Effects of Macromolecular Crowding on Gene Expression Studied In

Effects of macromolecular crowding on gene expression studied in protocell models The work described in this thesis was supported by a European Research Council (ERC), Advanced Grant (246812 Intercom) and a VICI grant from the Netherlands Organisation for Scientiﬁc Research (NWO)

ISBN: 978-94-6295-151-8

Cover design: Denis Arslanov and Ekaterina Sokolova

Printed & Lay Out by: Proefschriftmaken.nl || Uitgeverij BOXPress

Published by: Uitgeverij BOXPress, ’s-Hertogenbosch Effects of macromolecular crowding on gene expression studied in protocell models

Proefschrift

ter verkrijging van de graad van doctor aan de Radboud Universiteit Nijmegen op gezag van de rector magniﬁcus prof. dr. Th.L.M. Engelen, volgens besluit van het college van decanen in het openbaar te verdedigen op dinsdag 12 mei 2015 om 14.30 uur precies

door Ekaterina Alexandrovna Sokolova geboren op 30 augustus 1987 te Angren (Uzbekistan) Promotor: Prof. dr. W. T. S. Huck

Manuscriptcommissie: Prof. dr. ir. J. C. M. van Hest (voorzitter)

Prof. dr. H. N. W. Lekkerkerker (Universiteit Utrecht)

Dr. V. Noireaux (University of Minnesota, VS) Effects of macromolecular crowding on gene expression studied in protocell models

Doctoral Thesis

to obtain the degree of doctor from Radboud University Nijmegen on the authority of the Rector Magniﬁcus prof. dr. Th.L.M. Engelen, according to the decision of the Council of Deans to be defended in public on Tuesday, May 12, 2015 at 14.30 hours

by Ekaterina Alexandrovna Sokolova Born on August 30, 1987 in Angren (Uzbekistan) Supervisor: Prof. dr. W. T. S. Huck

Doctoral Thesis Committee: Prof. dr. ir. J. C. M. van Hest (chairman)

Prof. dr. H. N. W. Lekkerkerker (University of Utrecht)

Dr. V. Noireaux (University of Minnesota, USA) To my parents - Tatyana and Aleksandr

Contents

1 Macromolecular crowding and its inﬂuence on the chemistry of the cell 1 1.1 What is macromolecular crowding? ...... 1 1.2 Theory of the effects of crowding ...... 3 1.2.1 Thermodynamical framework ...... 4 1.2.2 Molecular interpretations (hard spheres approximation and depletion forces) ...... 6 1.2.2.1 BMCSL (Boublik-Mansoori-Carnahan-Starling-Leland) model ...... 7 1.2.2.2 Scaled-particle theory (SPT) ...... 10 1.2.2.3 Depletion interactions ...... 12 1.2.3 Dynamics effects of crowding ...... 14 1.3 Experimental investigations of crowding ...... 16 1.3.1 Approaches to mimic crowded media in vitro ...... 16 1.3.2 Type of crowders: advantages & disadvatages ...... 16 1.3.3 Reported effects of macromolecular crowding ...... 17 1.4 Biochemical model reactions ...... 18 1.4.1 Association equilibria ...... 18 1.4.2 Enzymatic activity ...... 18 1.4.3 Conformational changes ...... 19 1.4.4 Protein folding ...... 20 1.4.5 Other effects ...... 20 1.4.6 Attractive interactions ...... 21 1.5 Aim of the research presented in this thesis ...... 21

v vi Contents

2 A microfluidics platform for quantitative characterization of IVTT reaction in droplets 27 2.1 Introduction ...... 28 2.1.1 The course of gene expression in the cell ...... 28 2.1.2 Transcription and translation ...... 29 2.1.3 Cell-free gene expression ...... 29 2.1.4 Cell free expression in vesicles ...... 31 2.1.5 Cell free expression in microdroplets ...... 33 2.1.5.1 Overview of microfluidics ...... 34 2.1.5.2 In vitro gene expression in droplets ...... 35 2.2 Experimental protocols ...... 36 2.2.1 Materials ...... 36 2.2.2 Laser-induced fluorescence (LIF) and microfluidic setup operation 37 2.2.3 Device fabrication ...... 38 2.2.4 Operation of chamber and bilayer devices ...... 40 2.3 Results and Discussion ...... 41 2.3.1 In vitro gene expression from a single copy of DNA in droplets 42 2.3.2 Kinetic measurements of IVTT in droplets ...... 44 2.3.3 Control over water contents of the droplets ...... 45 2.4 Conclusion: From “bulk” to “cell-like” ...... 45

3 Cell lysate coacervates - possible models of crowding in cells 49 3.1 Introduction ...... 49 3.1.1 Spatial organization of the cell ...... 49 3.1.2 Compartmentalization induced by phase separation ...... 50 3.1.3 Coacervation as a route to crowding ...... 51 3.2 Results and Discussion ...... 52 3.2.1 Physical properties of cell lysate coacervates ...... 53 3.3 Materials and methods ...... 60 3.3.1 Phase separation of cell-free expression kit containing fluorescent PEG and lysate ...... 61 3.3.2 Phase separation of PEG in the absence of cell-free expression kit 61 3.3.3 Calculation of the distribution of PEG over the two phases . . . 61 3.3.4 Inductively-coupled plasma optical emission spectrometry (ICP- OES) ...... 62 3.3.5 Covalent labeling of the plasmid ...... 62 3.3.6 Data Acquisition and Analysis ...... 63 3.3.7 Calculation of partition coefficients of PEG and fluorescently labelled lysate ...... 63 3.3.8 Fluorescence recovery after photobleaching experiments . . . . 63 Contents vii

3.4 Conclusion ...... 65

4 Gene expression in crowded membrane-free protocells 69 4.1 Introduction ...... 69 4.1.1 Overview of the protocell models ...... 69 4.2 Results and Discussion ...... 73 4.2.1 Modeling of transcription and translation in cell lysate with deterministic rate equations ...... 73 4.2.2 Transcription and translation in membrane-free protocells formed by coacervation of cell lysate ...... 76 4.3 Materials and Methods ...... 81 4.3.1 Materials ...... 81 4.3.2 Home made in vitro transcription translation system ...... 82 4.3.3 Molecular beacons for mRNA labeling ...... 83 4.3.4 Data acquisition and analysis ...... 84 4.3.5 Method for mRNA production experiments in cell lysate . . . . 84 4.3.6 Method for GFP production experiments in cell lysate . . . . . 84 4.3.7 Method for transcription in bulk at various PEG concentrations . 85 4.3.8 Method for translation in bulk at various PEG concentrations . . 86 4.4 Conclusion ...... 86

5 Understanding the effect of macromolecular crowding on genetic networks in synthetic cellular systems 89 5.1 Introduction ...... 90 5.2 Results and Discussion ...... 91 5.3 Conclusion ...... 97

6 Stochastic gene expression in a crowded environment 101 6.1 Introduction ...... 102 6.1.1 Stochastic gene expression in vivo ...... 102 6.1.2 In vitro system for cell mimics ...... 104 6.2 Materials & Methods ...... 105 6.2.1 IVTT system ...... 105 6.2.2 Data acquisition and analysis ...... 106 6.3 Results and Discussion ...... 108 6.3.1 Aim of this study and limitations of approach taken ...... 108 6.3.2 Inﬂuence of different molecular weights dextran on expression of β-glucuronidase ...... 108 6.3.3 Stochastic β-glucuronidase expression in droplets ...... 109 6.3.3.1 Time of arrivals and kinetics distributions ...... 110 viii Contents

6.3.3.2 Individual droplet analysis ...... 114 6.4 Conclusions and future perspectives ...... 116 6.4.1 Microﬂuidics and IVTT system ...... 116 6.4.2 Implications of noise and crowding ...... 117

Summary 119

Samenvatting 123

Dankwoord 127 Chapter 1 Macromolecular crowding and its inﬂuence on the chemistry of the cell

1.1 What is macromolecular crowding?

A cell is the common structural unit shared by all living organisms. From a physico- chemical point of view cells are extremely complex systems, characterized by small volumes, highly concentrated in large molecules and complexes, but low copy numbers of each individual component, all together surrounded and subdivided into compartments by ubiquitous interfaces. The typical concentration of macromolecules (for example enzymes, filaments, nucleic acids, glycoproteins, lipids, etc.), in the cell is 300-400 g/L [1]. Despite significant progress being made it is still unclear how high abundance of macromolecules in the cell can influence the outcome of biochemical reactions. An artistic impression of the complex environment inside a eukaryotic cell can be seen in Fig. 1.1. The outside surface area of a typical 20 µm human tissue cell is approximately 2400 µm2, but the total area covered by internal membranes (rough and smooth endoplasmic reticulum, Golgi complex, etc.) is around 50 times larger (>120000 µm2) [1]. As a result, the average separation between interfaces (i.e., membranes) within the cell is only about 50 nm, which, coupled to the high concentration of proteins inside a cell leads to a dense and highly confined environment. Cryo-electron tomography gives direct evidence of the crowded interior of the cells. In this technique cells are rapidly frozen and studied under electron microscope with a spatial resolution of 5-6 nm. The

1 2 Chapter 1: Macromolecular crowding and its inﬂuence on the chemistry of the cell

Figure 1.1: Cartoon of eukaryotic cytoplasm magniﬁed ×106 (reproduced from [2]), modi- ﬁed [3]

high density of actin filaments and ribosomes seen in reconstructed images (Fig. 1.2) of slime mould Dictyostelium indicates that cytoplasm is filled with large amounts of macromolecules assembled into complexes, rather than with freely diffusing and collid- ing macromolecules [4]. Figure 1.2 clearly shows that due to the high concentration of macromolecules, there is very little‘free space’ in the cell. It is important to mention that it is not the high concentration of a single macromolecule that causes the cell to be ’fully packed’, but the total concentration of macromolecules. Those macromolecules are commonly called ‘crowders’ if they display no specific interactions. This phenomenon was termed ‘macromolecular crowding’ by Minton in 1981 [5]. The number and type of molecules in the cell depend on the cell type and probably on the cell cycle stage [6]. The total protein content of the cell is estimated to be 50-400 g/L corresponding roughly to 5- 40 % of the total cell volume [2, 3]. Zimmerman and Trach estimated the protein content of E. coli to be around 10-40 % in units of weight/volume [7]. Similarly, Lanni et. al. obtained a value of 200-300 mg/mL for 3T3 fibroplasts [8]. Since most of the space in the cell is already occupied by macromolecules, it is tempting to ask how proteins express and function in such an environment. This is particularly important because most of the research on proteins was performed in vitro in dilute solutions. In fact, experimentalists often make some efforts to use the most dilute solutions in order to avoid ‘non-idealities’ and to focus on the properties of pure protein solutions. This raises another question: to what extent can the conclusions derived from in vitro studies be applied to in vivo situa- tions? Various non-idealities could arise in the cell, such as excluded volume effects and non-specific interactions. In addition, the interior of the cell is much more viscous than most dilute solutions used for in vitro studies. 1.2. Theory of the effects of crowding 3

Figure 1.2: Visualization of actin network, membranes, and cytoplasmic macromolecular complexes. Volume of 815 nm by 870 nm by 97 nm, corresponding to the area of Fig. 1.2 framed in black, was subjected to surface rendering. Colors were subjectively attributed to linear elements to mark the actin ﬁlaments (reddish); other macromolecular complexes, mostly ribosomes (green); and membranes (blue) [4]. Reproduced from[4]

1.2 Theory of the effects of crowding

The consequences of excluded volume effects are theoretically the easiest to demonstrate and derive some equations for, which could help in understanding the experiments. Ex- cluded volume effects occur with all macromolecules and are particularly important in vivo due to high concentrations in the cytosol. The concept of volume exclusion was first introduced by polymer chemist Kuhn to explain the observation that real polymer chains tend to show less compaction than would be expected in the absence of volume effects [9]. The description of non-ideal gases (using the van’t Hoff isobar) also relies on the concept of an excluded volume [10]. The simplest explanation of this phenomenon is that two molecules, each having a finite molecular volume, cannot occupy the same space at the same time. Consequently, there is a zone surrounding each molecule from which other molecules are physically excluded. This can be illustrated by a pair of solid spheres (Fig. 1.3) whose centres must always be separated by at least the sum of their radii, if the molecules are assumed to have a true hard core repulsion. 4 Chapter 1: Macromolecular crowding and its influence on the chemistry of the cell

Figure 1.3: Excluded (pink and black) and available (blue) volume in a solution of spherical background macromolecules. (A) volume available to a test molecule of inﬁnitesimal size; (B) volume available to a test molecule of size comparable with background molecules

1.2.1 Thermodynamical framework Qualitatively, crowding is believed to have two main effects:

1. changing energy levels of reactants and complexes

2. changing diffusion constants

To understand the ﬁrst effect more quantitatively, we write a model chemical reaction, such as site-binding (LS) or bimolecular association (AB) (see Fig. 1.4). This thermodynamic cycle relates the free energy of association in the absence and presence of background interactions. The apparent equilibrium constant for association is related to the standard free energy of association by thermodynamic relationship. Back- ground interaction will increase the apparent equilibrium constant for association when the sum of free energies of individual background interactions A and B is more positive than the free energy of background interactions of the heterodimer AB. The reactions that increase available volume should be favoured by crowding conditions. The magnitude of these effects depends on the size and shapes of the reactants. The effect of crowding on one of these equilibria can be analysed by constructing a simple thermodynamic cycle as shown in Fig. 1.4. Using this cycle and the corresponding formulas, Zhou et. al. [11] showed that

∆FAB ≡ −RT lnKAB,

0 crowd crowd crowd ∆FAB − ∆FAB = ∆FAB − (∆FA + ∆FB ), 1.2. Theory of the effects of crowding 5

Figure 1.4: Thermodynamic cycles showing the link between free energy of transfer of reactants and products from dilute solution to crowded medium and standard free energy of (top) association and (bottom) site binding [11]

C crowd crowd crowd KAB ∆FA + ∆FB − ∆FAB 0 = exp , KAB RT

C ex KAB ∆µ 0 = exp , KAB kBT

−∆∆F K = K0 exp X , (1.1) X X RT where X stands for AB (bimolecular association) or LS (site-binding), K0 is the reaction equilibrium constant in dilute solution, R is the molar gas constant and T the absolute temperature, ∆µex is the change in chemical potential of excluded volume free energy of 6 Chapter 1: Macromolecular crowding and its inﬂuence on the chemistry of the cell

reaction AB. ∆∆FX is the free energy of association or the free energy of site binding in a crowded environment, respectively:

0 ∆∆FAB ≡ ∆FAB−∆FAB = C C C (1.2) = ∆FAB − ∆FA −∆FB or 0 ∆∆FLS ≡ ∆FLS − ∆FLS = C C (1.3) = ∆FLS−∆FL C where ∆FX is the standard free energy change of the transfer of X from bulk solution to the crowded medium. In this way the magnitude of the effect of crowding on reaction equilibria can be indirectly found by comparing the free energies of transfer of reactants and transfer of products from bulk solution to the crowded medium. Theory suggests that, when two reactive species, such as an enzyme and a ligand, are in a crowded medium, entropy favours complexation. All other macromolecules will have higher entropy when the two reactants are bound together because their available volume increases, thus increasing the chance of enzyme and ligand interacting. It should be noted that classical theory of crowding [12–14], predicts that the size ratio between the crowding molecules and the reacting species is the key parameter that determines the effect on reaction equilibria and kinetics. This can be understood as follows: generally, two effects must be distinguished when studying the effect of the volume fraction of crowders on reactions: 1. an increase in the amount of crowders leads to an increase in volume fraction

2. an increase in the size of the crowders also leads to an increase in their volume fraction At equal weight/volume concentrations, the low molecular weight crowders should lead to the largest enhancement of binding constants, since more of the smaller crowders are present per unit volume. This effect of the size of the crowders, in relation to the size of the probe, has been conﬁrmed experimentally [15]. The importance of size and concentration of crowders will be discussed in more detail below. Due to strong excluded volume interactions reactions that increase available volume, such as molecular assosia- tion, protein folding and aggregation, should be at least theoretically favored by crowding conditions [11].

1.2.2 Molecular interpretations (hard spheres approximation and depletion forces) The magnitude of the excluded volume effect between two molecules depends on their relative sizes and shapes. Non-spherical objects can have much larger exclusion volumes than spherical objects of the same volume [5, 16]. A theoretical modelling study 1.2. Theory of the effects of crowding 7 examining real fluids was conducted by MacMillan & Mayer to investigate a related phenomenon [17]. They calculated the work required to place one new particle into a fluid containing a large number of other particles. Their solution relies on viral coefficients and a given interaction function between the molecules. An important restriction is that the model uses only hard core repulsion, i.e., it assumes that the only effect of crowding is that the potential becomes infinite if the distance between two particles is less than the overlap distance. The advantage of this restriction is that the whole set of possible molecular interactions can be modelled by representing each particle as a single object with defined shape and size. However, to address the general case, it is necessary to adapt this approach to systems with multiple types of molecules. The most common way of addressing this issue is to use an equation of state, such as BMCSL (Boublik- Mansoori-Carnahan-Starling-Leland model), or the approximate models based on scaled particle theory (SPT). This approach was initially developed to describe changes in activity for fluids consisting of hard spherical particles [18] but was later extended to cover non-spherical particles as well [19, 20].

1.2.2.1 BMCSL (Boublik-Mansoori-Carnahan-Starling-Leland) model

The BMCSL equation of state describes the volumetric interactions in mixtures of spheres of different sizes and is considered to be accurate up to volume fraction of 0.6 [21, 22]. As a first approximation, we assume that the solution contains two types of spheres (1 and 2), 1 are reactants, 2 are crowders, and they interact only via hard-sphere volume exclusion. The volume fractions are indicated by ϕ, the size by σ, the chemical potential in the system by µ. The analytical expression for the chemical potential of a species i (molar Gibbs free energy) is given in [23] and is a measure for the excluded volume energy associated with each spherical particle i in the mixture. For mixtures in the so- called tracer limit (few particles of type 1, the reactants in a dispersion of many identical particles of type 2, the crowders) this expression simplifies to: µex 1 = −(1 − 3α2 + 3α3)ln(1 − ϕ )3+ k T 2 B (1.4) αϕ2 2 2 + 3 [5α − 3αϕ2 + 2 + (3ϕ2 − 6ϕ2 + 1)(1 + α − α )] (1 − ϕ2) or, more compact: ex µ1 2 3 3 = −(1 − 3α + 3α )ln(1 − ϕ2) + kBT 2 (1.5) α(3 + 2α − 3ϕ2) 1 + α − α +αϕ2 3 + 3 , (1 − ϕ2) 1 − ϕ2 where ϕ2 is the volume fraction of crowders and α is the ratio of diameters of the reactants and the crowders: α = σ1/σ2. 8 Chapter 1: Macromolecular crowding and its influence on the chemistry of the cell

This expression with only two parameters (volume fraction of crowders and the size ratio of reactants to crowders) can be used to estimate the effect of crowding (notably, the effect of volume fraction of crowders) on a model chemical reaction, such as site-binding (LS) or bimolecular association (AB). We write c ex K = K0exp(−∆µ /kBT), (1.6) ex where K0 is the reaction equilibrium constant in dilute solution and ∆µ is the change in excluded volume energy of the reaction (AB or LS). For site-binding, we can make an assumption that the excluded volume energy of the c c binding site with and without substrate is the same ( ∆FS = ∆FLS), such that the total ex ex reaction energy change is simply ∆µ = −∆µL (Eq. 1.4). For bimolecular association, we assume that two identical reactants of size σ1 and 3 volume v1 = πσ1 /6 react to form a single complex of volume vcomplex = 2v1 (conserva- 1/3 1/3 tion of volume), such that σcomplex = 2 σ1 and αcomplex = σcomplex/σcrowders = 2 α, with α = σ1/σcrowders. The change in excluded volume energy of the reaction is ∆µex µex 2µex AB = complex − 1 = kBT kBT kBT √3 ! (1.7) 2 δα 2 + δα − 2 = −(3δα − 1)ln(1 − ϕ2) − 3αϕ2 2 + , (1 − ϕ2) 1 − ϕ2 where δ = (2 − 22/3) is a numerical constant. One can visualize the effect of crowding (i.e., the volume fraction of crowders ϕ2) on the equlibrium constant Kc by plotting 10− ex c log(e)∆µ /kBT = log(K /K0) as a function of ϕ2 (Fig. 1.5).

Figure 1.5: Equilibrium constant for bimolecular association (AB) and site binding (LS) at different volume fractions α for binding association equilibria AB and site binding LS reactions

Clearly, the conclusion that small crowders have larger effects at the same volume fraction is recovered. Alternatively, the effect of crowding can be estimated at equal 1.2. Theory of the effects of crowding 9

weight concentration of the crowders ω2. If the crowders are treated as solid hard spheres with a constant mass density ρ2, the plots will not change, apart from a constant, since ϕ2 = ω2/ρ2 as depicted in Fig. 1.6.

Figure 1.6: Equilibrium constant for bimolecular association (AB) and site-binding (LS) at equal weight concentrations (g/L) for binding association equilibria reaction AB

This is argued to be the case for small globular proteins as crowders, with an average mass density ρ=1.43 g/mL [24] (however, note that [25] argues that ρ depends on σ for proteins). If, on the other hand, the crowders are treated as ﬂexible polymers, which are strongly solvated coils in solution, the plots change slightly. We note that the volume fractions in Equations 1.5 and 1.7 can be written as

3 πσ2 ω2NAν ν2ω2NAv ϕ2 = = (1.8) 6M2 M2 √ As the size of an ideal polymer chain scales as σ ∼ b N, with b the polymer Kuhn length and N the number of Kuhn monomers, and the polymer mass M2 = NMmon, the polymer mass can be calculated from the ratio of the size to the Kuhn length:

2 M2 = (σ2/b) Mmon (1.9)

1/3 and the volume fraction will scale with σ ∼ ν2 at equal weight concentrations: 2 πb σ2ω2NAν ϕ2 = (1.10) 6Mmon If we take a typical Kuhn length (1 nm) and monomer weight (100 g/mol) for PEO, we find that the plots change. For small sizes of polymers the conclusion that small crowders have larger effect on the binding constants is justified. Figure 1.7 shows that for large polymeric crowders, however, the crowding effect at equal weight concentration starts to become more important again (there seems to be a minimal effect at intermediate polymeric crowder size (at ω2 ≈5 g/L). 10 Chapter 1: Macromolecular crowding and its influence on the chemistry of the cell

Figure 1.7: Equilibrium constant for bimolecular association (AB) and site-binding (LS) for ﬂexible crowders at equal volume fraction 0.5 for binding association equilibria reaction AB

It must be noted, however, that the approximations in Equations 1.9 and 1.10 of non-overlapping ideal chains are violated already at very low weight concentrations for long polymers. At that point, alternative models for the excluded volume by overlapping polymers should be employed to estimate the effect of polymeric crowders of different molecular weights. For other (more realistic) particle shapes than spheres one should use the more empirical scaled particle theory or even numerical simulations.

1.2.2.2 Scaled-particle theory (SPT)

Scaled particle theory for calculating chemical potential of hard particles can be applied in two special cases:

1. all particles are spherical and of equal size

2. mixtures of particles of different shapes and sizes at low volume fraction (initial deviations from ideal behaviour)

The basic assumption in this theory is that long-range interactions between macromolecules are strongly screened. As a consequence, hard-particle interactions sufﬁce to quantitatively describe the excluded-volume effects in solutions of proteins. By using the scaled particle theory (SPT) of hard-particle ﬂuids [18], Zhou et. al. [11] could give a theoretical example of the effect of crowding on site-binding equilibria. Let

∆Fcrowd sc = −ln(1 − Φ) + A Q + A Q2 + A Q3, (1.11) RT 1 2 3 with Φ Q = , 1 − Φ 1.2. Theory of the effects of crowding 11

3 2 2 A1 = Rsc + 3Rsc + 3Rsc + 1.5Lsc(Rsc + 2Rsc + 1),

3 2 2 A2 = 1.5(2Rsc + 3Rsc) + 4.5Lsc(Rsc + Rsc),

3 2 A3 = 3Rsc + 4.5LscRsc,

crowd where ∆Fsc represents the free energy of placing a spherocylinder (sc), with a radius rsc = Rscrc and a cylindrical length of 2Lscrsc into a ﬂuid containing a volume fraction Φ of inert hard spheres (crowders) with radius rc. Combining this Equation 1.11 with Equations 1.1 and 1.3 and representing the free ligand as a sphere (Lsc = rsc) produces Fig. 1.8. The assumption that bound ligand is buried and thus inaccessible to crowders crowd (∆Fsc = 0) gives an upper boundary of the effect of crowding on the free ligand.

Figure 1.8: Effect of the volume fraction Φ of crowders upon the equilibrium constant for site-binding by a spherical ligand, for MWligans/MWcrowder=2 (a), 1 (b) or 0.5 (c)

Figure 1.8 shows an enhancement on site-binding by crowding in the theoretical (ideal) case. The comparison with Fig. 1.5 is almost quantitative (note that in Fig. 1.8 the 3 ratio of masses of solid hard spheres Mligans/Mcrowder = α is used instead of the ratio of sizes α). The divergence between dilute and crowded reactions becomes much more pronounced at higher volume fractions (>0.3). Furthermore, the effect is also strongly dependent on the relative size of the crowder and the ligand. At equal weight/volume concentrations, the low molecular weight crowders should lead to the largest enhancement of binding constants, since more copies of the smaller crowders are present per volume unit as indicated by the difference between three traces in Fig. 1.8. 12 Chapter 1: Macromolecular crowding and its inﬂuence on the chemistry of the cell

1.2.2.3 Depletion interactions

A depletion force or depletion attraction has been extensively investigated in colloid- polymer mixtures [26–29]. Depletion forces are often regarded as entropic forces, as was ﬁrst explained by the established Asakura-Oosawa-Vrij model [26, 27]. In this theory the depletion force arises from an increase in osmotic pressure of the surrounding solution when colloidal particles get close enough such that the excluded cosolutes (depletants) cannot ﬁt in between them (Fig. 1.9). Thus this phenomenon involves an entropy increase in crowding particles when two reactants associate.

Figure 1.9: The depletion interaction. Polymer coils are excluded from a depletion zone close to the surface of colloidal particles; when the depletion zones of two particles overlap there is a net attractive force between the particles arising from unbalanced osmotic pressure

An alternative approach to quantitatively estimate the effect of crowding on an association reaction is to treat the association as a process in which the volume depleted for crowders changes. In other words, the crowders favour the association by inducing a depletion attraction between the reactants. In general the free energy associated with this depletion interaction is given by:

Fdep = −∆VdepΠ, (1.12) where Vdep is the change in depleted volume and Π is the osmotic pressure due to the crowders. The change in depleted volume of two spheres of diameter σ1, surrounded by a crowder spheres of diameter σ2, approaching each other to a center-to-center distance D (from inﬁnity) is given by

3 π 3 3D D ∆Vdep = (σ1 + σ2) 1 − + 3 (1.13) 6 2(σ1 + σ2) 2(σ1 + σ2) 1.2. Theory of the effects of crowding 13

Since the closest approach center-to-center distance of two identical hard spheres of type 1 is Dmin = σ1, the maximum change in depleted volume is

π 3α α3 ∆V ∗ = (σ + σ )3 1 − + , (1.14) dep 6 1 2 2(1 + α) 2(1 + α)3 where α = σ1/σ2, as in the previous section. 3 The osmotic pressure of a solution of spheres with weight concentration ω2 (g/m or g/L) can be calculated from:

2 ! ω2RT ω2 ω2 Π = 1 + B2 +C2 + ... , (1.15) M2 M2 M2

3 where B2 (C2, ...) is the second (third, ...) virial coefﬁcients (m /mol), which is B2 = βNAν /2 with β the excluded volume of a single molecule. For hard spheres, β = 8ν, i.e., a single sphere excludes 8 times its own volume for other spheres of the same size [30]. Combining Equations 1.12, 1.14 and 1.15, gives the following equation for free energy associated with the depletion interaction between two spheres in a crowded solution:

F ω N π 2σ 3ω N π dep = − 2 Aν 1 + 2 2 Aν × k T 6M 3M B 2 2 (1.16) 3α α3 ×(σ + σ )3 1 − + 1 2 2(1 + α) 2(1 + α)3

Or, for solid hard spheres (no polymeric crowders), using the volume fraction ϕ2 = ν2ω2NAν /M2, the resulting Equation 1.16 becomes:

Fdep 2 = −(2ϕ2 + ϕ2/2)(3α + 2) (1.17) kBT This approach gives very similar results to the approach in the previous section, as can be seen from Fig. 1.10. Only for very small crowders at high weight concentration, the depletion interaction approach underestimates the effect of crowding, probably due to the fact that third and higher order virial coefficients have been neglected, whereas three-body interactions are likely to occur in these solutions (many small particles). The increase in the equilibrium constant and the induced effective attraction are both nonspecific effects induced by macromolecular crowding, irrespective of specific interactions between the associating macromolecules. When the crowder density is increased, macromolecular associations are expected to increase because of the increase in the equilibrium constant or, equivalently, because of the amplified effective attraction. It is obvious, however, that the macromolecules in the cell not only exhibit hard-sphere-like 14 Chapter 1: Macromolecular crowding and its influence on the chemistry of the cell

Figure 1.10: Equilibrium constant for bimolecular association (AB) for solid crowders at equal weight concentrations (g/L) (left) and flexible crowders at equal volume fraction (right) using depeletion interaction approach excluded volume interactions but may also additional nonspecific repulsive or attractive interactions such as van der Waals interactions, electrostatic interactions, hydrogen bonding interactions, and hydrophobic interactions. When the crowders have an intrinsic attraction towards the reactants, the crowding effect becomes less significant and even unfavourable for molecular associations in the case of very strong attractions. Therefore, overall crowding effects will derive from volume exclusion effects that can be compen- sated to a different extent by these additional nonspecific interactions [31, 32].

1.2.3 Dynamics effects of crowding

The macromolecular crowding affects the thermodynamics of reactants by volume exclusion effects, but also leads to hydrodynamic effects (such as change in the viscosity). A change in viscosity should slow down all (diffusion-controlled) kinetic processes and decrease the association rates. In a homogeneous solvent where solute size is comparable to or greater than that of the solvent, solute diffusion is described well by statistical equations and determined by solute size and shape. This type of diffusion is called "normal". From Fick’s laws or from Einstein’s relations, the mean squared displacement (MSD), hr2i, of a solute particle in three dimensions is related to the, D, by

hr2i = 6Dt (1.18)

Smoluchowski calculated the relationship between the rates of chemical reactions and diffusion coefﬁcient of participating molecules, setting the diffusion limit of reaction rates [33]. The diffusion coefﬁcient for biological molecules strongly depends on the 1.2. Theory of the effects of crowding 15 size of the molecules according to Einstein-Stokes equation:

kT D = , (1.19) f where f is the solvent friction coefficient. For a spherical particle in a hydrodynamic solvent of shear viscosity η, f = 6πηrh, where rh is the hydrodynamic radius of the particle. Small molecules, (such as sugars and nucleotides) of approximately 0.5 nm diameter, diffuse quickly with a diffusion coefficient of about 100 µm2/s; molecules of the size of the protein (3-5 nm) diffuse more slowly (D∼3-10 µm2/s); whereas larger vesicles diffuse as slowly as D∼0.1 µm2/s, taking hours to travel across a cell of 15 µm in diameter [34]. Molecular crowding and stickiness of the cellular environment can lead to sub-diffusion or anomalous diffusion [35], making such journeys even longer and significantly slowing all reactions in the cell. Solute diffusion that cannot be described by Einstein’s equations for Brownian motion (Equations 1.18 and 1.19) is defined as anomalous diffusion. The non-Brownian behavior is usually described by the equation:

hr2i = 6Dtα (1.20)

If α<1, the diffusion is called anomalous subdiffusion, and if α>1, the diffusion is called anomalous superdiffusion.

High abundancy of various macromolecules and lipids constituting up to 40 % of the cell’s volume [36, 37] explains its large macroviscosity. Yet, at the molecular scale, small proteins move freely by diffusion in various cell compartments, including the cell nucleus [38], cytoplasm [39–41], endoplasmic reticulum [42], and mitochondria [43]. All of these observations imply that crowding environment of cells exhibit different rheological characteristics at the molecular- and macroscales. In 1951 Mooney suggested that the Stokes-Einstein equation was only valid only at infinite dilution [44], (i.e. where the diffusing species only interacts with solvent). He suggested that the bulk viscosity of a solution was not always the same as the viscosity sensed by diffusing molecule. This means that crowded solutions would exhibit different viscosities depending on the nature of the crowding agent and diffusing species. He defined viscosity as a function of the size of a crowding agent, its concentration, and a substance specific constant. Since then multiple studies have come out supporting this idea of microviscosities that differ from macroviscosity [45–47], to further complicate the matter, recent work has suggested that crowding may cause solutions to exhibit multiple detectable microviscosities [48]. 16 Chapter 1: Macromolecular crowding and its influence on the chemistry of the cell

1.3 Experimental investigations of crowding

1.3.1 Approaches to mimic crowded media in vitro

The effects of crowding can be mimicked in vitro by adding high concentrations of inert polymers to assays containing purified components. However, using such artificial crowding agents can be complicated, as these crowding molecules can sometimes interact in other ways with the process being examined, such as by binding weakly to one of the components. One approach to produce more accurate measurements would be to use highly concentrated extracts of cells, to try to maintain the cell contents in a more natural state. However, when using such extracts it is very difficult to study one process in isolation.

1.3.2 Type of crowders: advantages & disadvatages

To test theoretical prediction of the effects of crowding, it is necessary to create controlled crowded environments in vitro. An ideal crowder should

1. be soluble

2. form no attractive interactions with the protein of interest

3. not interfere with the spectroscopic techniques used to study the protein

Crowding with another protein may seem to be the most straightforward option since that would most closely represent the situation encountered in the cell. However, protein crowders are usually not soluble in sufﬁciently high concentrations and form numerous charge-charge interactions as most proteins have many charged residues distributed over their surface. It is therefore necessary to either screen these charges with either high salt concentrations or to just use low protein concentrations. Another important concern is that spectroscopic techniques used to probe target protein will be subject to interference from the protein crowder. Since the protein crowder is present at a much higher concentration, it may dominate the signal and complicate the analysis. An alternative option is to use synthetic polymers, to induce the effects of macromolecular crowding. Polymers that have been used for this purpose include polyethylene glycol (PEG), dextrans, Ficoll and polyvinylpyrrolidone (PVP). These crowding agents offer an advantage that they can be prepared in different sizes. PEG is a polymer of ethylene glycol, PVP- of N-vinylpyrrolidone, dextran - of glucose and Ficoll - of sucrose (Table 1.1). These crowding agents offer an advantage that they can be prepared in different sizes. 1.3. Experimental investigations of crowding 17

Table 1.1: Chemical structure of polymers typically used as crowders

PEG PVP

Dextran Ficoll

They are highly soluble (up to 400 mg/mL or more in water) and bear no charge at neutral pH. They do not have strong absorption above 210 nm nor are they fluorescent upon excitation at 280 nm. When studying excluded volume effects, it is desirable to avoid attractive interactions between the crowding agent and the protein of interest. There is evidence that PEG forms attractive interactions with proteins in addition to inducing volume exclusion [49–51]. PVP has not been used widely and the only group that had used it for protein stability studies found that it forms unwanted attractive interactions with the proteins [52]. Another important property of the crowding agents is their molecular shape. PEG and PVP are likely to be very flexible polymers [53]. In contrast, Ficoll has a more specific spherical shape. This is because Ficoll is highly branched copolymer of sucrose and epichlorohydrin, which gives it a relatively compact and often sphere-like structure [54–57]. However, DLS studies have shown that Ficoll 70kDa adopts a shape that is intermediate between a sphere and a random coil [58]. In another study, Ficoll was modelled as a spherocylinder with a radius of 1.4 nm [59]. Dextran is a polymer of D-glucose with a lower degree of branching than Ficoll that adopts a more elongated, flexible shape [57, 60, 61]. Hydrodynamic radius values for different dextrans have been determined by light scattering [61].

1.3.3 Reported effects of macromolecular crowding

Excluded volume effects on proteins due to macromolecular crowding had been investigated for some time, even before Minton proposed the term in 1981. For example, it was demonstrated that the addition of PEG and Ficoll promotes the formation of HIV [62] and bacteriophage 29 [63] capsids, which are large macromolecular assemblies. For in- 18 Chapter 1: Macromolecular crowding and its inﬂuence on the chemistry of the cell dividual proteins, the volume changes associated with folding or binding will be smaller than those for viral capsids, but are still predicted to be sufﬁciently large to give a macromolecular crowding effect. The following section discusses the experimentally observed effects of macromolecular crowding on phenomena such as association equilibria, enzymatic activity and the folding equilibria and structure of proteins.

1.4 Biochemical model reactions

1.4.1 Association equilibria

In the case of association equilibria, there are two parameters that can potentially change during the reaction: the volume and shape of the monomer and multimer. Depending on how these parameters change upon association, volume exclusion may stabilize the associated state. The advantage of using association equilibria to study crowding is that they have well-defined start and end states. Snoussi and Halle reported a 30-fold increase in the association equilibrium constant for the formation of bovine trypsin inhibitor de- camers based on NMR analyses [64]. Similarly, Diaz-Lopez et al. estimated a 10-fold increase in the equilibrium association constant for a RepA-DNA complex when using bovine serum albumin (BSA) as crowder [65]. In another study involving protein crowders (Ribonuclease A, RNase A and human serum albumin), Zorilla et al. used steady state and time-resolved fluorescence anisotropy to show that the self -association of apomyoglobin increased in the presence of RNase A, but was unaffected by human serum albumin [66]. The free energy change for the conversion of human co-chaperonin into a heptameric species increased by 14 kJ/mol in response to crowding with 300 mg/ml Ficoll 70. It was further shown that this was primarily caused by effects on the stability of the individual monomers and that effects on the monomer-monomer interfaces were comparatively unimportant [67]. In 2010, Jiao et al. measured the binding of catalase to Superoxide dismutase (SOD) using dextran and Ficoll 70 as crowders and found that the binding affinity increased by 3.6 kJ/mol in the presence of either crowder but concluded that the crowders’ volume exclusion effects were tempered by attractive interactions [68].

1.4.2 Enzymatic activity

When analysing enzymatic activity, it is important to understand how crowding affects the reaction mechanism and whether the reaction is diffusion- or transition state-controlled. Higher viscosity of crowded solutions will increase diffusion times, which will reduce the rate of diffusion-controlled reactions and so would reduce enzymatic activity [11]. Thus overcoming the effects of volume change due to crowding. This is especially relevant for reactions involving small substrates,where the changes in volume of free protein and substrate, substrate-protein complex and the free protein and products can be very 1.4. Biochemical model reactions 19 small. Indeed, Homchaudhuri et al. reported an increase in the catalytic rate for alkaline phosphatases in the presence of dextran and Ficoll [69]. Moran-Zorzano et al. also found that high concentrations of PEG increased the rate of the reaction catalyzed by AspP from E. coli. [70]. However, Derham and Harding observed a linear decrease in the activity of urease in the presence of increasing concentrations of dextran or PEG, although the use of protein crowders caused a non-linear increase [71]. Similar non- linear crowding effects on enzymatic activity have also been reported by Pozdnyakova and Wittung-Stafshede for multi-copper oxidase Fet3p. In this case, the addition of Fi- coll or dextran with a molecular weight of 20 kDa initially increased the enzyme’s Km and Kcat values, which peaked at crowder concentrations of ∼150 mg/mL [72]. The effects of crowding on enzyme kinetics have been reviewed by Vopel¨ and Makhadatzde, who reported that the addition of Ficoll 70 did not generally change the Michaelis constant or catalytic turnover number, but that some exceptions have been presented [73]. Overall, the effect of macromolecular crowding on enzyme activity remains dependent on the precise conditions of experiment, and more studies in this area are needed to make conclusive statements.

1.4.3 Conformational changes

Macromolecular crowding can also promote structural transformations. Most crowding theories predict a change in the relative free energies of the folded and unfolded states, assuming that the structures of the two states do not also change at the same time. The most obvious such change that might occur is that the expanded unfolded state may become compacted. Minton postulated a compaction of the unfolded state in the presence of crowding agents [64]. For adenylate kinase, Ittah et al. showed that adding dextran 40kDa caused the distance between two residues in the unfolded chain to decrease but observed no such effect on the folded state [74]. Two other studies also reported similar unfolded state compaction in the presence of crowders based on two different techniques and two different proteins (CRAB I [75] and ribosomal protein S16 [76]). A more striking and unpredicted result was the finding that crowding by dextran 70 or Ficoll 70kDa affected the folded structures of three proteins: apoflavodoxin, VlsE and phosphoglycerate kinase (PGK). Far-UV CD spectroscopic analysis indicated that crowding increased the secondary structure content of apoflavodoxin and VlsE in addition to affecting their equilibrium properties. These results were rationalized with the help of coarse-grained simulations [77], and are important because they suggest that the folded structure observed in vitro is not necessarily that adopted in vivo. A subsequent in vivo study by Dahr et al. provided some important support for this idea, showing that PGK also adopted a more compact tertiary structure in vivo than had been observed in vitro [78]. 20 Chapter 1: Macromolecular crowding and its influence on the chemistry of the cell

1.4.4 Protein folding

Finally, a number of studies have reported crowding effects on protein folding equilibria and kinetics. Majority of the reports of crowding effects on thermal or chemical protein unfolding reactions were based on studies using synthetic crowding agents. However, it is difficult to compare these studies directly because they generally used different types and concentrations of crowding agents. In most cases, the crowders increased the tested protein’s equilibrium stability and resistance to thermal or chemical denaturation. However, the magnitude of the increased resistance to thermal denaturation varied significantly from protein to protein. The midpoint of thermal denaturation increased by around 20 ◦C for the molten globular form of apomyoglobin in the presence of 270 mg/mL dextran 30 kDa [79] while that of DNase I rose by around 15 ◦C in 200 mg/mL PEG [80]. Much more modest changes have also been reported: the midpoint for the thermal denaturation of PGK increased by only ∼1.5 ◦C in 150 mg/mL Ficoll [78] while that of maltose binding protein increased by ∼1.0 ◦C in the presence of 300 mg/mL Ficoll [81]. Compared to equilibrium studies, the effects of macromolecular crowding on protein folding kinetics have received relatively little attention. The refolding rate constant of carbonic anhydrase increased in the presence of Ficoll 70 but the total amount of refolded protein decreased [82]. Similarly, crowding caused reduced lysozyme to exhibit a slightly increased refolding speed but a reduced level of correct refolding due to aggregation [83, 84]. The refolding rate constants of VlsE110, apoflavodoxin[85] and apocytochrome b562 [86] increased in the presence of crowders such as Ficoll, Dextran or PEG, but their unfolding rate constants were unaffected.

1.4.5 Other effects

Crowding also seems to affect the way in which bacteria adapt to large changes in concentration of osmotically active molecules in their environment [87]. Other processes that are influenced by crowding include: the regulation of metabolic pathways associated with signal transduction [88], the extraordinary stability of crystalline proteins in the lens of the eye [49]. One of the more dramatic influences of crowding is the stimulation of the rate and extent of formation of rod-like structures. Examples are amyloid fibres formation of alpha-synuclein, implicated in Parkinson’disease [87], microtubules [89], and large arrays of polymers of protein FtsZ, essential for bacterial division [59]. Another observation is that the efficiency of the bacterial chaperonins GroEL and GroES - which work together to help certain proteins to fold - is enhanced by crowding agents. This might be due to an enhancement of the association between GroEL and GroES; the latter caps the internal cavity of GroEL and prevents the escape of an encapsulated polypeptide [90]. And it might be because crowding reduces the probability that an encapsulated polypeptide can diffuse away from an uncapped GroEL before folding completely [52]. 1.5. Aim of the research presented in this thesis 21

1.4.6 Attractive interactions Volume exclusion is based only on repulsive steric interactions between crowders and protein. The possibility of attractive interactions between protein and crowders has been proposed and might explain the discrepancies between different experimental data sets [32, 88]. Since Ficoll and dextran are uncharged molecules, ionic interactions with proteins are not expected. Studies have indicated that unlike PEG [49, 50], Ficoll and dextrans do not form speciﬁc interactions with most proteins. However, this claim has recently been questioned [32, 87, 89]. One way of investigating the existence of an enthalpic component is to probe for a temperature dependence of the crowding effect according to [68]. The change in the free energy of unfolding due to crowding can be split into enthalpic and entropic contributions according to Equation 1.1: Kcrowded ∆∆GU = −RT ln (1.21) Kbu f f er This expression can be substituted to yield Equation 1.22:

K ∆∆H ∆∆S ln crowded = − U + U (1.22) Kbu f f er RT R According to Equation 1.22, enthalpic contributions due to attractive interactions that favor the unfolded state should cause ln(Kcrowded/Kbu f f er) to decrease with decreasing temperature. Jiao et al. reported that the crowding effects of Ficoll and PEG on the association between SOD and catalase depends on the temperature and concluded that an enthalpic component exists in that system [68]. However in a study on the interaction between bovine serum albumin (BSA) and Ficoll using equilibrium sedimentation, no temperature dependence of the observed changes in buoyant mass was observed and so it was concluded that only entropic effects were relevant in this case [59]. However, a temperature dependence was observed for protein crowders [90] .The Pielak group reported that the ∆∆HU values for CI2 and ubiquitin changed following the addition of crowders (PVP, Ficoll) and interpreted that as a consequence of non-speciﬁc attractive crowder-protein interactions [46, 52, 91].

1.5 Aim of the research presented in this thesis

Living cells are complex chemical reactors, characterized by small volumes, concentrated contents and the presence of internal membranes and compartments. Despite increasing number of studies on macromolecular crowding, how this complex environment impacts the biochemical reactions taking place within is largely unknown. In recent years it has become clear that crowding can considerably alter the reactivity of individual 22 References macromolecules, both qualitatively and quantitatively. The quantitative characterization of crowding effects on macromolecular reactions presents special challenges to the researchers who are involved in the study of these reactions in dilute solution. Here we propose that the influence of crowding on macromolecular reactivity in vivo may be best explored by means of bottom-up-approaches. Such an approach would ultimately lead to the construction of cytomimetic environments incorporating all the major components thought to be present in the cell. The ability to control and independently manipulate temperature, pH, salt, concentration of test species, structural elements, and compartments in this model system will provide a rigorous approach to the characterization of nonspecific interactions influencing behaviour of enzymatic reaction in cell-like environments.

References

[1] K. Luby-Phelps, “Cytoarchitecture and physical properties of cytoplasm: Volume, viscosity, diffusion, intracellular surface area,” Int. Rev. Cytology, vol. 192, no. 189, 1999. [2] A. Fulton, “How crowded is the cytoplasm,” Cell, vol. 30, pp. 345–347, 1982. [3] R. Ellis and A. Minton, “Cell biology - Join the crowd,” Nature, vol. 425, pp. 27–28, 2003. [4] O. Medalia, I. Weber, A. Frangakis, D. Nicastro, G. Gerisch, and W. Baumeister, “Macromolecular ar- chitecture in eukaryotic cells visualized by cryoelectron tomography,” Science, vol. 298, pp. 1209–1213, 2002. [5] A. Minton, “Excluded volume as a determinant of macromolecular structure and reactivity,” Biopolymers, vol. 20, pp. 2093–2120, 1981. [6] B. Alberts, A. Johnson, J. Lewis, M. Raff, K. Roberts, and P. Walter, Molecular Biology of the Cell. 5th edition ed., 2007. [7] S. Zimmerman and S. Trach, “Estimation of macromolecule concentrations and excluded volume effects for the cytoplasm of escherichia-coli,” J. Mol. Biol., vol. 222, pp. 599–620, 1991. [8] F. Lanni, A. Waggoner, and D. Taylor, “Structural organization of interphase 3t3-fibroblasts studied by total internal-reflection fluorescence microscopy,” J. Cell Biol., vol. 100, pp. 1091–1102, 1985. [9] W. Kuhn, “Uber die gestalt fadenformiger molekule in losungen,” Kolloid-Zeitschrift, vol. 68, pp. 2–15, 1934. [10] G. Wedler, “Lehrbuch der physikalischen chemie,” KGaA, Darmstadt: Wiley-VCH Verlag GmbH & Co, 5th edition ed., 2004. [11] H.-X. Zhou, G. Rivas, and A. P. Minton, “Macromolecular crowding and confinement: Biochemical, biophysical, and potential physiological consequences,” Ann. Rev. Biophys., vol. 37, pp. 375–397, 2008. [12] A. Minton, “The influence of macromolecular crowding and macromolecular confinement on biochemical reactions in physiological media,” J. Biol. Chem., vol. 276, no. 14, pp. 10577–10580, 2001. [13] A. Minton, “The effect of volume occupancy upon the thermodynamic activity of proteins - some biochemical consequences,” Mol. Cell. Biochem., vol. 55, pp. 471–483, 1983. [14] J. Wenner and V. Bloomfield, “Crowding effects on EcoRV kinetics and binding,” Biophys. J., vol. 77, pp. 3234–3241, 1999. [15] J. Batra, K. Xu, S. Qin, and H.-X. Zhou, “Effect of Macromolecular Crowding on Protein Binding Sta- bility: Modest Stabilization and Significant Biological Consequences,” Biophys. J., vol. 97, pp. 906–911, 2009. References 23

[16] A. Minton, “Molecular crowding: Analysis of effects of high concentrations of inert cosolutes on biochemical equilibria and rates in terms of volume exclusion,” Methods Enzymol., vol. 295, pp. 127–149, 1998. [17] W. McMillan and J. Mayer, “The statistical thermodynamics of multicomponent systems,” J. Chem. Phys., vol. 13, pp. 276–305, 1945. [18] H. Reiss, H. Frisch, and J. Lebowitz, “Statistical mechanics of rigid spheres,” J. Chem. Phys., vol. 31, pp. 369–380, 1959. [19] T. Boublik, “Statistical thermodynamics of convex molecule ﬂuids,” Mol. Phys., vol. 27, pp. 1415–1427, 1974. [20] R. Gibbons, “The scaled particle theory for particles of arbitrary shape,” Mol. Phys., vol. 17, pp. 81–86, 1969. [21] T. Boublik, “Hard-sphere equation of state,” J. Chem. Phys., vol. 53, pp. 471–&, 1970. [22] G. Jackson, J. Rowlinson, and F. Vanswol, “Computer-simulation of mixtures of hard-spheres,” J. Phys. Chem., vol. 91, pp. 4907–4912, 1987. [23] E. Spruijt and P. M. Biesheuvel, “Sedimentation dynamics and equilibrium proﬁles in multicomponent mixtures of colloidal particles,” . Phys.-Condes. Matter, vol. 26, pp. 075101–16, 2014. [24] M. Quillin and B. Matthews, “Accurate calculation of the density of proteins,” Acta Crystallogr. Sect. D-Biol. Crystallogr., vol. 56, pp. 791–794, 2000. [25] H. Fischer, I. Polikarpov, and A. Craievich, “Average protein density is a molecular-weight-dependent function,” Protein Sci., vol. 13, pp. 2825–2828, 2004. [26] S. Asakura and F. Oosawa, “Interaction between particles suspended in solutions of macromolecules,” J. Polym. Sci., vol. 33, pp. 183–192, 1958. [27] A. Vrij, “Polymers at Interfaces and the Interactions in Colloidal Dispersions,” Pure Appl.Chem., vol. 48, 1976. [28] J. S. Kim and I. Szleifer, “Depletion Effect on Polymers Induced by Small Depleting Spheres,” J. Phys. Chem. C, vol. 114, pp. 20864–20869, 2010. [29] D. Marenduzzo, C. Micheletti, and P. Cook, “Entropy-driven genome organization,” Biophys. J., vol. 90, pp. 3712–3721, 2006. [30] R. de Vries, Equation list, accompanying the course Colloid Science. Wageningen University, 2011. [31] A. H. Elcock, “Models of macromolecular crowding effects and the need for quantitative comparisons with experiment,” Curr. Opin. Struct. Biol., vol. 20, pp. 196–206, 2010. [32] A. P. Minton, “Quantitative assessment of the relative contributions of steric repulsion and chemical interactions to macromolecular crowding,” Biopolymers, vol. 99, pp. 239–244, 2013. [33] M. Smoluchowski, “Drei vorträge über diffusion, brownsche molekularbewegung und koagulation von kolloidteilchen,” Physik. Z., vol. 17, pp. 557–571, 585–599, 1916. [34] L. Mirny, “Biophysics: Cell commuters avoid delays,” Nature Phys., vol. 4, pp. 93–95, 2008. [35] J. A. Dix and A. S. Verkman, “Crowding effects on diffusion in solutions and cells,” Ann. Rev. Biophys., vol. 37, pp. 247–263, 2008. [36] M. Record, E. Courtenay, S. Cayley, and H. Guttman, “Biophysical compensation mechanisms buffering E-coli protein-nucleic acid interactions against changing environments,” Trends Biochem.Sci., vol. 23, pp. 190–194, 1998. [37] G. Rivas, F. Ferrone, and J. Herzfeld, “Life in a crowded world,” EMBO reports, vol. 5, pp. 23–7, 2004. [38] R. Phair and T. Misteli, “High mobility of proteins in the mammalian cell nucleus,” Nature, vol. 404, 24 References

pp. 604–609, 2000. [39] M. Weiss, M. Elsner, F. Kartberg, and T. Nilsson, “Anomalous subdiffusion is a measure for cytoplasmic crowding in living cells,” Biophys. J., vol. 87, pp. 3518–3524, 2004. [40] M. Elowitz, M. Surette, P. Wolf, J. Stock, and S. Leibler, “Protein mobility in the cytoplasm of Escherichia coli,” J. Bacteriol., vol. 181, pp. 197–203, 1999. [41] R. Swaminathan, C. Hoang, and A. Verkman, “Photobleaching recovery and anisotropy decay of green fluorescent protein GFP-S65T in solution and cells: Cytoplasmic viscosity probed by green fluorescent protein translational and rotational diffusion,” Biophys. J., vol. 72, pp. 1900–1907, 1997. [42] M. Dayel, E. Hom, and A. Verkman, “Diffusion of green fluorescent protein in the aqueous-phase lumen of endoplasmic reticulum,” Biophys. J., vol. 76, pp. 2843–2851, 1999. [43] A. Partikian, B. Olveczky, R. Swaminathan, Y. Li, and A. Verkman, “Rapid diffusion of green fluorescent protein in the mitochondrial matrix,” J. Cell Biol., vol. 140, pp. 821–829, 1998. [44] M. Mooney, “The viscosity of a concentrated suspension of spherical particles,” J. Colloid Sci., vol. 6, pp. 162–170, 1951. [45] P. Ross and A. Minton, “Hard quasi-spherical model for viscosity of hemoglobin solutions,” Biochem. Biophys. Res. Commun., vol. 76, pp. 971–976, 1977. [46] Y. Wang, L. A. Benton, V. Singh, and G. J. Pielak, “Disordered Protein Diffusion under Crowded Condi- tions,” J. Phys. Chem. Lett., vol. 3, pp. 2703–2706, 2012. [47] D. Lavalette, C. Tetreau, M. Tourbez, and Y. Blouquit, “Microscopic viscosity and rotational diffusion of proteins in a macromolecular environment,” Biophys. J., vol. 76, pp. 2744–2751, 1999. [48] A. B. Goins, H. Sanabria, and M. N. Waxham, “Macromolecular Crowding and Size Effects on Probe Microviscosity,” Biophys. J., vol. 95, pp. 5362–5373, 2008. [49] J. Bloustine, T. Virmani, G. Thurston, and S. Fraden, “Light scattering and phase behavior of lysozyme- poly(ethylene glycol) mixtures,” Phys. Rev. Lett., vol. 96, 2006. [50] G. Tubio, B. Nerli, and G. Pico, “Relationship between the protein surface hydrophobicity and its partitioning behaviour in aqueous two-phase systems of polyethyleneglycol-dextran,” J. Chromatogr. B, vol. 799, pp. 293–301, 2004. [51] D. Winzor and P. Wills, “Molecular crowding effects of linear polymers in protein solutions,” Biophys. Chem., vol. 119, pp. 186–195, 2006. [52] A. C. Miklos, M. Sarkar, Y. Wang, and G. J. Pieak, “Protein Crowding Tunes Protein Stability,” J. Am. Chem. Soc., vol. 133, pp. 7116–7120, 2011. [53] J. Armstrong, R. Wenby, H. Meiselman, and T. Fisher, “The hydrodynamic radii of macromolecules and their effect on red blood cell aggregation,” Biophys. J., vol. 87, pp. 4259–4270, 2004. [54] L. Stagg, S.-Q. Zhang, M. S. Cheung, and P. Wittung-Stafshede, “Molecular crowding enhances native structure and stability of alpha/beta protein flavodoxin,” Proc. Natl. Acad. Sci. U. S. A., vol. 104, pp. 18976–18981, 2007. [55] D. Homouz, L. Stagg, P. Wittung-Stafshede, and M. S. Cheung, “Macromolecular Crowding Modulates Folding Mechanism of alpha/beta Protein Apoflavodoxin,” Biophys. J., vol. 96, pp. 671–680, 2009. [56] D. Homouz, H. Sanabria, M. N. Waxham, and M. S. Cheung, “Modulation of Calmodulin Plasticity by the Effect of Macromolecular Crowding,” J. Mol. Biol., vol. 391, pp. 933–943, 2009. [57] M. Bohrer, G. Patterson, and P. Carroll, “Hindered diffusion of dextran and ficoll in microporous membranes,” Macromolecules, vol. 17, pp. 1170–1173, 1984. [58] D. Venturoli and B. Rippe, “Ficoll and dextran vs. globular proteins as probes for testing glomerular permselectivity: effects of molecular size, shape, charge, and deformability,” Am. J. Physiol.-Renal Phys- References 25

iol., vol. 288, pp. F605–F613, 2005. [59] A. A. Fodeke and A. P. Minton, “Quantitative Characterization of Polymer-Polymer, Protein-Protein, and Polymer-Protein Interaction via Tracer Sedimentation Equilibrium,” J. Phys. Chem. B, vol. 114, pp. 10876–10880, 2010. [60] M. Davidson and W. Deen, “Hindered diffusion of water-soluble macromolecules in membranes,” Macro- molecules, vol. 21, pp. 3474–3481, 1988. [61] K. Granath, “Liquid-solid extraction fractionation of hydrolyzed dextran,” Makromolekulare Chemie, vol. 28, pp. 1–9, 1958. [62] M. del Alamo, G. Rivas, and M. Mateu, “Effect of macromolecular crowding agents on human immun- odeficiency virus type 1 capsid protein assembly in vitro,” J. Virol., vol. 79, pp. 14271–14281, 2005. [63] C.-y. Fu, M. C. Morais, A. J. Battisti, M. G. Rossmann, and P. E. Prevelige, Jr., “Molecular dissection of O29 scaffolding protein function in an in vitro assembly system,” J. Mol. Biol., vol. 366, pp. 1161–1173, 2007. [64] K. Snoussi and B. Halle, “Protein self-association induced by macromolecular crowding: A quantitative analysis by magnetic relaxation dispersion,” Biophys. J., vol. 88, pp. 2855–2866, 2005. [65] T. Diaz-Lopez, C. Davila-Fajardo, F. Blaesing, M. P. Lillo, and R. Giraldo, “Early events in the binding of the pPS10 replication protein RepA to single iteron and operator DNA sequences,” J. Mol. Biol., vol. 364, pp. 909–920, 2006. [66] S. Zorrilla, G. Rivas, A. Acuna, and M. Lillo, “Protein self-association in crowded protein solutions: A time-resolved fluorescence polarization study,” Protein Sci., vol. 13, pp. 2960–2969, 2004. [67] X. Aguilar, C. F. Weise, T. Sparrman, M. Wolf-Watz, and P. Wittung-Stafshede, “Macromolecular Crowd- ing Extended to a Heptameric System: The Co-chaperonin Protein 10,” Biochemistry, vol. 50, pp. 3034– 3044, 2011. [68] M. Jiao, H.-T. Li, J. Chen, A. P. Minton, and Y. Liang, “Attractive Protein-Polymer Interactions Markedly Alter the Effect of Macromolecular Crowding on Protein Association Equilibria,” Biophys. J., vol. 99, pp. 914–923, 2010. [69] L. Homchaudhuri, N. Sarma, and R. Swaminathan, “Effect of crowding by dextrans and ficolls on the rate of alkaline phosphatase-catalyzed hydrolysis: A size-dependent investigation,” Biopolymers, vol. 83, pp. 477–486, 2006. [70] M. T. Moran-Zorzano, A. Miguel Viale, F. Jose Munoz, N. Alonso-Casajus, G. Gabriel Eydallin, B. Zu- gasti, E. Baroja-Fernandez, and J. Pozueta-Romero, “Escherichia coli AspP activity is enhanced by macromolecular crowding and by both glucose-1,6-bisphosphate and nucleotide-sugars,” FEBS Lett., vol. 581, pp. 1035–1040, 2007. [71] B. K. Derham and J. J. Harding, “The effect of the presence of globular proteins and elongated polymers on enzyme activity,” BBA-Proteins Proteomics, vol. 1764, pp. 1000–1006, 2006. [72] I. Pozdnyakova and P. Wittung-Stafshede, “Non-linear effects of macromolecular crowding on enzymatic activity of multi-copper oxidase,” BBA-Proteins Proteomics, vol. 1804, pp. 740–744, 2010. [73] T. Voepel and G. I. Makhatadze, “Enzyme Activity in the Crowded Milieu,” PLoS One, vol. 7, 2012. [74] V. Ittah, E. Kahana, D. Amir, and E. Haas, “Applications of time-resolved resonance energy transfer measurements in studies of the molecular crowding effect,” J. Mol. Recognit., vol. 17, pp. 448–455, 2004. [75] J. Hong and L. Gierasch, “Macromolecular crowding remodels the energy landscape of a protein by favoring a more compact unfolded state,” J. Am. Chem. Soc., vol. 132, pp. 10445–10452, 2010. [76] T. Mikaelsson, J. Aden, L. Johansson, and P. Wittung-Stafshede, “Direct observation of protein unfolded 26 References

state compaction in the presence of macromolecular crowding,” Biophys. J., vol. 104, pp. 694–704, 2013. [77] M. Perham, L. Stagg, and P. Wittung-Stafshede, “Macromolecular crowding increases structural content of folded proteins,” FEBS Lett., vol. 581, pp. 5065–5069, 2007. [78] A. Dhar, A. Samiotakis, S. Ebbinghaus, L. Nienhaus, D. Homouz, M. Gruebele, and M. Cheung, “Struc- ture, function, and folding of phosphoglycerate kinase are strongly perturbed by macromolecular crowding,” Proc. Natl. Acad. Sci. U. S. A., vol. 107, pp. 17586–17591, 2010. [79] P. McPhie, Y. Ni, and A. Minton, “Macromolecular crowding stabilizes the molten globule form of apomyoglobin with respect to both cold and heat unfolding,” J. Mol. Biol., vol. 361, pp. 7–10, 2006. [80] Y. Sasaki, D. Miyoshi, and N. Sugimoto, “Regulation of dna nucleases by molecular crowding,” Nucleic Acids Research, vol. 35, pp. 4086–4093, 2007. [81] S. Kulothungan, M. Das, M. Johnson, C. Ganesh, and R. Varadarajan, “Effect of crowding agents, signal peptide, and chaperone secb on the folding and aggregation of e. coli maltose binding protein,” Langmuir, vol. 25, pp. 6637–6648, 2009. [82] B. Monterroso and A. Minton, “Effect of high concentration of inert cosolutes on the refolding of an enzyme: carbonic anhydrase b in sucrose and ficoll 70,” J. Biol. Chem, vol. 282, pp. 33452–33458, 2007. [83] B. van den Berg, R. Wain, C. Dobson, and R. Ellis, “Macromolecular crowding perturbs protein refolding kinetics: implications for folding inside the cell,” EMBO J., vol. 19, pp. 3870–3875, 2000. [84] B. van den Berg, R. Ellis, and C. Dobson, “Effects of macromolecular crowding on protein folding and aggregation,” EMBO J., vol. 18, pp. 6927–6933. [85] D. Homouz, M. Perham, A. Samiotakis, M. Cheung, and P. Wittung-Stafshede, “Crowded, cell-like environment induces shape changes in aspherical protein,” Proc. Natl. Acad. Sci. U. S. A., vol. 105, pp. 11754– 11759, 2008. [86] L. Stagg, A. Christiansen, and P. Wittung-Stafshede, “Macromolecular crowding tunes folding landscape of parallel alpha/beta protein, apoflavodoxin,” J. Am. Chem. Soc., pp. 646–648, 2010. [87] X. Ai, Z. Zhou, Y. Bai, and W.-Y. Choy, “15n nmr spin relaxation dispersion study of the molecular crowding effects on protein folding under native conditions,” J. Am. Chem. Soc., vol. 128, pp. 3916– 3917, 2006. [88] H. Zhou, “Influence of crowded cellular environments on protein folding, binding, and oligomerization: biological consequences and potentials of atomistic modeling,” FEBS Lett., vol. 587, pp. 1053–1061, 2013. [89] M. Sarkar, C. Li, and G. Pielak, “Soft interactions and crowding,” Biophys. Rev., vol. 5, pp. 187–194, 2013. [90] A. Fodeke and A. Minton, “Quantitative characterization of temperature-independent and temperature- dependent protein-protein interactions in highly nonideal solutions,” J. Phys. Chem. B, vol. 115, pp. 11261–11268, 2011. [91] L. Benton, A. Smith, G. Young, and G. Pielak, “Unexpected effects of macromolecular crowding on protein stability,” Biochemistry, vol. 51, pp. 9773–9775, 2012. Chapter 2 A microfluidics platform for quantitative characterization of IVTT reaction in droplets

Abstract

Living cells offer poor versatility and flexibility to address specific research questions concerning the role of the crowded environment of the cell in the chemistry of life. De- spite substantial efforts, a suitable cell model to extrapolate findings to an in vivo situation is still lacking. The research presented here, will fully exploit microdroplets as a tool to quantitatively follow biochemical reactions at the cell-size scales. By carrying out experiments on a large population of droplets under identical conditions, we gain insights on how ensembles of enzymes respond to their physical environment. This chapter demonstrates how droplets can be used as artificial cells where the effects on biochemical reactions of parameter such as volume, concentration and number of molecules can be studied.

27 Chapter 2: A microﬂuidics platform for quantitative characterization of IVTT reaction 28 in droplets

2.1 Introduction

Living cells differ greatly in size and shape, depending on their function and the organism they belong to. Even simple cells are made-up of highly complex mixtures of components, like DNA, RNAs, ions, energy sources, proteins and organelles. Genetic networks ensure that this composition is constantly monitored and regulated in order for the cell to function. In this Chapter, we propose to study the influence of the intracellular macromolecular crowding on gene expression by means of artificial models. One can ask whether it is necessary to build an artificial cell. Although the structure and function of most cells is widely understood, working with a model including the minimum number of components required for a particular biological process to occur, would enable us to carry out quantitative measurements of isolated cellular mechanisms. As components can be added at will, we can understand how complexity is established while avoiding the difficulties associated with quantitative measurements in vivo that often limit the interpretation of results to a partial description of the process.

2.1.1 The course of gene expression in the cell Gene expression is used by all known life- eukaryotes (including multicellular organisms), prokaryotes (bacteria and archaea), and utilized by viruses- to generate the macromolecular machinery for life. The demand for the production of a certain type of protein produces a signal which activates the gene encoding for this protein. Once the gene is activated, the gene expression process required for protein synthesis proceeds in two steps, transcription and translation. The transcription of the genetic information coded in the DNA generates messenger RNA (mRNA) containing a genetic code that, aided by nucleotides, enzymes, and energy sources is translated to a speciﬁc sequence of amino acids [1] (Fig. 2.1).

Figure 2.1: Schematic representation of the course of gene expression through the process of transcription/translation 2.1. Introduction 29

2.1.2 Transcription and translation

Transcription is the first step of gene expression, in which a particular segment of DNA is copied into mRNA by the enzyme RNA polymerase. RNA polymerase binds to the specific binding site at the start of the gene, the promoter region, and starts reading it. For a gene on the DNA strand to be transcribed, first the two strands (coding strand and template or noncoding strand) must be pulled apart by the RNA polymerase, as nucleotides are only added to the 3’ end of the RNA molecule. The polymerase opens up the double helix by the formation of a transcription bubble. As the RNA polymerase moves through the DNA, the RNA elongates simultaneously by the addition of complementary nucleotides. When the backbone of RNA is formed, the RNA separates itself from the DNA. Transcription is followed by translation, which is catalyzed by ribosomes. The ribosome provides a binding site at which transfer RNAs (tRNA) carrying amino acids can recognize their specific codon (sequence of 3 bases) on the mRNA and use it as a reading pane for the recruitment of the amino acids. Subsequently, peptide bonds are formed between amino acids and a polypeptide chain grows until specific stop codons on the mRNA are reached. The ribosome then disassembles from the mRNA and the translation process terminates. The linear polypeptide chain, that represents the primary structure of proteins, has to fold into the correct three-dimensional configuration of the target protein. Due to its length and flexibility one can expect many possibilities of con- formations in which protein can fold. This number is severely limited by the non-covalent interactions (hydrogen bonds, ionic bonds, van der Waals attractions) between different parts of the chain. The conformation that is ultimately adopted is the one with the maximum strength of these interactions and minimal free energy. Protein folding takes place in the crowded environment of cells and often requires assistance of special helper proteins called “molecular chaperones” in order to avoid misfolding or aggregation of the proteins [2].

2.1.3 Cell-free gene expression

Cell-free gene expression systems were designed to produce proteins that are toxic to cells. Later these systems became widely used for in vitro transcription/translation studies. Cell-free gene expression systems, as a tool for molecular biology, were already developed in the 1950s. It was found then that encapsulation in a cell is not necessary for the gene expression reaction [3]. In vitro transcription and translation systems may consist of an extracted cell lysate from bacterial or eukaryotic cells, or they can be reconstituted from purified components necessary for transcription and translation. Usually in vitro protein synthesis systems, commercially available as kits, do not contain endogenous polymerases but are loaded with highly promoter specific and fast working polymerases mainly from T7 phage. Protein synthesis systems made up from cell ex- Chapter 2: A microfluidics platform for quantitative characterization of IVTT reaction 30 in droplets tracts may exhibit several disadvantages like short reaction lifetime due to depletion of energy resources and sub optimum protein yield due to the presence of residual nucleases and proteases. It was reasoned that a system reconstituted solely from purified components could circumvent these problems. Indeed, in 2001 Shimizu et al. presented such a system which they termed the PURE system ("Protein synthesis using recombinant elements") [4, 5]. Such a reconstituted system guaranteed that the outcome of the reaction would not be distorted by interference of unwanted or unknown components. A minimal cell-free system usually consists of about 100 individual components, which can be divided roughly into four groups: transcription, translation, energy resources and other components (see Table2.1).

Table 2.1: Components of a minimal cell-free system; concentrations of some key elements in a cell-free system and in E. coli are given for comparison [6]

E.coli cell-free system E.coli bacteria DNA (3 nM) 5 nM Transcription RNA Polymerases (100 nM) 500-800 nM NTP mix (1 - 2 mM) 1.3-7.0 mM Ribosomes (2.4 µM) 30 µM 20 amino acids (300 µM) 1.5 mM Translation Transl. factors tRNA mix Aminoacylation comp. Energy resources NTP mix (1 - 2 mM) 1.3 - 7.0mM Regeneration system Buffer Other components (Chaperons) (Glycosylation comp.) (Phosphorylation comp.)

Cell-free systems have a reduced complexity of composition and are much more diluted than living cells. Polymerase and ribosome concentrations are about one order of magnitude lower, and concentration of proteins is about two orders of magnitude lower. Whichever IVTT system is used, the macromolecular concentration usually ranges between 10 mg/ml and 20 mg/ml. An immediate question is: can such a system be used as a cell model to investigate effects of crowding? In addition to crowding, the fact that cells occupy a very small total volume prompted researchers to build cellular mimics that better represent the complexity of cellular life from the bottom up. To date there are a number of paths that could be taken to build compartmentalized cellular mimics, including encapsulation of gene expression machinery into water-in-oil emulsions, mi- 2.1. Introduction 31 crodroplets via microﬂuidics, and vesicles [7–12].

2.1.4 Cell free expression in vesicles

In 2003 Nomura et al. [7] showed that cell-sized liposomes of about 5 µm enabled the encapsulation of a gene-expression system and the production of a fluorescent protein (rsGFP). They showed that at the early stage of expression efficiency inside vesicles was significantly higher than in the outside solution (Fig. 2.2a). This suggested that the macromolecules involved in transcription/translation could be more concentrated inside than outside the vesicles and that the size of the enclosed environment might influence the rate of expression. Addition of proteinase K, a protein degrader, to the outside solution showed that the encapsulated protein was protected from the attack of the protease, as the fluorescence intensity inside the vesicles did not decrease as in the outside solution (Fig. 2.2b). Apparently the lipid bilayer acted as a barrier to proteins.

Figure 2.2: a) Scheme of the experimental setting (left) and relative ﬂuorescence emitted from inside and outside individual vesicles containing expressed rsGFP protein (right). Shown are ﬂuorescence over time, effects of addition of proteinase K and two negative control experiments in the absence of DNA and methionine. b) Fluorescence image of rsGFP in a vesicle and in external solution 60 minutes after addition of proteinase K to the external solution [7]

A few years later, Souza et al. [8] described an experiment to find the minimal size of liposomes that could still sustain gene expression. Higher expression efficiency inside liposomes compared to bulk was found. In accordance with the aforementioned example, this observation indicated that all molecular components necessary for functional gene expression should be inside the liposome upon encapsulation. However, this is statisti- cally unlikely at the used concentrations for small liposomes of 100 nm radius, since they cannot contain all of the more than 75 different ingredients required for IVTT. Assuming that the solutes in the liposomes are protected from interactions with the surrounding solution, Souza et al. proposed that, the concentration of solutes inside liposomes became a factor twenty times larger. Although the results hinted on the role of molecular crowding, no exact mechanism was proposed. In large liposomes, some surface enhancement effect Chapter 2: A microfluidics platform for quantitative characterization of IVTT reaction 32 in droplets could be present, but the experiments did not allow a quantitative analysis and the role of surface to volume ratio remained unclear. As was mentioned earlier, depletion of energy resources might be a limiting factor in the performance of cell-free expression systems for certain applications. Noireaux et al. [9] overcame this problem by exploiting selective permeability of the vesicle membrane. They demonstrated how an E. coli based cell-free expression system can be encapsulated in a phospholipid vesicle in order to build a cell-like bioreactor. They increased the capacity of the reactor by expressing the membrane pore protein alpha-hemolysin inside the vesicle, thus prolonging the protein expression up to 4 days by inducing the selective permeability for nutrients (feeding solution) (Fig. 2.3). Still some factors like oxygen limitation or vesicle bursting due to high osmotic pressure could disrupt efficient gene expression.

Figure 2.3: eGFP inside vesicles and kinetics of expression. (A) Fluorescence images of a single vesicle and a doublet with eGFP after 5 h. The E. coli extract (50 % extract, 50 % feeding solution) is encapsulated in the vesicles with the plasmid pIVEX2.3d-18L-eGFP (0.5 nM) surrounded by a feeding solution supplemented with 4 % extract. (Scale bar, 15 µm) (B) Kinetics of the expression of the eGFP. 2.1. Introduction 33

Saito et al. [10] developed a method to generate and track transcription/translation systems inside giant liposomes. They saw the formation of hundreds of stable, but polydisperse (1-100 µm range), liposomes that could effectively express fluorescent proteins after encapsulation of the IVTT. When they studied the possibility of monitoring the expression using GFP, it was noted that there was significant variability in the levels of expression between individual liposomes, and no significant correlation between gene expression levels and liposome sizes was found (Fig. 2.4).

Figure 2.4: Confocal ﬂuorescence images over time of liposomes encapsulating the GFP expression system. DNA template concentration is 200 µg/mL. t=0 corresponds to bringing the emulsion inside the observation chamber and the initiation of the reaction [10]

Clearly, a better experimental platform is needed. We believe that microdroplets via microﬂuidics can be a powerful platform to produce monodisperse cell-sized droplets that can act as individual reaction vessels [11, 12].

2.1.5 Cell free expression in microdroplets

Microdroplets via microfluidics is an emerging field of science and technology that deals with small amounts (nano- to femtoliters) of liquids in channels with dimensions between 10 and 100 micrometers [13, 14]. One of its advantages is miniaturizing components and scaling-down processes, therefore leading to reduced reagent consumption, in addition to expanding a few possible trials into a much larger number of high throughput screenings Chapter 2: A microfluidics platform for quantitative characterization of IVTT reaction 34 in droplets

[15]. Other advantages of using microfluidics include fast system response and processing of biological materials such as DNA, proteins, cells and bacteria. From a biological perspective, microfluidics can be especially relevant, since most biological processes involve small-scale fluidic transport. Given their size and structure, microdroplets represent an attractive tool for ‘cell-like’ experimentation.

2.1.5.1 Overview of microﬂuidics

A variety of designs has been developed for creating and manipulating droplets in recent years (Fig. 2.5) [15–22]. Flow-focusing devices allow production of the droplets at a very high frequency (30 Hz) [15]. For certain applications droplets can be fused via electrocoalescence or passively by designing a proper geometry of the device [16, 17]; droplets can be sorted [18], split [19] or broken into the emulsion via electric field [20]; efficient mixing inside of droplets is achieved in winding channels; and finally droplets can be stored for long time in different kinds of chambers, delay lines, or traps [15, 22].

Figure 2.5: An overview of commonly used droplet manipulation modules. a) droplet formation in flow focusing device. b) droplet formation from jetting in flow focusing device. c) delay line/storage area [15]. d) droplet fusion by electrocoalescence [16]. e) passive, geometry- mediated droplet fusion [17]. f) electrosorting of droplets [18]. g) geometry-mediated droplet splitting [19]. h) electric-field induced emulsion separation [20]. i) mixing inside droplets in winding channels [21]. j) droplet storage [22]. Reproduced from [15–17, 21]

To improve the monodispersity of the generated droplets and prevent unwanted coa- lescence during their manipulation surfactant is typically added to the continuous phase. A surfactant is an amphiphile composed of a hydrophilic head (water soluble) and a hydrophobic tail (non-water soluble) (Fig. 2.6). It can lower the interfacial surface suf- 2.1. Introduction 35

Figure 2.6: Schematic diagram of a micelle of oil in aqueous suspension, such as might occur in an emulsion of oil in water. In this example the oil soluble tails of the surfactant molecules project into the oil, while the water-soluble ends remain in contact with the water phase

ﬁciently, compared to the viscous force, to preferentially form droplets [23]. Although droplet formation does not require surfactants, without stabilizing agents droplets rapidly coalesce inside the devices. One has to be especially concerned when choosing the surfactant for biological experiments. For example, adsorption of protein to the water-oil interface can be an important issue [24, 25]. Holtze et al. [15] synthesized a series of non-ionic perﬂuoropolyether-derived (PEG-PFPE) surfactants designed to stabilize droplets at high formation rates (30 kHz) and provide biocompatibility with in vitro transcription and translation of plasmid DNA and cultivation of yeast and mammalian cells within droplets.

2.1.5.2 In vitro gene expression in droplets

Microdroplets via microfluidics offer a quantitative route to follow large numbers of IVTT reactions at very small scales. The high-throughput formation of monodisperse subpicolitre size microreactors for in vitro expression of green fluorescent protein (GFP) was first reported by Dittrich et al. [11]. This system has been extended by Courtois et al. who developed a reservoir device to store up to 106 subpicoliter droplets for several hours and followed the in vitro expression of GFP therein (Fig. 2.7) [12]. High efficiency of GFP expression from single DNA molecules was observed. The measurements were done by opening the reservoir at various time intervals and recording the fluorescence intensity of the droplets passing through the outlet channel. Stabilization Chapter 2: A microfluidics platform for quantitative characterization of IVTT reaction 36 in droplets

Figure 2.7: Reservoir device design consisting of an oil inlet (A) joined later on by two other inlet channels (B1 and B2) at a narrow junction (10 µm) followed by a wiggle channel to allow rapid mixing (C). Droplets then ﬂow into a reservoir (D) with four pillars for support and a narrow outlet channel (30 µm) for droplet exit and subsequent detection (E) of droplets up to 6 hours was achieved by adding 3 % Abil EM 90 surfactant. However, droplets adhere to each other as the oil was absorbed into the walls of the device. Experi- mentation with droplets containing GFP showed that after 6 hours ∼10 % of the droplets had shrunk and another 10 % had fused. Nevertheless, the platform described in [12] provided an exciting starting point for our studies in droplets. In the following sections we will introduce and characterize cell-free protein expression in microdroplets as an experimental platform to study the effects of crowding on transcription/translation reactions. We describe scoping experiments to test the efﬁciency of IVTT in droplets and to set the detection limit of our set-up and the design of devices for controllable long-term storage of droplets.

2.2 Experimental protocols

2.2.1 Materials

Poly(dimethylsiloxane) (PDMS, Sylgard 184) was purchased from Dow Corning (UK). Glass coverslips (microscope slides) were obtained from VWR international. Fluorinated oil (FC-40) was supplied by Sigma. The RTS 100 Escherichia coli HY kit was supplied by Roche Diagnostics GmbH (Mannheim, Germany). Plastic syringes (B. Braun, Ger- many) were used for storage of droplets.Surfactant (Triblock high Mw Krytox+2x Jef- famine) was synthesized by Dr. Sergey Semenov (see Fig. 2.8). DNA plasmid pRSet5d- GFPHIS (3501bp) was a gift from Dr. Kerstin Blank (Cluster for Molecular Chemistry, Radboud University Nijmegen). 2.2. Experimental protocols 37

Figure 2.8: Chemical structure of krytox–jeffamine ED-900–krytox surfactant

2.2.2 Laser-induced fluorescence (LIF) and microfluidic setup operation Fluorescence detection was performed using a laser-induced fluorescence setup including laser (488, Coherent) and a low-noise photomultiplier tube (PMT; H8249, Hamamatsu Photonics) (Fig. 2.9). The laser beam was expanded to give a spot size of the droplet size, ∼20 µm. Acquisition was controlled by home written data acquisition program in LabVIEW 8.2 (National Instruments). Multifunctional DAQ card (National Instruments) was used to regulate PMT sensitivity and measure PMT output. Data was filtered and further analyzed with Diadem software (National Instruments).

Figure 2.9: Schematic drawing of laser-induced ﬂuorescence setup

Solutions were pumped by syringe adjustable infusion pumps (Harvard Apparatus; PHD 2000 infusion) using glass syringes (SGE, Australia) connected to polyethylene tubing (Smiths, International). Temperature in the bilayer device was controlled via Bipolar Temperature Controller (CL-100). Droplets were stabilized with 2 % krytox–jeffamine ED-900–krytox surfactant dissolved in ﬂuorinated oil FC-40. The devices were mounted on the inverted microscope (Olympus IX71) equipped with a motorized stage (Prior; Optiscan II), connected to a laser from Coherent and a pho- Chapter 2: A microﬂuidics platform for quantitative characterization of IVTT reaction 38 in droplets tomultiplier tube (H8249) from Hamamatsu Photonics. Fluorescent images were taken with the sensitive EMCCD camera (iXon; Andor) using illumination from the mercury lamp or the laser (Fig. 2.10). Analysis of images was done by ImageJ (National Institutes of Health) or home-written MATLAB routine (version R2012a, MathWorks, Natick, MA, USA).

Figure 2.10: Schematic drawing of the operation setup

2.2.3 Device fabrication

Devices were fabricated using soft photolithography (Fig. 2.11) [13]. The first step in the manufacturing process of a microfluidic device is mold fabrication. Mold fabrication started with creation of a device design in an AutoCAD program. A high-resolution commercial image setter then printed this design onto a transparency (JD Photo Tools), that served as the photomask in contact photolithography to produce a negative relief in a SU-8 negative photoresist (SU8-2025, Micro Resist Technology) spin-coated on a round silicon wafer substrate of 50-mm diameter (Si-Mat Silicon Materials) (Fig. 2.11). Spin coating parameters were optimized to achieve the desired film thickness. Subsequently, the sample was soft-baked for 1 min at 65 ◦C, 3 min at 95 ◦C, and 1 min at 65 ◦C to evaporate the solvent and densify the film. Then, the samples were exposed to UV light (λ = 365 nm) in the mask aligner (Karl Suss MJB 3 UV 300/400) for 21 s through the photomask. After exposure, the sample was post-baked for a time that depended on the thickness of the photoresist (1 min at 65 ◦C, 2 min at 95 ◦C, and 1 min at 65 ◦C). The samples were rinsed with developer solution (mr-Dev600; Micro Resist Technology) to 2.2. Experimental protocols 39 remove the non–cross linked regions. The resulting height of the channels was 25 µm. Thus, the fabrication of the SU-8 mask for drop production and storage chamber was accomplished on a silicon wafer. In a similar way, SU-8 masters for the reservoir channels were obtained with a thicker photoresist SU8-100 to achieve 175-µm–deep channels. Depending on the design used in the mask, masters for droplet production and reservoir channels were obtained (Fig. 2.12).

Figure 2.11: Schematic overview of the microﬂuidic chip fabrication by photolithography

Bilayer PDMS microfluidic chips were produced following the protocol in Table2.2. For the production of single-layer flow focusing or trap devices a mixture of poly(dimethylsiloxane) and cross linker (ratio 10:1 w/w) was poured over the master, and the system was degassed and then cured for 3 h at 70 ◦C. The cured device was cut and peeled from the master, and holes for tubing were cut with a biopsy punch. After treatment with air plasma for 9 s, the device was sealed against a glass slide and baked for 30 min at 110 ◦C. The next step of the device fabrication was hydrophobic coating with 2,5 % solution of trichloro (1H,1H,2H,2H-perfluorooctyl)silane in FC-40. Chapter 2: A microfluidics platform for quantitative characterization of IVTT reaction 40 in droplets

Figure 2.12: AutoCAD design of the microﬂuidic device: (A) drop spot layer, (B) zoomed part of the droplet production site, (C) osmotic reservoir layer

Table 2.2: Soft photolithography protocol for fabrication of the bilayer device

◦ Pour a 5:1 ratio PDMS:cross linker on a SU-8 master with reservoir channels (5 mm thickness) Thick PDMS ◦ Bake at 65 ◦C for at least 30 min reservoir layer ◦ Peel off the PDMS ◦ Punch holes with 1.5 mm diameter biopsy punch ◦ Spin coat a 21:1 ratio PDMS:cross linker on a SU-8 master Thin PDMS with droplet production channel droplet layer (10 sec at 1000 rpm, 60 sec at 1250 rpm) ◦ Bake at 90 ◦C for at least 5 min ◦ Treat both of the layers with plasma cleaner (9 sec, 65 % power) ◦ Carefully align reservoir channels over the drop storage Alignment of channels of the thin PDMS layer PDMS layers and ◦ Bake at 90 ◦C for 2 hours plasma bonding ◦ Peel off the resulting bilayer PDMS slab carefully from the master and punch holes (1 mm diameter biopsy punch) ◦ Bond PDMS bilayer with glass coverslip (plasma at 65 % power, 9 sec)

2.2.4 Operation of chamber and bilayer devices The design of the bilayer chip is based on work by Shim et al. [26, 27] (Fig. 2.13, Table 2.2). The bilayer chip is a poly(dimethylsiloxane) (PDMS) device, which utilizes 2.3. Results and Discussion 41

Figure 2.13: Schematic drawing of the microfluidic device in which two droplet populations with identical contents coexist: non-shrunk, homogeneous droplets, and shrunk, phase- separated droplets hydrodynamic focusing to produce drops of aqueous solution inside a continuous oil stream [28]. The device is composed of two layers of microchannels. The top thin layer (40 µm) contains traps [29], where droplets are captured and stored. It is sealed over a permeable PDMS membrane (15 µm), through which small molecules (water and low molecular weight organic solvents) can diffuse. The bottom thick layer (5 mm) contains flow channels for flushing MilliQ or a saturated salt solution. Each droplet stored in the traps of the top layer is therefore in contact with the bottom layer through the PDMS membrane This design allows modifying the solvent conditions in the droplets by changing the ionic strength of the liquid flowing through the reservoir channels, shrinking them using the saturated salt solution and comparing with the control droplets (MilliQ) [26, 27]. Liquids that were flushed through the bottom layer of our devices were first flushed through a Peltier temperature control element (Warner Instruments), to assure accurate control over the temperature of the droplets (Peltier outflow temperature ±0.5 ◦C, trap layer temperature ±3 ◦C). See reference [30] for a similar device set-up.

2.3 Results and Discussion

In this section we demonstrate the possibilities of studying in vitro transcription/translation reaction in droplets. Special attention is given to the accuracy of detection of very low concentrations, possibility of long term kinetic measurements and control over concentration of IVTT. Chapter 2: A microﬂuidics platform for quantitative characterization of IVTT reaction 42 in droplets

2.3.1 In vitro gene expression from a single copy of DNA in droplets First we demonstrate the implementation of microfluidics in respect to the detection of green fluorescent protein (GFP) signal expressed from a single DNA copy per droplet. Monodisperse droplets of 20 µm diameter (4.19 pL) were produced in flow-focusing devices (Fig. 2.14). This device contains a main channel, which carries an oil phase, and two lateral channels which carry the “water” phase (IVTT or dye solution), both phases meeting at the narrow junction and breaking into droplets due to shear stress.

Figure 2.14: Flow-focusing device, diameter of the channels junction 20 µm

Each droplet passing through the laser beam (excitation 488 nm) gave a peak of fluorescence intensity related to the concentration of fluorescein inside the droplet. To determine the fluorescent background of the IVTT reaction, a control experiment was performed without plasmid inside the droplets. After calibrating our set-up, droplets containing IVTT ingredients and DNA encoding for GFP were formed in a flow-focusing device, collected into a plastic syringe and incubated for 4 hours at room temperature for the protein synthesis reaction to be completed. Afterwards droplets were re-injected back from the syringe into the fluidic device to measure the fluorescence intensity when passing through the outlet channel. In order to distinguish individual intensity peaks of droplets (Fig. 2.15), oil was co-injected, thus separating droplets. To investigate the possibility of expressing and detecting GFP from a single copy of template in the droplet reservoir, the plasmid solution was diluted to a final concentration 0.23 pM. At these concentrations, less than one molecule of the template per droplet is present (on average). The percentage of droplets expected to contain a single copy of e−µ (µx) the template was calculated by Poisson statistics: P(x; µ) = x! , where µ is an average number of successes within a given region and x is the actual number of successes that result from the experiment. According to this calculation, in droplets formed from a solution of 0.23 pM plasmid there should be 78 % empty droplets, 19 % with only one plasmid and 3 % with more than one copy of DNA. It was found that the percentage of droplets expressing GFP was in very close agreement (19 %) with the calculated percentage of droplets expected to contain plasmid, coinciding exactly with 0.23 pM plasmid (Fig. 2.16). This method allowed storage of large numbers of droplets for several hours during 2.3. Results and Discussion 43

Figure 2.15: On-chip detection of droplets containing IVTT reaction mixture (plasmid DNA concentration 1 pM). Optical readouts of ∼40 ms windows. Fluorescent peaks pronounced, each peak represents an individual droplet. Applied settings: PMT sensitivity-900V, laser power: 50 mW. Incubation time 6 h

Figure 2.16: Histograms of area under the peak obtained for droplets in experiments where GFP was expressed in vitro in droplets with µ=0.25 DNA molecule per droplet on average which no significant decrease of the droplet quality was observed. Most importantly, we were able to detect functional GFP expressed in droplets containing one copy of plasmid DNA and distinguish them from droplets without plasmid DNA. The availability of monoclonal droplets could be a major step towards future experiments aimed at determining the role of molecular crowding, stochasticity and noise in gene expression using microfluidics as a tool. However, this technique makes it impossible to measure kinetics of gene expression in time of individual droplets, as only the end-point of GFP expression was measured. Chapter 2: A microfluidics platform for quantitative characterization of IVTT reaction 44 in droplets

2.3.2 Kinetic measurements of IVTT in droplets

In order to follow the IVTT reaction in time, in droplets, we used chamber devices with traps (see section 2.2 Experimental protocols for operational details). It allowed us to track GFP expression in droplets trapped at speciﬁc positions in the device by continuously monitoring the development of ﬂuorescence in each droplet (Fig. 2.17) [31]. Each droplet has a very similar composition (except for the Poisson distribution of low copy number components) and the variation between individual droplets thus provides information about the levels of noise and variance observed in cell-free gene expression experiments.

Figure 2.17: Left: Droplet entrapment array, ﬂuorescent image of IVTT droplets, scale bar 20 µm; Right: GFP expression curves from [1nM]DNA in individual droplets (each symbol corresponds to an individual droplet)

However, during droplet storage, shrinking due to water diffusing into the PDMS matrix was observed (Fig. 2.18). Shrinkage would jeopardize quantitative measurements of droplet contents and therefore an alternative design device had to be employed.

Figure 2.18: Shrinkage of droplets ﬁlled with GFP over time in the trap device 2.4. Conclusion: From “bulk” to “cell-like” 45

2.3.3 Control over water contents of the droplets

We designed a bilayer fluidic device (see section 2.2 Experimental protocols) where droplets can be stored, while their volume is controlled via a reservoir layer, which also introduces temperature control. This allowed us to prevent unwanted reaction initiation before starting the fluorescence measurements, by incubating droplets at 4 ◦C. Our device is able to change the water content of the trapped droplets by controlled osmostic shrinking or swelling of the droplets, allowing us to study simultaneously monodisperse droplets with identical starting conditions as in Fig. 2.19 A but with different final volumes. When the reservoir channels are filled with air or a solution of salt with a concentration higher than that in the stored droplets, water permeates from the drops through the membrane (saturated NaCl induces droplet shrinking from 27 to 20 µm within 20 min), thus increasing salt concentration of the PDMS-impermeable solutes inside the drop. If, in contrast, pure water is introduced in the reservoir channels, then the chemical potential gradient is reversed and water permeates through the membrane from the reservoir into the drop, thus diluting PDMS-impermeable solutes inside the drop (see Fig. 2.19 B). Figure 2.19 C shows the evolution of droplet size as a function of time after changing conditions in the reservoir channel, giving an indication of the rates of evaporation and swelling of the droplets in these bilayer devices. Control over water content of droplets, and consequently on the concentration of IVTT in droplets, as well as temperature control made the bilayer devices the platform of choice for our further investigations on transcription/translation in droplets (see Chap- ter 4).

2.4 Conclusion: From “bulk” to “cell-like”

Some 50 years after Lederberg’s pivotal experiments with polydisperse droplets sprayed onto glass slides [32], microdroplets via microfluidics has become a rapidly developing research field emerging into a powerful toolbox to study complex chemical and biological reactions. Microfluidics offers a number of advantages over conventional methods including monodispersity of the droplets, possibility of following large numbers of reactions, using very small amounts of sample; control over size, concentration and external conditions, like temperature; sensitive detection down to a single-molecule level. It allows conducting fundamental studies of biochemical reactions in confined spaces and concentrated solutions, thus making microdroplets via microfluidics a promising platform to mimic critical features of the chemical environment in living cells. In this Chap- ter, we have established the key ingredients required for all further experiments in this thesis: first of all, we have established that cell-free gene expression using commercially available kits provides robust protein production inside microdroplets stabilized by a polymeric surfactant. Using laser-induced fluorescence, we can detect protein produc- Chapter 2: A microfluidics platform for quantitative characterization of IVTT reaction 46 in droplets

Figure 2.19: Microfluidic device robustness. (A) Left: Bright-field image of just-formed droplets. Right: Bright-field image taken 40 min after droplet production. The top half of the droplet traps in both images is covered with a PDMS layer containing reservoir channels filled with a concentrated salt solution. The bottom half of the droplet traps is covered with a PDMS layer containing reservoir channels filled with Milli-Q. (B) Osmotic shrinkage of droplets can be reversed by exchanging the saturated NaCl solution for Milli-Q

tion from a single copy of DNA in a droplet, but we loose information on the kinetics of the process. Secondly, we have developed a bilayer devices where the volume and temperature of IVTT-containing droplets can be controlled for prolonged periods of time. Two different droplet volumes can be produced on the same device by splitting the reservoir layer into two areas: one containing saturated NaCl solution, the other milliQ water. By ﬁlling these devices with monodisperse droplets at 4 ◦C we can prevent initiation of protein production until the droplet volumes are stabilized. These devices are the platform used in the next chapters of this thesis.

Acknowledgements Dr. Aigars Piruska and Agata Rakszewska are acknowledged for their experimental contribution to this chapter (section 2.3.1). Dr. Venkatachalam Chokkalingam is kindly acknowledged for his help in fabrication of the devices. References 47

References

[1] F. H. C. Crick, “On protein synthesis,” Symp. Soc. Exp. Biol., vol. 12, pp. 138–163, 1958. [2] B. Alberts, A. Johnson, J. Lewis, M. Ra, K. Roberts, and P. Walter, “On protein synthesis,” Molekular- biologie der Zelle, vol. 4 ed., 2004. [3] A. S. Spirin, “Cell-free protein synthesis,” In RTS application manual for cell-free protein synthesis, vol. 9, pp. 7–12, . [4] Y. Shimizu, I. Akio, Tomari, T. Suzuki, T. Yokogawa, K. Nishikawa, and T. Ueda, “Cell-free translation reconstituted with purified components,” Nature Biotechnology, vol. 19, pp. 751–755, 2001. [5] Y. Shimizu, T. Kanamori, and T. Ueda, “Protein synthesis by pure translation systems,” Methods, vol. 36, pp. 299–304, 2005. [6] “CyberCell Database: CCDB. E. Coli Statistics,” http://www.wishartlab.com/CCDB/. [7] S. M. Nomura, K. Tsumoto, T. Hamada, K. Akiyoshi, Y. Nakatani, and Y. K., “Gene expression within cell-sized lipid vesicles,” ChemBioChem, vol. 4, p. 1172, 2003. [8] T. P. de Souza, P. Stano, and P. L. Luisi, “The Minimal Size of Liposome-Based Model Cells Brings about a Remarkably Enhanced Entrapment and Protein Synthesis,” ChemBioChem, vol. 10, p. 1056, 2009. [9] V. Noireaux and A. Libchaber, “A vesicle bioreactor as a step toward an artificial cell assembly,” PNAS, vol. 101, p. 17669, 2004. [10] H. Saito, Y. Kato, M. Le Berre, A. Yamada, T. Inoue, K. Yosikawa, and D. Baigl, “Time-Resolved Tracking of a Minimum Gene Expression System Reconstituted in Giant Liposomes,” ChemBioChem, vol. 10, p. 1640, 2009. [11] P. Dittrich, K. Tachikawa, and A. Manz, “Micro total analysis systems. Latest advancements and trends,” Anal. Chem., vol. 78, no. 12, pp. 3887–3907, 2006. [12] F. Courtois, L. F. Olguin, G. Whyte, D. Bratton, W. T. S. Huck, C. Abell, and F. Hollfelder, “An integrated device for monitoring time-dependent in vitro expression from single genes in picolitre droplets,” ChemBioChem, vol. 9, no. 3, pp. 439–446, 2008. [13] G. M. Whitesides, “The origins and the future of microfluidics,” Nature, vol. 442, no. 7101, pp. 368–373, 2006. [14] D. Beebe, G. Mensing, and G. Walker, “Physics and applications of microfluidics in biology,” Annu. Rev. Biomed. Eng., vol. 4, pp. 261–286, 2002. [15] C. Holtze, A. C. Rowat, J. J. Agresti, J. B. Hutchison, F. E. Angile, C. H. J. Schmitz, S. Koster, H. Duan, K. J. Humphry, R. A. Scanga, J. S. Johnson, D. Pisignano, and D. A. Weitz, “Biocompatible surfactants for water-in-fluorocarbon emulsions,” Lab Chip, vol. 8, no. 10, pp. 1632–1639, 2008. [16] C. Priest, S. Herminghaus, and R. Seemann, “Controlled electrocoalescence in microfluidics: Targeting a single lamella,” Appl. Phys. Lett., vol. 89, no. 13, 2006. [17] N. Bremond, A. R. Thiam, and J. Bibette Phys. Rev. Lett., vol. 100, 2008. [18] K. Ahn, J. Agresti, H. Chong, M. Marquez, and D. Weitz, “Electrocoalescence of drops synchronized by size-dependent flow in microfluidic channels,” Appl. Phys. Lett., vol. 88, no. 26, 2006. [19] D. Link, S. Anna, D. Weitz, and H. Stone, “Geometrically mediated breakup of drops in microfluidic devices,” Phys. Rev. Lett., vol. 92, no. 5, p. 054503, 2004. [20] L. M. Fidalgo, G. Whyte, D. Bratton, C. F. Kaminski, C. Abell, and W. T. S. Huck Angewandte Chemie- International Edition, vol. 47, p. 2042, 2008. [21] H. Song, M. Bringer, J. Tice, C. Gerdts, and R. Ismagilov, “Experimental test of scaling of mixing by chaotic advection in droplets moving through microfluidic channels,” Appl. Phys. Lett., vol. 83, no. 22, 48 References

pp. 4664–4666, 2003. [22] W. W. Shi, J. H. Qin, N. N. Ye, and B. C. Lin, “A vesicle bioreactor as a step toward an artificial cell assembly,” Lab Chip, vol. 8, p. 1432, 2008. [23] H. Stone, “Dynamics of drop deformation and breakup in viscous fluids,” Annu. Rev. Fluid Mech., vol. 26, pp. 65–102, 1994. [24] L. Roach, H. Song, and R. Ismagilov, “Controlling nonspecific protein adsorption in a plug-based microfluidic system by controlling interfacial chemistry using fluorous-phase surfactants,” Anal. Chem., vol. 77, no. 3, pp. 785–796, 2005. [25] Y. Liu, S.-Y. Jung, and C. P. Collier, “Shear-Driven Redistribution of Surfactant Affects Enzyme Activity in Well-Mixed Femtoliter Droplets,” Anal. Chem., vol. 81, no. 12, pp. 4922–4928, 2009. [26] J.-u. Shim, G. Cristobal, D. R. Link, T. Thorsen, Y. Jia, K. Piattelli, and S. Fraden, “Control and measurement of the phase behavior of aqueous solutions using microfluidics,” J. Am. Chem. Soc., vol. 129, no. 28, pp. 8825–8835, 2007. [27] J.-u. Shim, L. F. Olguin, G. Whyte, D. Scott, A. Babtie, C. Abell, W. T. S. Huck, and F. Hollfelder, “Simultaneous Determination of Gene Expression and Enzymatic Activity in Individual Bacterial Cells in Microdroplet Compartments,” J. Am. Chem. Soc., vol. 131, no. 42, pp. 15251–15256, 2009. [28] T. Squires and S. Quake, “Microfluidics: Fluid physics at the nanoliter scale,” Rev. Mod. Phys., vol. 77, no. 3, pp. 977–1026, 2005. [29] C. H. J. Schmitz, A. C. Rowat, S. Koester, and D. A. Weitz, “Dropspots: a picoliter array in a microfluidic device,” Lab Chip, vol. 9, no. 1, pp. 44–49, 2009. [30] G. Velve Casquillas, C. Fu, M. Le Berre, J. Cramer, S. Meance, A. Plecis, D. Baigl, J.-J. Greffet, Y. Chen, M. Piel, and P. T. Tran, “Fast microfluidic temperature control for high resolution live cell imaging,” Lab Chip, vol. 11, no. 3, pp. 484–489, 2011. [31] A. Huebner, D. Bratton, G. Whyte, M. Yang, A. J. deMello, C. Abell, and F. Hollfelder, “Static microdroplet arrays: a microfluidic device for droplet trapping, incubation and release for enzymatic and cell-based assays,” Lab Chip, vol. 9, no. 5, pp. 692–698, 2009. [32] J. Lederberg, “A simple method for isolating individual microbes,” J. Bacteriol., vol. 68, no. 2, pp. 258– 259, 1954. Chapter 3 Cell lysate coacervates - possible models of crowding in cells

3.1 Introduction

3.1.1 Spatial organization of the cell

The organization of basic biochemical reactions into compartments is a major hallmark of a living cell [1]. Compartmentalization allows cells to control local concentrations of biomolecules. A compartment separates self from non-self, creates a linkage between genotype and phenotype, partitions solutes and macromolecules, and optimizes conditions for biochemical reactions separated from a reservoir. The most natural way for the cell to compartmentalize components is by the use of membranes. Membranes are also used inside the cell to create compartments, such as organelles. This allows each organelle to carry out its own function, without mixing its contents with the rest of the cell. Nonetheless, subcellular compartments may exist without physical boundaries of a membrane. Recent work has identiﬁed liquid phase transitions in vivo in the formation of intracellular non-membrane-bound compartments exhibiting liquid-like properties, slowed down diffusion, and strongly interacting macromolecular components [2, 3]. Well-studied examples are the intracellular localization of DNA or RNA and proteins in Cajal bodies, P granules, and nucleoli [4–6], which can contain over 100 components. Protein-RNA bodies such as Cajal bodies in the nucleus (implicated in RNA metabolism) or nuclear promyelocytic leukemia (PML) bodies are formed under stress conditions in certain cells. Recent studies on P granules and nucleoli suggest that protein- RNA com-

The work in this Chapter was published in Proc. Nat. Acad. Sci. USA 2013, 110, 11692-11697

49 50 Chapter 3: Cell lysate coacervates - possible models of crowding in cells plexes are liquids that form by liquid phase transitions from cytoplasm. P granules are protein-RNA complexes that are involved in germline formation in the nematode C. elegans. P granules have been shown to exhibit liquid-like behavior; that is, they form ﬂuid droplets [7], suggesting that they arise through liquid-liquid demixing from the cytoplasm (see Fig. 3.1).

Figure 3.1: Phase transitions in the membrane and cytoplasm. Membrane domains (dark regions) are shown in the plasma membrane of a rat basophilic leukemia cell. (B) P granules (labeled in green) in a C. elegans one-cell-stage embryo are imaged by ﬂuorescence microscopy, as described in [6]. Liquid P granules consist of proteins and RNAs. (C) A single P granule in a C. elegans embryo is imaged by stimulated emission depletion microscopy, as described in [7] Images were reproduced from [7]

Nucleoli, which are sites of ribosome synthesis, were also shown to behave like liquid droplets of protein-RNA complexes, exhibiting viscous-like ﬂuid dynamics [2]. A further discovery in the structure and function of ribonucleoproteins (RNPs) [3, 8, 9] demonstrated that in mouse brain and human cell extracts, proteins with low-complexity sequence domains (regions with low amino acid diversity) separate into a different phase together with RNA by liquid-liquid demixing. The idea of liquid-like states that either separate from the cytosol or occur in cell membranes is a powerful way to think about cellular compartments [10].

3.1.2 Compartmentalization induced by phase separation

The idea, that phase separation may be important for intracellular organization dates back many years, and has cycled in and out of favor [11, 12]. As the importance of macromolecular crowding in cells became clear, the interest in biological implications of coexisting aqueous phases has grown again. In mixtures of neutral, random coil-type polymers, phase separation typically occurs at concentrations greater than a few weight percent of each species [13–15]. These concentrations are relatively low because every 3.1. Introduction 51 polymer segment contributes to the overall interaction between two unstructured polymers, maximizing the effect of chemical incompatibility and leading to phase separation. Because proteins and nucleic acids are more structured and compact, higher concentrations are typically required to induce phase separation of the type observed in incompatible polymer mixtures. Protein-protein two-phase systems generally do not form until the concentration of each reaches 7-10 % [16], concentrations seldom achieved by single macromolecular species in cytoplasm. Single proteins, therefore, would not be expected to be present in high enough concentration to form phases if separation occurred as it does in mixtures of two puriﬁed proteins. The total concentration of proteins in cytoplasm of E. coli is in the range 17 to 35 % w/w, the nucleic acid concentration is about 7.5 % w/v and the total macrosolute concentration 27.5 % w/v [17]. In the high concentration environment of the cytoplasm another factor comes into play. In the presence of macromolecular crowding, phase separation is expected to occur at much lower concentrations of the species involved than is predicted from phase diagrams determined in the absence of crowding [18]. Hence, any two incompatible macromolecular species or two species which have a net attraction for each other, could, in principle, drive phase separation and produce small, localized phases of distinct composition bounded by an interface with a characteristic interfacial tension. Because of the large number of protein species present, multiple phases could well occur, each localized by interfacial tension and/or nearby solid bounding surfaces with which it is in contact. However, our understanding of the effect of spatial heterogeneity on chemistry of the cell has been limited by the difﬁculty to experimentally reproduce in vitro the crowded and compartmentalized nature of the cellular environment. Examples in which compartmentalization and high local concentrations are obtained concurrently, include DNA brushes [19], aqueous two-phase systems [20], and liquid coacervates [21].

3.1.3 Coacervation as a route to crowding

Coacervation is usually defined as the spontaneous formation of a dense liquid phase from a macromolecular solution of poor solvent affinity. In ‘coacervation’, the loss of sol- vation arises from the interaction of complementary macromolecular species. Following the pioneering work of Bungenberg-de Jong, coacervates are classified into simple and complex coacervates, depending on the process that leads to coacervation [22]. A simple coacervate is usually caused by partial miscibility, and addition of salt normally promotes coacervation. The properties of simple coacervates were studied most thoroughly in con- nection with their preparation from aqueous gelatin solutions [23–25]. Both layers of a simple coacervate are rich in colloids, but each layer contains mainly one of them. When a neutral salt such as sodium chloride or sulfate, which removes water from the gelatin molecules, is added to such solutions, the latter separate into two layers, and a coacervate is formed (on heating to +50 ◦C). The amount of gelatin in both the coacervate layer and 52 Chapter 3: Cell lysate coacervates - possible models of crowding in cells the equilibrium liquid may change and depends on the total gelatin content of the solution from which the coacervate was formed. On the other hand, in complex coacervation, two oppositely charged macromolecules, such as proteins and polyelectrolytes undergo coacervation through associative interactions. Both oppositely charged macromolecules ac- cumulate in the liquid coacervate phase. The other liquid phase, the supernatant, contains mainly solvent and is in equilibrium with the coacervate phase [22]. These two liquid phases are immiscible and hence, are incompatible. Coacervation may occur in polymer solutions containing as little as 0.01 weight percent of polymer, whereas the polymer concentration in the drops of coacervate that form can be as high as 30-40 weight percent. For this reason, coacervation is used as a means of concentrating and fractionating native and denatured biopolymers (in particular, water-soluble proteins) and synthetic polymers [25]. This makes coacervates an attractive mimic of macromolecular crowding, provided that we understand the coacervate structure and the transport of macromolecules inside this phase.

3.2 Results and Discussion

Phase separation or coacervation occurs in a wide range of polymer and protein solutions, often triggered by changes in temperature or salt concentration, or by the addition of coacervating agents [13]. The coacervate droplets that are formed in such systems present macromolecularly crowded, aqueous, physical compartments, 1–100 µm in diameter [26]. We took advantage of microfluidics technology to form monodisperse picoliter water-in-oil droplets, which allow for reactions to be studied systematically under precisely controlled conditions (see Chapter 2) Among the most important complex sets of reactions in the cell is protein synthesis (the result of transcription and translation), a process that functions in vitro despite almost two orders of magnitude lower total protein concentration than found in vivo. We and others have previously shown how droplet- based microfluidic devices can be used to study the kinetics of in vitro expression of a reporter protein using commercial in vitro transcription and translation kits [27–29]. The volume of droplets can be controlled via osmotic transport of water to and from reservoir channels filled with concentrated salt solutions [30, 31] and separated from the droplet traps [31] by a thin (15-µm) polydimethylsiloxane (PDMS) membrane (Fig. 2.14 Chap- ter 2). Here, we have adapted this approach to raise the salt concentration and concentrate the contents in the droplets, and thus induce coacervation, via controlled withdrawal of water from water-in-oil droplets that are trapped inside microfluidic channels made out of PDMS. This approach allows us to study monodisperse droplets (27-µm diameter, 10.3 pL) with identical starting compositions (E. coli cell-free expression kit and plasmid DNA for GFP production) but with different final volumes simultaneously, as droplets in osmotic contact with the saturated NaCl reservoir shrunk from 27 to 20 µm within 20 3.2. Results and Discussion 53 min. When following the production of GFP using fluorescence microscopy, we find that in the shrunk droplets a second, much smaller (∼3 µm in diameter), highly fluorescent, liquid droplet appeared, before significant fluorescence due to GFP production was observed in the non-shrinking droplets (Fig. 3.2). In this Chapter, we investigate the origin and properties of this coacervate droplet.

Figure 3.2: (A and B) Optical microscopy images (Upper) and ﬂuorescence images (Lower) of non-shrunk, homogeneous droplets (A) and shrunk, phase-separated droplets (B) in droplets traps.(C) Zoomed optical microscopy (Upper) and false-color confocal microscopy (Lower) image of a phase-separated droplet, showing that the coacervate is homogeneous on length scales down to the resolution of the microscope. (All scale bars: 20 µm)

3.2.1 Physical properties of cell lysate coacervates In home-made lysate [32], we did not observe the emergence of two phases, unless small amounts (∼2 wt%) of PEG 8000 were added, similar to the presence of PEG in commercial cell-free expression kits (Fig. 3.3A). PEG is known to form aqueous two-phase systems in the presence of certain inorganic salts [33]. Reaction mixtures of the home made in vitro transcription translation kit were composed of one third cell lysate and two thirds reaction buffer. Both were prepared with slight modiﬁcations according to Shin and Noireaux [32]. 54 Chapter 3: Cell lysate coacervates - possible models of crowding in cells

Figure 3.3: Phase separation in buffer droplets and cell-free expression mixtures. (A) In home-made lysate, we did not observe the emergence of two phases, unless small amounts of PEG 8000 were added, similar to the presence of 2 wt% PEG 8000 in commercial cell-free expression kits. All scale bars represent 20 µm. (B) Phase separation of cell free expression kit containing lysate and rGFP. (C, D) Fluorescence microscopy of droplets containing HEPES buffer (100 mM, pH 8.0), 300 mM potassium glutamate, 30 mM magnesium glutamate, 40 g/L PEG 8000 and 0.3 g/L α, ω-di-fluorescein PEG 8000 at t = 0.00 h (C) and t = 2.45 h (D). Phase separation of the solution after droplet shrinkage occurred to form a PEG-rich fluorescent phase and a non-fluorescent phase. All scale bars represent 20 µm

To elucidate the role of PEG in the coacervation process, we covalently labeled all proteins in commercial cell lysate with DyLight 550 and added 0.1 wt% additional fluorescein-labeled PEG 8000. Upon shrinking the droplets, we first observed the formation of small PEG-rich droplets, which subsequently merged to form a single large coacervate (Fig. 3.4A-D). The (fluorescently labeled) cell lysate then accumulated in the liquid coacervate, as demonstrated by the rapid increase of DyLight 550 fluorescence intensity in the coacervate. Upon further incubation, more and more macromolecules partitioned into the coacervate and eventually 75 % of the protein content of the lysate and 78 % of the PEG was compartmentalized in the coacervate, corresponding to a total macromolecule concentration of 375 g/L, of which 290 g/L is PEG 8000 and 85 g/L are lysate proteins. In contrast, the total macromolecule concentrations of the dilute phase is 45 g/L, of which 33 3.2. Results and Discussion 55

Figure 3.4: Coacervates are cell-like compartments. (A and B) Process of coacervation in shrinking droplets showing the concentrations of PEG (A) and DyLight550-stained cell lysate (B) by false-color ﬂuorescence microscopy. (Scale bars: 20 µm.) The labels indicate time (h:mm). (C and D) Zoomed images of the nucleation the coacervate for PEG (C) and cell lysate (D). (Scale bars: 10 µm.) Images were processed by adjusting the brightness/contrast in Photoshop CS 5.1. False color images were generated by applying a linear gradient map in Photoshop CS 5.1

g/L is PEG 8000 and 12 g/L are lysate proteins. In a complementary experiment in which we spiked the same, commercial, non-labeled cell lysate with recombinant GFP (rGFP) [200nM], we observed that 81 % of the rGFP molecules partitioned into the coacervate, which shows that partitioning is not driven by the fluorescent labeling (Fig. 3.3 B). Inter- estingly, the coacervation of fluorescently labeled PEG 8000 in buffer alone, took place at an internal salt concentration around 1.9 M, leading to coacervate droplets that contain ∼180 g/L PEG (Fig. 3.3C and D). The coacervate and "dilute" phase are osmotically balanced, and as the coacervate contains most of the macromolecules (PEG and proteins combined) we expect the dilute phase to be enriched in salts, in agreement with classical examples of liquid-liquid phase separation in PEG-salt mixtures [34]. Measurements of the concentrations of various salts using inductively coupled plasma-optical emission spectrometry confirms that salt concentrations in the coacervate are a little lower (∼1.06 M for K+, Na+, and Mg2+ combined, compared with 1.22 M in the dilute phase), which means that 77 % of the salts remain in the dilute phase. Table 3.1 summarizes the resulting concentrations of K, Na, Mg, Zn, P and S in the coacervates, the dilute phase and in the cell-free expression kit prior to phase separation. 56 Chapter 3: Cell lysate coacervates - possible models of crowding in cells

Table 3.1: ICP-OES analysis of the cell-free expression kit prior to shrinkage and phase separation, and the coacervates and the dilute phase after phase separation. The average weight of samples before phase separation was 51.7 ± 1.1 mg and after phase separation 18.7 ± 3.3 mg. Note (1): the Zn signal of all samples was below the detection limit

Element Concentration prior Concentration in Concentration in to shrinking (mM) coacervate (mM) dilute phase (mM) K 370 957±136 1071±172 Na 36 84±15 94±34 Mg 20 22±20 62±20 Zn(1) 0.01 0.03±0.01 0.03±0.03 P 55 111±41 132±54 S 94 258±33 283±35 K, Na, Mg, Zn 426 1063±171 1227±226 combined

We found that a small amount of DNA adsorbed to the oil-water interface (Fig. 3.5) and, somewhat unexpectedly, no strong preference of the DNA for either phase [54 % vs. 46 % in the dilute and coacervate phases, respectively (Fig. 3.5C)].

Figure 3.5: Partitioning of covalently labeled plasmid DNA. (A and B) Fluorescence micrographs of ﬂuorescein-labeled plasmid taken at high gain to visualize the adsorption of a fraction of the plasmids to the oil-water interface. (C) Fluorescence micrograph of ﬂuorescein- labeled plasmid partitioning in droplets. The image is taken 40 min after initial phase separation. Image was processed after analysis by adjusting the brightness/contrast and levels in Photoshop CS 5.1 (All scale bars: 20 µm) 3.2. Results and Discussion 57

The characterization of the coacervates reveals that combination of E. coli lysate and PEG has a remarkable effect on the distribution of the biologically active components. This system presents a unique example of a salt-induced phase separation in a complex mixture where most components partition in a single crowded coacervate compartment. Such a process could also drive the formation of nucleoli and other complex non-membrane-bound intracellular compartments, where coacervation is driven by interaction between multivalent proteins, followed by accumulation of other classes of proteins via selective interactions or physical properties [35]. We ﬁnd that the phase separation of droplets from the commercial gene expression kit is triggered at an internal salt concentration in the range of 1.0 M, with little dependence on the temperature at which the controlled evaporation is carried out (Fig. 3.6).

Figure 3.6: Phase diagram of cell-free expression kit in droplets as a function of ionic strength and temperature. The open symbols represent single-phase droplets, and the closed symbols represent phase-separated droplets

Coacervate droplets formed from the kit have an 8- to 10-fold smaller volume than the initial droplets (Table 3.2), corresponding to a total macromolecule concentration of ∼200 g/L, of which ∼155 g/L is PEG 8000 and ∼45 g/L are lysate proteins, assuming the partition coefficients determined in the above-mentioned experiment using fluorescently labeled PEG and proteins. The difference between these values and those for coacervates containing fluorescently labeled lysate proteins and PEG is likely due to the presence of dye leading to the formation of smaller coacervates (Table 3.2). There is a clear correlation between coacervate volume and amount of lysate in the initial droplet (Fig. 3.7). 58 Chapter 3: Cell lysate coacervates - possible models of crowding in cells 80 50 70 50 50 60 180 separation, g/L (PEG) at phase 1.9 1.9 1.2 1.7 1.2 1.1 1.4 Ionic strength at phase separation, M c = 9 V C 8 9 / 20 11 13 12 i ≥ K V s = V S / i 4.6 3.7 2.6 3.7 2.5 2.4 2.9 K V ∗ 0.6 0.2 0.2 0.5 0.45 0.45 0.39 Because the shape of the coacervates in this experiment (crescent moon) ∗ ± ± ± ± ± ± ± 12.4 13.1 13.0 11.5 Coacervate 10.00 10.50 diameter, µm 14.40 14.4 17.3 20.8 16.5 19.5 18.5 20.55 diameter, µm Shrunk droplet 0.2 0.4 0.2 0.4 1.2 0.7 0.6 ± ± ± ± ± ± ± Initial 24.0 27.7 28.5 25.5 26.5 27.5 26.5 diameter, µm Diameter measurements and volume ratios of initial droplets and coacervates in various experiments. Internal total salt and PEG Sample IVTT kit, IVTT kit, IVTT kit, IVTT kit, IVTT kit, and lysate composition PEG 8000 in buffer droplets 80 pM plasmid 60 pM plasmid 40 pM plasmid made an accurate determination of thethe coacervate volume volume of impossible, the we shrunk take droplet. the We coacervates note to that be the hemispheres volume with of a these volume coacervates equal is to not half used in any further calculation Table 3.2: concentrations at the moment ofat coacervate the formation moment were of calculated coacervatefrom from formation. the the volume Internal volume ratio total ratio between salt initial between andstrength droplet initial of PEG (Vi) droplet the and concentrations and shrunk cell-free at expression droplet shrunken the at kit, droplet moment the as of moment measured coacervate of by formation coacervate ICP-OES. were formation calculated (Vs) and the known initial ionic 160 pM plasmid 120 pM plasmid Fluorescent PEG 3.2. Results and Discussion 59

Figure 3.7: The volume of the coacervate depends linearly on the volume of the original droplet and scales with the concentration of the cell-free expression kit. Different datasets correspond to different dilutions (1×, 2×, 5×) of the kit before droplet formation. The solid lines are linear ﬁts of the data

We performed series of ﬂuorescence recovery after photobleaching experiments probing the diffusion of GFP added to the coacervates in order to conﬁrm that the coacervate was a liquid phase.

Fluorescence recovery after photobleaching experiments probing the diffusion of two variants of GFP [enhanced GFP (eGFP) and recombinant (rGFP)] added to the coacervates confirm that the coacervate is a liquid phase with a density and viscosity strongly resembling the environment within living cells (Fig. 3.8). We find a diffusion coefficient of 2.9±0.7 µm2·s−1 for eGFP (1.4±0.5 µm2·s−1 for rGFP) compared with 3.6±0.7-7.7±2.5 µm2·s−1 (depending on the levels of expression) in E. coli [36]. Before phase separation, the diffusion coefficient in the dilute lysate is around 77±28 µm2·s−1 (59±24 µm2·s−1 for rGFP) compared with 87 µm2·s−1 in pure water. 60 Chapter 3: Cell lysate coacervates - possible models of crowding in cells

Figure 3.8: Fluorescence recovery of eGFP in single-phase droplets and coacervates after photobleaching. The solid lines are ﬁts of the recovery curves to a 1D diffusion problem (see Materials and Methods section 3.3.8 for derivation of the equation). Small images are false- color confocal microscopy images of coacervates, taken at times as indicated by the labels (in seconds)

3.3 Materials and methods

We used RTS 100 Escherichia coli HY kit (5PRIME). Alpha-omega-di-fluorescein PEG 8000 (150 mg/mL) (Chemicell) was used to locate presence in the phases. DyLight 550 staining kit (Thermo Fisher scientific) was used to stain protein in E. coli cell lysate. Recombinant enhanced GFP (eGFP) (1 mg/mL) was purchased from Cell Biolabs. Re- combinant GFP (rGFP) (1 mg/mL) was purchased from Roche Applied Science. Sylgard 184 silicone elastomer kit polydimethylsiloxane (PDMS) (Dow Corning) was used for 3.3. Materials and methods 61 the microfluidic device fabrication. Fluorinert FC-40 oil (Sigma-Aldrich) was used as a continuous phase in the flow. GFP-His 9 plasmid pRSET5d-GFPHis) was a gift from Kerstin Blank (Radboud University Nijmegen).

3.3.1 Phase separation of cell-free expression kit containing ﬂuorescent PEG and lysate

Reaction mixture consisted of 10 µL reaction mix, 5 µL reconstitution buffer, 12 µL amino acids, 1 µL methionine, 12 µL DyLight 550 stained E. coli lysate, 0.3 µL of 150 mg/mL α,ω-di-fluorescein PEG 8000, 9.7 µL Milli-Q. Commercial lysate was stained using DyLight 550 NHS ester kit following the manufacturer’s protocol. The final sample volume was 50 µL with a total PEG concentration of 20.9 g/L. Droplets were generated in a double layer device (oil flow: 120 µL/h, sample flow: 30 µL/h, 6 M NaCl solution: 100 µL/h) and observed for 2 hours by fluorescence microscopy.

3.3.2 Phase separation of PEG in the absence of cell-free expression kit

Reaction mixture consisted of 11.7 µL 428 mM HEPES pH 8.0, 10 µL 1.5 mg/mL α,ω- di-fluorescein PEG 8000, 8 µL 250 mg/mL PEG 8000, 3 µ5 M potassium glutamate, 3 µL 500 mM magnesium glutamate, 14.3 µL Milli-Q. The final sample volume was 50 µL with a total PEG concentration of 40.3 g/L. Droplets were generated in a single layer device (oil flow: 120 µL/h, sample flow: 30 µL/h) and observed for three hours by bright field and fluorescence microscopy. The area of the droplets was determined using ImageJ. Droplet radii and volumes were calculated from the measured areas. The volume ratio between the size of the initial droplet and the droplet size when phase separation occurs was used to calculate critical PEG and internal salt concentrations.

3.3.3 Calculation of the distribution of PEG over the two phases

A mixture of HEPES buffer (100 mM, pH 8.0), PEG 8000 (40 g/L), α,ω-di-fluorescein PEG 8000 (0.3 g/L), potassium glutamate (300 mM) and magnesium glutamate (30 mM) was analyzed to investigate if salt-induced phase separation of PEG 8000 can occur in the absence of components from the cell-free expression kit. Potassium glutamate and magnesium glutamate were used as they are the main salts present in most cell-free protein expression mixtures [32]. Droplets were generated in single-layer devices and evaporation induced droplet shrinkage was followed for three hours. Upon shrinking, the mixture phase separated into a highly fluorescent PEG-rich phase and a non-fluorescent phase (Fig. 3.3A and 3.2B). The ratio between the initial droplet volume and the volume of shrunk droplets were used to calculate PEG concentration and internal ionic strength upon separation. The PEG-rich phase strongly localizes to the droplet/oil interface, either 62 Chapter 3: Cell lysate coacervates - possible models of crowding in cells appearing as bright circles when localized to the top or bottom of the droplet or crescent moon shaped when localized to the sides.

3.3.4 Inductively-coupled plasma optical emission spectrometry (ICP-OES)

The total K, Na, Mg, Zn, P and S-content of the cell-free expression kit, the coacervates and the dilute phase coexisting with the coacervates was determined by inductively- coupled plasma optical emission spectrometry (ICP-OES, Thermo Scientific iCAP 6300). We prepared five samples of reaction mixture (50 µL total volume) in small volume eppendorf tubes. One of these samples was kept at 4 ◦C and used for the analysis of the reaction mixture prior to phase separation. The other four samples were brought to phase separation by evaporating part of the water under a constant flow of argon at 4 ◦C. We measured the mass of all samples before and after phase separation. A small volume (2- 5 µL) of the coacervate (bottom) and dilute (top) phase was taken from each eppendorf tube and transferred to a new eppendorf. Then, 100 µL concentrated nitric acid (HNO3, 65 %) was added on top of these ICP samples and the total volume was transferred quantitatively into Milli-Q (total volume 5.0 mL, final nitric acid concentration 1.3 % (v/v)) in polystyrene sample tubes. To the non-phase separated sample we added 100 µL concentrated nitric acid as well, and then transferred it to Milli-Q (total volume 5.0 mL) quantitatively. ICP-OES analysis was carried out for K, Na, Mg, Zn, P and S. The in- strument was calibrated using one blank and four standards of known concentration K, Na, Mg, Zn, P and S in Milli-Q. The detection limit for each element was 1 ppm or lower.

3.3.5 Covalent labeling of the plasmid

We also measured DNA distribution over the phase separated droplets by covalently at- taching a dye to plasmid DNA. Plasmids were covalently labeled with fluorescein using the Label IT® Tracker™ Intracellular Nucleic Acid Localization Kit, Fluorescein from Mirus Bio (Madison, WI). The following protocol was used: 5 µL 10× Labeling Buffer A, 8 µL Label IT® Tracker™ Reagent, 6.3 µL pIVEX GFP plasmid (1263.9 ng/µL), 30.7 µL Milli-Q. The reaction mixture was incubated for one hour at 37 ◦C. After addition of 0.1 volumes 5 M NaCl and 2 volumes of ice cold 100 % ethanol, the sample was incubated overnight at -20 ◦C. The labeled plasmid was pelleted by centrifugation at 14000 rcf for 10 minutes, followed by a washing step by centrifugation at 14000 rcf for 10 minutes with 0.5 ml 70 ethanol to remove residual salts. The pellet was dissolved in µL Milli-Q. Plasmid concentration and labeling density were determined by UV-Vis spectrophotometry. The final plasmid concentration was 145 ng/µL and the plasmid was labeled with a fluorescein every 336 bases on average. 3.3. Materials and methods 63

3.3.6 Data Acquisition and Analysis

Liquids were pumped using adjustable pumps (Harvard Apparatus; PHD 2000 infusion). Temperature in the device was controlled via Bipolar Temperature Controller (CL-100). Droplets were stabilized with 2 % krytox–jeffamine ED-900–krytox surfactant dissolved in ﬂuorinated oil FC-40. The devices were mounted on the inverted microscope (Olym- pus IX81) equipped with a motorized stage (Prior; Optiscan II). Fluorescent images were taken with the sensitive EMCCD camera (iXon; Andor) using illumination from the mercury lamp or the laser. Analysis of images was done by ImageJ or home-written MAT- LAB routine.

3.3.7 Calculation of partition coefﬁcients of PEG and ﬂuorescently labelled lysate

The fluorescence intensity was calculated using ImageJ, measuring the area, integrated density and mean gray value [37]. Radii and volumes of droplets and coacervates were calculated from the measured area. The corrected fluorescence intensity (Fcorr) was calculated as follows: ZZ Fcorr = f dx dy − A f¯b, (3.1) A where f is the measured fluorescence intensity, f¯b is the mean background fluorescence intensity, A is the cross-sectional area of the fluorescent droplet and Fcorr is the background-corrected integrated fluorescence intensity, which is proportional to the total amount of fluorescent molecules. The volume and Fcorr values of the PEG-poor phase were calculated by subtracting the coacervate values from the total droplet values. The ratio between the Fcorr values (normalized to volume) of the PEG-rich and PEG-poor phase provides the partition coefficient P: v0 [PEG]coac = P × × [PEG]0, (3.2) vcoac where V is the volume and the subscripts 0 and coac denote the initial droplet and the coacervate phase, respectively. The lysate partition coefficient and final lysate concentrations were calculated identically.

3.3.8 Fluorescence recovery after photobleaching experiments

Photobleaching experiments were performed on an Olympus IX81 confocal microscope, equipped with an Andor iXon3 camera, Andor 400-series solid state lasers, a Yokogawa CSU-X1 spinning disk and an Andor FRAPPA photobleach module. Microdroplets of cell lysate, to which 200 nM eGFP (Cell Biolabs, Inc.) was added, were bleached before and after phase separation at 100 % laser intensity (λ = 488 nm), using three subsequent pulses of 20 µs. For bleaching, a thin stripe was selected in the middle of the droplet 64 Chapter 3: Cell lysate coacervates - possible models of crowding in cells or the coacervate, such that the intensity proﬁle perpendicular to the stripe, across the droplet was symmetric. Recovery of the ﬂuorescence intensity was monitored at a 5- fold reduced laser intensity (λ = 488 nm), using an exposure time of 15 ms. Taking into account the readout time, the imaging frequency was 1/51 ms−1 or 1/105 ms−1 for droplets before and after phase separation, respectively.

For the analysis of the fluorescence images, we use a custom-made routine in Matlab. Following Elowitz et al. [38] and Konopka et al. [39], we consider the recovery of the spatial eGFP concentration distribution to be a one-dimensional diffusion problem. We assume that the concentration c of eGFP is proportional to the fluorescence intensity f . At every slice perpendicular to the bleaching stripe, the equation governing the diffusion of eGFP can be written as ∂c ∂ 2c = D , (3.3) ∂t ∂x2 where the droplet edges are located at x = 0 and x = b. At the boundaries, reflecting boundary conditions apply

∂c ∂c = 0, = 0, (3.4) ∂ x=0 ∂ x=b

The general solution to this equation can be written as a Fourier series:

∞ nπx Dn2π2t c(x,t) = ∑ Ancos exp − 2 (3.5) n=0 b b

We assume the initial bleached proﬁle can be written as a step function

b b c(x,0) = c + (c − c ) H x − − a + H − a − x , (3.6) b 0 b 2 2 where c0 is the uniform concentration of eGFP in the droplet before bleaching and cb is the concentration of eGFP in the bleached area, directly after bleaching. H(z) denotes the Heaviside step function. The amplitudes of the Fourier series solution can be written 3.4. Conclusion 65 as b 2 Z πnx A = c(x,0)cos dx = n b b 0 (3.7) ∞ 2a(c0 − cb) 2(c0 − cb) nπ(b − 2a) nπ(b + 2a) = c0 − + ∑ sin − sin b n=1 nπ 2b 2b Finally, the recovery proﬁle follows from integration over the bleached area

b/2+a Z F(y,t) = f (x,y,t)dx (3.8) b/2−a The experimental fluorescence recovery F was calculated from the time-dependent confocal microscope images of bleached droplets. For every image a background intensity, measured far outside the droplet, was subtracted from the fluorescence intensity, and the net fluorescence intensity was integrated over the bleached area in x-direction, and averaged over 10-20 y-slices. The experimental recovery data were fitted to the predicted recovery based on the Fourier series solution, with fixed values for a, b, c0 and cb that were taken from the images and using the diffusion coefficient D as a fitting parameter.

3.4 Conclusion

Here, we demonstrate the formation of crowded coacervate compartments composed of cell lysate and show that coacervation creates an artificial cell-like environment. The characterization of the coacervates reveals that combination of E. coli lysate and PEG has a remarkable effect on the distribution of the biologically active components. This system presents a unique example of a salt-induced phase separation in a complex mixture where most components partition in a single crowded coacervate compartment. Such a process could also drive the formation of nucleoli and other complex non–membrane- bound intracellular compartments, where coacervation is driven by interaction between multivalent proteins, followed by accumulation of other classes of proteins via selective interactions or physical properties [26]. We find that the phase separation of droplets from the commercial gene expression kit is triggered at an internal salt concentration in the range of 1.0 M, with little dependence on the temperature at which the controlled evaporation is carried out (Fig. 3.5). Coacervate droplets formed from the kit have an 8- to 10-fold smaller volume than the initial droplets (Table 3.2), corresponding to a total macromolecule concentration of ∼200 g/L, of which ∼155 g/L is PEG 8000 and ∼45 g/L are lysate proteins, assuming the partition coefficients determined in the above-mentioned experiment using fluorescently labeled PEG and proteins. The difference between these values and those for coacervates containing fluorescently labeled 66 References lysate proteins and PEG is likely due to the presence of dye leading to the formation of smaller coacervates (Table 3.2). There is a clear correlation between coacervate volume and amount of lysate in the initial droplet (Fig. 3.6). Fluorescence recovery after photobleaching experiments probing the diffusion of GFP added to the coacervates confirm that the coacervate is a liquid phase with a density and viscosity strongly resembling the environment within living cells (Fig. 3.8). We have shown that the coacervates are crowded compartments, which, despite the large changes in buffer and lysate concentrations during droplet shrinking and phase transition, form a functional transcription and translation compartment as evidenced by the observed production of GFP.

Acknowledgements Joost Groen is kindly acknowledged for his experimental contribution (Fig. 3.4) to this chapter and useful discussions. Dr. Evan Spruijt is thanked for his help with ICP-OES analysis and FRAP experiments.

References

[1] J. Szostak, D. Bartel, and P. Luisi, “Synthesizing life,” Nature, vol. 409, no. 6818, pp. 387–390, 2001. [2] C. P. Brangwynne, T. J. Mitchison, and A. A. Hyman, “Active liquid-like behavior of nucleoli determines their size and shape in Xenopus laevis oocytes,” Proc. Natl. Acad. Sci. U. S. A., vol. 108, pp. 4334–4339, MAR 15 2011. [3] S. C. Weber and C. P. Brangwynne, “Getting RNA and Protein in Phase,” Cell, vol. 149, no. 6, pp. 1188– 1191, 2012. [4] T. E. Kaiser, R. V. Intine, and M. Dundr, “De Novo Formation of a Subnuclear Body,” Science, vol. 322, no. 5908, pp. 1713–1717, 2008. [5] R. Parker and U. Sheth, “P bodies and the control of mRNA translation and degradation,” Mol. Cell, vol. 25, no. 5, pp. 635–646, 2007. [6] P. Albertsson, “Partition of proteins in liquid polymer-polymer 2-phase systems,” Nature, vol. 182, no. 4637, pp. 709–711, 1958. [7] C. P. Brangwynne, C. R. Eckmann, D. S. Courson, A. Rybarska, C. Hoege, J. Gharakhani, F. Juelicher, and A. A. Hyman, “Germline P Granules Are Liquid Droplets That Localize by Controlled Dissolu- tion/Condensation,” Science, vol. 324, no. 5935, pp. 1729–1732, 2009. [8] T. W. Han, M. Kato, S. Xie, L. C. Wu, H. Mirzaei, J. Pei, M. Chen, Y. Xie, J. Allen, G. Xiao, and S. L. McKnight, “Cell-free Formation of RNA Granules: Bound RNAs Identify Features and Components of Cellular Assemblies,” Cell, vol. 149, no. 4, pp. 768–779, 2012. [9] M. Kato, T. W. Han, S. Xie, K. Shi, X. Du, L. C. Wu, H. Mirzaei, E. J. Goldsmith, J. Longgood, J. Pei, N. V. Grishin, D. E. Frantz, J. W. Schneider, S. Chen, L. Li, M. R. Sawaya, D. Eisenberg, R. Tycko, and S. L. McKnight, “Cell-free Formation of RNA Granules: Low Complexity Sequence Domains Form Dynamic Fibers within Hydrogels,” Cell, vol. 149, no. 4, pp. 753–767, 2012. [10] A. A. Hyman and C. P. Brangwynne, “Beyond Stereospeciﬁcity: Liquids and Mesoscale Organization of Cytoplasm,” Dev. Cell, vol. 21, no. 1, SI, pp. 14–16, 2011. [11] K. Luby-Phelps, “The physical chemistry of cytoplasm and its inﬂuence on cell function: an update,” Mol. Biol. Cell, vol. 24, no. 17, pp. 2593–2596, 2013. References 67

[12] E. B. Wilson, “The structure of protoplasm,” Science, vol. 10, pp. 33–45, 1899. [13] H. Walter, D. Brooks, and D. Fisher, “Partitioning in Aqueous Two-Phase Systems: Theory, Methods, Uses, and Applications to Biotechnology,” Academic Press, Orlando, FL, vol. , no. , p. , 1985. [14] P. A. Albertsson, “Partition of Cell Particles and Macromolecules,” Wiley-Interscience, New York, NY, vol. , no. , p. , 1985. [15] H. Walter and G. Johansson, “Partitioning in aqueous 2-phase systems - an overview,” Anal. Biochem., vol. 155, no. 2, pp. 215–242, 1986. [16] S. Cayley, B. Lewis, H. Guttman, and M. Record, “Characterization of the cytoplasm of escherichia-coli- k-12 as a function of external osmolarity - implications for protein dna interactions invivo,” J. Mol. Biol., vol. 222, no. 2, pp. 281–300, 1991. [17] A. Fulton, “How crowded is the cytoplasm,” Cell, vol. 30, no. 2, pp. 345–347, 1982. [18] V. B. Tolstoguzov, “Some physico-chemical aspects of protein processing into foodstuffs,” Food Hydro- colloids, vol. 2, no. 5, pp. 339–370, 1988. [19] S. S. Daube, D. Bracha, A. Buxboim, and R. H. Bar-Ziv, “Compartmentalization by directional gene expression,” Proc. Natl. Acad. Sci. U. S. A., vol. 107, no. 7, pp. 2836–2841, 2010. [20] C. A. Strulson, R. C. Molden, C. D. Keating, and P. C. Bevilacqua, “RNA catalysis through compartmentalization,” Nat. Chem., vol. 4, no. 11, pp. 941–946, 2012. [21] S. Koga, D. S. Williams, A. W. Perriman, and S. Mann, “Peptide-nucleotide microdroplets as a step towards a membrane-free protocell model,” Nat. Chem., vol. 3, no. 9, pp. 720–724, 2011. [22] H. G. Bungenberg-de Jong and H. R. Kruyt, “Coacervation (partial miscibility in colloid systems),” Proc Koninklijke Nederlandse Akademie Wetenschappen, vol. 32, no. , pp. 849–856, 1929. [23] M. W. Beijerinck, “Emulsionsbildung bei der Vermischung wassriger Losungen gewisser gelatinieren- den Kolloide (Emulsion formation during mixing of aqueous solutions of some gel-forming colloids),” Kolloid. Z., vol. 16, no. , p. , 1910. [24] H. G. Bungenberg-de Jong, I. Van der Horst, and A. Lafleur, “Zur KennCnis der Komplexkoazervation. XVII. Spezifische EnflGsse beim Koazervations-mischiypus 4-1 des Gummiarabicum Sols (On complex coacervates. XVII. Specific effects in the coacervate-type 4 : 1 mixture of the gumarabic sol),” Biochem. Z., vol. 161, no. , p. , 1933. [25] D. Dervichian and C. Magnant, “Formation et conditions dexistence des coacervats contenant des pro- teines .1. Action des precipitants habituels et des detersifs lipidiques,” Bull. Soc. Chim. Biol., vol. 29, no. 7-9, pp. 655–659, 1947. [26] P. Li, S. Banjade, H.-C. Cheng, S. Kim, B. Chen, L. Guo, M. Llaguno, J. V. Hollingsworth, D. S. King, S. F. Banani, P. S. Russo, Q.-X. Jiang, B. T. Nixon, and M. K. Rosen, “Phase transitions in the assembly of multivalent signalling proteins,” Nature, vol. 483, no. 7389, pp. 336–U129, 2012. [27] F. Courtois, L. F. Olguin, G. Whyte, D. Bratton, W. T. S. Huck, C. Abell, and F. Hollfelder, “An integrated device for monitoring time-dependent in vitro expression from single genes in picolitre droplets,” ChemBioChem, vol. 9, no. 3, pp. 439–446, 2008. [28] A. Fallah-Araghi, J.-C. Baret, M. Ryckelynck, and A. D. Griffiths, “A completely in vitro ultrahigh- throughput droplet-based microfluidic screening system for protein engineering and directed evolution,” Lab Chip, vol. 12, no. 5, pp. 882–891, 2012. [29] J. J. Agresti, E. Antipov, A. R. Abate, K. Ahn, A. C. Rowat, J.-C. Baret, M. Marquez, A. M. Klibanov, A. D. Griffiths, and D. A. Weitz, “Ultrahigh-throughput screening in drop-based microfluidics for directed evolution,” Proc. Natl. Acad. Sci. U. S. A., vol. 107, no. 9, pp. 4004–4009, 2010. [30] J.-u. Shim, G. Cristobal, D. R. Link, T. Thorsen, Y. Jia, K. Piattelli, and S. Fraden, “Control and mea- 68 References

surement of the phase behavior of aqueous solutions using microfluidics,” J. Am. Chem. Soc., vol. 129, no. 28, pp. 8825–8835, 2007. [31] J.-u. Shim, L. F. Olguin, G. Whyte, D. Scott, A. Babtie, C. Abell, W. T. S. Huck, and F. Hollfelder, “Simultaneous Determination of Gene Expression and Enzymatic Activity in Individual Bacterial Cells in Microdroplet Compartments,” J. Am. Chem. Soc., vol. 131, no. 42, pp. 15251–15256, 2009. [32] J. Shin and V. Noireaux, “Efficient cell-free expression with the endogenous E. Coli RNA polymerase and sigma factor 70,” J. Biol. Eng., vol. 4, p. 8, 2010. [33] K. Ananthapadmanabhan and E. Goddard, “Aqueous biphase formation in polyethylene oxide - inorganic salt systems,” Langmuir, vol. 3, no. 1, pp. 25–31, 1987. [34] L. A. Ferreira and J. A. Teixeira, “Salt Effect on the Aqueous Two-Phase System PEG 8000-Sodium Sulfate,” J. Chem. Eng. Data, vol. 56, no. 1, pp. 133–137, 2011. [35] A. A. Hyman and K. Simons, “Beyond Oil and Water-Phase Transitions in Cells,” Science, vol. 337, no. 6098, pp. 1047–1049, 2012. [36] T. Baumgart, A. T. Hammond, P. Sengupta, S. T. Hess, D. A. Holowka, B. A. Baird, and W. W. Webb, “Large-scale fluid/fluid phase separation of proteins and lipids in giant plasma membrane vesicles,” Proc. Natl. Acad. Sci. U. S. A., vol. 104, no. 9, pp. 3165–3170, 2007. [37] T. A. Potapova, S. Sivakumar, J. N. Flynn, R. Li, and G. J. Gorbsky, “Mitotic progression becomes irreversible in prometaphase and collapses when Wee1 and Cdc25 are inhibited,” Mol. Biol. Cell, vol. 22, no. 8, pp. 1191–1206, 2011. [38] M. Elowitz, M. Surette, P. Wolf, J. Stock, and S. Leibler, “Protein mobility in the cytoplasm of Escherichia coli,” J. Bacteriol., vol. 181, no. 1, pp. 197–203, 1999. [39] M. C. Konopka, I. A. Shkel, S. Cayley, M. T. Record, and J. C. Weisshaar, “Crowding and confinement effects on protein diffusion in vivo,” J. Bacteriol., vol. 188, no. 17, pp. 6115–6123, 2006. Chapter 4 Gene expression in crowded membrane-free protocells

4.1 Introduction

4.1.1 Overview of the protocell models

In the previous Chapter 3 we have shown that the coacervates of E. coli cell lysate form crowded, cell-like compartments. We also hypothesize that a functional coacervate forming spontaneously due to salt-driven liquid-liquid phase separation could be of interest in new theories for the compartmentalization of protocellular components. The idea that phase separation in polymer solutions may have played a role in the early stage of cell development is not new. Already in the 1930s, Soviet biochemist Oparin hypothesized that ‘coacervate droplets’ formed from organic macromolecules could have developed into the ‘simplest primary organisms’ [1]. Oparin performed reactions in coacervate systems, for example, enzymatically preparing polyadenyllic acid in RNA/histone coacervate droplets. Inspired by progress in abiotic synthesis, Fox prepared “thermal pro- tenoids”, microspheres formed when abiotic polypeptides, prepared by heating amino acids, were added to water [2]. Oparin, Fox and their followers did not address the need for heritable genetic material, and it should be noted that DNA was not known at the time Oparin was developing his theory of the origin of life. In light of clear importance of genetic information, coacervate theory was largely dismissed as an explanation of the origin of life by modern scientists [3].

The work in this Chapter was published in Proc. Nat. Acad. Sci. USA 2013, 110, 11692-11697

69 70 Chapter 4: Gene expression in crowded membrane-free protocells

The origin of the cell is an intriguing and continuous question in the field of evolutionary biology. In conventional analytical study of the cell one would destroy the cell and isolate its components. However it not necessarily that the final step in the formation of the cell is the reverse of that process. In order to understand the origin of modern cells it is important to accept the need for evolution from an original cell, so-called ‘protocell’, a precursor to the contemporary cell. Although we are still far from complete understanding of the emergence of first cell, some pieces of puzzle have been reported. Here we will focus on just a few examples of new protocell models. As mentioned earlier modern cells have evolved into systems rich in intracellular structure and functional complexity. All known cells use membranes composed of lipid bilayers as their compartment barriers, and therefore one of the ways to construct a simple protocell is to surround it with lipid membrane. Bilayer membranes provide semiperme- able boundary between inside and outside of the cell, which is important to retain macromolecules inside of the cell, while allowing smaller molecules, like nutrients and waste to travel in and out. Luisi and co-workers proposed that encapsulation during vesicle formation could be a possible route for compartmentalization of the complex solute mixtures [4]. They encapsulated cell free expression system in 100 nm lipid vesicles under conditions where statistical entrapment of 80 components was extremely unlikely; however they observed protein expression in part of the vesicles indicating that they contained all of the components. Although phospholipids are the primary membrane components in modern cell, it might be the case that the early cells used other amphiphiles. Mansy and co-workers have demonstrated that fatty-acid-based membranes provide the necessary permeability to small molecules such as nucleotides, while maintaining macromolecules (Fig. 4.1) [5]. Externally supplied activated nucleotides permeate across the protocell membrane and act as substrates for the non-enzymatic copying of internal templates. Complete template replication followed by random segregation of the replicated genetic material leads to the formation of daughter protocells.

Figure 4.1: Growth of the protocell membrane results from the incorporation of environmen- tally supplied amphiphiles, whereas division may be driven by intrinsic or extrinsic physical forces. Reproduced from [5] 4.1. Introduction 71

Despite these advances, it is still unclear how the many different molecules needed for an early life became concentrated in the same precellular compartment. And how did early cells acquire further intracellular complexity? The highly crowded interior of living cells with total macromolecule concentrations in excess of 300 g/L, [6] cannot be mimicked in vesicles as they would swell and rupture due to the resulting osmotic pressure. Alternative models based on liquid–liquid phase transitions that lead to the emergence of compartments have been proposed in recent years [1, 7]. As was discussed in Chap- ter 3, interior of the cell is partitioned into different thermodynamic phases. Keating and co-workers took a bottom-up approach to create a phase separated model of the protocell and showed that separation can affect the distribution of macromolecules (Fig. 4.2). They also demonstrated encapsulation of aqueous two-phase systems within lipid bilayers forming primitive model of cytoplasmic organization. Their work is mostly focused on PEG/dextran systems because their phase behavior is well understood and they can be easily encapsulated within cell-sized giant lipid vesicles [8]. Although synthetic polymers used in this work are not found in real cells, this research might be seen as a challenge to create more realistic models.

Figure 4.2: Reversible microcompartmentalization in an ATPS-containing vesicle. (A) Op- tical microscopy images during heating and subsequent cooling show phase transitions in the interior ATPS. (B) Fluorescence microscopy indicates the location of the lipid membrane (red) and a protein concanavalin A (green) before and after temperature changes. Scale bars are 10 µm. Reproduced from [8]

Mann et al. reported reaction compartmentalization in coacervate microdroplets formed by mixing nucleotides with low molecular weight cationic peptides [9]. Peptide- nucleotide microdroplets form spontaneously in water, are stable for changes in temperature and salt concentration, promote alpha-helical peptide structure and selectively sequester small molecules (Fig. 4.3). Later they demonstrated an example of hybrid protocell model which combines formation of coacervate microdroplets with the assembly of fatty acid membrane [10]. Coacervate microdroplets were prepared from cationic 72 Chapter 4: Gene expression in crowded membrane-free protocells peptides/polyelectrolytes and adenosine triphosphate or oligo/polyribonucleotides. They supported the membrane, which mediated selective uptake or exclusion of small and large molecules (see Fig. 4.4). Given this researchers claimed that these coacervate mi-

Figure 4.3: (a, b) Optical micrographs of droplets prepared at pH 8 in water from 20 mM solutions of peptides and ATP: oligolysine–ATP (1–5 kDa, 5–24 monomer units; bright ﬁeld) (a), polylysine–ATP (24 kDa, 115 monomer units, phase contrast) (b). Droplets were stained with cationic methylene blue. Reproduced from [9]

Figure 4.4: Molecular uptake/exclusion from fatty-acid-coated PDDA/ATP and oligolysine/RNA coacervate microdroplets. Reproduced from [10] crodroplets can be considered as a new membrane-free protocell models. Although significant progress has been made towards the synthesis of cell mimicking systems in the laboratory, these advances only partially fulfill the requirements necessary for the real- ization of a fully developed protocell model. Majority of the experiments to date have in- 4.2. Results and Discussion 73 vestigated only one or two component subsystems in isolation, such as vesicle properties (growth and division), compartmentalization of biochemical reactions, or transformation of encapsulated genetic molecules. Clearly, a protocell model that could successfully combine these elements in one integrated system would represent a major breakthrough in the field.

4.2 Results and Discussion

In this section we will show that coacervates of cell lysate form not only crowded, but also functional compartments as they are capable of carrying out protein synthesis reaction. Our experimental platform allows us study the effects of crowding on gene expression in a quantitative manner, and directly compare the kinetics of gene expression in non- crowded and crowded (coacervates) environments.

4.2.1 Modeling of transcription and translation in cell lysate with deterministic rate equations

The combination of the cell-free system with properly characterized transcription/ translation kinetics and mathematical predictive model is a valuable tool for quantitative studies of gene expression and genetic networks. Knowledge of kinetic parameters, expected protein yields, concentration of template DNA and speed of reaction of particular cell- free system, which is not provided generally, is essential. Here the procedure of how to specify the dynamics of gene expression through combination of calibration experiments and mathematical model is established. Deterministic rate equations are the standard approach to model biochemical processes like transcription and translation. To quantify the effect of compartmentalization and crowding on the production of mRNA and GFP, we constructed a deterministic model, based on the underlying biochemical reactions. Inspired by similar models in recent experiments on cell-free expression kits [11, 12] we write both transcription and translation as a two-step reaction: after a fast-equilibrium complexation of DNA to T7 RNA polymerase (KTS) or mRNA to the ribosome (KTL), transcription and translation proceeds with an overall rate of catalysis kTS and kTL, respectively. The key features of this model are shown schematically in Fig. 4.5. For example Karzbrun et al. used rate equations to describe the initial rising phase of protein synthesis in cytoplasmic E. coli extract [11]. The model uses ten free parameters for protein synthesis, degradation rates, as well as concentration of components. We constructed a deterministic model of transcription and translation in the cell-free expression kit, based on the underlying biochemical reactions, using a similar framework as outlined by Karzbrun et al. [11], Stögbauer et al.[12] and Stricker et al. [13]. Both transcription (TS) and translation (TL) occur in two steps. First, a rapid-equilibrium binding of the 74 Chapter 4: Gene expression in crowded membrane-free protocells

Figure 4.5: Schematic reaction pathway of transcription and translation, as used to model our data enzyme (T7 polymerase or the Ribosomal complex) and the template molecule (DNA or mRNA, respectively) is established

k f DNA + Pol )−*− DNA − Pol, (4.1) kr where the equilibrium constant KTS = k f /kr = [DNA − Pol]/([DNA] × [Pol]), or equivalently, KTL = [mRNA − Rib]/([mRNA] × [Rib]), and [A] denotes the actual concentration of free A. The complex, DNA − Pol or mRNA − Rib, then produces mRNA or protein, respectively, at an overall catalysis rate

d[mRNA] = k × [DNA − Pol] − k × [mRNA], (4.2) dt TS XS where kTS (or kTL) is the catalysis rate constant of transcription (or translation), respectively, which can be converted into an effective rate of nucleotide incorporation or amino acid incorporation, if the length of the mRNA and GFP is known. The actual concentration of complex is obtained as the real, positive root of the quadratic equation derived from Eq. 4.3:

2 KTS × [X] − (KTScDNA + KTScPol + 1) × [X] + KTScDNAcPol, (4.3) where cA denotes the overall concentration of component A and X is the complex DNA− Pol. The incorporation of nucleotides into the growing mRNA consumes NTPs. The incorporation of amino acids into the growing GFP consumes two GTPs per amino acid. To account for the slowdown of transcription and translation due to depletion of NTPs and amino acids, we add normalized prefactors containing [NTP] and [aa] to Equation 4.2 and the equivalent expression for translation. Equation 4.2 also contains a degradation term of mRNA by RNAse with rate constant kXS (kXL for protein degradation by proteases 4.2. Results and Discussion 75 in translation). We ﬁnd that the rate of both degradation processes can be neglected in the commercial cell-free expression kit on the timescale of the expression experiments (see Fig. 4.6), in contrast to the model used by Karzbrun et al. [11], and we have used kXS = kXL = 0 and [RNAse] = [Protease] = 0 in all model calculations.

Figure 4.6: Control experiment for mRNA and protein degradation. (A) Addition of RNAse inhibitors to the commercial cell-free expression kit does not change the expression of proteins. mRNA degradation is negligible during the experimental timescales. (B) Addition of chloramphenicol to the commercial cell-free expression kit inhibits translation of deGFP. No signiﬁcant degradation of already formed deGFP is observed on experimental timescales

After production of the protein, an additional, first-order maturation (folding) step can be included, with maturation rate constant kmat , leading to the maturated, fluorescent GFP∗. For studies on fast-folding deGFP production, we use a fixed maturation rate −1 constant kmat = 1/8.5min [24]. Finally, degradation or inactivation of the translation machinery could account for the fact that protein production levels off at a constant point in time (100-120 minutes after start), independent of DNA concentration and independent of the amount of protein produced, as was argued by Stögbauer et al. [12]. The inactivation could be caused by misfolding of the rRNA or incorrect reassembly of the ribosomal complex. Follow- ing Stögbauer et al., we include a first-order inactivation of the translation machinery, governed by an inactivation constant kiTL. In summary, our model contains the components DNA, mRNA, GFP, GFP∗, Pol, Rib, NTPC/U , NTPA/G, and amino acids, and constants KTS, KTL, kTS, kTL, kmat and kiTL and the time-evolution of these components can be described using the following set of differential equations.

d[mRNA] = k × α × [DNA − Pol] − k × [mRNA] (4.4) dt TS NTP XS d[GFP] = k × α × α × [mRNA − Rib] − k × [GFP] − k × [GFP] (4.5) dt TL GTP aa mat XL 76 Chapter 4: Gene expression in crowded membrane-free protocells

d[GFP]∗ = k × [GFP] − k × [GFP]∗ (4.6) dt mat XL

d[NTP]U/C k L α = − TS mRNA NTP × [DNA − Pol] (4.7) dt 2 d[NTP] k L α A/G = − TS mRNA NTP × [DNA − Pol] dt 2 (4.8) −akTLLGFPαNTPαaa × [mRNA − Rib] d[aa] − k L α α × [mRNA − Rib] + k L × ([GFP] + [GFP∗]) (4.9) dt TL GFP GTP aa XL GFP d[Pol] = −k × ([Pol] + [DNA − Pol]) (4.10) dt iTS d[Rib] = −k × ([Rib] + [mRNA − Rib]) (4.11) dt iTL The actual concentrations of the complexes DNA-Pol and mRNA-Rib are obtained from Eq. 4.3. The prefactors αNTP and αaa indicate the fraction of nucleotides and amino acids relative to their initial concentrations. The lengths of mRNA (LmRNA) and GFP (LGFP) are taken to be 3×238 and 238, respectively. We solve the above set of differential equations for a given set of initial concentrations and rate constants using Matlab. We have chosen the binding constants relevant for transcription and translation the same as reported previously [12]. The initial concentrations of ATP, RNA-polymerase and Ribosomes were estimated from comparison to experiments with our home-made cell-free expression kit (see Materials and Methods section), yielding cPol = 0.5nM, cRib = 12nM, cUTP = cCTP = 1.0M, and caa = 1.0M for all amino acids. In the case of mRNA production, all data points were shifted by a ﬁxed delay time, such that the steepest slope, expressed as ∆[mRNA]/∆t, measured over at least three subsequent points, was located at t = 0, analogous to the approach by Karzbrun et al.[21],who denote this time τ0. In the case of protein production, all data points were shifted by the same ﬁxed delay time as for mRNA production and no additional protein production delay τ f was included.

4.2.2 Transcription and translation in membrane-free protocells formed by coacervation of cell lysate

To directly compare mRNA and protein synthesis rates in single-phase droplets and in the more crowded coacervates, we proceed as follows: droplets were formed at 4 ◦C to prevent transcription and translation, and only after phase separation and partitioning of the cell lysate and the DNA (typically after 40 min), the temperature was raised to 25 ◦C and transcription and translation were initiated. 4.2. Results and Discussion 77

We ﬁrst focused our studies on the kinetics of mRNA production as a function of DNA concentration using a molecular beacon that is complementary to a part of the GFP mRNA sequence. As was mentioned earlier (Chapter 3) combination of E. coli cell lysate and PEG has a remarkable effect on unique distribution of biological components in coacervates. We elucidated the role of PEG on transcription rates, since high fraction of polymer is present in the coacervates. Preliminary experiments showed that increasing the amount of PEG from 0 wt% to between 5 and 8 wt% increases transcription rates, but that mRNA production is strongly inhibited when the concentration of PEG is raised further (>10 wt%), even at high salt concentrations, as shown in Fig. 4.7.

Figure 4.7: (A) PEG inhibits mRNA transcription at high concentrations. No appreciable mRNA production is observed for PEG concentrations of 8 wt% and above. This observation suggests inhibition of transcription at high PEG concentrations. We note that low concentrations of PEG (up to ∼2 wt%) seem to promote transcription, as can been seen by comparing the lanes with 0 wt% and 2 wt% PEG. (B) Combination of high PEG and low and high salt concentrations without compartmentalization into coacervates inhibits mRNA transcription in the same system as used in A, and detected by the molecular beacon

Surprisingly, in coacervates (where the PEG concentration exceeds 12-15 wt%) transcription is fully active: Fig. 4.8 shows the increase in ﬂuorescence intensity for dilute lysate and coacervates as a function of time for a DNA concentration of 160 pM in the starting droplet (see Fig. 4.9 for mRNA production from 40, 60, 80, 120 and 160 pM DNA). All curves exhibit a delay time of approximately 10-20 min and a slowdown in mRNA production after 80-120 min, probably due to depletion of transcription resources and inactivation of some of the components from the lysate [11, 12]. To allow comparison between different DNA concentrations, all mRNA concentrations in the coacervates have been normalized to a reference droplet diameter of 27 µm and coacervate of 13 µm. The lines are model predictions for equilibrium T7 RNA polymerase binding constants KTS and transcription rate constants kTS as indicated by the 78 Chapter 4: Gene expression in crowded membrane-free protocells

Figure 4.8: Compartmentalization enhances transcription. (A) mRNA production in droplets and coacervates. The open symbols represent single-phase droplets, and the closed symbols represent coacervates. Inset shows a zoom-in of the production in single-phase droplets. (B) Data (symbols) and model predictions (lines) of the initial rate of mRNA production in droplets and coacervates as a function of the initial plasmid DNA concentration (see Fig. 4.9 for corresponding mRNA production curves)

labels (Fig. 4.8). The single-phase droplet data are modeled with a binding constant −1 −1 KTS = 0.12 nM and a transcription rate constant KTS = 25 min (solid blue); the coacervate data are modeled with KTS= 0.12 (dotted), 1.0 (dashed), 10 (dash-dot), and −1 −1 100 (solid red) nM and kTS= 143 min . The black dotted line shows the predicted mRNA production for the case in which all macromolecular concentrations are increased to their actual value in the coacervates, but all binding constants and rate constants are unchanged with respect to single-phase droplets. We also assumed that a small amount of DNA (0.095 µm−2, corresponding to 35 pM in a droplet of 27 µm) is adsorbed to the oil–water interface (see Chapter 3 for details of adsorption of DNA). Fits to the experimental data were obtained by solving set of differential equations describing our model for a given set of initial concentrations and rate constants using MATLAB. We have chosen binding constants relevant for transcription and translation the same as reported −1 −1 previously [12]. Brieﬂy for single-phase droplets, KTS = 0.12 nM , KTL = 0.015 nM , −1 −1 −1 −1 kTS = 25min , kTL = 0.80min , kmat = 0.12 min , kiTS = 0.02 min , and kiTL = 0.1 −1 0 2 −1 −1 min . For phase-separated droplets, KTS = 10 – 10 nM , KTL = 0.015 nM , kTS −1 −1 −1 −1 = 43 min , kTL = 0.12 min , kmat = 0.12 min , and kiTS = 0.5 min and kiTL = 0.1 min−1 .Clearly the rate of mRNA production in the coacervate is increased about 50-fold compared to the non-shrunk droplets, and µM concentrations of mRNA are obtained in coacervate droplets from sub-nM concentrations of DNA in the starting droplet. The fact that the coacervates act as compartments in which most of the relevant macromolecular components of transcription have been accumulated (Fig. 3.4, Chapter 3), accounts for part of the increase in transcription rate. However, our kinetic model shows that this 4.2. Results and Discussion 79

Figure 4.9: mRNA production from 40 (E), 60 (D), 80 (C), 120 (B), and 160 pM DNA (A). The data points represent average concentrations of at least 40 droplets, and the error bars represent their standard deviations. Insets show zoomed part of the data from single-phase droplets. The solid lines are fits of the combined data set to the deterministic model. All data points are corrected for the actual size of the droplets (Table 3.2, Chapter 3), and the increased concentrations due to droplet shrinkage. The values shown in this figure are normalized to a reference droplet of 27 µm diameter and a coacervate of 13 µm, respectively, to allow comparison of the different DNA concentrations 80 Chapter 4: Gene expression in crowded membrane-free protocells concentrating effect is not all. Even if we take into account the actual concentrations of DNA, T7 RNA polymerase (and the ribosomes and all other proteins) in the coacervates as determined above, we cannot predict as high a rate of transcription as we find experimentally. Two additional effects are likely to play a role in the coacervates. Crowding generally alters the association constants of complexation equilibria [14] such as the T7 RNA polymerase binding here. Secondly, the transcription rate constant kTS may be increased in the coacervates. The fits in Fig. 4.8A show that a combination of these effects, that is, an increase in the DNA-association constant by two orders of magnitude, and a −1 −1 5-6-fold increase in kTS, from 25 min to 143 min , leads to a predicted transcription rate that matches our experiments. The concentration of DNA strongly affects the rate of mRNA production, both in single-phase droplets and in coacervates. Figure 4.8B shows that our model correctly predicts the mRNA production rates for all DNA concentrations we investigated. In all cases the rate of mRNA production in the coacervates is enhanced by nearly two orders of magnitude, compared to the single-phase droplets. At very high DNA concentrations the rate of mRNA production will level off because every RNA polymerase is already bound to a DNA molecule. The DNA concentration at which leveling off occurs is similar to previous bulk experiments using another E. coli-based cell-free expression kit [12]. At very low DNA concentrations the mRNA production in single-phase droplets is too small to be detected reliably. Remarkably, both the RNA polymerase binding constant and the transcription rate constant in coacervates are of the same order of magnitude as the values typically found for E. coli in vivo [15], whereas two orders of magnitude lower constants are usually found in bulk experiments on cell- free expression kits with the same RNA polymerase in vitro [11, 12]. This supports our conclusion that the coacervates are crowded environments that mimic the conditions necessary for in vivo transcription. In contrast to the enhanced transcription, protein synthesis rates in coacervates are adversely affected by the high PEG concentrations. Although small amounts of PEG are typically added to optimize cell-free protein synthesis rates, higher concentrations completely prevent protein synthesis [16]. Indeed, we observed no appreciable production of GFP in home-made lysate when >10 % PEG 8000 was added (see Fig. 4.10). Nevertheless, we did observe higher concentrations of GFP in the coacervates than in dilute droplets (Fig. 3.2 B and C, Chapter 3), as indicated by the high fluorescence intensity. As GFP is a slow-folding protein, the increased concentrations observed in Fig. 3.2 C, Chapter 3 could be due to delayed folding of already produced proteins in the coacervate. We therefore extended our studies to include a fast folding mutant, deGFP with a maturation time of 8.5 min, [17] to decouple translation and fluorophore maturation. Figure 4.11 shows that the rate of fluorescence increase resulting from the fast folding deGFP is significantly larger in the coacervates, although the difference with single phase droplets is smaller than for transcription. In fact, if we again assume that all protein and RNA components necessary for translation have been accumulated to the same extent in 4.3. Materials and Methods 81

Figure 4.10: (A) No appreciable GFP production is observed at 10 wt% PEG, whereas lower PEG concentration shows signiﬁcant GFP production. This observation suggests inhibition of transcription or translation at high PEG concentrations. (B) Combination of high PEG and high salt concentrations without compartmentalization into coacervates inhibits mRNA transcription in the same system as used in (A) the coacervates, we can account for the increased rate of translation with a translation binding constant KTL that is unchanged and an overall rate constant kTL that is 6-7-fold lower, as expected from the adverse effect of PEG. We emphasize that this assumption is much stricter in case of translation, since tens of proteins and RNA fragments need to assemble before translation can take place. Nevertheless, the coacervates clearly form fully functional compartmentalized systems capable of protein synthesis, which is truly remarkable considering the high concentrations of PEG.

4.3 Materials and Methods

4.3.1 Materials

We used RTS 100 Escherichia coli HY kit (5PRIME) for all activity and kinetics studies. Alexa Fluor 647 (Invitrogen) was used for calibration curve of molecular beacon. Recombinant enhanced GFP (eGFP) (1mg/mL) (Cell Biolabs) was used for calibration curve of eGFP production. Sylgard 184 silicone elastomer kit polydimethylsiloxane (PDMS) (Dow Corning) was used for the microﬂuidic device fabrication. Fluorinert FC-40 oil (Sigma-Aldrich) was used as a continuous phase in the ﬂow. GFP-His 9plas- mid pRSET5d-GFPHis) was a gift from Kerstin Blank (Radboud University Nijmegen). Fast folding eGFP (deGFP) (plasmid pRSET5d-UTR1-eGFP-Del6-229: from plasmid pRSET5d-GFPHis, insertion of UTR1-eGFP-Del6-229 from pBEST-OR2-OR1-UTR1- eGFP-Del6-229-T500) was a gift from Vincent Noireaux and Jonghyeon Shin (Univer- 82 Chapter 4: Gene expression in crowded membrane-free protocells

Figure 4.11: Protein expression in coacervates. deGFP production in droplets and coacervates. The open symbols represent single-phase droplets, and the closed symbols represent coacervates. The solid lines are model predictions for equilibrium ribosome binding constants and translation rate constants as indicated by the labels. The coacervate data are modeled with the actual concentrations of all macromolecular components and enhanced −1 −1 transcription binding and rate constants KTS = 10 nM and kTS = 143 min , respectively (Fig. 4.8 A), compared with the single-phase droplets sity of Minnesota, Minneapolis, MN), and T7 RNA polymerase (50 units/µL) (New Eng- land Biolabs).

4.3.2 Home made in vitro transcription translation system

Reaction mixtures of the home made in vitro transcription translation kit were composed of one third cell lysate and two thirds reaction buffer. Both were prepared with slight modiﬁcations according to Shin and Noireaux [17]. The lysate was prepared with E. coli BL21 cells grown at 37 ◦C in 2YT medium up to OD600 = 1.5. After cell growth all further lysate puriﬁcation steps were performed on ice or at 4 ◦C. The cells with a typical wet weight of 5 g were re-suspended in 15 mL of S30 buffer A containing: 50 mM Tris, 60 mM potassium glutamate, 14 mM magnesium glutamate, 2 mM DTT, adjusted to pH 7.7, and centrifuged (3000 g for 10 minutes) twice. After a third re-suspension with 5 mL of S30 buffer A, the cells were broken with a bead – beater (Biospecs Products 4.3. Materials and Methods 83

Inc, mini bead-beater-1) using 0.1 mm glass beads, followed by centrifugation (30000 g for 30 minutes). The supernatant was dialyzed with S30 buffer B (5 mM Tris, 60 mM potassium glutamate, 14 mM magnesium glutamate, pH 8.2, 1 mM DTT) for 3 times 45 minutes followed by one dialysis overnight (4 x 250 mL) at 4 ◦C. A concentration of 22 mg/mL of proteins in the cell lysate was obtained, and the lysate was stored at -80 ◦C. The reaction buffer consisted of 50 mM Hepes adjusted to pH 7.5, 1.5 mM ATP and GTP each, 0.9 mM CTP and UTP each, 1mM spermidine, 0.75 mM cAMP, 0.33 mM NAD, 0.26 mM coenzymeA, 30 mM 3-PGA, 0.068 mM folinic acid, 0.2 mg/mL tRNA, 1 mM IPTG, 1.5 mM of each amino acid, 20 mM magnesium glutamate, 200 mM potassium glutamate and PEG 8000 (2% w/v).

4.3.3 Molecular beacons for mRNA labeling

If mRNA levels are to be measured fluorescently it is necessary to apply a fluorescent label. Therefore a dye should be non-toxic to cell-free transcription machinery and bind specifically to mRNA. Molecular beacons fulfill these requirements. These molecular beacons are oligonucleotides with the stem-loop structure. The stem forms because two complementary sequences form and the ends of the beacons anneal at the temperatures lower than their melting temperature. The loop sequence is complementary to a part of the target mRNA and hybridizes with it when it comes in close proximity. Since the loop sequence is longer than the stem the hybrid is more stable than the hairpin structure and it is therefore the prevalent configuration. This separates two ends of the beacon therewith the fluorescent donor-quencher pair attached to said ends. From this point on the fluorescence of the donor can be detected.

Figure 4.12: Schematic representation of molecular beacon probe

The sequence of the molecular beacon used to probe the GFP mRNA (SI Text, S6) concentration was as follows: 5’-(Alexa 647)-GCGCAAAUAAAUUUAAGGGUAAGCGC-(Iowa Black Quencher)-3’. The backbone of the molecular beacon was composed of 2’-O-methylribonucleotides. The molecular beacon was designed using the mfold program [18] according to the method of Bratu et al. [19], and by using the IntaRNA, version 1.2.5, program, accessible on the Freiburg RNA Tools server (http://rna.informatik.uni-freiburg.de:8080/IntaRNA.jsp). 84 Chapter 4: Gene expression in crowded membrane-free protocells

The molecular beacon was synthesized by Integrated DNA Technologies (optical density at 260 nm = 5.3), resuspended in autoclaved Milli-Q water to a concentration of 50.2 µM, and stored in light-protected tubes at -20◦C.

4.3.4 Data acquisition and analysis

Liquids were pumped using adjustable pumps (Harvard Apparatus; PHD 2000 infusion). Temperature in the device was controlled via Bipolar Temperature Controller (CL-100). Droplets were stabilized with 2 % krytox–jeffamine ED-900-krytox surfactant dissolved in ﬂuorinated oil FC-40. The devices were mounted on the inverted microscope (Olym- pus IX81) equipped with a motorized stage (Prior; Optiscan II). Fluorescent images were taken with the sensitive EMCCD camera (iXon; Andor) using illumination from the mercury lamp or the laser. Analysis of images was done by ImageJ or home-written MAT- LAB routine.

4.3.5 Method for mRNA production experiments in cell lysate

The reaction mixture contained the following: 10 µL of reaction mix, 5 µL of reconstitution buffer, 12 µL of amino acids, 1 µL of methionine, 12 µL of E. coli lysate, 10 µL of pRSET5d-GFPHis, and 1 µL of molecular beacon (1 µM). Droplets were produced in the double-layer device (oil flow: 120 µL/h; mixture flow: 30 µL/h). Constant temperature was maintained by flushing liquids through Peltier unit at the high flow rate of 100–500 µL/min. One part of the device was covered with channels containing Milli-Q, and the other part, with saturated salt solution at 4 ◦C. When phase separation had occurred in the droplets that were covered with salt-rich reservoir channels and distribution of all macromolecules (including DNA) over the two phases had reached equilibrium, the temperature of the device was raised to 25◦C (within a minute) to start up the reaction and followed the next 2 h using fluorescence microscopy. Device was mounted on Olympus IX81 inverted microscope, where sample was excited using mercury lamp. Fluorescence readout was performed using Andor iXon3 EMCCD. Calibration curve with the same acquisition settings for Alexa Fluor 647 was made to obtain concentrations of mRNA in the droplets (Fig. 4.13A).

4.3.6 Method for GFP production experiments in cell lysate

The reaction mixture contained the following: 10 µL of reaction mix, 5 µL of reconstitution buffer, 12 µL of amino acids, 1 µL of methionine, 12 µL of E. coli lysate, and 10 µL of pRSET5d- UTR1-eGFP-Del6-229. Droplets were produced in the double-layer device (oil flow: 120 µL/h; cell lysate flow: 30 µL/h). Constant temperature was maintained by flushing liquids through Peltier unit at the high flow rate of 100–500 µL/min. One part 4.3. Materials and Methods 85

Figure 4.13: mRNA and deGFP detection. (A) Calibration curve for the molecular beacon using Alexa Fluor 647 (Invitrogen) in droplets. (B) Calibration curve of EGFP (Cell Biolabs) in droplets, measured on Olympus IX81 confocal microscope. The detection settings are indicated in the Figure of the device was covered with channels containing Milli-Q, and the other part, with saturated salt solution at 4◦C. When phase separation had occurred in the droplets that were covered with salt-rich reservoir channels and distribution of all macromolecules (including DNA) over the two phases had reached equilibrium, the temperature of the device was raised to 25 ◦C (within a minute) to start up the reaction and followed the next 2 h using ﬂuorescence microscopy. The device was mounted on Olympus IX81 confocal microscope, where the sample was excited using 20 % laser power (λ = 488 nm). Fluo- rescence readout was performed using an Andor iXon3 EMCCD camera at an exposure time of 0.1 s. Calibration curve with the same acquisition settings for eGFP was made to obtain concentrations of the protein in the droplets (Fig. 4.13B).

4.3.7 Method for transcription in bulk at various PEG concentrations

See Fig. 4.7 A. The reaction mixture contained the following: Tris·HCl, pH 8.1, 40 mM; DTT, 5 mM; spermidine, 1 mM; MgCl2, 25 mM; ATP, GTP, CTP, UTP, 4 mM each; guanosine monophosphate, 5 mM; T7 RNA polymerase, 1 unit/ µL of reaction mixture; PEG 8000 concentrations of 0 %, 2 %, 5 %, 8 %, 10 %, 11.5 %, and 16.5 %; and 3 nM pRSET5d-GFPHis. The reaction was run at 30 ◦C for 3 h and analyzed on a 8 % acrylamide gel (TBE buffer, 8 M urea). See Fig. 4.7 B. The reaction mixture contained the following: Tris·HCl, pH 8.1, 40 mM; DTT, 5 mM; spermidine, 1 mM; MgCl2, 25 mM; ATP, GTP, CTP, and UTP, 4 mM each; guanosine monophosphate, 5 mM; T7 RNA polymerase, 1 unit/µL of reaction mixture; PEG 8000 concentrations of 0 %, 2 %, 5 %, 8 %, 10 %, 11.5 %, and 16.5 %; pRSET5d-GFPHis, 3 nM for low- and 1 nM for high-salt experiments; and 1 µL of 86 Chapter 4: Gene expression in crowded membrane-free protocells molecular beacon (1 µM). K+ was brought to 975 mM (equal to the salt concentration inside the coacervates; see Chapter 3) by addition of solid potassium glutamate before incubation. Transcription in bulk was followed by fluorescence measurements in a microplate (Nunclon 96 Flat Bottom Black Polystyro), using a microplate reader (Tecan Infinite 200Pro) with excitation λ = 630 nm and emission λ = 667 nm. The reaction was carried out at 30 ◦C and followed for the next 4 h. The home-made in vitro transcription and translation system was prepared as described in Section 4.3.2 with pRSET5d- GFPHis (1 nM) and PEG 8000, 2 %, 5 %, 8 %, and 10% (wt/vol). Before incubation, 1 µL of molecular beacon was added to the reaction mixture to allow detection of mRNA. In addition, solid potassium glutamate was added to reach a final K+ concentration of 975 mM (equal to the potassium concentration inside the coacervates; see Chapter 4). Transcription in bulk was followed by fluorescence measurements on a microplate reader (same as above) with excitation λ = 630 nm and emission λ = 667 nm. The reaction was carried out at 30 ◦C and followed for the next 4 h.

4.3.8 Method for translation in bulk at various PEG concentrations

See Fig. 4.10 The home-made in vitro transcription and translation system was prepared as described in Chapter 4 with pRSET5d-GFPHis (5 nM) and PEG 8000, 2 %, 4 %, 7 %, and 10 % (wt/vol). Translation in bulk was followed by ﬂuorescence measurements in a microplate (Nunclon 96 Flat Bottom Black Polystyro), using a microplate reader (Tecan Inﬁnite 200Pro) with excitation λ = 395 nm and emission λ = 509 nm. The reaction was carried out at 37 ◦C and followed for the next 5 h.

4.4 Conclusion

In this Chapter a mathematic model of cell-free gene expression kinetics based on differential equations for individual kinetics rates was presented. Furthermore, the model assumes a finite pool of resources for transcription and translation to account for the observed cessation of in vitro protein synthesis after a fixed time. This way cell-free gene expression kinetics could be fitted from the early to the late phase and this over several orders of magnitude of DNA concentration. This properly characterized cell-free system in combination with the predictive model presented here is a valuable tool for studies on transcription and translation kinetics or on in vitro gene regulation networks. The phase separation of E. coli cell lysate into a dense liquid coacervate containing all macromolecular components has enabled a direct comparison between mRNA synthesis in dilute and crowded environments. Coacervation creates an artificial cell-like environment in which the rate of mRNA production is increased significantly. Measured transcription rates were fitted with mathematical model and showed a two orders of mag- References 87 nitude larger binding constant between DNA and T7 RNA polymerase, and five to six times larger rate constant for transcription in crowded environments, strikingly similar to in vivo rates. We demonstrate that crowding significantly enhances the binding constant of T7 RNA polymerase to DNA and the transcription rate constant, which are direct results of crowding affecting the kinetics of the rate-determining steps of the fundamental machinery of gene expression in the cell. Our results thus impact on our understanding of how biochemical networks function, as rates of enzymatic reactions determined in dilute solution do not necessarily reflect in vivo rates. We show the intrinsic potential of cellular components to facilitate macromolecular organization into membrane free compartments by phase separation. It is striking that both the mRNA (and the molecular beacon) and the GFP remain associated with the coacervate during their production, even in the absence of lipid bilayer membrane around the droplets [20]. Our experimental platform enables a systematic study into the effects of crowding on key cellular processes such as transcription and translation in membrane-free protocells.

Acknowledgements Dr. Evan Spruijt is gratefully acknowledged for building the mathematical model and productive discussions. Emilien Dubuc is thanked for the design of the molecular beacon and experimental (Fig. 4.7) contribution to this chapter. Maike Hansen is acknowledged for her experimental contribution (Fig. 4.10) to this chapter.

References

[1] A. I. Oparin, The origin of Life. 1953. [2] S. Fox, “A theory of macromolecular and cellular origins,” Nature, vol. 205, no. 4969, pp. 328–&, 1965. [3] A. Lazcano, “Historical Development of Origins Research,” Cold Spring Harbor Perspect. Biol., vol. 2, no. 11, 2010. [4] P. Stano, T. Pereira de Souza, M. Allegretti, Y. Kuruma, and P. L. Luisi, New and unexpected insights on the formation of protocells from a synthetic biology approach: the case of entrapment of biomacro- molecules and protein synthesis inside vesicles. 2011. [5] S. S. Mansy, J. P. Schrum, M. Krishnamurthy, S. Tobe, D. A. Treco, and J. W. Szostak, “Template-directed synthesis of a genetic polymer in a model protocell,” Nature, vol. 454, no. 7200, pp. 122–U10, 2008. [6] A. Fulton, “How crowded is the cytoplasm,” Cell, vol. 30, no. 2, pp. 345–347, 1982. [7] A. A. Hyman and K. Simons, “Beyond Oil and Water-Phase Transitions in Cells,” Science, vol. 337, no. 6098, pp. 1047–1049, 2012. [8] M. Long, C. Jones, M. Helfrich, L. Mangeney-Slavin, and C. Keating, “Dynamic microcompartmentation in synthetic cells,” Proc. Natl. Acad. Sci. U. S. A., vol. 102, no. 17, pp. 5920–5925, 2005. [9] S. Koga, D. S. Williams, A. W. Perriman, and S. Mann, “Peptide-nucleotide microdroplets as a step towards a membrane-free protocell model,” Nat. Chem., vol. 3, no. 9, pp. 720–724, 2011. [10] T.-Y. D. Tang, C. R. C. Hak, A. J. Thompson, M. K. Kuimova, D. S. Williams, A. W. Perriman, and S. Mann, “Fatty acid membrane assembly on coacervate microdroplets as a step towards a hybrid protocell model,” Nat. Chem., vol. 6, no. 6, pp. 527–533, 2014. 88 References

[11] E. Karzbrun, J. Shin, R. H. Bar-Ziv, and V. Noireaux, “Coarse-Grained Dynamics of Protein Synthesis in a Cell-Free System,” Phys. Rev. Lett., vol. 106, no. 4, 2011. [12] T. Stoegbauer, L. Windhager, R. Zimmer, and J. O. Raedler, “Experiment and mathematical modeling of gene expression dynamics in a cell-free system,” Integr. Biol., vol. 4, no. 5, pp. 494–501, 2012. [13] J. Stricker, S. Cookson, M. R. Bennett, W. H. Mather, L. S. Tsimring, and J. Hasty, “A fast, robust and tunable synthetic gene oscillator,” Nature, vol. 456, no. 7221, pp. 516–U39, 2008. [14] A. Minton, “Macromolecular crowding,” Curr. Biol., vol. 16, pp. R269–R271, APR 18 2006. [15] S. Sastry and B. Ross, “Nuclease activity of T7 RNA polymerase and the heterogeneity of transcription elongation complexes,” J. Biol. Chem., vol. 272, no. 13, pp. 8644–8652, 1997. [16] X. Ge, D. Luo, and J. Xu, “Cell-Free Protein Expression under Macromolecular Crowding Conditions,” PLoS One, vol. 6, no. 12, 2011. [17] J. Shin and V. Noireaux, “Efﬁcient cell-free expression with the endogenous E. Coli RNA polymerase and sigma factor 70,” J. Biol. Engineer., vol. 4, pp. e28707–1–10, 2010. [18] M. Zuker, “Mfold web server for nucleic acid folding and hybridization prediction,” Nucleic Acids Res., vol. 31, no. 13, pp. 3406–3415, 2003. [19] D. P. Bratu, I. E. Catrina, and S. A. E. Marras, “Tiny molecular beacons for in vivo mRNA detection,” Methods Mol. Biol., vol. 714, no. 13, pp. 141–157, 2011. [20] J. Szostak, D. Bartel, and P. Luisi, “Synthesizing life,” Nature, vol. 409, no. 6818, pp. 387–390, 2001. Chapter 5 Understanding the effect of macromolecular crowding on genetic networks in synthetic cellular systems

Abstract As discussed in previous Chapters, the significance of effects of macromolecular crowding has been a subject of numerous discussions in recent years [1]. Many publications have shown that it has an influence on reactions of association and enzyme activity [2]. Meanwhile, other reports have indicated that it has no or insignificant effect on DNA hybridization kinetics [3] and on protein–protein interactions [4]. In a recent study Tan et al. [5] reported on the effects of molecular crowding on gene expression. Tan et al. investigated the effect of molecular crowding on gene expression and complex gene networks. The researchers found that crowding made gene expression more robust with respect to perturbations by ions, which tend to influence enzyme function. Furthermore, they found that crowding alters the behaviour of a simple gene regulatory module that contained a negative feedback loop. Finally, Tan et al. constructed an artificial cell from lipid membrane vesicles that encapsulated a synthetic expression system and a genetic construct. The liposomes could express green fluorescent proteins using the genetic construct, and enhanced gene expression was claimed when macromolecular crowding agents were added to the liposomes. This is an important topic, as the impact of the

The work in this Chapter was published in Nature Nanotechnology, 2014, 9, 406-407

89 Chapter 5: Understanding the effect of macromolecular crowding on genetic networks 90 in synthetic cellular systems physical, crowded environment within the cell on key enzymatic reaction networks, such as transcription and translation, is far from understood. Unfortunately, Tan et al. base their work on a wrong interpretation of the theory of crowding, and draw false conclusions from their experimental work due to a poor data collection method. In this Chapter, we present an alternative interpretation of the data gathered by Tan et al. using kinetic model we described in the previous Chapter 4.

5.1 Introduction

As discussed earlier, macromolecules occupy over 30% of a cell’s volume, lowering protein diffusion by order(s) of magnitude and inﬂuencing enzymatic reactions via increased binding constants of macromolecular complexes [6, 7]. The term macromolecular crowding (graphically shown in Fig. 5.1) is used to describe this condition of sig- niﬁcant volume-occupancy and the associated thermodynamic consequences for binding and rate constants [8, 9]. For a detailed discussion of macromolecular crowding see Chapter1.

Figure 5.1: Excluded (pink and black) and available (blue) volume in a solution of spherical background macromolecules. A,volume available to a test molecule of inﬁnitesimal size;B, volume available to a test molecule of size comparable with background molecules

Crowding can be mimicked experimentally, by adding high concentrations of inert synthetic or natural macromolecules, termed crowders. The magnitude of these effects (Fig. 5.2) can be estimated by using equations developed to describe the dynamics of ﬂuid containing hard spheres or coiled polymers. According to the classical theory of crowding: size ratio between the crowding molecules and the reacting species is the key parameter that determines the effect on 5.2. Results and Discussion 91

Figure 5.2: Effect of the volume fraction Φ of crowders upon the equilibrium constant for site-binding by a spherical ligand, for MWligand/MWcrowder = 2 (a), 1(b) or 0.5 (c) reaction equilibria and kinetics (Fig. 5.2) [6, 10–12]. At equal weight/volume concentrations, the low molecular weight crowders should lead to the largest enhancement of binding constants, since more of the smaller crowders are present per unit volume. This effect of the size of the crowders, in relation to the size of the probe, has been conﬁrmed experimentally [13].

5.2 Results and Discussion

To elucidate the effects of crowding on gene expression the authors [5] used dextrans of different molecular weights (Dex-Small of 6kDa, and Dex-Big of 2000kDa) as crowders. However, the authors made a fundamental error when they claim that the effects of the molecular weight of the crowder they observe (Fig. 5.3) are “consistent with a general theory of molecular crowding that suggest a larger enhancement of molecular binding by large crowding molecules than by small crowding molecules”, thereby referring to [10]. This is the opposite what theory predicts (Fig. 5.2) [6, 10–12]. Any comparison of crowders should be made on the basis of their occupied volume, and not on the basis of their weight concentration (which the authors have given the misleading name “molecular density”). A crude estimate of volume occupied by the “inert” low molecular weight ‘small’ dextran at 10 wt% (Fig. 5.3), based on hydrodynamic radius of dextran 6kDa, shows that no more than 16% [14] of the available volume is Chapter 5: Understanding the effect of macromolecular crowding on genetic networks 92 in synthetic cellular systems

Figure 5.3: a) Gene expression rates in environments containing small crowding molecules (Images taken from [5]). b) Gene expression rates in environments containing big crowding molecules (Images taken from [5]) occupied by the crowding agent, and one should not expect to see a significant effect of crowding on gene expression (Fig. 5.2). Any effect on gene expression at low volume fractions of crowder may result from subtle molecular interactions between the crowders and the transcription/translation machinery (a chemical effect). This is the case for most experiments in [5], including the studies on alternative crowders like Ficoll and PEG. As discussed in previous chapters, macromolecular crowding leads to a reduction in diffusion coefficients. The researchers studied the binding of RNA polymerase to its promoter binding site by fluorescence recovery after photobleaching experiments (FRAP). They found that in the presence of large macromolecular crowding agents the association rate of RNA polymerase to the promoter was increased, whereas its dissociation rate was reduced (Fig. 5.4). This in turn led to an increased gene expression rates in more crowded environments, which was indicated through the increased cell-free production of a green fluorescent protein. This is very important conclusion, which we attempted to verify by careful analysis of the FRAP data. The FRAP recovery rates published in Fig. 5.4 are very low. The limiting data at 0%Dex should be comparable to free diffusion of the RNAP-RFP fusion protein, since no DNA or other proteins are added to this solution. Crude estimate of the diffusion co- −1 ln(2)60 efficient from Fig. 5.5 gives: k = 0.5 min , t1/2 = k = 83s (makes sense, based on 2 trace in Fig. 5.5), D = w ln(2) = 0.8−1.2µm2/s (taking w, the half-width of the bleached 4t1/2 area, to be 20-25 µm). The free diffusion coefficient of RNAP in aqueous solution can be estimated using a conservative estimate of the Stokes radius [15] and should be approxi- kT 2 mately D f ree = 6πηR = 25 − 35µm /s. Even taking into account that RFP is attached to 5.2. Results and Discussion 93

Figure 5.4: Fluorescence recovery after photobleaching (FRAP) using RFP- T7RNAP [5]a)Initial recovery rates in presence of small- and big- dextrans; (Images taken from [5]) b) Immobile fractions of RFP–T7 RNAP. (Images taken from [5]) the RNAP, we cannot explain a more than 1 order of magnitude discrepancy between the data of Tan and theory. Most likely, the RNAP is aggregated or it is sticking to the glass surface (Fig. 5.5), which both lower the apparent diffusion coefficient. Furthermore, the FRAP experiments in [5] do not provide experimental support for enhanced binding of RFP-T7 RNAP polymerase to fluorescently labelled DNA as a result of crowding, because 0.2 wt% Dex-Big already shows a significant increase in binding events compared to 0.2 wt% Dex-Small. This doubt is reinforced by Tan’s unrealistic finding in Fig. 5.4 that a 4% (w/v) solution of 6 kDa dextran and 2 MDa dextran would slow down diffusion of a nanometer-sized probe molecule to the same extent, while the viscosities of these solutions differ by more than an order of magnitude [14].

Figure 5.5: a) A sample RFP time series in a FRAP experiment; b) Schematic of the FRAP experimental setup, (Images taken from [5]) Chapter 5: Understanding the effect of macromolecular crowding on genetic networks 94 in synthetic cellular systems

Since crowding generally affects binding constants and rate constants of the enzymatic reactions involved in gene expression, the rate at which proteins are produced is altered. To study this effect, one should carefully measure rates of protein production and compare those to predictions of the production rates from kinetic models. However, Tan et al. use single measurements after 1 hour of expression to represent the rate during this entire period of time. It is simply impossible to determine rates this way, as ever present fluctuations in delay times would change the apparent rate of protein production. More- over, in cell-free gene expression systems where only ATP and GTP are regenerated, resources are by definition limited, and their depletion will also change the apparent rate of protein production. Finally, it is of crucial importance to experimentally determine the rates of both transcription and translation, as the overall rate of protein production is a function of the rate of mRNA synthesis. We have attempted to reconstruct expression curves using the model we have recently developed to describe the production of both mRNA and proteins in dilute solution and in crowded coacervates [16]. We use the experimental concentrations listed in the Sup- plementary Information of Ref. [5] for plasmid DNA (0.85 nM), ATP (0.26 mM), and GTP, CTP and UTP (all 0.18 mM), and amino acids (presumably <0.5 mM total), and take the transcription and translation binding and catalysis rate constants from Ref. [5]; the latter were determined for a commercial cell-free expression system and were found to agree well with previous experiments [17, 18]. To estimate the effect of crowding on the binding constants and rate constants we use the experimentally determined activity coefficients of crowded protein solutions [19], instead of introducing additional power law scaling parameters with no direct link to chemical processes, like in Ref. [5]. We interpolate the RNA polymerase and ribosome binding constants, and the transcription and translation rate constants, between dilute solutions and crowded coacervates using volume-occupancy as our key variable [19]. Figures 5.6 a and b show our reconstructed mRNA and protein production curves and expression rates (from the maximum slope of the expression curve) for fractions of the total volume occupied by crowders ranging from 3% to 40%. Note that we did not include any degradation of the transcription and translation machinery yet [17], which is probably occurring in the system used by the authors (see Fig. 5.7, where a plateau is reached at the same time with and without RNAse present). We find that at the highest concentrations of crowders the solution is depleted of NTPs within the one hour observation time used by Tan et al., leading to a differential reaction rate that is significantly lower than the maximal protein production rate. Researchers also explored the influence of crowding on gene expression upon tuning the binding between transcription factors and promoter (Fig. 5.3d). We mimic their experiment by taking a 20-fold lower binding constant of RNA polymerase (pT7weak) to the T7 promoter sequence, we see a larger apparent effect of crowding on gene expression, even though the increase in binding and rate constants for crowded solutions is the same 5.2. Results and Discussion 95

Figure 5.6: a) Reconstruction of the mRNA expression curves at crowding volume fractions from 3%-32%. b) Reconstruction of the expression curves for CFP at crowding volume fractions of 3%-32%. c) Relative apparent reaction rate (signal after 60 minutes, divided by signal after 60 min for lowest crowding concentration) as a function of crowding volume fraction, for two choices of the RNA polymerase binding constant at low crowding concentration (0.12 nM−1, blue line and 0.006 nM−1= 20-fold lower, red line). The binding constant increases in the same way with increasing crowding concentration. The relative apparent rate increases more in the case of a lower binding constant, because the production of mRNA after 60 minutes is still very low in the case of a low binding constant

(Fig. 5.6c). It is immediately obvious that taking only one value of the production of ﬂu- orescent protein after a certain time (e.g., 1 hour) is completely misleading and gives rise to false conclusions such as a ‘biphasic response’ to crowding. Figure 5.6c demonstrates that apparent rates might lead one to believe that crowding gives rise to a plateau in protein expression (if we take our measurement after 1 hour) or a biphasic response (if we take our measurement after 2 hours), whereas in reality the solution is simply depleted of resources and much of the mRNA synthesized can no longer be translated into proteins. Going one step further, Tan et al. introduced a negative feedback loop via the production of a T7 lysozyme that binds to T7 RNA polymerase and inhibits its transcription activities (Fig. 5.8a). The introduction of such a negative feedback loop has many different effects as additional resources are utilized in the production of the T7 lysozyme and the effective concentration of T7 RNA polymerase available for the production of CFP is lowered due to the addition of plasmid DNA. The experimental results and model ﬁtting from [5] are presented in Fig. 5.8b. Chapter 5: Understanding the effect of macromolecular crowding on genetic networks 96 in synthetic cellular systems

Figure 5.7: GFP expression in cell-free systems with different [RNAse] (Image taken from [5])

The authors claim that these results follow from enhanced binding between T7 RNAP and DNA compared to T7 RNAP and lysozyme ‘..at a smaller crowding density’. We doubt how this enhanced binding impacts on the expression rates of the fluorescent protein. We therefore adapted our model to include the expression of lysozyme and binding of the lysozyme to T7. We assume that crowding affects all binding constants to the same extent (i.e., if crowding leads to five-fold enhancement of binding between RNAP and T7, it also leads to a five-fold increase in binding constant between lysozyme and T7). We also assume that transcription and translation speeds scale linearly with length of the transcripts [18]. The results of our model calculation are striking (Fig. 5.8c): crowding itself does not lead to any biphasic response or the appearance of a plateau if the available resources are unlimited. Instead, the introduction of the lysozyme leads to a lowered effective CFP production rate at all crowding densities. However, if we take into account that resources are limited, we do see a clear ‘biphasic’ response with increasing crowding, for exactly the same reasons as outlined above. Adding lysozyme plasmids again shifts the curve to lower effective CFP production rates, in line with expectations as an increasing fraction of the available RNA polymerase is bound to the Lysozyme and thus becomes less active [20]. To summarize these results, our model shows that apparent biphasic behavior does not result from crowding ‘shaping’ gene expression, but from the increased production of mRNA and protein leading to rapid depletion of resources. Researchers also performed gene expression in liposomes as an artificial cell model. They claim that ‘... molecular crowding would exert a larger impact on gene expression 5.3. Conclusion 97

Figure 5.8: a) Schematic of interactions between molecular crowding and a negative feedback loop [5]; b) Gene expression rates of a negative feedback loop with increasing crowding densities [5]; c) Comparison of the apparent CFP production rate in a gene network (see (a)) with unlimited and limited resources for transcription and translation. Relative CFP production rates are calculated in the same way as in Fig. 5.4c, following the method of Tan et al. The concentration of CFP plasmids is 0.85 nM in all calculations and the concentration of lysozyme-encoding plasmids is increased from 0 nM to 0.85 nM, as indicated by the labels in larger volumes than in small volumes.’ Without any evidence for such reasoning we can suggest in turn that what also could have happened is the adsorption of components to interfaces. This could lead DNA and/or protein machinery no longer being available for gene expression, thus hindering gene expression in liposomes with higher surface-to- volume ratio.

5.3 Conclusion

We conclude that crowding does affect gene expression, but any bimodality or correlation between liposome size and expression levels result from the depletion of resources. Indeed, the interior of the cell presents a highly crowded environment, and this envi- 98 References ronment leads to increased binding strengths between macromolecules, in turn leading to increased enzymatic activities. It is becoming increasingly clear that in order to understand transcription and translation processes, we must take into account the effect crowding has on these enzymatic reaction networks. This is even more urgent now that synthetic biology aims to provide us with a complete map of all connections and reaction routes in the cell; we must know if these routes are highways or byways.

Acknowledgements Prof. Wilhelm Huck, Prof. German Rivas, Dr. Evan Spruijt, Dr. Begona Monterosso are gratefully acknowledged for fruitful discussions and contribution to this chapter.

References

[1] Elcock, “Models of macromolecular crowding effects and the need for quantitative comparisons with experiment,” Curr. Opin. Struct. Biol., vol. 20, pp. 196–206, 2010. [2] H. X. Zhou, G. Rivas, and A. P. Minton, “Macromolecular crowding and confinement: Biochemical, biophysical, and potential physiological consequences,” Annu. Rev. Biophys, vol. 37, p. 375, 2008. [3] I. Schoen, H. Krammer, and D. Braun, “Hybridization kinetics is different inside cells,” Proc. Natl. Acad. Sci. U. S. A., vol. 106, no. 51, pp. 21649–21654, 2009. [4] Y. Phillip, E. Sherman, G. Haran, and G. Schreiber, “Common Crowding Agents Have Only a Small Effect on Protein-Protein Interactions,” Biophys. J., vol. 97, no. 3, pp. 875–885, 2009. [5] C. Tan, S. Saurabh, M. P. Bruchez, R. Schwartz, and P. LeDuc, “Molecular crowding shapes gene expression in synthetic cellular nanosystems,” Nat. Nanotechnol., vol. 8, no. 8, pp. 602–608, 2013. [6] M. T. Record, E. S. Courtenay, S. Cayley, and H. J. Guttman, “Biophysical compensation mechanisms buffering E-coli protein-nucleic acid interactions against changing environments,” Trends Biochem.Sci., vol. 23, no. 5, pp. 190–194, 1998. [7] A. P. Minton, “The effect of volume occupancy upon the thermodynamic activity of proteins - some biochemical consequences,” Mol. Cell. Biochem., vol. 55, no. 2, pp. 119–140, 1983. [8] A. P. Minton, “Molecular crowding: analysis of effects of high concentrations of inert cosolutes on biochemical equilibria and rates in terms of volume exclusion,” Methods Enzymol., vol. 295, pp. 127– 149, 1998. [9] R. Ellis, “Macromolecular crowding: an important but neglected aspect of the intracellular environment,” Curr. Opin. Struct. Biol., vol. 11, no. 1, pp. 114–119, 2001. [10] M. A. P., “The influence of macromolecular crowding and macromolecular confinement on biochemical reactions in physiological media,” J. Biol. Chem, vol. 276, pp. 10577–10580, 2001. [11] M. Alcorlo, M. Jimenez, A. Ortega, J. M. Hermoso, M. Salas, A. P. Minton, and G. Rivas, “Analytical Ultracentrifugation Studies of Phage phi 29 Protein p6 Binding to DNA,” J. Mol. Biol., vol. 385, no. 5, pp. 1616–1629, 2009. [12] J. R. Wenner and V. A. Bloomfield, “Crowding effects on EcoRV kinetics and binding,” Biophys. J., vol. 77, no. 6, pp. 3234–3241, 1999. [13] J. Batra, K. Xu, S. Qin, and H.-X. Zhou, “Effect of Macromolecular Crowding on Protein Binding Stabil- ity: Modest Stabilization and Significant Biological Consequences,” Biophys. J., vol. 97, no. 3, pp. 906– 911, 2009. References 99

[14] http://www.dextran.net/dextran-physical-properties.html. [15] L. J. Friedman, J. P. Mumm, and J. Gelles, “RNA polymerase approaches its promoter without long-range sliding along DNA,” Proc. Natl. Acad. Sci. U. S. A., vol. 110, no. 24, pp. 9740–9745, 2013. [16] E. Sokolova, E. Spruijt, M. M. K. Hansen, E. Dubuc, J. Groen, V. Chokkalingam, A. Piruska, H. A. Heus, and W. T. S. Huck, “Enhanced transcription rates in membrane-free protocells formed by coacervation of cell lysate,” Proc. Natl. Acad. Sci. U. S. A., vol. 110, no. 29, pp. 11692–11697, 2013. [17] E. Karzbrun, J. Shin, R. H. Bar-Ziv, and V. Noireaux, “Coarse-Grained Dynamics of Protein Synthesis in a Cell-Free System,” Phys. Rev. Lett., vol. 106, no. 4, 2011. [18] T. Stoegbauer, L. Windhager, R. Zimmer, and J. O. Raedler, “Experiment and mathematical modeling of gene expression dynamics in a cell-free system,” Integr. Biol., vol. 4, no. 5, pp. 494–501, 2012. [19] P. D. Ross and A. P. Minton, “Effect of non-aggregating proteins upon the gelation of sickle-cell hemoglobin - model calculations and data-analysis,” Biochem. Biophys. Res. Commun., vol. 88, no. 4, pp. 1308–1314, 1979. [20] N. M. Stano and S. S. Patel, “T7 lysozyme represses T7 RNA polymerase transcription by destabilizing the open complex during initiation,” J. Biol. Chem., vol. 279, no. 16, pp. 16136–16143, 2004.

Chapter 6

Stochastic gene expression in a crowded environment

Abstract Although cells typically contain >200g/L concentrations of macromolecules, individual components might be at very low concentration. Thus, many biochemical reactions in the cell involve low copy numbers and thus are prone to stochastic fluctuations. These fluctuations have been reported to play a critical role in diverse processes including cytoskeletal dynamics, cell polarization, signal transduction and neural activity [1]. Stochastic gene expression is a well-studied example as it is both central to almost all processes in the cell and, owing to the low copy number (usually 1-2) of genes, especially susceptible to noise [2].While this phenomenon has been observed in vivo, no detailed in vitro studies on gene expression noise have been reported. Here, we developed a microfluidic and analytical tool to study gene expression within a model cell population. We used in vitro transcription and translation system to express β-glucuronidase in picoliter droplets to emulate a cell population in which stochastic expression can be observed. We utilized dextran of different molecular weights as a crowder to simulate the in vivo situation of macromolecular crowding. We developed tools to analyze characteristics of our model cell population and showed that there was a profound difference in the noise behavior depending on the molecular weight of dextran used. This study sets the foundations to further investigations into stochastic gene expression in vitro and to draw conclusions about gene expression noise depending on gene copy number and crowding conditions.

101 102 Chapter 6: Stochastic gene expression in a crowded environment

6.1 Introduction

6.1.1 Stochastic gene expression in vivo

Along with macromolecular crowding, stochastic gene expression is an attribute of all living cells. The central idea behind the noise in gene expression is that stochastic events at the level of the single DNA molecule might propagate to the level of protein numbers. The stochastic events of protein production are often difficult to observe due to extremely sensitive and technically challenging measurements required. However, already in 1957 the first experimental evidence that suggested an effect of stochastic gene expression was observed in single cells by Novick and Weiner [3]. They induced β-galactosidase expression in E. coli with isopropyl β-D-1-thiogalactopyranoside (IPTG) and observed an “all or nothing” response to the inducer in single cells. The authors proposed that variation in permease level, a membrane protein increasing IPTG transport into the bacteria and therefore upregulating its own expression, was the cause of this “all or nothing” effect on cell population. Cells that expressed enough of the permease protein would rapidly be induced through a feedback loop. However, such early studies were hindered by the lack of reliable single-cell assays of gene expression. Since then the number of publications studying stochastic gene expression mechanism in numerous processes continuously increased every year. For example, the influence of variation in plasmid copy number on cell to cell variation has been extensively studied (Paulsson and Ehrenberg, 2001) [4]. High and low copy number plasmids were tested by changing the plasmid origin of replication and assaying for luciferase activity in cell lysates. This approach lacked the resolution to examine cellular heterogeneity due to differential partitioning of plasmids. The data indicated tighter control of gene expression using auto-repression, however, due to the number of variables unaccounted for in this study, it was impossible to argue for or against a mechanism of stochastic gene expression. Some of these issues were subsequently addressed by another group in an approach using chromosomally integrated reporters (Ozbudak et al., 2004) [5]. The expression of GFP was measured using B. subtilis strains containing point mutations in either ribosomal binding site, initiation codon or promoter sequences. The measurement device in this case was a fluorescence- activated cell sorter (FACS). It was important to be able to compare the mean expression levels under different experimental conditions. For example, an attempt to control for cellular heterogeneity was made using cell size gating during cell sorting. However, due to different rates of total protein production for each mutant, it was unclear how differences in the ratio of GFP concentration to cell size could be normalized. Unless the increase in cell volume due to growth was perfectly matched by the rate of protein production in each cell, the observed variation in GFP concentration could be attributed to potential dilution of GFP caused by variation in cell size. If 6.1. Introduction 103 production rates were assumed constant then the fluctuations in the system could only be explained by the variations in cell sizes. Shortly after the previous study a more robust setup was employed (Elowitz et al., 2002) [6] to examine stochastic gene expression in a single cell. This paper laid the foundation and defined the terminology to be used in many future investigations of noise. In the experimental setup two distinguishable fluorescent proteins were integrated into bacteria at loci equidistant from the chromosomal origin of replication, both under the control of identical promoters. This provided an internal calibration measure for the variation in expression between individual cells. As the genes were comparable in each aspect such as gene length, promoter strength and distance from the origin of replication, intrinsic as well as extrinsic noise could be observed (Fig. 6.1).

Figure 6.1: Fluorescent image of cell population expressing both YFP and CFP. Stochastic gene expression is visible; some cells mainly express CFP (red), while some mainly express YFP (yellow). From [6]

The authors referred to the difference in ﬂuorescence in a single cell as intrinsic noise, while the difference between cells was termed as extrinsic noise. Under the control of strong constitutively active promoters the cells exhibited little intrinsic noise. The expression of reporters was then examined under the control of a lactose repressor, with varying concentrations of IPTG added to the medium to induce expression. It was found that at low expression levels extrinsic noise was relatively high. The noise passed through a maximum at an intermediate IPTG concentration and was lowest under full expression conditions. The two color method developed by Elowitz et al. was an important advance in the study of stochastic gene expression because factors affecting protein production external to the cell (intracellular heterogeneity), could be separated from factors inside a cell (intercellular heterogeneity). However, conclusions were drawn based on several assumptions. The population of cells was assumed to be isogenic, lacking possible muta- 104 Chapter 6: Stochastic gene expression in a crowded environment tions during cell cycle. It has been shown that E. coli can optimize expression of lactose operon in less than a hundred generations (Dekel and Alon, 2005) [7], which raises questions about the rate of mutation in bacteria and its potential for unexplored effects on gene expression. A perfectly mixed homogeneous environment was also assumed in the modelling equations, while this is not the case (Golding and Cox, 2006) in reality [8]. None of those conditions could be provided, since they would both require a perfect experimental system impossible to establish. Further understanding makes it essential to build an experimental platform that will take into account possible effects of crowding on noise in gene expression. We think that such a platform can only be assembled in vitro.

6.1.2 In vitro system for cell mimics

In the research presented here we expressed β-glucuronidase in picoliter droplets to follow stochastic gene expression in a large population of droplets. β-glucuronidase is an endogenous E.coli enzyme with a molecular size of about 90 kDa that hydrolyzes β- D-glucuronic acid [9, 10]. It is known to be active as a tetramer (Fig. 6.2). While its association mechanism is not fully understood, in vitro studies have indicated a fourth- order dependence on monomer concentration [11]. β-glucuronidase synthesis was found to have two rate-limiting steps: monomer to dimer assembly and dimer to tetramer assembly. However, it is not known if the assembly from monomers to tetramers is coupled with the process of biosynthesis of protein. As the active enzyme is a tetramer we hypothesize that β-glucuronidase would be sensible to changes in excluded volume. Accordingly, added crowders would enhance the effective concentration of β-glucuronidase monomers. It is therefore a good model enzyme to investigate stochastic processes in our model cells, as the effect of additional crowder should be amplified by using β-glucuronidase activity as a readout. An important experimental consideration is that a single molecule of β-glucuronidase can produce a large amount of fluorescent product molecules by hydrolyzing a synthetic fluorogenic substrate, leading to amplification of the fluorescent signal. To investigate the activity of β-glucuronidase in an in vitro system, one has to consider the endogenous presence of β-glucuronidase, which is used for the formation of glucose. Here, we report an adjusted lysate protocol with supplemented glucose in order to lower the endogenous activity of β-glucuronidase. For our studies on stochasticity of gene expression in droplets we used an in-house in vitro transcription and translation (IVTT) system, derived from E.coli cell extract. To achieve a cell-like concentration of macromolecules in droplets synthetic crowders were added to induce depletion. Ideally, a crowder should resemble a hard sphere, it should be non-reactive towards its environment, and it should be available in different macromolecular sizes. Popular crowders include polyethylene glycol (PEG), bovine serum albumin (BSA), dextran and Ficoll. While PEG is available in a wide range of sizes, it is known 6.2. Materials & Methods 105

Figure 6.2: Model of β-glucuronidase formation, taken from [11] to inﬂuence transcription positively and translation negatively (see Chapter 4). BSA, a protein with a size of around 65 kDa, is relatively inert, but not available in wider size ranges. Dextran and Ficoll are both inert, are available in different sizes and can be modelled as a hard sphere (see Chapter 1). While Ficoll resembles a hard sphere to a higher extent than dextran, dextran was chosen for this study because it was available in a more suitable size range (9-70 kDa, which resembles the size of small to large proteins).

6.2 Materials & Methods

6.2.1 IVTT system

For the in vitro translation and transcription system, E.coli BL12 (DE3) cells were grown in 2 L broth (32 g tryptone, 20 g yeast extract, 10 g NaCl, 3 g NaH2PO4 • H2O, 7.8 g ◦ Na2HPO4 • 2H2O and 50 mM ﬁlter-sterilized glucose). Cells were grown at 37 C until an OD600 of 1.8 and collected by centrifugation for 10 min at 4 ◦Cand 5500 rpm. All subsequent steps were performed on ice or in a 4 ◦C cold-room. 106 Chapter 6: Stochastic gene expression in a crowded environment

The periplasmic proteins were removed by cold osmotic shock; the obtained pellets were dissolved in ice-cold 20 % w/v sucrose solution and incubated on for 10 min. After recollection (6800 rpm, 10 min), the pellet was resolved in 4 times pellet weight in volume of ice-cold MilliQ. Cells were collected (6800 rpm, 10 min) and the step repeated with 10 min incubation after dissolving the pellet. After centrifugation, the spheroplasts were dissolved in 1.5 times the pellet weight of ice-cold MilliQ and centrifuged again. Pellets were stored overnight in -80 ◦C freezer. For cell-lysis, the pellets were dissolved in 0.8 times the pellet weight in volume of ice-cold MilliQ and subjected to 10× of the following sonication cycle (10 s sonication at 10 µm amplitude followed by 30 s on ice. The cellular debris was collected by centrifugation (15700 rpm, 30 min). The supernatant was collected and residual debris and DNA were spun down (15700 rpm, 30 min). The supernatant was dialyzed 4 times against dialysis buffer (10 mM Tris, 60 mM potassium glutamate, 15 mM magnesium glutamate, 10 mM amino acid mix, 20 mM 3-PGA, 0.66 mM spermidine, 14 mM magnesium glutamate, 1 mM DTT) for 45 min each. The lysate was ﬂash-frozen with liquid nitrogen and stored at -80 ◦C in aliquots until use. Functional IVTT contained 20 % w/v cell lysate, 1 µL of T7 polymerase, 50 mM HEPES pH 8.0, 90 mM potassium glutamate, 15 mM magnesium glutamate, 10 mM amino acid mix, 20 mM 3-PGA, 0.66 mM spermidine, 1 mM of UTP, CTP and ATP each, 3 mM GTP. For β-glucuronidase expression, DNA encoding β-glucuronidase was added just before the reaction start. Additional crowders were added either from solution (dextran 35 kDa and 70 kDa) or, in the case of dextran 11 kDa, by weighting powder and centrifuging the lysate over the powder (2 min, 1000 rpm, 4 ◦C).

6.2.2 Data acquisition and analysis

To produce droplets we used ﬂow-focusing devices with 20 µm wide junction channel and an oil/aqueous ﬂow rate ratio of 6 (300 µl/h and 50 µl/h respectively) (Fig. 6.3). This allowed us for formation of monodisperse droplets about 30-40 µm diameters in separate experiments.

Figure 6.3: Flow-focusing device, diameter of the channels junction 20 µm 6.2. Materials & Methods 107

Produced droplets were stored in observation chambers for data acquisition. The devices were mounted on the inverted microscope (Olympus IX71) equipped with a motorized stage (Prior, Optiscan II). Fluorescent images were taken with the sensitive EMCCD camera (iXon, Andor) using illumination from the mercury lamp.

To follow the distribution of protein levels over a population of model cells we produced monodisperse droplets of 30 µm diameters and stored them in observation chambers for data acquisition. Typically, in one frame 200-400 droplets could be seen. For analysis, ImageJ (v.148) was used (see Fig. 6.4 for example).

Figure 6.4: Example ﬂuorescent image. Left: ﬂuorescent image of a population of droplets expressing β-glucuronidase. Right: ImageJ analysis of the frame. The “analyze particles” function is used to create a mask identifying droplets of the chosen size and measuring their intensity

To acquire the mean intensity and standard deviation for a population of droplets in each frame, an ImageJ macro was used. First, the relative occupancy of imaging space by droplets was analyzed by taking the last frame of an experiment. This allowed acquisition of the relatively faint droplets at the beginning of the experiment. Each frame was then analyzed by ImageJ by creating a binary mask, separating close droplets by the “wa- tershed” function of ImageJ, of the droplets within an area tolerance limit corresponding to droplets with a diameter of 30-40 µm. The mean intensity of each droplet was measured in the original picture and the standard deviation of the population calculated. This yielded a time versus mean intensity graph for each experiment. Droplet movement leads to the exclusion of some droplets from analysis (<3 % of all droplets). 108 Chapter 6: Stochastic gene expression in a crowded environment

6.3 Results and Discussion

6.3.1 Aim of this study and limitations of approach taken

The research presented here aimed at several targets initially. First was to reveal effects of crowding on expression of the enzyme β-glucuronidase at the broad range of initial DNA concentrations. Second was to shed light on influences of crowding and plasmid copy numbers on noise in gene expression. And third one was to be able to follow expression of the first enzyme molecules using sensitive microfluidics setup. Due to time limits we unfortunately could not meet all of our goals, but developed new detection and analytical tools that could help to reach them. Nevertheless, we think that this study can serve as a useful example for future work on investigating role of crowding in stochasticity of gene expression. To simulate the crowding conditions inside the droplets we have chosen dextran of different molecular weights. A significant limitation that occurred during this study was that we could not use concentrations of crowder much higher than 80 mg/ml due to the high viscosity of the resulting solutions, especially for low molecular weight dextran. Still, we observed profound effects of 80 mg/ml of dextran on the expression of β-glucuronidase (see below).

6.3.2 Inﬂuence of different molecular weights dextran on expression of β -glucuronidase

We studied the influence of different molecular weight dextrans (9-11kDa, 35-45kDa, 70kDa) on β-glucuronidase in droplets. Figure 6.5 shows the mean fluorescein production (which is the product of the catalytic activity of β-glucuronidase) within a population of droplets. In this case about 5000 plasmid copies were present in each droplet. The fluorescein production was measured in the presence of 80 mg/mL dextran of different molecular weights. The expression does not show a plateau due to fluorescence satura- tion of the detection camera. Figure 6.6 shows the mean fluorescein production with and without crowder at a) 200 on average plasmids per droplet, b) 500 on average plasmids per droplet, c) 1000 on average plasmids per droplets and d) 2500 on average plasmids per droplet. It is clear that with increasing DNA concentration the onset of β-glucuronidase activity happens earlier. While for 200 and 500 plasmids per droplet the population without any crowder shows fluorescein production much earlier, the population containing 80 mg/mL dextran 9-11 kDa shows similar activity at 1000 plasmids/droplet. For 5000 and 2500 plasmids per droplet, dextran 9-11 kDa seems to enhance the fluorescein production in comparison to the situation without crowder. It must be noted that for the small (9-11 kDa) dextran the volume fraction was higher than for the others, according to the classical theory of crowding. This might explain the increase in the kinetics for the higher starting concentrations of plasmid (Fig. 6.5, Fig. 6.6 c-d). Dextran 35-45 kDa and 70 kDa always 6.3. Results and Discussion 109 show either similar or lower fluorescein production than without crowder, indicating that these crowding agents could hamper transcription and/or translation reactions. On the other hand, data for lower DNA copy numbers shows slow down for the small dextran. The assumption was made, however not verified, that at higher volume fraction of dextran possible adsorption at the interface of DNA molecules could occur, thus causing decrease in β-glucuronidase expression.

Figure 6.5: Graph showing the mean ﬂuorescein production within a population (250-400) of droplets. The ﬂuorescein production was measured in the presence of 80 mg/ml dextran of different molecular weights. Each droplet contained on average 5000 plasmid copies. (It should be noted, that “on average” means that on a population of 250-400 droplets certain variations of components during droplet production is expected)

6.3.3 Stochastic β -glucuronidase expression in droplets

From the previous section it is clear that depending on the molecular weight of dextran and plasmid copy number one can expect different effects on the kinetics of β- glucuronidase expression. Further investigation with the use of molecular beacon complementary to mRNA, for example, should indicate which step of expression transcription or translation kinetics is affected. To quantify noise, a different framework of ex- tracting data was developed. Since it is difﬁcult in the present system to identify the 110 Chapter 6: Stochastic gene expression in a crowded environment source of noise we decided to apply a descriptive statistics approach on a substantial population of expressing droplets.

Figure 6.6: Graphs showing the mean ﬂuorescein production within a population (250-400) of droplets. The ﬂuorescein production was measured in the present of 80 mg/mL dextran of different molecular weights. Each droplet contained a) 200 on average plasmids per droplet, b) 500 on average plasmids per droplet, c) 1000 on average plasmids per droplets and d) 2500 on average plasmids per droplet

6.3.3.1 Time of arrivals and kinetics distributions

To describe the population of droplets, they were analyzed by choosing two parameters; ‘Time of Arrival’ (ToA) and their kinetics (Kin). ToA indicates the time one individual droplet requires to reach a set threshold of fluorescence intensity, or fluorescein concentration. This threshold was chosen for each set of experimental frames at the onset of activity increase, i.e. after the stationary phase and before the linear phase of fluorescence increase. The population of droplets was then analyzed by how many droplets in 6.3. Results and Discussion 111 each frame reach the threshold, given as relative amount of the population versus the time at which the threshold is reached (fraction of droplets versus ToA). Kin indicates the rate at which the fluorescence increases within an individual droplet, measured by a linear fit of three consecutive frames, indicating a period of 4 minutes in which the fluorescence increase showed to be linear. The starting frame is equivalent with the frame of the ToA analysis in which more than 50 % of the droplets have reached the threshold. For this, a binary mask had to be created manually for the first respective frame and used subsequently as an overlay for the following frames. Essentially, ToA gives us information about the time it takes for the first active enzyme complex to arise in a droplet, and Kin gives an indication of the rate of protein production in a droplet. Figure 6.7 shows ToA for each population. The threshold level was chosen to be 10 µM of fluorescein for each experiment. The droplet population was between 250 and 400 droplets, the graphs show the ToA from the start of the experiment and how many of the droplets newly reached the threshold (relative quantity). The ToA times were fitted using a normal distribution function and the width of the function 2σ was plotted against the plasmid number per droplet (Fig. 6.7 f). With increasing plasmid copy number the distribution of ToAs becomes narrower for all crowding conditions; while the time span in which all droplets pass the threshold is 20 to 60 minutes for 100-250 plasmids/droplet, this time window decreases to 8 minutes for a high DNA concentration (5000 plasmids per droplet). On the other hand, no significant difference between the normal distribution can be seen depending on DNA concentration or crowder condition (Fig. 6.7). This finding would indicate that crowding does not significantly impact on the noise in the system. Figure 6.8 shows the kinetic distribution of the different droplet population for each copy number and crowding condition. The kinetics are measured in gain of fluorescent units per minute (a.u./min) and are plotted against the absolute counts of droplets with a binning of 1 a.u./min. The data was fitted with a normal distribution and the noise in the kinetic distribution (standard deviation over the mean) was plotted against the plasmid copy number (Fig. 6.8 f). Now we observe very large differences in enzyme activity per droplet depending on crowder condition and plasmid copy number. With the exception of dextran 9-11 kDa, the noise increases sharply going from 200 to 500 plasmids per droplet and then decreases again for higher plasmid copy number. While it is expected that the noise decreases with higher copy numbers due to a higher gene content which would “buffer” fluctuations in the stochastic process of transcription, the increase in noise going from 200 to 500 plasmids/droplet and the different behavior of dextran 9-11 kDa is unexpected. Also, the noise obtained with dextran 35-45 kDa is consistently higher compared to all other crowders. At this point, conclusions about the possible source of this noise are hard to draw due to the descriptive nature of the experimental setup, which can only monitor the last step in protein formation: enzyme activity. Possible sources of increasing and 112 Chapter 6: Stochastic gene expression in a crowded environment

Figure 6.7: ToA for all crowding conditions: a) 200 plasmids copy on average per 30 µm droplet, b) 500 plasmids copy on average per 30 µm droplet, c) 1000 plasmids copy on average per 30 µm droplet, d) 2500 plasmids copy on average per 30 µm droplet, e) 5000 plasmids copy on average per 30 µm droplet, f) The width of the normal distribution ﬁt on the range of plasmids concentrations

decreasing noise by crowders can be due to following mechanisms; dextran may enhance or decrease transcription and/or translation rate or have an inﬂuence on β-glucuronidase tetramerization. Inﬂuences due to reaction with the substrate or on the activity of the 6.3. Results and Discussion 113

Figure 6.8: Kinetics of the enzyme for all crowding conditions: a) 200 plasmids copy on average per 30 µm droplet, b) 500 plasmids copy on average per 30 µm droplet, c) 1000 plasmids copy on average per 30 µm droplet, d) 2500 plasmids copy on average per 30 µm droplet, e) 5000 plasmids copy on average per 30 µm droplet, f) Noise in enzyme kinetics depending on crowder and plasmid number per droplet

enzyme itself can be excluded; the substrate is present in vast excess (∼300 µM, while analysis is done after conversion of 10 µM), and the presence of dextran has no effect on β-glucuronidase activity. 114 Chapter 6: Stochastic gene expression in a crowded environment

6.3.3.2 Individual droplet analysis

In order to investigate the expression of β-glucuronidase in individual droplets, an Im- ageJ script was written to link individual droplets, their ﬂuorescence and their size between acquisition frames (see Fig. 6.9 for an example of individual tracks).

Figure 6.9: Individual expressions of β-glucuronidase in the population of 300 droplets

Their individual paths’ were fitted according to the following formula: i f X < A, Y = B · X +C; else Y = B · X + D · (X − A)2, where X is the time in minutes, Y is the fluorescence intensity, A is the time before any protein expression occurs (‘lag time’), B is the background activity of the lysate, C is the background fluorescence and D is the protein expression rate (dE/dt). Hereby it is assumed that the glucuronidase activity is independent of the concentration of the crowder present and that the protein expression rate is constant in the first 30 minutes after onset of expression. Tetramerization kinetics of β-glucuronidase are not taken into account. The data of ∼750 to 10000 copies of DNA per droplet were analyzed, with no crowder and with 80 mg/ml of 9-11 kDa, 35-45 kDa and 70 kDa dextran. The distribution of lag time and dE/dt and their noise were plotted against plasmid copy number. With single-droplet analysis, we find similar tendency regarding noise in the lag time (compare to ToA) and enzyme production rate (first derivative of kinetics). The distribution in the lag time is much better resolved than in the initial analysis due to the use of a fitting function for each individual droplet (Fig. 6.10, 6.11). After an initial increase, the noise drops with increasing plasmid copy number. The noise in enzyme production rate dE/dt increases from low to intermediate copy numbers as in Fig. 6.11, and then 6.3. Results and Discussion 115

Figure 6.10: Plasmid copy number vs. enzyme production rate dE/dt vs. lag time. Top left: 3D plot. Top right: plasmids per droplet vs. lag time. Bottom left: plasmids per droplets vs. dE/dt. Bottom right: lag time vs. dE/dt

Figure 6.11: Noise measured as standard deviation over the mean for lag time and enzyme production rate dE/dt vs plasmids per droplet via single-droplet analysis. Top: noise in lag time. Bottom: noise in enzyme production rate dE/dt 116 Chapter 6: Stochastic gene expression in a crowded environment decreases with increasing DNA concentration. Here the overall noise is much higher for 35-45 kDa than with other crowders. The reason for the increase in noise for the intermediate plasmid concentrations would be an interesting point for further studies; one could speculate that the given plasmid concentration may be optimal regarding transcription and translation resources and would therefore be most prone to stochastic ﬂuctuations [12]. Plotting plasmid copy number versus lag time versus production rate reveals an interesting effect of the crowders used in the experiments. While the enzyme production rates depends linearly on plasmid concentration (Fig. 6.10, bottom left), the lag time decreases much more rapidly in the presence of 9-11 kDa dextran than for 35-45 kDa or no crowder, which show a linear trend as well (Fig. 6.10, top right). 9-11 kDa is the smallest crowder and is therefore expected to have the largest depletion effect on the IVTT which could enhance binding of polymerase or ribosomes to DNA/RNA, thereby reducing the lag time of the system and enhancing protein production rate, which is the case in Fig. 6.10, bottom left. Whether this effect is dependent on the concentration of 9-11 kDa dextran is the subject of ongoing research.

6.4 Conclusions and future perspectives

The three most important observations of this study were the following: ﬁrst, in the presence of dextran 35-45 kDa noise in the enzyme production and kinetics was the largest. Second, the lag time needed to start enzyme production decreases most rapidly with increasing DNA concentration in the presence of 9-11 kDa dextran. Third, an intermediate DNA concentration of 1800 copies of plasmid per droplet yielded the highest noise values for all crowders except dextran 9-11 kDa. The analysis based on averaging upon population of droplets and single-droplet analysis yield comparable results, although the single- droplet analysis is more suitable as the error of measurements becomes smaller due to individual tracking. It will be possible as well to distinguish within one droplet population the small variation in droplet size and their possible implications on noise. Further studies will focus on the inﬂuence of 9-11 kDa dextran regarding the lag time and on the differences between 9-11 kDa and 35-45 kDa dextran.

6.4.1 Microﬂuidics and IVTT system

Quantitative understanding of the gene expression dynamics and noise is one of the key problems in quantitative biology. During these studies we have combined experimental and modelling approaches to answer some unresolved questions related to the source of noise in E. coli gene expression and gene expression dynamics. We have shown the development of an IVTT system in which β-glucuronidase activity can be measured. References 117

We have developed microﬂuidic tools to obtain populations of model cells with varying DNA concentrations and crowder conditions. Parallel automated chip technology was utilized to robustly acquire data of parallel model cell populations. We have developed analytical tools to display stochastic behavior of these populations to investigate changes upon altering experimental parameters.

6.4.2 Implications of noise and crowding

The data obtained here show that low copy numbers favors heterogeneity in ToA, but homogeneity in Kin. It is interesting to try and make an analogy to living systems. For example, cell division depends on multiple factors such as small molecules, which have to be altered enzymatically before a sufficient concentration of product is reached to initi- ate cell division. This can be compared to the ToA analysis of droplet populations, where a given threshold of fluorescent product needs to be reached. If selection pressure in the form of alternating nutrient supply is applied, a heterogeneity is advantageous for the survival of the species; fast-dividing cells may be “caught” in a starvation period and die, while slow(er) proliferating cells survive. If our findings are correct and could be translated to a biological system, one would expect that genes coding for threshold-type enzymes would have a low copy number within the genome. On the other hand, a certain metabolic rate may be sustained at a very similar level throughout a cell population. This can generally applied to house-hold enzymes for metabolism. Here a low copy number would be advantageous for a species as well, if the observed effects of this study translate to the in vivo situation. Hence low copy number may have different stochastic effects if product threshold or kinetic rates are investigated, but future research is required to investigate this link.

Acknowledgements Dr. Aigars Piruska is kindly acknowledged for his help in writing ImageJ scripts. The experimental work in this chapter was carried out together with Jan Pille.

References

[1] A. Eldar and M. B. Elowitz, “Functional roles for noise in genetic circuits,” Nature, vol. 467, pp. 167–173, 2010. [2] M. Kaern, T. C. Elston, W. J. Blake, and J. J. Collins, “Stochasticity in gene expression: from theories to phenotypes,” Nature Rev. Genetics, vol. 6, pp. 451–464, 2005. [3] A. Novick and M. Weiner, “Enzyme induction as an all-or-none phenomenon,” Proc. Natl. Acad. Sci. USA, vol. 43, pp. 553–566, 1957. [4] J. Paulsson and M. Ehrenberg, “Random signal ﬂuctuations can reduce random ﬂuctuations in regulated components of chemical regulatory networks,” Phys. Rev. Lett., vol. 84, pp. 5447–5450, 2000. 118 References

[5] E. M. Ozbudak, M. Thattai, H. N. Lim, B. I. Shraiman, and A. van Oudenaarden, “Multistability in the lactose utilization network of Escherichia coli,” Nature, vol. 427, pp. 737–740, 2004. [6] M. B. Elowitz, A. J. Levine, E. D. Siggia, and P. S. Swain, “Stochastic gene expression in a single cell,” Science, vol. 297, pp. 1183–1186, 2002. [7] E. Dekel and U. Alon, “Optimality and evolutionary tuning of the expression level of a protein,” Nature, vol. 436, pp. 588–592, 2005. [8] I. Golding and E. C. Cox, “Physical nature of bacterial cytoplasm,” Phys. Rev. Lett., vol. 10, p. 098102, 2006. [9] J. D. McCarter and S. G. Withers, “Mechanisms of enzymatic glycoside hydrolysis,” Current opinion in structural biology, vol. 4, pp. 885–892, 1994. [10] B. Henrissat, “A classiﬁcation of glycosyl hydrolases based on amino acid sequence similarities,” The Biochemical journal, vol. 282, pp. 309–316, 1991. [11] T. Matsuura, K. Hosoda, N. Ichihashi, I. Kazuta, and T. Yomo, “Kinetic analysis of beta-galactosidase and beta-glucuronidase tetramerization coupled with protein translation,” The Journal of biological chemistry, vol. 286, pp. 22028–34, 2011. [12] T. M. Ozbudak, E. M. and, I. Kurtser, A. D. Grossman, and A. van Oudenaarden, “Regulation of noise in the expression of a single gene,” Nat. Genet., vol. 31, pp. 69–73, 2002. Summary

In this thesis we describe our research into the influence of macromolecular crowding on biochemical reactions by means of a bottom-up approach. For this purpose we assembled the reaction of interest (gene transcription/translation) in artificial protocell models and created a highly crowded environment. Crowded interior of all living cells is their unavoidable property. In Chapter 1 we summarized physical principles behind the phenomena of macromolecular crowding and explained this phenomenon in the terms of volume exclusion. A study showing the magnitude of the effect of the size and volume fraction of the crowders on volume exclusion was performed solving corresponding equations of state. Several models were taken into account: BMCSL (Boublik-Mansoori-Carnahan-Starling-Leland) model, scaled-particle theory and depletion interactions. Experimental evidence of observed effects of crowding was reviewed. In Chapter 2 microdroplets were shown to act as artificial cells which allows us to study effects of crowding on biochemical reactions. Gene expression was chosen as a reaction of interest to study the effects of volume and concentration on its kinetics. The course of gene expression is explained in details, focusing on in vitro assembly of this reaction in micro-containers and its differences from the situation in vivo. We reviewed several encapsulation methods of gene expression such as entrapment in water-in-oil emulsions, giant vesicles, liposomes and microdroplets. Microdroplets via microfluidics was shown to be a robust, sensitive technique exceeding all the other methods. In this Chapter we have established the key ingredients required for all further experiments in the thesis. We established that cell-free gene expression using commercially available kits provides robust protein production inside microdroplets stabilized by a surfactant. Using laser-induced fluorescence, we can detect protein production from a single copy of DNA in a droplet, but we lose information on the kinetics of the process. Therefore, we have developed bilayer devices where the volume and temperature of in vitro transcription translation (IVTT)-containing droplets can be controlled for prolonged periods of time. Two different droplet volumes can be produced on the same device by splitting the reservoir layer into two areas: one containing saturated NaCl solution, the other

119 120 milliQ water. By filling these devices with monodisperse droplets at 4 ◦C we can prevent initiation of protein production until the droplet volumes are stabilized. These devices are the platform used in the next chapters of this thesis. In Chapter 3 we exploited bilayer microfluidic devices to create a crowded environment inside microdroplets. By controlled evaporation we formed so-called “coacervates” which served as protocell models for further studies. We demonstrated the formation of crowded coacervate (∼200 g/L macromolecular concentration) compartments composed of cell lysate and show that coacervation creates an artificial cell-like environment. The characterization of the coacervates revealed that combination of E. coli cell lysate and PEG has a remarkable effect on the distribution of the biologically active components. We showed that the coacervates represented crowded compartments, which, despite the large changes in buffer and lysate concentrations during droplet shrinking and phase transition, formed a functional transcription and translation compartment as evidenced by the observed production of GFP. In Chapter 4 a mathematic model of cell-free gene expression kinetics based on differential equations for individual kinetics rates was presented. The model assumed a finite pool of resources for transcription and translation to account for the observed cessation of in vitro protein synthesis after a fixed time. In this way cell-free gene expression kinetics could be fitted from the early to the late phase and over several orders of magnitude of DNA concentration. The phase separation of E. coli cell lysate into a dense liquid coacervate containing all macromolecular components enabled a direct comparison between mRNA synthesis in dilute and crowded environments. Coacervation created an artificial cell-like environment in which the rate of mRNA production was increased significantly. Measured transcription rates were fitted by a kinetic model and showed a two orders of magnitude larger binding constant between DNA and T7 RNA polymerase, and five to six times larger rate constant for transcription in crowded environments, strikingly similar to in vivo rates.We demonstrated that crowding significantly enhanced the binding constant of T7 RNA polymerase to DNA and the transcription rate constant, which were direct results of crowding affecting the kinetics of the rate-determining steps of the fundamental machinery of gene expression in the cell. We showed that our experimental platform enables a systematic study into the effects of crowding on key cellular processes such as transcription and translation in membrane-free protocells. In Chapter 5 we revisited the work by Tan et al. ([5], Chapter 5), who investigated the effect of molecular crowding on gene expression and complex gene networks. The researchers found that crowding alters the behaviour of a simple gene regulatory module that contained a negative feedback loop. Finally, Tan et al. constructed an artificial cell from lipid membrane vesicles that encapsulated a synthetic expression system and a genetic construct. The liposomes could express green fluorescent proteins using the genetic construct, and enhanced gene expression was claimed when macromolecular crowding agents were added to the liposomes. This topic is important, as the impact of the phys- 121 ical, crowded environment within the cell on key enzymatic reaction networks, such as transcription and translation, is far from understood. Unfortunately, Tan et al. base their work on a wrong interpretation of the theory of crowding, and draw false conclusions from their experimental work due to a poor data collection method. In the Chapter 5, we present an alternative interpretation of the data gathered by Tan et al. using the kinetic model we described in the previous Chapter 4. In Chapter 6 we investigated the influence of crowding on stochasticity in gene expression. We developed a microfluidic and analytical tool to study gene expression within a model cell population. We used an in vitro transcription and translation system to express β-glucuronidase in picoliter droplets to emulate a cell population in which stochastic expression can be observed. We utilized dextran of different molecular weights as a crowder to simulate the in vivo situation of macromolecular crowding. We developed tools to analyze characteristics of our model cell population and showed that there was a profound difference in the noise behavior depending on the molecular weight of dextran used. This study paves the way for further investigations into stochastic gene expression in vitro and to the effect of gene copy number and crowding conditions on genetic noise.

Samenvatting

In dit proefschrift wordt een onderzoek naar de invloed vancrowding door macromoleculen op biochemische reacties beschreven. Dit onderzoek is gekenmerkt door een bottom- up aanpak: in een kunstmatige, rudimentaire cel (hierna: protocel) hebben we alle in- grediënten voor de transcriptie-translatie van genetische informatie, een van de meest relevante biochemische reacties, samengebracht, en vervolgens hebben we daarin een toestand van crowding gecreëerd. Crowding is een term die aangeeft dat het in een bepaald volume een drukte van be- lang is. De drukte wordt gevormd door alle opgeloste moleculen, groot en klein, samen. Alle levende cellen zijn van binnen crowded. In Hoofdstuk 1 worden allereerst de fysis- che principes die aan crowding ten grondslag liggen samengevat, en wordt het fenomeen crowding beschreven met behulp van het concept uitgesloten volume. We laten zien wat de invloed van de grootte en de volumefractie van de druktemakers (crowders) op het uitgesloten volume is aan de hand van de toestandsvergelijking van de gehele oplossing. Daarvoor hebben we verschillende benaderingen vergeleken: het BMCSL-model (ver- noemd naar Boublik, Mansoori, Carnahan, Starling en Leland), Scaled Particle Theory, en een model gebaseerd op depletie-interacties. Tot slot worden het bestaande experimentele bewijs voor de effecten van crowding besproken. In Hoofdstuk 2 laten we zien dat microdruppels geschikt zijn om kunstmatige cellen van te maken. In de microdruppels kan de invloed van crowding op biochemische reacties op gecontroleerde wijze worden bestudeerd. Zo bestuderen we de invloed van celvolume, en van de concentratie en het aantal aanwezige moleculen op de kinetiek van genexpressie als modelreactie. Eerst wordt het verloop van genexpressie uitgelegd, met een nadruk op de mogelijkheid om deze reactie in vitro te laten verlopen in microcontainers, en de verschillen met dezelfde reactiein vivo. Daarna worden verschillende manieren beschouwd om reacties als genexpressie in microcontainers in te kapselen, zoals bi- jvoorbeeld met emulsies van water in olie, reuzevesicles, liposomen en microdruppels. Uit deze beschouwing volgt dat microdruppels die met microﬂuïdica zijn gemaakt, het gevoeligste en meest robuuste systeem opleveren. Met deze achtergrondkennis worden tot slot de basisingrediënten die voor de experimenten in de rest van dit proefschrift nodig

123 124 zijn, vastgesteld. We laten zien dat genexpressie in microdruppels zonder cellen met behulp van commercieel verkrijgbare pakketten tot een robuuste productie van eiwit leidt. We kunnen de eiwitproductie van een enkel DNA-molecuul in een microdruppel meten met behulp van laser-geïnduceerde fluorescentie, maar we verliezen de informatie over de reactiekinetiek. Om eiwitproductie gedurende langere tijd te kunnen volgen, hebben we een dubbellaags microfluïdische chip ontwikkeld, waarin de grootte en temperatuur van microdruppels constant gehouden kunnen worden. We kunnen populaties van microdruppels met twee verschillende groottes maken in dezelfde chip door de laag met het reservoir in twee helften te splitsen: in de ene helft stroomt een verzadigde zoutoplossing, in de andere MilliQ-water. Het laatste ingrediënt is de controle over de aanvang van de genexpressie. Eiwitexpressie kan worden voorkomen door de gehele chip met behulp van de vloeistof in de reservoirlaag tot 4 ◦C te koelen tijdens de vorming en stabilisatie van monodisperse microdruppels. De in dit hoofdstuk beschreven dubbellaags chips vormen het experimentele platform voor de rest van de hoofdstukken van dit proefschrift. In Hoofdstuk 3 gebruiken we dubbellaags microfluïdische chips om een crowded fase binnenin microdruppels te maken. Door gecontroleerde verdamping ontstaan zoge- naamde coacervaten in de microdruppels. De coacervaten in dit hoofdstuk staan model voor kunstmatige cellen. We laten zien hoe de coacervaten, die onder meer cellysaat en een hoge totaalconcentratie (ongeveer 200 g/L) macromoleculen, door fasescheiding ontstaan, en dat ze in veel opzichten lijken op inhoud van levende cellen. Uit karakter- isatie van de coacervaten blijkt dat de combinatie van cellysaat van E. coli en PEG een bijzonder effect heeft op de verdeling van (re)actieve biomoleculen in de microdruppels. Aldus zijn de coacervaten crowded compartementen, waarin transcriptie en translatie plaats kunnen blijven vinden blijkens de waargenomen productie van GFP, ondanks de grote veranderingen in buffer- en lysaatconcentraties tijdens het krimpen en de fasescheiding van de druppels. In Hoofdstuk 4 presenteren we een mathematisch model om de reactiekinetiek van genexpressie zonder cellen te beschrijven aan de hand van differentiaalvergelijkingen voor elke component afzonderlijk. In het model houden we rekening met de beperkte hoeveelheid voor transcriptie en translatie benodigde bouwstenen om de stop van eiwitproductie na een vaste tijd te helpen verklaren. Met het beschreven model kan de experimenteel gemeten kinetiek van genexpressie worden beschreven vanaf het vroeg- ste begin tot aan het laatste stadium voor DNA-concentraties die meerdere ordes van grootte variëren. De fasescheiding van E. coli cellysaat in geconcentreerde vloeibare coacervaten maakt een rechtstreekse vergelijking mogelijk van de synthese van mRNA in een crowded omgeving enerzijds en een verdunde omgeving anderzijds. De mRNA productie is significant hoger in de crowded coacervaten. Uit fits van de data aan het kinetische model bleek dat de bindingsconstante van DNA en T7 RNA-polymerase twee ordes van grootte hoger is, en dat de transcriptiesnelheid vijf- tot zesmaal hoger is in een crowded omgeving, en opmerkelijk dichtbij de transcriptiesnelheid in vivo ligt. Deze 125 significante verhoging van de bindingsconstante van DNA en T7 RNA-polymerase en de transcriptiesnelheid door crowding zijn een direct gevolg van het feit dat crowding de snelheidsbepalende stappen in de machinerie die bij genexpressie betrokken is, beïn- vloedt. Met het experimentele platform dat wij beschreven hebben is het mogelijk om dergelijke effecten van crowding op fundamentele cellulaire processen als transcriptie en translatie op systematische wijze in membraanloze protocellen te bestuderen. In Hoofdstuk 5 beschouwen we het onderzoek van Tan et al.([5], Hoofdstuk 5), waarin de invloed van crowding op de genexpressie in complexe netwerken van genen werd bestudeerd. De onderzoekers vonden dat crowding het gedrag van een regulerende module met een negatieve terugkoppeling in een genennetwerk verandert. Daarnaast rap- porteerden de onderzoekers een methode om genetische constructen en een synthetisch expressiesysteem in lipide vesicles (liposomen) in te kapselen. In de gevulde liposomen werd GFP geproduceerd, en de genexpressie was volgens de onderzoekers noe- menswaardig hoger wanneer tevens macromoleculaire crowders in de liposomen aan- wezig waren. Het onderwerp van dit onderzoek is belangrijk, omdat het nog altijd verre van duidelijk is hoe de crowded omgeving die zich in levende cellen bevindt de netwerken van enzymatische reacties, zoals transcriptie en translatie, daar beïnvloedt (zie ook Hoofdstuk 1-4). Jammer genoeg baseren Tan et al zich op een onjuiste interpretatie van de theorie van crowding, en trekken zij verkeerde conclusies uit hun experimentele bevindingen als gevolg van ontoereikende metingen. In dit hoofdstuk presenteren we een alternatieve interpretatie van de data van Tan et al, die gebaseerd is op het kinetische model dat we in Hoofdstuk 4 ook hebben gebruikt. In Hoofdstuk 6 ten slotte, onderzoeken we welke invloed crowding heeft op stochasticiteit van genexpressie. Daartoe hebben we hulpmiddelen op het gebied van microfluïdica en data-analyse ontwikkeld, waarmee genexpressie in een populatie van modelcellen kan worden bestudeerd. Met behulp van een in-vitro-transcriptie-translatie-systeem brengen we β-glucurodinase tot expressie in druppels van de orde van grootte van picoliters. De druppels staan model voor een celpopulatie waarin de stochasticiteit van genexpressie kan worden waargenomen. De typische situatie van macromoleculaire crowding in vivo wordt nagebootst met dextraan van verschillend molecuulgewicht als crowder. Met de ontwikkelde analytische hulpmiddelen kunnen we de populatiestatistiek van de modelcellen karakteriseren. Hieruit volgt dat er belangrijke verschillen zijn in de ruis van genexpressie voor de verschillende molecuulgewichten van dextraan. Deze studie vormt een belangrijke basis voor verder onderzoek naar de stochasticiteit van genexpressie in vitro en de invloed van het aantal aanwezige genen en de crowding op genetische ruis.

Dankwoord

Today is the 28th of February 2015, which means that I have spent 5 years (!) in the Netherlands. Now looking back, I recall that time with warmth and smiles, although it took me a while to get used to Nederlandse wisselvallig weer ..Days summed up in years and steadily I felt in love with Nijmegen and Radboud University., I can call myself lucky, because this place changed my life into a completely new exciting turn! Do I beleive in fate afterall?!.. Now it is a perfect moment to thank all people who supported and helped me during this period. In a first place, I would like to thank professor Wilhelm Huck for giving me the opportunity to conduct my PhD research in his group. Thank you for your guidance, continuous encouragement, trust on me and patience! You taught me to see an opportunity in every challenge I faced and “onmogelijk is mogelijk” attitude to life. You were always ready to listen and help. You told me that PhD is not only about research and working in scientific environment, but it is also a way to learn who you are. Now, I can say that I have learnt a lot and achieved much more than I expected. Thank you for that and the lesson of trust you gave to me! I would like to thank Jung-uk Shim, Xin Liu, Wolfgang Bauer, Shaohua Ma and all the members of Wilhelm’s group in Cambridge in 2010 for introducing me to the field of microfluidics technology and teaching me how to make IVTT microdroplets. It was a beautiful spring I visited Cambridge and you enriched it with a warm welcome in your group! This manuscript wouldn’t be in its shape without contribution of Evan Spruijt. I am very grateful to you for your help, fruitful discussion and sharing your knowledge and expertise with me! Thank you for your work on PNAS and Nature Nanotech. papers and accelerating my project. Thank you for stepping in it just in time! Thank you for correcting the chapters of this thesis and translating the summary! I would like to thank Begoña Monterroso for reading the chapters of this manuscript and useful comments. It was a pleasure for me to meet and talk to you. Thank you for your support and kind attitude to me! I would like to thank manuscript committee for accepting to be the part of it and con-

127 128 structive criticism. Thank you: Jan van Hest, Henk Lekkerkerker and Vincent Noireaux. Starting in a newly-formed group was quite a challenge and I want to thank people who shared that challenge with me: Sonia, Aigars, Venkat and Sergey. Thank you for helping to get settled and starting the working lab! Aigars- thanks for your humor and readiness to help with all hardware issues! Venkat- thank you for your always cheerful mood, willingness to help and of course introducing me to samosas! Sonia, my plan- ning and organizational skills are still not perfect, and it is something, I tried to learn from you - thanks for that! Sergey, thank you for being a true friend, for our laughters and jokes, for your patience listening to me when something didn’t work and of course for the stock of chocolate! I would like to thank our, “IVTT sub-group” -Maike, Emilien and Joost for their hard work on home-made IVTT kit and converting me from just a physicist into a physicist with more affinity to biology Thanks to all colleagues of the Physical Organic Chem- istry group for your help, scientific, discussions, funny lunch conversations and creating friendly and kind atmosphere in the group! Thanks to Julian, Albert, Ilya, Sjoerd (hartelijk bedankt voor vertaling!), Esra, Rao, Marlies, David, Stephanie, Agata and Yujie! I would like to thank my student Jan Pille for the work we did together on β- glucuronidase expression in droplets. Thanks for your hard work, optimistic attitude and your support during DSM event. Thanks to Desiree van der Wey, Peter van Dijk, Theo Peters, Jan Dommerholt, Marieke Egbertzen and all the members of the Cluster of Molecular Chemistry for helping me in numbers of issues I encountered. I want to thank Denis, my husband, for being with me at all times. This thesis would not be ever completed without your dedication, your Tex expertise (I can be honest here ) and your passion for 3D-drawing! Thank you for painting my life in beautiful and, joyful colors! Thank you for making me happier and better person. Ëþáëþ òåáÿ! Thanks to my family-in-law! Thanks to my grandfather-in-law Febus Mincer- one of the most remarkable personalities I had a chance to know. It was an honour for me to meet you! Thanks to Nina and Danis, Dina, Sasha and Vika, Alexander and Ira, Anya, Pasha and Alenka! Thanks to my uncle Igor, aunt Nataliya, and my cousins Kate and Stepan! I would like to thank my brother Dima for being an inspiring example to me as a person and as a scientist. Thank you for helping me in difficult decision moments which I faced during my studies and work in Netherlands. You were a pioneer in our family to travel abroad and succeed there by only means of your talent and hard work. Thanks to your wife Sasha and your adorable daughters Capitolina and Vasilisa! Òû ñàìûé ëó÷øèé áðàò íà ñâåòå! Ñïàñèáî òåáå áîëüøîå! I want to thank my parents Tatyana and Aleksandr for their love, care and commit- ment to each other and our family! Dad, we don’t see you anymore, but I feel your invisible support and I think of you every single day. Ìàìà, ïàïà, ÿ Âàñ î÷åíü ëþá- ëþ! Ñïàñèáî Âàì çà Âñ¼! 129

Finally I thank my guardian angel for keeping me away from wrong decisions and giving me the power to persevere.

28.02.2015