Microfluidic time‐lapse analysis and reevaluation of the Bacillus subtilis cell cycle

Abstract Recent studies taking advantage of automated single‐cell time‐lapse analysis have reignited interest in the bacterial cell cycle. Several studies have highlighted alternative models, such as Sizer and Adder, which differ essentially in relation to whether cells can measure their present size or their amount of growth since birth. Most of the recent work has been done with Escherichia coli. We set out to study the well‐characterized Gram‐positive bacterium, Bacillus subtilis, at the single‐cell level, using an accurate fluorescent marker for division as well as a marker for completion of chromosome replication. Our results are consistent with the Adder model in both fast and slow growth conditions tested, and with Sizer but only at the slower growth rate. We also find that cell size variation arises not only from the expected variation in size at division but also that division site offset from mid‐cell contributes to a significant degree. Finally, although traditional cell cycle models imply a strong connection between the termination of a round of replication and subsequent division, we find that at the single‐cell level these events are largely disconnected.


| INTRODUC TI ON
The cell division cycle is one of the most extensively studied processes in biology. In bacteria, the classic view was established in the 1950s and 1960s, based largely on studies of Escherichia coli (Cooper & Helmstetter, 1968;Donachie, 1968;Kubitschek, 1966Kubitschek, , 1968Kubitschek, , 1969Perry, 1959) but thought generally to be similar in other symmetrically dividing rod-shaped bacteria (e.g., Bacillus subtilis; Sharpe, Hauser, Sharpe, & Errington, 1998). The scheme is summarized briefly in Figure 1. In rod-shaped organisms cell length is roughly proportional to cell mass or volume and because length is convenient to measure (Donachie, 1968;Grover, Woldringh, Zaritsky, & Rosenberger, 1977), critical moments in the cell cycle tend to be defined based on this (e.g., Sharpe et al., 1998;Taheri-Araghi et al., 2015;Sauls, Li, & Jun, 2016;Zheng et al., 2016). For any given growth condition, the newborn cell has a length L b . The cell grows exponentially at a rate proportional to its length. At a critical length L i , corresponding to the "initiation mass" (Donachie, 1968), the cell initiates a round of chromosome replication. After a fixed period of time (the C period), which was thought largely to be independent of growth rate, the round of chromosome replication terminates, at which time the cell has reached a length L t . Cell division follows, a fixed period of time later (called the D period; or D* -see below) at a length L d , approximately twice that of L b . This model, often called the Cooper-Helmstetter model (Cooper & Helmstetter, 1968), satisfied many observed features of the E. coli cell cycle, particularly changes in average cell size according to growth rate (faster growing cells tend to be larger than slow growing cells) (Cullum & Vicente, 1978). Its central assumptions included the ability of the cell to sense the initiation mass and dependence of division timing on constancy of the C and D periods. Note that, although Figure 1 shows a simple cell cycle representative of slow growing cells, at faster growth rates, initiation of chromosome replication occurs prior to the previous cell division, so that fast growing cells can contain multiple chromosome origins.
The early experiments providing the evidence that led to the Cooper-Helmstetter model were mainly based on population studies, in which the average behavior of cells was measured (Cooper & Helmstetter, 1968). In the last few years, the advent of accurate, automated time-lapse microscopy has enabled the model to be tested at the level of individual cell behavior (Campos et al., 2014;Potvin-Trottier, Luro, & Paulsson, 2018;Sauls et al., 2016;Taheri-Araghi et al., 2015;Wallden, Fange, Lundius, Baltekin, & Elf, 2016;Yu et al., 2017;Zheng et al., 2016). Surprisingly, these experiments have led to a reevaluation of cell cycle behavior. The classical model falls into a class of models sometimes called "Sizer," in which the cell can sense its mass and use this information to control cell cycle events. An interesting alternative to this model, which has received considerable attention recently, is sometimes called "Incremental" or "Adder" and is based on the notion that the cell is oblivious to its starting mass but instead divides after the addition of a constant amount of new material (Amir, 2014;Campos et al., 2014;Fantes & Nurse, 1981;Modi, Vargas-Garcia, Ghusinga, & Singh, 2017;Sauls et al., 2016).
The two models both assume that the cell can somehow "measure" mass -Sizer measures the mass appropriate for division and Adder the amount of new growth. Both models can account for population-based measures of cell cycle progression. A third model called Timer, assumes that cells grow for a relatively fixed period of time between divisions but this is not particularly effective at supporting size homeostasis.
As the cell separation time is quite variable in B. subtilis, depending both on growth conditions and cell to cell differences (Holmes et al., 1980;Nanninga et al., 1979), we previously defined a D* period, corresponding to the interval between termination of replication and membrane scission, which is relatively constant when measured at the population average level (Sharpe et al., 1998). Moreover, because the FtsZ-based division machine, which is almost universal in bacteria, operates during membrane scission rather than cell separation, D* is probably functionally equivalent to the D period of E. coli (Errington et al., 2003;Harry, 2001). The only report of time-lapse analysis on individual growing cells of B. subtilis, was based on phase contrast imaging (Taheri-Araghi et al., 2015), which detects cell separation rather than scission, and which we show are temporally separated events.
Here, we take advantage of an automated system for measurement of the growth and division of B. subtilis cells, over many generations, in an agarose-based microfluidic device (Eland, Wipat, Lee, Park, & Wu, 2016;Moffitt, Lee, & Cluzel, 2012). We have also developed fluorescent tools for measuring DNA replication and particularly the membrane steps of the cell cycle. We find that for two growth media, conferring different growth rates, the B. subtilis cycle tends to follow an Adder-like model, but that the accuracy of cell size homeostasis depends on the growth rate. We also report an unexpected contribution to cell birth size variation through division F I G U R E 1 Schematic view of the bacterial cell cycle. Blue ovals represent chromosomes. O and T represent, respectively, the origin and terminus sites for chromosome replication. The red dot indicates initiation or termination events. Note that in many bacteria growing rapidly, rounds of DNA replication can overlap, creating more complicated cell cycle patterns. Unlike Gram-negative bacteria, in which constriction at the division site and separation of sister cells occur more or less simultaneously, in Gram-positive bacteria, cells can remain connected together via common wall material in the division septum for a protracted and relatively variable period of time. We therefore previously defined the completion of septation in Bacillus subtilis as equivalent to division in Escherichia coli, and defined the period between completion of replication and septation as D* (Sharpe et al., 1998) asymmetry, despite the common assumption that E. coli and vegetative B. subtilis cells divide with considerable precision at mid-cell.
Finally, we have tested the dependence of cell division on the chromosome cycle and found, surprisingly, that in individual cells there is little or no apparent coupling between the termination of chromosome replication and cell division, so the nucleoid is unlikely to provide the cell mass or length "ruler" needed for an Adder model.

| Strains
Strains and plasmids used are listed in Table A2.
The WALP23 artificial transmembrane helix (sequence AW 2 L(AL) 8 W 2 A) was previously shown, when fused to mGFP, to provide a good general membrane label when expressed in B. subtilis (Scheinpflug et al., 2017). The Scheinpflug construct was under xylose-inducible control. To be able to express the fusion constitutively, we removed part of the xylose operator sequence from the original construct. First, nine nucleotides in the middle of the operator sequence (TTTGGGCAA) in plasmid pL015 were removed by sitedirected mutagenesis (SDM) using primers 5'-GATTAAAATAAGT TAGTTTGCAAACTAATGTGCAACTTACTTAC-3' and 5'-GTAAGTAA GTTGCACATTAGTTTGCAAACTAACTTATTTTAATC-3'. The resulting plasmid (pL062) was transformed into the B. subtilis wild-type strain 168CA, resulting in strain sL105. sL105 was subsequently transformed with plasmids pL006 (tetR-mCherry) and pCRW10 (dac-C::tetO at 171°, termination site) to generate strain sL099.

| Microscopy
Microscopy experiments were performed on an inverted fluorescence microscope (Nikon Eclipse Ti body with Perfect Focus) with F I G U R E 2 Stable growth of Bacillus subtilis in agarose-based microfluidics. A derivative of the wild type B. subtilis strain (sL105) was grown in SM (a, b) and FM (c, d) in agarose-based microfluidic channels at 32°C. Strains were recorded in SM for about 6 hr through 6 generations (n = 524) and in FM for about 6 hr through 11 generations (n = 568). (a, c) show plots of cell length at birth against elapsed time. Gray dots show each single-cell data point, and blue crosses and lines show the average of binned data and errors (standard deviation). (b, d) show histograms of cell length distribution at birth

| Agarose-based microfluidics
An agarose-based microfluidic system, consisted of a PDMS chamber that housed an agarose pad, a syringe pump and a microscope (Eland et al., 2016;Moffitt et al., 2012)  to the glass bottom of the microfluidic chamber, then an agarose pad was placed on the culture to trap cells in the channels on the agarose pad. The agarose pad, prepared using 5% low melting point agarose (Lonsza, SeaPlaque GTG Agarose) containing growth medium, had channels of three different widths (0.8 µm, 0.9 µm, and 1.0 µm wide). Subsequently, the top of the PDMS chamber was sealed by bonding to a plasma-cleaned plastic cover slip (Agar Scientific, plastic cover slips 22 × 22 mm). Then, through microbore tubing (Tygon, 0.020" × 0.060"OD), one of the two buffer reservoirs in the chamber was connected to a 50ml syringe that contained the growth medium and was controlled by a syringe pump (New Era Pump Systems Inc, NE-1000), and the other to a flask to collect spent medium and overgrown cells.
The microscope was enclosed in an incubator chamber (Solent Scientific) maintained at a constant temperature of 32°C (by allowing equilibration to temperature for at least 5 hr before the microfluidic device was introduced into the chamber). Cells grew along straight microfluidic channels molded into the agarose pad (Supporting Information S1, S2 and S3). Time-lapse imaging was initiated after 30 min to allow cells to acclimatize to the growth conditions.

| Image processing and analysis
Cell membrane and termination data from the microfluidic experiments were image-processed using FIJI-ImageJ (Schindelin et al., 2012). Images of cell membrane fluorescent tag were first processed to remove background signal, adjust contrast and brightness, and to apply edge processing. After that, the membrane signal was selected as a bold line, to provide a clear cell boundary. Following image processing, even unclear division septa were well defined (see Appendix Figure A1). Images of the replication termination fluorescent tag were also processed to remove background signal, adjust contrast and brightness, and apply gamma processing. Following processing, the termination spots could be seen clearly without blurred boundaries ( Figure 6a4).
Cell division time was recorded as the time difference between birth of mother and daughter cells. Cell division (membrane scission) was verified by the appearance of a clear, intact membrane ring at mid-cell ("0P" in Appendix Figure A1). Cell birth length

Cell length at birth (µm)
and termination length were measured using MicrobeTracker (Sliusarenko, Heinritz, Emonet, & Jacobs-Wagner, 2011) and verified, where necessary, by visual inspection. Because termini could move in three dimensions, two termini separated in the Z plane could be mistaken for an unreplicated terminus. We interpreted elongated or unusually bright foci in frames just preceding those with two clearly separated foci as containing two overlapping terminus foci, to partially compensate for delayed recognition of separation.

| Data processing
With the microfluidic images, we measured the cells collected over a series of generations to give a continuous cell pedigree. Recordings

| Long-term cell cycle analysis by time-lapse fluorescence imaging
As phase contrast microscopy does not detect septum formation with precision in B. subtilis (see above), we tested a range of fluorescence-based methods for ability to detect septal closure while minimizing phototoxicity and bleaching, to enable long-term timelapse imaging. We eventually fixed on a constitutively expressed monomeric GFP (mGFP) fusion to an artificial model transmembrane helix, WALP23-mGFP (Scheinpflug et al., 2017;see Experimental Procedures), which clearly labeled membranes and showed the progression of septation. A UV filter was used to reduce DNA damage, and a highly sensitive objective lens (TIRF ×60, NA1.49), which was approximately 30% more sensitive was used instead of the standard 100× lens. In some experiments, we also used a TetR-mCherry fusion targeted to the extreme terminus region of the chromosome, so that we could use focus duplication to time the completion of chromosome replication relative to cell division events. We also developed an automated system for tracking growth and division of individual cells through multiple generations. Two low background fluorescence media recipes were developed giving different growth rates, "Fast Medium" (FM) and "Slow Medium" (SM), while minimizing image capture times and thus phototoxic effects (Supporting Information S1, S2 and S3). In control experiments using bright field imaging only (i.e., no fluorescence irradiation) the measured growth rates were indistinguishable from those of the irradiated cultures.
Appendix Figure A1 shows that septation was detected by fluorescence and image processing several frames before it became visible by bright-field microscopy.

| Steady-state growth under microfluidic conditions
To check that the microfluidic conditions could support long-term steady-state growth of B. subtilis we grew the cells in FM or SM at 32°C and measured various cell cycle parameters over a prolonged time period (320 min in SM and 280 in FM), during which up to 6 (SM) and 11 (FM) sequential cell division events were captured (Appendix Table   A1). Figure 2a,c shows that steady-state growth over many hours was achieved under both media conditions. The average interdivision (i.e., generation or doubling) times were 25 ± 5 min for FM and 57 ± 11 min for SM. Average lengths at birth (<L b >) were 4.36 ± 1.02 µm and 3.13 ± 0.50 µm, respectively. Figure 2a,c also shows how these values varied during a long (6 hr) time-lapse experiment. Apart from a slight reduction in cell size toward the end of the experiment, especially in FM, growth parameters were well maintained. immediately, at the next division (Appendix Figure A2a). In contrast,

| Cell size homeostasis and the Adder model
Adder is oblivious to size at birth or division and works by allowing a fixed increment of growth, corresponding to the preferred (average) population size, before next division. This model predicts that size correction will occur over several generations, with a recurring "memory" of previous cell size (Appendix Figure A2b). In practice, both models need to be adjusted to take account of an inevitable level of stochastic variation in the precision of the timing and positioning (central or off-central) of division. Figure 3 plots the relationship between cell length at birth (L b ) and elapsed time to next division (interdivision time; ΔT), or added length (ΔL) for cells grown in FM or SM. In both media, the interdivision time was negatively correlated with the birth length (Figure 3a,c). This was expected, as short new born cells need to grow for a longer period than long newborn cells by any plausible homeostatic mechanism. For added length (Figure 3b, this as a mechanism for growth homeostasis at the faster growth rate.
Timer gave a poor fit to the data under both conditions, as anticipated.

| Imprecision in mid-cell division placement
The requirement for homeostatic mechanisms such as Adder or Sizer implies the existence of sources of variation in cell size. As shown in Figure 3, an important source of this variation is probably imprecision in division timing -the cell is longer or shorter than its "preferred" size (determined by Sizer or Adder effects) at division. However, in principle, it can also arise by variation in placement of the division septum. Some organisms, such as Caulobacter, have an intrinsically asymmetric division process (Campos et al., 2014) Figure A3 shows that the frequency of asymmetric division did not change appreciably during the 6-hr time course in the microfluidic channels.
The histograms in Figure 5c,d plot the frequency with which division was placed at different distances from mid-cell (ignoring, for now, the length of the dividing cell). Taking an arbitrary cutoff of 200 nm as the boundary between central and offset division (based on the approximate level of resolution in our image processing), it appeared that in both SM and FM a substantial proportion of divisions (20% and 50%, respectively) were detectably off center. Thus, by both measures, imprecision in division site placement appears to make a small but significant contribution to cell length variability.
We noted that Migocki, Freeman, Wake, and Harry (2002)   were consistent with this model (Sharpe et al., 1998). However, virtually all of the evidence for this coupling has been based on population studies. To test for coupling of termination and division in individual cells, we took the WALP23-mGFP strain and introduced a tetO/ mCherry-TetR system that would label a site close to the terminus of chromosome replication, similar to constructs used to measure terminus position previously (Bogush, Xenopoulos, & Piggot, 2007;Lemon, KURTSER, I., & GROSSMAN, A.D., 2001;Teleman, Graumann, Lin, Grossman, & Losick, 1998;Webb et al., 1997).

| CON CLUS IONS
We have revisited the cell cycle of B. subtilis using a microfluidic system and semiautomated time-lapse imaging, using fluorescent makers for cell division and chromosome termination. The results are generally in line with previous population-based studies. In light of the recent emphasis on time-lapse imaging and single-cell analysis, we can draw several important conclusions. First, our results provide general support for the Adder model for cell size homeostasis, although the relatively low overall precision of division makes it difficult to exclude alternative models. Given that cell cycle progression must be highly important to cells, it would seem to us to be surprising if they did not use a plethora of overlapping regulatory systems to control the decision to divide, and thus not to fit rigorously to any simple growth law. Also, it seems likely that some mechanisms will only come into play in cells that have strayed, for whatever reason, outside of the "normal" range. Second, we find that vegetative B. subtilis cells divide off-center at an appreciable frequency, and that this contributes to the overall variability in cell size. Finally, and probably our most surprising finding, we found little connection between the timing of chromosome termination and subsequent division.
Since the D or D* periods, measured at the population level, have historically been shown to be relatively constant across a range of growth rates in both E. coli and B. subtilis, we expected that this would be reflected in single-cell analysis. Perhaps the lack of connection is not so surprising, given that the D* period is relatively long (most of the cell cycle) in B.
subtilis. To conclude, it seems that much remains to be learned about the timing and localization of the bacterial division machinery.

ACK N OWLED G M ENTS
We are grateful to A. Wipat, S. Park, and L.E. Eland for the initial set-

CO N FLI C T O F I NTE R E S T S
The authors declare no conflict of interest.

AUTH O R CO NTR I B UTI O N S
S.L. performed all the experiments and most of the data analysis.
J.E., L.J.W., and S.L. designed the study and wrote the paper.

E TH I C S S TATEM ENT
None required.

DATA ACCE SS I B I LIT Y
All data are provided in full in the results section and the Appendices. F I G U R E A 1 Detection of cell septation by fluorescence and comparison with bright field images. Bright field images (B), fluorescent membrane protein images (M) and post-processed fluorescence images (P) of strain sL105 in a microfluidic channel. Images are in time order, with 2 minutes between frames. The septation event visible in the fluorescence image and defined by the clear, closed, membrane ring in the post-processed image defines time zero.

O RCI D
Other frames are arranged in sequence with times indicated relative to that of the zero frame (min). Note that division was undetectable by bright field microscopy until about + 12 or + 16 min [Correction added on 5 July 2019 after first online publication: Figure