Imaging Organization of RNA Processing within the Nucleus

Abstract

Within the nucleus, messenger RNA is generated and processed in a highly organized and regulated manner. Messenger RNA processing begins during transcription initiation and continues until the RNA is translated and degraded. Processes such as 5′ capping, alternative splicing, and 3′ end processing have been studied extensively with biochemical methods and more recently with single-molecule imaging approaches. In this review, we highlight how imaging has helped understand the highly dynamic process of RNA processing. We conclude with open questions and new technological developments that may further our understanding of RNA processing.

Nuclear organization occurs at several levels, ranging from cellular to molecular. At the cellular level, the nucleus must reestablish itself with each cell division. Additionally, large-scale nuclear reorganization, such as flexing of the envelope in response to cellular stress or formation of nuclear lobes during cell differentiation, must occur through dynamic processes (for review, see Newport and Forbes 1987). At an intermediate scale, chromosomes and chromatin are spatially organized to regulate gene expression at different loci (Lieberman-Aiden et al. 2009). This organization is dynamic, allowing chromosomes to restructure during the process of differentiation and to permit phenomena such as tissue-specific gene expression. At the molecular level, RNA processing machinery is assembled in highly stochastic yet ordered processes. With each new RNA molecule being synthesized, protein factors assemble and disassemble to dictate all aspects of an RNA's life (for review, see Tutucci et al. 2018a). With assistance from single-molecule imaging techniques, the order and dynamics of this process is clear for multisubunit complexes such as the transcription preinitiation complex (PIC) and the spliceosome (DeHaven et al. 2016).

In this review, we will discuss how single-molecule imaging has shed light upon organization of RNA processing. We begin by introducing the single-molecule imaging techniques and their strengths and limitations. Then we sequentially discuss the steps of RNA processing, transcription, splicing, 5′ capping, polyadenylation, and nuclear export (Fig. 1). We will focus on how single-molecule imaging has furthered our understanding of these dynamic processes.

Figure 1.

Overview of RNA processing within the nucleus. (A) Steps of nuclear RNA processing. Blue, sequences involved in RNA processing. Pink, core components of the processing machinery. Green, factors that are loaded onto the transcript as a result of RNA processing. Beginning from upper left and moving clockwise: DNA (black lines) is organized into domains that are actively relocated throughout the nucleus. Reorganization of DNA through factor deposition or chromatin opening (pink ovals) allows for assembly of the preinitiation complex (PIC) (pink circles) at the promoter (blue arrow). After assembly, RNA polymerase II is recruited, initiates, and pauses after approximately 100 nucleotides. After pause release, capping occurs shortly after the 5′ end of the RNA exits RNA polymerase II. Cotranscriptional loading of splicing factors (green ovals), cleavage, and polyadenylation machinery occurs as RNA polymerase progresses through the transcript. After the transcript is released from the transcription site, it diffuses through the nucleoplasm until it finds a pore (pink ovals) suitable for export. Shortly after export, ribonucleoprotein (RNP) remodeling occurs. (B) (Left) Factors involved in RNA processing. (Right) The carboxy-terminal domain (CTD) of RNA polymerase II provides a platform for coupling transcription elongation to other RNA-processing events. The phosphorylation state of the CTD changes to permit coupling (right downward arrows). (Panel created from data in Hocine et al. 2010.)

INTRODUCTION TO TECHNIQUES

New imaging technologies have allowed for the observation of gene expression, both at the single-cell and single-molecule level (for review, see Vera et al. 2016; Biswas et al. 2018; Tutucci et al. 2018a; Sato et al. 2020). This review will highlight findings from techniques including single-molecule fluorescence in situ hybridization (smFISH), MS2/PP7 RNA imaging, fluorescence recovery after photobleaching (FRAP), and single-particle tracking (SPT) (Fig. 2). Each of these techniques has unique advantages for measuring different aspects of transcription dynamics (Liu et al. 2015).

Figure 2.

Imaging methods to study RNA processing. (A) Imaging splicing in cells. (Top left) To visualize splicing in living cells, two different RNA stem loop cassettes (red PP7 and green MS2 boxes, respectively) are engineered into a respective intron and exon. During transcription, both the intron and exon are visualized by binding of different fluorescent coat proteins (PP7 coat protein-mCherry and MS2 coat protein-green fluorescent protein [GFP]) to the stem loop arrays. After splicing occurs and the intron is removed, the mature mRNA is labeled with a single-stem loop array. (Top right) By tracking the color change at the transcription site, splicing kinetics can be measured in living cells. As the RNA is synthesized, increasing signal from the first intron (red line) occurs prior to the second exon (green line). A sudden loss of intronic signal (red line, downward arrow) occurs when the intron is no longer colocalized with the exon. (Bottom left) To visualize splicing in living cells, two sets of single-molecule fluorescence in situ hybridization (smFISH) probes (red and green circles, respectively) are designed against one intron and exon. During transcription, both the intron and exon are visualized as dual-colored smFISH spots. After splicing occurs and the intron is removed, the mature mRNA is labeled with a single set of smFISH probes. (Bottom right) Dual-colored smFISH can find overlapping diffraction-limited spots at the transcription site with the red signal corresponding to the first intron and the green signal corresponding to the second exon. (B) Imaging alternative polyadenylation in cells. (Top) Multicolor smFISH can be used to label individual isoforms in either two colors (long isoform) or one color (short isoform). Two sets of smFISH probes (red and green circles, respectively) are designed against the conserved coding sequence (thin line) and long isoform-specific sequence (thick line). Short isoforms are singly labeled (green only) and long isoforms are dual colored (green and red probes). (Bottom) In cases where alternative polyadenylation leads to differing localizations, abundances, or stabilities, two-color smFISH identifies these differences within cells.

Subcellular and single-molecule resolution of RNA processing was first achieved with smFISH (Femino et al. 1998). By using multiple dye-conjugated DNA oligos that hybridize to an RNA of interest, specific RNA molecules can be imaged within the cell as diffraction-limited spots (Fig. 2A; Femino et al. 1998; Raj et al. 2006). Transcription sites appear as nondiffraction limited and more intense spots because multiple nascent RNAs are being synthesized along the gene. Complementary oligos can be designed against the coding sequence, introns, or alternative untranslated regions (UTRs) to study different aspects of RNA processing (Fig. 2).

Multicolor smFISH (Femino et al. 1998; Levsky et al. 2002; Shah et al. 2018) is a powerful tool to image the process of splicing and alternative polyadenylation (APA). By labeling the introns with probes fluorescing in one color and the exons in another color, precursor mRNAs that contain introns are visualized as dual-colored spots while mature mRNAs that do not contain introns are single-colored spots (Fig. 2A). In the case of cotranscriptional splicing, FISH probes against the intron will localize to the nascent RNA transcript and colocalize around the brighter transcription site (Levsky et al. 2002; Shah et al. 2018). When long and short 3′ UTRs are produced from the same gene by APA, they too can be distinguished by two-color smFISH. Specifically, one set of smFISH probes detects the shared regions of the two isoforms while a second set of smFISH probes detects the unique region of the long mRNA isoform. As a result, the short mRNA isoform is only detected by the shared smFISH probes (single-colored) while the long mRNA isoform is detected by both sets of smFISH probes (dual-colored) (Fig. 2B).

To image RNAs in living cells, stem loop sequences from bacteriophages (MS2, PP7, lambda; for review, see Tutucci et al. 2018a) are inserted into the gene of interest (Bertrand et al. 1998; Daigle and Ellenberg 2007; Chao et al. 2008; Tutucci et al. 2018b). Stem loop cognate proteins, modified viral capsid proteins, are genetically fused to a fluorescent protein and coexpressed in the same cells. Multiple capsid proteins fluorescently label the RNA after binding to the stem loop sequences. As a result, each RNA molecule can be tracked as it moves throughout the cell (Bertrand et al. 1998; Tutucci et al. 2018b). RNA imaging can track the appearance and maturation of nascent RNA molecules at endogenous transcription sites in live cells in real time (Larson et al. 2011; Lionnet et al. 2011) and in living animals (Park et al. 2014; Nwokafor et al. 2019). Multiple orthogonal stem loop sequences can be used to interrogate processing of the RNA in real time (Hocine et al. 2013; Martin et al. 2013; Coulon et al. 2014).

Live-cell splicing kinetics have been studied by tagging either introns or exons with stem loop/capsid protein systems and measuring both the fluctuations and residence time of RNAs at the transcription site (Fig. 2A). The kinetics of splicing can also be observed on individual transcripts using two orthogonal stem loop/capsid protein systems to tag different introns and exons. When one set of stem loops is present within the intron and the second is within a constitutive exon, the diffraction-limited spot will lose one set of stem loops during the process of splicing and therefore go from dual-colored to single-colored (Coulon et al. 2014). Alternative configurations of stem loop arrays have also allowed for dynamic observation of translation (Halstead et al. 2015) and RNA decay (Horvathova et al. 2017).

To image individual proteins in vivo, fusions of fluorescent protein (for review, see Lippincott-Schwartz et al. 2001; Snapp 2005) and more recently ultrabright and photostable protein tags (Keppler et al. 2004; Gautier et al. 2008; Encell et al. 2012; Tanenbaum et al. 2014; Grimm et al. 2015, 2020) can be used. Methods such as FRAP or fluorescence correlation spectroscopy (FCS) can determine the diffusion coefficients of both protein and RNA (Wu et al. 2012, 2015) within cells. For example, FRAP can be used to probe changes in diffusion and the chromatin-binding residence time of a collection of factors including transcriptional activators as shown during the heat shock response in Drosophila polytene cells (Yao et al. 2006). With the advent of long-term SPT, proteins can be followed as they bind, unbind, and perform their regulatory activities. SPT can examine both the diffusion and stable binding of a single transcription factor on target sites in vitro or even on the genome in live mammalian cells (Chen et al. 2014; Teves et al. 2016).

TRANSCRIPTION INITIATION—A COMPLEX POTPOURRI OF FACTORS

Over last 50 years, biochemical experiments have identified and characterized most of the factors required for initiating transcription (Cramer 2019; Roeder 2019). Early steps in this process involve the formation of a PIC, consisting of over 80 different proteins (e.g., TBP/TFIID and TFIIA/IIB/IIE/IIF/IIH) along with Mediator and RNA polymerase II (RNA Pol II). Factors comprising the PIC coordinately assemble in a highly ordered fashion onto a ∼60 base pair stretch of the core promoter (Lemon and Tjian 2000). Once the PIC is assembled, the carboxy-terminal domain (CTD) of the largest subunit of RNA Pol II must be phosphorylated at select residues for RNA Pol II to escape the promoter region (CTD-Ser5ph) and begin elongating downstream (CTD-Ser3ph). During promoter escape, the PIC is partially disassembled when RNA Pol II disengages from interacting factors such as TFIIB and TFIIH. As RNA Pol II transcribes further into the coding sequence, the phosphorylated CTD of RNA Pol II serves as a landing platform for numerous factors involved in subsequent RNA processing of the mRNA including splicing and polyadenylation (Saldi et al. 2016).

Now that most of the players involved in transcription initiation have been revealed, researchers have begun unraveling the dynamics of gene regulation. Recent development of bright organic fluorescent dyes (Grimm et al. 2015, 2020), orthogonal labeling systems (e.g., HaloTag, SNAP-tag, CLIP-tag) (Keppler et al. 2004; Gautier et al. 2008; Encell et al. 2012), and advanced microscopes (e.g., multifocus, lattice light-sheet, and target lock stimulated emission depletion [STED] microscopy) (Abrahamsson et al. 2013; Liu et al. 2014; Li et al. 2019) allow transcription dynamics to be visualized in live cells. When combined with live-cell RNA-labeling technologies (e.g., MS2/PP7 tagging) (Hocine et al. 2013) and CRISPR gene editing (Spille et al. 2019), it is now possible to track the activity of a transcription factor at a single transcribing endogenous gene inside a live cell (Donovan et al. 2019; Li et al. 2019).

TRANSCRIPTION INITIATION DYNAMICS—INEFFICIENT AND BURSTY

Over the last 20 years, single-molecule imaging studies have described the dynamics and efficiency of transcription initiation. In eukaryotes, transcription was determined to be a stochastic process marked by short periods (minutes) of intense activity or bursts of transcription interspersed between long intervals (>30 min to hours) of inactivity (for review, see Rodriguez and Larson 2020; Tunnacliffe and Chubb 2020). The cyclical turning on and off of transcription initiation suggests that transcription factors controlling this process must also act in a similar manner. Indeed, transcription factors rapidly cycle on and off the genome in hubs or clusters of activity (Cisse et al. 2013; Liu et al. 2014; Tsai et al. 2017; Chong et al. 2018; Mir et al. 2018). Furthermore, the clustering dynamics of RNA Pol II has been proposed to dictate the dynamics of transcription initiation (Cho et al. 2016). Therefore, most if not all transcription factors dynamically bind and unbind in hubs of activity reinforcing the cyclical nature of gene-specific transcription initiation and bursting.

Transcription initiation is an inherently inefficient process both in vitro and in vivo (Darzacq et al. 2007; Revyakin et al. 2012). Both FRAP of a gene array and single-molecule imaging of an endogenously tagged locus indicate that ∼1% of RNA Pol II–binding events at a gene array leads to productive transcription initiation (Darzacq et al. 2007; Li et al. 2019). Transcription initiation by RNA polymerase I (RNA Pol I) and III (RNA Pol III) require fewer factors (for review, see Engel et al. 2018). Yet surprisingly, live-cell imaging studies also indicate that RNA Pol I transcription initiation is also inefficient (∼1%–11%) in vivo (Dundr et al. 2002). This is likely due to the complexity and dynamic instability of PIC assembly. Adding to the complexity, transcription factor–binding sites in enhancers and promoters exist within chromatin, which mitigates the assembly of the PIC. Complementary genomic studies in yeast show that rapid conditional removal of individual general transcription factors (GTFs) from the nucleus severely compromises the stability of the PIC (Petrenko et al. 2019). This inherent instability of partial PIC assemblies likely leads to a low probability for complete PIC assembly and an overall low efficiency of transcription initiation in vivo.

TRANSCRIPTION STABILITY DYNAMICS AT PROMOTER—RAPID AND MULTISCALE

In contrast to textbook schema, in vitro and live-cell single-molecule tracking studies indicate that transcriptional activators, chromatin-associated factors, and GTFs (e.g., Sox2, glucocorticoid receptor [GR], p53, BRD4, and TFIIB) typically cycle on and off their genomic targets on the order of seconds (Fig. 3A; Chen et al. 2014; Morisaki et al. 2014; Zhang et al. 2016; Coleman et al. 2017; Li et al. 2019). Other factors, such as TBP/TFIID, still cycle on and off, but bind the core promoter DNA for much longer periods of time (∼100 sec) (Revyakin et al. 2012; Zhang et al. 2016; Coleman et al. 2017; Teves et al. 2018). The enhanced dynamics of activators likely reflects their roles in recruiting multiple factors (TFIIA, TFIIB, TFIIF, Mediator, and RNA Pol II) onto a stable TFIID/core promoter scaffold to complete PIC formation that is then competent for transcription initiation (for review, see Roeder 2019). The relatively unstable interaction of TFIIB (∼1.5 sec) with a stable TFIID/promoter DNA complex (minutes) likely facilitates multiple rounds of transcriptional reinitiation within very short time periods (e.g., seconds) via successive recruitment of RNA Pol II onto a semi-stable partial PIC assembly.

Figure 3.

Dynamics of transcription initiation and histone eviction. (A) Imaging the dynamics of preinitiation complex (PIC) formation, RNA Pol II convoy formation, and transcriptional bursting. Fluorescently labeled transcriptional activators and PIC components (general transcription factor [GTFs]/TBP/TFIID) (orange circles) dynamically assemble onto enhancer regions (upstream activation sequence [UAS]) and the transcription start site (TSS) of the core promoter (blue boxes). The different stabilities of the factors at the core promoter allows rapid successive recruitment of RNA Pol II (pink circles) that forms convoys during a transcriptional burst. As RNA Pol II elongates past the MS2/PP7 repeats in the gene (green box), the emanating RNA forms stem loops that are subsequently recognized by green fluorescent protein (GFP)-tagged MCP/PCP coat proteins. (B) Imaging the temporal acetylation of histones, phosphorylation of RNA Pol II carboxy-terminal domain (CTD), and disruption of a histone variant during transcription initiation. After addition of hormone, fluorescently labeled Fab antibody fragments (green ovals) indicate histone H3K27 acetylation precedes FP-glucocorticoid receptor binding (yellow star) at a gene reporter. Subsequent recruitment of RNA Pol II to the TSS leads to phosphorylation (ph) of Serine 5 (Ser5) of the RNA Pol II CTD. Upon promoter escape, Serine 2 (Ser2) within the RNA Pol II CTD is phosphorylated. Disruption of a fluorescently labeled H2A.Z (blue circle) containing nucleosome immediately downstream of the TSS requires phosphorylation of RNA Pol II CTD Ser5. (GR) Glucocorticoid receptor, (Trxn act.) Trxn activation.

Consistent with such a model is the observation of RNA Pol II convoys (sometimes consisting of 19 polymerases) that rapidly load onto the core promoter during a transcriptional burst (Tantale et al. 2016). During convoy formation, RNA Pol II can rapidly load every 4 sec likely owing to the dynamic assembly and disassembly of GTFs on a stable TFIID/core promoter scaffold (Fig. 3A). RNA Pol II convoys last for ∼100 sec followed by a brief period of inactivity (∼100 sec) where presumably the factors required for reinitiation are reloaded onto the core promoter. It is tempting to speculate that the length of RNA Pol II convoy formation is dictated by the overall stability of TFIID on the core promoter given equivalent timescales (∼100 sec). Eventually, successive rounds of RNA Pol II convoys stop forming and the promoter shuts down for longer periods of time (>30 min to hours) where presumably a more repressive chromatin state is reestablished on the enhancer, core promoter, and gene body.

TRANSCRIPTION INITIATION PTMs AND HISTONE EVICTION—AN ORDERED PROCESS

Fluorescently labeled antigen-binding fragments (Fabs) can image select posttranslational modifications (PTMs) such as histone marks and changes in the CTD of RNA Pol II in live cells (FabLEM) (Hayashi-Takanaka et al. 2009, 2011). FabLEM showed that histone H3K27 acetylation preceded RNA Pol II CTD phosphorylation during zygotic genome activation in live zebrafish embryos (Sato et al. 2019). Using green fluorescent protein (GFP)-tagged GR, RNA Pol II CTD phosphorylation onto a gene array was found to contain hyperacetylated histone H3K27 before hormone stimulation (Fig. 3B; Stasevich et al. 2014). Sorting cells with differing amounts of H3K27 acetylation at the gene array showed that histone hyperacetylation enhanced levels of GR recruitment and elongating RNA Pol II containing the CTD phosphorylated at Serine 2.

Histone eviction at the promoter region may also affect transcription initiation. In yeast, the +1 nucleosome contains the histone variant H2A.Z immediately downstream of the transcription start site (TSS) (Fig. 3B; Albert et al. 2007). H2A.Z is rapidly turned over and linked to a chromatin state that is permissive to transcription initiation (Weber and Henikoff 2014). Previously, it was thought that PIC formation led to displacement of H2A.Z at the promoter (Tramantano et al. 2016). In contrast, live-cell SPT imaging using fluorescently tagged H2A.Z combined with a rapid conditional depletion system (Haruki et al. 2008) found that incorporation of RNA Pol II into the PIC per se was not sufficient for H2A.Z eviction from the genome (Ranjan et al. 2020). Rather, downstream events were required since conditional depletion of CDK7 that phosphorylates Serine 5 of the RNA Pol II CTD led to enhanced stabilization of fluorescent H2A.Z on the genome (Fig. 3B). This finding suggested that promoter escape of RNA Pol II was necessary to displace the first nucleosome that RNA Pol II encountered after leaving the TSS. Future studies may combine FabLEM and conditional depletion of factors alongside SPT and MS2/PP7 RNA imaging to examine chromatin remodeling and transcription initiation during transcriptional bursting.

RNA SPLICING

The process of mRNA splicing is an important step of the mRNA maturation process. The biochemical mechanisms of spliceosome function have been elucidated by both biochemistry and structural biology (Will and Lührmann 2011; Wilkinson et al. 2020). The splicing machinery (the spliceosome) is composed of five small nuclear RNAs (snRNAs) and several core proteins that are assembled into small nuclear ribonucleoprotein (RNP) particles (U1, U2, and the U4/U5/U6 tri-snRNP) as well as the protein Prp19-complex (NineTeen complex [NTC]). Recognition of the 5′ splice site occurs by binding to U1; subsequently the branch site sequence is recognized by U2 and the 3′ splice site by U2 auxiliary factor (U2AF).

In the past 15 years, single-molecule imaging methods were widely applied to the field of splicing. Single-molecule tracking of snRNPs has been possible using cell extracts (Crawford et al. 2008; Hoskins et al. 2011) and inside living cells (Grünwald et al. 2006). Below, we summarize the mechanistic insights of splicing uncovered by single-molecule imaging methods.

DYNAMIC AND REVERSIBLE SPLICEOSOME ASSEMBLY

Fixed cell imaging has been able to colocalize intronic RNAs with different components of the splicing machinery (Zhang et al. 1994; Brody et al. 2011). Dynamics of the assembly process, such as residence time and order of assembly, can be observed using fluorescently tagged snRNP proteins. FCS was used to calculate the lifetimes of the free diffusing and bound populations of different snRNPs. Ninety percent of the associations between splicing factors and pre-mRNA was brief (15–30 sec) (Huranová et al. 2010) while the remaining 10% of the associations was significantly longer than 30 sec. Further investigation of spliceosome assembly using single-molecule imaging found assembly of the spliceosome on pre-mRNA, while ordered, was also highly reversible (Hoskins et al. 2011).

Like the stochastic nature of other complex assembly processes within a cell (such as the PIC), association of a snRNP with the pre-mRNA does not commit the spliceosome; in fact, each complex has multiple on and off binding events on an individual pre-mRNA before the pre-mRNA is committed to undergoing splicing. As these events progress and the spliceosome assemble, the probability of splicing increases (Hoskins et al. 2011). Further work showed that a key step in splicing, formation of the lariat, did not occur during the early assembly stage of splicing. Rather, the 5′ splice site and the branchpoint sequence only came together after the NTC has bound, thus providing a checkpoint for proper assembly (Crawford et al. 2013). Spliceosomal assembly is therefore a highly dynamic and stochastic process with commitment to splicing occurring after proper assembly (for review, see DeHaven et al. 2016). Similar to the PIC, low efficiency of splicing may arise from the large number of factors that are involved and their stochastic assembly/disassembly. Additionally, questions remain about how inefficient splicing processes can be coordinated with the movement of RNA Pol II.

SPLICING AND TRANSCRIPTION KINETICS—CO- OR POSTTRANSCRIPTIONAL

The highly dynamic process of spliceosomal assembly must be performed with each newly transcribed intron. The kinetics of this process may lead to different splicing rates across cells and individual transcripts. To understand how transcription, spliceosome assembly, and proper splicing are interrelated, the process of splicing has been imaged in both fixed and live cells.

Two predominant splicing modes—co- and posttranscriptional splicing—have been observed. RNAs are either efficiently spliced at or near the transcription site or spliced after pre-mRNAs are released into the nucleoplasm (Vargas et al. 2011).

Cotranscriptional RNA splicing may be facilitated by binding of the splicing machinery to the CTD of the RNA Pol II (Das et al. 2007; Brody et al. 2011). Reverse transcription polymerase chain reaction (RT-PCR) measurements of several nascent RNAs point toward cotranscriptional splicing being present in a wide variety of terminal introns (Schmidt et al. 2011). While many introns have been visualized to undergo cotranscriptional splicing using RNA FISH (Levsky et al. 2002; Shah et al. 2018), live imaging (Sheinberger et al. 2017; Wan et al. 2019), and PCR-based methods (Schmidt et al. 2011), the levels of posttranscriptional splicing vary among transcripts (Schmidt et al. 2011).

Cases of posttranscriptional splicing have been described. Disruption of the splicing machinery, either through chemical inhibition (Schmidt et al. 2011), knockdown of splicing factors (Waks et al. 2011), mutations of splicing factors (Coulon et al. 2014), mutation of the splice site (Schmidt et al. 2011), or RNA structural changes at splicing factor-binding sites (Vargas et al. 2011), all have shown to contribute to increased, unprocessed transcripts within the nucleoplasm that eventually are spliced prior to export.

In the absence of splicing perturbation, is cotranscriptional splicing de facto for all transcripts? It is likely that several competing rates (RNA Pol II elongation rate, splicing rate, pausing and transcription termination rate) can contribute to either co- or posttranscriptional splicing. For instance, placing the MS2 array with a strongly spliced intron upstream of a slowly processed polyadenylation sequence may strongly increase the chances of cotranscriptional splicing (Schmidt et al. 2011).

Alternative scenarios may require posttranscriptional splicing. For example, increasing the number of introns within a reporter RNA can lead to accumulation of nascent, unspliced transcripts at the transcription site (Brody et al. 2011). Imaging showed that this increase in nascent transcript was not due to increased levels of polymerase but rather prolonged splicing as it was coincident with increased loading of splicing factors but not RNA Pol II and reversed by splicing inhibition. This suggests that when excess introns are present, the excess transcripts that are laden with spliceosomal components remains close to the transcription site as it completes processing (Brody and Shav-Tal 2011; Brody et al. 2011). The presence of partially spliced transcripts near the transcription site may be a generalized finding. By combining smFISH and expansion microscopy, it was observed that newly synthesized RNAs dwell near the transcription site after transcription. This allows them to complete and undergo continuous splicing as they traverse a zone of ∼300 nm around the transcription site (Coté et al. 2020). For a given RNA, it is likely that both co- and posttranscriptional processing occur with characteristics of the gene tilting the scales in favor of one or the other. Using two-color, live-cell imaging combined with fluctuation correlation analysis revealed a kinetic competition between splicing and transcription with posttranscriptional splicing occurring more often than cotranscriptional splicing (Coulon et al. 2014).

The race between polymerase and the splicing machinery contributes to heterogeneity of splicing. On a case-by-case basis, single-molecule methods are most apt to tease apart the individual contributions of co- and posttranscriptional splicing. Particularly exciting are transcriptome-wide, single-molecule imaging and sequencing methods that may allow for higher throughput interrogation of endogenous splicing rates (Oesterreich et al. 2016; Herzel et al. 2018; Shah et al. 2018).

MEASURING THE DYNAMICS OF TRANSCRIPTION ELONGATION AND 3′ END PROCESSING

Transcription elongation and termination dynamics have been measured in vivo using both sequencing- and imaging-based approaches (for review, see Jonkers and Lis 2015). While imaging RNA Pol II in single cells has the advantage of high spatial and temporal resolution, it is difficult to distinguish RNA Pol II molecules that are transcribing a gene of interest from the rest of the RNA Pol II molecules in nucleoplasm. Two methods overcome this challenge by amplifying RNA Pol II signal at a locus. The first method is to randomly integrate tens or hundreds of the same gene at a gene locus, forming a tandem gene array. The large number of fluorescent RNA Pol II molecules that are recruited to the tandem gene array appear as a bright spot (Janicki et al. 2004). By applying FRAP or fusing RNA Pol II to photoactivated GFP, the dynamics of transcription elongation (Boireau et al. 2007; Darzacq et al. 2007), splicing (Brody et al. 2011; Martin et al. 2013), and 3′ end processing (Boireau et al. 2007) have been determined. The second method is to use polytene chromosomes, which are large chromosomes in Drosophila salivary glands and have thousands of DNA strands (see the section Cotranscriptional Loading of Factors). The transcription of HSP70 was used to determine the transcription dynamics because its expression is inducible by heat shock (Yao et al. 2006; Ardehali et al. 2009; Buckley et al. 2014). The aforementioned methods both require gene multiplication; therefore, measurements are averaged across the gene cluster.

More recently, endogenous gene expression has been imaged in real time (Larson et al. 2011). The transcription dynamics were determined using a mathematical model that analyzes the fluctuation of fluorescently labeled nascent mRNAs. By tagging the longest yeast gene, MDN1, with PP7 stem loops, the prolonged RNA Pol II transcription elongation time can be accurately measured. The dwell time of a fluorescently labeled MDN1 mRNA at its transcription site equals the summation of two time spans: (1) the transcription elongation time for RNA Pol II to finish transcribing the part of MDN1 downstream of PP7 stem loops, and (2) the 3′ end processing time. When the PP7 stem loops are inserted at the 5′ or 3′ UTR of MDN1, the difference of the mRNA dwell times at the transcription site equals the time to transcribe MDN1 open reading frame (ORF). Thus, transcription elongation and 3′ end processing time were determined.

Using the above methods, RNA Pol II elongation rates range between 14 nt/sec and 61 nt/sec (Hocine et al. 2013). The 3′ end processing time ranges between 30 sec and 70 sec. Sequencing-based methods showed that the RNA Pol II elongation rate varies between genes and within the same gene. For example, RNA Pol II pauses 30 to 60 bp downstream of the transcription start site, accelerates through the gene body, and again pauses near the termination site (for review, see Kwak and Lis 2013; Proudfoot 2016; Mayer et al. 2017). Such variability may explain the wide range of results determined by different methods.

MONITORING mRNA ISOFORMS AS A RESULT OF ALTERNATIVE POLYADENYLATION

About 60% of genes in mammalian cells have more than one transcription termination site. As a result, these genes produce mRNA isoforms with different 3′ UTRs, the process called APA. The genomic landscape of APA has been determined by microarray- and deep-sequencing-based approaches in multiple model organisms (microarray-based methods [Flavell et al. 2008; Sandberg et al. 2008; Ji and Tian 2009; Ji et al. 2009] and deep-seq-based methods [Ozsolak et al. 2010; Shepard et al. 2011; Derti et al. 2012; Ulitsky et al. 2012; Mata 2013; Wang et al. 2018]). Different mRNA isoforms produced by APA can have different stability, localization, or altered protein-coding regions. Thus, APA regulates protein expression and cellular function (Sandberg et al. 2008; Mayr and Bartel 2009).

Unlike sequencing-based methods, smFISH can determine both the abundance and spatial distribution of different mRNA isoforms within individual cells (Fig. 2A). In neuronal cells, the short isoforms of myo-inositol monophosphatase-1 (lmpa1), Importin β1 and brain-derived neurotrophic factor (BDNF) mRNA are enriched in soma while their long isoforms are enriched at the distal axon (lmpa1 and Importin β1 mRNA) or dendrites (BDNF mRNA). A cis-regulatory element, which is present in the long 3′ UTR and absent in the short 3′ UTR, directs the differential mRNA localization (An et al. 2008; Andreassi et al. 2010; Perry et al. 2012). A sequencing-based approach revealed that 593 genes have differentially localized 3′ UTR isoforms, which resulted from either APA or alternative splicing (Ciolli Mattioli et al. 2019). As a result, the translational products of these mRNA isoforms have different roles in neuronal responses. For example, the BDNF proteins that are produced by the long mRNA isoforms at dendrites regulate spine morphology and synaptic plasticity in hippocampal neurons (An et al. 2008).

Recent studies uncovered a novel function of long 3′ UTRs, which act as a scaffold and recruit proteins that regulate the mRNA translational product (Berkovits and Mayr 2015; Lee and Mayr 2019). For example, the long 3′ UTR of CD47 mRNA recruits a protein complex that localize CD47 protein to the cell surface. In contrast, the short 3′ UTR of CD47 mRNA cannot recruit the protein complex. And its translational products are located at the endoplasmic reticulum (ER). Thus, while the long and short CD47 mRNAs have the same cellular distribution, their protein products are differentially distributed (Berkovits and Mayr 2015).

COTRANSCRIPTIONAL LOADING OF FACTORS

Cotranscriptional processes include 5′ capping, splicing, and 3′ end processing. RNA-binding proteins (RBPs) bind cotranscriptionally to regulate these processes, forming an RNA–protein (mRNP) complex that travels to nuclear pores for export.

In salivary gland cells from Drosophila melanogaster and Chironomus tentans, polytene chromosomes provide a highly repetitive array that is suitable for visualizing transcription with electron microscopy (Miller and Beatty 1969). By labeling a protein of interest with immunogold, overlapping signals from nascent mRNAs and gold particles represent cotranscriptionally assembled mRNPs (summarized in Mehlin and Daneholt 1993; Daneholt 1997, 2001; Björk and Wieslander 2015). Using this approach, it was noted that as RNA polymerase progresses, it loops with small particles, which are now understood to be the splicing machinery (Beyer and Osheim 1988). After this initial loading of RBPs during transcription, RBPs are classified into two categories: (1) RBPs that accompany the mRNA into cytoplasm, such as the exon junction complex (Visa et al. 1996a, 1996b; Percipalle et al. 2001; Zhao et al. 2002); and (2) RBPs that are retained within the nucleus (Alzhanova-Ericsson et al. 1996; Sun et al. 1998). Further studies are needed to understand why some RBPs, but not the rest, can accompany mRNA through nuclear pores (future directions).

EXPORT COMPETENCE AND RNP IDENTITY—THE RNP's NEW CLOTHES

After the completion of RNA processing, RNAs are released from the transcription site and exported through the nuclear pore complex (NPC) to reach the cytoplasm. Movement of mRNA through the nucleoplasm occurs through interchromatin spaces and by an energy-independent, diffusion-based mechanism (Politz et al. 1999; Shav-Tal et al. 2004; Vargas et al. 2005; Siebrasse et al. 2008; Mor et al. 2010).

Export times range between 2.5 min and 1 h depending on the model system used (Shav-Tal et al. 2004). The mRNA searches until it finds a pore suitably primed for export (Siebrasse et al. 2012) with only one in three or four attempts leading to successful export (Grünwald and Singer 2010; Ma and Yang 2010; Siebrasse et al. 2012). Once found, passage of RNA through NPC is extremely fast and occurs in three discrete steps, docking (80 msec), translocation through the NPC (5–20 msec), and then release (80 msec) (Grünwald and Singer 2010; Grünwald et al. 2011; Ma et al. 2013). Longer dwell times have been noted at the nuclear basket, possibly correlating with RNA remodeling (Grünwald and Singer 2010; Montpetit et al. 2011; Siebrasse et al. 2012).

Given the inefficient process of export, further studies will determine how gating or priming of the mRNP is accomplished. While questions still exist for why certain pores can engage the mRNA, the random diffusion of RNA from the transcription site facilitates interactions with multiple pores and may prevent clogging of the export machinery (Mor et al. 2010). RNAs may transiently interact with pores that are occupied and therefore may be unsuitable for export (Grünwald and Singer 2010; Grünwald et al. 2011). Delays in RNA movement at the nuclear and cytoplasmic faces are due to exchange of proteins as well as mRNP remodeling (Siebrasse et al. 2012). In yeast, Dbp5 helicase is involved in this exchange through its interactions with RNA (Montpetit et al. 2011). By remodeling the RNP as it exits the NPC, nuclear proteins are shed, and cytoplasmic proteins are added with exit from the nuclear pore (Ledoux and Guthrie 2011). Strikingly, deletion of Dbp5 in yeast prevents RNP export, causing the mRNP to reverse directions at the cytoplasmic face of pore, presumably though inability to shed nuclear identity (Hodge et al. 2011).

COUPLING OF RNA PROCESSING

Each step of RNA processing must be coordinated with the binding and unbinding of hundreds of factors. This process occurs faithfully throughout development, yet how the different steps are coupled to one another is an area of active investigation. For example, gene promoters orchestrate cotranscriptional loading of RBPs that accompany the mRNA to the cytoplasm and regulate mRNA's localization, translation, or stability (Bregman et al. 2011; Trcek et al. 2011; Vera et al. 2014; Zid and O'Shea 2014). The mechanisms underlying promoter-based recruitment are under investigation but may involve either RNA sequence elements or RNA Pol II subunits (for review, see Haimovich et al. 2013).

RNA polymerase plays a central role in organizing the different steps of processing. Phosphorylation of the CTD serves as a recruitment platform (for review, see Bentley 2014; Saldi et al. 2016). Additionally, the elongation speed of RNA Pol II also functions as a regulatory step for multiple aspects of gene expression. In the case of RNA splice site selection both recruitment and kinetics are at play (de la Mata et al. 2003; Dujardin et al. 2014). Recruitment coupling by RNA Pol II brings processing factors to the transcription site. Kinetic coupling generates a window of splicing opportunity during the process of polymerase elongation, depending on the time windows, different splice sites become available.

RNA Pol II also regulates termination and APA through similar kinetic (torpedo model) or recruitment (allosteric model). In the torpedo model, polymerase continues to transcribe after cleavage and is chased down by the 5′-3′ exonuclease XRN2. The allosteric model suggests that as protein factors are lost from the CTD, polymerase processivity decreases until it releases the elongating strand.

In addition to RNA Pol II, hundreds of RNA-binding proteins contribute to the layers of regulation that determine alternative splicing outcomes of a single gene. RBPs can bind to short sequence motifs within the pre-mRNA to influence spliceosome assembly or disassembly. RNA-binding proteins also can dictate alternative splicing changes, not only by competing with other RBPs for binding sequences on the pre-mRNA but also by causing structural changes to the RNA. Looping the mRNA or promoting new double-stranded regions can mask or unmask motifs bound by splicing factors. The kinetics of RBP association appear to be highly dynamic (Jobbins et al. 2018) and further studies will elucidate how the RBP code functions within cells (discussed in future directions).

In addition to regulation by RNA-binding proteins or modification of the core transcriptional machinery, regulatory control may be linked to the biophysical properties of RNA. Emerging work suggests that phase-separated compartments within the nucleus may promote the above processes and be coupled to the process of transcription (Hnisz et al. 2017; Henninger et al. 2020).

THE FUTURE OF IMAGING THE PROCESSES OF TRANSCRIPTION AND RNA PROCESSING

Great strides have been made in developing imaging approaches to understand the mechanisms of transcription and immediate downstream events regulating the processing of the transcript: splicing, 3′ end formation, termination. However, these approaches so far have relied on detection mainly of the nascent chains during elongation and termination (Brody et al. 2011; Larson et al. 2011; Coulon et al. 2014). Much more needs to be learned about the temporal aspects of each of the steps of initiation, elongation, and termination by observing single genes in action, and what happens to the transcripts during and subsequently. This will require developing microscopes and reagents designed for better detection over longer periods of time in living cells. The rapidly moving discoveries of proteins involved in various stages of these processes makes the need for imaging compelling since it will be the only way to characterize their molecular effects with sufficient temporal resolution to postulate the effects they have on the process.

What is currently needed are two developments that will advance our knowledge substantially. The first is to identify and characterize specific proteins that act during these events and then observe them in action. The second is to develop a high-resolution means to detect specific genes actively expressing in isolation from others and without perturbing their activity. The first of these developments is nearer at hand, since specific dyes have been developed that allow the detection of single proteins in the nucleus, and their binding as revealed by their dwell time at a specific site (Chen et al. 2014; Grimm et al. 2015, 2020). Doping in a small number of these molecules into the cell allows the process to be followed and tracked without the complexity of a myriad of tagged molecules confounding the analysis. However, while the state of the molecules can be inferred, and their exact positions determined by their Gaussian point spread function, their binding sites cannot because of uncertainties in the presumed target site. Uncertainties can arise from the microscope's point spread function, localization errors, or the size of the tag used (Fig. 4). Even a super-resolved spot in the nucleus (10–20 nm in diameter) will contain dozens of genes, and those genes will be physically mobile making it impossible to follow them for more than a few milliseconds. By tagging an individual gene using a CRISPR approach (e.g., with MS2 stem loops), it will be possible to obtain some temporal information, for instance how long a specific splicing factor can stay on a transcript, using high-resolution colocalization and dwell times, how long and frequent pauses occur during transcription, or how long 3′ end formation takes, or the duration of splicing events. All of this, while important, is only observing outcomes, not cause and effect relationships.

Figure 4.

Dynamic imaging of multiple factors loading onto a single RNA, in real time. (A) High-speed multicolor imaging is necessary to capture how proteins are dynamics loaded and unloaded on a DNA or RNA molecule. Labeled molecules may include the DNA (blue circles), histone marks (red circles), RNA (green circles), and processing factors (pink and yellow circles). Simultaneous imaging of the DNA locus (blue circles) and nascent RNA (green circles) can be used to find processing factors (pink and yellow circles) that colocalize to a given spot. (Left to right) With each microscope exposure, the movement of RNA and processing factors occurs. (B) (Left) The point spread function (PSF) of the microscope limits the spatial resolution. Conventional wide-field epifluorescence microscopes have a larger PSF (outer circle) than their super-resolved counterparts (inner circle). (Middle) When imaging multiple fluorescent molecules (red and green circles) that are within a diffraction-limited spot, their resolution may be limited by the resolution of the microscope. (Right) Gaussian fitting can be used to localize the positions of molecules within a diffraction-limited spot. Accurate localization will require using ultrabright and bleach-resistant fluorophores.

Imaging in high spatiotemporal resolution with two or more colors—one for the gene and one for a putative regulatory protein—would be the most accessible approach (Fig. 4). By super-registration of the tagged protein with the gene, one can correlate temporal relationships between the arrival (or departure) of the protein on the gene and its effect on an immediate event such as transcription or termination (Donovan et al. 2019; Li et al. 2019). We would expect a dwell time to accompany the association of the protein with the gene, or the RNA nascent chain. Proteins not involved in binding to the gene, or its transcripts would diffuse through the space and hence not be detectable at any location.

Additionally, this approach could be applicable to chromatin changes that precede gene expression: histone modifiers that influence the gene availability to factors. What are needed are the reagents that would allow the histone methylations or acetylations to be detected in real time. The development of nanobodies (Fridy et al. 2014), which can be expressed and labeled intracellularly, against a panel of specific histone modifications would go a long way to facilitate imaging chromatin dynamics. These key reagents would allow the addition or removal of select histone modifications to be correlated with the dwell time of the corresponding fluorescently labeled chromatin modification enzymes.

One of the downstream indicators of changes in chromatin is the decondensation of genetic loci during gene activation (Tumbar et al. 1999; Tsukamoto et al. 2000). Advances in CRISPR-based imaging technologies, such as MS2/PP7 stem loop engineering into guide RNAs (Ma et al. 2016; Wang et al. 2016b) or SunTag labeling of Cas9 (Tanenbaum et al. 2014), which increase the signal-to-noise ratio, will be necessary to detect decondensation in a way that does not perturb gene activation. Likewise, enzymes responsible for decondensation can be tagged and identified to be coincident with the activation of a locus as judged by its dwell time, but it will be difficult to distinguish the many of these events happening rapidly all over the genome. To achieve this goal, we will need to develop imaging methods to simultaneously and dynamically track tens to hundreds of loci at sufficient temporal and spatial resolution in a live cell (Ma et al. 2016; Wang et al. 2016b).

Underlying all of these challenges is where to place a particular label on the genome. Recent advances in understanding the 3D folding of the genome via both imaging and genomic approaches help to suggest key regions to tag (Wang et al. 2016a; Bintu et al. 2018; Kempfer and Pombo 2020). The broad diversity of chromatin structures inside the cell as seen by chromatin electron multitilt microscope tomography (ChromEMT) suggests that choosing select regions to tag based on euchromatic versus heterochromatic distinctions will be difficult (Ou et al. 2017). ChromEMT uses illumination of DNA-binding fluorescent dyes to deposit a contrast agent, OsO₄, onto DNA, which is visualized by electron microscopy tomography (EMT). Therefore, ChromEMT is limited to visualizing an outline of nucleosomes as DNA wraps around the histones without information on the underlying DNA sequence.

Ideally, one would like to determine the structure of chromatin around specific genes that are transcriptionally active or repressed. One approach to achieve this goal would be to combine CRISPR-based imaging with correlated fluorescence microscopy and ChromEMT. In this manner, CRISPR-based fluorescent maps could serve as “signposts” to tell researchers where to begin examining specific regions of the nucleus using ChromEMT. Advances in ChromEMT could also eventually allow enhanced contrast of select proteins (e.g., enhancer-binding factors, PIC components, RNA Pol II, and cleavage and polyadenylation factors, etc.) and nascent transcripts for use as additional “signposts.” In this manner, one can begin to determine the chromatin structures and accessible genomic regions immediately surrounding key regions of interest (e.g., enhancers, promoters, and termination sites) at high enough spatial resolution, which would not be possible with the genomically tagged CRISPR-based fluorescent maps alone. Using this strategy, researchers will then have a chance to calibrate the position of fluorescently tagged genomic regions and chromatin structures in live cells in a similar manner as has been achieved using DNA-FISH and genomic approaches (Wang et al. 2016a; Bintu et al. 2018; Kempfer and Pombo 2020).

The future of imaging the dynamics of transcription and RNA processing looks bright. Researchers now must coordinately take advantage of multiple expertise from different groups throughout the world to seize upon this opportunity.