Single-molecule tracking of transcription protein dynamics in living cells: seeing is believing, but what are we seeing?

Abstract

A universe of transcription factors (TFs), cofactors, as well as chromatin remodeling and modifying enzymes combine or compete on chromatin to control transcription. Measuring quantitatively how these proteins dynamically interact is required in order to formulate models with predictive ability to elucidate transcription control mechanisms. Single molecule tracking (SMT) provides a powerful tool towards this goal: it is a fluorescence microscopy approach that measures the location and mobility of individual TF molecules, as well as their rates of association with and dissociation from chromatin in the physiological context of the living cell. Here we review SMT principles, and discuss key TF properties uncovered by live-cell SMT, such as fast turnover (seconds), and formation of clusters that locally increase activity.

The transcription cycle

Transcription starts with the binding of sequence-specific TFs to the promoter of a gene or enhancer sequences regulating its expression (Figure 1) [1]. In eukaryotes, nucleosome organization can impede access to cognate DNA sequences, requiring cross-talk between ‘pioneer’ TFs and chromatin remodeling and histone modification enzymes to increase site accessibility (e.g. DNAse hypersensitive sites; DHS) [2]. TFs also recruit a host of co-activator proteins, as well as components of the pre-initiation complex (PIC). The PIC is an assembly of proteins that act as a recruitment platform for RNA Polymerase II (Pol II), responsible for transcription of protein-coding genes. Once the PIC has fully assembled onto the promoter, transcription initiation starts with the phosphorylation of the C-terminal Domain (CTD) of Pol II at the serine 5 position. The CTD is a long unstructured domain that consists of repeats of the consensus motif YSPTSPS and is required for transcription activity and viability [3^••,4^•]. Shortly after initiation, metazoan Pol II undergoes a short pause ~50–80 bp downstream of the promoter. Release from this promoter-proximal pause requires phosphorylation of the CTD at the serine 2 position by the kinase CDK9, part of the P-TEFb complex. After exit from the pause, Pol II transcribes the gene while processing events occur on nascent RNA (capping, splicing, polyadenylation), transiently displacing downstream nucleosomes until termination.

Figure 1.

(a) Stages of the transcription cycle. Transcription occurs in a series of steps, starting with the binding of sequence-specific Transcription Factors (TFs), that recruit co-activators, chromatin remodelers (for instance depositing histone variants such as H2A.Z, marked with Z) and eventually members of the Pre-Initiation Complex (PIC). Upon formation of the complete PIC containing Pol II, transcription starts, marked by Ser5 phosphorylation of the CTD of Pol II. Upon release from the promoter-proximal pause (marked by phosphorylation of the CTD at Ser2), productive elongation of RNA transcripts. (b) Single-Molecule Tracking workflow. A Transcription Factor (TF) of interest is fused to the self-labeling tag HaloTag. Cells containing the fusion protein are labeled with a HaloTag ligand conjugated to a photoactivatable fluorophore, and unbound ligands are washed out. Upon a low intensity pulse of blue light excitation, ~1–20 TF molecules turn fluorescent (example image shows individual molecules of H2B-HaloTag labeled with PA-JF646; cell nucleus outlined in yellow). The motion of those molecules is subsequently recorded until they leave the focal plane or photobleach; a new cycle is initiated by another pulse of blue light that turns on a fresh subset of molecules. Subsequent analysis of molecule trajectories (x,y positions over time) enables classifying TF molecules into different mobility populations.

Multiple control points punctuating the transcription cycle provide numerous opportunities for regulation, opening the door to rich dynamic behaviors. Indeed, transcripts encoded by an individual gene are not produced as a steady stream, but rather synthesized during bursts of activity interspersed by long periods of inactivity [5^•]. A major challenge is to understand how transcription factors collectively generate complex bursting dynamics. As most players typically only dwell for seconds at the promoter [6^•], one key question is how relatively short-lived interactions can lead to expression programs that are sustained over days.

What can live-cell SMT reveal about transcription, and what are its limitations?

Single-Molecule Tracking (SMT) in living cells provides precise measurements of the location and motions of nuclear factors, complementing genome localization-sequencing technologies that can be subject to side-effects of chemical cross-linking [7]. It is highly technical, requiring experience in optical physics, biophysics, computer science, molecular genetics and cell biology [8,9]. SMT directly observes the kinetic behaviors of fluorescently labeled protein factors at an x–y resolution approaching the 10 nm-diameter nucleosomes on chromatin fibers (~25 nm × 25 nm) but considerably lower z-resolution (>70 nm). At a minimum, a wide field epifluorescence microscope with additional high laser power and ultrasensitive camera components monitors the location over time of an individual protein (spot) undergoing free diffusion in the nucleoplasm, or paused due to association with chromatin or other nuclear structures. Movies recording a temporal sequence of spots (a trajectory) within a circumscribed nuclear area – under conditions where neighboring spots are absent or suffi-ciently distant to avoid confounding identities – are limited to a few minutes because of fluorophore photo-bleaching and chromatin motions out of the focal plane, which leads to disproportionate loss of long trajectories. Nonetheless, one can extrapolate binding events beyond the SMT observable window by calibrating photo-bleaching kinetics [10], or by normalizing the measured lifetime distribution to a chromosomal histone with low bulk turnover outside of S-phase [11, 12]. Repeated imaging of multiple cells over an hour or two records a statistically significant number of molecules, usually >1000, to inform the overall distribution of kinetic behaviors of the population.

From raw trajectories at fast frame rates (30 ms and under), one can extract the number and frequency of dynamic states [13^•]. Because histones are known to be largely incorporated in chromatin, the slowest diffusive histone fraction provides a reference for the average diffusion coefficient D of bulk chromatin, although D values can vary depending on specific gene states or chromosomal locations (see below) [14]. Tightly associating chromatin and transcription factors may have similar diffusive properties as incorporated histones, but factor binding to other nuclear structures should also impede free diffusion. The use of DNA-binding or histone-binding mutants is necessary to assist functional assignment of diffusive populations [14]. In addition, transcription factors could exhibit a spectrum of diffusive behaviors between stably incorporated and free. Diffusion Coefficients alone do not fully capture the complexity of single-molecule behaviors, and complementary spatial metrics can be computed from trajectories in order to provide deeper insight into TF behavior. For instance, trajectories of chromatin-bound factors remain confined within a certain radius (in the 20–100 nm range) that tends to be shorter in nuclear lamin-associated domains compared to other chromatin regions [20,21,22^•,25], and observation or recurring motion may allow calculation of a spring constant [16]. Displacement anisotropy (i.e. the tendency of a trajectory to feature U-turns) is a signature of how a factor explores the nucleus: either in a compact fashion that involves local oversampling, or more global exploration [15]. Anisotropy biases that appear at distinct time intervals, as observed for CTCF, suggest that the underlying nuclear space is heterogenous [16]. Since individual molecules might sample different diffusive states in their trajectory (e.g. binding, followed by dissociation and free diffusion), the transitions between states can be identified using Hidden Markov Models (e.g. vbspt [17]), or more recently using deep learning approaches [18,19].

Slower tracking (100 ms and above) motion blurs freely diffusing molecules in order to preferentially capture the chromatin-bound fraction and measure stable and transient dwell times of chromatin-bound molecules. Together, these allow one to calculate the frequency and duration of TF occupancy [18]. Because of the complexity of the data and the stochasticity of the pro-cesses at play, computational modeling and simulations are required to predict how the spatiotemporal dynamics of multiple transcription-related components can produce a defined RNA output [3^••].

Visualization of the entire nuclear distribution of molecular behaviors by SMT can uncover novel subpopulations that would otherwise be masked by population averaging [9]. Furthermore, two-color, region-specific imaging (e.g. heterochromatin/euchromatin) and locus-specific marking and imaging with new target-lock [27^••] or orbital tracking optical technologies [28^••,29^••] point the way to visualizing dynamics of gene-specific chromatin and interacting factors (see chapter by BC Chen). Continuous improvements in labeling chemistry enabling brighter fluorophores with higher affinity and specificity for protein tags extend SMT precision and raise the temporal detection limit [30^•]. However, for abundant proteins such as histones or RNA polymerases, the necessity for sparse labeling to image individual molecules makes it difficult to simultaneously visualize molecular interactions between two different species. A constant concern for DNA damage imposed by short-wavelength laser light is partially alleviated by orange and red illumination [31], but SMT data interpretation is conditioned by other explicit and implicit biophysical assumptions that complicates biological conclusions. For example, intrinsic ‘blinking’ of a single fluorophore can produce apparent clusters from the same molecule, requiring a corrective computational approach [32^•]. Especially worrisome is the real possibility that fusion of a ~30 kD protein tag (chemically labeled or genetically engineered fluorescent protein) and/or protein overexpression can adversely perturb function and produce misleading diffusive behaviors [33,34]. The advent of gene editing by CRISPR technology now enables rigorous functional validation of protein fusions expressed as the sole source under natural promoter/enhancer control [11,35^•].

Dynamic chromatin binding and dissociation is compatible with high target occupancy

Recent SMT experiments have provided measurements on the residence times of chromatin-bound factors averaged across the entire nucleus. While most TF interactions with chromatin typically last <1 s (‘transient’), a small fraction display longer binding by an order of magnitude (‘stable’), often approximating an exponential decay with a half-life on the order of 2–20 s in yeast [28^••,36^•] and 10–20 s in metazoans [12,14,29^••,37-39,40^•]. Factors lacking a functional DNA-binding domain usually interact only transiently, suggesting that these represent non-specific interactions. Residence times are visualized using long exposure times in order to motion-blur chromatin-free molecules (100–500 ms per frame: slow tracking), while shorter exposures (10–50 ms per frame: fast-tracking) allow imaging of fast diffusing TFs through the nucleoplasm. These two regimes enable estimation of the search time t_search: the time it takes upon dissociation for a molecule to find the next accessible genomic target based on the classic facilitated diffusion model [14,41], which posits that TFs find their targets by sequential searches in 1D (sliding along the double helix for a duration of t_1D) and 3D (diffusing in nuclear space for a duration of t_3D). After N_Trials trials, the TF eventually finds a cognate site, where it resides longer (10–20 s). t_search = (N_Trials − 1) * (t_3D + t_1D) + t_3D. In contrast With other modes of nuclear exploration (e.g. 3D diffusion or 1D scanning alone), facilitated diffusion successfully predicts the high association rates observed experimentally [41]. Search times range from ~7 s in yeast [36^•] to ~6 min for a mammalian Sox2 molecule to find any of its ~7000 target sites in ES cells [14]. SMT of the lac repressor provided the first evidence supporting 1D sliding during facilitated diffusion in live bacteria [42], and a new in vitro study now reveals that LacI rotates as it slides on a microsecond time scale, hopping in and out of the DNA groove [43^•].

The short-lived stable residence time of TFs on chromatin measured by SMT seems at first paradoxical [6^•]: sequence-specific TFs display robust steady-state binding profiles in genomic assays such as ChIP-seq [14], and can drive developmental programs lasting days. Recent competition experiments for Ascl1, a TF in the nondividing Xenopus oocyte suggest a residence time of hours [44]. However, this does not necessarily conflict with SMT studies in cultured cells, because chromatin and transcription biology might differ between systems. Indeed, studies using endogenous knock-in of fluorescent tags validate that tags do not induce phenotypes, indicating that 10–20 s binding times are sufficient for TF function [11,35^•]. Orthogonal techniques imaging for hour-long periods (e.g. FRAP, Fluorescence Recovery After Photobleaching) also show that TF binding is highly dynamic, convergent with SMT results [11,45]. These findings argue that the 10–20 s stable residence times do not represent an artifactual temporal boundary set by SMT. Moreover, locus-specific imaging of yeast Gal4 molecules measures binding events lasting on average 12–35 s [28^••], peaking ~14 s before the onset of transcription, demonstrating that seconds-long interactions are sufficient to trigger transcription [28^••]. Collectively, SMT results indicate that binding for just a few seconds might be sufficient for a TF to efficiently initiate PIC recruitment, whereas interactions <1 s are unproductive. The stable dwell times of <5 s for most PIC components in yeast provide a consistent timescale for this process [101]. TF residence times can be regulated by signaling [28^••,29^••,39], and longer residence times lead to higher transcription output [29^••,37,38,102]. Interestingly, some TFs show a continuum of dwell times [12,29^••,46], which could reflect binding to a variety of DNA sequences with graded affinities.

At any given accessible, DNase hypersensitive (DHS) promoter, the TF occupancy, defined as % occupancy in a particular time period, and not necessarily the residence time, is likely the key parameter regulating transcription levels. For a simple binding reaction TF + DHS → TF*DHS, the equilibrium dissociation constant K_d = [TF] [DHS]/[TF*DHS] indicates that the concentration of the TF-promoter complex [TF*DHS], is proportional to [TF] and [DHS] and inversely proportional to K_d. The important contribution of SMT is to enable calculation of K_d in living cells from measurable rate constants (k_off /k_on = K_d), where k_off is the inverse of the residence time and k_on is inversely proportional to t_search, and N_sites the number of cognate DHSs in the genome. [TF] can be quantified roughly [47^•] and varies depending on expression level, nuclear import, nuclear volume, and inhomogeneity or local clustering near a promoter [37,48^•], while N_sites can be estimated from genomic technologies and is dependent on ‘pioneer’ TFs, and chromatin remodeling and modification enzymes regulating transitions between DHSs and canonical nucleosomes. Thus, TF occupancy depends on any combination of changes in factor concentration, number and accessibility of cognate sites, residence and search times, and additional influences such as TF association with a ligand partner or protein partner, or competitive displacement by a nuclear enzyme. Accordingly, even relatively short residence times (10–20 s) are compatible with target motifs being occupied a large percentage of the time, provided that there are compensatory mechanisms.

Protein clustering depends on IDRs in live cells: is phase transition the underlying mechanism?

The growing evidence that TFs cluster spatially as foci in the nucleus via multi-valent, low-affinity interactions between intrinsically disordered regions (IDRs), also called low complexity domains (LCDs), a ubiquitous feature of TFs, suggests a mechanism to increase local concentration and thus drive up occupancy at key regulatory sites [27^••,49^••,50^•,51^•,52-58]. Clustering near DNA or in the nucleoplasm may foster local niches or ‘hubs’ with unique diffusive properties [23,59^••]. Promoter/Enhancer clustering increases the effective size of a target site, potentially attracting other TFs harboring IDRs, although the rules governing the specificity or promiscuity of IDR–IDR interactions are not well understood [3^••,35^•,49^••,52,60^•]. Clustering would likely provide an extra assist to the search process beyond 1D scanning and 3D diffusion [6^•]

Short-lived clustering (seconds to minutes) [35^•,50^•,61] parallels the propensity of IDRs to phase separate in vitro into ‘condensates’ or droplets [49^••,50^•,51^•,52,62^•,63^•,64]. However, in vitro phase separation assays often require TF concentrations orders of magnitude higher than physiological levels, and thus as a group, should be interpreted with caution, since alternative mechanisms can also increase local TF concentration at physiological levels in the cell, for example, locally enhanced DNA binding [59^••,65^•,66^•,67].

The CTD of Pol II offers an important paradigm of how clustering impacts transcription [53]. CTD propensity to cluster scales with its length [62^•]. CTDs shorter or longer than their WT lengths display dramatic phenotypes, while truncated CTDs fused to an unrelated IDR are functional, suggesting that the transcription machinery operates in a narrow range of clustering potential [3^••,4^•]. CTD phosphorylation regulates Pol II affinity for itself and other IDR-containing partners such as Mediator [53,62^•,64,68^•]. These observations suggest that transient and stable Pol II clusters [27^••,50^•] are recruited via CTD interactions, and that subsequent CTD phosphorylation frees elongating Pol II molecules [69]. The observation that the number of mRNAs produced per burst scales with CTD length and Pol II cluster size further support this model [3^••,61]. Of interest, nuclear actin, whose function has long been speculated, is implicated in Pol II clustering [70], while nuclear myosin movement on actin filaments appears to facilitate long-range chromosome rearrangements during transcription [20,71].

For other chromatin regulators, conserved IDR residues of the Cbx2 subunit of the Polycomb Repressive Complex PRC1 play a key role in forming foci in living cells and condensates in vitro [72]. SMT studies reveal dynamics of PRC1 and PRC2 subunits [40^•,73], showing stable dwell times of ~10 s, comparable to mammalian TFs [40^•], An unexpected low level of PRC1 site occupancy argues against a physical mode of repression [74^•]. Similarly, the groucho/TLE family of transcription repressors exhibits puncta in living Ciona with properties of phase-separated condensates [75], and the silencing factor Sir3 promotes long-range contacts between distant loci [76]. Alanine or glutamine repeat expansions in IDRs of human sequence-specific transcription factors and the TATA-Binding Protein TBP change the capacity to form phase-separated condensates in vitro, and alter the composition of heterotypic puncta in live cells [77^•]. The HP1 heterochromatin protein oligomerizes via interactions between IDRs and forms phase-separated droplets in vitro [78,79], but another study reports a weak propensity for droplet formation in cells and proposes an alternative polymer collapse model [65^•,80].

Local and global chromatin movements on short and long timescales

Even when chromatin-bound, individual transcription factor molecules exhibit a wide range of mobility and confinement behaviors [24^•], possibly reflecting local dynamics of chromatin sites. Chromatin mobility can be assessed locally, by fluorescent labeling of a genomic locus [81,82^••], or globally, by tracking fluorescent histones [24^•,83^•,84^•]. Transcription activity at a locus generally correlates with increased mobility at short timescales (<5 s) and is dependent on Pol II initiation and elongation activity [82^••]. However, an ectopically integrated locus responding to Estrogen Receptor activation shows constrained mobility at longer timescales (>5 s) independent of Pol II elongation [81].

Global tracking of chromatin motion provides a complementary view. An approach termed Dense Flow reConstruction and Correlation (DFCC) shows that at short time scales, small chromatin domains move independently from one another, while at longer time scales (>10 s) micron-scale domains (up to the size of an entire chromosome) appear to move coordinately [85]. Nucleosomes within lamin-associated and heterochromatic regions move slower and are more confined [24^•]. Counterintuitively, shutting down transcription globally induces an increase in histone mobility [83^•,86^•]. These apparently conflicting observations may be reconciled by a model where active transcription – and associated chromatin remodeling [26,67] – stiffens the periphery of chromosomal domains decorated with RNA [87-89], which constrains the motion of internal heterochromatin domains, forcing large regions (possibly entire chromosomes) to move in sync. Blocking transcription releases that motion constraint. A caveat common to many chromatin tracking experiments is that ectopically expressed fluorescent H2B might display behavior reflecting incorporation into chromatin domains that do not fully represent endogenous H2B synthesized in S-phase. Indeed, DNA fluorescently labeled by intercalating dyes displays distinct nuclear mobility from a fluorescent H2B fusion expressed in mammalian cells [85]. Ideally, a fluorescent H2B fusion should be the sole source of the histone, expressed under natural promoter control as shown for yeast [84^•] but this is technically challenging for expression in metazoans, due to multiple H2B gene copies.

Future directions: temporal order of events and convergence with in vitro SMT biochemistry

Despite recent progress in SMT at select loci [27^••,28^••,29^••,61], sparse labeling constraints make it impractical so far to measure interactions between different protein species at the single-molecule level. Future progress in tracking two or more factors simultaneously will help address longstanding questions of how accessible promoters and enhancers integrate inputs from multiple TFs [90], the order of in vivo assembly of the transcription machinery, and how promoter-enhancer dynamics [91^•,92^•,93^•], TF binding [28^••,35^•,94], and chromatin modification or remodeling [26,67,84^•,95^•,96,97^•] are dynamically coupled with transcription activation. In vitro approaches using DNA or reconstituted nucleosome templates offer a tantalizing glimpse of how multi-color SMT and well-controlled experimental perturbations can deepen understanding of fundamental transcription mechanisms lying beyond the reach of current live-cell imaging [98^•,99,100]. Furthermore, recapitulating the crowded nuclear environment and native chromatin templates for SMT in vitro will provide powerful means to uncover key reaction intermediates transiently populating the transcription pathway. This is an inescapable challenge for the future.