E. coli Interrogating the cell cycle by cell dimension perturbations HaiZhenga,b,Po-YiHoc,MeilingJianga,BinTangd,WeirongLiua,b,DengjinLia,XuefengYue,NancyKlecknerf,ArielAmirc,1, andChenliLiua,b,1 aCenterforSyntheticBiologyEngineeringResearch,ShenzhenInstitutesofAdvancedTechnology,ChineseAcademyofSciences,Shenzhen518055,P.R.China; bUniversityofChineseAcademyofSciences,Beijing,100049,P.R.China;cSchoolofEngineeringandAppliedSciences,HarvardUniversity,Cambridge,MA02138,USA; dDepartmentofMaterialsScienceandEngineering,SouthernUniversityofScienceandTechnology,Shenzhen518055,P.R.China;eInstituteofBiomedicineand Biotechnology,ShenzhenInstitutesofAdvancedTechnology,ChineseAcademyofSciences,Shenzhen518055,P.R.China;fDepartmentofMolecularandCellularBiology, HarvardUniversity,Cambridge,MA02138,USA 7 ThismanuscriptwascompiledonJanuary4,2017 1 0 Bacteriatightlyregulateandcoordinatethevariouseventsintheir wheretheconstantv0 setstheaveragecellsizeatbirth. This 2 cell cycles to duplicatethemselves accuratelyand to controltheir is known as the “incremental" or “adder" model, and cells cellsizes.GrowthofEscherichiacoli,inparticular,followsarelation followingthisbehavioraresaidtoexhibit“addercorrelations” n a knownasSchaechter’sgrowthlaw. Thislawsaysthattheaverage [1–7]. Importantly,thesemodelspostulatethatcelldivisionis J cellvolumescalesexponentiallywithgrowthrate,withascalingex- governedbyaphenomenologicalsizevariable,withnoexplicit 3 ponentequaltothetimefrominitiationofaroundofDNAreplication reference to DNAreplication. tothecelldivisionatwhichthecorrespondingsisterchromosomes A second class of models for control of cell division pos- ] segregate.Here,wesoughttotesttherobustnessofthegrowthlaw tulates that cell division is governed by the process of DNA B tosystematicperturbationsincelldimensionsasachievedbyvary- replication,whichcanbedescribedasfollows. Thetimefrom C ingtheexpressionlevelsofmreBandftsZ.Wefoundthatdecreas- a replication initiation event to the cell division that segre- . ingmreBlevelresultedinincreasedcellwidth,withlittlechangein gates thecorresponding sister chromosomes can be split into o celllength,whereasdecreasingftsZ levelresultedinincreasedcell the C period, from initiation to termination of replication, i b length. Furthermore,thetimefromreplicationterminationtocelldi- and the D period, from termination of replication to cell di- - visionincreasedwiththeperturbeddimensioninbothcases. More- vision [8, 9]. Both the C and D periods remain constant at q over, thegrowthlawremained validovera rangeof growthcondi- approximately40and20minutes,respectively,forcellsgrown [ tionsanddimensionperturbations. Thegrowthlawcanbequanti- invariousgrowthmediasupportingarangeofdoublingtimes 1 tativelyinterpretedasaconsequenceofatightcouplingofcelldivi- between 20 and 60 minutes [10, 11]. We will refer to growth v siontoreplicationinitiation. Thus,itsrobustnesstoperturbations rateswithinthisrangeasfast. Allexperimentsdescribedhere 7 incelldimensionsstronglysupportsmodelsinwhichthetimingof are carried out under such fast growth conditions. Note that 8 replicationinitiationgovernsthatofcelldivision,andcellvolumeis C+D is approximately 60 minutes and larger than thetime 5 thekeyphenomenologicalvariablegoverningthetimingofreplica- between divisions at fast growth. This situation is achieved 0 tion initiation. These conclusionsare discussed in the context of by the occurrence of multiple ongoing rounds of replication. 0 ourrecentlyproposed“adder-per-origin”model,inwhichcellsadda Thatis,undertheseconditions,acellinitiatesaroundofrepli- . 1 constantvolumeperoriginbetweeninitiationsanddivideaconstant cationsimultaneouslyatmultipleoriginsthatultimatelygive 0 timeafterinitiation. rise to the chromsomes of their grand- or even great-grand- 7 daughters[12]. ExtendingthebasicdefinitionoftheC andD 1 Cellcycle|Cellgrowth|E. coli|DNAreplication periods,CooperandHelmstetterspecificallyproposedthatan : v i Bacteria can regulate tightly and coordinate the various X events in their cell cycles in order to accurately dupli- r cate their genomes and to homeostatically regulate their cell a sizes. This is a particular challenge under fast growth condi- SignificanceStatement tions where cells are undergoing multiple concurrent rounds How bacteria regulate cell division toachieve cell size home- of DNA replication. Despite much progress, we still have ostasis, with concomitant coordination of DNA replication, is anincompleteunderstandingoftheprocessesthatcoordinate afundamentalquestion. Currently,thereexistseveralcompet- DNA replication, cell growth, and cell division. This lack ingmodelsforcellcycleregulationinE.coli.Weperformedex- of understandingis manifested, for instance, in discrepancies perimentswherewesystematicallyperturbedcelldimensions, among various recent studies that propose different models and found that average cell volume scales exponentially with for control of cell division in thebacterium Escherichia coli. the product of the growth rate and the time from initiation of Oneclass of modelssuggests that cell division istriggered DNA replication to the corresponding cell division. Our data bytheaccumulationofaconstantsize(e.g. volume,length,or support a model inwhich cells initiatereplication on average surface-area) between birth and division [1–3]. Such models at a constant volume per origin, and divide a constant time aresupportedbyexperimentsmeasuringcorrelationsbetween thereafter. cell size at birth and cell size at division, which showed that, .. HZ,MJ,BT,WL,DL,XY,andCLperformedtheexperiments.PH,NK,AA,andCLanalyzedthe when averaged over all cells of a given birth size vB, cell size dataandwrotethemanuscript.CLdesignedtheresearch. at division vD approximately follows: Theauthorsdeclarenoconflictsofinterests. 1Towhomcorrespondenceshouldbeaddressed.E-mail:[email protected],CL- vD =vB+v0, [1] [email protected] www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX PNAS | January4,2017 | vol.XXX | no.XX | 1–7 initiation [15]. Becausecells growexponentiallyatthesingle- celllevel[16],cellswillthendivideonaverageatavolume∆ I per origin times a scaling factor S = 2(C+D)/τ. The average cell volume then follows V =∆2(C+D)/τ =S∆, [2] with∆=log(2)∆ ,becausethecellvolumeaveragedoveran I exponentially growing population is the average cell volume at birth times 2log(2) [17]. In Schaechter’s experiments, the C and D periods were approximately constant, giving rise to theexponentialscalingobserved. However,thederivationfor Eq. 2 holds regardless of the values of the C and D periods, andincaseswheretheyarenotconstant,averagecellvolume is not expected to scale exponentially with growth rate. Eq. 2isknownasSchaechter’sgrowthlaw,butwillbereferredto simply as thegrowth law for the rest of this manuscript. Recent single-cell analyses found that cells indeed initiate replication on average at aconstant volumeper origin orper somelocusclosetotheorigin[11]. Whilefurtherexperiments are required, the fact that introduction of an origin onto a plasmid doesnot affect cellcycletimingsorcell size suggests thatthelatterpossibilityiscorrect[11,18,19](SI).Below,for Fig.1. (A)Schematicillustrationoftheexperiment. MreBandFtsZareinvolvedin simplicity, we will use the phrase “per origin,” while keeping cellwallsynthesisandseptumformation,respectively.UsingmreB-orftsZ-titratable thiscomplexity in mind. strains,weareabletotunetheirexpressionlevelscontinuously,andperturbcelldi- Clearly, the two classes of models for control of cell divi- mensions. Inbothexperiments,theDperiodincreasedwithcellwidthandlength. sion differ fundamentally. In the first class, division depends TheCperiodanddoublingtimeτremainedconstant. (B)Schematicillustrationof ourmodel. TheperturbedDperiodsetstheaveragenumberoforiginspercell, only on the accumulation of size from birth, and DNA repli- whichisequaltothescalingfactorSbecausereplicationinitiationtriggerscelldi- cation plays no explicit role. In the second class, division is vision. Theaveragenumberoforiginspercellthensetstheaveragecellvolume downstream of the preceding initiation of DNA replication. followingthegrowthlaw. FortitratedmreBlevels,cellvolumechangesmanifested Importantly,also,theexperimentsleadingtothefirstclassof mostlyascellwidthchanges. FortitratedftsZ levels,celllengthchangedinstead, becauseFtsZdidnotaffectcellwidth. models defined cell size differences by measuring cell length. Since for a constant growth environment the widths of bac- terial cells are very narrowly distributed with a coefficient of initiationeventtriggersadivisionafteratimeC+D,thereby variation (CV) less than 0.05 [2], these analyses cannot dis- ensuringthatcellsdivideonlyafterthecompletionofaround tinguish whether cell size in a given environment is set by ofDNAreplication[8,10]. ThisCooper-Helmstetter(CH)for- a constant volume, surface-area, or length. This ambiguity mulation,hereaftertheCHmodel,belongstothesecondclass raises the question of what is the key phenomenological vari- of models for control of cell division. able governing cell cycle progression. The CH model is supported by the phenomenon of rate Here, we sought to test these models by perturbing cell maintenance [13]: after a change from one growth medium dimensionsinE.coli,andassayingtheeffectsofthosepertur- to a richer one (a shift-up), cells continue to divide at the bations on both replication events and cell division. In our rate associated with the poorer medium for a period of 60 study, shape perturbations were achieved by systematically minutes. According to the CH model, in which division is varying expression levels of the protein MreB, an actin ho- triggered by initiation, all cells that have already initiated mologue involvedincell wall synthesis,and theprotein FtsZ, replication beforeashift-upwillhavealsoalreadycommitted a tubulin homologue involved in the formation of the divi- to their ensuing divisions, and thus the rate of division will sionseptum[20,21]. Ourapproachisindicatedschematically remain unchanged for a time C+D following a shift-up. in Fig. 1A. It extends and complements Schaechter’s experi- Thesamevalueof60minuteshadalso emergedinaseem- ingly different context a decade earlier, in the seminal study ofSchaechteretal. [14]. Intheirwork,cellvolumes,averaged A %*5# B 2.5 mreB couveltruarengerxopwoinnegntuianldlyergdroowzeinngs opfopduifflaetrieonnt,gwreorwetmhemaseudriaedsufopr- "#$%& level AN12..50 ftsZ wpoarstiwngellfadsetscgrriobwetdh.byItanwaesxpfoounnednttiahlatrealvateiroangewcietlhl vgoroluwmthe *+# !!"*!$#*+# ()#’()#*,# .,/0#1# Rm .leR01..50 rate V = ∆eλT, where V is the average cell volume, ∆ is /0)1#234## ,-%&-.!"# $%&’"#"#$%& 0.01 10 100 a constant with dimensions of volume, λ is the growth rate, ()#’()& aTc (ng ml-1) τ =log(2)/λ is the doublingtime, and T ≈60 mins. Fig.2. TitratablemreBorftsZ expression. (A)ThegeneticcircuitofthemreB-or Donachie showed that this exponential scaling of average ftsZ-titratablestrains.TheexpressionofmreBorftsZisunderthecontrolofaPtet- cellvolumewithgrowthratecanalsobeexplainedbytheCH tetR feedbackloopandthenativemreB orftsZ wasseamlesslyreplacedwitha modelifitisfurtherassumedthatcellsinitiatereplicationon kanamycinresistancegene.(B)RelativemreBandftsZmRNAlevelinthetitratable average at a constant volume ∆ per origin of replication at strainsinbulkculturecontainingvariousconcentrationsofaTc(3-50ngml−1). I 2 | www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX Zheng etal. ments,inwhichgrowthratewasperturbed,andisreminiscent volume growth rates, or OD doubling rates (Fig. S1), of 600 of thework of Harris et al. [3], in which cell dimensions were the titratable strain remained approximately constant (CV perturbed, but with the important addition to both studies of 0.03) (Fig. 3D). Together, these results suggest that the thatwealsomeasuredthecellcycleperiodsC andDbecause titratable system is suitable for characterizing the functions theyplayanimportantroleintheCHmodel. Forthispaper, of MreB in a quantitative manner with no need to consider we definedivision as completion of septation. complications dueto differing growth rates. We next measured cell dimensions by phase contrast mi- croscopy (Fig. 3A), and found that with decreasing mreB Results expression level, the cell width increased (Fig. 3B). The cell Decreased mreB level resulted in increased cell width, with lengthchangedslightly,whichwediscussbelow(Fig. 3C,SI). littlechangeincelllength.Cell length and cell width are two Although the software used for image processing is designed majorcharacteristicsofarod-shapedcell. Asacellgrows,cell to define bacterial cell dimensions to subpixel precision [25], length increases exponentially, while cell width remains con- wewantedtoverifyourresultsusinganindependentmethod. stant. Howcellwidthisdeterminedandmaintainedislargely WeshowedthattheODpercellislinearlycorrelatedwiththe unknown. However,severalmutationsofmreB werereported cell volume in µm3 (Fig. S1). We also verified that neither to result in altered cell dimensions [20, 23, 24], thus raising thepresenceofinducernorthepresenceofthegeneticcircuit thepossibilitythatalterationsinmreBexpressionlevelwould construct has any effect on wild type cells within the ranges altercellwidth. Tocontinuouslyandsystematically varycell of inducer concentrations studied here (Fig. 3BC, and SI). width, we constructed a strain in which the level of mreB Similar results were obtained in all other growth media, sup- could be experimentally controlled. We employed a system porting various fast growth rates, tested in this study (Fig. inwhichaP -tetR feedbacklooptriggeredmreB expression S2). All of our experiments, and thus the resultant conclu- tet (Fig. 2A, and SI). The modulated copy of mreB was the sions, concern fast growth conditions as definedabove. sole version of the gene in the genome as the native copy of mreB was replaced by a kanamycin-resistance gene. In this Cellwidthmaintenancebycellwallstiffness.Lastly,we char- construct, expression of mreB could be tightly controlled by acterized in detail the morphological properties of the mreB adjustingtheconcentrationofanappropriateinducerofPtet- titratedcellsusingscanningelectronmicroscopy. Here,wider tetR, anhydrotetracycline (aTc). cells showed a slightly flattened, dumpling-like morphology We found that cell size increased with decreasing inducer (Fig. S3A). It is known that MreB is involved in bacterial concentration until, at very low mreB levels, the cells even- cell wall synthesis [20, 23, 24]. Thus it seemed possible that tually lysed (at an aTc concentration below 1 ng ml−1 in thecell wall might havebecomesofter in cells expressinglow Rich Defined Medium (RDM) + glucose). Above this mini- levels of MreB. Due to the difficulty in directly measuring mum threshold, the expression level of mreB varied linearly cell wall elasticity, we instead measured the effective cellular with the concentration of inducer (Fig. 2B). We also found stiffness (ECS) of the mreB titrated cells using atomic force that within a certain range of mreB expression levels, the microscopy. As expected, the ECS significantly decreased as Fig.3.TitratablemreBorftsZexpressiontosystematicallyperturbcellwidthorcelllength,respectively,withoutaffectingthevolumegrowthrates.(C)Representativephase contrastimagesofthemreB-titratableandwildtypestrains. (DandE)Scatterplotpresentstheaverage,meancellwidth(averagedalongthelongaxisofacell)(D)or averagecelllength(E)oftheindividualwildtype(WT)ormreB-titratablecells.(F)VolumegrowthratesinbulkcultureversusmeancellwidthformreB-titratablecells.(G,H, I,andJ)Thesameas(C,D,E,andF)butfortheftsZ-titratablestrain.ErrorbarsrepresenttheSEMofthreereplicates. Zheng etal. PNAS | January4,2017 | vol.XXX | no.XX | 3 X = 2(C(1−m)+D)/τ [26]. We used this relation to extract A 60 B 50 the lengths of the C and D periods from qPCR data of the )n50 )n40 copy numbers of different chromosome loci (including oriC, im im ( d40 ( d30 terC and a series of different loci between them; SI). The oire30 oire20 C period was also measured independently by the methods Dp 20 Cp 10 in Refs. [27, 28] which gave values consistent with those ob- tainedbytheqPCRmethod(SI).WealsomeasuredhOiusing 10 0 0.8 1.0 1.2 1.4 0.8 1.0 1.2 1.4 replication-run-out experiments (SI). Mean cell width (µm) Mean cell width (µm) These analyses revealed that, over the analyzed ranges in C 60 D 50 thelevelsof bothmreB andftsZ expression, theC period re- )n50 )n40 mained unchanged (CV of 0.09 and 0.05 for mreB- and ftsZ- im im titratablestrains, respectively),while theD period increased ( do40 ( do30 with increasing cell width or length, respectively (Fig. 4). irep30 irep20 Note that the relation between the D period and cell length D 20 C 10 predicted by our model below is not linear, but appears ap- 10 0 3 4 5 6 7 3 4 5 6 7 proximately linear given the particular values of the relevant Cell length (µm) Cell length (µm) parameters undertheconditions of these experiments. Thegrowthlawholds inface ofperturbations tocell dimen- Fig.4. Changesinthecellcycleparametersascelldimensionsareperturbed. (A) sions.We find that the growth law, Eq. (2), holds in our TheDperiodincreasedmonotonicallywithcellwidthinmreB-titratablestrains.The lineisthebestlinearfit.(B)TheCperiodremainedapproximatelyconstantascell experiments, both across a range of growth media (Fig. 4) widthchangedinresponsetotitratedmreBexpressionlevels.Thelineisthemean andacrossthetwotitratableperturbationsthatdrasticallyaf- valueofC averagedovermreB expressionlevels. (C)TheD periodincreased fectedcellwidthandcelllength(Fig. 5A).Thebestfitpropor- monotonicallywithcelllengthinftsZ-titratablestrains. (D)TheCperiodremained tionality constant is ∆=0.55±0.04µm3. Theplus-minusin- approximatelyconstantascelllengthchangedinresponsetotitratedftsZ expres- sionlevels.Thecircle,triangle,andsquareindicatesmreB-titratable,ftsZ-titratable, dicatesthe95%confidenceintervalofthefit. Fromthederiva- andwildtypestrains, respectively. Differentcolorsdenotegrowthmedia: redis tionofthegrowthlawintheIntroduction,wefindtheaverage RDM+glucose,blueRDM+glycerol.TheSEMsofthreereplicatesweresmallerthan cellsizeperoriginatinitiationtobe∆ =0.79±0.06µm3. In I thesizeofthesymbols. contrast, theaverage cell area, cell length, andcell widthare not proportional to thescaling factor S (Fig. 5B, Fig. S4). cellwidthincreased(Fig. S3B).Thisresultraisesthepossibil- Wefurthertested thevalidity of aconstant ∆ byfixing∆ itythatincreasedcellwidthreflectstheforcebalancebetween to 0.55µm3 and calculating the ratio of log2(V/∆) and τ−1, turgor pressure and thetensile resistance of thecell wall. which is equal to C+D according to Eq. (2) (Fig. 6A). We found that the values of C +D obtained in this way agree DecreasedftsZ levelresultedinincreasedcelllength,andno well with independent measurements in both the titratable, changeincellwidth.Weconstructedandcharacterized,using as well as wild type, strains (Fig. 6B). the same method as above, an ftsZ-titratable strain to allow perturbationofcelllength(SI).Wefound,inagreementwith Discussion previous work [21], that cell length increased with decreased ftsZ expression levels, but that both cell width and growth Here,we perturbedcell dimensions and thenobserved theef- rateremainedrelativelyconstantwithintherangesofinducer fectsoftheseperturbationsonthecellcycleinordertointer- concentration studied here (CV of 0.01 and 0.03 respectively, rogate the mechanism of cell cycle regulation in E. coli. The across different experiments, Fig. 3EFGH). most important finding of this work is that the growth law, that average cell volume is proportional to the scaling factor Correlationsbetweenperturbedcelldimensionsandcellcy- S = 2(C+D)/τ, remained valid across large perturbations in cletimings.We next investigated how the mreB and ftsZ ex- celldimensions. AsdiscussedintheIntroduction,thegrowth pressionlevelsaffecttheC andDperiods. Forabulkculture law can bequantitativelyderived undertwo assumptions: (i) in steady-state exponential growth, the CH model predicts theCHmodel,thatreplicationinitiationtriggerscelldivision that the average number of origins per cell, hOi, scales expo- after a constant time C +D, and (ii) that the average cell nentiallywithgrowth rate[22,26]. Thisisbecauseinsteady- volumeatinitiationofDNAreplicationisproportionaltothe state exponential growth, the number of cells and the total numberoforiginsatinitiation. Therobustnessofthegrowth number of origins in the population must both grow expo- lawasdocumentedabovesuggeststhatbothassumptionshold nentially at the same rate. However, because the CH model in face of theperturbations studied here. postulatesthatdivisiononlyoccursafteratimeC+Dfollow- Correspondingly,giventhefirstassumption,thepresented ing the corresponding initiation, the total number of origins results support the second class of models for control of cell will be larger than the number of cells by the scaling factor division (Introduction) in which the timing of initiation gov- S, defined above. Therefore, hOi = S. This relation holds erns that of division, and opposes the first class of models regardless of the valueof C+D. in which the timing of division is governed by the accumula- An expression for the average number of copies X of a tionofcell sizewithnoexplicitreference toDNAreplication. gene per cell as a function of the location m of the gene The presented findings do not rule out models in which, op- along the chromosome (m = 0 for oriC and m = 1 for positely to the CH model, division triggers initiation. How- terC) can be derived similarly. Under the assumption that ever, these models are heavily challenged by the finding that replication forks travel at a constant speed, theexpression is C+Disessentiallyconstant(CVofapproximately0.1)onthe 4 | www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX Zheng etal. A 6 B 6 A 3 8273 B100 3)mµ( emulov lleC0240 2 ∆4=0.565 µm8310 2)m( aera lleCµ 0240 2 4 6 8 10 logemuloVgoL2∆2-00121.00 0.02 0.04 6516 DC)nim( +46800040 60 80 100 S S 1/τ (min-1) Slope in panel A Fig.5. Thegrowthlawholdsinfaceofperturbationtocelldimensions. (A)The Fig.6.Theaveragecellvolumeperoriginatinitiationisconstantinfaceofperturba- averagecellvolumeisproportionalto thescalingfactorS = 2(C+D)/τ. The tionsincelldimensions.(A)Coloredlinesconnectlog2(∆)toboldedsymbolsasex- blacklineshowsthegrowthlawV = S∆forthebestfitproportionalityconstant amples.Therespectiveslopesofthelinesarecomputedastheratiooflog2(V/∆) ∆ = 0.55±0.04µm3. Theplus-minusindicatesthe95%confidenceintervalof over1/τ,andshownasnumbers(minutes).(C)MeasuredC+Dvaluesareplotted thefit. ThecoefficientofdeterminationR2ofthefitis0.81. (B)Theaveragecell againsttheratioscalculatedinpanel(A).Thecircle,triangle,andsquareindicates areaisnotproportionaltothescalingfactor. Theblacklineshowsthebestfitwith mreB-titratable, ftsZ-titratable, andwildtypestrains,respectively. Differentcolors interceptforcedtozero. TheR2ofthefitis0.40. Thecircle,triangle,andsquare denotegrowthmedia:redisRDM+glucose,blueRDM+glycerol.TheSEMsofthree indicatesmreB-titratable,ftsZ-titratable,andwildtypestrains,respectively.Different replicatesweresmallerthanthesizeofthesymbols. colorsdenotegrowthmedia:redisRDM+glucose,blueRDM+glycerol. TheSEMs ofthreereplicatesweresmallerthanthesizeofthesymbols. this situation, happens to manifest as an increase in length alone, with no change in width. This asymmetric change in single-celllevelatfastgrowth[11]. Tobeconsistentwiththis length,versuswidth,matchesthefactthatcellwidthismain- finding,thesemodelsmustsomehowcoordinatethetriggered tained normally even in filamentous cells where septation is initiationeventwiththedivisioneventinthenextgeneration. completely eliminated [21, 29]. Furtherstudieswill berequiredtoinvestigate thispossibility. Reduced mreB expression also increases the D period. In Our findings also provide information about the question one possibility, MreB would have two direct roles: both in of what keyphenomenological variablegovernscell cyclepro- septation via local effects at the division site [30], and in de- gression. Wefindthatonlycellvolume,andnotsurface-area, termining cell width. Here, at reduced MreB levels, delayed length, or width, is proportional to the scaling factor S (Fig. septation would increase the D period. Due to the second 5, Fig. S4). Together with the second assumption (above), roleofMreB,thecorrespondingincreaseincellvolumeinthis therobustnessofthegrowthlawthensupportsthehypothesis case is implemented mostly by the increase in cell width. Al- that volume per origin, rather than other geometric features, ternatively,MreBmightplayonlytheroleofdeterminingcell is the key phenomenological variable that is invariant at ini- width. In this scenario, the increased cell width then results tiation. Hence, cell volume governs the timing of initiation in aprolonged septation process andthus,alonger D period. and, bythe CH model, thetiming of division. In both scenarios, however, given experimentally-determined Figs. 5 and 6 also show that the proportionality constant values for the D period and cell width, the cell volume pre- ∆, which links average cell volume to the scaling factor S in dicted by the growth law also dictates small changes in cell thegrowth law (Eq. 2), remained constant across thepertur- length. Themagnitudesofthesepredictedchangesarejustat bationsstudiedhere. Inthederivationofthegrowthlaw,∆is the level of detection of the current study but are consistent proportionalto∆ ,ortheaveragevolumeperoriginrequired with observed values (Fig. S5). I forinitiation. Theconstancyof∆thereforesuggestthatnone In general, the growth law specifies cell volume without oftheperturbationsstudied(MreB,FtsZ,andvariousgrowth referencetotheaspectratiooftherod-shapemorphology. As media) affected themolecular mechanism underlyingthereg- aresult,agivenchangeincellvolumeduetoachangeinthe ulation of initiation. Despite drastic changes in cell shape, parameters of the growth law can be manifested as diverse thetitratablestrainsstillinitiatedreplicationonaverageata combinations of changes in cell length and cell width. constant 0.79±0.06µm3 perorigin just asin wild type. This Ouranalysishasshownthat,inanalyzingsize-relatedmea- value for the average cell volume per origin at initiation is surements, the growth law and its underlying tenets imply similar to the recently reported valueof 0.9−1.0µm3 [11]. thatanyperturbationstotheC orD periods,orgrowthrate Althoughthepresentstudyexaminedalteredgeneticcondi- will affect cell volume - even if the perturbations are not af- tionsthatresultinalteredcelllengthandcellwidth,ouranal- fectingthecoremechanismofsizeregulation thatdetermines ysis implies that these changes in cell length and cell width the value of the invariant average cell volume per origin at did not affect cell volume directly. Rather - since neither ge- initiation. Thus, with respect to determining cell size, an netic perturbation affected growth rate, the C period, or ∆ important distinction can be made between “primary” and - changes in the D period alone were responsible for the ob- “secondary” regulators, as highlighted in Ref. [31]. In the served changes in cell volume, which in turn are manifested context of the above analysis, MreB and FtsZ appear to be as changes in cell length and cell width. secondary regulators in E. coli because ∆ remained constant Fig. 1B illustrates schematically how the two genetic per- across titrated mreB and ftsZ levels. turbationsmightexerttheireffects. Inonecase,reducedftsZ Alloftheconsiderationsabovebuildontheclassicalworks expression increases the D period. Since FtsZ is a direct me- of Schaechter, CH, and Donachie, which consider population diator of septum formation, this effect could result directly average behaviors. Donachie further proposed a single-cell from a prolongation of the septation process. The increased interpretation of his idea in which initiation occurs in a cell D period dictatesanincreasedaveragecellvolume,which,in whenthecellreachesaconstantvolumeperorigin [15]. How- Zheng etal. PNAS | January4,2017 | vol.XXX | no.XX | 5 ever,theproposalisnotcompatiblewithexperimentalsingle- References cell measurements showing adder correlations between cell sizes at births and at divisions (see Introduction) because it 1. CamposMetal(2014)Aconstantsizeextensiondrivesbacterialcellsizehomeostasis.Cell 159(6):1433–1446. predicts no such correlations under conditions where growth 2. Taheri-AraghiSetal (2015) Cell-sizecontrol andhomeostasis inbacteria. Curr.Biol. rate is essentially constant [2, 16, 22]. 25(3):385–391. 3. HarrisLK,TheriotJA(2016)Relativeratesofsurfaceandvolumesynthesissetbacterialcell Asresolutiontothisconflict,someofushaverecentlypro- size.Cell165(6):1479–1492. posed a phenomenological adder-per-origin model in which a 4. AmirA(2014)Cellsizeregulationinbacteria.Phys.Rev.Lett.112(20):208102. 5. DeforetM,vanDitmarschD,XavierJB(2015)Cell-sizehomeostasisandtheincrementalrule constant volume is added between two rounds of initiations, inabacterialpathogen.Biophys.J.109(3):521–528. rather thanbetween tworoundsof divisions[4,22]. Wehave 6. SoiferI,RobertL,AmirA(2016)Single-cellanalysisofgrowthinbuddingyeastandbacteria shown that this adder-per-origin model leads both to adder revealsacommonsizeregulationstrategy.Curr.Biol.26(3):356–61. 7. KennardASetal(2016)Individualityanduniversalityinthegrowth-divisionlawsofsingleE. correlations and a constant average cell volume per origin colicells.Phys.Rev.E93(1):012408. at initiation. Furthermore, adder-per-origin can also explain 8. HelmstetterC,CooperS,PierucciO,RevelasE(1968)Onthebacteriallifesequence.Cold rate maintenanceand thegrowth law. SpringHarb.Symp.Quant.Biol.33(0):809–822. 9. ChienAC,HillNS,LevinPA(2012)Cellsizecontrolinbacteria.Curr.Biol.22(9):R340–R349. We conclude by discussing several unsolved questions. 10. Cooper S,Helmstetter C(1968) Chromosome replicationandthe divisioncycleof Es- cherichiacolib/r.J.Mol.Biol.31(3):519–540. First,wehaveignored single-cellfluctuationsintheC andD 11. WalldenM,FangeD,LundiusEG,BaltekinO,ElfJ(2016)Thesynchronizationofreplication periodsinourdiscussion,butrecentworksshowthattheyare anddivisioncyclesinindividualE.colicells.Cell166(3):729–39. important to understanding single-cell correlations [22, 32]. 12. SkarstadK,BoyeE,SteenHB(1986)Timingofinitiationofchromosomereplicationinindi- vidualEscherichiacolicells.EMBOJ5(7):1711–1717. Wehavealso not discussed E. coli in slow growth conditions, 13. CooperS(1969)CelldivisionandDNAreplicationfollowingashifttoarichermedium.J.Mol. but several works raise the possibility that E. coli might be- Biol.43(1):1–11. have qualitatively differently there. For one, previous works 14. SchaechterM,MaaloeO,KjeldgaardN(1958)Dependencyonmediumandtemperatureof cellsizeandchemicalcompositionduringbalancedgrowthofSalmonellatyphimurium. J. suggested that average cell volume per origin at initiation is gen.Microbiol.19(3):592–606. not constant in such conditions [33, 34]. Recent works also 15. DonachieWD(1968)RelationshipbetweencellsizeandtimeofinitiationofDNAreplication. Nature219(5158):1077–1079. suggested that at slow growth, E. coli does not show adder 16. GodinMetal(2010)Usingbuoyantmasstomeasurethegrowthofsinglecells.Nat.Methods correlations [11]. In light of these observations, it will be im- 7(5):387–390. portant to further study the single-cell physiology of E. coli 17. PowellEO(1956)Growthrateandgenerationtimeofbacteria,withspecialreferenceto continuousculture.J.gen.Microbiol.15:491–511. undercell dimension perturbations at slow growth. 18. HelmstetterCE,LeonardAC(1987)Coordinateinitiationofchromosomeandminichromo- somereplicationinEscherichiacoli.J.Bacteriol.169(8):3489–3494. There has recently been much interest in the question of 19. WangX,LesterlinC,Reyes-LamotheR,BallG,SherrattDJ(2011)Replicationandsegrega- cell size homeostasis across all domains of life [35], and we tionofanEscherichiacolichromosomewithtworeplicationorigins.PNAS108(26):243–250. note that adder-per-origin may be applicable to other organ- 20. GarnerECetal(2011)Coupled,circumferentialmotionsofthecellwallsynthesismachinery andMreBfilamentsinB.subtilis.Science333(6039):222–225. isms as well. For example, it is known that the bacterium 21. HillN,KadoyaR,ChattorajD,LevinP(2012)CellsizeandtheinitiationofDNAreplicationin Bacillus subtilis also exhibits the growth law and adder cor- bacteria.PLoSGenet.8(3):e1002549. relations. Repeating the experiment here in B. subtilis may 22. HoP,AmirA(2015)Simultaneousregulationofcellsizeandchromosomereplicationin bacteria.Front.Microbiol.6:662. helpprobetherelationsbetweencelldimensionsandcellcycle 23. WangS,Arellano-SantoyoH,CombsPA,ShaevitzJW(2010)Actin-likecytoskeletonfila- timings in a Gram-positive bacterium. Adder correlations in mentscontributetocellmechanicsinbacteria.PNAS107(20):9182–9185. cell volumewerealso found recentlyinbuddingyeastdiploid 24. TeeffelenSVetal.(2011)ThebacterialactinMreBrotates,androtationdependsoncell-wall assembly.PNAS108(38):15822–15827. daughter cells [6]. Given the different morphology of these 25. SliusarenkoO,HeinritzJ,EmonetT,Jacobs-WagnerC(2011)High-throughput,subpixel- cells, which changes dramatically throughout the cell cycle precisionanalysisofbacterialmorphogenesisandintracellularspatio-temporaldynamics. Mol.Micro.80(3):612–627. and particularly at budding, it is plausible that cell volume, 26. BremerH,DennisP(1996)EscherichiacoliandSalmonella,ed.NeidhardtF.(ASMPress; and not cell shape, is thekey phenomenological variable gov- WashingtonD.C.),pp.1553–1569. erning cell cycle regulation also in this case. 27. BremerH,ChurchwardG(1977)Deoxyribonucleicacidsynthesisafterinhibitionofinitiation ofroundsofreplicationinEscherichiacolib/r.J.Bacteriol.130(2):692–697. Importantly, although our study here has suggested a 28. BipatnathM,DennisPP,BremerH(1998)Initiationandvelocityofchromosomereplication inEscherichiacolib/randk-12.J.Bacteriol.180(2):265–273. coarse-grained, phenomenological model on the level of cell 29. AmirA,BabaeipourF,McIntoshDB,NelsonDR,JunS(2014)Bendingforcesplastically dimensions, it has not alluded to the molecular players in- deformgrowingbacterialcellwalls.PNAS111(16):5778–5783. volved. Several hypothetical molecular mechanisms, such as 30. FentonAK,GerdesK(2013)DirectioninteractionofFtsZandMreBisrequiredforseptum synthesisandcelldivisioninEscherichiacoli.EMBOJ32(13):1953–1965. theaccumulationofathresholdamountofaninitiatorprotein 31. BoyeE,NordströmK(2003)Couplingthecellcycletocellgrowth.EMBORep.4(8):757–760. perorigin[36]orthedilutionofaninhibitorprotein[37],were 32. AdiciptaningrumA,OsellaM,MoolmanMC,LagomarsinoMC,TansSJ(2015)Stochasticity previously shown to implement molecularly the phenomeno- andhomeostasisintheE.colireplicationanddivisioncycle.Sci.Rep.5(18261). 33. BatesD,KlecknerN(2005)Chromosomeandreplisomedynamicsine.coli:lossofsister logical model discussed here [6, 22]. Yet despite decades of cohesiontriggersglobalchromosomemovementandmediateschromosomesegregation. work, the molecular mechanism for cell cycle regulation in Cell121(6):899–911. bacteria remains a fundamentalunsolved question. 34. WoldS,SkarstadK,SteenHB,StokkeT,BoyeE(1994)TheinitiationmassforDNAreplica- tioninEscherichiacolik-12isdependentongrowthrate.EMBOJ13(9):2097. 35. Ginzberg MB, Kafri R, Kirschner M (2015) On being the right (cell) size. Science 348(6236):1245075. 36. SompayracL,MaaloeO(1973)AutorepressormodelforcontrolofDNAreplication.Nature 241(109):133–135. ACKNOWLEDGMENTS. Thisworkwasfinanciallysupportedby 37. FantesPA,GrantWD,PritchardRH,SudberyPE,WhealsAE(1975)Theregulationofcell NSFC (31471270), 973 Program (2014CB745202), 863 Program sizeandthecontrolofmitosis.J.Theor.Biol.50:213–244. (SS2015AA020936), theGuangdongNaturalScienceFundsforDis- 38. DattaS,CostantinoN,CourtDL(2006)Asetofrecombineeringplasmidsforgram-negative tinguished Young Scholar (S2013050016987), Shenzhen Peacock bacteria.Gene379:109–115. Team Plan (KQTD2015033ll7210153) to C.L and by a National 39. LiuCetal.(2011)Sequentialestablishmentofstripepatternsinanexpandingcellpopulation. Science334(6053):238–241. Institutes of Health grant NIH RO1 GM025326 to N.K. AA ac- 40. NeidhardtFC,BlochPL,SmithDF(1974)Culturemediumforenterobacteria. J.Bacteriol. knowledgessupportfromtheA.P.SloanFoundation. Theauthors 119(3):736–747. acknowledge the reviewers, Terry Hwa and Arieh Zaritsky for in- 41. Lobner-OlesenA,SkarstadK,HansenFG,vonMeyenburgK,BoyeE(1989)TheDnaA sightfuldiscussions. proteindeterminestheinitiationmassofEscherichiacolik-12.Cell57(5):881–889. 6 | www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX Zheng etal. 42. ChurchwardG,BremerH(1977)Determinationofdeoxyribonucleicacidreplicationtimein exponentiallygrowingEscherichiacoliB/r.J.Bacteriol.130(3):1206–1213. Zheng etal. PNAS | January4,2017 | vol.XXX | no.XX | 7