control parameters. SCF cycling was triggered upon this concurrence unless an extended cycle strategy was applied. In antibiotic production using Streptomyces aureofaciens ATCC 12416c (Zenaitis and Cooper 1994), phenol degradation using P. putida ATCC 12633 (Hughes and Cooper 1996), and cultivating Bacillus subtilis ATCC 21332 (Sheppard and Cooper 1991), DO minimum occurred concomitantly with nitrogen depletion or the complete removal of phenol. Moreover, the co-occurrence of the completion of cell replication with the aforementioned two key events was observed in a wealth of SCF studies, and this is summarized as Trend A in Fig. 5A. Moreover, during polyhydroxybutyrate (PHB) production using Alcaligenes eutrophus DSM 545 (Marchessault and Sheppard 1997) and the growth of B. subtilis ATCC 10774 (Sheppard 1993), while the minimum in DO and the nitrogen source depletion coincided, synchronized cell replication was completed much earlier -in the middle of the SCF cycles. As the timing of the cell cycle end point showed a signi cant difference, these studies serve as representations for Trend B in Fig. 5B. In summary, for most of the microbial systems used in earlier SCF studies, the depletion of a limiting nutrient and characteristic values of control parameters (DO minimum, CER maximum, or ORP in ection point) occurred concurrently at the end of each SCF cycle (Trends A and B in Fig. 5A and 5B). One round of synchronous cell doubling was completed at the same time (Trend A in Fig. 5A) or, in some instances, in the middle of the cycles (Trend B in Fig. 5B). (re ected by exit gas mass ow rate in anerobic conditions), though cell counts were not reported due to clumping of the yeast cells. As mentioned earlier, the same trend was observed in the present study when cultivating E. coli MG1655 or engineered S. cerevisiae CEN.PK 113-1A Matα (except that, for E. coli undergoing SCF long cycles, the end of cycle was due to an inhibitory effect rather than glucose depletion). Trend C in Fig. 5C is used to describe the pattern observed in these studies.

Compared to Trends
Moreover, transcriptional evidence during S. cerevisiae SCF short cycles presented in this study revealed a likely cell replication pattern in short cycles -partially synchronized cell cycle starting and ending in the middle of short cycles. This cycle-spanning mode of cell replication is presented by the black dashed line in Fig. 5C. Overall, in studies displaying Trend C, the attening of CER decrease or DO increase coincided with the depletion or a plateau of the limiting nutrient at the end of SCF long cycles, but synchronized cell replication was completed in the middle of the long cycles, corresponding to CER maximum or DO minimum. SCF short cycles ended at CER maximum or DO minimum, but partially synchronized cell replication likely started and ended in the middle of the short cycles. The limiting nutrients were not depleted by the end of the SCF short cycles.
The discrepancies amongst the three major trends in the characteristic events during SCF operation were likely derived from intrinsic differences in the microorganisms and nutrient environments used. A. eutrophus and B. subtilis ATCC 10774 following Trend B, and C. alkanolyticum, E. coli, and S. cerevisiae following Trend C likely sensed nutrient conditions more actively and adopted a feed-forward strategy (Levy and Barkai 2009) -in which cells proactively sensed external changes and regulated gene transcription and expression prior to the alteration of the growth rate (Levy and Barkai 2009). Completing one generation of the synchronized cell cycle but deciding not to continue the proliferation at the expense of the remaining limiting nutrient seemed to be the growth strategy of these microorganisms (Fig. 5B and  5C). On the contrary, for a number of microorganisms following Trend A, all the available limiting nutrient was used in completing the cell doubling (Fig. 5A). The difference between A. eutrophus and B. subtilis ATCC 10774 in Trend B, and C. alkanolyticum, E. coli, and S. cerevisiae in Trend C is expected to lie in the respiratory intensity between the end of the cell cycle and the time at which the limiting nutrient was depleted or reached a plateau. For the former group, the intensity of respiration increased even after synchronized cell replication. Therefore, the characteristic value of the control parameter (DO minimum) co-occurred with the exhaustion of the limiting nutrient but not with the end of cell doubling (Fig. 5B). For microbes displaying Trend C, respiration slowed signi cantly after synchronized cell replication (during the consumption of residual limiting nutrient), and therefore CER maximum or DO minimum occurred at the completion of synchronized cell doubling but not at the depletion or a plateau of the limiting nutrient Limiting nutrient depletion has been one of the original premises of SCF, but a broader picture is emerging. The Trend C observed in some studies suggests a deviation from the original description of SCF -SCF does not necessarily require limiting nutrient depletion. Consequently, a novel description of SCF is proposed below, taking into consideration all SCF scenarios presented in Fig. 5. This new SCF de nition excludes the requirements of limiting nutrient depletion and joint occurrence of the key events.
SCF is a semi-continuous fermentation approach that allows the completion of one generation of microbial cell replication during each cycle. The cycling procedure comprises harvesting precisely one half of the working volume and then replenishing with the equivalent amount of fresh medium. Cycling is dictated by microbial growth and metabolism and is triggered automatedly based on monitoring one or more growth-and/or metabolism-associated sensing elements (e.g., DO, CER, ORP, exit gas mass ow rate, etc.). SCF cycling takes place directly after the completion of one generation of cell proliferation or with a delay, depending on the microorganism, the initial nutrient conditions, and the control parameter conditions for cycling being implemented. SCF cycling is not necessarily related to the time at which the limiting nutrient is depleted or reaches a plateau. If limiting nutrient depletion or a plateau does not cooccur with the cell cycle completion, we identify SCF operation that cycles in advance of exhaustion or a plateau of the limiting nutrient as "short cycle", and correspondingly, SCF operation that cycles upon depletion or a plateau of the limiting nutrient as "long cycle" (Fig. 5). "Extended cycle" is generally referred to as SCF operation that cycles beyond exhaustion or a plateau of the limiting nutrient (Fig. 5).

Conclusions
Previous SCF operation of E. coli and S. cerevisiae triggered cycling upon glucose depletion when the CER attened. In the present study, SCF short cycle operation of E. coli and S. cerevisiae were cycled at the maximum in CER and led to stable and reproducible short cycles. Notably, compared to SCF long cycles, volumetric biomass productivity was signi cantly improved during SCF short cycles. Further transcriptional analysis for selected S. cerevisiae cyclin genes inferred a cycle-spanning mode of cell replication during SCF short cycles. Viable SCF short cycles also helped identify the maximum in CER as the endpoint of cell replication during long cycles. Moreover, a thorough overview of previous SCF studies summarized the occurrence of three SCF characteristic events, (1) the completion of synchronized cell replication, (2) the depletion or a plateau of the limiting nutrient, and (3) the characteristic points of control parameters, into three typical trends. A novel description of SCF was then proposed to include all scenarios of SCF operation and clear de nitions for SCF "short cycle", "long cycle", and "extended cycle".
This work highlights a diversity in SCF operation and shows the potential of SCF as a research tool to explore microbial physiological properties -including nutrient use, proliferation strategies, and respiration intensity. It also demonstrates that short cycle schemes, in particular, can be used to improve performance of bioconversion. With trends in the key events summarized and the establishment of a clear de nition of SCF, the present work consolidates and deepens our understanding of the SCF technique and its in uences on microbial populations. Finally, it provides a solid framework to guide the further design and implementation of SCF-based processes.

Methods/experimental
Strains, Media, and Pre-cultures Escherichia coli MG1655 (CGSC 6300) was used in E. coli studies. Luria-Bertani (LB) broth (all chemicals used in this study were purchased from Fisher Scienti c and Sigma Aldrich, Canada) was used for agar plates (1.5% w/v of agar). Semi-de ned liquid medium was used in Erlenmeyer asks or fermenters, containing 6 g/L sodium phosphate dibasic, 4 g/L ammonium nitrate, 4 g/L potassium phosphate monobasic, 0.014 g/L disodium EDTA, 0.05 g/L yeast extract, 0.01 g/L calcium chloride dihydrate, 0.01 g/L iron sulfate heptahydrate, 6 g/L glucose, and 0.2 g/L magnesium sulfate heptahydrate. Pre-cultures were grown in 250-mL Erlenmeyer asks at 37 • C, 250 rpm for 12 h. Approximately 4×10 10 cells (10 mL) were withdrawn from pre-cultures and used to inoculate the 1-L fermenter working volume to achieve 1% v/v inoculation. Approximately 8×10 8 cells (10 mL) from pre-cultures were added to the 1-L working volume in the fermenter to achieve 1% v/v inoculation.

SCF Con guration and Operation
The SCF con guration was previously described in (Sauvageau et  The fermenter temperature was maintained at 37°C during bacterial growth and 30°C during yeast growth. Agitation at 250 rpm with a Rushton impeller (4-cm diameter) and aeration at 400 mL/min for E. coli and 845 mL/min for S. cerevisiae provided su cient mixing and aerobic conditions. During the SCF cycling procedure, agitation was ceased to maintain liquid level stability. Cell culture drainage driven by gravity stopped when the liquid level reached the low-level sensor. Fresh medium was then pumped into the bioreactor until the 1-L working volume was reached.
The following conditions were used to trigger automated cycling. For the E. coli SCF long cycle operation: (1) cycle time was greater than 90 min; (2) the absolute value of short dCER was less than 0.02 ppm/min; (3) long dCER was less than 0. For the E. coli SCF short cycle scheme: (1) cycle time was greater than 60 min; (2) short dCER was less than -0.02 ppm/min. For the S. cerevisiae SCF long cycle scheme: (1) cycle time was greater than 300 min; (2) CER was less than 3000 ppm; (3) the absolute value of short dCER was less than 0.05 ppm/min; (4) long dCER was less than 0. For the S. cerevisiae SCF short cycle strategy: (1) cycle time was greater than 110 min; (2) short dCER was less than -0.02 ppm/min; (3) CER was more than 3000 ppm. Batch Reactor Con guration and Operation Figure S2 depicts the BR set-up, which was adapted from the SCF set-up and used when BR operation was decoupled from SCF operation in cultivating E. coli. Cultivation conditions during BR were congruent with those used for SCF operation. Additionally, the rst cycles of SCF operation are analogues of BR.
Measurement of Optical Density, Glucose, Ammonium, and Nitrate & Nitrite A spectrophotometer (Ultrospec 50, Biochrom) was used to measure optical density of culture samples at a wavelength of 600 nm (OD 600 ).
Glucose concentration was determined using the reducing sugar method (Miller 1959). Dinitrosalicylic acid (DNS) reagent was used, containing 10 g dinitrosalicyclic acid, 2 g phenol, 0.5 g sodium sul te, and 10 g sodium hydroxide in 1 L deionized water. 20 µL of the ltered samples were mixed with 140 µL of DNS reagent, followed by a 5-min incubation at 95°C. Samples were then cooled on ice for 5 min to stop the reactions. After that, 840 µL of deionized water was added. Samples were nally measured through a spectrophotometer (Ultrospec 50, Biochrom) set to a wavelength of 540 nm. A standard curve was established based on standards and used for quanti cation.
Nitrogen measurement followed methods detailed in (Bollmann et al. 2011). To measure ammonium, a solution of 12 g/L of sodium hydroxide was mixed with another containing 85 g/L sodium salicylate and 0.6 g/L sodium nitroprusside at a 2:1 volume ratio. 375 µL of this freshly prepared mixture was added to 750 µL of every sample. 150 µL of 0.2 g/L sodium dichloroisocyanurate was then added, followed by 30 min of incubation in a dark environment. After incubation, absorbance of samples was measured at 660 nm using a spectrophotometer (Ultrospec 50, Biochrom). A standard curve was established based on standards. The concentration of ammonium in samples was determined based on the standard curve. To measure nitrate and nitrite, 75 µL of a catalyst solution containing 35.4 mg/L copper sulfate pentahydrate and 0.9 g/L zinc sulfate monohydrate was added to 500 µL of every sample. Then, 75 µL of 40 g/L sodium hydroxide and 75 µL of 1.71 g/L hydrazine sulfate were added sequentially, and samples were incubated in the dark for 15 min. After incubation, 250 µL of 10 g/L sulfanilamide dissolved in 3.5 M hydrochloric acid and 75 µL of 1 g/L naphthylethylene diamine dichloride were added sequentially, and samples were incubated in the dark for an additional 10 min. Samples were nally assessed by measuring absorbance at 540 nm using a spectrophotometer (Ultrospec 50, Biochrom). A standard curve was established based on standard solutions and used for quanti cation.

Calculation of Yield and Productivity for Biomass Production
Eq. 1 and 2 were used to calculate the yield and productivity in the production of E. coli or S. cerevisiae cells.
Y X/S is the yield of E. coli and S. cerevisiae biomass (assessed by OD 600 ) on glucose in L•g glucose −1 . r P represents the volumetric productivity of E. coli and S. cerevisiae biomass (assessed by OD 600 ) in h −1 .

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.