The purpose of this article is to review techniques used to visualize RNA localization in living cells, however FISH deserves a brief treatment here. Like all hybridization-based techniques, FISH detects endogenous or engineered RNA (or DNA) molecules using fluorescently-labeled probes complimentary to the sequence of interest. Two approaches used to achieve the signal-to-noise ratio required for single mRNA visualization include the use of 4–10 multiply-labeled probes [20], or using many (40+) singly-labeled probes [21]. Each of these approaches offers relative advantages and disadvantages in terms of sensitivity, specificity, and quantitation; however they share common drawbacks. The first limitation is that FISH requires samples to be fixed and therefore does not provide dynamic information about RNA trafficking or transient interactions with host factors. Additionally, fixation itself can affect signal intensity and may disrupt the integrity of certain organelles [22]. Furthermore, non-hybridized probe can remain within the sample, reducing the signal-to-noise ratio. Despite these limitations, single molecule FISH can provide quantitative data regarding the expression and localization of endogenous transcripts, which is advantageous over some of the other systems discussed here.

Similarly to FISH, RNAs can be visualized in living cells using fluorescently-labeled, linear probes complementary to the sequence of interest (Table 1) [23]. However, in live cells, unbound probe cannot be removed by washing. As a result, this approach requires the introduction of multiple unique probes targeting the same transcript so that localized concentrations of probes can be differentiated from high levels of background fluorescence [7]. Linear probes have been useful in visualizing RNAs associated with small nuclear ribonucleoprotein complexes (snRNPs) and ribosomal RNAs associated with nucleoli [6,7]. However, limitations including low signal-to-noise ratios, the need to microinject probes into cells of interest, and the tendency of probes to rapidly accumulate in the nucleus following microinjection have discouraged more widespread use of this approach [22].

To increase the signal-to-noise ratio associated with linear probes, some groups have utilized Förster resonance energy transfer (FRET)-based systems [8,9]. In this approach, two probes complementary to adjacent regions on the transcript of interest are labeled with donor and acceptor fluorophores, respectively. Hybridization of each probe to the transcript brings the fluorophores into close proximity, resulting in FRET when the donor fluorophore is excited (Table 1) [8]. Both probes must bind the transcript of interest to produce a FRET signal, increasing the specificity of this system over single linear probes [8]. The use of FRET also increases the signal-to-noise ratio of the system, because FRET only occurs when both probes are bound to the same transcript. The FRET signal can be further increased through the addition of multiple donor fluorophores [24]. Alternatively, background FRET signal can be reduced through the use of autoligation FRET probes. In this system, the donor probe is labeled with a fluorophore and a quencher, preventing fluorescence of the donor. Hybridization of the donor and acceptor probes to the transcript results in autoligation of the donor and acceptor probes with excision of the quencher, producing a FRET signal (Table 1) [10,11]. While FRET-based systems may overcome some of the sensitivity issues associated with single linear probes, they share the same delivery obstacles. As a result, these systems have not been used extensively to visualize viral RNA in living cells.

Unlike linear probes, molecular beacons form stem-loop structures. The 5’ and 3’ ends of the beacon form a self-complimentary stem, while the central loop region is complimentary to the transcript of interest [12]. The 5’ and 3’ ends of the molecule are conjugated to a fluorophore and a quencher, which prevents fluorescence when the beacon is in a stem-loop conformation. Binding of the beacon to the transcript induces melting of the molecular beacon stem structure, moving the fluorophore away from the quencher and resulting in fluorescence (Table 1). A further increase in the signal-to-noise ratio associated with molecular beacons can be achieved using FRET between two molecular beacons labeled with donor and acceptor fluorophores [25]. Owing to the decreased background fluorescence associated with molecular beacons compared to linear probes, molecular beacons have been used more extensively to visualize viral RNAs in living cells. This approach has been used to monitor the spread of bovine and human respiratory syncitial viruses [26,27] and coxsackievirus B6 [28] in tissue culture. Additionally, molecular beacons have been used to study the intracellular trafficking dynamics of poliovirus positive-stranded RNA [29] and influenza A virus mRNA [30].

To study viral RNA trafficking, molecular beacons offer some advantages over the fluorescent protein-based RNA labeling techniques discussed below. Primarily, because molecular beacons hybridize to a sequence of interest, this approach can be used to visualize native (i.e., non-engineered) viral RNAs. As a result, molecular beacons can be used to visualize viral RNAs in clinical isolates [26], obviating the need to introduce a foreign RNA sequence into the viral RNA of interest. However, the use of molecular beacons to study retroviral RNA trafficking may be complicated by the presence of endogenous proviruses, which frequently share a high degree of sequence homology with exogenous retroviruses. For example, many strains of laboratory mice express up to eight endogenous retroviruses with high homology to exogenous mouse mammary tumor virus (MMTV) [31]. Consequently, care must be taken to design molecular beacons that effectively distinguish exogenous retroviral RNA from RNA expressed by endogenous retroviruses.

An additional obstacle to the use of hybridization-based probes, including molecular beacons, is the delivery of probes into living cells. Microinjection is the most direct delivery method and allows for very rapid visualization of target RNAs [32], but poses technical challenges. Accordingly, other methods are frequently used to deliver molecular beacons into cells. One approach is to reversibly permeabilize cells using streptolysin O [26,30]. This technique offers rapid visualization of target RNAs without the technical requirements of microinjection [23]. Rapid introduction of oligonucleotide probes or molecular beacons into cells can also be accomplished by conjugating the probe to a cell-penetrating peptide, including HSV-1 VP22 or HIV-1 Tat [22]. Alternatively, cationic lipid-based transfection reagents have been used to deliver molecular beacons into cells [29]. Unlike permeabilization or cell-penetrating peptides, however, these reagents may deliver molecular beacons to the endocytic pathway, resulting in nuclease degradation of the beacon and increased background fluorescence [22]. However, non-specific fluorescence associated with molecular beacon degradation can be reduced using 2’O-methylated probes [7] or by using specific quantum dot-molecular beacon conjugates [33]. Other transfection methods such as electroporation have also been used to introduce molecular beacons into live cells, though all transfection-based delivery methods impart a lengthy delay before target transcripts can be visualized [22]. As a result, molecular beacons and other hybridization-based techniques can be powerful tools to track viral RNAs in living cells, but all of these approaches share some common limitations and are not universally applicable to every model system.

The MS2 RNA-binding protein forms the viral capsid of the MS2 bacteriophage. During the late phase of phage replication, an MS2 dimer binds a 19-nucletide stem-loop structure on the phage genomic RNA, which initiates phage capsid assembly and encapsidation of the genome [34]. In 1998, Bertrand et al. [13] took advantage of this system to visualize RNA localization by linking GFP to an MS2 mutant that binds RNA but is incompetent for capsid assembly [35]. A nuclear localization signal (NLS) was added to the MS2-GFP fusion protein to reduce background fluorescence in the cytoplasm. To visualize RNA localization in yeast, six MS2 stem-loops were inserted between the LacZ coding region and the ASH1 3’ untranslated region (LacZ-MS2-ASH1) (Table 1) [13]. When MS2-GFP was expressed alone, the fluorescent signal accumulated in the nucleus by virtue of the NLS. However, when MS2-GFP was coexpressed with LacZ-MS2-ASH1, discrete MS2-GFP foci were visualized in the nucleus and cytoplasm. Subsequent refinements demonstrated that inserting 24 copies of the MS2 stem-loop sequence into the RNA yielded single-molecule sensitivity [14]. Interestingly, on average only 33 molecules of GFP were associated with each RNA molecule [14]. Because MS2 binds as a dimer, this result suggests that on average only about two-thirds of the 24 MS2 stem-loops were occupied by MS2-GFP. In fact, Fusco et al. [14] visualized single RNA particles containing as few as 20 GFP molecules, indicating that single RNAs can be visualized using smaller numbers of fluorophores.

The use of MS2-GFP offers several advantages to hybridization-based approaches; most notably RNA can be tracked in living cells, obviating the concern for fixation artifacts. Additionally, the high affinity interaction between the MS2 coat protein and the MS2 RNA (Kd = 0.4 nM) results in specific labeling of RNAs containing the MS2 stem loops [36] and the MS2-GFP NLS allows newly synthesized RNA to be bound by the reporter co-transcriptionally [37,38]. Together, these features result in a high signal-to-noise ratio in the cytoplasm [14]. This system does, however, have the disadvantage that MS2-GFP only binds to RNAs engineered to contain MS2 stem-loops, so it cannot be used to track native RNAs. Furthermore, achieving an optimal ratio of MS2-GFP to RNA expression may require titrating different amounts of each expression construct [13]. The localization of free MS2-GFP in the nucleus may partially obscure visualization of nuclear RNAs, although the punctate intranuclear MS2-GFP signal can be brought out by carefully titrating MS2-GFP expression [37,38,39]. Another potential concern arises from the possibility that the RNA-bound MS2-GFP protein could alter normal RNA trafficking. However, RNAs bound by MS2-GFP are efficiently exported from the nucleus and properly localized in the cytoplasm [14,40,41]. Furthermore, retroviral RNAs bound by MS2-GFP are efficiently encapsidated into virus particles [15]. Together, these data suggest that MS2-GFP does not interfere with proper RNA trafficking or binding by the retroviral Gag protein.

Due to its ease of use and relatively few drawbacks, the MS2-GFP system has been used to study retrovirus RNA biology in living cells. Boireau and colleagues [38] took advantage of the high affinity interaction between MS2-GFP and HIV-1 RNAs containing 24 copies of the MS2 stem-loop to study the kinetics of HIV-1 RNA transcription using fluorescence recovery after photobleaching (FRAP). MS2-GFP at sites of HIV-1 transcription was subjected to photobleaching, and fluorescence recovery occurred when free MS2-GFP proteins bound to newly transcribed RNAs bearing the MS2 stem-loops [38]. This allowed the authors to measure the rate of HIV-1 RNA transcription. In addition, due to very slow exchange of RNA-bound MS2-GFP with free MS2-GFP, this system can be used to track single RNA molecules over time. By using a photoactivatable form of GFP, RNA molecules can be differentially labeled at sites of transcription [37]. In theory, this approach could allow the tracking of single RNA molecules throughout the entire process of retrovirus particle formation.

The MS2-GFP system has also provided valuable insight into the viral and cellular factors that regulate the trafficking of retroviral RNAs. The canonical view of retroviral assembly holds that for C-type retroviruses, Gag and the gRNA interact at the plasma membrane to assemble new virus particles. However, using MS2-GFP, Gag-RNA interactions have been visualized at earlier locations in the assembly pathway. For example, the feline immunodeficiency virus (FIV) Gag protein co-localizes with the gRNA on the cytoplasmic face of the nuclear envelope [42], whereas HIV-1 and murine leukemia virus (MLV) Gag-gRNA interactions were visualized on the cytoplasmic face of endosomes [42,43]. The high sensitivity of the MS2-GFP system has also permitted a detailed kinetic analysis of individual HIV-1 virus particle formation at the plasma membrane [44]. Together, these data have provided insight into the intracellular trafficking of retroviral RNAs and the mechanism underlying the incorporation of RNA genomes into virus particles.

Another informative application of the MS2-GFP system is to visualize protein-RNA and RNA-mediated protein-protein interactions in living cells. This approach, called trimolecular fluorescence complementation (TriFC), utilizes MS2 conjugated to a non-fluorescent N-terminal fragment of Venus fluorescent protein (MS2-VN). Binding of an MS2 stem-loop containing RNA by MS2-VN and a target protein containing the C-terminal fragment of Venus (VC) results in reconstitution of the Venus fluorophore and detectable fluorescent signal [45]. Recently, Milev et al. [46] used TriFC to probe Gag-Gag and Gag-Staufen interactions on viral RNA. As expected, Gag-VC complemented MS2-VN on psi-containing RNAs, however Staufen-VC failed to complement MS2-VN in the absence of Gag [46]. To examine whether Gag could recruit Staufen to viral RNAs, the authors took advantage of the fact that the MS2 protein binds to RNA as a dimer. By co-expressing MS2-Gag with MS2-VN and VC-labeled target proteins, they found that HIV-1 Gag actively recruits Staufen to the viral RNA [46]. This study demonstrates that MS2-GFP is a powerful system to study the dynamic trafficking of viral RNAs and to probe host-virus interactions in living cells.

The Bgl-mCherry RNA labeling system, analogous to the MS2-GFP system, is derived from the E. coli anti-termination protein BglG. The first 58 amino acids comprise the BglG RNA-binding domain, and this protein fragment is sufficient for anti-termination activity [47]. To create a live-cell RNA labeling system, Chen et al. [15] fused the N-terminal 52 amino acids of the BglG RNA-binding domain to mCherry and an NLS to create Bgl-mCherry. Bgl-mCherry binds a 29-nucleotide stem-loop structure, and single molecule sensitivity has been obtained using only 18 Bgl stem-loops (Table 1) [15]. However, it is not clear whether all 18 stem-loops are required to achieve this level of sensitivity.

This system shares many of the advantages and disadvantages of the MS2 RNA labeling system. Bgl-mCherry binds with high affinity to RNAs containing BglG stem-loops, allowing selective labeling of the target RNA [15]. However, like MS2, the multiple, highly repetitive stem loop sequences can be unstable, although this problem can be managed by using recombination-deficient bacterial strains. The Bgl-mCherry system has been used successfully to label HIV-1 RNAs, and viral genomes bound by Bgl-mCherry are incorporated into retrovirus particles with >90% efficiency, suggesting that Bgl-mCherry is unlikely to adversely affect viral RNA trafficking [15].

Due to the highly specific binding of MS2-GFP and Bgl-mCherry to their respective RNA sequences, separately labeled RNAs can be discriminated within the same cell [15,48]. Moreover, because MS2-GFP and Bgl-mCherry are efficiently encapsidated into retrovirus particles, the genomic RNA content of single virions can be compared [15,40,48,49]. The Hu laboratory has used this system to study genome encapsidation by HIV-1. Because retroviral genomic RNAs (gRNAs) are encapsidated as a non-covalent dimer, packaging of two heterologous RNAs can lead to recombination during reverse transcription [50]. When psi-containing HIV-1 RNAs labeled with MS2-GFP or Bgl-mCherry were co-expressed, the genomic RNAs were randomly assorted into virus particles with the expected 1:2:1 ratio [15]. However, mutations within the genomic RNA dimerization initiation sequence disrupt this ratio, suggesting that dimerization occurs prior to encapsidation [15]. Using the same system, the Hu laboratory also demonstrated that HIV-1 genome dimerization appears to occur in the cytoplasm and that the RNA nuclear export pathway influences the assortment of genomic RNAs virus particles [40].

The Bgl-mCherry system has also been used to study the mechanism of encapsidation of genomic RNA by HIV-2 [49] and to investigate cross-encapsidation of HIV-2 RNA by HIV-1 Gag [48]. Other retroviruses, including MMTV and Mason-Pfizer Monkey Virus (MPMV), can also cross-package the genomes of orthologous retroviruses [51] and exogenous retroviruses can recombine with endogenous retroviruses to create novel pathogens [52]. As a result, single particle RNA analysis using MS2-GFP and Bgl-mCherry may provide valuable insight into the frequency and mechanisms of cross-packaging events, and may shed light onto the mechanisms used by different retroviruses to select genomic RNAs for packaging.

One of the difficulties associated with the MS2-GFP and Bgl-mCherry systems is the insertion of long (e.g., ~1,300 nucleotides for 24 MS2 stem-loops), highly repetitive sequences into the RNA of interest. To circumvent this problem, Daigle and Ellenberg recently developed the λ_N-GFP system to study intracellular RNA trafficking in living cells [16]. The λ_N-GFP reporter protein is composed of four tandem repeats of the N-terminal 22 amino acids of the bacteriophage λ N coat protein (λ_N) fused to three eGFP molecules and an M9 localization signal. The λ_N-GFP reporter binds four 15-nucleotide stem-loop structures called BoxB, which are inserted into the target RNA (Table 1) [16]. Each set of four BoxB loops adds only ~80 nucleotides, reducing the amount of exogenous sequence within the study RNA. Like other RNA-binding protein-based tracking systems, λ_N-GFP binds its target RNA with high affinity and specificity; each λ_N repeat binds a single BoxB stem-loop with Kd = 22 nM [53]. As with MS2-GFP and Bgl-mCherry, λ_N-GFP binding could influence the trafficking of the target RNA molecule. However, actin zipcode-containing mRNAs bound by λ_N-GFP are properly localized, providing evidence that λ_N-GFP may not alter the trafficking of mRNAs to which it binds [16].

To visualize MMTV RNA trafficking in living cells, we created a previously unpublished subviral RNA (svRNA) construct containing the MMTV Ψ packaging sequence [54]; the Rem response element, which is bound by the MMTV Rem protein to export full-length viral RNA from the nucleus [55,56,57]; and four BoxB loops (Figure 1A) [16]. When expressed alone in normal mouse mammary epithelial (NMuMG) cells, λ_N-GFP was localized to the nucleus as expected (Figure 1B, left). Co-expression of the svRNA with λ_N-GFP resulted in no detectable cytoplasmic GFP signal (Figure 1B, middle) although co-expression of Rem, svRNA, and λ_N-GFP resulted in export of λ_N-GFP from the nucleus, supporting the finding that Rem is required for the nuclear export of MMTV RNA (Figure 1B, right) [56]. Interestingly, the MMTV RNA accumulated in cytoplasmic granules reminiscent of mRNA processing sites known as P-bodies. To address whether MMTV RNA was associated with P-body proteins, NMuMG cells were transfected with svRNA, Rem, λ_N-GFP, and RFP-Dcp1a, a P-body component [58]. Living cells were imaged on a DeltaVision Elite microscope with a 60X 1.4 NA objective in a 37 °C environmental chamber. A subset of MMTV RNA granules transiently trafficked to P-bodies containing RFP-Dcp1a (Figure 1C). Together, these data demonstrate the feasibility of the λ_N-GFP system for tracking retroviral RNAs in living cells.

One of the major drawbacks to the RNA-binding protein-based systems discussed to this point is that all of these systems require the insertion of specific stem-loop structures into the RNA of interest. On the other hand, although endogenous transcripts can be visualized using the hybridization-based techniques discussed earlier, these techniques are technically challenging. These limitations have resulted in the development of a genetic probe based on the RNA-binding protein PUMILIO1 to track endogenous RNAs in living cells [17]. The PUMILIO1 homology domain (PUM-HD) is composed of eight repeating units, each of which binds a single RNA nucleotide [59,60]. The wild-type PUM-HD recognizes the RNA sequence 5’ UGUA(U/C)AUA, however PUM-HD can be engineered to recognize different sequences by mutating individual units based on a predictable set of rules [61,62].

To visualize endogenous RNA molecules in living cells Ozawa et al. created two PUM-HD mutants that recognized adjacent sequences on the mitochondrial NADH dehydrogenase 6 RNA [17]. The PUM-HD constructs were fused to the N- and C-terminal fragments of a split GFP reporter, so GFP fluorescence was only produced when both PUM-HD constructs were bound to the same RNA (Table 1) [17]. Due to the high signal-to-noise ratio achieved through the split GFP reporter, this system has been used to visualize single, endogenous RNAs in living cells [18]. The PUM-HD system has also been used to study the trafficking of tobacco mosaic virus RNA in living plant cells, and RNA localization appears to be unaffected by PUM-HD binding [63]. As a result, the PUM-HD system may be a powerful technique to study trafficking of RNAs expressed from endogenous retroviruses and during authentic retroviral infection.

As with all RNA tracking systems, however, the PUM-HD system has some limitations. Background fluorescence may result from non-specific complementation of the split GFP fluorophore, reducing the signal-to-noise ratio [63]. Also, because PUM-HD recognizes endogenous RNAs, it has the potential to bind off-target RNAs much like an oligonucleotide probe. However, by recognizing two different eight nucleotide sequences in tandem, the PUM-HD system is theoretically able to distinguish its target from 4.3 × 10⁹ different transcripts [17]. Accordingly, PUM-HD probes should provide the specificity required to recognize the target RNA even in the presence of transcripts with similar sequences. As with hybridization-based approaches, proper controls are necessary to ensure that the PUM-HD probes specifically bind the sequence of interest.