The process by which a gene is expressed as a protein consists of two stages: transcription, which involves the DNA of the gene being copied into messenger RNA (mRNA); and translation, in which the mRNA is used as a template to assemble amino acids into a protein. Transcription and translation are controlled by many interlinked pathways, which ensures that genes are expressed when and where required.

One of these regulatory pathways involves steroid receptors. The binding of a steroid molecule to its receptor causes the receptor to move into the nucleus and interact with a specific gene, triggering transcription of that gene. When measured at the level of the whole organism, this transcriptional response is dose-dependent—the more steroid molecules that are present, the greater the amount of transcription. However, this is not the case in single cells, in which transcription is either activated or not. This ‘on/off’ behaviour is also seen over time: steroid-activated transcription occurs in bursts, separated by periods of inactivity.

To unravel the molecular mechanism behind this phenomenon, Larson et al. created a light-activated form of the ligand that activates a specific steroid receptor. Using this molecule, they were able to switch transcription of the gene controlled by that receptor on and off. They then used fluorescent proteins to label the mRNA and protein molecules that were produced as a result.

They found that activating the steroid receptor increases the likelihood of transcription occurring inside a cell, but not the duration of individual bursts of transcriptional activity, nor the amount of mRNA produced during each burst. Activation of a steroid receptor seems to control transcription by reducing the length of time each cell spends in the ‘off’ state between bursts.

Larson et al. incorporated their findings into a model that also takes into account the natural variability in levels of transcription between cells, and found that this could explain how the digital (on/off) control of transcription at the cellular level leads to analogue, dose-dependent control at the level of a whole organism. These findings should lead to further insights into how transcription is controlled at the molecular level.

Steroid receptors coordinate a diverse range of responses in higher eukaryotes and are involved in a wide range of human diseases (Gronemeyer et al., 2004). Steroid receptor response elements are present throughout the human genome and modulate chromatin remodeling and transcription in both a local and long-range fashion (John et al., 2011). As such, steroid receptor-mediated transcription is a paradigm of genetic control in the metazoan nucleus. Moreover, the ligand-dependent nature of these transcription factors makes them appealing targets for therapeutic intervention, necessitating a quantitative understanding of how receptors control output from target genes.

The classic sigmoidal dose response of steroid-regulated gene products belies the complexity of a system which relies on an intricate, multi-step sequence of events to initiate transcription from a target gene (McKenna et al., 2009). Depending on the particular steroid receptor, a partial list of events required to activate the target gene includes ligand-binding, receptor dimerization, nuclear translocation, eviction of co-repressors (i.e., histone deacetylases), recruitment of chromatin modifying enzymes (histone acetyltransferases, ATP-dependent remodeling enzymes, methyltransferases), and eventually recruitment of the basal transcription machinery. Despite this complexity, the dose-response is deceptively simple: it often takes the form of a simple Hill function with a coefficient of unity, which has traditionally been interpreted to imply that ligand-binding is the only rate-limiting step in the activation pathway (Ong et al., 2010). This description models the dose response as a continuum where each cell transcribes RNA at rates proportional to the dosage level.

However, single-cell studies of gene expression demonstrate that this description is incorrect. First, gene expression is not uniform over a population for a given dose, but shows variation from cell to cell, an observation which was first made for a steroid-responsive MMTV reporter gene (Ko et al., 1990). The authors demonstrated that the observed analog dose response was in fact a digital dose response (on or off) when viewed in single cells. Likewise, previous studies initially demonstrated that enhancers increase the probability of activation of a cell but not the strength of activation in the cell (Moreau et al., 1981; Weintraub, 1988; Moon and Ley, 1991; Walters et al., 1995). Second, expression is only a snapshot of temporally evolving gene activity (Ross et al., 1994; White et al., 1995; Harper et al., 2011; Suter et al., 2011). Thus, a cell is counted as activated or not activated dependent on the moment it is observed. Thus far, all single-cell measurements on metazoans suggest that genes are transcribed in ‘bursts’ of transcription, meaning that short periods of RNA synthesis are interspersed by long periods of inactivity. The causes of transcriptional bursting are unknown. Third, since molecules involved in regulating transcription are usually present at low copy number, this leads to stochastic fluctuations (‘noise’) and hence gene expression variation across the population (Larson et al., 2009). Finally, dynamic interactions between upstream regulators and chromatin add another level of complexity to the molecular events occurring during transcriptional activation (McNally et al., 2000; Darzacq et al., 2009; Ong et al., 2010). Under such conditions, the observed dose response does not result solely from ligand-binding but rather the composite result derived from many coupled reactions. In summary, population models of gene activation are too coarse to explain activation in single cells. Moreover, tools do not exist whereby the activity of single genes in single cells can be directly manipulated and measured.

In this work, we describe an approach for activating a steroid-receptor in order to achieve high temporal and spatial precision and to measure the activity of a responsive gene in the same cell over time. In contrast to the ensemble approach derived from observations of cell populations, we have developed a single-molecule kinetic approach for interrogation of a single gene. This approach is based on photoactivation of a steroid receptor ligand followed by observation of pre-mRNA synthesis at an active locus. The system consists of an exogenous reporter gene under control of the ecdysone receptor which is activated by the agonist ponasterone A (No et al., 1996). The real-time behavior of the gene is visualized using a bacteriophage capsid protein which binds MS2 RNA stem loops with high affinity to label nascent pre-mRNA in living cells (Bertrand et al., 1998). We demonstrate experimentally how the ensemble steroid dose response arises from the stochastic behavior of individual genes. These results suggest that the response element controls the frequency of gene activity but affects neither the duration of the active period nor the actual rate of transcripts produced during an active period. By using a caged ligand that could be uncaged by a light pulse (Lee et al., 2009; Lin et al., 2002) we measured the impulse-response of the gene and determined that a single pulse of active ligand resulted in a corresponding burst of polymerase activity several hours later. Further, this photoactivatable ligand has the property of being an antagonist in the caged state and an agonist in the uncaged state, enabling a precise window for kinetic perturbation in single cells. Thus, we were able to propose and validate a stochastic model of steroid-receptor activity for a reporter gene which provides a new framework for studying this ubiquitous mechanism of eukaryotic gene regulation.

We sought to design a reporter system that would enable visualization of multiple steps in gene expression: 1) the nascent pre-mRNA as it is synthesized by RNA polymerase II at an activated locus, 2) the completed mRNA in the cytoplasm and 3) the protein product produced from the mRNA. Furthermore, we desired a gene that would be unresponsive to endogenous steroid receptor (SR) ligands but also recapitulate the basic mechanism of SR transcriptional regulation. For these reasons, we chose to reconstitute ecdysone receptor-mediated transcription in human U2-OS cells (No et al., 1996). We first introduced a chimeric ecdysone receptor into the cells and subcloned a cell line which showed a strong response of a luciferase reporter. We next introduced an ecdysone-responsive reporter gene. The gene consists of a multimerized E/GRE response element (hybrid ecdysone/glucorticoid response element) in front of an Sp1 activator and a minimal heat shock promoter (mHSP, Figure 1A) (No et al., 1996). The coding region is bi-cistronic with the upstream region coding for CFP with a C-terminal SKL amino acid peroxisome-targeting motif (Janicki et al., 2004), and the downstream region coding for DsRED expressed in the cytosol from an IRES. Located in the 3′ UTR of the upstream cistron are 24 MS2 stem loops. The entire construct is 7.1 kb in length and is stably infected into the target U2-OS cells using lentiviral vectors (Mostoslavsky et al., 2006). Infections were carried out at MOI = 0.1 to obtain cells enriched for single integrations, and multiple clones were isolated and expanded for further analysis. The MS2 RNA stem loops are bound by the high affinity bacteriophage MS2 capsid protein dimer, where each monomer is labeled with YFP (Figure 1B) (Bertrand et al., 1998; Chao et al., 2008). Since the MS2-YFP is constitutively expressed and localized to the nucleus with an NLS, as soon as the RNA stem-loops are transcribed at an active locus, the MS2-YFP protein binds, thus making the transcription site (TS) visible in the microscope (Figure 1C, white arrow). The mature mRNAs carry the bound fluorescent proteins into the cytoplasm and are visible in the microscope as punctuate diffraction-limited fluorescent spots (Figure 1C, green, inset) (Shav-Tal et al., 2004; Mor et al., 2010). The translated protein is targeted to peroxisomes which are likewise visible as punctuate diffraction-limited fluorescent spots consisting of many copies of the CFP-SKL (Figure 1C, blue). Therefore, the reporter gene provides a live cell fluorescence readout that reflects both the history of the gene (total amount of mRNA and protein) and the current state of the gene (presence of the active transcription site).

Figure 1

Download asset Open asset

Alternatively, these cells when fixed can also be used for fluorescence in situ hybridization (FISH). In this case, we hybridize a fluorescently labeled 20-mer oligodeoxynucleotide to the region between each MS2 RNA stem loop (Figure 1D) (Femino et al., 1998; Zenklusen et al., 2008). FISH is a complementary approach to the live-cell MS2 system and likewise enables direct observation of both nascent RNA at an active transcription site (Figure 1E, white arrow) and total cellular reporter mRNA (Figure 1E, inset). The benefit of FISH is that it allows very high signal to noise observation of mRNA in fixed cells, while the advantage of the MS2 system is that one can directly observe dynamics in living cells. Both approaches utilize the polymeric nature of mRNA to label single molecules with large numbers (>40) of fluorophores, and both approaches enable quantification of the earliest steps in gene expression on single molecules, single genes and in single cells (Figure 1F) (Larson et al., 2009).

Direct measurement of transcriptional bursting on single genes in human cells

First, we directly observed RNA synthesis at an active locus in a clonal cell line using single-molecule live-cell microscopy. Consistent with previous studies which directly observe transcriptional activity of single genes in eukaryotes we observed transcriptional bursting (Chubb et al., 2006; Stevense et al., 2010; Yunger et al., 2010). We then measured the bursting dynamics in the microscope for three different concentrations of the steroid Ponasterone A (PA) (3.125 μM, 12.5 μM, 50 μM, control: 0.0 μM), which is the high affinity ligand for the chimeric ecdysone receptor. An example time series is shown for 50 μM PA (Figure 2A–F, Videos 1–5). The first transcription sites were visible several hours after induction (Figure 2—figure supplement 1). At steady state (24 hr after induction) for each dose, single gene transcription sites were captured at 15 min intervals for ∼ 15 hr, resulting in representative transcription intensity traces shown in Figure 2E,F (red, green lines, Video 4). As is evident from the images and the intensity analysis, individual cells show punctuated periods of transcriptional activity. The intensity traces were fit using a two-state hidden Markov model (Figure 2E,F, black lines, 50 μM; Figure 2—figure supplement 2, 12.5 μM, 3.125 μM) (Lee, 2009), allowing the distribution of on-times and off times to be determined (Figure 2G,H). We observed that the duration of inactivity (off-time) decreased in a dose-dependent manner (Figure 2I, gray bars). However, the duration of transcription activity (the on-time) did not significantly change over the dose response curve (Figure 2I, black bars). Moreover, the average peak brightness did also not change significantly with [PA] (Figure 2J). In the case of treatment with a vehicle ([PonA] = 0.0 μM), occasional spurious spots are detected by the tracking algorithm (Figure 2I, 0.0 μM, black bar), and the off-time is statistically indistinguishable from the duration of the entire time-lapse video (Fig. 2I, 0.0 μM, gray bar, compared to red-dashed line). Since the brightness of the nascent RNA signal reflects the number of individual RNA at the site of transcription (Zenklusen et al., 2008; Larson et al., 2011) (Figure 1F), these data indicate the average number of polymerases which fire during an active period is insensitive to [PA]. This live-cell analysis was also repeated for an additional clone with equivalent results (Figure 2—figure supplement 3 and below). In summary, direct measurements of transcription activity suggest that changing steroid levels affects the frequency of active periods but neither the duration of the active period nor the number of polymerases fired during the active period. Even for this highly-induced artificial reporter gene, the gene is active approximately 50% of the time at saturation (average on time = 38 ± 6 min; average off time = 35 ± 4 min) and shows a total gene cycling time of approximately 80 min.

Figure 2 with 3 supplements see all

Download asset Open asset

Video 1

Download asset

Video 2

Download asset

Video 3

Download asset

Video 4

Download asset

Video 5

Download asset

We then repeated this experiment on the same plasmid that was transiently transfected instead of genomically integrated by lentivirus (Figure 3, Videos 6 and 7). The transfected template, which is present at many copies in the nucleus and which is not integrated into chromatin, shows an extremely rapid induction (<20 min) and robust transcription throughout the time series. We were unable to detect transcriptional bursting. These data indicate that chromatin integration is a primary determinant of transcription kinetics for the hormone response, as has been suggested in previous studies (Archer et al., 1992).

Figure 3

Download asset Open asset

Video 6

Download asset

Video 7

Download asset

Abundance and variation of cellular mRNA indicates that nuclear receptors are frequency modulators

Using the lentivirus approach results in quasi-random insertion of the transgene throughout the genome. Since the kinetics of transcription depend strongly on transgene insertion (Figure 3), we sub-cloned three different lines, measured the dose response to [PA], and fit these data to a three-parameter Hill equation for the steroid response (Figure 4A):

(1) $\mathit{activity}=\frac{A_{\text{max}}{\left[\mathit{PA}\right]}^h}{{\mathit{EC}}_{50}+{\left[\mathit{PA}\right]}^h}$

where A_max is the amount of reporter mRNA present at saturation, [PA] is the concentration of ligand, EC₅₀ is the half-maximal response, and h is the Hill coefficient (Ong et al., 2010). Although the saturation value for the number of mRNA/cell differed substantially for each insertion site (Figure 4A, 960 ± 60 mRNA/cell, 430 ± 50 mRNA/cell, 180 ± 10 mRNA/cell, blue, green, red curves respectively), the Hill coefficient did not change significantly (1.2 ± 0.2, 1.3 ± 0.3, 1.2 ± 0.2), nor did the EC₅₀ (8.0 ± 1.4 μM, 10 ± 3.2 μM, 9.1 ± 1.5 μM). We also compared the protein dose response to the RNA dose response for one of the cell lines and observed similar behavior (Figure 4B). Since the population dose response for this reporter gene is statistically indistinguishable from a first-order Hill response, we also show a fit to a two-parameter Hill equation with the Hill coefficient fixed at unity (Figure 4B, black line, A_max = 190 ± 10 mRNA, EC₅₀ = 9.6 ± 2.0 μM). In summary, the population dose response is well-described by a first-order Hill response function where the Hill coefficient and the EC₅₀ are invariant with insertion site, but the saturation level (A_max) shows fivefold variation among different clones.

Figure 4

Download asset Open asset

We next sought to build and test a quantitative kinetic model capable of elucidating how the classical analog dose response in a population (Figure 4A,B) relates to the time-dependent digital behavior observed directly in single cells (Figure 2). We used the ‘Random Telegraph model’, which is a paradigm for stochastic transcription and quantitatively describes the distribution of RNA/cell in a variety of organisms (Peccoud and Ycart, 1995; Larson et al., 2009). The key feature of this model is the hypothesis that the gene toggles between an active state where transcripts can be synthesized and an inactive state where no transcripts are produced. It is important to note that the gene can be ‘active’ even when no nascent RNA is detected, owing partly to the stochastic initiation events within an active state (Figure 4C) and partly to the placement of the MS2 cassette in the middle of the gene (Figure 1A). The biochemical determinant(s) of this active state is unknown and is likely to vary between genes. Consequently, it is not possible at present to do a direct measurement of the active state. However, the distribution of total cellular reporter mRNA reflects the dynamics of this active-inactive switching, so we measured the steady state distribution of total cellular reporter mRNA over a population using single-molecule FISH in fixed cells.

For each cell line, the variation of total cellular mRNA within a population is indicative of infrequent bursts of transcriptional activity (Golding et al., 2005; Chubb et al., 2006; Raj et al., 2006; Voss et al., 2006; Zenklusen et al., 2008) (Figure 4D–F). At low PA induction levels (3.12 μM, Figure 4D), there is a large heterogeneity of cellular mRNA in the population: some cells contain dozens of mRNAs and others have none. Transcription sites are infrequent. At medium induction levels (12.5 μM, Figure 4E), most cells contain mRNA, and many are actively transcribing. At high levels of induction (50 μM, Figure 4F), cells contain a high abundance of mRNA in the cytosol, and most cells show an active transcription site in the nucleus. At each concentration of the steroid, we fit the full distribution of mRNA/cell to the steady-state solution for the random telegraph model (Figure 4G–L, N = 60 cells at each dose). The steady state distribution function used for fitting is a function of three unitless kinetic parameters (a/d, b/d, c/d) corresponding to the rate of activation (a), the rate of inactivation (b), and the rate of firing once the gene is active (c), each normalized by the RNA decay rate, d (Figure 4C). These parameters correspond to the frequency of bursts, the duration of bursts, and the initiation rate from a burst, respectively. At steady state, the mean level of RNA 〈N〉 is:

(2) $\langle N\rangle =\frac{c}{d}\left(\frac{a}{a+b}\right)$

where a/(a+b) corresponds to the fraction of time the gene spends in the active state.

There is a functional similarity between the stochastic random telegraph equation (Equation 2) and the deterministic Hill equation (Equation 1): A_max corresponds to c/d, EC₅₀ corresponds to b, and [PA] corresponds to a. This correspondence indicates that non-cooperative behavior for the hormone response (h = 1, Equation 1) is mathematically equivalent to frequency modulation. In other words, the rate of switching to the active state (a) is the only variable which changes with ligand concentration. In Figure 4G–L we show the distribution of mRNA/cell compared to the theoretical fit arising from the telegraph model as a function of dose. Indeed, these data are consistent with a model where the only variable which varies with dose is the rate of activation (a), while the other parameters (b,c) are globally fit for the entire dose response for a particular cell line (b/d = 1.5, c/d =340). When the same analysis is carried out on cell lines with different insertion sites (i.e., Figure 4A), the dose-dependent activation rate (a) and the dose-independent inactivation rate (b) remain invariant, but the initiation rate (c) changes with the cell line. Thus, the absolute number of polymerases loaded during an active period does not depend on steroid level but rather some extrinsic factor such as local chromatin environment or abundance of some component of the core transcriptional machinery. This is evident when comparing the episomal genes, which do not show frequency modulation. In summary, both fixed-cell and live-cell single-molecule measurements are consistent with a model where the steroid is a robust frequency modulator.

Triggering a single activation event in living cells with UV photolysis

We emphasize that the frequency change observed here is in response to changes in the time-invariant level of steroid. Conversely, there is a growing body of evidence that the dynamics of the upstream activator itself may in fact carry biological information and modulate downstream expression levels (Cai et al., 2008; Ashall et al., 2009; Stavreva et al., 2009; Purvis et al., 2012). To investigate the temporal relationship between steroid availability and the kinetics of transcription, we required a ligand that could be activated at a precise time to induce transcription. We utilized a caged PA that consists of PA modified at a critical OH residue with a photolabile organic ligand (Figure 5A) (Lin et al., 2002).

Figure 5 with 2 supplements see all

Download asset Open asset

We observed empirically that we were unable to activate single genes in the presence of the caged reagent and hypothesized that the caged PA possessed antagonist properties. We therefore examined the dose-dependent induction properties of the reporter gene in the presence and absence of the caged reagent by performing an in vivo competition experiment. First, we confirmed that both forms of the caged PA were inactive by measuring a dose response of the caged ligands (Figure 5—figure supplement 1). Second, the cells were incubated with two different forms of the caged PA (DMNB-PonA and CNB-PonA) (Figure 5A). After overnight incubation, the cells were exposed to 25 μM of the active, unmodified PA for 30 min. The cells were then washed and returned to the incubator for 8 hr to allow the buildup of protein (Figure 5B). At 100 μM of DMNB-PA, we observed a fivefold reduction in protein levels, indicating that the caged ligand blocks the activity of the active, uncaged form in vivo (Figure 5C, gray, Figure 5—figure supplement 2). When the DMNB-PonA and uncaged PA are present in equimolar amounts, the competition is alleviated, and the protein output from the reporter gene is close to the levels observed in the absence of the caged ligand. Thus, we show that this ligand has the unique property of being an agonist in the uncaged state and an antagonist in the caged state.

The benefit of a reagent that can be switched from repressive to activating is that the effect of small amounts of the agonist are suppressed by the presence of the antagonist. Such a feature is especially important for a reagent that is membrane-permeable and might diffuse to neighboring cells in the tissue. Thus, the caged compound has built-in protection against ‘leakiness’ and is capable of acting in a highly localized spatial manner when uncaged. The method of activation, therefore, has to be all or none: the balance between agonist/antagonist must be shifted past a threshold in order for gene activation to occur. A typical experimental design is schematized in Figure 6A. The nucleus is activated with a UV laser at 349 nm. The above-threshold dose of uncaging light (40 pulses, 100 Hz, 2 J/cm² total dose at the sample) is delivered in approximately 400 ms over the area of the nucleus (∼80 μm²). Now, all the receptors in the nucleus are active, leading to the initiation of transcription at the single locus. Concurrently, any activating ligand that diffuses away encounters neighboring cells that are in the repressed state due to the presence of the antagonist. Once the activation occurs in the target cell, the diffusion into the cell of unactivated antagonist coupled with the rapid turnover of receptor-bound agonist (Stavreva et al., 2004) results in antagonist-bound receptors that bind to the response element and repress transcription.

Figure 6

Download asset Open asset

The time evolution of a typical activated cell is shown in Figure 6 (Video 8). The cell is uncaged at t = 0 (Figure 6B); the transcription site is visible at 2.5 hr (Figure 6C) and persists in the on state for 1.5 hr afterwards (Figure 6D), consistent with the duration of the on-state measured previously (Figure 2G). At 15 hr, the CFP peroxisomes are observed to assay whether the cell has activated (Figure 6E). In this example, only the central cell was photolyzed, and is the only cell that displays expression of the CFP protein construct. By contrast, the two non-photolyzed neighboring cells show no evidence of CFP expression. The photolysis protocol above results in: 1) an average 106 +/− 36 min between uncaging and the appearance of a transcription site which then persisted for 35 +/− 12 min, 2) 40% success rate for protein expression, and 3) a single on-state in 70% of those cells which activated, with the remainder showing either no detectable on-period (13%) or multiple periods (17%) (N = 30). Moreover, we also observe cell division after uncaging, which is a measure of viability after UV photolysis (Video 9). In sum, the photoactivation data suggest that a single pulse of ligand results in a single active period of transcription.

Video 8

Download asset

Video 9

Download asset

The work presented here addresses the specific question as to how transcription factor availability translates into transcriptional activity. By a variety of approaches, we show that the frequency of initiation is the variable that determines RNA levels, but only when the gene is integrated into the chromatin. By measuring the synthesis of pre-mRNA directly, we connected the dynamics of the gene with the availability of an upstream transcription factor, the ligand-bound steroid receptor. We demonstrated how steroid receptors, which are present in all metazoans, can control transcription through frequency modulation. This digital frequency modulation of transcription in single cells manifests as an analog dose response in cellular mRNA and protein in a population. Importantly, the ability to activate the transcription factor (steroid receptor) using light provided a precise temporal start to measure the time needed for the gene to respond, an unexpectedly long time (2 hr). Hence our observations clarify the mechanics behind these observations and extend the understanding to the intrinsic cycling on and off for a gene. One dose of a transcription factor yields one cycle of transcription.

The precise timing of the transcriptional response places constraints on the regulation of the gene: one can infer regulatory principles based on the observed synthesis of pre-mRNA (Pedraza and Paulsson, 2008). A stochastic model of kinetic rates allowed us to break down the steroid transcriptional response into distinct processes, which could be measured independently in the microscope. Although a number of studies have inferred transcriptional bursting by measuring events which are downstream of transcription, such as cellular RNA or protein production, (Blake et al., 2003; Raser and O’Shea, 2004; Becskei et al., 2005; Bar-Even et al., 2006; Newman et al., 2006; Raj et al., 2006; Zenklusen et al., 2008; Singh et al., 2010; Batenchuk et al., 2011; Skupsky et al., 2010; So et al., 2011; Suter et al., 2011), we are able to unambiguously resolve the properties of individual bursts and relate these measured kinetic steps to steroid abundance. For this ecdysteroid-responsive reporter gene in mammalian cells, the time between on-states decreased at higher doses of PA. The duration of the active state and the transcription rate from the active state were invariant. However, the ‘on’ times and ‘off’ times were both exponentially distributed (Figure 2G,H), which suggested that both processes were determined by a single rate-limiting step. The simplest biochemical model to explain this result is that the state of the gene is determined entirely by the on rate and off rate of the active ligand-receptor complex from the response element. The corollary to this model is that the actual transcription rate (c) from the active state is determined by factors other than the response element. For example the ability of trans-acting factors to access cis-acting sequences in the core promoter may determine the firing rate from the active state. In agreement with this supposition, the dynamic frequency modulation by the steroid was invariant with the insertion site, but the transcription rate varied over fivefold with insertion site. In this way, genetic regulation can be conceptualized as the superposition of individual parts, with distal enhancers and response elements controlling frequency of activation and proximal promoters determining the firing rate. Future experiments will be able to directly test this hypothesis using further development of the single-molecule approaches described here.

This simple model must address the fact that the exchange rate of SRs on chromatin has been observed to be extremely dynamic. Fluorescence photobleaching recovery measurements on tandem gene arrays in single cells record dwell times on the order of seconds to minutes (Darzacq et al., 2009), while the duration of the active transcriptional state in this study is on the order of an hour. One possibility is that SR binding is actually much longer, and FRAP data on arrays is dominated by non-specific interactions and is not capable of detecting the small percentage of receptors which might be stably bound. Another possibility is that SR binding is in fact short-lived but initiates a cascade of events involved in activation of the gene (Herschlag and Johnson, 1993; Hager et al., 2006; Metivier et al., 2006). In support of this latter view, we observe that there was a delay between photoactivation of the ligand and the appearance of the first transcription sites (Figure 6). This delay was comparable in size to the average duration of the off period. In contrast, a non-chromatinized or weakly chromatinized template showed extremely rapid induction (<20 min). It is plausible that chromatin remodeling is inefficient at this reporter gene, possibly because the promoter lacks cis-acting elements which might aid in a more rapid chromatin remodeling response. Taken together, these results support the notion that opening of chromatin interposes a delay between the docking of active ligand-receptor complex on the response element and the eventual synthesis of pre-mRNA (Janicki et al., 2004).

The synergy between DNA binding factors and chromatin modifying proteins is essential for control of gene expression (Segal and Widom, 2009), and the role of chromatin in a bimodal model of gene regulation has been proposed previously (Archer et al., 1992; Di Croce et al., 1999). Moreover, recent evidence points strongly toward chromatin as directly controlling the duration of the on and off states. Histone methylation has been implicated in the transmission of transcriptional frequency between mother and daughter cells (Muramoto et al., 2010). Manipulation of histone acetylation changes transcriptional cycle timing (Harper et al., 2011; Suter et al., 2011). Histone deacetylases and nucleosome-remodeling complexes are cyclically recruited during transcriptional activation of an estrogen-responsive gene (Métivier et al., 2003). Importantly, FRAP studies on histones and genome-wide pulse-chase measurements on histones suggest a timescale of turnover (∼hour) that is more consistent with the time scales of transcriptional activity observed in this study (Kimura and Cook, 2001; Dion et al., 2007; Deal et al., 2010). The bimodal ‘telegraph’ model is likely a simplification of the number of biological steps involved in gene activation, but the quantitative agreement between this model and our data suggests that at steady state, the requirements for re-activating a gene that has cycled off may depend on only a few rate-limiting steps.

In summary, we have described a comprehensive approach for unraveling the dynamic behavior of genetic control in single cells. By using a combination of nascent RNA visualization, single molecule microscopy, computational modeling, and light activation, we were able to construct a quantitative model of steroid-receptor mediated activation in which SRs control the frequency of the transition to an active transcriptional state in a dose-dependent manner. This model is based on a stochastic, kinetic description that reflects the molecular events occurring in single-cells.

1. 1. Archer T
   2. Lefebvre P
   3. Wolford R
   4. Hager G

   (1992) Transcription factor loading on the MMTV promoter: a bimodal mechanism for promoter activation

   Science 255:1573–1576.

   https://doi.org/10.1126/science.1347958

   • Google Scholar
2. 1. Ashall L
   2. Horton CA
   3. Nelson DE
   4. Paszek P
   5. Harper CV
   6. Sillitoe K

   et al. (2009) Pulsatile stimulation determines timing and specificity of NF-{kappa}B-dependent transcription

   Science 324:242–246.

   https://doi.org/10.1126/science.1164860

   • Google Scholar
3. 1. Bar-Even A
   2. Paulsson J
   3. Maheshri N
   4. Carmi M
   5. O’Shea E
   6. Pilpel Y

   et al. (2006) Noise in protein expression scales with natural protein abundance

   Nat Genet 38:636–643.

   https://doi.org/10.1038/ng1807

   • Google Scholar
4. 1. Batenchuk C
   2. St-Pierre S
   3. Tepliakova L
   4. Adiga S
   5. Szuto A
   6. Kabbani N

   et al. (2011) Chromosomal position effects are linked to Sir2-mediated variation in transcriptional burst size

   Biophys J 100:L56–L58.

   https://doi.org/10.1016/j.bpj.2011.04.021

   • Google Scholar
5. 1. Becskei A
   2. Kaufmann BB
   3. van Oudenaarden A

   (2005) Contributions of low molecule number and chromosomal positioning to stochastic gene expression

   Nat Genet 37:937–944.

   https://doi.org/10.1038/ng1616

   • Google Scholar
6. 1. Bertrand E
   2. Chartrand P
   3. Schaefer M
   4. Shenoy SM
   5. Singer RH
   6. Long RM

   (1998) Localization of ASH1 mRNA particles in living yeast

   Mol Cell 2:437–445.

   https://doi.org/10.1016/S1097-2765(00)80143-4

   • Google Scholar
7. 1. Blake WJ
   2. Kaern M
   3. Cantor CR
   4. Collins JJ

   (2003) Noise in eukaryotic gene expression

   Nature 422:633–637.

   https://doi.org/10.1038/nature01546

   • Google Scholar
8. 1. Cai L
   2. Dalal CK
   3. Elowitz MB

   (2008) Frequency-modulated nuclear localization bursts coordinate gene regulation

   Nature 455:485–490.

   https://doi.org/10.1038/nature07292

   • Google Scholar
9. 1. Chang C
   2. Niblack B
   3. Walker B
   4. Bayley H

   (1995) A photogenerated pore-forming protein

   Chemi Biol 2:391–400.

   https://doi.org/10.1016/1074-5521(95)90220-1

   • Google Scholar
10. 1. Chao JA
    2. Patskovsky Y
    3. Almo SC
    4. Singer RH

    (2008) Structural basis for the coevolution of a viral RNA-protein complex

    Nat Struct Mol Biol 15:103–105.

    https://doi.org/10.1038/nsmb1327

    • Google Scholar
11. 1. Chubb JR
    2. Trcek T
    3. Shenoy SM
    4. Singer RH

    (2006) Transcriptional pulsing of a developmental gene

    Curr Biol 16:1018–1025.

    https://doi.org/10.1016/j.cub.2006.03.092

    • Google Scholar
12. 1. Crocker JC
    2. Grier DG

    (1996) Methods of digital video microscopy for colloidal studies

    J Colloid Interf Sci 179:298–310.

    https://doi.org/10.1006/jcis.1996.0217

    • Google Scholar
13. 1. Darzacq X
    2. Yao J
    3. Larson D
    4. Causse S
    5. Bosanac L
    6. de Turris V

    et al. (2009) Imaging transcription in living cells

    Annu Rev Biophys 38:173–196.

    https://doi.org/10.1146/annurev.biophys.050708.133728

    • Google Scholar
14. 1. Deal RB
    2. Henikoff JG
    3. Henikoff S

    (2010) Genome-wide kinetics of nucleosome turnover determined by metabolic labeling of histones

    Science 328:1161–1164.

    https://doi.org/10.1126/science.1186777

    • Google Scholar
15. 1. Di Croce L
    2. Koop R
    3. Venditti P
    4. Westphal HM
    5. Nightingale KP
    6. Corona DF

    et al. (1999) Two-step synergism between the progesterone receptor and the DNA-binding domain of nuclear factor 1 on MMTV minichromosomes

    Mol Cell 4:45–54.

    https://doi.org/10.1016/S1097-2765(00)80186-0

    • Google Scholar
16. 1. Dion MF
    2. Kaplan T
    3. Kim M
    4. Buratowski S
    5. Friedman N
    6. Rando OJ

    (2007) Dynamics of replication-independent histone turnover in budding yeast

    Science 315:1405–1408.

    https://doi.org/10.1126/science.1134053

    • Google Scholar
17. 1. Femino AM
    2. Fay FS
    3. Fogarty K
    4. Singer RH

    (1998) Visualization of single RNA transcripts in situ

    Science 280:585–590.

    https://doi.org/10.1126/science.280.5363.585

    • Google Scholar
18. 1. Fusco D
    2. Accornero N
    3. Lavoie B
    4. Shenoy SM
    5. Blanchard J-M
    6. Singer RH

    et al. (2003) Single mRNA molecules demonstrate probabilistic movement in living mammalian cells

    Curr Biol 13:161–167.

    https://doi.org/10.1016/S0960-9822(02)01436-7

    • Google Scholar
19. 1. Golding I
    2. Paulsson J
    3. Zawilski SM
    4. Cox EC

    (2005) Real-time kinetics of gene activity in individual bacteria

    Cell 123:1025–1036.

    https://doi.org/10.1016/j.cell.2005.09.031

    • Google Scholar
20. 1. Gronemeyer H
    2. Gustafsson J-A
    3. Laudet V

    (2004) Principles for modulation of the nuclear receptor superfamily

    Nat Rev Drug Discov 3:950–964.

    https://doi.org/10.1038/nrd1551

    • Google Scholar
21. 1. Hager G
    2. Elbi C
    3. Johnson T
    4. Voss T
    5. Nagaich A
    6. Schiltz R

    et al. (2006) Chromatin dynamics and the evolution of alternate promoter states

    Chromosome Res 14:107–116.

    https://doi.org/10.1007/s10577-006-1030-0

    • Google Scholar
22. 1. Harper CV
    2. Finkenstädt B
    3. Woodcock DJ
    4. Friedrichsen S
    5. Semprini S
    6. Ashall L

    et al. (2011) Dynamic analysis of stochastic transcription cycles

    PLOS Biol 9:e1000607.

    https://doi.org/10.1371/journal.pbio.1000607

    • Google Scholar
23. 1. Herschlag D
    2. Johnson FB

    (1993) Synergism in transcriptional activation: a kinetic view

    Genes Dev 7:173–179.

    https://doi.org/10.1101/gad.7.2.173

    • Google Scholar
24. 1. Janicki SM
    2. Tsukamoto T
    3. Salghetti SE
    4. Tansey WP
    5. Sachidanandam R
    6. Prasanth KV

    et al. (2004) From silencing to gene expression: real-time analysis in single cells

    Cell 116:683–698.

    https://doi.org/10.1016/S0092-8674(04)00171-0

    • Google Scholar
25. 1. John S
    2. Sabo PJ
    3. Thurman RE
    4. Sung M-H
    5. Biddie SC
    6. Johnson TA

    et al. (2011) Chromatin accessibility pre-determines glucocorticoid receptor binding patterns

    Nat Genet 43:264–268.

    https://doi.org/10.1038/ng.759

    • Google Scholar
26. 1. Kimura H
    2. Cook PR

    (2001) Kinetics of core histones in living human cells

    J Cell Biol 153:1341–1354.

    https://doi.org/10.1083/jcb.153.7.1341

    • Google Scholar
27. 1. Ko MS
    2. Nakauchi H
    3. Takahashi N

    (1990)

    The dose-dependence of glucocorticoid-inducible gene exprssion results from the change in the number of transcriptionally active templates

    EMBO J 9:2835–2842.

    • Google Scholar
28. 1. Larson DR
    2. Singer RH
    3. Zenklusen D

    (2009) A single molecule view of gene expression

    Trends Cell Biol 19:630–637.

    https://doi.org/10.1016/j.tcb.2009.08.008

    • Google Scholar
29. 1. Larson DR
    2. Zenklusen D
    3. Wu B
    4. Chao JA
    5. Singer RH

    (2011) Real-time observation of transcription initiation and elongation on an endogenous yeast gene

    Science 332:475–478.

    https://doi.org/10.1126/science.1202142

    • Google Scholar
30. 1. Lee T-H

    (2009) Extracting kinetics information from single-molecule fluorescence resonance energy transfer data using hidden Markov models

    J Phys Chem B 113:11535–11542.

    https://doi.org/10.1021/jp903831z

    • Google Scholar
31. 1. Lin W
    2. Albanese C
    3. Pestell RG
    4. Lawrence DS

    (2002) Spatially discrete, light-driven protein expression

    Chem Biol 9:1347–1353.

    https://doi.org/10.1016/S1074-5521(02)00288-0

    • Google Scholar
32. 1. Lee H
    2. Larson DR
    3. Lawrence DS

    (2009) Illuminating the chemistry of life: design, synthesis, and applications of “Caged” and related photoresponsive compounds

    ACS Chem Biol 4:409–427.

    https://doi.org/10.1021/cb900036s

    • Google Scholar
33. 1. McKenna NJ
    2. Cooney AJ
    3. DeMayo FJ
    4. Downes M
    5. Glass CK
    6. Lanz RB

    et al. (2009) Minireview: evolution of NURSA, the nuclear receptor signaling atlas

    Mol Endocrinol 23:740–746.

    https://doi.org/10.1210/me.2009-0135

    • Google Scholar
34. 1. McNally JG
    2. Müller WG
    3. Walker D
    4. Wolford R
    5. Hager GL

    (2000) The glucocorticoid receptor: rapid exchange with regulatory sites in living cells

    Science 287:1262–1265.

    https://doi.org/10.1126/science.287.5456.1262

    • Google Scholar
35. 1. Métivier R
    2. Penot G
    3. Hübner MR
    4. Reid G
    5. Brand H
    6. Kos M

    et al. (2003) Estrogen receptor-[alpha] directs ordered, cyclical, and combinatorial recruitment of cofactors on a natural target promoter

    Cell 115:751–763.

    https://doi.org/10.1016/S0092-8674(03)00934-6

    • Google Scholar
36. 1. Metivier R
    2. Reid G
    3. Gannon F

    (2006) Transcription in four dimensions: nuclear receptor-directed initiation of gene expression

    EMBO Rep 7:161–167.

    https://doi.org/10.1038/sj.embor.7400626

    • Google Scholar
37. 1. Moon A
    2. Ley T

    (1991)

    Functional properties of the beta-globin locus control region in K562 erythroleukemia cells

    Blood 77:2272–2284.

    • Google Scholar
38. 1. Mor A
    2. Suliman S
    3. Ben-Yishay R
    4. Yunger S
    5. Brody Y
    6. Shav-Tal Y

    (2010) Dynamics of single mRNP nucleocytoplasmic transport and export through the nuclear pore in living cells

    Nat Cell Biol 12:543–552.

    https://doi.org/10.1038/ncb2056

    • Google Scholar
39. 1. Moreau P
    2. Hen R
    3. Wasylyk B
    4. Everett R
    5. Gaub MP
    6. Chambon P

    (1981) The SV40 72 base repair repeat has a striking effect on gene expression both in SV40 and other chimeric recombinants

    Nucleic Acids Res 9:6047–6068.

    https://doi.org/10.1093/nar/9.22.6047

    • Google Scholar
40. 1. Mostoslavsky G
    2. Fabian AJ
    3. Rooney S
    4. Alt FW
    5. Mulligan RC

    (2006) Complete correction of murine Artemis immunodeficiency by lentiviral vector-mediated gene transfer

    Proc Natl Acad Sci USA 103:16406–16411.

    https://doi.org/10.1073/pnas.0608130103

    • Google Scholar
41. 1. Muramoto T
    2. Müller I
    3. Thomas G
    4. Melvin A
    5. Chubb JR

    (2010) Methylation of H3K4 is required for inheritance of active transcriptional states

    Curr Biol 20:397–406.

    https://doi.org/10.1016/j.cub.2010.01.017

    • Google Scholar
42. 1. Newman JRS
    2. Ghaemmaghami S
    3. Ihmels J
    4. Breslow DK
    5. Noble M
    6. DeRisi JL

    et al. (2006) Single-cell proteomic analysis of S. cerevisiae reveals the architecture of biological noise

    Nature 441:840–846.

    https://doi.org/10.1038/nature04785

    • Google Scholar
43. 1. No D
    2. Yao T-P
    3. Evans RM

    (1996) Ecdysone-inducible gene expression in mammalian cells and transgenic mice

    Proc Natl Acad Sci USA 93:3346–3351.

    https://doi.org/10.1073/pnas.93.8.3346

    • Google Scholar
44. 1. Ong KM
    2. Blackford JA
    3. Kagan BL
    4. Simons SS
    5. Chow CC

    (2010) A theoretical framework for gene induction and experimental comparisons

    Proc Natl Acad Sci USA 107:7107–7112.

    https://doi.org/10.1073/pnas.0911095107

    • Google Scholar
45. 1. Peccoud J
    2. Ycart B

    (1995) Markovian modeling of gene-product synthesis

    Theor Population Biol 48:222–234.

    https://doi.org/10.1006/tpbi.1995.1027

    • Google Scholar
46. 1. Pedraza JM
    2. Paulsson J

    (2008) Effects of molecular memory and bursting on fluctuations in gene expression

    Science 319:339–343.

    https://doi.org/10.1126/science.1144331

    • Google Scholar
47. 1. Purvis JE
    2. Karhohs KW
    3. Mock C
    4. Batchelor E
    5. Loewer A
    6. Lahav G

    (2012) p53 dynamics control cell fate

    Science 336:1440–1444.

    https://doi.org/10.1126/science.1218351

    • Google Scholar
48. 1. Raj A
    2. Peskin CS
    3. Tranchina D
    4. Vargas DY
    5. Tyagi S

    (2006) Stochastic mRNA synthesis in mammalian cells

    PLOS Biol 4:e309.

    https://doi.org/10.1371/journal.pbio.0040309

    • Google Scholar
49. 1. Raser JM
    2. O’Shea EK

    (2004) Control of stochasticity in eukaryotic gene expression

    Science 304:1811–1814.

    https://doi.org/10.1126/science.1098641

    • Google Scholar
50. 1. Ross IL
    2. Browne CM
    3. Hume DA

    (1994) Transcription of individual genes in eukaryotic cells occurs randomly and infrequently

    Immunol Cell Biol 72:177–185.

    https://doi.org/10.1038/icb.1994.26

    • Google Scholar
51. 1. Segal E
    2. Widom J

    (2009) From DNA sequence to transcriptional behaviour: a quantitative approach

    Nat Rev Genet 10:443–456.

    https://doi.org/10.1038/nrg2591

    • Google Scholar
52. 1. Shav-Tal Y
    2. Darzacq X
    3. Shenoy SM
    4. Fusco D
    5. Janicki SM
    6. Spector DL

    et al. (2004) Dynamics of single mRNPs in nuclei of living cells

    Science 304:1797–1800.

    https://doi.org/10.1126/science.1099754

    • Google Scholar
53. 1. Singh A
    2. Razooky B
    3. Cox CD
    4. Simpson ML
    5. Weinberger LS

    (2010) Transcriptional bursting from the HIV-1 promoter is a significant source of stochastic noise in HIV-1 gene expression

    Biophys J 98:L32–L34.

    https://doi.org/10.1016/j.bpj.2010.03.001

    • Google Scholar
54. 1. Skupsky R
    2. Burnett JC
    3. Foley JE
    4. Schaffer DV
    5. Arkin APHIV

    (2010) Promoter integration site primarily modulates transcriptional burst size rather than frequency

    PLOS Comput Biol 6:e1000952.

    https://doi.org/10.1371/journal.pcbi.1000952

    • Google Scholar
55. 1. So L-H
    2. Ghosh A
    3. Zong C
    4. Sepulveda LA
    5. Segev R
    6. Golding I

    (2011) General properties of transcriptional time series in Escherichia coli

    Nat Genet 43:554–560.

    https://doi.org/10.1038/ng.821

    • Google Scholar
56. 1. Stavreva DA
    2. Müller WG
    3. Hager GL
    4. Smith CL
    5. McNally JG

    (2004) Rapid glucocorticoid receptor exchange at a promoter is coupled to transcription and regulated by Chaperones and Proteasomes

    Mol Cell Biol 24:2682–2697.

    https://doi.org/10.1128/mcb.24.7.2682-2697.2004

    • Google Scholar
57. 1. Stavreva DA
    2. Wiench M
    3. John S
    4. Conway-Campbell BL
    5. McKenna MA
    6. Pooley JR

    et al. (2009) Ultradian hormone stimulation induces glucocorticoid receptor-mediated pulses of gene transcription

    Nat Cell Biol 11:1093–1102.

    https://doi.org/10.1038/ncb1922

    • Google Scholar
58. 1. Stevense M
    2. Muramoto T
    3. Müller I
    4. Chubb JR

    (2010) Digital nature of the immediate-early transcriptional response

    Development 137:579–584.

    https://doi.org/10.1242/dev.043836

    • Google Scholar
59. 1. Suter DM
    2. Molina N
    3. Gatfield D
    4. Schneider K
    5. Schibler U
    6. Naef F

    (2011) Mammalian genes are transcribed with widely different bursting kinetics

    Science 332:472–474.

    https://doi.org/10.1126/science.1198817

    • Google Scholar
60. 1. Thompson RE
    2. Larson DR
    3. Webb WW

    (2002) Precise nanometer localization analysis for individual fluorescent probes

    Biophys J 82:2775–2783.

    https://doi.org/10.1016/S0006-3495(02)75618-X

    • Google Scholar
61. 1. Trcek T
    2. Chao JA
    3. Larson DR
    4. Park HY
    5. Zenklusen D
    6. Shenoy SM

    et al. (2012) Single-mRNA counting using fluorescent in situ hybridization in budding yeast

    Nat Protocols 7:408–419.

    https://doi.org/10.1038/nprot.2011.451

    • Google Scholar
62. 1. Voss TC
    2. John S
    3. Hager GL

    (2006) Single-cell analysis of glucocorticoid receptor action reveals that stochastic post-chromatin association mechanisms regulate ligand-specific transcription

    Mol Endocrinol 20:2641–2655.

    https://doi.org/10.1210/me.2006-0091

    • Google Scholar
63. 1. Walters MC
    2. Fiering S
    3. Eidemiller J
    4. Magis W
    5. Groudine M
    6. Martin DI

    (1995) Enhancers increase the probability but not the level of gene expression

    Proc Natl Acad Sci USA 92:7125–7129.

    https://doi.org/10.1073/pnas.92.15.7125

    • Google Scholar
64. 1. Weintraub H

    (1988) Formation of stable transcription complexes as assayed by analysis of individual templates

    Proc Natl Acad Sci USA 85:5819–5823.

    https://doi.org/10.1073/pnas.85.16.5819

    • Google Scholar
65. 1. White MR
    2. Masuko M
    3. Amet L
    4. Elliott G
    5. Braddock M
    6. Kingsman AJ

    et al. (1995)

    Real-time analysis of the transcriptional regulation of HIV and hCMV promoters in single mammalian cells

    J Cell Sci 108:441–455.

    • Google Scholar
66. 1. Yunger S
    2. Rosenfeld L
    3. Garini Y
    4. Shav-Tal Y

    (2010) Single-allele analysis of transcription kinetics in living mammalian cells

    Nat Meth 7:631–633.

    https://doi.org/10.1038/nmeth.1482

    • Google Scholar
67. 1. Zenklusen D
    2. Larson DR
    3. Singer RH

    (2008) Single-RNA counting reveals alternative modes of gene expression in yeast

    Nat Struct Mol Biol 15:1263–1271.

    https://doi.org/10.1038/nsmb.1514

    • Google Scholar

Author details

1. Daniel R Larson

   Laboratory of Receptor Biology and Gene Expression, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, United States

   Contribution

   DRL, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article, Contributed unpublished essential data or reagents

   For correspondence

   dan.larson@nih.gov

   Competing interests

   No competing interests declared.

2. Christoph Fritzsch

   Institute of Molecular Biology, Johannes Gutenberg University of Mainz, Mainz, Germany

   Contribution

   CF, Acquisition of data, Analysis and interpretation of data

   Competing interests

   No competing interests declared.

3. Liang Sun

   1. Department of Chemistry, Division of Chemical Biology and Medicinal Chemistry, University of North Carolina, Chapel Hill, United States
   2. Department of Pharmacology, University of North Carolina, Chapel Hill, United States

   Contribution

   LS, Acquisition of data, Contributed unpublished essential data or reagents

   Competing interests

   No competing interests declared.

4. Xiuhau Meng

   Department of Anatomy and Structural Biology, Albert Einstein College of Medicine, New York, United States

   Contribution

   XM, Acquisition of data, Contributed unpublished essential data or reagents

   Competing interests

   No competing interests declared.

5. David S Lawrence

   1. Department of Chemistry, Division of Chemical Biology and Medicinal Chemistry, University of North Carolina, Chapel Hill, United States
   2. Department of Pharmacology, University of North Carolina, Chapel Hill, United States

   Contribution

   DSL, Conception and design, Drafting or revising the article

   Competing interests

   No competing interests declared.

6. Robert H Singer

   1. Department of Anatomy and Structural Biology, Albert Einstein College of Medicine, New York, United States
   2. Gruss-Lipper Biophotonics Center, Albert Einstein College of Medicine, New York, United States

   Contribution

   RHS, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   RHS: Reviewing editor, eLife.

• 3,042

  views

• 479

  downloads

• 129

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.