Single-Cell and Single-Molecule Analysis of Gene Expression Regulation

Snapshot Approaches to Single-Cell Analysis

The rationale for imaging is to obtain information on the spatial position of cellular components inside their native environment. Immunostaining and in situ hybridization have long been used for studying gene expression in the context of tissue structures. Single-mRNA molecules are detected in fixed cells by smFISH (50, 121) (Figure 1a). By hybridizing each mRNA with multiple probes labeled with fluorescent dyes, it is possible to detect single-mRNA molecules and count their number (Figure 2a). Variations of smFISH have been developed using 50-mer (50) or 20-mer (121) oligodeoxynucleotides (ODNs), branched DNA (bDNA) probes (6), and sequential tethered and intertwined ODN complexes (FISH-STICs) (131). In addition, specific protocols have been optimized for rapid diagnostics (128) and for detecting transcribed gene fusions (125), single-nucleotide variants (SNVs) (84), noncoding RNAs (21), and nascent mRNAs (85). Analysis of the image obtained by smFISH requires software, such as Localize and Air-Localize (89, 142, 159) or FISH-QUANT (104), to automatically count the number of transcripts in the nucleus and the cytoplasm and to quantify nascent mRNAs at active transcription sites (TSs) in three dimensions. The relationship of the mRNA counts with their cellular distribution provides valuable information on mRNA retention in the nucleus (2, 7), mRNA half-life (144), and transcriptional output (69, 159).

Although smFISH provides extensive information on the metabolism of mRNA, it can assess only a few genes at a time. To overcome this limitation, a combinatorial probe labeling approach has been introduced to visualize 11 mRNAs simultaneously for multiplexed gene expression profiling (86). This approach has been combined with super-resolution microscopy (95), sequential probe hybridization and stripping (96), and a Hamming algorithm (30) to further increase the number of genes, potentially up to a thousand RNA species, for simultaneous detection. Moreover, fluorescence in situ RNA sequencing (FISSEQ) allows transcriptome sequencing by oligonucleotide ligation and detection (SOLiD) of reverse-transcribed cDNA inside cells (82). Although the current form of FISSEQ produces many fewer reads than RNA sequencing and is less accessible to genes involved in RNA and protein processing, it has potential for measuring transcriptomes in single cells within their spatial context.

Overall, smFISH provides accurate information on the number and localization of single-mRNA molecules. Quantification of large numbers of cells informs cell-to-cell variability, and the application of algorithms, such as FISH-QUANT, infers gene expression dynamics. Nonetheless, it lacks the capacity to follow the same particle over time, necessitating the development of live imaging of single molecules.

Real-Time Approaches in Live Cells

A high temporal resolution is required to understand single-molecule dynamics and cell fates in heterogeneous cell populations. To achieve single-mRNA resolution inside living cells, a genetically encoded stem-loop array has been successfully used. The MS2 system exploits a highly specific binding of MS2 capsid protein (MCP) and MS2 binding site (MBS) RNA stem-loop (SL) from MS2 bacteriophage (11). A gene tagged with multiple copies of MS2-SL produces a reporter mRNA that binds MCP-GFP (green fluorescent protein) fusion proteins expressed in the same cell. As a result, each reporter mRNA is labeled with many GFPs and becomes bright enough for single-mRNA imaging in live cells. Tagging an endogenous mRNA with MS2-SL is relatively easy in yeast using homologous recombination (68). Studies in mammalian cells have been done using transient transfection or cells from a β-actin mRNA MBS knock-in mouse (89) and a transgenic mouse expressing MCP-GFP (115). This technique has been extensively used to monitor real-time dynamics of transcription, nuclear export, and trafficking of mRNAs in live cells (reviewed in 20, 114) (Figure 1a). Also, recent advances in CRISPR-Cas technology have greatly facilitated genome editing (129) and RNA labeling (108) for live cell imaging.

The evaluation of each aspect of mRNA metabolism has required the development of specific microscopy or analytical tools. For example, one of the great advantages of the MS2-MCP system is that it allows single-molecule tracking of RNA in living cells. The goal of single-particle tracking (SPT) is to assign a specific trajectory to individual mRNAs (Figure 1b and Table 1). An important parameter for SPT is the spacing-displacement ratio, which refers to the average distance between particles divided by the average displacement of a particle between two successive image frames. If this ratio is larger than one, particle tracking can be performed by searching for the nearest neighbor in successive image frames. However, with a lower spacing-displacement ratio, it becomes more difficult to connect the trajectories (114). Thus, various tracking algorithms have been developed to find the most probable set of particle trajectories (31). mRNA molecules exhibit complex dynamics switching between diffusive motion and directional transport (52), which can be analyzed by using a software for inferring transient transport states (75, 101). Simultaneous tracking of single mRNAs and ribosomes by multispectral live cell imaging has revealed that mRNAs loaded with polyribosomes move slower than nontranslating mRNAs (75). This observation was confirmed by simultaneously tracking single mRNA molecules and nascent polypeptides as they are synthesized (155). The translational state of an mRNA is indistinguishable based on its dynamics but can be identified by the visualization of the nascent peptides (103, 150, 155, 157). Another example, discussed below, is the development of a super-registration method that resolves the kinetics of β-actin mRNA export through the nuclear pore complex (NPC) (60). By combining this method with newly developed multifocus microscopy, one can capture 3D single-molecule real-time images and perform an analysis that determines the position as well as the diffusion rate of β-actin mRNAs in the nucleus (134).

The resources to see real-time biological processes, such as transcription and translation, have increased with the design of the orthologous stem-loop systems such as PP7 (27) (Figure 1a,c) and other systems like λ boxB RNA (78) or a U1A tag (16, 138) (specific for yeast). Multicolor RNA imaging has been applied to measure transcription elongation by tagging the same mRNA with PP7-SL in the 5′ end and MS2-SL in the 3′ end and quantifying, over time, the fluorescence signal from nascent transcripts binding PP7 coat protein fused to GFP (PCP-GFP) and MCP-mCherry (68). Alternatively, the first round of translation has been localized to single-molecule resolution within a live cell by tagging a single mRNA in the open reading frame (ORF; where the PCP is displaced by translation) and the 3′ untranslated region (3′ UTR), and monitoring where the change from two colors to one color occurs in the cell (63).

Although mRNA is a central molecule to analyzing the regulation of gene expression, gene function is usually performed by proteins. Imaging of single proteins has been limited by the number of photons emitted by one fluorophore. To increase the signal, several solutions have been developed. A repeating peptide array named SunTag has been engineered to recruit multiple copies of an antibody fused to a fluorophore, such as GFP (139) (Figure 1a). It is possible to achieve single-molecule resolution for any protein tagged with the SunTag system and track the proteins in live cells (Figure 1c). In addition, N-terminal tagging with the SunTag system visualizes the synthesis of a nascent peptide from single mRNAs labeled with the MS2 or PP7 systems (103, 150, 155, 157) (Figure 1b,c). An alternative approach has been to use brighter dyes than the conventionally used fluorescent proteins. Especially remarkable for its versatility is the HaloTag technology, a modular tag system based on a modified haloalkane dehalogenase (Halo) from bacteria designed to covalently bind to synthetic ligands (94). This ligand can be a reactive linker attached to a fluorescent dye. The HaloTag fluorescence ligands are at least an order of magnitude brighter than conventional fluorescent proteins like GFP (58). This technology has been extremely useful to perform SPT of transcription factors (TFs) (discussed below). Additionally, advances in two-photon fluorescence correlation spectroscopy (FCS) have made possible the analysis of single-mRNA and single-protein association in the cell (154). This approach is based on the statistical analysis of fluctuations in fluorescence signals from two molecules of different colors (RNA and protein) when passing through the excited spot. In this application, the total amount of proteins bound to the RNA can be calculated, on the basis of the amount of correlated brightness, by fluorescence fluctuation spectroscopy (FFS) (Figure 1b) (154).

Combinations of snapshot and real-time approaches have been used to elucidate different steps of gene expression. To survey the latest advances, we describe recent discoveries on the transcriptional and post-transcriptional processes that regulate gene expression, and the coupling between transcription and mRNA fate.

Promoter Sequence

A promoter contains specific sequences recognized by TFs to modulate transcription. It is now well established that transcription is inherently stochastic and occurs predominantly as an episodic process or burst (90). Nonetheless, constitutive modes have been described (159) in which transcription is not limited to a short period of time. Characterization of the transcriptional activity by smFISH in yeast has indicated that several transcriptional profiles are encoded by the promoter sequence without being influenced by the genomic locus, gene length, or gene sequence (69). Despite the strong influence of the promoter sequence, transcription from alleles of the same gene is not coordinated in time and strength (68, 89, 141, 147). For example, in mouse embryonic stem cells (ESCs) transcription of Nanog, a key regulator of pluripotency, occurs from both alleles in 14% of the cells and 45% of them have only one active allele (141). These observations underscore the stochasticity of transcription and the influence of other regulatory factors.

Transcription Factor Dynamics

TFs transmit information on the cell state and bind to specific DNA sequences on gene promoters to control the rate of transcription by RNA polymerase II (Pol II). Cell-to-cell variation in TF concentration leads to differential cell-fate decisions even within isogenic populations (4, 126, 141). This is especially relevant for ESCs, where heterogeneity provides an advantage to maintain the pluripotency network (141). In addition to concentration, TFs such as ERK, p53, or NF-κB can transfer information through changes in localization rather than global concentration (25, 36, 109, 118). In this case, tracking of fluorescence proteins in live cells provides a more powerful and robust approach for assessment of TF dynamics than does analysis of fixed cells. For example, p53 fused to a fluorescence protein shows a series of discrete pulses of proteins into the nucleus. The number of pulses varies among cells and the mean of pulses increases with DNA damage, but the individual pulses are of fixed duration and amplitude (77). At the c-Fos gene, the TF concentration determines the burst frequency, the duration of TF binding on chromatin affects the burst duration, and the strength of the transactivation domain of TF affects RNA Pol II initiation frequency (126).

Recently, microfluidics coupled to quantitative time-lapse fluorescence microscopy to control TF dynamics and measure the dynamic gene expression response of individual genes have been applied to the analysis of single yeast cells (64, 65, 88, 100). This analysis has revealed that the TF Msn2 determines multiple and distinct gene expression profiles depending on the amplitude threshold and timescale of activation, which is influenced by the nucleosome occupancy (64). Interestingly, the gene expression program interpreted by a specific promoter can be modulated by modifying promoter cis elements (65). Moreover, usually several TFs operate simultaneously, and the promoter should integrate different sources of information. In the case of Msn2, it differentially modulates the expression of target genes, depending on the temporal overlap or nonoverlap with the pulse of activity of the transcriptional repressor Mig1 (88). These examples indicate that TF dynamics transmit the information necessary to regulate transcription and therefore require analysis with the proper spatial and temporal resolution.

Transcription Factor Target Search

Ultimately, each TF molecule has to find its specific locus and bind to its target site on genomic DNA to trigger gene expression. Fluorescence recovery after photobleaching (FRAP) and FCS complement each other and provide further detailed information on TF dynamics in the nucleus (23, 117; reviewed in 37, 92). These techniques can extract kinetic parameters on the movement and distribution of proteins and the rate of exchange between different cellular compartments or immobilization of proteins in large structures. They have, for example, established that TF binding to chromatin is generally transient (61) (Figure 1c).

Although FCS and FRAP experiments measure the average movement of many molecules, super-resolution fluorescence microscopy allowed the development of SPT techniques for the quantification of TF dynamics with molecular resolution (29, 46, 73, 93). SPT has started to reveal target-search strategies, differentiate specific from nonspecific residence times, and define a hierarchical behavior of TFs in the whole nucleus. In ESCs, Sox2 and Oct4 are essential TFs to maintain pluripotency. Sox2 binds first on the DNA, followed by Oct4. Sox2 locates its cognate binding sites using 3D diffusion (duration ~3 s) followed by 1D sliding along DNA (duration ~0.8 s). Once bound to a cognate recognition site, the average residency time of Sox2 alone is ~12 seconds before it arrests Oct4, which stabilizes the enhanceosome assembly (29). Clusters of Sox2 bound in the nucleus have been localized by a lattice light-sheet–based single-molecule strategy (93). These clusters represent clusters of enhancers and are located in nonheterochromatin regions and overlap with subsets of Pol II.

Interestingly, chromatin structure influences the dynamics of Sox2 (29, 93), and this TF behaves differently in heterochromatic regions and enhancer clusters. Its search mode as well as its distribution change as histone acetylation is globally increased. In particular, histone H3K27 acetylation has been shown to accelerate the search kinetics of TFs and accelerates the transition of Pol II from initiation to elongation (135).

The impact of nuclear architecture on the diffusion of TFs has also been evaluated by tracking two TFs of comparable sizes: c-Myc and positive transcription elongation factor (p-TEFb) (73). These TFs explore the nucleus in different manners. c-Myc moves almost everywhere inside the nucleus (a noncompact exploration), whereas P-TEFb movement is constrained to a specific path through the nucleus (a compact exploration). This geometry-controlled kinetics of exploration could have a strong impact on how TF activity is controlled in space and time in the nucleus (153).

Chromatin Structure and Dynamics

Gene expression requires that gene regulatory elements and coding regions be accessible to TFs and to the transcriptional machinery. The nucleus is a crowded environment with a hierarchy of organized structures in which chromosomes fold into complex 3D conformations (55). Chromosome conformation capture (3C) and 3C-related technologies (4C, 5C, and Hi-C) have allowed the mapping of average intra- and interchromosomal interactions within populations of cells. These technologies, combined with computational modeling, have provided a picture of the topological association of individual domains as well as a genome-wide perspective on functional and structural genomic subdomains in the eukaryotic nucleus (45, 71). These technologies are applied mainly to cell populations and require cross-linking of macromolecular components; thus, they cannot be used to determine the physical distances between interacting regions (8, 107). DNA-FISH is the main alternative for resolving spatial relationships in single cells. Although DNA-FISH can assess only a few loci at the time, it established that chromosomes occupy discrete territories in the cell nucleus (39). In mammalian cells, gene-rich chromosomes generally cluster at the nuclear center, and gene-poor chromosomes frequently localize near the nuclear periphery (152). Moreover, gene-rich regions decorate the outside of their own chromosome territories (15) and certain genes loop outside of their own chromosome territory (152). A recently developed high-throughput 3D mapping platform has unexpectedly shown that DNA replication, rather than mitosis, establishes gene position relative to the nuclear periphery (127). A combination of 5C data, physical modeling, 3D single-molecule DNA, and smFISH for the X chromosome has shown that chromatin structures underlying topological associated domains are highly variable and that these structural fluctuations can contribute to transcription (56).

A limitation of all these methods is that they are applied to fixed cells where dynamic aspects of chromatin organization can only be inferred from snapshots. Tagging of single loci with Lac operon (LacO) arrays with Lac Repressor (LacI) fused to photoactivatable fluorescent protein-tagged histones (102) has demonstrated that chromatin is locally mobile but rarely moves over long distances (105). Nonetheless, the heat shock protein 70 (HSP70) loci move unidirectionally from the nuclear membrane toward nuclear speckles upon induction of transcription by heat shock in mammalian cells (76). The effect of gene positioning on transcriptional levels is controversial and appears to be organism dependent. Relocation of a specific human chromosome to the nuclear periphery reduces the expression levels of some endogenous human genes located near the tethering sites (51). In contrast, active inducible genes relocate to the NPC in yeast (10, 135). Because most of these studies have been performed using the LacO and LacI system, it is possible that some of the results are biased by such integrations in the genome (43, 74). Recent development of the CRISPR-Cas9 system has produced new tools to follow endogenous sequences. The use of a fluorescently labeled nuclease-deficient Cas9 (dCas9) protein for live imaging of the MUC4 loci supports the notions that genes assume nonrandom positions and exhibit confined movements at short timescales (28).

Many studies have shown that transcription in the mammalian nucleus occurs at discrete locations. These locations have been called “transcription factories” (137), but the idea of stable transcription factories has been recently challenged. To determine the dynamic nature of these factories, a quantitative single-cell and single-molecule approach has resolved the spatiotemporal organization of Pol II in the nucleus (34). Pol II was fused to the photoswitchable fluorescent protein Dendra2 and examined at single-molecule resolution with PALM. Pol II clusters were found to dynamically assemble and disassemble, with an average lifetime of 5.1 s, supporting a model of self-organizing clustering. Furthermore, these dynamically assembled clusters could explain the long-range chromosomal contacts observed for some co-regulated genes (48). These transient Pol II clusters have been linked to the number of mRNAs synthesized during serum stimulation, suggesting they constitute a pre-transcriptional regulatory event that controls nascent mRNA output (32). The origin of these Pol II clusters remains unknown but could be linked in the future to TF clustering or factors involved in the Pol II preinitiation complex (PIC) assembly.

Transcription Is Predominantly Bursting

It is now well established that transcription predominately occurs in bursts in eukaryotic cells (22, 33, 120, 126, 159), viruses (130), and bacteria (22, 140). Multiple quantitative models have been developed to provide a description of stochastic bursting (79, 120, 126, 159). In the main model, called the telegraphic model, transcription of a gene can switch randomly between an inactive OFF state and an active ON state (Figure 1c). While the allele is in the ON state, it can initiate transcription at a specific rate. For example, during serum induction, the TF concentration modulates the burst frequency of c-Fos (i.e., the transition between an OFF state and an ON state), but the other parameters, including the rate of initiation by individual Pol II complexes, remain unchanged (126). Steroids also drive the burst frequency of transcription to produce the well-known analog dose response across the population (79). The existence of a transcriptional memory (a period after a burst during which cells cannot switch from the OFF state to the ON state) that lasts up to several hours has been described in mammalian cells (66, 130). Conversely, the social amoeba Dictyostelium is more likely to re-express an endogenous developmental gene than initiate new expression (the probability to switch from OFF state to ON state is higher) (33). Large-scale studies suggest that the transcription of a reporter transgene encoding a fluorescent protein occurs as a burst regardless of its location in the human genome. The frequency and size of the transcriptional burst are equally modulated across the genome (41, 132, 148).

Different steps of transcription can potentially generate bursts. For example, chromatin may switch between a permissive and a repressive state, a TF may bind to a promoter or enhancer, or a loop may be formed so that the PIC can reinitiate transcription (26, 38, 136). Models of transcription regulation have been built after the simultaneous observation of pairs of these regulatory events in fixed and live cells, and the application of correlation techniques to define kinetics and provide models of transcription regulation (38, 56, 76, 80, 111, 126, 135, 156). A better picture of the components orchestrating the spatiotemporal regulation of gene transcription would come with the simultaneous imaging of chromatin, TFs, and mRNA. A step toward observing biological processes in real time has been the visualization of transcript formation by combining the MS2-MCP and PP7-PCP tags in the same transcript (68). The advantage of this technology is that it enables an mRNA to be followed from the genomic site of its transcription to the cytoplasm and can provide valuable understanding of its dynamic response to factors that influence its biogenesis and translation.

mRNA Maturation

Some insights on how and where mRNA splicing occurs have emerged from imaging the process of splicing, achieved by tagging the intronic sequences on the pre-mRNA (Figure 1c). Splicing is cotranscriptional, and it takes place, at least partially, at the TS. Kinetic studies have shown that splicing is completed 5 to 10 minutes after transcription, regardless of the length or number of introns (17, 85, 146). The elongation rate of Pol IIhas been described as uncoupled from ongoing splicing, and multiple spliceosome complexes can assemble simultaneously on active TSs (17). The speed of Pol II elongation and the presence of pause sites can result in alternative splicing (24, 44). However, details of splicing differ from gene to gene. For example, dual-color mRNA imaging in living human cells has shown that β-globin mRNA splicing can occur stochastically both before and after transcript release (38).

An mRNA undergoes maturation by processes in addition to splicing. N6-Methyladenosine and the addition of the polyA tail regulate the stability and translation of the mRNA (91, 151). Additionally, mRNA-binding proteins and nuclear assembly factors are cotranscriptionally recruited to form the messenger ribonucleoprotein (mRNP), which needs to be released from the TS after polyadenylation (119). Some mRNA maturation processes take place at the TS and occur during time the transcribed transcripts remain at the TS (approximately 2 minutes) (38), as shown by using probes or inserting the MS2-MCP in the 3′ end of the gene (17, 126). Nonetheless, there are exceptions. For example, to regulate translation, the CPEB family of proteins modifies the length of the polyA tail of specific mRNAs once in the cytoplasm (151).

Nuclear Diffusion and Export

Translation requires that the mRNPs exit the nucleus. Once released from the TS, it takes the mRNA molecule ~20 minutes to localize in the cytoplasm of mammalian cells (9, 60). The movement of the mRNA in the nucleus has been analyzed by following the trajectories of single mRNAs tagged with the MS2-MCP system. Using a high-speed multifocus microscope for simultaneous imaging of mRNA in 3D and an internal registration marker for the NPC showed that >60% of total β-actin mRNAs localize within 0.5 μm of an NPC. β-actin mRNAs seem to diffuse freely in the nucleus, although the distribution is not uniform (134).

Entry into the cytoplasm requires that the mRNA successfully finds, interacts, and moves through the NPC. A combination of super-registration and high-speed microscopy has resolved the kinetics of mRNA export (60). A three-step model with docking (80 ms), translocation through the NPC (5–20 ms), and release (80 ms) has clarified the steps involved in mRNA export in mammalian cells (60) (Figure 1c). These experiments suggested that not all pores are equally active and the transcript scans along pores until it finds one suitable for mRNA export. Interestingly, the pores transporting β-actin are repeatedly active over time (134). These facts suggest that the selection of a specific NPC could be determined by the interaction of proteins of the mRNP and the nuclear basket of the NPC. In yeast, deletion of the gene encoding the nuclear basket protein Mlp1/2 or the mRNA binding protein Nab2 alters the scanning behavior and shortens the residency times inside the NPCs (123). Additionally, cells achieve directionality of export by selectively removing the essential export factor Mex67 from the mRNP, thereby preventing further interaction of the now cytoplasmic mRNP with the NPC (123, 134). Once in the cytoplasm, the mRNPs are confined near the nuclear envelope. This suggests that mRNPs, once exported, remain with multiple NPCs at their cytoplasmic face; most likely, the mRNP composition is remodeled for future events in the cytoplasm (133). Visualization of the interaction between specific RNA-binding proteins (RBPs) and single-mRNA molecules is essential to understanding the metabolism of the mRNA in the cytoplasm.

Cytoplasmic Mobility and Localization

The composition of mRNPs is dynamic. Some factors associate cotranscriptionally and remain throughout the life of the mRNA and others interact briefly with the mRNA molecule. The stoichiometry of each of these interactions influences the movement, localization, and fate of the mRNA (47). Usually, an mRNA has one or two binding sites for each factor, which makes it especially challenging to localize, follow, and observe these interactions. However, molecular interactions between mRNA and protein have been recently resolved and localized by fluorescence fluctuation spectroscopy (FFC) and heterospecies partition analysis (154) (Table 1) (Figure 1b,c). This technology has been used in cells expressing β-actin mRNA tagged with the MS2-MCP system and the well-characterized binding protein ZBP1, tagged with mCherry. This approach uses a fluorescent spot, generated by a two-photon microscope, to detect the brightness and diffusion of molecules passing through a femtoliter volume (Figure 1b,c). Quantitative measurements of the amount of mRNA, amount of ZBP1, and the different locations of their association proved that β-actin mRNA was preferentially bound to ZBP1 at the nuclear periphery, although this interaction did not form at the TS (154). This association supports the well-known function of ZBP1 in mediating localization and translational repression of this mRNA until activation by the proper stimulus (18, 47). The method also correlates association with specific cellular behavior or morphology.

Localization of mRNAs in different cellular compartments is a widespread mechanism to regulate several cellular processes; it is mRNA and cell-type specific. For example, in the yeast Saccharomyces cerevisiae, transport of ASH1 mRNA along the actin cytoskeleton to the bud tip inhibits mating-type switching in the daughter cell (11, 18). In Drosophila melanogaster oocytes, oskar mRNA is actively transported in ribonucleoprotein granules to be anchored in the posterior end where it is translated (see below) (143). Localization and translational repression in the granules are essential for Drosophila development and patterning. Structural illumination microscopy and smFISH have indicated the homotypic distribution of the mRNAs in these granules (143). β-actin mRNA is also packed in neuronal granules to exit the soma and be actively transported in dendrites (19). Synaptic stimulation induces the temporal release of the β-actin–mRNA from the granules, probably to be locally translated and contribute to synaptic plasticity (19, 115). In addition, β-actin mRNA localization in fibroblast protrusions promotes directed cell migration, as shown by an analysis that correlates mRNA localization in a living cell with motility (116). Finally, localization of mRNAs in the cytoplasm also responds to changes in environmental conditions. For example, mRNAs are packed in processing or stress bodies under stress conditions, and after stress the mRNA can be released and return to translation or become degraded (42).

Single-Cell Analysis of Translation: Where, When, and How?

The importance of spatial localization is most apparent in the context of translation. By combining spatial and temporal regulation, the cell allows transcripts to be in the right place at the right time, ensuring that the resulting proteins are used efficiently when needed. This saves cellular resources and prevents possible protein-associated toxicities.

To determine when the pioneer round of translation occurs in mRNAs, the MS2 and PP7 systems have been used to tag the same transcript with two spectrally different fluorescent proteins. The rationale is that the fluorescence signal from the MS2-MCP, inserted in the 3′ UTR, localizes the transcript at all times and the fluorescence signal from the PP7-PCP, inserted in the ORF, is released by the elongating ribosomes melting the PP7-SL. The technique, known as TRICK (translation RNA imaging by coat protein knockoff), is based on the switch from a two- to one-color labeled transcript, therefore distinguishing the moment when mRNA converts from an untranslated to translated state (63) (Figure 1c). In mammalian cells, TRICK has confirmed that little translation occurs in the nucleus and that the pioneer round of translation is completed before the mRNA has diffused more than a few micrometers away. Interestingly, mRNAs sequestered in processing bodies upon stress remain untranslated while trapped (63). Because the method assesses the spatiotemporal control of translation, it has been used to localize the expression of oskar in the posterior pole of the Drosophila oocyte at stage 10. In this case, the change of the fluorescence labeling of the reporter has been correlated with the detection of the accumulation of the oskarprotein in the posterior pole (63). To provide further information on translation events, this technique needs to be complemented with live-cell imaging of the translational status of the mRNA after the pioneer round.

To analyze where translation takes place and how it alters the dynamics of mRNAs, one can image both ribosomes and transcripts within the same cell. FFC has been used to determine the association of ribosomes with β-actin transcripts (154). The subsequent heterospecies analysis showed differences in ribosome loading upon individual β-actin transcripts, with a higher number of ribosomes per transcript at the cell periphery. Therefore, association of β-actin transcripts with ZBP1 and ribosomes is anticorrelated, and these results corroborate prior observations that sequestration by ZBP1 leads to translational repression at the nuclear periphery (70). Translation of β-actin at the cell periphery has also been suggested by tracking the comovement of the transcript with labeled ribosomes at high spatiotemporal resolution. HMM-Bayes analysis (Table 1) of the trajectories has indicated the presence of two diffusion states (slow and fast) (75). Transition from slow to fast states correlates with the disappearance of the ribosomal signal (75) (Figure 1c). Therefore, this method is a good way to distinguish translating (polysome-associated) from nontranslating transcripts and suggests a burst mode of translation similar to that described for transcription.

This proposition was recently confirmed with a method to directly quantify the number of proteins made per burst of translation and the frequency and speed of translation (103, 150, 155, 157). A major hurdle to accomplishing this has been the observation of single-protein molecules. The diffusion of most proteins is too fast to observe with standard microscopy. Recently, two different approaches have addressed this problem through the use of multiplexed antibody conjugated fluorophores, which not only increase the signal but also decrease the diffusion rate of the target protein. Through the use of Fabs (fragments antigen-binding) targeting either a short peptide epitope or GFP itself, multiple groups have been able to increase the fluorescence intensity from a single protein. Additionally, these techniques detect the nascent peptides as they are being synthesized because they do not require additional maturation time to fluoresce, as is the case for GFP. One example was the development of Spaghetti monster (149), which used exogenous, bead-loaded Fab fragments targeting multimeric HA and FLAG epitopes on a GFP molecule. The use of single-chain variable fragment (scFv) antibodies fused to GFP makes the development of a fully genetically encoded system possible (SunTag). Both systems (Spaghetti monster and SunTag) allowed the addition of 24 × GFPs onto a single target protein, providing a bright enough signal to image translation sites as well as single protein molecules over long periods of time (103, 139, 150, 155, 157) (Figure 1a,c). By combining these approaches, mRNAs tagged with the MS2-MCP or PP7-PCP systems have been co-tracked with newly formed proteins within the same translation sites, allowing for a number of observations to be made about translation. The first conclusion is that translation undergoes bursty kinetics (periods of continuous translation followed by periods of quiescence), much like in transcription (Figure 1b,c). After exposure to oxidative stress, global translation decreases; however, some transcripts, such as the stress response transcription factor Atf4, have a sharp and transient increase in translation (150).

Through different combinations of FRAP, drug treatment, and modeling of ribosome runoff after harringtonine treatment, the four studies directly measured ribosome elongation rate (3–10 AA/s) and calculated that ribosomes occur approximately every 200–300 nucleotides on an mRNA (103, 150, 155, 157). Although codon optimization increases the rate of translation, it was also noticed that 5–10% of mRNAs had ribosomes fail to run off because of chemical damage of the mRNA (157). In hippocampal neurons, translation occurs more often in the proximal regions of the neuron and can still happen while the mRNA is moving. Interestingly, visualization of the translational dynamics of ER-targeted mRNAs has shown that tethered mRNAs can be translated only by ER-bound ribosomes, supporting the co-translocon model (155). These studies will pave the way for further insights on translation in different contexts, such as cell cycle regulation, development, and disease.

Single-Cell Analysis of Decay

Finally, at the end of its lifetime, the mRNA molecule has to be degraded. This is a fast and highly regulated process used by cells to quickly adapt their transcriptome to their metabolic requirements. Single-cell analysis has also been used to clarify longstanding debates about the location at which nonsense-mediated decay (NMD) destroys transcripts bearing nonsense codons. SmFISH identified transcripts within the same cell containing either a normal or premature termination codon (PTC) using a bidirectional promoter expressing β-globin (Figure 1c). Surprisingly, NMD degradation of the majority of mRNAs that contain a premature stop codon takes place in the cytoplasm and not in the nucleoplasm as suspected (144). Export of these mRNAs is thus not affected by the stop codon. Through the use of TRICK (see above), the pioneer round has been visualized (63); thus, it is now possible to track transcripts that escape NMD, which should give interesting insights into what happens to the small population of surviving transcripts. Results thus far suggest that the small number of transcripts that escape this process appear to have decay rates similar to their normal counterparts (62).

Cell-to-Cell Heterogeneity: Noise in Gene Expression

Single-cell analysis of gene expression has revealed significant variations among cells. This variation, defined as noise, refers to heterogeneity in a population of isogenic cells that cannot be explained by the low rate of mutation. Noise can be of extrinsic or intrinsic origin (106). Extrinsic noise derives from environmental fluctuations, arbitrary partitioning of molecules between daughter cells, or cell size. It is relatively easy to picture that cells in a population could be differentially exposed to environmental changes and therefore respond differently to a stimulus. Intrinsic noise, derived from the stochasticity of transcription and translation, could enhance or compensate the diversification of the response (40, 98). For example, increases in cell volume are accompanied by adjustments in transcriptional burst size and frequency during the cell cycle (113). At essential genes that encode products needed to operate at defined concentrations, highly bursty transcription is compensated by inefficient mRNA translation to achieve low intrinsic noise (83).

Nuclear retention of specific mRNAs represents an alternative mode of regulation that allows cells to express proper protein levels independent of transcription and translation (2). Although transcription in metazoan cells is mainly stochastic, the cytoplasmic content of mRNAs can be determined, considering both the cellular phenotypic state and context (7). These experiments uncovered a new function for the nuclear envelope: to buffer the transcriptional noise in the cytoplasm of higher eukaryote cells. It still remains to be determined whether this or other mechanisms represent an evolutionary path to control noise in gene expression programs required for proper functioning in multicellular organisms.

Single-Cell Analysis in Multicellular Organisms: Spatiotemporal Resolution of Cell-to-Cell Variability

In multicellular organisms, each cell must adapt its transcriptome and proteome to the demands of the whole organism. One example of such adaptation is the liver, which absorbs glucose after a meal and synthesizes and releases glucose during starvation (3). SmFISH has revealed that hepatocytes switch from glucose absorbers to secreters by coordinating transcription and mRNA decay of the gluconeogenic genes Pck1 and G6pc. During fasting states, gluconeogenic genes achieve higher levels of expression by a synchronized increase of the transcriptional burst and decrease of mRNA degradation rates. This synchronization also allows for a prompt degradation of transcripts to adapt to the fast increase in glucose levels after meals. Surprisingly, there is a high heterogeneity in the transcriptional output among hepatocytes. This cell-to-cell variability also applies to other genes considered constitutive, such as β-actin. In the case of β-actin, transcription is burstier and the mRNAs are more stable. Interestingly, polyploid hepatocytes are less noisy than diploid ones. Hence, a combination of cell polyploidy and regulation of the mRNA half-life seems to control transcriptional noise in the liver (3).

The heterogeneous transcription among cells must be shaped to ensure tissue functionality and development. In the case of the pituitary gland, the length and intensity of the prolactin transcriptional burst become shorter and less intense over age and the frequency is slightly unchanged. Interestingly, in the adult gland, cells that are proximal to each other have a better coordination in expression than distal ones, suggesting a role of the spatial organization and inter-cellular communication in dampening the noise (49).

The most noteworthy example of gene expression coordination occurs during embryogenesis. From fertilization, gene expression is tightly delimited in time and space to determine the identity and fate of each individual cell and to orchestrate the formation of the new body. The MS2 system provides the ability to image individual cells in the whole Drosophila embryo, unraveling spatiotemporal and mechanistic insights behind its development (57). Three hours after fertilization, the Drosophilaembryo is composed of ~6,000 cells with pattern formation information that determines the Drosophila body segments (81). The maternal mRNA bicoid (bcd) determines the anterior-posterior axis by activating the gene hunchback (hb) in a protein gradient–dependent manner. The formation of this sharp boundary between hb-expressing and -nonexpressing cells has been characterized in real time by monitoring the activity of the hbpromoter with the MS2-MCP system (53, 97). Interestingly, cells in the posterior initiate transcription at the same time as cells in the anterior, but the concentration of the BCD protein determines the strength and length of transcription. Spontaneous activation of the hb promoter becomes repressed in the posterior from nuclear division 11. It is also remarkable that all active cells exhibit synchronized initiation of transcription after mitosis and that once transcription is on in a cell, there is a single peak of activity (53, 97). This deterministic expression differs from the highly variable bursty profile displayed by the even-skipped (eve) stripe 2 gene. Remarkably, this well-defined pattern shows a broad activation in interphase of nuclear divisions 11 to 13, which is refined to the stripes at cellularization and disappears at gastrulation (13).

These examples highlight the relevance of visualizing individual cells and their activity over time to unmask mechanisms such as Drosophilapattering. Nonetheless, this strategy has only begun to characterize the initial processes in the life of a multicellular organism. In Drosophila, the complex interaction of ~1,000 enhancers defines the activity of a few genes. The strategies used to examine single enhancers and promoters at the hb and eve stripe 2 genes, cited above, have been extended to characterize the interplay of pairs of enhancers for hb, kni, and snapromoters. Each case showed its own individual scenario to reach precise and reliable levels of expression (14).

If there are many combinations of factors to consider, it is necessary to integrate thousands of experiments to take into account the variability of gene expression. Hence, it is a necessity to explore new technologies that will allow the visualization of hundreds of genes at once. These experiments must consider not only transcription, the first step of gene expression, but also post-transcriptional regulatory events that control localization, stability, and translation of an mRNA.

Single-Cell Analysis in Rare Cells: Role of Stem Cells in Tissue Homeostasis

Single-cell analysis has the unique advantage of being able to characterize low-abundance cells, such as stem cells, in the context of their environment (145). A representative example is the LGR5⁺ epithelial stem cells that orchestrate the homeostasis of the small intestine. Tracking of differentially labeled stem cells has resolved the structure, dynamics, and plasticity of the small intestine in time and space, and provided a platform to analyze the expression patterns of individual cells (35). This has been done by two complementary methods: smFISH in tissue and RNA-seq. SmFISH has revealed the spatial distribution of stem cell markers in the crypt, regulatory coexpression of transcripts, and changes in expression upon irradiation. Interestingly, cell positioning in the tissue defines its pattern of gene expression, which is similar among intestinal crypts and mice and over time (59, 72). The most relevant consequence of these single-cell studies is their therapeutic implications. In vitro production of mini-guts from single stem cells facilitates the regeneration of the damaged intestine and sheds light on the molecular origin of adenocarcinoma (124).