|Home | About | Journals | Submit | Contact Us | Français|
DNA is subject to large deformations in a wide range of biological processes. Two key examples illustrate how such deformations influence the readout of the genetic information: the sequestering of eukaryotic genes by nucleosomes and DNA looping in transcriptional regulation in both prokaryotes and eukaryotes. These kinds of regulatory problems are now becoming amenable to systematic quantitative dissection with a powerful dialogue between theory and experiment. Here, we use a single-molecule experiment in conjunction with a statistical mechanical model to test quantitative predictions for the behavior of DNA looping at short length scales and to determine how DNA sequence affects looping at these lengths. We calculate and measure how such looping depends upon four key biological parameters: the strength of the transcription factor binding sites, the concentration of the transcription factor, and the length and sequence of the DNA loop. Our studies lead to the surprising insight that sequences that are thought to be especially favorable for nucleosome formation because of high flexibility lead to no systematically detectable effect of sequence on looping, and begin to provide a picture of the distinctions between the short length scale mechanics of nucleosome formation and looping.
In its role as the chief informational molecule of the living world, DNA is subjected to a wide variety of physical manipulations. Examples include the looping events that occur during DNA replication (1,2), bending of DNA during recombination (1,2), the bending and twisting induced by a variety of different architectural proteins such as IHF, H-NS and HU in bacteria (3), the bending induced by the histones responsible for packing the genetic material in eukaryotes (4,5) and the physical rearrangements of genomic DNA induced by transcription factors (1,2,4,6). In fact, one of the most ubiquitous classes of regulatory architecture found in all domains of life is often referred to as ‘biological action at a distance’ where transcription factors bind several sites on the DNA simultaneously, thus looping the intervening DNA (7–9).
Interestingly, many of the biological manipulations experienced by DNA, but especially many cases of ‘action at a distance’ in transcriptional regulation, involve bending and twisting the DNA on length scales that are short in comparison with its natural scale of deformation, that is, the persistence length (6). Eukaryotic DNA is subjected to enormous deformations when packed in nucleosomes, with 147bp of DNA (already smaller than the persistence length) wrapped one and three-quarters times around the histone octamer (4,5). Similarly, in the context of prokaryotic transcription factor-mediated DNA looping, not only are such lengths the default in naturally occurring transcriptional networks, but the optimal in vivo lengths as determined by the maximal regulatory effect are often at loop lengths smaller than 100bp (6,10). Despite the clear importance of the short-length-scale mechanical properties of DNA, however, there remains both uncertainty and controversy about the ease with which such short DNAs can be deformed, and also about the role of sequence at these short scales, particularly in the context of protein-mediated bending [reviewed recently in (11,12)].
Here, we exploit insights about DNA flexibility garnered from one class of genetic regulation where it has been studied extensively, that of nucleosome formation, to make predictions about how a different class of mechanical deformations in regulatory biology, that of DNA looping by a transcription factor, will be altered by these same sequences. We test these predictions experimentally with a single-molecule assay in conjunction with ideas from statistical mechanics for the case of one of the most well-known transcriptional regulators in bacteria, that of the Lac repressor, though there are clear implications for other prokaryotic and eukaryotic regulatory motifs as well.
As shown schematically in Figure 1, we have combined tethered particle motion (TPM), in which the Brownian motion of a reporter bead is the readout of the state of its DNA ‘leash’ (13,14), with a statistical mechanical model and the systematic variation of four biologically relevant parameters. The most important of these parameters for the purpose of this study is the flexibility of the DNA in the loop, which is captured in a parameter called the looping J-factor. The looping J-factor is analogous to the cyclization J-factor obtained in the ligation-mediated cyclization assays that are commonly used to measure DNA flexibility at short lengths, and can be thought of as the effective concentration of one end of the loop in the vicinity of the other (15,16), providing a measure of the energetics of bending the DNA into the loop. The approach we have developed here allows us to measure these looping J-factors in a way that provides quantitative insights into how each of the four biologically important parameters we tested affects DNA looping and permits us to contrast the role of sequence in DNA cyclization and nucleosome formation with that of looping. We find that two sequences with significantly different propensities for forming DNA minicircles in in vitro cyclization assays or for forming nucleosomes create a more complicated sequence dependence in the context of DNA loop formation than has been previously appreciated.
A key tool for making the measurements presented here is the concentration titration (see Figure 2): by tuning the repressor concentration and measuring the looping probability, we can fit for other parameters that affect looping probability, namely the operator dissociation constants (Kd’s) and more importantly the looping J-factors for different DNA sequences and lengths. Intuitively, at low protein concentrations, the probability of forming a loop is small. Similarly, at high concentrations, the looping probability is low because the two operators are each occupied by separate transcription factors. At intermediate concentrations, the looping state has its highest probability. These intuitions can be captured mathematically by statistical mechanical models that take into account all of the different ways that the operators can be decorated with repressors. These models make very strict predictions about the functional form of the looping probability curves as a function of the various biological parameters.
Our model states that if the operators have dissociation constants Ki and Kii, and the intervening DNA has looping J-factor Jloop, the looping probability ploop will be
where [R] is Lac repressor concentration. [See (17) and Section S1 in the Supplementary Data for derivation and details.] Although this model was first derived in our earlier work in (17), as a result of the fact that we here explore the analytic consequences of this model, we consider the results presented here to be the first rigorous and successful test of its applicability to DNA looping experiments and its robustness under numerous experimental variations.
In Equation 1 Jloop is the sum of the J-factors for each of the four possible loop configurations that have different DNA-binding orientations, as well as for any additional loop conformations arising from protein flexibility (diagrammed in the legend of Figure 4). The J-factor depends on the length, phasing and flexibility of the DNA, as well as the precise shape of the looped complex (18–20). In fact, we observe two looped states in almost all of our DNA constructs (see Figure 3B and E), as have other studies with Lac repressor (17,21–26). Modifications to Equation 1 that account for these multiple looped states, as well as for experimental issues which may affect the Kd's and J-factors we report, such as the tetramer-to-dimer dissociation at low repressor concentrations, are discussed in Section S1 in the Supplementary Data. However, Equation 1 is the main workhorse of the paper since we found it to be sufficient to account for the data presented here. Similarly, in Section S5, we note a number of experimental controls that were performed to ensure that the parameters we fit to this model were not affected by the effects of the reporter bead size on loop formation, the large amount of surface area in the TPM sample chamber which could cause a difference between the pipetted and actual concentrations of repressor, or the particular repressor batch used in these experiments.
As discussed in Section S5 in the Supplementary Data, we obtained reproducible TPM results only with Lac repressor purified in-house. We used a protocol modified from one received from Kathy Matthews in May 2009, essentially described in (27). The Escherichia coli lacI− BLIM cells and pJCI plasmid used for the purification were kind gifts from the Matthews lab. After elution from the phosphocellulose column, our protein was found to have a concentration between 1 and 2 mg/ml, using a monomer extinction coefficient of 0.6 (mg/ml)−1cm−1 (28), and was ≥99% pure by SDS-PAGE. In one case, some repressor was also purified over a Superdex 200 10/300 GL size-exclusion column (GE Healthcare) using an AKTA system and eluted as a single peak at a molecular weight corresponding to the expected weight of a LacI tetramer.
Plasmids pZS25′ Oid-E/T(89–116)-O1−45-YFP, where ‘E/T(89–116)’ indicates that the sequence of the loop is either from the random E8 sequence or the 601TA sequence from (29) and has a length of 89–116 bp, were constructed by site-directed mutagenesis as described in (17). Jonathan Widom kindly provided the E8 and TA sequences used in (29), which are a subset of those studied here and from which the other E8 and TA lengths were derived. The operator and loop sequences used in this work can be found in Section S3 in the Supplementary Data; schematics of the constructs without the lacUV5 promoter are shown in Figure 1B. QuikChange site-directed mutagenesis (Agilent Technologies) was used to make the operator changes Oid to O1 and Oid to O2, additional loop lengths, and the promoter-containing constructs. Linear labeled DNAs used in tethering assays were created by polymerase chain reaction (PCR) with primers labeled at the 5′ ends with digoxigenin (forward primers) or biotin (reverse primers) (Eurofins MWG Operon); a PCR of the pZS25 plasmid resulted in ~450bp tethers. Primer sequences can be found in Table 3 of (17). See Figure 1B for flanking DNA lengths for the no-promoter PCR products; the promoter-containing constructs of Figure 3D–F are identical to the no-promoter constructs shown in Figure 1B, except that the O1 operator closest to the bead was replaced by O2, 36bp of the loop closest to this O2 operator were replaced by the lacUV5 promoter sequence, and the length of the flanking DNA between O2 and the bead was 139bp rather than 172 bp.
Our TPM protocol was essentially that of (17), with the following modifications:
We first explore how the Lac repressor concentration and its affinity for several known binding sites alter the looping probability, and how these alterations may be used to extract the looping J-factor of the DNA, as well as the repressor–operator dissociation constants. Looping by the Lac repressor has been studied by TPM (17,24–26,30–32), as well as by other single-molecule techniques such as Förster resonance energy transfer (FRET) (21–23), but in all cases only one or a couple loop lengths, operators and repressor concentrations were studied. In many cases, therefore, the repressor–operator dissociation constants were assumed (as opposed to measured) in order for a looping J-factor to be calculated. Here, we describe a new way of measuring both the operator dissociation constants and the relative flexibilities of different DNA sequences as contained in the looping J-factor, by tuning both repressor concentration and operator strengths, with a rigorous comparison between these experiments and theory. We find that the most accurate and logically consistent way of measuring both the J-factors and operator dissociation constants involves a global fit of our model to multiple data sets with different combinations of operators simultaneously.
As described in the ‘Materials and Methods’ section, we can use the tools of statistical mechanics to relate J-factors, operator dissociation constants and transcription factor concentrations to the experimentally observable looping probability through the expression in Equation 1. The main workhorse of our approach to test this statistical mechanical description of looping probability is the repressor concentration curve, where we measure this probability at different repressor concentrations, and then fit Equation 1 to obtain dissociation constants (Kd’s) and J-factors. Equation 1 makes very specific and falsifiable predictions for how these repressor concentration curves should change as the model parameters change. Figure 2 shows a suite of previously untested predictions based upon this statistical mechanical model (as well as the comparison of these predictions to experiment). We consider first the effect of changing the affinity of the repressor for its operators, and in the next section we consider the effect of changing the J-factor.
Figure 2A shows the prediction of our model for how the concentration curves should change as the dissociation constant for one of the operators is varied: changing the strength of one of the operators should change both the concentration at which looping is maximal, and the amount of looping at that maximum, but the curves should overlap at high repressor concentrations. These observations can be formalized by appealing to Equation 1. The concentration at the maximum in the looping probability can be found by differentiating Equation 1 with respect to [R] and results in
Note that the concentration at which the looping probability is maximized does not depend upon the DNA flexibility as captured in the parameter Jloop. The looping probability at this maximum, however, does depend on Jloop, according to
and will, therefore, be discussed in more detail in the next section where our measurements of the J-factors of two different sequences are directly addressed. Finally, we note that at high concentrations, Equation 1 approaches the limit Jloop/(2[R]), which is independent of operator strength, explaining why the curves in Figure 2A overlap at high concentrations. As an experimental consequence, data at low concentrations are essential for determining operator strengths, whereas high concentration data are sufficient for determining J-factors.
Figure 2D shows experimental results for a loop containing 94bp of a synthetic random sequence called E8, described previously (29,33), flanked by three different combinations of the operators Oid, O1 and O2, which are known to have distinct affinities for the Lac repressor. As predicted by our model, increasing the binding strength of one of the operators (i.e. decreasing the value of one Kd) shifts the maximum of the curve to the left and increases its amplitude: that is, stronger operators allow more looping at lower concentrations. Similarly, since the J-factor is a property of the DNA loop length and sequence, we would expect all three curves to be fit by the same J-factor, and for the fits to reflect the reality that they share O1 as one of the operators. This is indeed what we find, as shown in the fit parameters listed in Table 1: fits to the individual data sets (dashed lines in Figure 2D) and a global fit to all three data sets simultaneously (solid lines), where we have enforced the constraint that all three data sets share the same J-factor and dissociation constant of the O1 operator, are comparable in their fidelity. We find that the fitted values for the Kd's agree well with values in the literature obtained through bulk biochemical techniques (see references cited in Table 1), as well as for the most part agreeing between individual fits to different data sets; and that the fitted J-factor also agrees well between data sets, with a value of 300±20 pM. We are, therefore, confident that this combined concentration titration plus statistical mechanical model approach provides us with reasonable parameter values for both dissociation constants and J-factors, and that the global fit supplies the most reliable parameter estimates.
The looping J-factor for E894 is higher than the corresponding cyclization J-factor of 54 pM reported in earlier work (29), and significantly higher than cyclization J-factors for other sequences of similar lengths (34). However, since the looped geometry imposes less stringent constraints on the DNA than does cyclization (discussed in more detail below), we would expect the looping J-factor to be larger than the cyclization J-factor.
Though the role of DNA sequence has not been extensively studied in the particular case of transcription factor-mediated looping, it has become a key parameter in the discussion of a different mechanism of transcriptional regulation, that of nucleosome positioning in eukaryotes (35). A number of sequences with very different nucleosome affinities have been identified, some isolated from natural sources and others from nucleosome affinity assays with synthetic sequences (35). It has been argued for both classes that their nucleosomal affinities stem from different intrinsic flexibilities, and not in response to some other in vivo condition or to a property specific to nucleosome binding, which in turn has led not only to many theoretical and experimental studies on the relationship between sequence and flexibility (11,12,36,37), but also to the determination of certain sequences that are claimed to be highly flexible. For example, Cloutier and Widom characterized a sequence, 601TA, which has a significantly higher affinity for nucleosomes and a J-factor for cyclization 5–30 times greater than the random E8 sequence described in the previous section, depending on the phasing discussed in the next section (29,33,38). If 601TA and E8 differ in mechanical bendability in some general sense, then 601TA should increase looping by a bacterial transcription factor just as it increases nucleosome binding and cyclizes more readily than E8.
As derived in Equations 2 and 3 and shown graphically in Figure 2B, if the 601TA and E8 sequences have different J-factors, then the concentration at which looping is maximal should be the same for both sequences, but looping should increase at all concentrations with the more flexible sequence. This is indeed what we find experimentally in Figure 2E, which shows results for the looping probability as a function of repressor concentration for a loop with 94bp of a sequence derived from 601TA (henceforth abbreviated to ‘TA’), flanked by the Oid and O1 operators. In analogy with the case of different operators discussed in the previous section, the agreement between the individual fit to the TA data (red dashed line) and the global fit to both the E8 and TA data (solid lines) demonstrates that the two data sets can be fit by the same operator dissociation constants but different J-factors (Table 1). The outcome of this measurement is a looping J-factor of 4.2±0.6nM for the TA sequence, about 10 times higher than the random E8 sequence. This is again higher than the cyclization J-factors in (29) and (34) in terms of absolute magnitude, and significantly so: if we use Equation 3 and the cyclization J-factors of (29) to predict maximal looping probabilities, we would expect the maximal looping probability for Oid-E894-O1 to be 0.25±0.3 (compared to the experimentally observed 0.62±0.01), for Oid-TA94-O1 to be 0.87±0.2 (compared to 0.95±0.01) and the O2-E894-O1 construct to show essentially no looping at all. The looping J-factor we measure for the TA sequence is not, however, as much higher than E8 as the 30-fold difference measured in cyclization (29), hinting that the constraints imposed on the DNA in cyclization versus loop formation may lead to a different dependence on sequence, as indeed we find below.
One of the signatures of looping by transcription factors both in vitro and in vivo is a significant modulation of transcription factor activity as the distance between the transcription factor binding sites is varied (2,10,39,40). A similar phasing effect has been observed in cyclization data with the E8 and TA sequences (29). Our experiments, in conjunction with our model that allows us to extract J-factors, permit us to explore this phasing behavior for both of the sequences discussed in the previous section and to compare with several recent theoretical predictions of the looping J-factor.
In the spirit of the kinds of theoretical predictions of Figure 2B, we can use the cyclization results of (29), which looked at the differences between E8 and TA across multiple DNA lengths, to make a naïve prediction of how we would expect the sequence dependence to looping shown in Figure 2E to manifest as the loop length is changed. Such a prediction is shown as a red hatched region in Figure 3A. However, as shown in that figure, to our surprise our experimental results for the looping probabilities for the two sequences, at a constant repressor concentration of 100 pM, show no sequence dependence to looping, with the exception of one or two lengths around the length shown in Figure 2B. The modulation of looping due to phasing is observed in both the E8- and TA-containing sequences, and, with the exception of the 94bp loop length, it appears that this phasing is the same for both sequences. Yet again, surprisingly, not only does the nucleosome positioning sequence not fall within the hatched predicted region, in fact the nucleosome positioning sequence has comparable or smaller looping probabilities compared with the random sequence at most loop lengths.
Even more surprising is that a difference in loopability between the E8 and TA sequences can be restored when the last 36bp of the loop is replaced with the bacterial lacUV5 promoter sequence, as shown in Figure 3D. We were motivated to make this change since in parallel work we have measured how this sequence-dependent looping affects gene expression in vivo and the presence of the promoter is a natural part of the full regulatory network. Though these loops contain 36bp of the loop that are identical between the E8 and TA constructs, the TA-containing DNAs now loop more than the E8-containing DNAs and at some lengths are even as much more flexible than the E8-containing DNAs as predicted based on cyclization assays, as shown by the red hatched region in Figure 3D. Interestingly, the J-factors for the E8 sequence with and without the promoter are comparable—that is, the inclusion of the promoter increases the flexibility of the TA-containing loops only (Figure 3F).
Before discussing the implications of these complex sequence dependencies, we note several additional features of these length data in light of recent theoretical works on the length dependence of Lac repressor-mediated looping, which are plotted in Figure 4. We and others observe two looped states with any pair of operators, which have been hypothesized to arise from the four distinct topological states of the looped DNA and/or several distinct repressor conformations schematized in the legend of Figure 4 (see also the ‘Materials and Methods’ section) (17,18,21–26,41). Regardless of their underlying molecular origins, in Figure 2F we show that the two looped states we observe can be modeled as differing only in effective J-factor; so in Figure 4, we compare the recent theoretical works plotted there with our experimental looping J-factors, but we do so for the two looped states separately, as each of the theoretical results make assumptions about the loop conformation that surely must differ between the two looped states we observe. As can be seen in that figure, different assumptions about the loop and protein geometry, and potential protein flexibility, lead to orders of magnitude differences in the predicted J-factors, reflecting our current uncertainty about the structure of the loop. Moreover, no single theoretical work captures both the magnitude and the phasing of our experimental J-factors, suggesting that none of the theories accurately represents the loop structure yet.
We caution the reader, however, that a detailed direct comparison between these theoretical predictions and with our data may not be possible for several reasons: (i) assumptions about experimental conditions such as salt concentrations differ between references and from the conditions in this work, (ii) it is possible, as argued in (17,18), that the experimentally observed states correspond to superpositions of two or more theoretically predicted states for different loop topologies and/or repressor conformations and (iii) as suggested by FRET data (22), TPM with cross-linked repressor (26) and molecular dynamics simulations (42), the protein conformation in both states may involve some degree of rearrangement relative to the V-like conformation observed in the crystal structure [at the least, rotation of the DNA binding domains, as in (42)]. In these cases, our data would not align with any single theoretical curve. However, we do make some general observations below and in Section S2 in the Supplementary Data.
We find experimentally that the J-factors for the two states have opposite phasings, at least without the promoter, as shown in Figure 3B, and this phasing does not change between sequences except near 94 bp. Such out-of-phase behavior for two different loop structures has been observed for other DNA looping proteins (43), and has been used to explain key features of in vivo repression data (44). However, it is not captured by all of the theoretical models in Figure 4 [e.g. the ‘va’ and ‘e’ states of (19)]. Intriguingly, the promoter changes the relative probabilities of the two looped states: as shown in Figure 3E, the promoter-containing constructs result almost exclusively in the middle state, whereas without the promoter, the two looped states alternate in prevalence (Figure 3B). As these measurements represent the first single-molecule study on the phasing of these two looped states at single base pair resolution, over two helical periods of DNA, at the short loop lengths where the models in Figure 4 show the most pronounced differences in J-factors due to repressor and loop conformations, we hope that our data will help shed light on the molecular origins of the two looped states.
We have shown here that the looping J-factors for 94bp of a random sequence and a nucleosome positioning sequence differ by an order of magnitude, with the nucleosome positioning sequence being more flexible than the random sequence, as expected based on previous cyclization and nucleosome formation assays. To our surprise, however, this sequence dependence occurs only at 94 bp, unless a bacterial promoter sequence is added to the loop, in which case a consistent length-independent sequence dependence is restored.
It is clear that data on more sequences are needed before any definitive conclusions can be drawn. However, we present here one possible hypothesis that we have considered: that the sequence-dependent free energy of bending a DNA depends more strongly than has been previously appreciated upon the specific details of how the DNA double helix is deformed when forming loops versus nucleosomes versus DNA circles. Drew and Travers (45) argued that a DNA minicircle formed by cyclization shares structural similarities with the DNA wrapped around a histone octamer, explaining the usefulness of cyclization assays for understanding the sequence preferences of nucleosome formation. Cyclization has often been cited as a model by which to understand looping as well (8,33,37,46). However, as diagrammed in Figure 4, for DNA loop formation by the Lac repressor, there are multiple looped configurations allowed for a given loop length, most of which are probably quite far from circular as a result of the distinct boundary conditions imposed by repressor binding, and which should have large effects on the associated looping J-factor. We argue that although DNA cyclization may share characteristics with DNA looping such as length-dependent phasing, it apparently does not share other characteristics such as trends in sequence-dependent flexibility, possibly because of this difference in boundary conditions.
We also suspect that the strong sequence dependence at 94bp without the promoter, and with the promoter at all lengths, is due to a change in the preferred loop conformations of these constructs, compared with the majority of the no-promoter constructs. Indeed, the change in the predominant looped state (the ‘bottom’ and ‘middle’ states alternating without the promoter, but the ‘middle’ state predominating at all lengths with the promoter) supports this hypothesis that the promoter alters the preferred conformation of the loop. Such a change in the preferred loop conformation could arise, for example, because of an intrinsic curvature to the lacUV5 promoter sequence. To further unravel these subtleties, we believe a high-throughput approach that makes it possible to look at many sequences might be necessary. We also hope that additional theoretical analyses, perhaps involving the observed tether lengths of the looped state with and without the promoter given in Section S6 in the Supplementary Data, may shed further light on the conformations of looping for these different sequences.
As discussed in the Introduction, the mechanics of loop formation at these short loop lengths that are so prevalent in cellular processes is a subject of much debate, regardless of their sequences (6,11). However, the question of how flexible we expect short DNAs to be is more complicated to answer in the case of protein-mediated DNA looping than in the case of cyclization. As shown in Figure 4, varying the boundary conditions of the loop or the assumed protein flexibility can lead to enormous differences in predicted looping J-factors. Some of these predicted J-factors, using canonical assumptions about DNA flexibility, and without invoking anharmonic elasticity, are in fact consistent with the J-factors we measure, so perhaps it should not be surprising that short transcription factor-mediated loops can form readily in vitro.
Transcription factor-mediated loops are a common motif in both prokaryotic and eukaryotic gene regulation. Here, we have presented a combined single molecule plus modeling approach that allows us to explore how such looping is influenced by four distinct, tunable biological parameters: transcription factor binding site strength, transcription factor concentration, DNA loop length and DNA loop sequence. We have demonstrated that this approach explains how the looping probability depends upon the strength of the operator dissociation constants and that our measured Kd's agree well with values previously obtained by bulk biochemical methods. Further, our model accounts well both quantitatively and qualitatively for the effects of varying the loop flexibility, as well as for details of our single molecule looping experiments such as the presence of two looped states. Our method provides a way of measuring J-factors that is orthogonal to, and therefore complementary to, current methods in use, which we argue has led to important new insights into the role of sequence in DNA flexibility. In particular, we have argued here that the sequence-dependent free energy of bending a DNA must depend more strongly than has been previously appreciated upon the specific details of how the DNA double helix is deformed when forming loops versus nucleosomes versus DNA circles. It is not the case that the TA sequence can be claimed to be more flexible in some general sense, nor can cyclization assays be used to determine DNA flexibility for all biological contexts, as we have shown here that loop formation does not necessarily follow the same sequence rules as cyclization. Measurements of looping J-factors with many more sequences, and further theoretical explorations of the possible effects of sequence on these looping J-factors, will be necessary to understand the initial results presented here. Continuing decades of work on the sequence-dependent mechanics of DNA, the influence of sequence on DNA looping by transcription factors now demands the same kind of scrutiny that has already been given to nucleosome formation.
Supplementary Data are available at NAR Online: Supplementary Tables 1–3, Supplementary Figures 1–12, Supplementary Methods and Supplementary References [52–59].
National Institutes of Health (NIH) [DP1 OD000217A (Director's Pioneer Award), R01 GM085286, R01 GM085286-01S1, and 1 U54 CA143869 (Northwestern PSOC Center)]; National Science Foundation through a graduate fellowship (to S.J.); Wenner-Gren foundation (to M.L.); Fondation Pierre Gilles de Gennes (to R.P.); and the foundations of the Royal Swedish Academy of Sciences (to M.L.). Funding for open access charge: NIH [1 U54 CA143869].
Conflict of interest statement. None declared.
This work is dedicated to Jon Widom with warmth and appreciation for years of scientific advice and insight including for this project. We thank Kathy Matthews, Jia Xu, Kate Craig, Lin Han, Hernan Garcia, Phil Nelson, John Beausang, Laura Finzi, Jane Kondev, Shimon Weiss, Bob Schleif, Dave Wu, Matthew Johnson, Seth Blumberg, and Luke Breuer for insightful discussions and technical help, and the Mayo, Shan and Bjorkman labs for borrowed equipment and advice on the LacI purification.