Influenza A virus acutely transcribes viral mRNAs from the eight segmented viral genome when it infects. The kinetics of viral transcription, nuclear export of viral transcripts, and their potential variation between the eight segments are poorly characterised. Here we introduce a statistical framework for estimating the nuclear export rate of each segment from a snapshot of mRNA in situ localisation at a single time point, exploiting the cell-to-cell variation observed by an imaging-based in situ transcriptome assay. A mathematical modelling indicated that the relationship between the nuclear ratio and the total count of mRNAs in single cells is dictated by a proxy for the nuclear export rate. Using this model, we showed that the two influenza viral antigens hemagglutinin and neuraminidase were the slowest segments in the nuclear export, suggesting that influenza A virus uses the nuclear retention of viral transcripts to delay the expression of antigenic molecules. Our framework presented in this study can be widely used for investigating the nuclear retention of nascent transcripts produced in a transcription burst.
This important study combines virology experiments and mathematical modeling to determine the nuclear export rate of each of the eight RNA segments of the influenza A virus, leading to the proposal that a specific retention of mRNA within the nucleus delays the expression of antigenic viral proteins. The proposed model for explaining the differential rate of export is compelling, going beyond the state of the art, but the experimental setup is incomplete and would benefit from additional approaches. The insight so far is interesting, but because in the end it is left as an observation, the overall advance remains limited.
Temporal gene expression is of vital importance for viral pathogens to proliferate in the host. Human exogenous retroviruses insert their RNA genome into the host chromatin and use host factors to minimise the viral protein production to evade the immune response and persist in the host 1. In contrast, influenza A virus acutely infects the respiratory tract and produces thousands of viral progenitors in a single round of infection to spread in the surrounding epithelial cells by transcribing and replicating its RNA genome 2. Influenza A virus genome is segmented into eight single-stranded RNAs, and expression of each gene is essential for productive infection 3. Unlike the human exogenous retroviruses, influenza A virus does not seem to have a sophisticated mechanism of controlling the viral gene expression at the transcription level. However, it has been known for decades that hemagglutinin (HA) and neuraminidase (NA), the two influenza viral antigens targeted by the host immune response, are produced late during the course of infection 4,5,6, a mechanism of which is currently unclear.
Eight viral RNA segments are transcribed by the same set of molecular machinery 7. Viral genome is contained in the viral envelope in association with viral proteins. The 5′ and 3′ ends of each viral genomic RNA hybridise to form an intramolecular loop 8. The RNA-dependent RNA polymerase of influenza A virus, comprising PB1 (polymerase basic 1), PB2 (polymerase basic 2), and PA (polymerase acidic) subunits, is associated with the 5′ and 3′ joint end; the rest of the loop is bound by the oligomeric viral nucleoprotein (NP) 9. The viral ribonucleoprotein complex (vRNPs) initiates the viral transcription in the nucleus by deriving a 5′-capped short RNA fragment from the host nascent RNA on the stalled RNA polymerase II (Pol II) 10. At the end of transcription, the viral polymerase adds a poly-A tail to the viral mRNA by stuttering on a short uridine repeat in the viral RNA template 11. Thus, eight viral segments sharing the common mechanism of viral transcription with no additional layers of transcriptional controls e.g., transcription-factor binding, the transcription of eight influenza viral genes is likely to occur at the same rate across the segments that is determined by the collision of the vRNP and the stalled Pol II.
Another possible factor that would determine the rate of transcription is the length of viral RNA template 12, 13, given that the duration of transcription is rate-limiting. However, this does not agree with the order of protein expression in the influenza virus infection: PB2 (2.4 kb), PB1 (2.3 kb) and PA (2.2 kb), the three longest viral genes, and NP (1.8 kb) are expressed first, followed by the expression of three segments HA (1.8 kb), NA (1.5 kb) and M (1.0 kb) 4, 6.
A third possibility is the nuclear retention of viral mRNAs. Nuclear retention of mRNAs has gained an attention as a point of control on the protein expression in the context of transcription burst 14. Previous studies reported the accumulation of M mRNAs in the nuclear speckles 15, NXF1-dependent nuclear export of HA, NA and M mRNAs 16, 17, and the nuclear export inhibition by the host factor hnRNPAB for M, HA, NP and NS mRNAs 18. It is becoming clear that the influenza viral segments show differential dependencies on the nuclear export factors; and this dependency is not accounted for by the RNA structure (i.e., whether they are intron-less, unspliced or spliced). Because an incomplete set of viral segments was studied in each of these studies, the influenza viral mRNA transport remains elusive.
In this paper, we introduce a statistical framework for systematically quantifying the nuclear transport across all the eight viral segments. We localised eight viral mRNAs simultaneously in single cells in situ. Intracellular distribution of viral mRNAs indicated that the nuclear export rate of viral transcripts varied according to the segments. A statistical model, exploiting cell-to-cell variation in the abundance of transcripts due to the stochastic nature of multiple virological processes prior to the onset of viral transcription, allowed for the estimation of nuclear export rate from the intracellular distribution of mRNAs observed at a single time point. With this model, we show that hemagglutinin and neuraminidase mRNAs were the slowest in exiting the nucleus, suggesting that influenza A virus uses nuclear retention of mRNAs to delay the production of antigenic molecules in the course of viral infection. This study demonstrates that the nuclear export rate of nascent transcripts can be estimated from a snapshot of RNA in situ measurement in single cells, and our result serves as a reference for the nuclear export rate of viral transcripts across the segments, shedding light on the comprehensive understanding on the nuclear export of viral transcripts.
Diffraction-limited, single-colour imaging of viral mRNAs
First, we used single-molecule RNA FISH (smFISH) to study the viral transcription of PB1, HA, NP and NS. The lung carcinoma cell line A549 was infected with the WSN strain of influenza A virus at the multiplicity of infection (M.O.I.) of ∼2 for two hours, and the viral mRNAs were visualised with a tile of short oligonucleotide probes, each labelled with a single fluorophore 19 (Fig. 1A). Viral mRNAs were detected as single diffraction-limited spots in the three-dimensional image stacks, allowing for absolute mRNA quantification (Fig. 1B).
We localised the mRNAs in 3D and counted the total number of mRNAs in each cell (Fig. 1C). The median number of mRNAs per cell was 145, 231, 310 and 303 molecules for PB1, HA, NP and NS, respectively (Fig. 1D), in line with the theoretical prediction that the rate of mRNA production is partly determined by the gene length.
What came intriguingly was the intracellular distribution of mRNAs of the four segments. The intracellular distribution of mRNAs projected on the XY plane (Fig. 1C) indicated that the HA mRNAs were more abundant in the nucleus compared to the other three segments (Fig. 1E).
We wished to quantify the nuclear ratio of mRNAs in each cell. The mRNA count within the 2D nuclear projection is considered to be a good approximation for the number of mRNAs contained in the nucleus 20, 21. We validated this approximation for the lung epithelial A549 cells used in this study by looking at the intracellular Z positions of mRNAs: The mRNAs within the nuclear projection would have higher Z positions over the cell baseline, considering that the nucleus has a diameter of several microns, much thicker than the cytoplasm in the cell periphery. To this end, mRNAs were segmented into three regions in the 2D projection, that is, nuclear, perinuclear and peripheral, depending on the relative location to the nucleus (Fig. 1F). In contrast to the perinuclear (orange) and peripheral (green) mRNAs, mRNAs within the nuclear projection (blue) were distributed around the nuclear focal plane Z = 0 (Fig. 1G); and about 80% of mRNAs in the nuclear projection (blue) were above the cell baseline (Fig. 1H), defined by the median of the Z distribution of peripheral mRNAs (Fig. 1G, red dashed line). Thus, in this paper, we consider the spots within the nuclear 2D projection to be the nuclear fraction for the sake of simplicity.
With this approximation, about 30% of the PB1, NP and NS mRNAs were nuclear, while ∼70% of HA mRNAs were in the nuclear fraction (Fig. 1I), confirming the visual inspection that the HA mRNAs are more abundant in the nucleus at the early time point of infection (Fig. 1E). This result indicates that the velocity of viral mRNA export from the nucleus varies according to the viral segments. Note that our fluorescent probes could also detect complementary RNA (cRNA), an intermediate RNA that is produced for the viral genome replication, as it has the same polarity as viral mRNA. However, the fluorescent signals detected at 2 hours post-infection are deemed to be from the mRNAs, because cRNA production occurs much later than the mRNA synthesis 13, 22.
Mapping eight viral mRNAs in single cells by multiplex RNA FISH
Next, we wish to simultaneously quantify all the eight viral segments in situ. To this end, we applied MERFISH (multiplex error-robust RNA FISH) 23, an imaging-based transcriptome assay in which multiple rounds of probe hybridisation and imaging are performed to identify hundreds to thousands of mRNA species in situ by decoding binary fluorescent signals. For example, hundreds of distinct mRNAs can be encoded in 16-bit binary signals, in which case eight rounds of hybridisation and imaging would be required to read out the signals with two fluorophores 23. In our case of detecting eight influenza A virus segments, 6 bits were sufficient to accommodate all the eight mRNAs, with each code being separated by two or more binary flips from others (Fig. 2A). Thus, only three rounds of hybridisation and two-colour imaging were required. We sought for an economical way of performing sequential rounds of hybridisation and imaging, instead of setting up a custom- built, automated liquid pumping system used elsewhere 23. Inspired by an in vitro reconstitution assay of molecular motors 24, we assembled a flow cell on a slide glass, comprising two stripes of double-sided tape and the coverslip, with cells on the coverslip facing inwards (Fig. 2A). Liquids were introduced from the right open side and wicked from the left. During image acquisition, the two open sides were sealed with rubber cement. This easy implementation of flow cell allowed us to comfortably perform three rounds of hybridisation and two-colour imaging to read out the 6-bit binary tags; and by decoding the 6-bit encodings, the eight viral segments were identified in each cell (Fig. 2B). On average, about 100-200 molecules of each segment were contained per cell (Fig. 2C, top row). The difference in the total number of spots from the result presented in Fig. 1D is likely due to the batch effect (experimental variation).
The nuclear fraction of mRNAs varied between the viral segments (Fig. 2C, bottom row): While ∼30% of mRNAs were nuclear for PB1, NP and NS segments, about 70% of mRNAs remained in the nucleus for HA and NA segments; for the other segments (PB2, PA and M), around half of mRNAs were in the nucleus. This reiterates the idea that the velocity of viral mRNA export from the nucleus varies according to the viral segments. These results were confirmed by smFISH using probes for each segment that were synthesised economically in-house (Supplementary figure 1A, B and C).
Single-cell clustering according to the abundance of eight viral segments revealed that the cell population was heterogeneous in terms of the abundance of viral mRNAs (Fig. 2D). A pair-wise analysis of the abundance of viral segments indicated inequal distribution of viral mRNAs, regardless of the segments (Fig. 2E): Cells that carry abundant mRNAs from one segment tend to also carry abundant mRNAs from the other segments. These observations suggest that the viral transcription is susceptible to the stochastic variation in each infection 25, 26, which we subsequently used in our statistical model to estimate the nuclear export rate.
Estimating the nuclear export rate using population statistics
We suspected that the higher nuclear ratio of HA and NA mRNAs is due to the slower nuclear export rate for these mRNAs compared to other segments such as NP and NS. Thus, we wished to quantify the nuclear export rate of eight viral mRNA species from the observed nuclear-to-cytoplasmic distribution. To this end, we conceptualised a statistical model using the cell-to-cell variation in the abundance of viral transcripts of each segment (Fig. 3A). In this model, influenza viral particles are added to the cell population at time 0 h, and are allowed to undergo viral attachment to the cell surface, membrane fusion, and vRNP migration to the nucleus. The waiting time for these processes vary in each infection due to the stochastic nature of biochemical events 25, 27, and therefore, the onset of viral transcription varies from cell to cell in the population. In other words, each cell will have a varying duration for viral transcription by the time the observation is made at time 2 h (Fig. 3A, top). As a result, the population contains front runners in which the viral transcription was initiated early, and therefore abundant viral transcripts are carried; slow starters in which the transcription began late, containing fewer transcripts; and many other cells that lie along the spectrum between the two. The cell population at the single point in time, therefore, forms a trajectory of the time development of viral transcription and subsequent nuclear export on the scale of nuclear fraction against the abundance of transcripts (Fig. 3A, bottom): In a spontaneous transcription burst, the nascent transcripts emerge in the nucleus at the early stage of the burst, and they are exported to the cytoplasm as more transcripts are being synthesised 20.
Indeed, when the nuclear ratio of viral mRNA was plotted against the total quantity of mRNA molecules observed by multiplex RNA FISH in each cell, the plots formed a curve that originated from top left (i.e., nearly 100% of the nuclear ratio when a cell contained a few mRNA molecules), and descended as the number of mRNA molecules increased (30% to 70%, depending on the segments) (Fig. 3B and Supplementary figure 2).
We further validated this concept by performing smFISH measurements along the time-course with 40-min interval (Fig. 3C). A549 was infected with WSN at the M.O.I of ∼0.02. At each time point, the NP mRNAs were stained and quantified. At the beginning (0 and 40 min), the mRNA count was almost zero. As the mRNA count increased (80 min to 200 min post-infection) (Fig. 3C, top), the nuclear ratio against the total mRNA count derived from the single cells progressed along a curve, from the top left to the bottom right (Fig. 3C, bottom), supporting the concept of our statistical model.
The curvature of the trace appeared to vary between segments (Fig. 3B). We explored to see if the curvature could represent a proxy for the nuclear export rate. Thus, we formulated a simple kinetic model in which mRNAs are synthesised at the rate k in the nucleus, and are exported with the reaction constant ε (Fig. 4A). In this model, the molar concentrations of mRNAs in the nuclear and cytoplasmic compartments are described, respectively,
Omitting the degradation of mRNAs, the analytical solutions of these equations were:
Thus, the mRNA nuclear fraction φ was described as a function of total mRNA molecules Ntotal,
The nuclear fraction φ described in equation (5) is time-independent, and the parameter α determines the curvature of the line that the single cells would trace in the time-course (Fig. 4B). Thus, we estimated the parameter α by fitting to the observed data points derived at the time in which cells were fixed (Fig. 4C). By obtaining the synthesis rate k from the total number of mRNAs produced by the time of fixation (Fig. 2C, top row), the nuclear export rate ε for each segment was derived according to equation (6) (Fig. 4D). The nuclear export rate ε agreed with the historical classification of the early and late influenza A virus genes 4, 6, indicating that the timing of influenza A virus protein expression in the initial stage of infection is determined by the velocity of the nuclear export. Notably, the HA and NA mRNAs had the slowest nuclear export rate, suggesting that influenza A virus delays the viral antigen production by nuclear retention of viral transcripts.
Based on the stochastic nature of influenza viral transcription studied by single-cell sequencing 25, 26, we introduced a quantitative framework for estimating the nuclear export rate of viral mRNAs to advanced our current understanding on the kinetics of viral gene expression. This was achieved by in situ localisation of eight influenza A virus segments simultaneously, coupled with the statistical model that allowed for estimating the nuclear export rate from the intracellular distribution of mRNAs at a single time point. This is based on the concept that the cells in the population are at various stages of transcription at the time the observation is made, due to the multiple stochastic processes the virus needs to overcome before it begins transcription. We demonstrated that the initiation of each viral transcription is concurrent in the eight viral segments; it is instead the nuclear export rate that controls the temporal protein expression at the initial phase of infection. The early production of the polymerase subunits (PB2, PB1 and PA) and nucleoprotein would facilitate the formation of vRNP, and hence the viral transcription and genome replication. On the other hand, pushing back the expression of viral antigens HA and NA would be beneficial for the virus to delay the host immune response against the infected cells in which the virus is being replicated.
Instead of time-series measurements, the estimation of mRNA export rate was made from a snapshot of in situ RNA localisation data taken at a single time point. We anticipate that the statistical framework for quantifying the nuclear export rate proposed in our study can be widely used and benefits the research area on the regulation of gene expression, especially since nuclear retention has been recognised as a point of control against the transcription burst 14. For example, it is applicable for the cell population in which de novo synthesis of mRNAs occurs concurrently, such as a stimuli-induced activation of immediate-early genes 21, 28. On the contrary, this approach is not appropriate for scenarios in which asynchronous transcription bursts are taking place randomly in each single cell 20, 29; cells at the late stage of transcription burst, having residual amount of transcripts in the cytoplasm 20, would obscure the time-course trajectory of the nuclear export.
The experiment in this study was conducted at the early stage of infection (up to 200 min post-infection). This conferred two advantages over investigating influenza viral mRNA export at the late stage of infection (e.g., 6-12 hours post- infection). First, our study allowed for the absolute quantification of viral transcripts at the single-molecule level in each cell, and thereby enabled us to formulate the model and fit the equation to the numerical data points. Otherwise at the late stage of infection, accurate quantification of mRNAs by the imaging- based single-molecule detection would not be feasible due to the high density of mRNAs. Second, at the late stage of infection, NS1 protein would be expressed in the cytoplasm. NS1 protein would then migrate into the nucleus and alter the nuclear export of viral mRNAs 30, obscuring the inherent nature of nuclear export of each viral segment. Thus, in our study, we elucidated the quantitative measure for the influenza viral mRNA nuclear export that is intrinsic to each viral segment.
Host factors that relay the influenza viral mRNAs to the nuclear pore remain elusive. NXF1 and CRM are two major cellular mRNA export pathways for the 5′-capped mRNAs 31. Of these two pathways, influenza A virus mRNAs, also carrying 5′ cap structure derived from the host nascent mRNA and the poly-A tail as a result of viral polymerase stuttering the uridine repeat, are dependent on the NXF1 pathway 16, 17. However, the eight viral segments show differential dependencies on this export pathway. The segments HA, M and NS is exported by NXF1 and UAP56, whereas PB2, PB1 and PA appear to be independent of the NXF1 pathway 17. The nuclear export rate derived in our study does not agree with the dependencies on NXF1. According to the previous study, both HA and NS are dependent on NXF1 and UAP56 16; however, in our study, HA and NS had significantly different nuclear export rate.
Two viral segments M and NS are subject to splicing, yielding unspliced and spliced transcripts. Our current system lacks the ability of discriminating between the unspliced and spliced forms of these segments; although we infer that the majority of NS segments that we detected in this study were in the unspliced form because we obtained similar result with probes exclusively detecting the unspliced form of NS (Supplementary figure 3). An observation that the unspliced form of M and NS dominates throughout the infection supports this 13. In either case, the RNA structure does not appear to accord with the NXF1 dependency, nor the nuclear export rate, either.
Our study suggests that the nuclear export of influenza viral transcripts is intrinsic to the nucleotide sequence of each segment. Recently, the human nuclear factor hnRNPAB was shown to associate with the nucleoprotein of influenza A virus, and inhibit the NXF1 pathway. It is tempting to speculate that the host factors (e.g., hnRNPAB) have differential binding affinities to each segment, thus determining the velocity of nuclear export (or the degree of nuclear export inhibition). It is currently unknown what factors binds to the early genes such as PB2, PB1 and PA. Further studies will be required to identify the host factors that bind to these transcripts for the nuclear export. Our study serves as a reference for the nuclear export rate of each segment, and thus sheds light to comprehensive understanding on the nuclear export of viral segments. Having revealed the nuclear retention of influenza viral transcripts at the unprecedented resolution, this study opens up a new research direction on how the nuclear retention is regulated and its implication in viral pathogenesis.
Materials and Methods
Cells and virus preparation
The human lung carcinoma cell line A549 was maintained in Dulbecco’s Modified Eagle’s Medium (DMEM) (Sigma, D5796) supplemented with 10% FBS (Nichirei Biosciences, 175012), 100 U/ml penicillin and 100 µg/ml streptomycin (Gibco, 15140-122) at 37°C in 5% CO2. Madin-Darby canine kidney (MDCK) cell line was maintained in Minimum Essential Medium (MEM) (Sigma, M4655) supplemented with 10% FBS, 100 U/ml penicillin and 100 µg/ml streptomycin at 37°C in 5% CO2. Influenza A virus strain WSN was propagated in MDCK cells in a 75 cm2 flask. The plaque-forming unit of the harvested virus was determined using MDCK cells grown confluent in a 6-well plate.
An 18-mm round coverslip (0.17±0.005 mm thickness, Matsunami Glass) was used. For MERFISH, two straight sides were created by cutting two edges off the glass using a laser-wheel glass cutter (Nikken Dia, NC-X03 LASER 110°). The coverslip was rinsed in 70% ethanol and placed in a 12-well plate. A549 cells were seeded and attached onto the coverslip overnight. On the following day cells were rinsed twice with PBS and infected with WSN at the indicated multiplicity of infection in MEM for 2 hours (unless indicated otherwise in Fig. 3C) in the absence of FBS. After the infection, cells were fixed with 4% paraformaldehyde (Electron Microscopy Sciences, 15714-S) in PBS for 10 min and then rinsed three times with PBS. The coverslip was immersed in 70-75% ethanol and stored at −20°C until use.
The probes used in Figs. 1, 3C, Supplementary figure 1 (probes for NP) and Supplementary figure 3 were obtained from LGC Biosearch Technologies. Stellaris RNA FISH Probe Designer (https://www. biosearchtech.com/stellaris-designer) was used to design the probes against the coding region of WSN mRNAs.
The coverslip was equilibrated in 1 ml of Wash buffer A (Stellaris) for >5 min in a 12-well plate prior to hybridisation. Cells were hybridised to 125 nM probes in Hybridization buffer (Stellaris) on a Hybrislip (Sigma, GBL712222) at 37°C overnight in a humidified chamber. On the following day the coverslip was rinsed twice in 1 ml of Wash buffer A (Stellaris), each for 30 min. Cytoplasmic stain (HCS Cellmask Green Stain, Invitrogen, H32714) was included in the second wash at the concentration of 50 ng/ml in Figs 1, 3C and Supplementary figure 3. Then the coverslip was rinsed in 1 ml of Wash buffer B (Stellaris) for 5 min. The coverslip was mounted in anti-fade medium containing DAPI (Vector Laboratories, H-1200) on a slide glass and sealed with nail polish.
For an economical reason and to allow for more flexible design of probes to increase the number of fluorophores attached per mRNA molecule, the probes used in Supplementary figure 1 (except the NP probes) were synthesised in-house (Supplementary figure 1A). (The NP probe used in Supplementary figure 1 was purchased from LGC Biosearch.) Candidate probe-binding sites in the coding region of each viral segment were identified using OligoMiner 32, and were further selected against the human RefSeq. The sequences were flanked by P5 (5′ AAT GAT ACG GCG ACC ACC GA 3′) and P7 (5′ CAA GCA GAA GAC GGC ATA CGA GAT 3′), with P7 at the 5′ end and P5 at the 3′ end of the probe sequence (Supplementary figure 1A). For the probes against HA, NA, M and NS, additional 20-nt sequences RS0332 (5′ GGG AGA ATG AGG TGT AAT GT 3′) and RS0406 (5′ GAT GAT GTA GTA GTA AGG GT 3′) 23 were inserted between P7 and the probe, the probe and P5, respectively (Supplementary figure 1A). These two insertions were used for additional probes to bind to increase the number of fluorophores per molecule. An oligonucleotide pool comprising the set of probe-binding sites with these additional nucleotide sequences was synthesised at Integrated DNA Technologies (IDT). The oligonucleotide pool was subjected to PCR amplification (98°C for 30 sec; 18 cycles of 98°C for 5 sec, 64°C for 10 sec and 72°C for 20 sec; and 72°C for 5 min) (Phusion High-Fidelity DNA Polymerase, NEB, M0530S) using the P7 and a T7-fused P5 primer (5′ TAA TAC GAC TCA CTA TAG GGA ATG ATA CGG CGA CCA CCG A 3′) (Supplementary figure 1A). RNA was synthesised using T7 RNA polymerase (HiScribe T7 High Yield RNA Synthesis Kit, NEB, E2040S). Then RNA was reverse- transcribed using a recombinant Moloney leukemia virus reverse transcriptase (GeneAce cDNA Synthesis Kit, Nippon Gene, 319-08881) with an ATTO550- conjugated P7 primer at the 5′ end. The single-stranded DNA probe was purified using a spin column (Zymo Research, D7010) after RNA digestion with a cocktail of RNase H and RNase A. The probe was further purified by Urea-PAGE to remove any residual RNA and the fluorescent-labelled primer.
The coverslip was equilibrated in 1 ml of 30% formamide buffer composed of 2×SSC and 30% deionised formamide (Sigma, F9037) in a 12-well plate for >5 min prior to hybridisation. Cells were hybridised with the probes at the concentration of 5.6 nM per probe in 100 µl Hybridization buffer composed of 2×SSC, 30% deionised formamide, 0.1% yeast tRNA (Invitrogen, 15401-011), 1% (v/v) RNase inhibitor (NEB, M0314S) and 10% (w/v) Dextran Sulfate (Merck, S4030) 23 on a Hybrislip at 37°C overnight in a humidified chamber. On the following day the coverslip was rinsed twice in 1 ml of 30% formamide buffer, each for 30 min. DAPI was included at 5 ng/ml in the second wash if additional staining was not performed with the external probes (for PB2, PB1 and PA). In this case the coverslip was rinsed in 1 ml of PBS after the DAPI staining, mounted in anti-fade medium (Vector Laboratories, H-1900) on a slide glass and sealed with nail polish. For staining HA, NA, M and NS with the extra probes, the coverslip was first equilibrated with 10% formamide buffer composed of 2×SSC, 10% deionised formamide and 0.05% (v/v) RNase inhibitor, and then hybridised with 10 nM each of probes for RS0332 and RS0406 23, both labelled with Quasar 570 (LGC Biosearch) at the 3′ end, in Stellaris hybridisation buffer (containing 10% formamide) at 37°C in the humidified chamber for 1 hour. Cells were washed in 10% formamide buffer twice, with DAPI included in the second wash. The coverslip was rinsed in PBS, mounted in the anti-fade medium (Vector Laboratories, H-1900), and sealed with nail polish.
Sequential hybridisation and imaging in the flow cell
An oligonucleotide pool for synthesising the encoding probes used in Fig. 2, comprising the target-binding sequence, readout tags compatible with the two-step amplification probes 23, 33, and the P5 and P7 PCR tags, were obtained from IDT. The target-binding sites were identified in the coding region of each segment using OligoMiner 32 and selected against human RefSeq. The readout tags RS0332, RS0343, RS0406, RS0255, RS0015 and RS038 23, 33 were inserted adjacent to the upstream or downstream of the target-binding sites for the 6-bit encoding of each segment (Fig. 2A). Probes were synthesised by PCR, T7 transcription and reverse-transcription as described above.
A coverslip was equilibrated in 1 ml of 30% formamide buffer (2×SSC and 30% deionised formamide) 23. Cells were hybridised with 1.6 µM encoding probes in Hybridisation buffer composed of 2×SSC, 30% deionised formamide, 0.1% yeast tRNA, 1% RNase inhibitor, 10% dextran sulfate 23 on a Hybrislip at 37°C overnight in a humidified chamber. Cells were washed twice in 1 ml of 30% formamide buffer at 37°C, and then equilibrated in 1 ml of Readout-probe hybridisation and wash buffer composed of 2×SSC, 10% deionised formamide and 20 units/ml Murine RNase inhibitor 23. The first and second amplification probes for RS0332, RS0255, RS0343, RS0406, RS0015 and RS0384, described previously 33, were annealed sequentially at 10 nM of each probe in Readout-probe hybridisation and wash buffer for 2 hours at 37°C on a Hybrislip. The coverslip was washed for 30 min twice in Readout-probe hybridisation and wash buffer at each of the first and second probe hybridisation. At the final wash, 0.10 µm green-fluorescent carboxyl polystyrene beads (Bangs Laboratories, FCDG002) were included (∼0.1% v/v) as a fiducial marker for image registration. The cells and beads were fixed with 3.2% paraformaldehyde in PBS for 10 min. The coverslip was rinsed in PBS three times and stored in 2×SSC supplemented with 40 units/ml RNase inhibitor at 4°C overnight.
On the following day a flow cell was assembled on a slide glass, composed of two parallel stripes of double-sided tape (0.23 mm thickness, ∼5 mm apart) and the coverslip on top, with cells facing inwards (Fig. 2A). The cells at the top and bottom end of the coverslip were scraped using a cell scraper before the assembly and any trace amount of liquid remaining on the exposed surface was removed with filter paper to ensure that the coverslip is firmly attached to the tape. The chamber was immediately flushed with 2×SSC to avoid cells drying up.
Readout-probe hybridisation and wash buffer was flushed in the chamber to equilibrate the cells. Readout probes were introduced at 10 nM in Readout-probe hybridisation and wash buffer and incubated at 37°C in a humidified chamber for 30 min. Subsequently, Readout-probe hybridisation and wash buffer was flushed to wash away the probes. Imaging buffer, composed of 2×SSC, 50 mM Tris-HCl pH 8, 10% glucose (Fujifilm Wako, 049-31165), 2 mM Trolox (Sigma, 238813), 0.5 mg/ml glucose oxidase (Sigma, G2133), 40 µg/ml catalase (Sigma, C30) and 0.1% RNase inhibitor 23, was flushed to the flow cell prior to the image acquisition. The flow cell was sealed at the both sides with Fixogum rubber cement ejected through a 200-µl pipette tip during the image acquisition.
After the imaging, the disulfide bond linking the probe and the fluorophore was cleaved with Cleavage buffer comprising 2×SSC and 50 mM TCEP (Sigma, 646547) for 15 min. The chamber was flushed with Readout-probe hybridisation and wash buffer to wash away the free fluorophores prior to introducing the readout probes for the next round. Readout probes used in this study are listed in Table 1.
Images were acquired using a Nikon Ti2-E with a Nikon Plan Apo Lambda 60x (NA 1.40) (Fig. 1, Supplementary figure 3) or 100x (NA 1.45) (Fig. 3C) objective lens equipped with a Prime 95B 25MM back-illuminated sCMOS camera (Photometrics). Otherwise, an Olympus IX83 equipped with an Olympus UPlan XApo 60x (NA 1.42) objective lens and a cooled monochrome CCD camera (DP80, Olympus) was used (Fig. 2 and Supplementary figure 1).
Fluorescent spots detected by smFISH were quantified using either FISH-QUANT 34 (Figs. 1, 3C and Supplementary figure 3) or a more recent Python package Big-FISH 35 (Supplementary figure 1). Cytoplasmic and nuclear segmentation was performed using Cellpose 36, 37. Objects that were on the boundary of the image was eliminated. Cytoplasmic regions with no nucleus therein due to the erroneous segmentation was also eliminated. The identification numbers assigned to each object by Cellpose were rearranged from scratch so that the cytoplasm and nucleus of the same cell correspond for the downstream analyses. In order to make the Cellpose mask readable in FISH-QUANT, the Cellpose mask were converted into x and y coordinates by tracing each object in the mask using the Moore neighbourhood edge-finding algorithm.
For MERFISH signal decoding, signals from the green-fluorescent carboxyl beads were used to register the images. Images at the corresponding field-of-view from the three rounds of imaging were registered using phase_cross_correlation from the Python package skimage.registration. The registered images were filtered using Laplacian-of-Gaussian implemented in Big-FISH, and the image intensity was manually normalised. Fluorescent signals were then decoded using PixelSpotDecoder in DetectPixels implemented in starfish (https://github.com/spacetx/starfish").
Data analyses and presentation
Microscopy images were presented using the Python package microfilm (https://github.com/guiwitz/microfilm) (Fig. 1B). Cell clustering was performed with the method centroid implemented in scipy.cluster.hierarchy.linkage from SciPy (https://scipy.org/) (Fig. 2D). The parameter α in equation (5) was estimated using curve_fit implemented in scipy.optimize from SciPy (Fig. 4C).
We thank Imaging and Flow Cytometry Core, The Centre for PanorOmic Sciences (CPOS), LKS Faculty of Medicine of The University of Hong Kong for the use of Nikon Ti2-E widefield imaging system; and Central Research Institute of Kawasaki Medical School for the use of Olympus IX83 imaging system. This study was supported by Research Project Grant R04S-003 and R03S-001 from Kawasaki Medical School, The KAWASAKI Foundation of Medical Science and Medical Welfare, and Grant-in-Aid for Scientific Research (C) (22K07091) from Japan Society for the Promotion of Science (JSPS) to MM; Grant-in-Aid for Scientific Research (B) (21H03188) from JSPS to TN; the General Research Fund (17107019) of the Research Grants Council, Hong Kong SAR to HC.
Michi Miura: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data Curation, Writing - Original Draft, Writing - Review & Editing, Visualization, Supervision, Project administration, Funding acquisition. Naho Kiuchi: Software, Validation, Investigation, Writing - Review & Editing. Siu- Ying Lau: Resources, Writing - Review & Editing. Bobo Wing-Yee Mok: Resources, Writing - Review & Editing. Hiroshi Ushirogawa: Resources, Writing - Review & Editing. Tadasuke Naito: Resources, Writing - Review & Editing, Funding acquisition. Honglin Chen: Resources, Data Curation, Writing - Review & Editing, Supervision, Project administration, Funding acquisition. Mineki Saito: Resources, Data Curation, Writing - Review & Editing, Supervision, Project administration, Funding acquisition.
Declaration of interests
The authors declare no competing interests.
Data and code availability
The probe sequences, original images, and the custom codes used in this study are available upon request.
- 1.Regulation of Latency in the Human T Cell Leukemia Virus, HTLV-1Annu Rev Virol 6:365–385
- 2.SnapShot: Influenza by the NumbersCell 182:532–532
- 3.Influenza A Virus Cell Entry, Replication, Virion Assembly and MovementFront Immunol 9
- 4.The gene structure and replication of influenza virusAnnu Rev Biochem 52:467–506
- 5.Influenza virus gene expression: control mechanisms at early and late times of infection and nuclear-cytoplasmic transport of virus-specific RNAsJ Virol 61:764–773
- 6.Control of influenza virus gene expression: quantitative analysis of each viral RNA species in infected cellsJ Biochem 105:537–546
- 7.Influenza virus RNA polymerase: insights into the mechanisms of viral RNA synthesisNat Rev Microbiol 14:479–493
- 8.Characterization of the RNA-fork model of virion RNA in the initiation of transcription in influenza A virusJ Virol 69:4012–4019
- 9.The structure of native influenza virion ribonucleoproteinsScience 338:1634–1637
- 10.Interplay between Influenza Virus and the Host RNA Polymerase II Transcriptional MachineryTrends Microbiol 27:398–407
- 11.Direct evidence that the poly(A) tail of influenza A virus mRNA is synthesized by reiterative copying of a U track in the virion RNA templateJ Virol 73:3473–3476
- 12.Transcription and replication of eight RNA segments of influenza virusVirology 142:68–77
- 13.Segment-specific kinetics of mRNA, cRNA and vRNA accumulation during influenza infectionJ Virol 95
- 14.Nuclear Retention of mRNA in Mammalian TissuesCell Rep 13:2653–2662
- 15.Influenza virus mRNA trafficking through host nuclear specklesNat Microbiol 1
- 16.Individual influenza A virus mRNAs show differential dependence on cellular NXF1/TAP for their nuclear exportJ Gen Virol 91:1290–1301
- 17.Influenza polymerase encoding mRNAs utilize atypical mRNA nuclear exportVirol J 11
- 18.Cellular hnRNPAB binding to viral nucleoprotein inhibits flu virus replication by blocking nuclear export of viral mRNAiScience 24
- 19.Imaging individual mRNA molecules using multiple singly labeled probesNat Methods 5:877–879
- 20.Single-cell heterogeneity and cell-cycle- related viral gene bursts in the human leukaemia virus HTLV-1Wellcome Open Res 2
- 21.Kinetics of HTLV-1 reactivation from latency quantified by single-molecule RNA FISH and stochastic modellingPLoS Pathog 15
- 22.Strand-specific real-time RT-PCR for distinguishing influenza vRNA, cRNA, and mRNAJ Virol Methods 173:1–6
- 23.High-throughput single-cell gene-expression profiling with multiplexed error-robust fluorescence in situ hybridizationProc Natl Acad Sci U S A 113:11046–11051
- 24.Nucleotide- dependent behavior of single molecules of cytoplasmic dynein on microtubules in vitroFEBS Lett 584:2351–2355
- 25.Single-cell analysis and stochastic modelling unveil large cell-to-cell variability in influenza A virus infectionNat Commun 6
- 26.Extreme heterogeneity of influenza virus infection in single cellsElife 7
- 27.Modeling the intracellular dynamics of influenza virus replication to understand the control of viral RNA synthesisJ Virol 86:7806–7817
- 28.Histone H2A monoubiquitylation and p38-MAPKs regulate immediate-early gene-like reactivation of latent retrovirus HTLV-1JCI Insight 3
- 29.Time- course of host cell transcription during the HTLV-1 transcriptional burstPLoS Pathog 18
- 30.Influenza A Virus NS1 Protein Promotes Efficient Nuclear Export of Unspliced Viral M1 mRNAJ Virol 91
- 31.Sequence Determinants for Nuclear Retention and Cytoplasmic Export of mRNAs and lncRNAsFront Genet 9
- 32.OligoMiner provides a rapid, flexible environment for the design of genome-scale oligonucleotide in situ hybridization probesProc Natl Acad Sci U S A 115:E2183–e2192
- 33.Multiplexed detection of RNA using MERFISH and branched DNA amplificationSci Rep 9
- 34.FISH-quant: automatic counting of transcripts in 3D FISH imagesNat Methods 10:277–278
- 35.FISH-quant v2: a scalable and modular tool for smFISH image analysisRNA 28:786–795
- 36.Cellpose: a generalist algorithm for cellular segmentationNat Methods 18:100–106
- 37.Cellpose 2.0: how to train your own modelNat Methods 19:1634–1641