Time-lapse microscopy revealed that YgjD depletion lead to a gradual phenotypic transition in division and growth. Single cells that were transferred from permissive conditions to solid pads of LB medium with glucose first continued to divide regularly, forming microcolonies in which the number of cells initially increased exponentially. Then, after about four divisions, cell division slowed down and stopped (Figure 1b and Additional file 4 - movie 3). Analysis of the time-lapse images (Additional file 4 - movie 3) showed that, during this transition, cells size decreased (Figure 2a). This indicates a disturbance in of cell size homeostasis 19 - that cells divide before their cell size doubled.

We used elongation rates of single cells and the time interval between two divisions to analyze the change in cell size homeostasis during YgjD depletion. Since we were interested in how these parameters changed during depletion, we separated data from different cell generations of the depletion process. The first cell that is founding a microcolony is generation 0; this cell divides into two cells of generation 1, which divide into four cells of generation 2, and so on (also see Additional File 5 - Figure S2). To avoid comparisons between cells that are in different phases of their cell cycle, we only used cell size measurements (and later fluorescence intensities) of cells immediately before division. Also, to avoid incomplete and biased sampling, we removed data from above generation 6.

This analysis revealed that the small size of cells depleted for YgjD was a consequence of two effects: first, the rate of elongation (cell length increase over time) decreased (Figure 3a). Second, cells did not respond to the decrease in elongation rate by adjusting the frequency at which they divided; the interval between two cell divisions remained initially constant. As a direct consequence, cell length at division decreased continuously (Figure 2a).

The phenotype induced by YgjD depletion was specific, and depletions of other essential genes lead to different cellular morphologies. We analyzed time-lapse images of the depletion of three other essential genes (dnaT, fldA and ffh). Depletion of each protein resulted in cellular phenotypes that were different from each other and from YgjD when depleted (Additional file 6 - Figure S3; also see Additional Files 7, 8 and 9 - movies 4, 5 and 6). Also, the effects of YgjD depletion were different from the consequences of exposure to two antibiotics that we tested: we followed wildtype E. coli cells exposed to the translational inhibitors kanamycin and chloramphenicol at minimum inhibitory concentration (2.5 μg/ml for chloramphenicol, 5 μg/ml for kanamycin), and observed no decrease in cell size (Additional file 10 - Figure S4, and Additional Files 11 and 12 - movies 7 and 8).

For reference, we also analyzed images of growing microcolonies of wildtype E. coli MG1655 cells on LB medium supplemented with glucose. This experiments confirmed cell size homeostasis, as expected from normally growing cell: cells divided close to the moment when they had doubled their size, and small fluctuations in cell elongation rates were compensated for by adjustments in the time of cell division (Figure 2b and 3b).

What kind of disturbance of cell size homeostasis is induced by depletion of YgjD? We considered two possibilities. First, it is possible that the control that couples cell division to cell size is lost, so that cells divide in an uncontrolled way, irrespective of their size. Second, it is conceivable that cell division remains coupled to cell size, but the target size that a cell needs to reach before initiating division decreases over time.

If the decrease in cell size is the result of a controlled transition towards smaller cells, one would expect that, during the transition, the cell elongation rate and the timing of cell division would still be linked, but that this link would change quantitatively over time. In fact this is what we observed when we analyzed each generation of cells during the depletion process separately (inserts Figure 3a and 3b). Within a given generation the time interval between divisions and the rate by which a cell elongated was negatively correlated: cells that grew faster than the average of their generation tended to initiate division more quickly; cells that grew more slowly initiated division later. This suggests that cell growth and the timing of cell division are still linked within each generation in the depletion process, but that this link changes quantitatively over successive generations.

This analysis has, however, an important limitation: cells within a given generation are not independent from each other. Some of these cells are more closely related, because they derive from the same mother or grandmother. This can lead to spurious correlations between traits; in our case, this effect could lead to artificial correlations between cell elongation rates and interdivision intervals. This problem of relatedness in lineage trees is known from phylogenetic studies, where it is referred to as phylogenetic dependence 21 . In the context of phylogenetic studies, these dependencies can be resolved by analyzing differences between independent pairs of species, rather than calculating correlations on the basis of the whole phylogenetic lineage 21 .

We used a variation of this approach to get an unbiased view on the relationship between cell growth and the timing of cell division: for each generation, we analyzed pairs of cells emerging from the same cell division, and calculated the difference in growth rates and in the time to division for each pair. We refer to two cells emerging from the same division as 'sisters' (thereby ignoring that these two cells have cell poles of different ages, 22 23 ). The differences for all sister pairs represent independent data points, and we can use them to calculate the correlation between cell growth and time to division in an unbiased way.

The independent contrast analysis confirmed our earlier conclusions: comparing each cell to its sister cell, we found that cells that grew slower than their sisters also displayed a longer interval between cell divisions, while faster growing cells divided earlier. This manifests as a negative correlation between the difference in cell elongation rate and the difference in interdivision intervals between two sisters (inserts Figure 3c and 3d; see also Additional File 13 - Figure S5). This is consistent with the interpretation that, during YgjD depletion, the timing of cell division remained coupled to a given cell size - and that the target cell size declined.

The transition to decreased cell size is reminiscent of morphological changes that occur during the 'stringent response' 24 25 , a stress adaptation program that is elicited when cells encounter amino-acid or carbon-starvation 26 . The stringent response is induced by accumulation of the 'alarmone' guanosine tetra/penta phosphate ((p)ppGpp), e.g. in response to low concentrations of amino-acylated tRNAs 26 . We thus wanted to investigate this possible link to (p)ppGpp signaling more closely, and asked whether the changes in cell homeostastis upon YgjD depletion are mediated through (p)ppGpp.

We constructed a strain, TB84, that is deficient in (p)ppGpp synthesis ((p)pGpp⁰), due to deletions of relA and spoT 26 27 , and in which expression of ygjD was again under control of P_ara. We followed growing microcolonies of TB84 as described above and found that the consequences of YgjD depletion were profoundly different: cell elongation rate decreased during the YgjD depletion process as for the relA ⁺ spoT ⁺ strain TB80 (Figure 4a). In contrast to what we observed with this (p)ppGpp⁺ strain, the decrease in elongation rate was compensated for by an increase in the time interval between two divisions (Additional file 14 - movie 9, and Figure 4a). As a consequence, cell size at division was not reduced, and the final cell length of depleted (p)ppGpp⁰ cells (TB84) was on average twice that of depleted (p)ppGpp⁺ cells (TB80) (Figure 4b). This is reminiscent of the elongated cells found in populations of cells depleted for YgjD by Handford and colleagues 3 .

This suggests that the changes in cell size in response to YgjD depletion are mediated through the alarmone (p)ppGpp; an alternative explanation is that the absence of (p)ppGpp leads to cell elongation (as has been previously reported 27 ), and that this elongation compensates indirectly for reductive fission upon YgjD depletion. Importantly, TB84 cells still ceased cell division (Additional file 15 - Figure S6). Thus, ygjD is still essential even in the absence of (p)ppGpp, and termination of cell division is not solely a consequence of a diminished cellular growth rate.

To further test the idea that ygjD depletion triggers (p)ppGpp synthesis we measured, on a single cell level during YgjD depletion, the activity of two promoters known to respond to the intracellular level of (p)ppGpp: P_apt is repressed by (p)ppGpp, while P_rsd is induced by (p)ppGpp 28 . We transformed TB84 with plasmids carrying transcriptional promoter-gfp fusions 29 encoding P_apt-gfp and P_rsd-gfp, and measured gene expression from these promoters as fluorescence intensity over consecutive cell divisions. The level of GFP expression steadily decreased in the strains where gfp was controlled by P_apt (Figure 5a), and steadily increased when controlled by P_rsd (Figure 5c). Furthermore, this change in fluorescence was tightly linked to the rate by which cells elongated (Figure 5b and 5d). When the same strains were grown on L-arabinose containing medium no consistent changes of fluorescence could be observed (Additional file 16 - Figure S7). These observations are consistent with the scenario that YgjD depletion induces (p)ppGpp synthesis, and thus influences promoters whose expression depends on the levels of (p)ppGpp.

Single cell analysis indicated that, in the cells depleted for YgjD, there is a link between decreased cell elongation rate and (p)ppGpp levels. Using independent comparisons between sister cells in the microcolonies undergoing YjgD depletion, we found that if a cell had a lower elongation rate than its sister, it also tended to have lower levels of GFP expressed from P_apt (details not shown; for P_rsd-gfp, this pattern was not observed). These data support the idea that the link between (p)ppGpp levels and the cell elongation rate is direct; for example, it is possible that high levels of (p)ppGpp cause low elongation rates 30 .

Our results suggest further that YgjD depletion has two (possibly linked) effects: first, depletion triggers (p)ppGpp synthesis. Second, it leads to termination of cell division. To gain insights in which phase of the cell cycle YgjD-depleted cells are arrested we visualized the DNA-content of individual cells with DNA-staining and subsequent fluorescence microscopy (Additional File 17 - Figure S8). After YgjD depletion in (p)ppGpp⁺ cells (TB80), DNA was localized at midcell and filled large areas of the cell (Additional File 17 - Figure S8 b), possibly indicating that cells were unable to carry out additional cell divisions due to "nucleoid occlusion" 31 . This mechanism prevents premature cell division before chromosomes have been distributed to opposite cell halves. However, termination of cell division also manifests in a (p)ppGpp⁰ strain (Additional File 17 - Figure S8 c): depleted cells were elongated, and only a small fraction of the cell volume was filled with DNA. Thus, in the (p)ppGpp⁰ background, nucleoid occlusion alone cannot be responsible for termination of cell division. The elongated phenotype of YgjD depleted (p)ppGpp⁰ cells resembles filamentous cells blocked in cell division. However, since abrogating cell division is not inhibiting DNA replication or DNA segregation 32 it appears unlikely that YgjD directly affects cell division.

P1 transduction and TSS transformation were performed as described elsewhere 36 37 . Strain DY330 as well as strains harboring the plasmid pCP20 38 were grown at 32°. All other strains were grown at 37°. To grow TB80 and TB84 under permissive conditions, we used LB medium (Sigma) supplemented with 0.1% (batch culture) or 0.01% (before time-lapse microscopy) L-arabinose (Sigma). LB agar (1.5% agar) was from Sigma, and used for preparing agar plates and agar pads for time-lapse microscopy.

Strains containing more than one knockout or marker were generated by sequential P1-transductions. Resistance markers were removed by Flp recombinase mediated site-specific recombination 39 . To control expression of ygjD, we constructed a conditional mutant with a second copy of the promoter of the araBAD operon in front of the native chromosomal locus of ygjD by directly inserting a P_ara-construct in front of ygjD, as described previously 40 . Removal of L-arabinose and addition of glucose allows tight repression of target genes under control of P_ara 40 41 .

We closely followed the description of 40 and first constructed a strain in which a kanamycin resistance cassette is linked to the promoter of araBAD. We inserted such a kanamycin marker downstream of araC with the following primers:

5'_araC_yabI_insert

AATCAGACAATTGACGGCTTGACGGAGTAGCATAGGGTTTTGTGTAGGCTGGAGCTGCTTC;

3'_araC_yabI_insert

GCATAATGTGCCTGTCAAATGGACGAAGCAGGGATTCTGCCATATGAATA

TCCTCCTTAGTTCCTAT.

The insertion was done in DY330 following the protocol described by 42 , verified by PCR and moved to MG1655 by P1 transduction, thus generating TB55.

TB55 was subsequently used to generate a PCR product that spanned the kanamycin cassette adjacent to araC, araC, the full intergenic region between araC and araB, and 42 basepairs at the 5' and 3' -prime ends that were homologous to the upstream and 5'-coding region of ygjD, dnaT ,fldA or ffh. The sequence of the primers was

ygjD_insert5' AGTTTTACATCAACCCGCATTGGTCCTACACTGCGCGGTAATAATGTGCCTGTCAAATGGACG

ygjD_insert3' GCCGGTTTCATCGCAGGAAGTTTCAATACCCAGTACACGCATCGTTTCACTCCATCCAAAAAA

dnaT_insert5' TCCGTGTGTTACTATAAAAGTTATCTCCCTTCTCGTTCATCGAATGTGCC

TGTCAAATGGACG

dnaTC_insert3' GTCAATACCAACGACGTCCGGGGTCAAAACTCTGGAAGACATCGTTTCAC

TCCATCCAAAAAA

ffh_insert5'

GACGCCTTCATGTTATACTGCGGCAAAATACTGATGATGTGTAATGTGCC

TGTCAAATGGACG

ffh_insert3' GCGCAGCGTGCGCGACAAACGATCGGTTAAATTATCAAACATCGTTTCAC

TCCATCCAAAAAA

fldA_insert5' TGCCTTTATCCGTGGGCAATTTTCCACCCCCATTTCAATAAGAATGTGCC

TGTCAAATGGACG

fldA_insert3'

ATTACCGGTGTCGCTGCCGAAAAAGATGCCAGTGATAGCCATCGTTTCAC

TCCATCCAAAAAA

DY330 cells were grown in LB medium supplemented with 0.2% L-arabinose and made electro- and recombination competent 42 , and electroporated with the above described PCR product. After electroporation, cells were transferred to LB medium containing 0.1% L-arabinose and incubated at 32° for 1.5 hours prior to plating on LB plates agar containing L-arabinose (0.1%) and kanamycin (50 μg/ml, Sigma). Clones were checked on LB agar plates supplemented with 0.4% glucose to confirm that they were unable to grow in the presence of glucose.

The promoter fusions and the adjacent araC gene were verified by sequencing with the following primers:

araC_FW GCTACTCCGTCAAGCCGTCA;

ygjD_RW GGCAATTGGTCTGGGGAGCA.

dnaTC_RW AGAGTTGATCGTCCAGAGCG

ffh_RW ATTTTGACGAACTCCTGCCC

fldA_RW CGAGAGTCGGGAAGAAGTCA

The constructs were then moved by P1 transduction into MG1655. To construct TB80 the kanamycin cassette was removed with pCP20.

The knockout ΔrelA::kan was derived from the KEIO library clone JW2755 2 and P1-transduced into TB82. ΔspoT::kan was introduced using AB1058) as donor strain for P1 transduction.

To measure activity of the promoters P_ara, P_rsd and P_apt, MG1655 and TB80 were transformed 37 with plasmids that contain transcriptional promoter-gfp fusions 29 .

LB agar pads were prepared by filling a cavity of a sterile microscope cavity slide with a drop of freshly melted LB agar, and covering it with a cover slip to attain a flat surface. The cavity slide was transferred to a fridge for a short time to allow the agar to solidify. Upon removal of the cover slip and removal of excess LB agar, the pads can be inoculated with precultured cells as described in the next paragraph.

TB80 and TB84 were cultured over night at 37° in LB medium with 0.1% L-arabinose and diluted 1:100 into fresh LB medium containing 0.01% L-arabinose. In early exponential phase, cultures were washed at least twice in LB supplemented with 0.4% glucose to remove residual L-arabinose. Wildtype E. coli MG1655 was treated similar for control experiments. 1.5 μl of a washed and diluted culture were transferred to the surface of a pad of LB agar (supplemented with D-glucose, L-arabinose, chloramphenicol or kanamycin as indicated for individual experiments) in a microscope cavity slide. The agar pad was closed with a cover slip and sealed with vacuum grease. Under these conditions, cells can grow exponentially in a two-dimensional plane for many generations without restrictions 23 . The slide was mounted onto an automated microscope (Olympus BX81) and incubated at 37°C (Cube and Box incubation system, Life Imaging Services, Reinach, Switzerland). Images were recorded every 2 or 4 minutes. Intensity and exposure times to fluorescent light were minimized to avoid cellular damage. The resulting image sequences were analyzed with the Matlab based script package "Schnitzcell" (kindly provided by Michael Elowitz, CalTech, USA 18 ), and data was extracted with custom-made Matlab scripts (Table 1).

To quantify associations between phenotypic traits, we used non-parametric correlation analysis (Spearman's rank correlation in PASW Statistics 18.0).