A Plasmid System with Tunable Copy Number

1 A Plasmid System with Tunable Copy Number

2 Miles V. Rouches and Yasu Xu 3 Field of Biophysics, Cornell University, Ithaca, NY 14853, USA

4 Louis Cortes and Guillaume Lambert⇤ 5 School of Applied and Engineering Physics, Cornell University, Ithaca, NY 14853, USA 6 (Dated: July 16, 2021) Plasmids are one of the most commonly used and time-tested molecular biology platforms for genetic engineering and recombinant gene expression in bacteria. Despite their ubiquity, little consideration is given to metabolic e↵ects and fitness costs of plasmid copy numbers on engineered genetic systems. Here, we introduce two systems that allow for the finely-tuned control of plasmid copy number: a plasmid with an anhydrotetracycline-controlled copy number, and a massively parallel assay that is used to generate a continuous spectrum of ColE1-based copy number variants. Using these systems, we investigate the e↵ects of plasmid copy number on cellular growth rates, gene expression, biosynthesis, and genetic circuit performance. We perform single-cell timelapse measurements to characterize plasmid loss, runaway plasmid replication, and quantify the impact of plasmid copy number on the variability of gene expression. Using our massively parallel assay, we find that each plasmid imposes a 0.063% linear metabolic burden on their hosts, hinting at a simple relationship between metabolic burdens and plasmid DNA synthesis. Our plasmid system with tunable copy number enables the precise control of gene expression and highlight the importance of tuning plasmid copy number as a tool for the optimization of synthetic biological systems.

7 INTRODUCTION 38 cellular growth and metabolism. Specifically, we first in- 39 troduce a strategy for finely tuning plasmid copy number

40 8 Plasmids are an invaluable tool in biotechnology and in a way that is dependent on the concentration of an ex- 41 9 genetic engineering due to their small size and the ease ogenous inducer molecule. Then, we develop a method 42 10 with which they can be engineered for biotechnological which uses a massively parallelized assay for rapid screen- 43 11 applications. Recent studies have highlighted the poten- ing of the dependence of a system on plasmid copy num- 44 12 tial benefits of fine-tuning plasmid copy number to the ber. 13 task at hand. For example, runaway replication has been 45 We further characterize our systems at both the macro- 14 used as a method for increasing cellular protein produc- 46 scopic and single cell level to uncover a relationship be- 15 tion [1, 2]. Recent works have also leveraged plasmid 47 tween gene copy number, protein production, and cellular 16 copy numbers to amplify a signal through a large increase 48 growth rates. Our approach also leads to insights into the 17 in copy number [3, 4] and to increase the cooperativity 49 variation in plasmid copy numbers and the phenomena 18 and robustness of synthetic gene circuits [5]. Single cell 50 of plasmid loss and runaway replication at the single- 19 measurements have also been used to accurately quantify 51 cell level. We finally demonstrate that manipulations of 20 the prevalence of plasmid loss within a population [6]. 52 plasmid copy numbers can be used to tune the behav- 21 While recent studies have probed the e↵ect of plas- 53 ior of synthetic gene circuits such as a simple CRISPRi 22 mid copy number on host cell growth [7, 8] and the role 54 genetic inverter, and control the expression of complex 23 of plasmids on central metabolic processes [9], we still 55 biosynthetic pathways to optimize the production of the 24 lack a quantitative description of the relationship be- 56 pigment Violacein. Our work thus provides a straightfor- 25 tween plasmid copy number and host metabolic burden. 57 ward method for generating copy number variants from 26 Indeed, recombinant gene expression results in signifi- 58 a specific plasmid backbone, which we anticipate will ex- 27 cant metabolic burden on bacterial cells [10–12] which 59 pand the synthetic biology toolkit and constitute the ba- 28 may interfere with normal cellular processes and even 60 sis for more nuanced attention to plasmid copy number 29 lead to cessation of growth. The control of plasmid copy 61 in synthetic gene circuits, protein biosynthesis, and other 30 number can also drastically alter the behavior of simple 62 biotechnological applications. 31 repression systems [13, 14] which may significantly im- 32 pact engineered gene circuits and should be important 33 considerations for plasmid-based biotechnology and bio- 63 RESULTS 34 engineering applications

35 In this work, we use two novel approaches to actively 64 Model of ColE1 Plasmid Replication and Copy 36 control the copy number of pUC19 and other ColE1- 65 Number 37 based plasmids to study the impact of copy number on

66 The molecular mechanisms that regulate plasmid copy 67 number control have been well characterized [16]. In Corresponding author ⇤ 68 the ColE1 origin of replication, used extensively in this 2

A pUC19 Plasmid map D E 0.03 Inhibition window 100 Pbla Pp,kp (n steps) Plac RNA-p 0.02 +1 +111 +360 +555 MCS pUC-pTet 1 10 Ptet ampR ORI LacZa 0.01 P ,k OD 600 0 RNA-i i i 0 300 600 Time (mins) Growth rate (1/min)

Plasmid Copy Number 1 B RNA-p C 0 0.01 0.1 1 10 100 1 10 100 aTc Concentration (ng/ml) Copy number RNA-i 100 1000 RNA-i/p 10 transcription PCN 1 F pUC-pLac G 100 0 60 120 Division time 0.020 ρ 300 pUC19 copy number Plac 0.015 RNA-i/p 10 RNA-p hyperbolic (n=1) binding priming 200

Runaway replication 1 stableexponential plasmid copy (n=∞) number 0.010 Plasmid Copy Number

1 100 OD 600 0 0.005 0 300 600 Plasmid loss Time (mins) 30 Growth Rate (1/min) RNA-p Plasmid Copy Number 0.00 0.25 0.50 0.75 1.00 1.25 -4 -2 0 0.000 release 0 10 10 10 30 100 200 300 No Plasmid RNA production ratio IPTG Concentration (mM) Copy Number replication replication ( )

Fig. 1. Replication of ColE1 Origin Plasmids. A) Map of the pUC19 plasmid. The origin of replication consists of a 555bp region containing the priming RNA (RNA-p), the inhibitory RNA (RNA-i), the inhibition window, and the site of replication initiation (ORI). B) Diagram of the steps leading to ColE1 replication. In the ColE1 Origin, reversible binding of the inhibitory RNA-i to RNA-p exercises control over the copy number by inhibiting replication (left branch). Production of RNA-p transcript triggers plasmid replication through the hybridization of RNA-p with the ORI (right branch). C) Predicted plasmid copy number for hyperbolic (upper) and exponential (lower) initiation models [15] plotted as a function of RNA-p transcription initiation rate. Inset shows the predicted dependence of plasmid copy number on host cell division time. Model parameters: ✏ =0.545 / min, ⇢ =0.65, r =0.01 / min, ki =0.95 / min. D) Plasmid copy number measurements made by digital droplet PCR of the aTc-inducible pUC-pTet plasmid at increasing aTc concentrations. E) Growth rates of cells hosting the pUC-pTet plasmid at increasing plasmid copy numbers. We see a marginal e↵ect of plasmid copy number on cellular growth rates, inset shows individual growth curves colored by aTc concentration. F) Plasmid copy number measurements of a plasmid which contains an additional copy of the inhibitory RNA under the control of an IPTG-inducible promoter. G) Growth rate vs plasmid copy number of the IPTG-inducible plasmid shows a strong metabolic burden at high levels of RNA-i expression.

69 work, two RNAs control replication (Fig. 1A): the prim- 88 where ✏ is the RNA degradation rate, ⇢ is the RNA-p 70 ing RNA (RNA-p), which acts in cis as a primer near 89 priming probability, r is the cellular growth rate, and 71 the origin of replication, and the inhibitory RNA (RNA- 90 n is the number of rate-limiting steps in the inhibition 72 i) which is transcribed antisense to RNA-p and acts in 91 window [15]. The n 1 and n limits represent ! !1 73 trans to inhibit replication through binding to the prim- 92 hyperbolic and exponential inhibition, respectively (Fig. 74 ing RNA prior to hybridization near the origin (Fig. 1B) 93 1C). 75 [17]. 94 As the priming-to-inhibitory RNA production ratio 76 A simple macroscopic-scale model developed by Pauls- 95 kp/ki approaches 1, the number of inhibitory RNA will 77 son et al. recapitulates these observations and predicts 96 not be present in high enough number to sequester prim- 78 the functional dependence of plasmid copy number on the 97 ing RNA transcripts, thereby failing to limit plasmids 79 molecular details of this process [15] (Fig. 1C). This pro- 98 from duplicating in a process called runaway replication 80 cess is classiﬁed as inhibitor-dilution copy number control 99 (Fig. 1C). In addition, increasing the plasmid copy num- 81 [18], owing to the fact that the copy number is controlled 100 ber may increase the metabolic burden associated with 82 through a combination of replication inhibition and dilu- 101 plasmid expression and reduce the growth rate r,which 83 tion of plasmids and regulatory components due to cell 102 in turn increases the plasmid copy number further as 84 division. 103 r 0. ! 85 Speciﬁcally, as the ratio of RNA-p transcription rate 104 On the other hand, as the RNA production ratio kp/ki 86 kp to RNA-i transcription rate ki is increased, plasmid 105 decreases, the number of plasmids inside each cell will ap- 87 copy number Np will also increase according to 106 proach 1 and small variations in the number of plasmids 107 apportioned to each daughter cell during cell divisions 108 may lead to the catastrophic consequence of plasmid loss. 1 n 109 These two stresses, plasmid loss and metabolic burden, ✏ + r ⇢k n p 110 necessitate the evolution of copy number control systems Np = 1 (1) ki kp r 111 to ensure that neither stress results in the loss of the "✓ ◆ # 3

pUC-pTet-sfGFP A B 138 the absolute di↵erence in growth rates is small, a cell with Ptet 0.025 ~50 plasmids 139 a few copies of the plasmid grows roughly 10 % faster sfGFP per cell Pconst 140 than a cell with 50 plasmids, we can clearly see that in-

~1-3 plasmids 0.020 141 creasing the plasmid copy number has a weak, though per cell 142 consistent, e↵ect on the host cell growth rate.

[aTc] 143 To complement the work above, we sought to demon-

PCN Growth Rate (1/min) fitness 0.015 0 0.01 0.1 1 10 100 144 Low protein expression High protein expression strate that the transcription level of inhibitory RNA can Risk of plasmid loss Risk of runaway replication aTc Concentration (ng/mL) 145 be used to control ColE1 copy number as well. Seeking to 146 do this with minimal perturbation to the native pUC19 C D 147 origin of replication, we inserted a second copy of the 103 103 148 inhibitory RNA downstream of the IPTG-inducible Lac 149 promoter on the pUC19 plasmid. This gives us control 150 over the inhibition of replication through the addition of GFP (AU) 102 GFP (AU) 102 151 exogenous IPTG (Fig. 1F).

0 0.01 0.1 1 10 100 0 20 40 152 When no IPTG was added to the cells we measured aTc conc (ng/mL) Plasmid Copy Number 153 270 plasmids/genome, a number that agrees with what ⇠ 154 we have measured for the standard pUC19 plasmid, indi- 155 cating that our system eciently represses the additional Fig. 2. Control of Gene Expression through Induction of Plas- 156 inhibitory RNA transcript and does not a↵ect the copy mid Copy Number. A) Schematic of the e↵ects of adding aTc 157 number in the absence of induction. Induction of the in- to the pUC-pTet-GFP plasmid. As the concentration is in- 158 hibitory RNA greatly reduced plasmid copy number to creased the number of plasmids and GFP genes increases as 159 approximately 30 plasmids/genome at full RNA-i induc- well as the metabolic cost of the plasmid. B) Growth rates 160 tion, demonstrating that tuning RNA-i production rate of cells hosting the pUC-pTet-sfGFP plasmid as a function of aTc concentration. C) GFP Production of the pUC-pTet- 161 can be used to control the plasmid copy number. GFP plasmid grown at di↵erent aTc concentrations. D) GFP 162 It is interesting to note that we observed paradoxi- production of the pUC-pTet-GFP plasmid as a function of 163 cal e↵ects on the growth rate from cells harboring this plasmid copy number as measured by digital droplet PCR. 164 plasmid - at high plasmid copy numbers we see a faster 165 growth rate and at lower copy numbers we see a markedly 166 slower growth rate (Fig. 1G). These observations are

112 plasmid [19]. 167 markedly di↵erent from what we see in the pUC-pTet 168 plasmids, where increased plasmid copy number has a rel- 169 atively modest e↵ect on the growth rate of host cells. In-

113 Tuning Priming and Inhibitory RNA Transcription 170 stead, the reduction in the maximum cell density reached 114 to Control ColE1 Copy Number 171 by the IPTG-induced populations (Fig. 1G, inset) sug- 172 gests that overproduction of the inhibitory RNA may 173 completely abolish plasmid replication in some cells, pre- 115 The Paulsson model of ColE1-type replication predicts 174 venting them from dividing further. 116 that the RNA-p transcription initiation rate should de- 117 termine the plasmid copy number without drastically al- 118 tering the sensitivity of copy number control [18]. Thus, 175 119 in order to only control the activity of the promoter Gene Expression Scales with Plasmid Copy Number 120 upstream of the priming RNA, we ﬁrst replaced the 121 native priming promoter in the pUC19 plasmid with 176 Understanding the interplay between plasmid copy 122 an anhydrotetracycline (aTc) inducible promoter (Fig. 177 number, host cell growth rate, and gene expression is 123 1D). Liquid cultures of TetR- and LacI-overexpressing 178 key to optimizing protein production in bacterial cell cul- 124 E. coli carrying this plasmid were grown in aTc concen- 179 tures. If protein production is too low, one may have to 125 trations spanning 4 orders of magnitude and the plasmid 180 process massive amounts of cells to obtain a sucient 126 copy number was determined via digital droplet PCR 181 amount of their target protein. If too much protein is 127 (ddPCR) from total DNA isolates by measuring the ra- 182 produced, on the other hand, cells may stop growing or 128 tio of the bla gene on the plasmid to the genomic dxs 183 be out-competed by cells harboring plasmids that lack 129 gene [20]. 184 the correct insert. Understanding the interplay between 130 This plasmid showed a robust control of copy number 185 plasmid regulation and protein expression is key to solve 131 with aTc induction (Fig. 1D). Copy numbers ranged 186 protein production maximization problems. 132 from 1.4 copies per cell in the complete absence of aTc to 187 To directly measure the e↵ects of plasmid and gene 1 133 roughly 50 copies per cell at full induction (100 ng mL 188 copy number on gene expression and host cell growth 134 aTc). 189 rate, we ﬁrst inserted sfGFP downstream of a consti- 135 We subsequently measured the growth rate of cells har- 190 tutive promoter on our pUC-pTet plasmid, whose copy 136 boring the pUC-pTet plasmid as a function of the aTc 191 number is controlled through exogenous aTc induction. 137 concentration in a microplate reader (Fig. 1E). Though 192 This gives us a straightforward way of monitoring the ef- 4

A B C D 100 0.1 ng/mL 1.00 0.3 ng/mL 1 ng/mL 100 ng/mL N=59,757 0.75 <µ1> = 0.0148 10−1 1.0 10 ng/mL 0.50 N=81,511 0.5 FC

0.0 <µ0.3> = 0.0157 Fold-change 0.25 3 ng/mL −2 0 1 100 10 aTc N=71,885 <µ > = 0.0172

Rel. Prob. Density 0.1 0.00 0 0.01 0.1 1 10 100 −2.25 −2 −1.75 −1.5 1 ng/mL 10 10 10 10 N=35,277 aTc concentration (ng/mL) Growth Rate (1/min)

0.3 ng/mL E F N=33,156 60% 100% 100 3 0.1 ng/mL Run- 10 100 N=13,800 10 away 40% repli- 1 0.01 ng/mL 3 cation 1 0.1 N=10,276 10% 0.01 0.3 20% 0.1 Plasmid 0 ng/mL loss 0.3 N=1,455 0 0 0.01 Viable Cells Fraction 0%

10 2 10 4 Elongated Cells Fraction 0 500 1000 1500 0 500 1000 1500 Intensity Time (mins) Time (mins)

Fig. 3. Microscopic Analysis of sfGFP Expressed by the pUC-pTet. A) Representative snapshots of microfluidic chambers at 1 various aTc concentrations. Slow-growing cells at high copy number are indicated by orange arrows and cells at 0.3ngmL of aTc in an “activated” state are indicated by purple arrows. B) Distribution of cellular fluorescence intensities at increasing aTc concentrations. C) Normalized mean fluorescence as a function aTc concentration, log scale. Inset: Normalized mean 1 fluorescence vs aTc concentration on a linear scale. D) Growth rate distributions of cells grown at 0.1, 0.3and1ngmL aTc. E) Fraction of slow-growing, elongated cells over time at several aTc concentrations. This measurement is indicative of the extent to which cells are burdened by the high plasmid copy numbers. F) Proportion of viable cells as a function of time at various aTc concentrations. The sharp decrease in the number of viable cells at low aTc concentrations provides insight into the extent to which plasmid loss occurs at low copy numbers.

193 fects of plasmid copy number on gene expression. Cell 220 Metabolic Costs of the pUC-pTet Plasmid at the 194 cultures with this tunable copy number plasmid were 221 Single-Cell Level 195 grown in a microplate reader where simultaneous mea- 196 surements of fluorescence and growth rate could be made. 222 While macroscopic measurements show a clear interde- 223 pendence between gene expression and copy number, sin- 197 We first observe clear di↵erences in growth rates as 224 gle cell measurements can provide a more powerful path 198 a function of both plasmid copy number and aTc con- 225 towards understanding the relationship between plasmid 199 centration (Fig. 2B). We see a roughly 15 % decrease in 226 copy number and gene expression. In particular, single- 200 host growth rate at maximal induction of copy number 227 cell studies allow us to determine the distributions of gene 201 ( 50 copies per cell). This metabolic burden is signif- ⇠ 228 expression of single cells at varying plasmid copy num- 202 icantly higher than the pUC-pTet plasmid (Fig. 1E), 229 bers and directly observe the phenomena of plasmid loss 203 indicating that the majority of the reduction in growth 230 and runaway replication. 204 rate comes from the overexpression of GFP and not from 231 To first corroborate protein production measurements 205 the increased plasmid copy number. In this sense, addi- 232 with plasmid copy number, we observed the pUC-pTet- 206 tion of GFP to the plasmid provides a stronger coupling 233 sfGFP system for up to 150 generations in a microfluidic 207 between the host growth rate and plasmid copy number 234 device [21, 22]. 208 by taking up additional cellular resources. 235 In agreement with population-wide measurements 209 Despite the fitness costs associated with increasing the 236 (Fig. 2C), individual cells show an increase in the mean 210 plasmid copy number, we nevertheless observe an almost 237 fluorescence with plasmid copy number (Fig. 3A). We 211 perfect GFP induction curve when compared to the aTc 238 also observe a marked increase in the fluorescence of some 212 concentration (Fig. 2C). In addition, Fig. 2D shows a 239 cells at high aTc concentrations. Interestingly, cells with 213 nearly linear increase in the fluorescence with the plasmid 240 an increased fluorescence level are found to be elongated 214 copy number, suggesting that the metabolic costs associ- 241 (Fig. 3A and Fig. S2). The intensity distribution (Fig. 215 ated with protein overproduction only marginally impact 242 3B) undergoes a profound change as the plasmid copy 216 the overall GFP production rate. Our pUC-pTet plasmid 243 number increases: the distribution is peaked at a very 217 provides a simple way to control gene expression without 244 low intensity at low plasmid copy number but, as the 218 the need to insert an inducible promoter upstream of the 245 copy number increases, the distribution widens and a 219 gene of interest. 246 high-intensity tail emerges. 5

247 The most drastic shift in the distribution occurs be- 302 Pi library), giving us 2 separate libraries of plasmids: 1 1 248 tween 0.3ngmL and 1 ng mL ,wheretheaverage cell 303 one with a variable priming RNA transcription rate kp 249 intensity (Fig. 3C) is in good agreement with population 304 and another with a variable inhibitory RNA transcription 250 wide measurements (Fig. 2C), but each population sees a 305 rate ki. 251 steady increase in the number of individual cells with in- 306 These libraries were constructed using multiplexed 252 creased GFP levels (Fig. 3A, red arrows). As the plasmid 307 PCR and KLD assembly [23] and electroporated into 253 copy number increases further, a growing number of cells 308 competent E. Coli cells (Fig. 4A, step 2). Cell cultures 254 reach an “activated” state where they show fluorescence 309 were grown in selection conditions until they reached an 255 levels as high as those full pTet induction. Eventually, all 310 OD600 of 1.0 and subsequently diluted by a factor of 4 1 256 cells for aTc concentrations above 3 ng mL reach this 311 and grown again to an OD600 of 1.0 (Fig. 4A, step 3). 257 activated, fully-induced state (Fig. 3C). 312 Excess culture from each step of dilution was set aside 258 As the plasmid copy number sharply increases between 313 for sequencing (Fig. 4A, step 4). 1 1 259 0.1ngmL and 1 ng mL of aTc, we observe a clear 314 For each recovered mutant, the abundance of each con- 260 drop in average growth rate (Fig. 3D and Fig. S4) sim- 315 struct at each time point was used to infer the growth rate 261 ilar to bulk measurements (Fig. 2B). The distribution 316 of each individual construct. By coupling this informa- 262 of cellular growth rates (Fig. 3D) show that this drop is 317 tion with the fold change in sequencing counts from the 263 largely due to a fraction of cells that su↵er from reduced 318 library prior to transformation, we determined the rela- 264 growth rate. This is likely due to the metabolic burden of 319 tive copy numbers of every plasmid in this library (Figs. 265 plasmid maintenance and GFP expression severely lim- 320 4B-C) and their growth rates (Fig. 4D). Our method 266 iting the growth of a subset of the cellular population. 321 was able to generate 1,194 plasmids, each with a distinct 1 267 Above 1 ng mL aTc, the growth rate of healthy cells 322 plasmid copy number ranging from less than 1 copy per 268 seems to stabilize (Fig. S4). However, we notice an in- 323 cell to close to 800 copies per cell (Table S1). 269 creasing fraction of elongated cells (Fig. 3E), many of 324 To make sure our sequencing-based method yields ac- 270 which have lost the ability to grow and turned extremely 325 curate plasmid copy number measurements, we next 271 bright (supplementary movie M1). The inability of a 326 sought out to calibrate our sequencing-based numbers to 272 fraction of cells to grow may explain why growth rate 327 absolute copy numbers made using digital droplet PCR 273 continually decreases even though plasmid copy number 328 measurements. Individual constructs of 9 promoters were 274 saturates at high aTc levels in bulk measurements (Fig. 329 selected (light blue datapoints in Fig. 4B) and their ab- 275 2B). 330 solute copy number were determined via digital PCR by 276 At low plasmid copy numbers, we see a decreased num- 331 the ratio of dxs to bla genes [20]. There is a good agree- 277 ber of cells in each microfluidic chamber. In fact, nearly 332 ment between fold change in sequencing abundance and 278 every cell undergoes lysis within the first 1500 min of 333 the plasmid copy numbers as measured by digital droplet 1 279 tracking for aTc concentrations lower than 1 ng mL , 334 PCR (Fig. 4D), indicating that the fold change in abun- 280 suggesting that cells lost resistance to carbenicillin due 335 dance of sequencing counts with a modest correction ap- 281 to plasmid loss (Fig. 3F). 336 plied to correct for di↵erences in growth rates is a sound 282 These observations highlight the delicate balance be- 337 measure of plasmid copy number. 283 tween plasmid loss at low copy number and heightened 338 The simultaneous measurement of plasmid copy num- 284 metabolic burdens at high copy numbers. Our results 339 bers and the growth rates of their host cells provides 285 demonstrate that both phenomena occur stochastically 340 us with a powerful measurement of the interdependence 286 and on a per-cell basis. 341 of copy number and metabolic burden. While a high 342 plasmid copy number should reduce host cell growth, to 343 what extent that occurs is not known. Our measure- 287 Massively Parallelized Assay Determines Copy 344 ments of plasmid copy numbers and growth rates for the 288 Numbers of ColE1-derived Plasmids 345 830 plasmids in the priming promoter library, all with 346 nearly identical origins of replications, provide a power-

289 Though we were able to obtain robust control over 347 ful way to deduce this relationship. 290 plasmid copy number with our inducible plasmids, we 348 In Fig. 4E, we observe that, as the plasmid copy num- 291 next sought to exert control over plasmid copy number 349 ber increases, the growth rate of host cells decreases. 292 without the need for external inducer molecules. In order 350 A simple relationship in which the metabolic burden of 293 to survey the landscape of possible plasmid copy numbers 351 plasmid expression is linear with the copy number largely 294 that arise from changes of the transcription rates of the 352 agrees with these ﬁndings –ie. cost = a b x + b,where · · 295 RNAs controlling ColE1 replication, we simultaneously 353 x is the plasmid copy number, with a =0.063 % and 1 296 constructed 1024 di↵erent variants of the pUC19 plas- 354 b =0.023 min . This relationship can arise from a sim- 297 mid through site directed mutagenesis of the 35 and 355 ple resource allocation model in which each plasmid takes 298 10 boxes of the promoter controlling the priming RNA 356 up a small fraction of the resources necessary for cellu- 299 and at the +1 site of the priming RNA transcript (Fig. 357 lar growth proportional to the plasmid size (pUC19 is 300 4A, Pp library). The same procedure was performed on 358 2686 bp in size, which is approximately equal to 0.058% 301 the promoter controlling the inhibitory RNA (Fig. 1A, 359 the length of the E. coli genome). This results opens up 6

1) Library preparation: 2) Transformation: 3) Growth and selection: 4) Sequencing Pp Pi >103 and counting: Linearization . n

. f n . lig . . plasmids . + KLD assembly 0.25x

Electroporation 0.25x n3 Priming promoter Pp library -35 -10 |+1 OD 0.25x pUC19 : TTGAGAtcct...cgcgTAATCTgctgctT n2 Pp : TTKMMMTcct...cgcgTWWRMTgctgctN E. coli Fold-change: log2(n1/nlig) time Growth rate: (n /n )^0.25 Inhibiting promoter Pi library >300k transformants Recovery n1 i+1 i -35 -10 |+1 (1h) Plasmid copy number: pUC19 : TTGAAGtggt...cggcTACACTagaagA n1/nlig exp(-t/Tau) Pi : TTKMMMtggt...cggcTWWRMTagaagN

B C D 800

830 365 300 100

ddPCR 10

1 3 10 30 100 300 800 Sequencing Measurement

Pp Library E Pi Library Pp 0.030 Pi metabolic cost = 0.063% per Ranked promoter list Ranked promoter list 0.025 plasmid n=830 n=365

0.020

Plasmid loss Stable PCN 0.015 Linear Metabolic Burden

Growth Rate (1/min) Error > 25% 5% < Error < 25% 0.010 Error < 5%

−4 0 4 8 12 −4 0 4 8 12 1 3 0.1 10 30 Fold-change Fold-change 0.1 100 300 800 Plasmid Copy Number

Fig. 4. Massively Parallelized Copy Number Assay. A) Procedure for massively parallelized copy number assay. 1) The promoter upstream of the priming RNA was randomized with site directed mutagenesis. 2) Plasmids were ligated and cells were transformed with this library of plasmids with high eciency and grown until OD600 of 1.0. 3) At which point a fraction of the culture was diluted and regrown and the remaining fraction was saved for plasmid extraction. 4) Sequencing libraries were generated from isolated plasmids and sequenced using next generation sequencing. Growth rates and Plasmid Copy Numbers (PCN) were then calculated from the frequencies of sequencing counts present at each time point. B) Ranked list of the measured fold-change for plasmids generated through mutations to the promoter driving the priming RNA in this work (n = 830). Promoters with a fold-change below zero will result in plasmid loss (PCN¡1). (lower) The distribution of copy numbers arising from mutations to the priming RNA promoter. C) Ranked list of the measured fold-change for plasmids generated through mutations to the promoter driving the inhibitory RNA in this work (n = 365). D) Agreement between plasmid copy number as measured by sequencing counts and digital droplet PCR. E) The relationship between plasmid copy number and growth rate as measured across 830 variants of the pUC19 plasmid. The size of each data point is inversely proportional to the error in the measurement, with the top, middle two and lowest quartiles representing an error of less than 5%, between 5% and 25%, and more than 25%, respectively. Data from Fig. 1E is plotted (red to blue gradient points) A linear metabolic burden seems to account for the observed relation reduction in growth rate with increasing copy number.

360 the possibility for improved models of plasmid evolution 368 Plasmid Copy Number Controls Violacein 361 and replication, as they indicate that plasmids impose a 369 Biosynthesis 362 metabolic burden that is linear with their copy number

363 on their hosts. Knowledge of this relationship between 370 We sought to demonstrate the potential for using plas- 364 host growth and plasmid copy number may shed light on 371 mid copy number to control complex biosynthetic pro- 365 the evolution of cis and trans acting regulatory elements 372 cesses comprising several proteins and enzymes. We 366 as well as the stringency of plasmid copy number control 373 chose to focus on Violacein (Fig. 5A), a purple pigment 367 mechanisms. 374 formed by the condensation of two tryptophan molecules 375 [24] which is commonly used to demonstrate biosynthe- 376 sis optimization [25]. The violacein biosynthetic pathway 377 contains ﬁve enzymes, VioABCDE, each under the con- 378 trol of a di↵erent promoter and spanning approximately 7

396 A VioD ter this library was transformed into competent cells, we VioB O VioE VioC O O O O 397 picked individual colonies from a plate and sequenced N N N O O N N VioA VioB N VioE N N VioD VioC O N N 398 their plasmids in order to infer their copy number from N O N O O N VioA O O VioC N 399 the Pp library data (Table S1). We also measured the vi- Tryptophan VioB Violacein 400 olacein production from cell extracts and found a clearly 401 positive relation between the plasmid copy number and

B D 402 VioABCDE pathway VioABCDE pathway Violacein production (Fig. 5E). Interestingly, we ob- 403 served a marked decrease in violacein yields as the plas- VioA VioB VioC VioD VioE 404 Gibson assembly Pp library mid copy number is increased above 500 copies per cell, 7.754kb insert (1024 sequences) 405 and growth curve measurements performed using a plate

ampR 406 reader show that the decreased yield can be attributed 407 to a longer lag time (Fig. 5E, inset). Our results show pTet KLD assembly 408 that using tunable copy number thus constitutes a gen- ampR >10k transformants 409 High PCN eralizable method for the scaling-up and optimization of Low PCN High PCN Low PCN variants 410 complex biosynthetic pathways. [aTc] variants

Violacein production Violacein production

C E 411 Control of a CRISPRi system through sponge 100 412 binding sites 1.00 400

0.75 200 413 Controlling the plasmid copy number may also be use- −1 Lag Time (mins) 0.5 1.0 10 0.50 Viol. Yield 414 ful in synthetic biology and in the creation of robust 0.25 415 gene circuits [3, 5, 27, 28]. To test this, we used a Violacein Yield (AU) Violacein Yield (AU) 416 simple CRISPR-dCas12a inverter to demonstrate the ef- 10−2 0.00 0 0.01 0.1 1 10 100 5 417 fect of adding “sponge” binding sites to a simple genetic 50 aTc concentration (ng/mL) 100 300 500 Plasmid Copy Number 418 circuit through manipulation of plasmid copy numbers. 419 Sponge sites are additional binding sites used to titrate Fig. 5. Optimization of Violacein Production by Tuning Plas- 420 out (“soak up”) transcription factors, and they have been mid Copy Number. B) The complete VioABCDE pathway 421 used to redirect flux in metabolic pathways to optimize was inserted in the pUC-pTet plasmid. C) Controlling vio- 422 arginine production [29], to activate silent biosynthetic lacein biosynthesis by increasing aTc concentrations. D) A 423 gene clusters [30], to tune gene expression timing [31], library of variable copy number plasmids was created by per- 424 and mitigate protein toxicity [32]. In these studies, vari- forming mutagenesis on the priming promoter of a ColE1- 425 able numbers of sponge sites are placed on high or low based plasmid containing the VioABCDE pathway. E) Viola- 426 copy number plasmids. cein production peaks as the plasmid copy number nears 300, 427 Here, we added a sponge site to a few isolates from with a marked decrease in the yield as plasmid copy number is increased beyond 500. Inset: relationship between the vio- 428 our tunable copy number library to alter the qualita- lacein yield and lag time of the population when grown in a 429 tive behavior of a simple genetic circuit. Specifically, we plate reader. 430 use a two plasmid system in which a low-copy plasmid 431 (pSC101, 2-5 copies per cell) contains an aTc inducible 432 CRISPR-Cas12a guide RNA which can bind to an active 433 site Sa and turn OFF a promoter driving sfGFP, while 379 8kb of DNA (Fig. 5B). Hence, we reasoned that tuning 434 the second plasmid contains a decoy binding site Sd site 380 the plasmid copy number of the whole Violacein produc- 435 that sequesters CRISPR-Cas12a molecules away from the 381 tion pathway may serve as an e↵ective method to scale 436 active site Sa. The sponge plasmids were picked out from 382 up the expression of VioABCDE while keeping the pro- 437 our library of variable copy number plasmids generated 383 duction of each enzyme under the same stoichiometric 438 in Fig. 4B and span copy numbers from 30 to 270. 384 ratio. 439 Time-series measurements of the fully-induced 385 To achieve this, constructed a pTet inducible plasmid 440 CRISPRi inverter show that the OFF state of the 386 harboring the Violacein biosynthesis pathway using Gib- 441 inverter has a higher fluorescence level as the number of 387 son assembly [26] (Fig. 5B). We then grew cells with the 442 sponge sites is increased (Fig. 6A). This suggests that 388 pUC-pTet-Vio plasmid at increasing aTc concentrations 443 the addition of sponge sites reduces the occupancy of 389 and observed a clear increase in the violacein production 444 the active site Sa and decreases the repression eciency 390 with increasing aTc concentrations (Fig. 5C). 445 of the CRISPRi inverter. 391 Additionally, we inserted the VioABCDE pathway into 446 We also observe a drastic reduction in the dynamic 392 a pUC plasmid under the control of a moderate-strength 447 range of circuits as the number of sponge sites is increased 393 promoter, and we used site directed mutagenesis of the 448 (Fig. 6B), where cells with more than 150 sponge sites 394 origin of replication, in the manner described above, to 449 have at least a 40% reduction in dynamic range compared 395 generate a plasmid copy number library (Fig. 5D). Af- 450 to those with no sponge sites. In contrast, the addition 8

Competitive binding FnCas12a S d 1 470 tion, and synthetic biology applications. Using a combi- A S B d 2 Sponge sites 471 Sd nation of high-throughput measurements and single-cell CRISPRi 3 Tunable plasmid 1.2 Ptet copy number Nc

... with decoy binding crRNA Sd 472 Sa Nc sites S . experiments, we deduce relations between plasmid copy P d A 1.0 M sfGFP 473 number and host cell’s growth rate, protein expression, 0.8 150 474 biosynthesis, and the performance of simple regulatory Nc = 0 Nc = 30 0.6 475 elements. This system could help resolve known issues 100 Nc = 50 Nc = 170 0.4 Nc = 0 476 currently plaguing the design and scaling up of synthetic Nc = 270 Nc = 30 Nc = 50 477 50 Fluorescence Level 0.2 Nc = 170 gene circuits such as leaky expression, narrow ranges of Nc = 270 Fluo. Intensity 478 activation and repression control, and a large metabolic 0.0 0 [aTc] = 1ng/mL 0 2 4 6 10−4 10−2 100 102 479 burden due to competition over a common pool of re- Time (h) aTc (ng/mL) C D 480 sources [34–36]. 100 ON competitor site 481 Using a massively parallel directed mutagenesis ap- 200 75 50 482 proach, we ﬁnd that the most straightforward and re-

25 active site 483 OFF liable way to change the plasmid copy number in the Active site

On/Off level 0 0 0 100 200 300 occupancy (%) 0 100 200 300 484 ColE1 Origin is through mutations to the promoter con- Sponge copy number Nc Sponge copy number Nc 485 trolling the priming RNA. Further, changes to the ma- 486 chinery controlling inhibitory RNA levels are much less Fig. 6. Control of a CRISPRi Inverter with Competitor Bind- 487 likely to give stable constructs given the fact that muta- ing Sites. A) Top: Diagram showing how production of a 488 tions to the promoter controlling inhibition change the dCas12a CRISPR RNA (crRNA) is controlled by an aTc- 489 sequence of the priming transcript can have drastic ef- inducible promoter, and the dCas12a+crRNA duplex can ei- 490 fects on plasmid copy numbers through secondary struc- ther bind to the active site Sa, which turns o↵production of 491 ture changes [37]. Overall, these results suggest that the sfGFP, or to a decoy site Sd residing on a “sponge” plasmid. Bottom: Fluorescence intensity vs time for CRISPR-dCas12a 492 ColE1 replication machinery is more robust to changes in systems with variable numbers of sponge sites (Nc). B) In- 493 the priming RNA transcription initiation rate and much duction curves for a CRISPRi system with sponge sites on 494 less robust to changes in the parameters detailing inhibi- plasmid with varying copy numbers. C) ON-OFF levels for 495 tion. variable numbers of sponge sites. The ON and OFF ﬂuo- 496 Our multiplexed method of generating a plasmid copy rescence level of the inverter both increase as the number of 497 number library allows one to rapidly screen the e↵ects sponge sites varies from 0 to 270. The dynamic range de- 498 of plasmid copy number on a given system and could creases with increasing sponge copy number. D) Active site occupancy vs number of sponge sites, as the number of sponge 499 be tailored to select for genetic devices that perform the sites increases the active site becomes less occupied. The 500 best under a given set of conditions. A key ﬁnding of this gray area represents the numerical solution of a fold-change 501 facet of the work is that ColE1-type plasmids impose a model [14, 23] for dCas12a where the binding energy equal to 502 linear metabolic burden on their hosts proportional to 13.25 kBT(top)and 12.25 kBT(bottom). 503 the relative size of the plasmid, which may shed light on 504 the evolution of various regulatory elements in inhibitor- 505 dilution copy number control mechanisms.

451 of 30 to 50 sponge sites only reduces the dynamic range 506 Our single cell measurements also show that plasmid 452 by approximately 6-8% (Fig. 6C). 507 loss and runaway replication are quite common phenom- 453 While a large dynamic range is often desirable in ge- 508 ena. These ﬁndings are in line with recent single cell 454 netic devices, a reduction in dynamic range could be used 509 measurements of plasmid copy numbers [6] and illustrate 455 to divert the ﬂux in a metabolic pathway, ensure that the 510 that very careful attention should be given to the plas- 456 concentration of some protein stays within a given range, 511 mids used to house genetic circuits in order to ensure re- 457 or to reduce the toxicity of CRISPR-Cas proteins [33] due 512 liability. Since we observed an increasingly large fraction 458 to spurious o↵-target binding. Using a simple model of 513 of non-growing cells as we increased the plasmid copy 459 protein-DNA binding [14], we can discover that the occu- 514 number, our results further suggest that lower growth 460 pancy of the active binding site Sa (Fig. 6D) in the OFF 515 rate for populations carrying a high-copy number plas- 461 state is reduced from nearly 100 % with zero sponge sites 516 mid may be explained by non-growing cells as opposed 462 to 50 % when 270 sponge sites are present. Overall, we 517 to a universal reduction in cellular growth rates. While 463 demonstrate that plasmid copy number can be used to 518 this work is restricted to the ColE1 origin of replication, 464 accurately tune the level of transcription factor titration 519 similar principles can be used to tune the copy numbers 465 in a simple genetic inverter. 520 of other inhibitor-dilution controlled plasmids.

521 The implications of plasmid copy number go beyond 522 its e↵ects on gene expression and host cell ﬁtness that 466 DISCUSSION 523 we’ve presented thus far. As vehicles of horizontal gene 524 transfer, the prevalence of which increases with plasmid 467 This work details the development of a system of tun- 525 copy number, plasmids play a key role in bacterial ecol- 468 able copy number plasmids and demonstrates their util- 526 ogy and evolution[38]. On the other hand, a consequence 469 ity in protein expression, biosynthetic pathway optimiza- 527 of a high copy number is a reduced retention and ﬁxation 9

528 of novel plasmid variants [39]. Together, these two issues 582 of 1.0, a procedure that was repeated 4 times. Excess 529 create a complicated image of the role of plasmid copy 583 culture from each time point was set aside for plasmid 530 number in bacterial diversity and evolution, it’s possible 584 isolation. Plasmids were isolated by miniprep (Zymo) 531 that the methods for controlling plasmid copy number 585 and a 243 bp region of the ori was amplified (pBR322- 532 presented here can be used to help shed light on the role 586 5prime-promoter1-rev 5’-TCTACACTCT TTCCCTA- 533 of plasmid replication in bacterial evolution. 587 CAC GACGCTCTTC CGaTcTCAGA CCCCGTA- 534 The evolution of the fitness cost associated with 588 GAA AAGaTcAAAG GaTcTTC-3’, pBR322-3prime-for 535 plasmid-borne antimicrobial resistance genes is of cen- 589 5’-GACTGGAGTT CAGACGTGTG CTCTTCCGAT 536 tral importance in preventing the spread of drug-resistant 590 CTCGAGGTAT GTAGGCGGTG CTACA-3’) via PCR, 537 bacteria [40]. Our work presented here could provide 591 at which point universal primers and indices were at- 538 a clear path for investigating the role of plasmid copy 592 tached to the segment using an additional PCR step 539 number in the evolution of the cost of antimicrobial re- 593 (primers). Prepared libraries were then diluted to 50pM 540 sistance. Further, the fitness cost of plasmid expression 594 and sequenced on the Illumina iSeq using overlapping 541 is highly species-dependent and dependent on the diver- 595 2 150 bp overlapping reads. ⇥ 542 sity of a bacterial community [38] and recent findings 543 have shown that plasmids can quickly achieve fixation 544 in a population under non-selective conditions [41], fur- 596 Construction of inducible plasmids 545 ther understanding of plasmid replication mechanisms 546 and copy number e↵ects might allow for a more nuanced 597 547 understanding of these varied properties of plasmids. The pUC-pTet plasmid, whose copy number is in- 598 ducible in the presence of aTc was constructed using a 548 Together, these findings demonstrate that plasmid 599 PCR-based insertion as follows. Primers (pTet-tn10-for 549 copy number can have drastic e↵ects on host cellular pro- 600 5’-AAATAACTCT aTcAATGATA GAGTGTCAAG 550 cesses and provide straightforward ways of investigating 601 AAGaTcCTTT GaTcTTTTCT ACGGGGTCTG A- 551 these e↵ects in other systems. Our results underscore 602 3’, pTet-tn10-rev 5’-TACCACTCCC TaTcAGTGAT 552 the importance of tuning plasmid copy number for the 603 AGAGATaTcT GCAAACAAAA AAACCACCGC 553 optimization of biosynthetic pathways, gene regulatory 604 TACCAGC-3’) overlapping with the pUC19 origin of 554 circuits, and other synthetic biological systems. 605 replication and containing a pTet-inducible promoter 606 were used to linearize and amplify the pUC19 backbone. 607 The linearized plasmid was then circularized using a 555 MATERIALS AND METHODS 608 KLD reaction (NEB) and transformed into competent 609 cells via electroporation. During the recovery period 556 Next-generation sequencing library preparation 610 1x aTc was added to the recovery media to facilitate 611 replication of the plasmid. 557 The multiplexed libraries of plasmids was generated by 612 The pUC-pLac-RNAi plasmid, whose copy number 558 site-directed mutagenesis using the NEB Q5 High-fidelity 613 can be reduced through the addition of IPTG, was con- 559 polymerase. Primers (pBR322-promoter1-1024-rev 5’- 614 structed as follows. The inhibitory RNA (commonly 560 GAKKKMAAGA AGaTcCTTTG aTcTTTTCTA 615 called RNAI) from the pUC19 plasmid was isolated via 561 CGGGGTCTG-3’,pBR322-promoter1-1024-for 5’- 616 PCR. In the process 15 bp overhangs were added to facil- 562 CTTTTTTTCT GCGCGTWWRM TGCTGCTNGC 617 itate Gibson assembly. Another pUC19 plasmid was the 563 AAACAAAAAA ACCACCG-3, pBR322-promoter1- 618 linearized downstream of the pLac promoter (primers) 564 1024-rev 5’-GTTAGGCCAC CAKKKMAAGA ACTCT- 619 and the linearized plasmid together with the inhibitory 565 GTAGC ACCGCCTACA TACC-3’, pBR322-promoter1- 620 RNA insert were assembled via Gibson assembly at 50 C 566 1024-for 5’-TACGGCTWWR MTAGAAGNAC AG- 621 for 1 h and transformed into competent cells via electro- 567 TATTTGGT aTcTGCGCTC TGCTG-3’ ) containing 622 poration. 568 degenerate nucleotides in the -35, -10, and +1 sites of 569 the promoters controlling transcription of both RNAs 570 were used to amplify the standard pUC19 plasmid 623 571 (NEB). After PCR amplification samples were digested Copy number and growth rate measurement via 624 next-generation sequencing reads 572 with DpnI (NEB) restriction enzyme to remove excess 573 pUC19 template. 574 The PCR product was then circularized with electroli- 625 To determine the relative plasmid copy number from 575 gase (NEB) and transformed with high eciency ( 1.5 626 the abundance of next generation sequencing reads we 576 million CFU/mL) into Endura Electrocompetent cells. 627 compare the fraction of the sequencing reads at several 577 The transformation eciency was determined by plat- 628 time points with that in the initial library. The ratio of 578 ing of serial dilutions of recovered cells. After trans- 629 the two, with a small correction applied to account for 579 formation liquid cultures were grown in Terrific Broth 630 di↵erences in the growth rates of cells harboring di↵erent 580 (TB, VWR) until reaching an OD600 of 1.0, at which 631 plasmids, is what we report as a relative plasmid copy 581 point cultures were diluted 1:4 and regrown to OD600 632 number. 10

633 We measure the initial fraction of the reads in the liga- 670 Colonies hosting the appropriate constructs were picked 634 tion product (fi) and the final fractions of the reads of the 671 o↵of plates after transformation and grown in 3 mL of 1 1 635 plasmids harvested at di↵erent time points (↵,f2,f3,f4). 672 TB or H media (10 g L tryptone, 8 g L NaCl) to an 636 Di↵erences in the doubling time () can be measured from 673 OD600 of 1.0. Cells were then serially diluted 8000-fold 637 di↵erences in the fractions of reads from points separated 674 in nuclease free ddH2O. Cells were lysed by heating at 638 in time. These relationships can be described in the equa- 675 95 C for 20 min. After lysis, 1 µL of the lysate was used 639 tions below. First, to determine the doubling time from 676 in each digital PCR reaction. Two digital PCR reactions 640 adjacent time points: 677 were performed per sample - one targeting the genomic 678 dxs gene and another targeting the beta lactamase bla t( 1 1 ) fn = fn 12 ⌧ ⌧av 679 gene on the plasmid (primers). This defines the plas- 680 mid copy number as the number of plasmids per genome. 641 The above line states that the fraction at a given time 681 After generating droplets on the QX200 droplet genera- 642 point is equal to the fraction at the previous time point 682 tor (Biorad), reactions were thermal cycled at 95 C for 643 multiplied by the di↵erence in the number of doublings 683 10 min, followed by 35 cycles of 98 C for 30 s, 58 C for 644 from the bulk culture. Below we have rearranged the 684 30 s, and 72 C for 1 min. Samples were held at 4 C 645 equation to solve for the growth rate from the variables 685 after the reaction was completed and before droplets 646 we can either observe or control. 686 were read. Droplets were read with the QX-200 Droplet 687 Reader (Biorad) and concentrations were determined us- 688 ing the Quantasoft analysis pro software (Biorad). Log ( fn ) 2 fn 1 1 1 + = t ⌧av ⌧

689 PCN Sponge E↵ects on CRISPRi Inverter 647 Secondly, to determine the copy number from the 648 growth rate and the di↵erence in sequencing counts we 649 express the fraction at the ﬁrst time point as the product 690 CRISPR system with nuclease activity deactivated Cas 650 of the fraction in the ligation, the relative copy number, 691 protein(dCas) can be used to interfere with transcrip- 651 and the number of excess doubling times: 692 tion regulation (CRISPRi) by design crRNA targeting 693 at the promoter for the gene of interest which acts a 694 repressor of gene. However, dCas/crRNA complex can t( 1 1 ) ff = fiC2 ⌧ ⌧av 695 also bind to other sites that share the same DNA se-

652 696 quence as the target but does not directly regulate the 697 gene (these are called competitor sites). When the sec- fn t( 1 1 ) 2 ⌧ ⌧av = C 698 ond scenario happens, it will have an impact on the re- fi 699 pression level of the gene at a given concentration of 700 dCas/crRNA, i.e., a sponge e↵ect for the gene of inter- 653 By inserting the above expression for the doubling 701 est by consuming dCas/crRNA from the same pool. To 654 time, we can have the relative copy number solely in 702 test the sponge e↵ect at di↵erent numbers of competitor 655 terms of the fractions of each sequencing reads and the 703 sites, a dual-plasmid system was used. The main plas- 656 ratio of the time between time points and the time be- 704 mid(Kanamycin resistant) is a low copy(pSC101) pTet- 657 tween transformation and the ﬁrst time point. 705 inducible CRISPRi-dFnCas12a inverter, where the cr- 706 RNA transcribes by pTet target at a promoter that ff fn 1 t 707 drivers a green ﬂuorescence protein(sfGFP). The sec- 2 t = C 708 ond plasmid is edited from the above-mentioned pUC19- fi fn 709 PCN library by inserting a single competitor via standard 658 This analysis hinges on the assumption that upon 710 Site-Directed Mutagenesis(NEB, E0554). Two plasmids 659 transformation each cell receives a single plasmid, that 711 were co-transformed into the GL002 strain that has dFn- 660 cells do not transfer plasmids between each other, and 712 Cas12a and TetR integrated in its genome with Mix & 661 that the number of transformants of a given plasmid is 713 Go! E.coli Transformation Kit(T3001). Five di↵erent 662 proportional to the amount of that construct present in 714 PCN-plasmids were tested with the same main inverter 663 the initial electroligation product. Further, this analy- 715 plasmid. The expression of sfGFP from the dual-plasmid 664 sis assumes that cells harboring di↵erent constructs have 716 system with di↵erent PCN was individually measured at 665 identical lag times. 717 a gradient of 20 aTc concentrations on the BioTek (Syn- 718 ergy H1) plate reader in H medium. Three biological 719 replicates were used for each system, i.e., three di↵er- 666 Absolute Copy Number Determination via digital 720 ent colonies were picked up and grown in 2 mL H media 667 droplet PCR (ddPCR) 721 without aTc for approximately 18 hours. Then, 100 µL 722 of overnight cell culture was used and diluted into 2 mL 668 Digital Droplet PCR (ddPCR) was used to corroborate 723 fresh H media before transferring to a 96 well plate where 669 our sequencing measurements of plasmid copy number. 724 each of the middle of 10 x 6 wells contains 1 µL of diluted 11

725 cell culture and 200 µL fresh H media with corresponding 771 refrigerated. In order to start the experiment, 3 µL of 726 aTc concentration. 772 the stock colony was dispersed in 3 mL of H media with 773 plasmid selecting antibiotic carbenicillin and aTc. The 774 cells were grown to OD 0.5 0.1, concentrated by cen- 600 ± 727 Growth Rate Measurements 775 trifugation and inserted in plasma cleaned microﬂuidic 776 chips.

728 Growth rates of individual constructs were measured 777 After entering the chip bacteria populated and grew 729 on the BioTek (Synergy H1) plate reader. Overnight cul- 778 in 60 µm long, 7.5 µm wide and 1.05 µm thick chambers 730 tures of cells grown to saturation were serially diluted 779 connected to a supply channel flowing fresh media pres- 731 by a factor of 100,000 into well-mixed 200 µL aliquots of 780 surized to 4 psi [22]. On top of H media, carbenicillin 1 732 media containing the appropriate antibiotics and inducer 781 and aTc, 0.1gL of bovine serum albumin was added 733 concentrations. Cells were grown at 37 C for 36 h while 782 to the media to improve evacuation of cells from the sup- 734 to OD600 and fluorescence (if appropriate) was recorded 783 ply channel [42]. 735 every 3 min. Growth rates were calculated from the re- 784 Single cells were resolved by epi-fluorescence mi- 736 sultant time series data by averaging a window of 6min 785 croscopy of sfGFP with a 100 x, 1.4 NA apochromat Le- ± 737 around the maximal growth rate, calculated as a discrete 786 ica objective. Single cell detection and tracking enabled 738 di↵erence of OD600 readings between time points. 787 measurement of cell dimensions, intensity and growth 788 rate overtime. Cells can be classified in three categories: 789 (1) non dividing dark cells (loss of antibiotic resistance 739 Gene Expression Measurements 790 due to plasmid loss), (2) healthy cells and (3) non grow- 791 ing bright cells (likely due to plasmid overload). Whereas

740 The expression of green fluorescent protein (sfGFP) 792 in bulk pathological cells of type (1) or (3) would quickly 741 was used as a measure of gene expression in this work. 793 be overgrown by healthy cells, in chambers, we find that 742 Fluorescence measurements were taken on cells prepared 794 they can get stuck at the back or even populate the whole 743 as described above as they grew. Gene expression was 795 chamber, rendering the chamber unusable when the cells 744 calculated as follows - the average fluorescence in a small 796 eventually die. This is particularly true for non divid- 745 window surrounding the maximal growth rate was cal- 797 ing dark cells at low aTc concentration. By using 7.5 µm 746 culated and divided by the integral of the OD600 in the 798 wide chambers, we allow many cells to grow side by side, 747 same window. This can be interpreted as the fluores- 799 which increases the chances of healthy cells to repopulate 748 cence (ie gene expression) per cell and should be indepen- 800 the chambers. However, this is not enough to maintain 749 dent of the growth rate since we divide by the integral 801 healthy chambers for more than a few hours at low aTc 1 750 of the OD600. Six replicates of each inducer concentra- 802 concentration. For aTc concentrations of 1 ng mL and 751 tion were taken and the average and standard deviation 803 above a majority of chambers last for the whole experi- 752 of this measure of gene expression were reported. 804 ment time of 5000 min. We base intensity measurements 805 on growing cells only (growth rate larger than 1/3 of 806 mean growth rate) to prevent over representation of non

753 Violacein Production Measurements 807 growing cells that can reach up to 40% of the detected 808 cells (Fig. 3F). Selected cells are highlighted in the sup- 809 plementary movie. 754 Cultures of single colonies hosting plasmids containing 810 Due to the rapid loss of chambers we are unable to 755 the violacein biosynthesis pathway were grown overnight 811 measure a stabilized fluorescence level for aTc concen- 756 in Terrific Broth with appropriate antibiotics. First, 1 812 tration below 1 ng mL . However, we can still observe 757 200 µL of liquid culture was centrifuged and cells were 813 that the aTc concentration promotes plasmid replication. 758 resuspended in 50 µL of water. Then, 450 µL of methanol 814 Indeed, the chamber survives longer at higher aTc con- 759 was added to the resuspended cells and they were shaken 815 centration (plasmid loss is less frequent) and the average 760 for 1 h at room temperature. Cells were spun down and 816 fluorescence increases with aTc. For aTc concentration 761 200 L of the supernatant was transferred to a 96-well µ 1 1 817 of 1 ng mL to 100 ng mL time dependent fluorescence 762 plate where the absorbance at 650 nm was used to quan- 818 and growth rate curves (Figure S2) stabilize after about 763 tify violacein production. 819 600 min (i.e. 12 generations) and are expected to accu- 820 rately represent PCN in exponential growth condition.

764 Single cell microscopy 821 All experiments and cell cultures were performed at 822 37 C.

765 The single cell data was collected within a month over 766 8 subsequent experiments with varying aTc concentra- 823 ACKNOWLEDGMENTS 767 tions. Each experiment was run for 1300 to 5000 min, 768 corresponding to at least 20 to 80 generations. All exper- 769 iments were started from the same stock bacterial colony, 824 We wish to thank David Specht for helpful discussions 1 770 initially grown to OD600 0.5 at 10 ng mL aTc and kept 825 and critical review of the manuscript. We also thank 12

826 members of the Lambert laboratory for providing feed- 828 NIH funding under 1R35 GM133759 Maximizing Inves- 827 back on the manuscript. This work was supported by 829 tigators’ Research Award (MIRA).

830 [1] A. P. Togna, M. L. Shuler, and D. B. Wilson, Ef- 887 M. Rydenfelt, and R. Phillips, The Transcription Factor 831 fects of Plasmid Copy Number and Runaway Plas- 888 Titration E↵ect Dictates Level of Gene Expression, Cell 832 mid Replication on Overproduction and Excretion of - 889 156,1312(2014). 833 Lactamase from Escherichia coli, Biotechnology Progress 890 [15] J. Paulsson, K. Nordström, and M. Ehrenberg, Require- 834 9,31(1993). 891 ments for Rapid Plasmid ColE1 Copy Number Adjust- 835 [2] K. Nordström and B. E. Uhlin, Runaway-replication plas- 892 ments: A Mathematical Model of Inhibition Modes and 836 mids as tools to produce large quantities of proteins from 893 RNA Turnover Rates, Plasmid 39,215(1998). 837 cloned genes in bacteria, Bio/Technology (Nature Pub- 894 [16] G. del Solar, R. Giraldo, M. J. Ruiz-Echevarr´ıa, M. Es- 838 lishing Company) 10,661(1992). 895 pinosa, and R. D´ıaz-Orejas, Replication and Control of 839 [3] M. Dwidar and Y. Yokobayashi, Riboswitch Signal Am- 896 Circular Bacterial Plasmids, Microbiology and Molecular 840 plification by Controlling Plasmid Copy Number, ACS 897 Biology Reviews 62,434(1998). 841 Synthetic Biology 8,245(2019). 898 [17] J.-i. Tomizawa, Control of cole1 plasmid replication: Ini- 842 [4] R. U. Sheth, S. S. Yim, F. L. Wu, and H. H. Wang, Mul- 899 tial interaction of RNA I and the primer transcript is 843 tiplex recording of cellular events over time on CRISPR 900 reversible, Cell 40,527(1985). 844 biological tape, Science 358,10.1126/science.aao0958901 [18] J. Paulsson and M. Ehrenberg, Noise in a minimal regu- 845 (2017). 902 latory network: plasmid copy number control, Quarterly 846 [5] L. Potvin-Trottier, N. D. Lord, G. Vinnicombe, and 903 Reviews of Biophysics 34,1(2001). 847 J. Paulsson, Synchronous long-term oscillations in a syn- 904 [19] J. Paulsson, Multileveled selection on plasmid replica- 848 thetic gene circuit, Nature 538,514(2016). 905 tion., Genetics 161,1373(2002). 849 [6] B. Shao, J. Rammohan, D. A. Anderson, N. Alper- 906 [20] M. Plotka, M. Wozniak, and T. Kaczorowski, Quan- 850 ovich, D. Ross, and C. A. Voigt, Single-cell measurement 907 tification of Plasmid Copy Number with Single Colour 851 of plasmid copy number and promoter activity, Nature 908 Droplet Digital PCR, PLoS ONE 12, 10.1371/jour- 852 Communications 12,1475(2021). 909 nal.pone.0169846 (2017). 853 [7] J. C. Diaz Ricci and M. E. Hernández, Plasmid Ef- 910 [21] P. Wang, L. Robert, J. Pelletier, W. L. Dang, F. Taddei, 854 fects on Escherichia coli Metabolism, Critical Reviews 911 A. Wright, and S. Jun, Robust Growth of Escherichia 855 in Biotechnology 20,79(2000). 912 coli, Current Biology 20,1099(2010). 856 [8] A. S. Karim, K. A. Curran, and H. S. Alper, Charac- 913 [22] G. Lambert and E. Kussell, Memory and Fitness Op- 857 terization of plasmid burden and copy number in Sac- 914 timization of Bacteria under Fluctuating Environments, 858 charomyces cerevisiae for optimization of metabolic en- 915 PLoS Genet 10,e1004556(2014). 859 gineering applications, FEMS Yeast Research 13,107 916 [23] D. A. Specht, Y. Xu, and G. Lambert, Massively parallel 860 (2013). 917 CRISPRi assays reveal concealed thermodynamic deter- 861 [9] D. S.-W. Ow, P. M. Nissom, R. Philp, S. K.-W. Oh, 918 minants of dCas12a binding, Proceedings of the National 862 and M. G.-S. Yap, Global transcriptional analysis of 919 Academy of Sciences 117,11274(2020). 863 metabolic burden due to plasmid maintenance in Es- 920 [24] H. C. Lichstein and V. F. Van De Sand, Violacein, an 864 cherichia coli DH5 during batch fermentation, Enzyme 921 Antibiotic Pigment Produced by Chromobacterium Vio- 865 and Microbial Technology 39,391(2006). 922 laceum, The Journal of Infectious Diseases 76,47(1945). 866 [10] W. E. Bentley, N. Mirjalili, D. C. Andersen, R. H. Davis, 923 [25] M. Kholany, P. Trébulle, M. Martins, S. P. Ventura, 867 and D. S. Kompala, Plasmid-encoded protein: The prin- 924 J. Nicaud, and J. A. Coutinho, Extraction and purifica- 868 cipal factor in the “metabolic burden” associated with 925 tion of violacein from Yarrowia lipolytica cells using aque- 869 recombinant bacteria, Biotechnology and Bioengineering 926 ous solutions of surfactants, Journal of Chemical Tech- 870 35,668(1990). 927 nology & Biotechnology , jctb.6297 (2019). 871 [11] M. Pasini, A. Fernández-Castané, A. Jaramillo, 928 [26] D. G. Gibson, L. Young, R.-Y. Chuang, J. C. Venter, 872 C. de Mas, G. Caminal, and P. Ferrer, Using promoter 929 C. A. Hutchison, and H. O. Smith, Enzymatic assem- 873 libraries to reduce metabolic burden due to plasmid- 930 bly of DNA molecules up to several hundred kilobases, 874 encoded proteins in recombinant Escherichia coli, New 931 Nature Methods 6,343(2009). 875 Biotechnology 33,78(2016). 932 [27] M. R. Atkinson, M. A. Savageau, J. T. Myers, and A. J. 876 [12] A. Rozkov, C. A. Avignone-Rossa, P. F. Ertl, P. Jones, 933 Ninfa, Development of Genetic Circuitry Exhibiting Tog- 877 R. D. O’Kennedy, J. J. Smith, J. W. Dale, and M. E. 934 gle Switch or Oscillatory Behavior in Escherichia coli, 878 Bushell, Characterization of the metabolic burden on Es- 935 Cell 113,597(2003). 879 cherichia coli DH1 cells imposed by the presence of a plas- 936 [28] Y. Mileyko, R. I. Joh, and J. S. Weitz, Small-scale copy 880 mid containing a gene therapy sequence, Biotechnology 937 number variation and large-scale changes in gene expres- 881 and Bioengineering 88,909(2004). 938 sion, Proceedings of the National Academy of Sciences 882 [13] F. M. Weinert, R. C. Brewster, M. Rydenfelt, R. Phillips, 939 105,16659(2008). 883 and W. K. Kegel, Scaling of Gene Expression with 940 [29] T. Wang, N. Tague, S. A. Whelan, and M. J. Dunlop, 884 Transcription-Factor Fugacity, Physical review letters 941 Programmable gene regulation for metabolic engineer- 885 113,258101(2014). 942 ing using decoy transcription factor binding sites, Nucleic 886 [14] R. C. Brewster, F. M. Weinert, H. G. Garcia, D. Song, 943 Acids Research 49,1163(2021). 13

944 [30] B. Wang, F. Guo, S.-H. Dong, and H. Zhao, Activation of 945 silent biosynthetic gene clusters using transcription factor 946 decoys, Nature Chemical Biology 15,111(2019). 947 [31] M. Z. Ali and R. C. Brewster, Controlling gene expression 948 timing through gene regulatory architecture, bioRxiv , 949 2021.04.09.439163 (2021). 950 [32] X. Wan, F. Pinto, L. Yu, and B. Wang, Synthetic protein- 951 binding DNA sponge as a tool to tune gene expression 952 and mitigate protein toxicity, Nature Communications 953 11,5961(2020). 954 [33] S. Zhang and C. A. Voigt, Engineered dCas9 with re- 955 duced toxicity in bacteria: implications for genetic circuit 956 design, Nucleic Acids Research 46,11115(2018). 957 [34] C. Tan, P. Marguet, and L. You, Emergent bistability 958 by a growth-modulating positive feedback circuit, Nature 959 Chemical Biology 5,842(2009). 960 [35] Y. Qian, H.-H. Huang, J. I. Jiménez, and D. Del Vecchio, 961 Resource Competition Shapes the Response of Genetic 962 Circuits, ACS Synthetic Biology 6,1263(2017). 963 [36] F. Ceroni, R. Algar, G.-B. Stan, and T. Ellis, Quantifying 964 cellular capacity identifies gene expression designs with 965 reduced burden, Nature Methods 12,415(2015). 966 [37] J. Tomizawa and T. Itoh, Plasmid ColE1 incompatibility 967 determined by interaction of RNA I with primer tran- 968 script., Proceedings of the National Academy of Sciences 969 78,6096(1981). 970 [38] J. Rodr´ıguez-Beltrán, J. DelaFuente, R. León-Sampedro, 971 R. C. MacLean, and San Millán, Beyond horizontal gene 972 transfer: the role of plasmids in bacterial evolution, Na- 973 ture Reviews Microbiology 19,347(2021). 974 [39] J. Ilhan, A. Kupczok, C. Woehle, T. Wein, N. F. Hülter, 975 P. Rosenstiel, G. Landan, I. Mizrahi, and T. Dagan, Seg- 976 regational Drift and the Interplay between Plasmid Copy 977 Number and Evolvability, Molecular Biology and Evolu- 978 tion 36,472(2019). 979 [40] T. Vogwill and R. C. MacLean, The genetic basis of the 980 fitness costs of antimicrobial resistance: a meta-analysis 981 approach, Evolutionary Applications 8,284(2015). 982 [41] T. Wein, N. F. Hülter, I. Mizrahi, and T. Dagan, Emer- 983 gence of plasmid stability under non-selective conditions 984 maintains antibiotic resistance, Nature Communications 985 10,2595(2019). 986 [42] D. Yang, A. D. Jennings, E. Borrego, S. T. Retterer, and 987 J. Männik, Analysis of factors limiting bacterial growth 988 in pdms mother machine devices, Frontiers in Microbiol- 989 ogy 9,871(2018).