An Optimized Bistable Metabolic Switch To Decouple Phenotypic

Subscriber access provided by University of Sunderland

Article

An optimized bistable metabolic switch to decouple phenotypic states during anaerobic fermentation Naveen Venayak, Kaushik Raj, Rohil Jaydeep, and Radhakrishnan Mahadevan ACS Synth. Biol., Just Accepted Manuscript • DOI: 10.1021/acssynbio.8b00284 • Publication Date (Web): 30 Oct 2018 Downloaded from http://pubs.acs.org on October 30, 2018

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 26 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Synthetic Biology

An optimized bistable metabolic switch to decouple phenotypic states during anaerobic fermentation Naveen Venayak1,* , Kaushik Raj1,* , Rohil Jaydeep1 , and Radhakrishnan Mahadevan1,2,** 1

Department of Chemical Engineering and Applied Chemistry, University of Toronto, ON 2 The Institute of Biomaterials & Biomedical Engineering, University of Toronto, ON * These authors contributed equally. ** Corresponding author, [email protected]

Abstract Metabolic engineers aim to genetically modify microorganisms to improve their ability to produce valuable compounds. Despite the prevalence of growth-coupled production processes, these strategies can significantly limit production rates. Instead, rates can be improved by decoupling and optimizing growth and production independently, and operating with a growth stage followed by a production stage. Here, we implement a bistable transcriptional controller to decouple and switch between these two states. We optimize the controller in anaerobic conditions, typical of industrial fermentations, to ensure stability and tight expression control, while improving switching dynamics. The stability of this controller can be maintained through a simulated seed train scale-up from 5mL to 500 000L, demonstrating industrial feasibility. Finally, we demonstrate a two-stage production process using our optimal construct to improve the instantaneous rate of lactate production by over 50%, motivating the use of these systems in broad metabolic engineering applications. Keywords: dynamic control, bistable switch, two-stage production, anaerobic fermenation, lactic acid 1

Rising concern over human-impact on the environment and increasing volatility in petroleum

2

feedstock pricing has led to the pursuit of microbial chemical production processes. These processes

3

rely on renewable feedstocks, involve more mild reaction conditions, and exploit the natural diversity

4

of compounds producible in metabolic systems. However, microbes often have limited natural

5

performance when applied to producing valuable chemicals 1 . Metabolic engineers aim to improve

6

these microbes by optimizing three key performance indicators of the production process: titer, rate,

7

and yield (TRY). It is well understood that optimization of each of these metrics is crucial to a

8

feasible commercial bioprocess 2,3 , especially when producing low-value and high-volume commodity

9

chemicals.

1

ACS Paragon Plus Environment

ACS Synthetic Biology 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

10

The negligible yield of most important products by wild microbes often makes this the first pro-

11

duction target, with rate and titer improving naturally as a result. If the product is non-natural, a

12

heterologous pathway must first be discovered, expressed and balanced. Then, competing pathways

13

can be eliminated to redirect flux toward this production pathway. Most commonly, metabolic en-

14

gineers target a growth-coupled production phenotype, where product formation is necessary for

15

biomass accumulation. The identification of such growth-coupled strategies can be guided by strain

16

design algorithms, which use a metabolic model to predict genetic interventions 4,5 and many of

17

these algorithms have been experimentally validated 6,7,8,9,10 . These strategies are implemented by

18

knocking out genes or balancing pathways to optimize flux toward desired phenotypes. However,

19

growth-coupled production strategies lead to theoretical limits of overall process rates, especially

20

when producing chemicals at high yield 11,12,13 .

21

Recently, given the wealth of synthetic biological circuits which offer a new level of control, there

22

has been significant interest in decoupling growth and production into distinct stages 12,14,15,16,17,18 .

23

In the first stage, flux toward growth can be maximized to rapidly generate biomass, the required

24

catalyst for the production process. Then, the second stage can proceed with maximal flux toward

25

product, achieving improvements to overall production rates.

26

For some products, growth and production can be decoupled using natural regulatory systems,

27

by controlling environmental conditions such as pH, temperature or oxygen availability. However,

28

these natural regulatory systems often do not coincide with engineering objectives such as producing

29

non-natural chemicals. To extend this capability for arbitrary products, synthetic genetic circuits

30

can be applied to effectively control gene expression subject to numerous cues. For example,

31

gene expression can be controlled using inducers 19,20 , internal metabolites 21,22 or quorum sensing

32

molecules 23,24 . Genetic switches have been used in the past to decouple growth and production

33

stages to improve production of a variety of products such as isopropanol and γ-aminobutyric

34

acid 20,25 . However, due to the monostable nature of these decoupling strategies, removal of inducer

35

caused by dilution in large scale fermentations could result in cells being unable to retain a particular

36

phenotype. Hence, though such circuits have proven effective in improving biochemical production

37

processes at a laboratory scale, significant optimization will be required to implement such genetic

38

logic at an industrial scale.

39

Despite the range of implemented triggers and actuators, controllers have not received com-

40

parable attention. The complex dynamics of biological regulatory networks offer diverse natural

41

examples of controllers with important properties. For example, the bacteriophage λ switch 26 and

42

cyanobacterial circadian oscillator 27 could be applied to achieve desirable complex network dynam-

43

ics. Engineered applications of these systems have been described as some of the seminal works in

44

synthetic biology 28 , moving toward the design and application of more robust and modular genetic

45

circuits.

46

The genetic toggle switch 29 , one of the earliest implemented synthetic circuits, provides an ad-

47

dressable unit of cellular memory, which can be switched between states using transient induction.

48

This type of system, which is used by the λ phage to switch between lytic and lysogenic states,

2


Page 2 of 26

Page 3 of 26 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


49

could also be applied to switching microbes between phenotypes in the context of metabolic engi-

50

neering, by controlling the expression of genes responsible for critical pathways 30,31 . As opposed to

51

more typical expression systems which exhibit an analog response to inducer concentration, such

52

a bistable system should be less sensitive to heterogeneous bioreactor environments. Furthermore,

53

these switches can be induced transiently, with a pulse of some cue such as light 32 or nutrient

54

starvation 33 . These systems could also prove effective in decreasing population heterogeneity for

55

improved production performance 34 . In the past, a bistable controller has been used to improve

56

the production of production of polyhydroxybutyrate in E. coli using glucose starvation as a trig-

57

ger under aerobic conditions 33 . However, this controller showed potential for improvement in the

58

time taken to switch between growth and production phenotypes. A deeper understanding of the

59

various factors affecting the controller’s function is necessary to expand the implementation of such

60

controllers to a wide range of conditions.

61

Here, we implement a bistable state controller to decouple growth and production phenotypes.

62

These types of systems could be extended to a large number of states for more complex temporal

63

or spatial control of microbial phenotypes 35,36 . We developed a bistable transcriptional phenotypic

64

controller by applying a well-known mutually repressive design 37 and controlled critical pathways

65

using this switch. First, we assess this circuit in oxygen-limited conditions to characterize the bur-

66

den, stability and switching dynamics in an environment common to bioproduction. We believe that

67

impaired growth rates under anaerobic conditions could affect circuit function and consequently,

68

lead to ineffective decoupling of growth and production phenotypes if the circuit is not optimized 38 .

69

Next, we apply this system to decouple growth from lactic acid production in a range of evolved

70

E. coli mutants. This system requires the control of a small number of genes, or metabolic valves,

71

to decouple these phenotypes and thus allows us to study the interplay of growth and production

72

states in detail, as a function of evolutionary time. More complex metabolic engineering strategies

73

would require careful consideration in the choice of such metabolic valves.

74

We consider a library of switch architecture variants to minimize the negative effect of leaky

75

expression on production phenotypes. To ensure balanced pathway flux, we also considered a

76

library of controllers with a range of expression strengths. With the optimal controller chosen, we

77

assess its stability to understand the feasibility of such circuits for engineered applications, such

78

as prolonged production processes. Finally, we implement the optimal construct in a production

79

batch to assess the impact of switching time on the trade-off between yield and rate.

80

The lack of modularity and predictability of synthetic circuits has limited the application of

81

complex architectures for metabolic engineering. Here, we characterize a common bistable controller

82

in conditions relevant to metabolic engineering, motivate the use of promoters with tight expression

83

control, and demonstrate the application of an optimized circuit for lactate production. This work

84

presents a foundation for the application of bistable architectures in a generically applicable context

85

for metabolic engineering.

3



Page 4 of 26

86

Results and Discussion

87

Optimization of the bistable circuit in oxygen-limiting conditions

88

As the backbone for our phenotypic controller, we implemented a two-state bistable transcriptional

89

controller based on a seminal work in synthetic biology, the genetic toggle switch. Our first attempts

90

using one of the original toggle switch designs, pTAK132 29 , exhibited monostability in our host

91

(E. coli MG1655) and process conditions (Figure S1). We then implemented the pKDL071 toggle

92

switch 37 , built with pL(tet)O/tetR and pTrc2/lacI promoter/repressor pairs, using mCherry and

93

gfpmut3 as reporters, respectively. The two states of this controller will henceforth be referred to

94

as state A (to denote the state with lacI/gfpmut3 expression) and state B (to denote the state with

95

tetR/mCherry expression). State A can be induced with anhydrotetracycline (aTc), and state 2

96

can be induced with isopropyl β-D-1-thiogalactopyranoside (IPTG) analogues to tetracycline and

97

lactose, respectively (Figure 1). Phenotype A Pathways Ptet Ptet Ptet

pKDL071 state-controller (bistable toggle switch)

reporter

reporter

Phenotype B Pathways

State B Inducer: IPTG

gfpmut3

Ptet

lacI

Ptet

Ptrc2

tetR

Ptrc2

mCherry

Ptrc2 Ptrc2 Ptrc2

State A Inducer: aTc

STATE B

STATE A

Figure 1: A schematic of the bistable transcriptional controller used in this study. The mutually repressive promoter/repressor pairs consist of pTrc2/lacI and pTet/tetR. State A is driven by pL(tet)O, induced by aTc, and expresses lacI and gfpmut3. State B is driven by pTrc2, induced by IPTG, and expresses tetR and mCherry. Pathways for either growth or production phenotypes can be expressed in independent states to decouple, for example, growth and production processes. 98

We assessed the switch in oxygen limited conditions which are typical of high density industrial

99

production processes. Despite the effectiveness and wide use of GFP, mCherry, and other proteins

100

of the same family, their requirement for molecular-oxygen to exhibit fluorescence 39 can be limiting

101

in metabolic engineering applications. This can be overcome by sampling and maturing the proteins

102

in the presence of oxygen, before reading fluorescence. However, this process precludes the ability

103

for on-line reads in anaerobic environments and increases measurement variability. In experiments

104

where fluorescent proteins were used, we allowed these proteins to mature in oxygen-rich environ-

105

ment before quantification. Thus, since all subsequent experiments are performed in anaerobic

106

conditions, fluorescent proteins are matured where needed (see materials and methods, Inducer

107

toxicity and effectiveness). However, as will be discussed, these fluorescent probes have a weak

108

correlation to the expression level of critical metabolic pathways, and thus are not frequently used.

109

Improved probes which do not exhibit a strong dependence on oxygen availability for fluorescence

110

could be used to overcome some of these limitations 40 . 4


Page 5 of 26 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


111

Inducer concentration

112

First, we determined the effective range of both inducers, IPTG and aTc, by assessing the growth

113

profile of ∆lacI strains with pKDL071 (a control plasmid with only the core bistable controller)

114

between 0 and 100 times of typical inducer concentrations (0-100mM IPTG, 0-10mg/mL aTc).

115

Based on this screen, we established the maximum non-toxic inducer concentrations as 50mM

116

IPTG (Figure S2a) and 2.5mg/mL aTc (Figure S2b).

117

Using these bounds, we determined the optimal inducer concentration to use at specific switch-

118

ing optical densities (OD) in order to maximize the effectiveness of switching for both inducers.

119

For IPTG, the switching rate had a small dependence on inducer concentration, and the inducer

120

effectiveness decreased at higher induction densities (Figure S2c). For aTc, both the inducer

121

concentration and induction density had an impact on the inducer effectiveness(Figure S2d).

122

Although protein degradation and production have been reported in the stationary phase 41,42,43 ,

123

we noticed a distinct decrease in the inducer effectiveness at higher induction densities which were

124

not in an exponential growth phase. This is especially apparent when induced with IPTG, where

125

nearly no switching was observed in cultures inoculated and induced at OD600 = 1. These results

126

would indicate that protein dilution by cell division has an impact for the clearing of proteins in

127

this system - lacI and tetR. These results also indicate that IPTG reaches a saturating regime for

128

LacI binding at relatively low concentrations, while high concentrations of aTc are required to reach

129

saturation for TetR binding, which was not observed within the non-toxic range. Therefore, rapid

130

switching from state A to B would require relatively high concentrations of aTc (≥2.5mg/mL), while

131

the switch from state B to A would require more typical concentrations of IPTG (∼1mM). Hence,

132

in our subsequent experiments we used 2mM of IPTG throughout, 500ng/mL aTc for overnight

133

cultures, and 1mg/mL aTc when rapid switching from state B to state A was desired.

134

Plasmid burden

135

Although bistable circuits have attractive characteristics, these can oftentimes come at the expense

136

of metabolic burden. To obtain bistability, the expression and degradation rates of each repressor

137

must be balanced. In addition, high expression levels lead to a wider area of bistability 29 and

138

increased growth rates lead to a smaller area of bistability 38 . We characterized the growth of one

139

of our target strains, LAC002-D28 (Table 2), with our target transcriptional controllers. This

140

strain has a lower growth rate compared to wild-type, a typical consequence of most metabolic

141

engineering efforts. In this case, the reduced growth rate is attributed to the elimination of pathways

142

for preferred fermentation products, ethanol and acetate, which allow for a higher biomass yield

143

and growth rate.

144

We characterized three different strains: LAC002-D28 (no plasmid), LAC002-D28+pKDL071,

145

and LAC002-D28+pTB001 (Table 1), in three different media (see materials and methods for

146

recipes): M9 minimal medium, phosphate minimal medium (PMM), and rich defined medium

147

(RDM). pTB001 is a modified version of pKDL071, which has a considerably reduced translation

148

initiation rate (TIR, according to RBS calculator v2.0 44 ) of lacI (0.71% of pKDL071) and tetR 5



149

(2.7% of pKDL071). The comparison of burden between pKDL071 and pTB001 is indicative of the

150

burden owed to the expression of tetR and lacI alone, as opposed to other elements in the bistable

151

switch.

152

In M9 minimal medium, the burden of the pKDL071 plasmid is considerably high as compared

153

to the no plasmid control and the pTB001 strain (Figure S3a). This trend is even more evident in

154

PMM, where the no plasmid control has a growth advantage, followed by the pTB001 and pKDL071

155

strains, respectively (Figure S3b). These results suggest that the plasmid alone has a considerable

156

burden, which is substantially increased with higher repressor translation rates. Recently, a single-

157

copy toggle switch was developed which could alleviate this burden 45 ; however, further study is

158

required to understand the burden of these systems in strains which already suffer from a growth

159

defect, such as metabolically engineered strains, as well as their stability in diverse environments.

160

Finally, we were able to eliminate this burden by supplementing PMM with amino acids (RDM)

161

which can directly be used to produce the toggle switch peptides (Figure S3c).

162

Metabolic valves to decouple growth from lactate production

163

To implement this bistable architecture for metabolic engineering applications, a set of important

164

reactions should be identified and dynamically controlled; these reactions are referred to as metabolic

165

valves. The choice of such valves in complex engineered strains with a large number of genetic

166

perturbations requires significant consideration. We applied this system to an anaerobic lactic acid

167

production strain of E. coli, which required only two knockouts, simplifying the choice of these

168

valves.

169

We targeted a series of MG1655 ∆adhE,pta,lacI strains (LAC002), which have approximately

170

half the growth rate of wild-type but reach near 100% lactate yield 6 (Figure 2a-c). ∆lacI will be

171

referred to as a wild-type in the context of this manuscript, since it exhibits the same phenotype

172

under our growth conditions; lacI is deleted from all strains to prevent competition with the toggle

173

switch repressors. The two deleted genes, adhE and pta are responsible for ethanol and acetate

174

production, respectively, the preferred fermentation products in E. coli. By deleting these genes,

175

flux is naturally diverted to produce mostly lactic acid, one of the few remaining fermentation

176

products. This results in a sub-optimal biomass yield, and growth rate. By exploiting adaptive

177

laboratory evolution, the lactate yield of this strain improves by approximately 25% (Figure 2c).

178

In this case, stoichiometric models suggest that expression of either adhE or pta alone is insuf-

179

ficient to restore the optimal biomass yield (Figure S4). To realize an optimal growth phenotype,

180

adhE and pta are co-expressed to allow flux to ethanol and acetate, respectively, maximizing

181

biomass yield and growth rate. We explored a library of architecture variants and mutants to

182

optimize expression levels toward achieving a wild-type growth rates.

6


Page 6 of 26

Page 7 of 26

a

b μmax (1/h)

.5 GLC .5 ATP

Pentose phosphate pathway

.5 G6P

ldhA NAD+

LAC

Formate

CO2 + H2

AcCoA

pta ATP

AcP ackA

Ac

adhE

TCA

NAD+

ACALD

adhE NAD+

EtOH

BIOMASS

lactate titer (g/L)

c

GA3P PYR

GLC

LAC002-D1

0.6

Growth Valve ON

0.4

ldhA

2

BIOMASS

AcCoA

PYR

0.2 0

.5 ATP

NADH, 2ATP

d

0.8

pta

adhE

Ac EtOH

LAC

1.5

e

1

GLC

0.5 0

Δa d Δa hE dh ,pt Δ a l Δa E,p ,la acI dh ta cIE, ,la D1 pt cI a, -D la 2 cI 8 -D 59

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Production Valve OFF

LAC

BIOMASS

AcCoA

PYR ldhA

pta

adhE

Ac EtOH

Figure 2: Overview of E. coli mixed-acid fermentation metabolism, knockout metabolism, and dynamic control strategy. (a) Central metabolic map highlighting anaerobic pyruvate metabolism. Red genes indicate genes which are deleted to improve lactate flux through the ldhA pathway, indicated in green. (b) Growth rate of ∆lacI strains and evolved ∆adhE,pta,lacI strains. D1, D28, D59 refer to the duration of evolution, in days. (c) Lactate titer using 2g/L glucose. (d) Metabolic flux profile with valves ON, allowing flux toward acetate and ethanol. (e) Flux profile with valves OFF, redirecting flux toward lactate. Error bars represent s.d. (n ≥ 3).

184

Design, construction, and metabolic optimization of the phenotypic state-controller for metabolic engineering

185

The phenotypic state-controller was expressed on a medium-copy plasmid with ColE1 replicon,

186

and to ensure optimal expression in this genetic context we designed a library with varying ex-

187

pression levels of adhE and pta. It is well known that several parameters affect the steady-state

188

levels of proteins, and thus the steady-state flux distribution and phenotype of resulting strains.

189

These include transcription rate, translation rate, degradation rate, and dilution rate (by growth).

190

Using this canonical switch design, we were restricted to the specific promoters used in the state

191

controller, and thus controlled the translation rate. We modified the ribosome binding site (RBS)

192

to build our library and determine optimal expression levels of adhE and pta. The RBS translation

193

initiation rates were targeted to span a 4-fold range (Table 1) and were calculated using the RBS

194

calculator 44,46 .

195

Toggle switch driven by state B is limited by background expression

196

The symmetric nature of this switch leads to two potential architectures for the implementation

197

of the desired growth and production states. Here, state A is implemented for production (empty)

198

and state B for growth. Thus, adhE and pta were expressed in state B (pTSLAC10X, Figure 3b).

199

We characterized a range of TIR variants by modifying the RBS, in order to determine which set

200

optimized growth rate and product yield in their respective states (Figure 3c-f ).

183

201

The strain with the optimal growth phenotype, LAC002-D59+pTSLAC103, restored 89% of

7



Page 8 of 26

Table 1: Plasmids used in this study, distinguishing features, and RBS strengths. RBS strengths (translation initiation rates, T.I.R) were calulated using the Salis RBS calculator v2.0 46 . Plasmid

Generation

Distinguishing Feature

pKDL071

—

control

pTSLAC101 pTSLAC102 pTSLAC103 pTSLAC104

1

pTSLAC111

1.1


2

RBS Strength (x103 T.I.R.) adhE pta ldhA — — —

state B growth

1.4 2.5 4.2 8.5

0.26 1.38 0.82 8.9

— — — —

ldhA expression

4.2

0.82

3.6

state A growth

1.0 4.6 7.6 10

1.1 5.1 7.5 9.4

— — — —

1.1 5.1 7.5 9.4

— — — —


3

degradation tags

1.0 4.6 7.6 10

pTSLAC312 pTSLAC313 pTSLAC314

3.1

low pta expression

4.6 7.6 10

1.1 1.1 1.1

— — —

pTB001

—

low repressor expression

—

—

—

8


Page 9 of 26

State A

a

pKDL071 (empty control)

b

pTSLAC10X State A (production) Empty

State B

State B (growth) Growth pathways Ptrc2

c

f

0

Growth state end-point titer (g/L) lactate ethanol

Production state end-point titer (g/L) lactate ethanol growth rate

e

acetate

0.5

d

0.2 0.15 0.1 0.05 0 2 1.5 1 0.5 0

0.5

0.5 0 0.2 0.15 0.1 0.05 0 2 1.5 1 0.5 0

0.5

WT+pKDL071

pTSLAC104

pTSLAC103

pTSLAC102

Growth target

pTSLAC101

0

pKDL071

pTSLAC104

pTSLAC103

pTSLAC102

pTSLAC101

Production target

adhE

1

1

0

RBS X

1

growth rate

acetate

1

Ptrc2

pta

RBS X

pKDL071

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 3: Characterization of LAC002-D59 strains in growth and production states using 2g/L glucose and generation 1 plasmids, driven by state B (pTSLAC10X). Purple bars indicate controls, LAC002-D59+pKDL071 for the production state, and WT+pKDL071 for the growth state. Dashed lines indicate the growth and production targets based on these controls. (a) pKDL071 contains empty state A and B (i.e. only genes necessary for the controller’s function are present). (b) pTSLAC10X contains pta and adhE in state B, and an empty state A. (c) Acetate titer. (d) Ethanol titer. (e) Lactate titer. (f) Growth rate. Error bars represent s.d. (n ≥ 3). See also Figures S8,S9.

9



202

the wild growth rate and nearly eliminated lactic acid production. Despite this promising result,

203

none of the constructs achieved high lactic acid titer in the production state, with the exception

204

of pTSLAC104. Although pTSLAC104 achieved favorable lactic acid titer in the production state

205

(Figure 3e), this construct exhibited poor growth in the growth state (Figure 3f ). We anticipate

206

that the high expression of pta might have led to inactive protein aggregates, preventing optimal

207

expression and making this construct inadequate (Figure 3f ). This is further supported by the

208

unusually low titers of acetate for this strain in the growth state (Figure 3c). We also assessed the

209

impact of ldhA expression in the production state to determine if lactate yield could be improved;

210

however, this did not improve lactate yields, especially in our most evolved mutant (Figure S8-S9)

211

Generally, the high levels of ethanol and acetate in the production state, where adhE and pta

212

are anticipated to be OFF, led to low lactate yield. This indicates that leaky expression can be a

213

significant hindrance to decoupling growth and production phenotypes when low levels of enzyme

214

can lead to high pathway flux. This is corroborated by the concentration of fluorescent reporters,

215

where the dynamic range (the ratio of ON to OFF states) of the pTrc2 promoter is only 8.3-fold,

216

compared the the pL(tet)O promoter which is 17.5-fold (Figure S5). We anticipate the use of

217

promoters engineered to reduced leaky expression could alleviate this issue 45 , and will be of critical

218

importance in the implementation of synthetic genetic circuits for metabolic engineering.

219

Toggle switch driven by state A for tighter expression control

220

We also assessed an inverted transcriptional controller architecture (compared to generation 1 con-

221

structs), where state A drove the growth state, and state B drove the production state (empty).

222

We expected this architecture to significantly alleviate the impact of leaky expression due to the

223

improved dynamic range of the pL(tet)O promoter (Figure S5). Here, we note that the per-

224

formance of both states is improved (Figure 4). Ethanol production (Figure 4d) and growth

225

rate (Figure 4f ) are restored to 97% of wild-type in the pTSLAC202 variant. Most importantly,

226

pTSLAC202 also shows an improved lactate titer in the production state, 61% of the production

227

target (LAC002-D59+pKDL071) and a 3.2-fold increase compared to LAC002-D59+pTSLAC103

228

(our optimal construct from the first generation plasmids).

229

Although the growth state reached performance near the wild-type control and the production

230

state was improved over first generation plasmids, it only achieved 60% of the lactate titer of the

231

production control (LAC002-D59+pKDL071). This can be attributed to the high acetate titers,

232

since ethanol is undetectable in the production state. We anticipate that small amounts of pta are

233

responsible for significant flux through this pathway, further necessitating tighter control of gene

234

expression. The significant impact of leaky pta expression could be owed to its favorable enzyme

235

kinetics (kcat /Km = 659mM −1 s−1 ) 47 .

236

Using this strain, we performed preliminary pH controlled bioreactor trials to effect a switch

237

between growth and production. We monitored fluorescent proteins and product titers to charac-

238

terize the switching dynamics. Following induction, we noted that GFP fluorescence (which should

239

be co-expressed with adhE and pta) reached OFF levels within approximately 3 hours. However, 10


Page 10 of 26

Page 11 of 26

State A

a


b

pTSLAC20X State A (growth) Growth pathways adhE

c

RBS X

State B

State B (production) Empty PL(tet)O

pta

RBS X

PL(tet)O

1.5 acetate

acetate

1.5 1 0.5

1 0.5 0

growth rate

f

2 1.5 1 0.5 0 1

0.15 0.1 0.05 0 2 1.5 1 0.5 0 1

0.5

0.5 pTSLAC204

pTSLAC203

pTSLAC202

Production target

pTSLAC201

WT+pKDL071

pTSLAC204

pTSLAC203

pTSLAC202

0

pKDL071

0 pTSLAC201

Growth target

0.15 0.1 0.05 0

growth rate

e


d

Production state end-point titer (g/L) lactate ethanol

0

pKDL071

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 4: Characterization of LAC002-D59 strains in growth and production states using 2g/L glucose and generation 2 plasmids, driven by state A (pTSLAC20X). Purple bars indicate controls, LAC002-D59+pKDL071 for the production state, and WT+pKDL071 for the growth state. Dashed lines indicate the growth and production targets based on these controls. (a) pKDL071 contains empty state A and B (i.e. only genes necessary for the controller’s function are present). (b) pTSLAC20X contains pta and adhE in state A, and an empty state B. (c) Acetate titer. (d) Ethanol titer. (e) Lactate titer. (f) Growth rate. Error bars represent s.d. (n ≥ 3). See also Figures S10,S11.

11



Page 12 of 26

240

we did not notice a shift in metabolite production rates within the entire batch duration (Fig-

241

ure S6). This is likely due to significant differences in degradation rate between the large native

242

peptides (≥720AA) and the heterologous fluorescent reporters (≤240AA). In order for this switch

243

architecture to be practical in the target batch time scales, significant improvements to degrada-

244

tion rate of adhE and pta would be required. Furthermore, this experiment exemplifies the poor

245

correlation between levels of reporters and pathway enzymes, precluding the use of reporters to

246

study metabolic switching dynamics.

247

Degradation tags improve switching rate and reduce leaky expression

248

To improve phenotypic switching dynamics, degradation tags were added to adhE and pta in second

249

generation plasmids, to generate third generation plasmids (pTSLAC30X, Figure 5b). These tags

250

are natively added to peptides which have been incompletely translated, tethering them to the

251

ClpXP protease and leading to their rapid degradation 48 . We utilized a degradation tag with a

252

high affinity for the native SsrA protease, referred to as “LAA” for the terminal three amino acids

253

of the tag 49 . Due to the high production state titers of acetate in all generation 2 constructs, we

254

also generated a set of variants with pta fixed at its lower limit, while varying adhE expression

255

(pTSLAC31X, Figure 5c).

256

For this generation of plasmids, we noted that leaky expression seemed to be eliminated in all

257

constructs for the production state (Figure 5d-e). This is demonstrated by consistently low levels

258

of acetate and ethanol in the production state, very similar to the production controls (LAC002-

259

D59+pKDL071), in all tested constructs. These same constructs achieve a growth rate near growth

260

controls (∆lacI+pKDL071) in the corresponding state.

261

Based on a simple model of transcription, translation and degradation, we would expect leaky

262

expression to be unaffected by the inclusion of degradation tags for a given steady-state protein

263

concentration: dP = α − Pγ dt

(1)

264

where P is the protein concentration (M), α is the lumped transcription and translation rate (mol/(L

265

s)), and γ is the degradation rate (1/s).

266

At steady-state, the dynamic range is independent of degradation rate: PON αON = POF F αOF F

(2)

267

However, this model assumes a linear relationship between degradation rate and protein concen-

268

tration, which is unlikely owing to limited cytoplasmic concentrations of protease. Since ssrA-

269

dependent proteolysis ensures proteins which cannot be fully translated are degraded, and only

270

approximately 0.4% of all native peptides are ssrA tagged 50 , this is a likely consequence of high

271

ssrA-tagged protein expression. Considering Michaelis-Menten type kinetics for this saturation, the

272

model is refined as: 12


Page 13 of 26

State A

a


b

pTSLAC30X State A (growth) Growth pathways

LAA adhE RBS X

c

State B

State B (production) Empty

PL(tet)O

LAA pta RBS X

PL(tet)O

pTSLAC31X State A (growth) Growth pathways

LAA adhE RBS X

State B (production) Empty PL(tet)O

LAA pta

RBS 1

PL(tet)O

1

acetate

acetate

d 0.5

1 0.5 0

0

growth rate

0.1 0

2 1.5 1 0.5 0 1 0.5

0.2 0.1 0

2 1.5 1 0.5 0 1 0.5

pTSLAC314

pTSLAC313

pTSLAC312

pTSLAC304

pTSLAC303

pTSLAC302

pTSLAC301

pTSLAC314

WT+pKDL071

pTSLAC313

pTSLAC312

pTSLAC304

pTSLAC303

Production target

pKDL071

0 pTSLAC302

0

pTSLAC301

Growth target

Production state end-point titer (g/L) lactate ethanol

g

0.2

growth rate

f


e

pKDL071

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 5: Characterization of LAC002-D59 strains in growth and production states using 2g/L glucose and generation 3 (pTSLAC30X) and 3.1 (pTSLAC31X) plasmids, driven by State A. Purple bars indicate controls, LAC002-D59+pKDL071 for the production state, and WT+pKDL071 for the growth state. Dashed lines indicate the growth and production targets based on these controls. (a) pKDL071 contains empty state A and B (i.e. only genes necessary for the controller’s function are present). (b) pTSLAC30X contains pta and adhE in state A, and an empty state B. (c) pTSLAC31X contains pta and adhE in state A with the RBS strength of pta fixed at it’s lower limit, and an empty state B. (c) Acetate titer. (d) Ethanol titer. (e) Lactate titer. (f) Growth rate. Error bars represent s.d. (n ≥ 3). See also Figures S12,S13.

13



dP P = α − γmax dt P + Km 273 274

Page 14 of 26

(3)

where γmax is the maximum degradation rate (mol/(L s)) and Km is the Michaelis-Menten constant (mol/L).

275

With this model, assuming that the protease is saturated in the ON case, the relative degrada-

276

tion rate in the OFF state increases, leading to reduced leaky expression (Figure S7). This result

277

would motivate the use of degradation tags, not only to improve switching dynamics but also to

278

control leaky expression, in cases where extremely tight control is required. These systems could be

279

further improved by using degradation tags which can be tuned using the expression of an adapter

280

protein 48 , albeit oftentimes with lower maximal degradation rates.

281

Choice of evolved mutant influences phenotype

282

We had anticipated that the evolved mutant (LAC002-D59), despite having near 100% lactate

283

yield in the production state, would suffer from suboptimal growth in the growth state. Our initial

284

hypothesis was that silencing the effect of genetic perturbations (knockouts) would restore the

285

optimal wild-type growth rate. In this case, we expressed these genes (adhE and pta) on a plasmid

286

and screened a library to find the optimal construct with balanced expression. However, since the

287

evolved mutant has accumulated mutations in addition to the gene knockouts, its genotype is more

288

divergent from the parent strain than an unevolved mutant, and all of these mutations may need

289

to be silenced to effectively restore the parent phenotype.

290

Despite these genotypic variations, we did not observe considerable differences in growth state

291

growth rates in any of the evolved mutants, using our final construct (pTSLAC314, Figure S14).

292

This discrepancy could be attributed to the difference in medium used in this study (RDM), as

293

compared to the medium used for adaptive laboratory evolution (M9), or because the original

294

wild-type strain had not been evolved to its theoretical maximum growth rate. However, despite

295

the evolved mutant attaining favorable growth rate, the overall phenotype diverged from the wild-

296

type with increased evolutionary time, indicating that adaptively evolved strains require further

297

consideration to be restored to the original parent phenotype (wild-type), especially if several

298

phenotypic traits are of interest (Figure 6).

299

Phenotypic stability for scaled production

300

To determine if cell populations would remain phenotypically stable when cells are propagated be-

301

fore production, we assessed the phenotypic stability of our optimal construct and strain (LAC002-

302

D59+pTSLAC314) in the growth state. Since stable circuits should not require additional inducer,

303

this would allow for the use of minute inducer quantities for very small volume preculture steps,

304

used as inocculum to a seed fermentation train. We simulated a seed train from 0.005L to 500 000L,

305

a typical industrial fermentation volume, by using four 100x dilution steps (108 x total dilution).

306

We monitored end-point phenotypic traits including product and biomass titers, as well as 14


Page 15 of 26

Evolutionary time (days)

1

28

59

lactate

1 0.5 0

b acetate

0.2 0 −0.2

c

0.2

ethanol

Growth-state end-point titer deviation from wild-type (g/L)

a

0.1 0 −0.1 −0.2 pTSLAC314

pTSLAC313

pTSLAC304

pTSLAC303

pTSLAC302

pTSLAC301

pTSLAC314

pTSLAC313

pTSLAC304

pTSLAC303

pTSLAC302

pTSLAC301

pTSLAC314

pTSLAC313

pTSLAC304

pTSLAC303

pTSLAC302

pTSLAC301

Figure 6: Deviation of product titers of evolved variants in the growth state, compared to the wild-type control. (a) Lactate titer. (b) Acetate titer. (c) Ethanol titer. Error bars represent s.d. (n ≥ 3)

307

population distribution using fluorescent reporters (Figure 7). We identify that this phenotype

308

is relatively stable through this experiment as shown by the consistently low levels of mCherry

309

fluorescence in each panel, demonstrating the effectiveness of these bistable systems for maintaining

310

their state through scale-up processes. However, we observed that the GFP gluorescence decreased

311

slightly over the course of the experiment. We anticipate that this is due to the accumulation of

312

mutations that result in a lower expression of GFP over several generations, however a thorough

313

examination of fluorescent protein expression in either state needs to be conducted to ascertain the

314

exact cause. b 1

2

End-point OD (g/L)

End-point titer (g/L)

Growth state (aTc)

2.5

1.5 1 0.5 0 100

10k

1M

100M

Fold-dilution etoh

ac

0.8

100M

1M

10k

0.16

2

0.12

5

10k

1M

100M

5

0.08

2

100

0.04

5 2

1

10

100

Fold-dilution lac__D

10k

2

1000

0.6 0.4 100

100

Fold-dilution mCherry fluorescence (a.u.)

a

1000

1

10

100

1000

1

10

100

1000

1

10

100

1000

0

Fraction of total population

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


gfp fluorescence (a.u.)

OD600

Figure 7: Phenotypic stability of final construct, pTSLAC314, in the growth state. Cells were serially passed by diluting 100x every 24 hours into medium without inducer. (a) End-point product and biomass titers. Error bars represent s.d. (n = 3). (b) Cell populations by flow cytometry, mean of three biological replicates is shown.

15



315

Improved production rates using two-stage production

316

Finally, we use our optimal strain, LAC002-D59+pTSLAC314 for a two-stage production run in

317

pH controlled 500mL bioreactors sparged with nitrogen. We tested five conditions: growth only,

318

production only, and strains which were switched from growth to production at 0, 2 and 4 hours.

319

For the growth and production only conditions, strains are grown overnight in their corresponding

320

inducers, and inoculated into reactors also containing these inducers. The strains to be switched

321

are grown overnight in the growth inducer, and inoculated into reactors containing inducer-free

322

medium. This exploits the memory characteristic of this circuit to maintain the growth state, until

323

the production state is induced. We screen a range of switching times, to identify the optimal

324

switching time for growth and production rates.

325

The growth and production controls behaved as expected, with the growth control having the

326

highest biomass, acetate and ethanol titers with low lactate titer, and the production control having

327

the highest lactate titer, and lowest ethanol, acetate and biomass titers (Figure 8). The switched

328

strains resulted in titers between these controls, becoming more similar to the production strain

329

the earlier the switch was induced.

330

One of the most important benefits of two-stage production is the improvement to process rates,

331

and the choice of switching time will have a strong influence on this rate. Early switching will leave

332

insufficient time for biomass production, and late switching would result in unacceptably low titers.

333

We have shown that the optimal switching time given our batch conditions was 2 hours, leading

334

to an increase in the maximum instantaneous lactic acid production rate of 55% and an overall

335

improvement of 20%. Given longer batch times, this improvement would be further emphasized

336

owing to larger biomass concentrations.

337

In conclusion, we demonstrate the first reported application of a bistable state-controller in the

338

context of two-stage anaerobic production. These advanced controller architectures offer promising

339

characteristics for stability and induction profiles, at the expense of increased complexity. We first

340

characterize the state controller in anaerobic conditions to identify optimal induction parameters.

341

The low growth rates resulting from anaerobic conditions affect the rate of switching and conse-

342

quently the effectiveness of the controller. Next, we screen a range of switch variants to identify

343

an optimal architecture to overcome leaky expression and reach theoretical production and growth

344

targets. By doing this, we analyze the effect of varying the expression levels of the genes adhE and

345

pta during the growth stage in restoring the mutant strain’s phenotype to resemble that of the wild-

346

type strain. We show that degradation tags help to reduce leaky expression, thereby allowing tight

347

control of gene expression and allowing us to effectively decouple the growth and production stages.

348

In addition, we characterize the stability of this metabolic controller to show that this controller is

349

stable for a cumulative 100-million fold dilution (comparable to a scale-up from 5mL to 500000L).

350

Finally, we use our optimized strain to improve the instantaneous rate of lactate production by

351

55% over control strains with no dynamic control. This study and optimized circuit architecture

352

will drive the adoption of more complex and promising synthetic circuits to engineered systems,

353

and motivates the improved characterization of such circuits for engineering applications. 16


Page 16 of 26

Page 17 of 26

titer (g L-1)

Production phenotypes

rate (g L-1 h-1)

lactate

a

e

ethanol

10

Time (h)

f

1

j

0.2

g

1.5

k

0.3

0 0.2

0.5

0.1

0

0

0.1

h

2

0

(OD h )

l

0.4 0

0.2 0.1

4h

2h

induction time

0h

growth only

0

PRODUCTION

production only

0

GROWTH

10

0.1 4h

5 Time (h)

Time (h)

induction point

0.2

2h

0

(OD h-1)

p

0.3

0h

10

(OD h )

0.4

PRODUCTION

5

0

GROWTH

0

0.1 0.05

-1

−0.2

0 0.2 0.15

0 -1

0.2

1

o

0.3

0.2

(OD)

0.05

0.05

1

d

0.1

0.1

0

0

induction time

n

0.2 0.15

0.1

0.5

induction time

4h

5

2h

0

10

0h

0

5

PRODUCTION

0.2

0

GROWTH

0.5

0

4h

0.5

2h

0.4

2

0h

1

PRODUCTION

1

b

acetate

m

4

Growth phenotypes

biomass

max. overall rate (g L-1 h-1)

i

Time (h)

c

max. rate (g L-1 h-1)

GROWTH

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


induction time

Figure 8: Characterization of LAC002-D59+pTSLAC314 strains in pH controlled anaerobic bioreactors. Precultures were grown in corresponding inducers for growth and production controls and inoculated into reactors with matching inducer. Strains to be switched were grown in the growth state, washed, and inoculated into reactors without inducer. These reactors were induced with 2mM IPTG at designated switching times (blue arrows): 0, 2, or 4 hours. (a-d) Product titers. (e-h) Product rates were determined by numerical differentiation. (i-l) The maximum rate is the instantaneous maximum rate obtained from numeric differentiation. (m-p) The maximum overall rate is determined from the final titer (before substrate consumption) divided by the batch time at this point.

17



Page 18 of 26

354

Materials and Methods

355

Strains and plasmids

356

The parent strain, MG 1655 ∆adhE,pta and the resulting strains from adaptive laboratory evo-

357

lution, were generously donated by Dr. Stephen Fong 6 . The transcriptional repressor, lacI, was

358

deleted from all strains using the λ-red method to eliminate interference with the toggle switch 51 .

359

The deletion cassette was amplified from pKD4 using primers lacI RED F and lacI RED R. The re-

360

sistance marker was removed using the pCP20 plasmid 51 . Strains were transformed with designated

361

plasmids as indicated using electroporation. Table 2: Strains used in this study and their respective genotype. Name WT

Genotype MG1655 ∆lacI

ALE time

Reference Litcofsky et al. 37

LAC001-D1 LAC001-D28 LAC001-D59

MG1655 ∆adhE,pta MG1655 ∆adhE,pta MG1655 ∆adhE,pta

1 28 59

Fong et al. 6 Fong et al. 6 Fong et al. 6

LAC002-D1 LAC002-D28 LAC002-D59

MG1655 ∆adhE,pta,lacI MG1655 ∆adhE,pta,lacI MG1655 ∆adhE,pta,lacI

1 28 59

This study This study This study

362

Plasmids were transformed and maintained in DH10β. All enzymes were obtained from New

363

England Biolabs (NEB). Synthetic oligonucleotides were obtained from Integrated DNA Technolo-

364

gies. All PCR amplifications for plasmid construction were performed using Q5 High-Fidelity

365

Polymerase (NEB). Plasmids were constructed using standard molecular biology techniques and

366

ligase cycling reaction 52 . Plasmid parts were flanked by insulating sequences (UNS) to improve

367

assembly efficiency 53 . pKDL071 was generously donated by Dr. James Collins and used as the

368

backbone for the state controller plasmids in this study 37 . pTSLAC plasmids were constructed

369

from pKDL071, MG1655 gDNA and gBlocks using primers indicated in Supplemental Table S1.

370

Ribosome binding sites (RBS) were generated using the RBS Calculator 44 v2.0, and were added

371

on primer overhangs.

372

Media

373

Phosphate minimal medium (PMM) was composed of salts (10X, in 1L: 35g KH2 PO4 , 50g K2 HPO4 ,

374

35g (NH4 )2 HPO4 ) 54 , supplements (1000X: 1M MgSO4 , 20000X: 0.5M CaCl2 ), and trace metals

375

(1000X: 1.6g/L FeCl3 , 0.2g/L CoCl2 ·6H2 O, 0.1g/L CuCl2 , 0.05g/L H3 BO3 , 0.8g/L MnCl2 , 0.2g/L

376

NaMoO4 , 0.2g/L ZnCl2 ·4H2 O).

377 378

M9 minimal medium was composed of salts (5X in 1L: 64g Na2 HPO4 -7H2 O, 15g KH2 PO4 , 2.5g NaCl, 5.0g NH4 Cl), with supplements and trace metals from PMM.

18


Page 19 of 26 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Table 3: Truth table for implemented state controller. Both inputs would lead to an unstable state in which both repressors are expressed maximally and inactivated by inputs. State A is the growth state and state B is the production state. A third undesirable state 3 is achieved with both inputs, where both repressors are inactivated, leading to maximal expression of each - this state is unstable. Input aTc IPTG 0 0 0 0 1 0 1 0 0 1 0 1 1 1

State Current Next A A B B A A B A B B A B A or B unstable

379

Rich defined medium was composed of PMM, amino acid supplements (5X: 4mM alanine, 26mM

380

arginine, 2mM aspartate, 0.5mM cysteine, 3.3mM glutamate, 3mM glutamine, 4mM glycline, 1mM

381

histidine, 2mM isoleucine, 4mM leucine, 2mM lysine, 1mM methionine, 2mM phenylalanine, 2mM

382

proline, 50mM serine, 2mM threonine, 0.5mM tryptophan, 3mM valine) and nucleotide supplements

383

(10X: 2mM adenine, 2mM cytosine, 2mM uracil, 2mM guanine in 15mM KOH). Glucose was added

384

as the sole carbon source as indicated.

385

Phosphate buffered saline was obtained from Biobasic (PD8117) and used for cell resuspension

386

when reading fluorescence.

387

Automated 96 well plate preparation and evaluation

388

Experiments were prepared using a Tecan EVO 200 liquid handling system, equipped with an 8-

389

channel fixed tip arm and a Tecan M200 plate reader. Cells were grown overnight in 96-well plates

390

in 150µL LB+1% glucose including inducers and antibiotics as required. Plates were sealed with

391

a PCR film (Bio-Rad, MSB1001) in order to limit oxygen availability. Plates were centrifuged at

392

2000xg for 10 minutes and transferred to liquid handler. The liquid handler removed supernatant

393

from the plate, added fresh medium with 0.2% glucose, and resuspended cells. This wash procedure

394

was repeated twice, before an OD normalization procedure.

395

Plates were transferred to a Tecan M200 plate reader to begin OD normalization procedure.

396

The OD in each well was read at 600nm, and from this blank subtracted data the required volume

397

of cells, and balance volume of medium for a total of 150µL were calculated using Magellan (v7.2,

398

Tecan) and dispensed by the liquid handler. Plates were cycled 3x with nitrogen in an airlock

399

and transferred to an anaerobic chamber with a 100% N2 atmosphere. 50µL mineral oil (BioShop,

400

MIN444) was added to each well before being read to eliminate evaporation. Transparent films

401

were unsuitable in our setup due to condensation on the film. Plates were incubated at 37◦ C in a

402

plate reader (Spectramax M2) with shaking inside the anaerobic chamber and the optical density

19



403

was monitored at 600nm.

405

Automated preparation of samples for end-point analyte and reporter measurement

406

Following incubation, plates were transferred to the liquid handler. Medium was aspirated from

407

below the oil layer to a microtiter plate and centrifuged at 2000xg for 10 minutes. Supernatant was

408

transferred to a 0.2µm filter plate (Millipore MSGVN2210) and filtered at 300 psi for 120 seconds,

409

the flow through was collected for measurement of analyte concentration via HPLC. Cell pellets

410

were resuspended by repeated mixing in phosphate buffered saline and transferred to a microtiter

411

plate (Corning 3904). This plate was incubated at room temperature for thirty minutes, for a total

412

of one hour from oxygen exposure, before fluorescence determination (if required).

413

Inducer toxicity and effectiveness

414

Inducer toxicity was assessed by growing cells overnight in 5mL LB+1% glucose+50µg/mL kan

415

with corresponding inducer, either 2mM IPTG or 500ng/mL aTc. Cells were washed with PMM,

416

OD normalized to 0.05 and seeded into 96-well plates containing 150µL of medium with varying

417

inducer concentration. Cells were grown in an anaerobic chamber.

404

Page 20 of 26

418

Inducer effectiveness was assessed similarly to above, with OD being varied as an additional

419

variable. The effectiveness of switching was quantified using the ratio of fluorescent reporters in the

420

new target state to the previous state, 6 hours after induction. Cultures were grown overnight in

421

IPTG or aTc medium, before washing and diluting into fresh medium with the opposing inducer at

422

different cell densities. The cultures were then grown for 6 hours and the cells were resuspended in

423

PBS. This was followed by incubation under aerobic conditions at room temperature for 30 minutes

424

to allow the fluorescent proteins to mautre, and reading fluorescence.

425

500mL bioreactor trials

426

LAC002-D59+pTSLAC314 strains were inoculated from glycerol stocks into 5mL+1% glucose+50µg/mL

427

kan LB and grown overnight. Cultures were transferred to 200mL LB+1% glucose+50µg/mL kan

428

in 250mL sealed baffled flasks, for oxygen limitation. Cells were induced with 500ng/mL aTc for

429

the production state and 2mM IPTG for the growth state. Cells were washed 3x in RDM before

430

being transferred to nitrogen-sparged 0.5L bioreactors controlled at pH 7 with 15M KOH and 37◦ .

431

Bioreactor medium was RDM + 5g/L glucose. Growth control reactors contained 2mM IPTG,

432

production control reactors contained 500ng/mL aTc, switched reactors contained no-inducer, and

433

2mM IPTG was added at given induction points.

434

Switch stability

435

Switch stability was determined using our final strain and construct, LAC002-D59+pSLAC314.

436

Cultures were inoculated from glycerol stock into 5mL LB+1% glucose+50µg/mL kan. 50µL of 20


Page 21 of 26 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


437

culture was transferred to anaerobic tubes, prepared by sparging nitrogen for 5 minutes. Anaerobic

438

tubes were grown overnight in a shaker incubator at 250RPM and 37◦ for 24 hours. This process

439

was repeated 3x for a total 108 dilution. Samples were taken daily for HPLC and flow cytometry.

440

Analytical methods

441

For plate experiments, absorbance was measured every 15 minutes at 600nm to asses optical density.

442

Following incubation, plates were prepared as need for downstream HPLC or fluorescence applica-

443

tions. Product concentrations were measured by HPLC using an Aminex HPX-87H cation-exchange

444

mL column (BioRad, 1250140) at 0.6 min with 5mM H2 SO4 mobile phase, 60◦ column temperature and

445

20µL injection volume. Chromatograms were integrated using Chromeleon v7.

446

Fluorescence was read in a Tecan M200 plate reader equipped with a monochromater and the

447

following parameters: gfpmut3 (Ex: 488/9, Em: 525/20, K: 110) and mCherry (Ex: 570/9, Em:

448

610/20, K: 150).

449

Flow cytometry was performed using a LSRFortessa X-20 (BD Biosciences). GFP was acquired

450

using a 488nm laser and a 525/50nm emission filter, and mCherry was acquired using a 561nm

451

laser and 610/20nm emission filter. At least 25,000 single cell events were collected for each sample.

452

A FSC-A threshold of 1500 and a SSC-A threshold of 1000 were used in combination to improve

453

detection. Data was analyzed in python, using the fcsparser package. Data was gated as follows to

454

eliminate non-cell particles: side scatter 6000-20000, forward scatter 2000-15000. Two dimensional

455

log-histograms were generated using numpy, with 25 bins in each dimension, between 0.1 and 106 .

456

Analyte data analysis

457

Fermentation data was analyzed in python using the IMPACT framework, an open-source toolbox

458

for data analysis (http://github.com/nvenayak/impact). Since no glucose was remaining in the

459

cultures in any experiment, the trends observed in product/biomass titers are the same as those in

460

their respective yields.

461

Supporting Information

462

Figures S1-S14 in section A of Supporting Information are supplementary figures that have been

463

referenced in the manuscript. Sections B and C contain the DNA sequences of synthesized DNA

464

and primers used in this study respectively. Section D contains the RBS strengths as predicted

465

using the Salis RBS calculator v1.1 44 . Section E contains the nucleotide sequences of all plasmids

466

used in this study in GenBank format.

467

Author Contributions

468

NV and RM conceived the study. NV designed and constructed the plasmids and strains. NV

469

and KR performed experiments. NV analyzed data and wrote the manuscript. NV, KR, and RM 21



470

reviewed the manuscript. RJ performed preliminary studies for the manuscript.

471

Acknowledgments

472

We sincerely thank Dr. Stephen Fong (Virginia Commonwealth University, USA) for providing

473

the MG1655 ∆adhE,pta background strains and Dr. Raffi Afeyan and Dr. James Collins (Mas-

474

sachusetts Institute of Technology, USA) for providing the toggle switch backbone. We would also

475

like to acknowledge Nikolaos Anesiadis for insightful discussions regarding this work and Kayla

476

Nemr for assistance with analytics (HPLC). This work was supported by the Natural Sciences and

477

Engineering Research Council (NSERC), the NSERC Industrial Biocatalysis Network, BioFuelNet

478

Canada, the Ontario Ministry of Research and Innovation. NV would like to acknowledge funding

479

from an NSERC postgraduate scholarship.

480

References

481 482

483 484

485 486

487 488

489 490

491 492 493

494 495

496 497 498

499 500

501 502 503

1. Lee, S.Y. and Kim, H.U. (2015) Systems strategies for developing industrial microbial strains. Nat. Biotechnol. 33, 1061. 2. Stephanopoulos, G. (2012) Synthetic biology and metabolic engineering. ACS Synth. Biol. 1 (11), 514–525. 3. Van Dien, S. (2013) From the first drop to the first truckload: commercialization of microbial processes for renewable chemicals. Curr. Opin. Biotechnol. 24 (6), 1061–1068. 4. Klamt, S. and Mahadevan, R. (2015) On the feasibility of growth-coupled product synthesis in microbial strains. Metab. Eng. 30, 166–178. 5. Maia, P., Rocha, M., and Rocha, I. (2016) In Silico Constraint-Based Strain Optimization Methods: the Quest for Optimal Cell Factories. Microbiol. Mol. Biol. Rev. 80 (1), 45–67. 6. Fong, S.S., Burgard, A.P., Herring, C.D., Knight, E.M., Blattner, F.R., Maranas, C.D., and Palsson, B.O. (2005) In silico design and adaptive evolution of Escherichia coli for production of lactic acid. Biotechnol. Bioeng. 91 (5), 643–648. 7. Gawand, P., Hyland, P., Ekins, A., Martin, V.J.J., and Mahadevan, R. (2013) Novel approach to engineer strains for simultaneous sugar utilization. Metab. Eng. 20, 63–72. 8. Gold, N.D., Gowen, C.M., Lussier, F.X., Cautha, S.C., Mahadevan, R., and Martin, V.J. (2015) Metabolic engineering of a tyrosine-overproducing yeast platform using targeted metabolomics. Microb. Cell Fact. 14 (1). 9. Harder, B.J., Bettenbrock, K., and Klamt, S. (2016) Model-based metabolic engineering enables high yield itaconic acid production by Escherichia coli. Metab. Eng. 38, 29–37. 10. Harder, B.J., Bettenbrock, K., and Klamt, S. (2017) Temperature-dependent dynamic control of the TCA cycle increases volumetric productivity of itaconic acid production by Escherichia coli. Biotechnol. Bioeng. 115 (1), 156–164.

22


Page 22 of 26

Page 23 of 26 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

504 505 506

507 508

509 510

511 512 513

514 515

516 517 518

519 520

521 522 523

524 525

526 527 528 529

530 531 532 533

534 535 536

537 538

539 540 541


11. Mahadevan, R., Burgard, A.P., Famili, I., Van Dien, S., and Schilling, C.H. (2005) Applications of metabolic modeling to drive bioprocess development for the production of value-added chemicals. Biotechnol. Bioprocess Eng. 10 (5), 408–417. 12. Venayak, N., Anesiadis, N., Cluett, W.R., and Mahadevan, R. (2015) Engineering metabolism through dynamic control. Curr. Opin. Biotechnol. 34, 142–152. 13. Klamt, S., Mahadevan, R., and H¨adicke, O. (2018) When do two-stage processes outperform one-stage processes? Biotechnol. J. 13 (2), 1700539. 14. Gadkar, K.G., Doyle III, F.J., Edwards, J.S., and Mahadevan, R. (2005) Estimating optimal profiles of genetic alterations using constraint-based models. Biotechnol. Bioeng. 89 (2), 243– 251. 15. Holtz, W.J. and Keasling, J.D. (2010) Engineering static and dynamic control of synthetic pathways. Cell 140 (1), 19–23. 16. Cress, B.F., Trantas, E.A., Ververidis, F., Linhardt, R.J., and Koffas, M.A.G. (2015) Sensitive cells : enabling tools for static and dynamic control of microbial metabolic pathways. Curr. Opin. Biotechnol. 36, 205–214. 17. Brockman, I.M. and Prather, K.L.J. (2015) Dynamic metabolic engineering : New strategies for developing responsive cell factories. Biotechnol. J. 10, 1360–1369. 18. Burg, J.M., Cooper, C.B., Ye, Z., Reed, B.R., Moreb, E.A., and Lynch, M.D. (2016) Large-scale bioprocess competitiveness: the potential of dynamic metabolic control in two-stage fermentations. Curr. Opin. Chem. Eng. 14, 121–136. 19. Brockman, I.M. and Prather, K.L.J. (2015) Dynamic knockdown of E. coli central metabolism for redirecting fluxes of primary metabolites. Metab. Eng. 28, 104–113. 20. Soma, Y., Fujiwara, Y., Nakagawa, T., Tsuruno, K., and Hanai, T. (2017) Reconstruction of a metabolic regulatory network in Escherichia coli for purposeful switching from cell growth mode to production mode in direct GABA fermentation from glucose. Metab. Eng. 43 (A), 54–63. 21. Dahl, R.H., Zhang, F., Alonso-Gutierrez, J., Baidoo, E., Batth, T.S., Redding-Johanson, A.M., Petzold, C.J., Mukhopadhyay, A., Lee, T.S., Adams, P.D., and Keasling, J.D. (2013) Engineering dynamic pathway regulation using stress-response promoters. Nat. Biotechnol. 31 (11), 1039–46. 22. Xu, P., Li, L., Zhang, F., Stephanopoulos, G., and Koffas, M. (2014) Improving fatty acids production by engineering dynamic pathway regulation and metabolic control. Proc. Natl. Acad. Sci. 111 (31), 11299–11304. 23. Soma, Y. and Hanai, T. (2015) Self-induced metabolic state switching by a tunable cell density sensor for microbial isopropanol production. Metab. Eng. 30, 7–15. 24. Gupta, A., Reizman, I.M.B., Reisch, C.R., and Prather, K.L.J. (2017) Dynamic regulation of metabolic flux in engineered bacteria using a pathway-independent quorum-sensing circuit. Nat. Biotechnol. 35 (3).

23



542 543 544

545 546

547 548 549

550 551

552 553

554 555

556 557

558 559 560

561 562

563 564

565 566

567 568

569 570 571

25. Soma, Y., Tsuruno, K., Wada, M., Yokota, A., and Hanai, T. (2014) Metabolic flux redirection from a central metabolic pathway toward a synthetic pathway using a metabolic toggle switch. Metab. Eng. 23, 175–184. 26. Ptashne, M. (2004) A genetic switch : phage lambda revisited, volume 3. Cold Spring Harbor Laboratory Press, Cold Spring Harbor NY. 27. Ishiura, M., Kutsuna, S., Aoki, S., Iwasaki, H., Andersson, C.R., Tanabe, A., Golden, S.S., Johnson, C.H., and Kondo, T. (1998) Expression of a Gene Cluster kaiABC as a Circadian Feedback Process in Cyanobacteria. Science 281 (5382), 1519–1523. 28. Cameron, D.E., Bashor, C.J., and Collins, J.J. (2014) A brief history of synthetic biology. Nat. Rev. Microbiol. 12 (5), 381–390. 29. Gardner, T.S., Cantor, C.R., and Collins, J.J. (2000) Construction of a genetic toggle switch in Escherichia coli. Nature 403 (6767), 339–342. 30. Anesiadis, N., Cluett, W.R., and Mahadevan, R. (2008) Dynamic metabolic engineering for increasing bioprocess productivity. Metab. Eng. 10 (5), 255–266. 31. Anesiadis, N., Kobayashi, H., Cluett, W.R., and Mahadevan, R. (2013) Analysis and design of a genetic circuit for dynamic metabolic engineering. ACS Synth. Biol. 2 (8), 442–452. 32. Olson, E.J., Hartsough, L.a., Landry, B.P., Shroff, R., and Tabor, J.J. (2014) Characterizing bacterial gene circuit dynamics with optically programmed gene expression signals. Nat. Methods 11 (4), 449–455. 33. Bothfeld, W., Kapov, G., and Tyo, K.E.J. (2017) A Glucose-Sensing Toggle Switch for Autonomous, High Productivity Genetic Control. ACS Synth. Biol. 6 (7), 1296–1304. 34. Xiao, Y., Bowen, C.H., Liu, D., and Zhang, F. (2016) Exploiting non-genetic, cell-to-cell variation for enhanced biosynthesis. Nat. Chem. Biol. 12 (5), 339–344. 35. Roquet, N., Soleimany, A.P., Ferris, A.C., Aaronson, S., and Lu, T.K. (2016) Synthetic recombinase-based State machines in living cells. Science 353 (6297). 36. Lo, T.M., Chng, S.H., Teo, W.S., Cho, H.S., and Chang, M.W. (2016) A Two-Layer Gene Circuit for Decoupling Cell Growth from Metabolite Production. Cell Syst. 3 (2), 133–143. 37. Litcofsky, K.D., Afeyan, R.B., Krom, R.J., Khalil, A.S., and Collins, J.J. (2012) Iterative plugand-play methodology for constructing and modifying synthetic gene networks. Nat. Methods 9 (11), 1077–1080.

573

38. Klumpp, S., Zhang, Z., and Hwa, T. (2009) Growth rate-dependent global effects on gene expression in bacteria. Cell 139 (7), 1366–75.

574

39. Tsien, R.Y. (1998) The Green Fluorescent Protein. Proteins 67 (11), 509–544.

572

577

40. Drepper, T., Eggert, T., Circolone, F., Heck, A., Krauss, U., Guterl, J.K., Wendorff, M., Losi, A., Gartner, W., and Jaeger, K.E. (2007) Reporter proteins for in vivo fluorescence without oxygen. Nat. Biotechnol. 25 (4), 443–445.

578

41. Nystr¨om, T. (2004) Stationary-Phase Physiology. Annu. Rev. Microbiol. 58 (1), 161–181.

575 576

24


Page 24 of 26

Page 25 of 26 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

579 580

581 582 583

584 585

586 587 588

589 590 591

592 593 594

595 596

597 598 599

600 601

602 603

604 605 606

607 608 609

610 611 612


42. Navarro Llorens, J.M., Tormo, A., and Mart´ınez-Garc´ıa, E. (2010) Stationary phase in gramnegative bacteria. FEMS Microbiol. Rev. 34 (4), 476–495. 43. Shaikh, A.S., Tang, Y.J., Mukhopadhyay, A., Mart´ın, H.G., Gin, J., Benke, P.I., and Keasling, J.D. (2010) Study of stationary phase metabolism via isotopomer analysis of amino acids from an isolated protein. Biotechnol. Prog. 26 (1), 52–56. 44. Salis, H.M., Mirsky, E.A., and Voigt, C.A. (2009) Automated design of synthetic ribosome binding sites to control protein expression. Nat. Biotechnol. 27 (10), 946–950. 45. Lee, J.W., Gyorgy, A., Cameron, D.E., Pyenson, N., Choi, K.R., Way, J.C., Silver, P.A., Del Vecchio, D., and Collins, J.J. (2016) Creating Single-Copy Genetic Circuits. Mol. Cell 63 (2), 329–336. 46. Espah Borujeni, A., Channarasappa, A.S., and Salis, H.M. (2014) Translation rate is controlled by coupled trade-offs between site accessibility, selective RNA unfolding and sliding at upstream standby sites. Nucleic Acids Res. 42 (4), 2646–2659. 47. Campos-Bermudez, V.A., Bologna, F.P., Andreo, C.S., and Drincovich, M.F. (2010) Functional dissection of Escherichia coli phosphotransacetylase structural domains and analysis of key compounds involved in activity regulation. FEBS J. 277 (8), 1957–1966. 48. McGinness, K.E., Baker, T.A., and Sauer, R.T. (2006) Engineering Controllable Protein Degradation. Mol. Cell 22 (5), 701–707. 49. Andersen, J.B., Sternberg, C., Poulsen, L.K., Bjørn, S.P., Givskov, M., and Molin, S. (1998) New unstable variants of green fluorescent protein for studies of transient gene expression in bacteria. Appl. Environ. Microbiol. 64 (6), 2240–2246. 50. Lies, M. and Maurizi, M.R. (2008) Turnover of endogenous SsrA-tagged proteins mediated by ATP-dependent proteases in Escherichia coli. J. Biol. Chem. 283 (34), 22918–22929. 51. Datsenko, K.a. and Wanner, B.L. (2000) One-step inactivation of chromosomal genes in Escherichia coli K-12 using PCR products. Proc. Natl. Acad. Sci. 97 (12), 6640–6645. 52. Kok, S.D., Stanton, L.H., Slaby, T., Durot, M., Holmes, V.F., Patel, K.G., Platt, D., Shapland, E.B., Serber, Z., Dean, J., Newman, J.D., and Chandran, S.S. (2014) Rapid and reliable DNA assembly via ligase cycling reaction. ACS Synth. Biol. 3 (2), 97–106. 53. Torella, J.P., Boehm, C.R., Lienert, F., Chen, J.H., Way, J.C., and Silver, P.a. (2013) Rapid construction of insulated genetic circuits via synthetic sequence-guided isothermal assembly. Nucleic Acids Res. 42 (1), 681–689. 54. Causey, T.B., Zhou, S., Shanmugam, K.T., and Ingram, L.O. (2003) Engineering the metabolism of Escherichia coli W3110 for the conversion of sugar to redox-neutral and oxidized products: homoacetate production. Proc. Natl. Acad. Sci. 100 (3), 825–832.

25


Growth Phenotyope

O OH

ACS Synthetic Biology Page 26 of 26 O

OH O OH

OH

OH OH

Growth

1 2 3 4 5

Production

O

Production Phenotype


O OH O OH OH

Growth

Production

OH

OH

OH

An Optimized Bistable Metabolic Switch To Decouple Phenotypic

Recommend Documents