Illuminating RNA biology through imaging

Abstract

RNA processing plays a central role in accurately transmitting genetic information into functional RNA and protein regulators. To fully appreciate the RNA life-cycle, tools to observe RNA with high spatial and temporal resolution are critical. Here we review recent advances in RNA imaging and highlight how they will propel the field of RNA biology. We discuss current trends in RNA imaging and their potential to elucidate unanswered questions in RNA biology.

The transformation from DNA to protein is a complex, multi-stage process that revolves around RNA metabolism. After transcription, RNA molecules proceed to splicing, localization, translation and degradation. These steps are highly coordinated and tightly regulated in both spatial and temporal domains. Traditional biochemistry and genetic tools have elucidated some of the what and the how, such as the identities and functions of proteins and non-coding RNAs (ncRNAs) involved in each step of RNA processing. To delve deeper into the when and where, methods to visualize RNA within cells are required. Towards this goal, in the past four decades groups have developed and advanced RNA imaging tools for both fixed and live cells (Fig. 1 and Table 1). These RNA imaging tools take advantage of recent and rapid innovation in fluorescent microscopy, image processing, DNA chemistry and next-generation sequencing to achieve multiple milestones, including single-molecule sensitivity, super-resolution, multiplexing and live-cell RNA tracking. In this Review we discuss the developments in RNA imaging and the RNA biology they have and are poised to unravel.

Fig. 1 |. Timeline of subcellular RNA imaging technologies.

a, The development of fixed-cell RNA imaging from the development of fluorescent RNA detection in 1982. smFISH amplifies signals by utilizing multiple fluorescent DNA probes binding to a single RNA target. RCA-FISH, clampFISH, HCR-FISH, RNAscope and SABER-FISH enhance signals by amplifying the primary probes that hybridize to the RNA target by RCA, by secondary and tertiary probes, or by primer-exchange reaction (PER). Fluorescently labelled DNA probes bind to these amplified sites and emit much brighter signals compared to smFISH. Combinatorial FISH methods (MERFISH, seqFISH) and in situ sequencing (ISS, FISSEQ, STARmap, HybISS, ExSeq, Seq-Scope) methods enable multiplexing. b, The evolution of live-cell RNA imaging started with the microinjection of fluorescent RNA in 1997. The stem-loop system takes advantage of a fluorescent protein attached to a viral coat protein that can bind to an RNA stem loop, such as MS2, enabling single-molecule resolution for the first time. The molecular beacon, which remains dark until hybridized to a target RNA, can be delivered to cells to image endogenous RNA. Fluorogenic RNA takes advantage of an RNA aptamer that can encase the fluorophore structure 4-hydroxybenzylidene imidazolinone and emit light. Microinjected fluorescent dUTPs can be incorporated into endogenous RNA, enabling visualization. Recently, conjugates of fluorescent protein and RNA-binding Cas proteins were developed to visualize endogenous RNA in living cells. ‘N hybridizations’ indicates N rounds of hybridizations; Rehyb., rehybridization.

Table 1 |.

Current methods of RNA imaging

Method	Live/fixed	Throughput	Error detection	Optical de-crowding	Isoform	RNA species specificity	Genetically modify	Single-molecule sensitivity	Detection of endogenous unmodified RNA	Commercial product	Ref.
smFISH	Fixed	Single gene per colour	No	Not needed	No	Yes	No	Yes	Yes	Stellaris	2–5
smiFISH	Fixed	Single gene per colour	No	Not needed	No	Yes	No	Yes	Yes	No	10
SNV FISH	Fixed	Single gene per colour	No	Not needed	Limited	Yes	No	Yes	Yes	No	11
inoFISH	Fixed	Single gene per colour	No	Not needed	Limited	Yes	No	Yes	Yes	No	12
HCR-FISH	Fixed	Single gene per colour	No	Not needed	Limited	Yes	No	Yes	Yes	Molecular Instruments	8
SABER-FISH	Fixed	10	No	Not needed	Limited	Yes	No	Yes	Yes	No	9
ClampFISH	Fixed	Single gene per colour	No	Not needed	Limited	Yes	No	Yes	Yes	No	7
RNAscope	Fixed	12	No	Not needed	No	Yes	No	Yes	Yes	ACD Bio	6
seqFISH+	Fixed	10,000	Yes	Sparse labelling	Limited	Yes	No	Yes	Yes	Spatial Genomics	17
MERFISH	Fixed	10,000	Yes	Expansion microscopy	Limited	Yes	No	Yes	Yes	Vizgen	15,16,18
STARmap	Fixed	1,000	Yes	None	No	Yes	No	Yes	Yes	No	22
HybISS	Fixed	119	Yes	Sparse labelling	No	Yes	No	Yes	Yes	Cartana (10x Genomics)	20
FISSEQ	Fixed	Whole transcriptome	No	Expansion microscopy	Yes	Yes	No	Yes	Yes	Readcoor (10x Genomics)	23–25
Fluorescently labelled RNA	Live	Single gene per colour	No	Not needed	No	Yes	No	No	No	No	27
RNA stem-loop system	Live	Single gene per colour	No	Not needed	No	Yes	Yes	Yes	No	No	31,32
Fluorogenic RNA	Live	Single gene per colour	No	Not needed	No	Yes	Yes	Yes	No	No	46,48,52
Molecular beacon	Live	Single gene per colour	No	Not needed	No	Yes	No	Yes	Yes	No	53,54,56
Fluorescent dUTP	Live	Single gene per colour	No	Not needed	No	No	No	No	Yes	No	57,58,70
Cas system	Live	Single gene per colour	No	Not needed	No	Yes	Yes	No	Yes	No	61–63

RNA imaging technologies

RNA imaging technologies have been evolving rapidly for both fixed and live cells. In fixed cells, current methods have achieved substantial throughput and are capable of detecting localization and quantifying the expression level of the whole transcriptome. In live cells, throughput is limited to a single gene per colour; however, the temporal resolution of live-cell RNA imaging has significantly advanced our understanding of the dynamics of RNA processing.

Fixed cells and fluorescence in situ hybridization

In 1982, Singer and Ward were among the first to demonstrate fluorescence in situ hybridization (FISH) for RNA detection by probing actin messenger RNA (mRNA) with rhodamine-conjugated avidin binding to a DNA probe with incorporated biotinylated 2′-deoxyuridine-5′-triphosphate (dUTP)¹. In 1998, single-molecule FISH (smFISH) was developed using a complementary DNA (cDNA) oligonucleotide synthesized with five fluorochromes per probe². In 2008, the method was further refined to detect mRNAs at single-molecule resolution by probing each mRNA with 48 DNA probes, each labelled with single fluorochromes³. Rather than tiling multiple probes to a desired mRNA target, rolling-circle amplification (RCA)-FISH first hybridizes and ligates a padlock probe specific to the mRNA target and then amplifies the padlock probe using RCA^4,5.

Innovation of single-molecule RNA imaging continues to build on smFISH and RCA-FISH to further improve detection efficiency, increase brightness and reduce overall cost. RNAscope leverages multiple tiled primary, secondary and tertiary DNA oligonucleotide probes⁶. Similarly, click-amplifying FISH (clampFISH) iteratively hybridizes padlock probes to target mRNAs and ligates them using bio-orthogonal click chemistry before hybridization with a FISH probe, resulting in a >400-fold signal amplification per single molecule of RNA⁷. Rather than tiling probes along a transcript, hybridization chain reaction (HCR)-FISH⁸ and signal amplification by exchange reaction (SABER)-FISH⁹ amplify primary probes with hairpin probes and concatemers, respectively, to tile fluorescent secondary probes along a primary probe. Other groups have made smFISH more cost-effective (smiFISH)¹⁰ or have enabled the detection of single-nucleotide variants (SNVs) on transcripts (SNV FISH)¹¹ or adenosine-to-inosine-edited transcripts (inoFISH)¹².

Fixed-cell, multiplexed RNA imaging.

Subcellular multiplexed RNA imaging methods generally fall into two categories: combinatorial FISH and in situ sequencing.

Combinatorial FISH.

Combinatorial FISH assigns each unique RNA target a ‘spectral barcode’, with each bit in the barcode corresponding to a specific fluorochrome in a specific round of imaging. Increasing the number of bits in a barcode exponentially scales the number of unique transcripts that can be detected. In 2002, five pseudocolours and two rounds of imaging were leveraged to image ten unique transcripts¹³. In 2014, sequential FISH (seqFISH) used four colours and two rounds of imaging to detect 12 unique transcripts in budding yeast¹⁴. The advent of multiplexed error-robust FISH (MERFISH) represented the first time the combinatorial labelling of RNA had pushed beyond 100 unique transcripts¹⁵.

Subsequent developments to MERFISH¹⁶ and seqFISH^14,17 both enable the detection of 10,000 unique RNA targets, but differ in how they address the challenge of optical crowding. Whereas MERFISH leverages expansion microscopy (ExM)¹⁸, seqFISH+ opts for a sparse labelling approach¹⁷ by detecting a small fraction of targets at each round of imaging.

In situ sequencing.

In 2013, in situ sequencing (ISS)¹⁹ leveraged RCA-FISH and sequencing-by-ligation (SBL) to amplify and read out the barcode and identify the location of target mRNA. With modifications in probe design leading to a new barcoding system, the next iteration of ISS, hybridization-based ISS (HybISS), provided improved spatial detection of RNA transcripts²⁰. BaristaSeq²¹ followed a similar strategy but used Illumina sequencing-by-synthesis (SBS) chemistry. Recently, STARmap²² increased the fidelity of ISS by using two partially complementary probes to label each target, a new error-robust SBL scheme (SEDAL) to sequence 5-nt barcodes, and hydrogel embedding to remove background autofluorescence. These advances enable STARmap to measure 1,020 genes simultaneously in intact medial prefrontal cortex tissue with an error rate of only ~1.8%.

Fluorescent in situ RNA sequencing (FISSEQ)^23,24 attempted the unbiased single-molecule measurement of all RNAs. Rather than hybridization with a padlock probe, FISSEQ hybridizes random hexamer primers. After reverse transcription, the cDNA itself is circularized using CircLigase II, becoming a template for RCA. Using SOLiD sequencing, the cDNA is partially sequenced and aligned to the genome. Although the unbiased measurement of the whole transcriptome was a major technical advancement, optical crowding, dominance of rRNA in resulting reads, and low circularization efficiency remain substantial hurdles to its widespread adoption. Expansion sequencing (ExSeq) addressed some of these limitations by pairing FISSEQ with ExM and ex situ sequencing to improve the overall detection efficiency and fidelity²⁵.

A promising new front in the battle of multiplexed RNA imaging methods is the use of RNA captured on spatially barcoded slides. The recently developed Seq-Scope repurposes Illumina next-generation sequencing (NGS) chemistry to generate clusters from captured RNAs with a distance of 0.5–0.8 μm between clusters²⁶. Table 1 compares current methods of RNA imaging in fixed cells.

Live-cell, exogenous RNA imaging.

Fluorescently labelled RNA.

In 1997, Glotzer and colleagues microinjected fluorescently labelled oskar RNA into Drosophila oocytes to study its short-range and long-range transport²⁷. Using similar strategies, microtubule-dependent transport of other RNAs in Drosophila oocytes, including wingless and bicoid, was also examined^28,29. A drawback with microinjected RNA is the susceptibility to endosome entrapment³⁰.

RNA stem-loop systems.

In 1998, Singer and colleagues developed the RNA stem-loop system to visualize ASH1 mRNA localized to the bud tip in Saccharomyces cerevisiae³¹. This system consists of two plasmids. One plasmid encodes a green fluorescent protein (GFP) fused to the coding sequence for a single-stranded RNA phage capsid protein MS2, also called MS2 coat protein (MCP). The second plasmid expresses a reporter RNA containing the coding sequence of a protein of interest followed by six MS2 binding sites (MBSs). In 2003, single-molecule resolution of the MS2 system using 24 MBSs was demonstrated³². Several improvements on the first generation of MS2 have been developed to (1) overcome deletion of repetitive MS2 sequences³³, (2) improve the degradation and turnover of reporter mRNA carrying MS2³⁴, (3) enhance the signal-to-noise ratio and uniformity of RNA labelling³³ and (4) reduce background caused by unbound fluorescent protein by using split fluorescent protein or split Halotag^35–37. Besides MS2, other RNA stem-loop systems have also emerged, including PP7, λ_N22, U1A and BglG^38–42. In these systems, the stem-loop length varies from 15 to 29 nucleotides with their protein binding partners’ sizes ranging from 22 to 129 amino acids⁴³. The MS2/PP7 systems are relatively resistant to photobleaching as there are 48 GFPs on each mRNA, enabling RNA tracking to study the dynamics of mRNA processing^32,44. Furthermore, the MS2 system could be genetically integrated into endogenous genes to study mRNA dynamics in live mouse brain tissue⁴⁵.

Fluorogenic RNA.

In 2011, Jaffrey and colleagues reported an RNA aptamer that mimics GFP⁴⁶. In GFP, the three residues Ser⁶⁵-Tyr⁶⁶-Gly⁶⁷ form a fluorophore structure, 4-hydroxybenzlidene imidazolinone (HBI). Based on this principle, the authors performed systematic evolution of ligands by exponential enrichment (SELEX) and found an RNA aptamer, named Spinach, that can encase HBI, leading to fluorescence. To overcome the thermal instability and misfolding of Spinach, Spinach 2 was developed⁴⁷. Following similar SELEX approaches for different fluorophores, other fluorogenic RNA systems, such as Broccoli, Mango, Pepper and Peach, have been engineered^48–51. Recently, based on the bright and thermodynamically stable Mango aptamer, the Mango II array with 24 repeats of the aptamer sequence has been shown to achieve single-molecule resolution for live-cell RNA imaging⁵².

Live-cell, endogenous RNA imaging.

All three systems, fluorescently labelled RNA, RNA stem-loop and fluorogenic RNA, are among the earliest methods developed to visualize RNA in living cells and have elucidated multiple aspects of RNA biology. One drawback of these systems is the inability to image endogenous, non-genetically modified mRNA. Chemically synthesized probes and genetically encoded probes are alternatives that can overcome this limitation.

Chemically synthesized probes.

In 1996, Tyagi and Kramer invented a single-stranded oligonucleotide probe, named ‘molecular beacon’, that fluoresces upon hybridization to target RNA⁵³. Despite being proposed in the 1996 report that molecular beacon was suitable for RNA imaging in live cells⁵³, it was not until 2003 that this capability was demonstrated⁵⁴. To overcome their instability in living cells, multiple chemistry modifications have been applied to molecular beacons, including 2′-O-methylribonuclotides, phosphorothioate backbones and locked nucleic acids⁵⁵. In 2018, molecular beacon was shown to image endogenous RNA in living neurons with single-molecule resolution⁵⁶.

Another system that visualizes endogenous RNA involves incorporating fluorescently labelled dUTP into RNA during RNA synthesis. Typically, fluorescently labelled dUTPs are injected into early-stage embryos. RNAs with incorporated dUTPs are imaged in neurons differentiated from these embryos either in vitro or in vivo^57–59. A limitation of this system is the inability to track specific RNA, as fluorescently labelled dUTP can integrate into any RNA.

Genetically encoded probes.

Following the discovery of clustered regularly interspaced short palindromic repeats (CRISPR) and CRISPR-associated (Cas) proteins that target DNA, it was found that in vitro programmable targeting of RNA is possible with Cas9 (RCas9)⁶⁰. RCas9 can target RNA when the protospacer adjacent motif (PAM) sequence is provided in trans as a separate DNA oligonucleotide. In 2016, our laboratory demonstrated that RNA tracking in live cells was possible with RCas9 fused to a GFP⁶¹. In 2017, Zhang and colleagues showed that Cas13a can be engineered to target mammalian RNA and demonstrated live-cell RNA imaging with catalytically inactive Cas13a (dCas13a) fused to GFP⁶². A recent study in 2019 has compared the ability of multiple dCas13 proteins to image RNA in living cells and provide an improved signal-to-noise ratio by incorporating multiple fluorescent proteins into a single dCas13 protein⁶³. Despite these efforts to engineer Cas systems for live-cell RNA imaging, single-molecule resolution has yet to be achieved. Table 1 compares current methods of live-cell RNA imaging.

RNA biology gained via imaging technologies

The advances in RNA imaging described above have increased our understanding of RNA throughout its functional life-cycle: transcription, splicing, localization, translation and degradation (Fig. 2).

Fig. 2 |. Highlights of RNA biological insights gained through RNA imaging.

a, Multiplexed RNA imaging combined with chromatin tracing, such as with seqFISH+, can be used to elucidate nuclear domains with distinct chromatin states and gene expression¹⁵¹. b,c, Live-cell imaging using stem-loop systems can examine transcriptional properties such as bursting at transcription sites (TSs)⁶⁹ (b) and the temporal and spatial characteristics of splicing⁷⁵ (c). d, Stem-loop system and translating RNA imaging by coat protein knock-off (TRICK) assay revealed the dynamics of the initiation of protein synthesis on a single RNA¹⁰⁷. e,f, Stem-loop systems can also be leveraged to track the dynamics of RNA nuclear export⁸⁵ (e) and RNA transport over time (f) in highly asymmetric cells such as neurons⁵⁶.

Transcription.

Live-cell RNA imaging with MS2 systems can examine multiple transcriptional properties. For example, it has been used to describe transcriptional bursting^64,65. A combination of fluorescently tagged RNA polymerase II and MS2 labelling of nascent mRNA has been applied to measure the elongation rate⁶⁶. High-speed time-series measurements were able to discern elongation rate as well as observe multi-scale transcriptional bursting controlled via groups of closely spaced polymerases, termed ‘convoys’⁶⁷. MS2-based bursting measurements from a single gene have been achieved using fluorescence fluctuation microscopy⁶⁸. In vivo detection of transcriptional bursting was also demonstrated in acute brain slices from transgenic mice with 24 repeats of MS2 binding sites inserted into the β-actin gene⁴⁵. MS2 systems have also been used to correlate the binding of Gal4 transcription factor to chromatin with transcriptional bursting⁶⁹.

Fluorescently labelled dUTPs combined with fluorescence anisotropy imaging have shown that chromatin structures are more open at transcriptionally active compartments in living cells⁷⁰. Recently, MERFISH and seqFISH+ have been modified to study how chromosome three-dimensional (3D) organization affects transcriptional activity. seqFISH+ was modified to target the intronic regions of 10,421 genes and uncovered that nascent transcription sites were localized to the surfaces of chromosomes⁷¹. DNA-MERFISH was developed to trace chromatin itself at the genome scale. A combination of DNA-MERFISH, MERFISH and immunofluorescence has simultaneously imaged over 1,000 gene loci, nascent transcripts from these loci and nuclear structures (nuclear speckles and nucleoli)⁷².

Subcellular RNA imaging may continue to answer critical questions in transcription. The combined progress in live-cell RNA imaging at transcription sites and chromosomal architecture imaging with MERFISH and seqFISH+ may make headway towards understanding the mechanism of transcriptional bursting. The use of pooled genomic screens in concert with RNA imaging⁷³ can assess the contribution of different transcriptional activators and repressors.

Splicing.

In the early 1990s, following the finding that ~90% of pre-mRNAs are spliced during or after transcription⁷⁴, the next quest was to decipher the structural and kinetic coupling of splicing and transcription. Using smFISH, Tyagi and colleagues found that, when the intronic polypyrimidine tract is present within a strong secondary structure, splicing is uncoupled from transcription and delayed until transcription is completed⁷⁵. Using live-cell RNA imaging with the MS2 system in combination with fluorescence recovery after photobleaching (FRAP), Shav-Tal and colleagues showed that splicing events do not affect polymerase elongation kinetics⁷⁶. Using the MS2 system, it was shown that transcription is the rate-limiting step for the excision of long introns⁷⁷. Combining MS2 and PP7 systems, Larson et al. labelled the introns with one fluorescent colour and the exons with another fluorescent colour to track transcription and splicing events simultaneously, and found that the two processes are coordinated via kinetic competition⁷⁸. These studies have also shown that splicing occurs at variable timescales from 20 s to minutes. Furthermore, when quantifying at the level of a single cell, alternative splicing seems to occur stochastically, exhibiting cell-to-cell variability⁷⁹.

Transcriptomic studies have suggested alternative splicing as a mechanism for RNA localization^80,81. Recently, APEX-seq, developed by Ting and colleagues, has further pushed the spatial resolution of transcriptome-wide mapping of isoforms by enabling the mapping of RNA localization to nine different organelles⁸². In highly asymmetric cells like neurons, where RNA localization along the neurites is linked to precise function, an even higher spatial resolution of transcriptome mapping will help to understand the precise role of alternative splicing in subcellular localization.

RNA transport.

The first study using an RNA stem-loop imaging system showed that ASH1 mRNA exhibited bidirectional movement with occasional stalling in yeast and reported that the transport speed ranged from 200 to 400 nm s⁻¹, consistent with the speed of the myosin V motor³¹. In mammalian systems, Arc mRNA appeared to travel at 0.1–1 μm s⁻¹ (ref.⁸³). Tracking RNA movement in Cos cells revealed four types of mobility with different probability: (1) immobility (33–40%), (2) directional movement (2–5%), (3) restricted diffusion (40–45%) and (4) diffusion (15–25%). Interestingly, mRNA can dynamically switch from one type of mobility to another³².

RNAs synthesized within the nucleus exit through the nuclear pore complex (NPC) for further processing. Both the MS2 system⁸⁴ and molecular beacon⁸⁴ showed that ribonucleoprotein (RNP) complexes follow a diffusional model in the nucleus. mRNA export through the NPC appeared to be faster than a simple diffusion model based on imaging with the MS2 system⁸⁵. Consistently, nucleocytoplasmic transport was found to follow a three-step model consisting of (1) docking (80 ms), (2) transport (5–20 ms) and (3) release (80 ms), in which transport through NPC was not the rate-limiting step. This finding was made possible using the MS2 system in combination with a super-registration approach capable of resolving 20-ms time precision and 26-nm spatial resolution⁸⁶.

Although previous studies suggested that multiple mRNA species can be packaged and transported within the same RNP^87–89, a study using smFISH and quantitative imaging showed that mRNA molecules travelled singly in neuronal dendrites⁹⁰. A similar observation was made using a molecular beacon, which revealed that more than 70% of β-actin mRNA molecules travelled singly in neuronal axons⁵⁶. This study also teased out different transport models in axons, including (1) diffusion and entrapment, (2) directed transport by motor proteins, (3) localized protection of mRNA from degradation and (4) a ‘sushi belt’ model that incorporates directed transport with local entrapment at synapses.

RNAs are thought to be packaged into RNPs for transport. However, the composition of RNPs and the process of packaging remain unclear. Further investigations are required to figure out the biological rules that govern RNP packaging, such as identification of the RNA-binding protein (RBP) required for packaging and transporting a given RNA. After being packaged, transport RNPs are thought to be shuffled to their destination by motor proteins via microtubes⁹¹. The adaptors connecting RNPs and motor proteins remain unknown. A recent study suggests that RNPs can attach to the lysosome for long-distance transport⁹². To what extent this mode of transportation applies to the transcriptome requires future studies. It has been suggested that mRNA molecules travel singly in neuronal dendrites and axons^56,89. These studies, however, are limited to a small number of RNA species. Transporting a single mRNA at a time could guarantee specificity, although it appears to cost more energy than transporting multiple mRNAs at once to the same destination. A transcriptome-wide effort could give more insights into the universality of this process.

RNA localization.

The influence of subcellular RNA localization on cellular functions in a variety of cell types has been extensively reviewed^93–97. Here we briefly discuss critical findings in the context of the advancement of subcellular RNA imaging technology.

In 1986, Singer and Lawrence were the first to establish distinct localization patterns of actin, vimentin and tubulin mRNAs in intact somatic cells via ISH⁹⁸. Following this initial study, multiple investigations into the subcellular localization of individual mRNA species were conducted using smFISH^99–102. Recently, MERFISH was performed to interrogate the subcellular localization of transcripts from ~10,000 genes¹⁶. As a validation, transcripts from gold-standard secretome genes were found to be enriched in the endoplasmic reticulum¹⁶. Furthermore, the authors developed a pseudotime method based on nuclear/cytoplasmic RNA enrichment to indicate the cell-cycle state of individual cells.

Highly asymmetric cells such as neurons leverage localized translation to respond to stimuli with low latency. Local translation of β-actin mRNA following glutamate uncaging was demonstrated by a combination of FISH and a HaloTag-actin reporter construct to measure actin transcripts and proteins in dendritic spines¹⁰³. Similarly, smFISH demonstrated that intestinal epithelia cells leverage asymmetric subcellular localization to polarize translational efficiency¹⁰⁴. In axons, fluorescent-UTP labelling and SunTag nascent protein labelling were used to demonstrate that Rab7a endosomes carrying mRNA and ribosomes pause on mitochondria to translate mRNAs encoding mitochondrial proteins while traversing axons⁵⁷. Even within non-polarized cells, mRNA localization was found to depend on ongoing local translation, suggesting co-translational RNA targeting¹⁰⁵.

RNA mislocalization has been implicated in multiple neurodegenerative diseases¹⁰⁶, and transcriptomic sequencing studies have identified those mislocalized mRNAs. The advent of spatial transcriptomics and live-cell RNA imaging equips us with the ability to study mRNA mislocalization at higher spatial and temporal resolution.

Translation.

Using translating RNA imaging by coat protein knock-off (TRICK), a double labelling of PP7 in the coding sequence and MS2 in 3′UTR, Chao and colleagues observed that mRNAs are not translated in the nucleus, but are translated within minutes of export¹⁰⁷. Dual labelling of translating protein and RNA via SunTag and MS2 was used to understand translation in sub-dendrites¹⁰⁸, the number of ribosomes per polysome¹⁰⁹, as well as how ribosome occupancy decompacts mRNA¹¹⁰. Furthermore, the SunTag/MS2 strategy has enabled the discoveries that mRNAs resume translation during recovery from stress¹¹¹ and that mRNAs are translated in stress granules, arguing against a direct role of stress granules in the inhibition of protein synthesis¹¹².

Whereas live-cell imaging enables an understanding of the temporal dynamics of translation, fixed-cell RNA imaging allows the study of translation dynamics at a broader scale. A combination of smFISH and nascent protein staining by O-propargyl-puromycin revealed that global mRNA localization in the intestinal epithelium is polarized, which leads to a polarization in translational efficiency¹⁰⁴. A similar approach has shown that mRNA localization requires ongoing translation, leading to widespread co-translational RNA targeting¹⁰⁵.

RNA degradation.

Using smFISH, Singer and colleagues studied cell-cycle-regulated RNA degradation in yeast and found that promoter-dependent activity directly influences how and when an mRNA will be degraded in the cytoplasm¹¹³. To study mRNA degradation at higher temporal resolution in living cells, Chao and colleagues developed a technique called 3′-RNA end accumulation during turnover (TREAT)¹¹⁴, which utilizes a fluorescent reporter that leverages the orthogonality of MS2 and PP7 systems to label intact and degraded mRNAs. Using TREAT, they found that, unlike for transcription, mRNA degradation does not burst. By labelling processing bodies (P-bodies) simultaneously with TREAT, they found that a majority of TREAT mRNAs are not degraded in P-bodies. This provides a new understanding of P-bodies, which were previously considered to be the centre of RNA degradation¹¹⁵. Furthermore, mRNAs localized to stress granules and P-bodies when exposed to stress showed no difference in degradation dynamics during recovery compared with cytosolic mRNAs¹¹¹.

In addition to the degradation of normal transcripts, cells have developed nonsense-mediated decay (NMD) to eliminate transcripts harbouring a premature termination codon. Imaging translating mRNA with the MS2/SunTag system¹¹⁶ showed that NMD efficiency is affected by the number of introns and that, for the same RNA, each round of translation has an equal probability of inducing NMD¹¹⁷.

Besides active transport and diffusive models, degradation has been proposed as a mechanism to induce and maintain RNA localization¹¹⁸. mRNAs transported in RNPs are typically protected from degradation, ensuring proper delivery to their destination. Future studies with high spatial and temporal resolution will shed light on the interplay between RNA degradation and localization.

ncRNA.

Even though more than 85% of the genome is transcribed to RNA¹¹⁹, only <2% of the mammalian genome encodes proteins¹²⁰. Hence, a majority of transcribed RNAs are ncRNAs, such as microRNAs (miRNAs) and long ncRNAs (lncRNAs). Intracellular single-molecule, high-resolution localization and counting (iSHiRLoC) has been developed to track the localization of microinjected fluorescently labelled miRNAs in living HeLa and U2OS cells^121–124 and revealed two kinetically distinct pathways of miRNA assembly into large RNPs¹²¹. iSHiRLoC also showed that miRNA stability and nuclear retention were dependent on Argonaute (Ago) proteins and targets. Furthermore, miRNA unwinding, strand selection and cytoplasmic retention were dependent on Ago2¹²³. iSHiRLoC, together with tracking of fluorescently labelled P-bodies, revealed that miRNAs localized to P-bodies are mostly dysfunctional¹²⁴.

An smFISH survey on the localization of 61 lncRNAs found that nearly half exist in the cytoplasm¹²⁵. For those lncRNAs that localize in the nucleus, their distribution can be either diffuse, in foci or in speckles and paraspeckles, like MALAT1 and NEAT1, respectively. Live-cell imaging of NEAT1 using dCas13-GFP showed that paraspeckles underwent ‘kiss-and-run/fusion’ dynamics, where materials rapidly moved in and out of paraspeckles⁶³. smFISH revealed that lncRNAs exhibit cell-to-cell expression variability¹²⁵, like mRNAs. smFISH also showed that imprinting lncRNAs Kcnqlot1¹²⁶ and Air¹²⁷ localized at their target sites of transcription on the same allele, suggesting that these lncRNAs may silence their target genes in cis.

Viral RNA.

The human immunodeficiency virus (HIV) is one of the most well-studied RNA viruses. A study using the RNA stem-loop system and live-cell imaging found that more than 90% of HIV-1 particles contain viral RNA³⁸. By labelling individual RNA strands with different colours, it further showed that the HIV-1 structural protein Gag packages a dimeric RNA molecule, not two monomeric RNA molecules. Simultaneous imaging of the Gag protein and HIV-1 genome RNA has uncovered their dynamics and functional interactions during viral particle assembly at the plasma membrane¹²⁸. The MS2/SunTag system shows that ~50% of HIV-1 RNA is actively translated and that Gag only packages non-translating RNA¹²⁹.

RNA imaging methods such as FISH and RNAscope have also been applied to detect the presence of viruses¹³⁰ and SARS-CoV-2^131–133, a single-stranded RNA virus that led to the COVID-19 pandemic with over six million fatalities worldwide. smFISH has been applied to visualize host mRNA dynamics during SARS-CoV-2 infection, revealing that the biogenesis of interferon (IFN) I and II, a marker of immune responses, is inhibited at multiple stages, including transcription induction, transcription release and nuclear-cytoplasmic transport of IFN mRNAs¹³⁴. smFISH also confirms that activation of the cellular oxygen-sensing pathway inhibit SARS-CoV-2 entry and replication in lung epithelial cells¹³⁵.

outlook

As mRNA imaging in fixed cells has evolved from a single target to the transcriptome scale, imaging speed and image analysis have remained bottlenecks to the study of subcellular mRNA localization. Furthermore, the ability to resolve multiple mRNAs as diffraction-limited spots has become a challenge, inhibiting our understanding of whether different species of mRNA can be co-processing in the same place. Efforts in artificial intelligence to automate cell segmentation, RNA location assignment and spot detection and tracking will further push the boundary of our current understanding of RNA localization^{101,136–138}. Going beyond expanding the number of mRNA species, the ability to image endogenous small RNAs, such as miRNA, and RNA isoforms will greatly enhance our understand of RNA biology at subcellular resolution.

The current live-cell RNA imaging methods have provided a huge leap towards a high spatiotemporal understanding of multiple aspects of RNA processing. However, studies have been limited to a few mRNA species and relatively short-term tracking. Live-cell RNA imaging with Cas holds great promise by offering a flexible, easy-to-use system to target any endogenous gene in the transcriptome, although single-molecule resolution has yet to be seen. In addition, a future system with multiplexing capability can open the door to explore whether and how different RNA species are co-processed. A limitation to long-term single-molecule tracking in live-cell imaging has been phototoxicity. Future developments in systems to overcome this will enable tracking mRNA throughout its life-cycle.

In addition to RNA imaging, multiple RNA sequencing and computational methods have been developed to study subcellular RNA localization, including APEX-seq⁸², RNA-GPS⁸⁰, LncLocator¹³⁹, RNALocate¹⁴⁰, iLoc-lncRNA¹⁴¹, Axon-seq¹⁴², CeFrac-seq¹⁴³ and RNATracker¹⁴⁴. Methods that detect RBP targets, such as TRIBE^145,146 and STAMP¹⁴⁷, can potentially be adapted to study RNA subcellular localization. Although these methods do not have the high spatiotemporal resolution of RNA imaging, the ability to multiplex and sequence isoforms is unmatched by current RNA imaging methods. Slide-seq¹⁴⁸, Seq-scope²⁶ and ExSeq²⁵ have pushed the limits of transcriptomic imaging from known targets to unbiased profiling. Future methods incorporating the power of both RNA imaging and sequencing will help us to make a big leap forward in RNA biology.

In the past, high-resolution imaging was often thought of as a low-throughput method and not suitable for high-throughput screening, in comparison to other fluorescent methods such as flow cytometry and the fluorescent microplate reader. However, innovations in optical instrumentation, automation and image analysis have added high-throughput capability into fluorescent imaging. High-throughput screens via protein imaging have expanded our understanding of gene and protein functions^73,149,150. We expect that new methods enabling high-throughput screens via RNA imaging will further contribute to our knowledge of not just gene and protein functions, but also RNA functions.

Finally, RNA processing involves not only RNA but also DNA and proteins. Going beyond an RNA-centric outlook, combining RNA imaging with DNA and RBP imaging will greatly enhance our understanding of RNA biology, answering questions such as how chromosome organization affects gene expression and how RNPs form and organize. Furthermore, an integration with high-throughput screen studies such as large-scale RBP–RNA interactions and CRISPR screens will also expand our toolbox to explore the multidimensionality of RNA processing (Fig. 3).

Fig. 3 |. The outlook towards a multidimensional approach to study RNA biology.

Integration of the high spatiotemporal approach of RNA imaging with high-throughput methods (such as RNA sequencing and the CRISPR screen) and large-scale RBP–RNA interaction mapping techniques to build a complete picture of RNA processing from RNA, DNA and protein perspectives.