Multiplexed sequential imaging in living cells with orthogonal fluorogenic RNA aptamer/dye pairs

Abstract

Detecting multiple targets in living cells is important in cell biology. However, multiplexed fluorescence imaging beyond two-to-three targets remains a technical challenge. Herein, we introduce a multiplexed imaging strategy, ‘sequential Fluorogenic RNA Imaging-Enabled Sensor’ (seqFRIES), which enables live-cell target detection via sequential rounds of imaging-and-stripping. In seqFRIES, multiple orthogonal fluorogenic RNA aptamers are genetically encoded inside cells, and then the corresponding cell membrane permeable dye molecules are added, imaged, and rapidly removed in consecutive detection cycles. As a proof-of-concept, we have identified in this study four fluorogenic RNA aptamer/dye pairs that can be used for highly orthogonal and multiplexed imaging in living bacterial and mammalian cells. After further optimizing the cellular fluorescence activation and deactivation kinetics of these RNA/dye pairs, the whole four-color semi-quantitative seqFRIES process can be completed in ∼20 min. Meanwhile, seqFRIES-mediated simultaneous detection of critical signalling molecules and mRNA targets was also achieved within individual living cells. We expect our validation of this new seqFRIES concept here will facilitate the further development and potential broad usage of these orthogonal fluorogenic RNA/dye pairs for multiplexed and dynamic live-cell imaging and cell biology studies.

Graphical Abstract

Graphical Abstract.

Introduction

Cellular processes are often regulated by a group of interactive biomolecules. The distributions and dynamic correlations of these molecules play important roles in determining cell organization and the efficiency of cell signaling network (1–3). The ability to detect multiple targets, especially in situ inside living cells, is critical for our understanding of the molecular mechanisms and distinct signatures of different healthy and diseased cells. Fluorescence microscopy remains one of the most prominent tools for the live-cell detection of target analytes. However, the broad fluorescence spectrum and large spectral overlap among common fluorophores have dramatically hindered their capability to simultaneously detect multiple targets (4–6). Live-cell fluorescence imaging beyond two or three targets remains a common technical challenge. Although many efforts have been made towards multiplexed cellular detection, especially with recent development of vibrational (7–9) and spectral imaging techniques (10,11), as well as far-red fluorescent probes (12), simple and broadly applicable techniques are still needed that can image multiple biomolecules in living systems just based on regular fluorescence microscopes available in a typical life science laboratory.

In fixed and membrane-permeabilized cell and tissue samples, the problem of fluorescence spectral overlap has been elegantly solved via sequential imaging approaches, such as MERFISH, seqFISH, etc. (13–15). In these techniques, multi-target detection is achieved via dye-labeled DNA- or antibody-based fluorescent probes. Instead of imaging all the probes at a time, multiple rounds of labeling-imaging-and-stripping processes are conducted to allow highly multiplexed target detection. However, due to the requirement of cell permeabilization and fixation for the delivery and removal of these DNA/antibody-based fluorescent probes, current sequential multiplexing techniques are not capable for live-cell imaging or multiplexed molecular profiling in their native cellular context.

Herein, we report a sequential imaging platform that can be used for live-cell multiplexed detection, named sequential Fluorogenic RNA Imaging-Enabled Sensor (seqFRIES). Fluorogenic RNAs (FR) are single-stranded RNA aptamers that can specifically bind with otherwise non-fluorescent small-molecule dyes and activate their fluorescence signals (16). One unique feature of the FR/dye system is that the binding between RNA aptamers and dyes is non-covalent and highly reversible (Figure 1A). As a result, while the FR aptamers can be genetically encoded inside the cells, the corresponding membrane-permeable dyes can be externally synthesized, added to the cells for imaging, and then easily washed away to turn off fluorescence signals. Sequential multiplexed detection in living cells can thus be possibly achieved via rounds of rapid imaging-and-stripping processes using this non-covalent fluorogenic RNA/dye-based system (Figure 1B).

Figure 1.

Working principle and orthogonality of fluorogenic RNA (FR) and seqFRIES. (A) Upon specific reversible binding, orthogonal FR/dye pairs can turn on their fluorescence signals independent from each other. (B) After genetically encoding orthogonal FRs in living cells, sequential multiplexed detection can be achieved via rounds of imaging-and-stripping process by adding and removing the cognate membrane-permeable small-molecule dyes. (C) In vitro characterization of five identified orthogonal FR/dye pairs. More experimental details are shown in Supplementary Figure S1. Both the number and brightness of each block indicate the intensity of mean fluorescence signals as measured from at least three independent replicates. (D) Fluorescence images in BL21 Star (DE3) cells that express Broccoli, DNB, Corn, or Pepper RNA after a 30-min incubation with 200 μM DFHBI-1T, 1 μM TMR-DN, 10 μM DFHO or 1 μM HBC620. Scale bar, 5 μm. (E) Cellular fluorescence intensities as quantified in ∼50 individual cells in each case. Shown are the mean and standard deviation (SD) values from cellular images taken from at least three independent replicates.

The potential multiplexing capability of seqFRIES is determined by the number of FR/dye pairs that can function independently; while the overall imaging efficiency and throughput will be controlled by the time required for each round of labeling, imaging, and stripping. It is thus critical to identify orthogonal FR/dye pairs with fast cellular fluorescence activation/deactivation kinetics. In this current work, we identified the largest set of orthogonal fluorogenic RNA/dye pairs reported so far, including four FR/dye pairs that can function independently and be used for multiplexed sequential imaging in living bacterial and mammalian cells. We have further optimized the cellular fluorescence turn-on and turn-off rate for each of these FR/dye pairs. Under optimized imaging-and-stripping conditions, live-cell semi-quantitative four-color target detection can be achieved within ∼20 min.

In addition, these FR aptamers can be modularly engineered into sensors for the cellular recognition and detection of different target analytes (17–19). Using signaling molecules and mRNAs as sample targets, our results have further demonstrated the potential usage of seqFRIES for studying cellular correlations of different biomolecules in single living cells. We expect this new seqFRIES platform and the concept of live-cell sequential imaging will open the door for advanced applications in the field of live-cell multiplexed imaging and single-cell molecular profiling, all of which can be performed using conventional fluorescence microscopes.

Materials and methods

Reagents

All the chemicals were purchased from MilliporeSigma or Fisher Scientific unless indicated otherwise. Commercial reagents were directly used without additional purification. 4-(2-hydroxyethyl-methylamino)-benzylidene-cyanophenylacetonitrile analog, HBC620, was purchased from GlpBio (Montclair, CA). YO3-biotin, an oxazole yellow derivative, was purchased from Applied Biological Materials Inc (Vancouver, Canada). S-(5′-Adenosyl)-L-methionine chloride was purchased from Cayman Chemical (Ann Arbor, MI). 3,5-difluoro-4-hydroxybenzylidene imidazolinone-2-oxime (DFHO) was a gift from Samie R. Jaffrey's lab at Weill Cornell Medicine.

Synthesis of oligonucleotides

DNA oligonucleotides were synthesized and cartridge-purified by Integrated DNA Technologies (Coralville, IA) or W. M. Keck Oligonucleotide Synthesis Facility (Yale University School of Medicine). Oligonucleotides were then dissolved at 100 μM concentration in TE buffer (10 mM Tris–HCl, 0.1 mM EDTA, pH 7.5) and stored at –20°C. Double-stranded DNAs were made by PCR amplification using an Eppendorf Mastercycler. The PCR products were then cleaned using a QIAquick PCR purification kit (Qiagen, Germantown, MD). Concentration measurement of nucleic acids was performed using a NanoDrop One UV-vis spectrophotometer. RNA synthesis was conducted via in vitro transcription using a HiScribe™ T7 high yield RNA synthesis kit (New England BioLabs, NEB, Ipswich, MA). After DNase I (RNase-free) (NEB) treatment to remove the DNA strands, the transcribed RNAs were column-purified and further verified by running 10% denaturing polyacrylamide gel electrophoresis. All the synthesized RNA products were separated into aliquots and stored at –20°C for immediate usage or at –80°C for long-time storage. NUPACK and mFold online software were used to simulate and design all the RNA structures.

Synthesis of tetramethylrhodamine-PEG3- dinitroaniline (TMR-DN)

Tetramethylrhodamine-PEG3-dinitroaniline (TMR-DN) was synthesized following a previously reported procedure (20): dinitroaniline-PEG3-amine (DN-PEG3-amine) was first synthesized by adding a 10 ml dichloromethane (DCM) solution of dinitrofluorobenzene (1.0 g, 5.4 mmol) dropwise into a solution of 2,2′-(ethylenedioxy)bisethylenediamine (4.67 g, 32.4 mmol) in 20 ml of DCM at 0°C. The reaction mixture was then stirred at room temperature for 30 min. Afterwards, washed the reaction mixture with 100 ml water, and recovered DN-PEG3-amine in the DCM phase, which was then extracted to the aqueous phase in the presence of 0.1 M HCl. After adjusting the pH of the aqueous solution to ∼12–13 and mixing it with DCM, the DN-PEG3-amine product was extracted back to the DCM phase and dried overnight. The product was used without further purification for the next step of reaction. HR-ESI (positive): calculated 315.1299, found 315.1594 for C₁₂H₁₉N₄O₆. To synthesize TMR-DN, the above-prepared DN-PEG3-amine (1.8 mg, 5.7 μmol) was dissolved in 50 μl of dimethylformamide (DMF) and added dropwise to a solution of 5-carboxy-tetramethylrhodamine-N-hydroxysuccinimide (1.0 mg, 1.9 μmol) in 100 μl of DMF. The reaction mixture was then stirred at room temperature for 15 min. The reaction mixture was purified on a reversed-phase liquid chromatograph with a C18 column and a mobile phase composition of 60% acetonitrile and 0.1% trifluoroacetic acid to yield TMR-DN (0.53 mg, 38% yield). HR-ESI (positive): calculated 727.2722, found 727.3671 for C₃₇H₃₉N₆O₁₀.

In vitro fluorescence assay

All the in vitro fluorescence measurements were performed with a PTI fluorimeter (Horiba, New Jersey, NJ) at room temperature (23°C). Fluorescence signals were measured in a buffer consisting of 40 mM Tris, 5 mM MgCl₂, 100 mM KCl at pH = 7.6. The fluorescence emission spectra of Chili, Broccoli, Corn, DNB, Pepper and Mango II were collected by exciting at 415, 480, 505, 555, 577 and 580 nm, respectively. An Origin software was used to plot all the in vitro fluorescence data.

Vector construction for bacterial imaging

Pepper, DNB, Corn, and FDB were cloned into a pETDuet-1 vector (EMD Millipore, Burlington, MA), respectively. This vector was first double digested with SgrAI and NdeI restriction enzymes (NEB) and purified from 1% agarose gel using a QIAquick Gel Extraction Kit (Qiagen, Germantown, MD). A double-stranded insert was digested at the same restriction sites, followed by ligation with the digested vector, using T4 DNA ligase (NEB). Chili and FPC were cloned respectively into the pETDuet-1 by double digestion and ligation following the same procedure described above, and the restriction enzymes used were NdeI and PacI (NEB). The pETDuet-Broccoli plasmid was a gift from Samie R. Jaffrey's lab at Weill Cornell Medicine. F30-D-CDG/Pepper was cloned into the pETDuet plasmid via double digestion by SgrAI and SacII, and B-P₄G was cloned into the same vector via NdeI and XhoI digestion. All the ligated products were then transformed into TOP10 chemical competent cells (Invitrogen) and screened based on ampicillin resistance. All the plasmids were extracted using a GeneJET Plasmid Miniprep Kit (Thermo Fisher Scientific) and confirmed via Sanger sequencing performed by Eurofins Genomics.

Vector construction for mammalian imaging

For mammalian cell imaging, cFDB and cFPC were cloned into an AIO-Puro vector (Addgene). The DNA templates of cFDB and cFPC were respectively digested with BsaI and BbsI restriction enzymes (NEB) and ligated with the vector digested at the same restriction sites. cBroccoli, cDNB, cPepper and cCorn were cloned into an pAVU6 + 27-Tornado vector (Addgene). The DNA templates of cBroccoli, cDNB, cPepper, and cCorn were double digested by NotI-HF and SacII (NEB) and ligated with the vector digested at the same restriction sites. All the ligated products were then transformed into TOP10 chemical competent cells (Invitrogen) and screened based on ampicillin resistance. cBroccoli-7SK, P-ACTB, C-HR2, B-ST3 and D-TK1 sensors were respectively cloned into a pAV-U6 + 27-Tornado-Broccoli vector (Addgene), between the two ribozyme sequences (TTRz and TRz), by Gibson Assembly^® Cloning Kit (NEB). B-ST3 and C-HR2 were also respectively cloned into the above-mentioned pAV-U6 + 27-Tornado vectors that encode D-TK1 or P-ACTB, by double digestion with XbaI and XhoI. The pAV-U6 + 27–5xF30-tCorn plasmid was a gift from Jiahui (Chris) Wu's lab at UMass Amherst. All the plasmids were extracted using a GeneJET Plasmid Miniprep kit (Thermo Fisher Scientific) and confirmed by Sanger sequencing by Eurofins Genomics.

Bacterial cell imaging and data analysis

Confocal imaging of Escherichia coli cells was conducted according to a previously reported protocol (21). The BL21 Star (DE3) cells were grown to OD₆₀₀ of 0.4–0.5 at 37°C in LB media, and then 1 mM isopropyl β-d-1-thiogalactopyranoside (IPTG) was added for 4 h to induce cellular RNA transcription. All the fluorescence images were collected with NIS-Elements AR software using Yokogawa spinning disk confocal on a Nikon Eclipse-TI inverted microscope. Broccoli/DFHBI-1T and Corn/DFHO were excited with a 488 nm laser, and the emission was collected in the range of 500–550 nm via a filter set. Pepper/HBC620 and DNB/TMR-DN were imaged by a 561 nm laser, and the fluorescence emission was collected in the range of 575–625 nm. Chili was excited with a 405 nm laser, and the emission was collected in the range of 500–550 nm. All these images were obtained through a 60× or 100× oil immersion objective. Data analysis was performed with ImageJ software. Bacterial cells were identified by MicrobeJ plugin using the local default setting. For quantitative comparison, only the fluorescence signals from surface-attached rod-shaped bacterial cells were used. Data calculation and fitting were carried out using the Origin and GraphPad Prism software.

HPLC analysis of guanosine tetraphosphate (ppGpp)

HPLC analysis of ppGpp levels was performed following a previously described method (58) with slight adjustments. BL21 Star (DE3) cells expressing pETDuet-F30-D-CDG/Pepper-B-P4G were cultured overnight in LB medium at 37°C. After dilution to OD₆₀₀ of 0.1 and grown for 60 min, 1 mM IPTG was added for a 4-hour induction. Cells were then diluted to OD₆₀₀ = 1.0, and 3 ml of culture was harvested by centrifugation at 5000 g for 2 min, and then resuspended in imaging medium (DPBS with or without 100 μM tetracycline) for 45 min under shaking at 37°C. Following incubation, cells were centrifuged at 5000 g for 2 min, and the supernatant was discarded. Then 180 μl of 0.1 M KOH was added to the cell pellet for 30 min at 0°C to induce cell lysis. The lysates were subsequently neutralized by adding 1.8 μl of 85% H₃PO₄ and 180 μl of HPLC buffer (composed of 0.5 M NH₄H₂PO₄ and 2.5% acetonitrile, with a pH of 3.4), followed by centrifugation at 13000 rpm for 15 min to remove cellular debris. 10 μl of prepared cell lysates were then injected into a Hypersil™ SAX HPLC column (250 mm × 4.6 mm) within an Agilent 1100 HPLC system. An isocratic mobile phase of 100% above-mentioned HPLC buffer was run at a flow rate of 1 ml/min for 15 min. Analytes were detected using a 252 nm UV/Vis detector, and standard ppGpp samples were also tested to determine the target elution time.

HPLC analysis of cyclic diguanylate (c-di-GMP)

To measure cellular c-di-GMP levels using HPLC, we followed a previously reported procedure (59). BL21 cells expressing pETDuet-F30-D-CDG/Pepper-B-P4G were cultured overnight in LB medium at 37°C. The cultures were diluted to OD₆₀₀ = 0.2 in 10 ml of LB medium and allowed to grow until reaching an OD₆₀₀ of 0.4–0.5. Following this, the cells were induced with 1 mM IPTG for 4 hours at 37°C with continuous shaking. Cultures equivalent to 1 ml with an OD₆₀₀ of 1.8 were obtained for each condition. Cells were harvested by centrifugation at 5000 g for 2 min, and then resuspended in imaging medium (DPBS with or without 100 μM tetracycline) for a 45-min shaking at 37°C. The cultures were then centrifuged, after discarding the supernatant, the cell pellet was washed twice with ice-cold PBS, resuspended in 100 μl of ice-cold PBS, and then heated to 100°C for 5 min. After cooling, ice-cold ethanol was added to reach a final concentration of 65%, followed by vortexing for 15 s. The extraction procedure was repeated twice, and all the supernatants were combined in a tube and kept on ice. After vacuum drying the combined supernatants, the pellets were resuspended in 200 μl of DNase/RNase-free water. The samples were injected into a reversed-phase C18 column within an Agilent 1100 HPLC system, employing a mobile phase consisting of a mixture of buffer A (10 mM ammonium acetate in water) and buffer B (10 mM ammonium acetate in methanol), with a gradient ranging from 1% to 90% buffer B over 32 minutes. Elution was monitored using a 253 nm UV/Vis detector, and the elution time of c-di-GMP was determined using a standard c-di-GMP sample.

Mammalian cell culture and transfection

HEK293T/17 cells were purchased from ATCC and cultured in DMEM (high glucose) supplemented with 10% FBS and 1% penicillin-streptomycin under 5% CO₂ at 37°C. SKBR3 cells were also purchased from ATCC and cultured in McCoy's 5A Medium supplemented with 10% FBS and 1% penicillin-streptomycin under 5% CO₂ at 37°C. The cell lines were mycoplasma-free and split using 0.25% trypsin at ∼80% confluence. HEK293T cell transfection was performed using the FuGENE HD transfection reagent (Promega) according to the manufacturer's instructions. More specifically, 1 μg DNA plasmids and 3 μl FuGENE reagent were first mixed with 50 μl Gibco Opti-MEM medium and incubated at room temperature for 15 min. Then, 12 μl of the mixture was added to a chamber of poly-D-lysine-coated 8-chamber glass bottom slide (Cellvis, CA) for imaging. The same batch of HEK293T cells without transfection were used as the negative control and added to another chamber of the same imaging slide. SKBR3 cells used for sequential imaging were transfected with lipofectamine™ 3000 (Invitrogen) according to the manufacturer's instructions. In tube 1, 25 μl of Opti-MEM medium and 0.5 μl of Lipofectamine 3000 reagent were mixed. In tube 2, 25 μl of Opti-MEM medium, 500 ng of DNA, and 1 μl of P3000™ reagent were mixed. Afterwards, tube 2 solution was added to the tube 1, mixed well and incubate at room temperature for 10–15 min. 20 μl of the complex was then added to each well of cells and swirled gently to distribute it to the entire well. After 48 hours of transfection, these HEK293T and SKBR3 cells were ready for imaging.

Mammalian cell imaging and data analysis

HEK293T and SKBR3 cell imaging were also conducted using Yokogawa spinning disk confocal on the Nikon Eclipse-TI inverted microscope. All the images were obtained through a 40 × oil immersion objective. cBroccoli/DFHBI-1T and cCorn/DFHO were excited with a 488 nm laser, and the emission was collected in the range of 500–550 nm via a filter set. cPepper/HBC620 and cDNB/TMR-DN were imaged by a 561 nm laser, and the fluorescence emission was collected in the range of 575–625 nm. Data analysis was performed with ImageJ software. Cells were identified manually for quantitative analysis. Data calculation and fitting were carried out using the Origin and GraphPad Prism software. To distinguish fluorescence ‘on’ and ‘off’ cells, cellular fluorescence signals of untransfected HEK293T and SKBR3 cells after a 30-min incubation with the corresponding dye molecules at a concentration of 40 μM DFHBI-1T, 1 μM TMR-DN, 40 μM DFHO, or 5 μM HBC620 are first used to determine the background cellular fluorescence intensities for each imaging channel. The cellular fluorescence threshold values were then calculated as the average background fluorescence (μ) from 10 untransfected cells plus two times of the standard deviation (σ) values of these background signals, i.e. ‘μ+2σ’. In each imaging channel, HEK293T and SKBR3 cells that exhibit greater than μ+2σ cellular fluorescence intensities were considered as having this imaging channel ‘on’. Otherwise, the corresponding imaging channel was considered as ‘off’.

Cytotoxicity measurement

Cytotoxicity measurement was performed by incubating BL21 Star (DE3) with 1 μM SYTOX Blue nucleic acid stain (Invitrogen) for 5 min before imaging by a 405 nm laser irradiation. For HEK293T cells, live and dead cells were counted by a Luna II™ cell counter with trypan blue staining. More specifically, cells in a 12-well plate were washed with DPBS before incubating with 0.2 ml of 0.25% trypsin on each well. After the cells were detached, 0.8 ml of complete medium was added per well. Then 10 μl of cell suspension was mixed with 10 μl of 0.4% trypan blue solution, and 10 μl of the mixture was loaded a Luna™ cell counting slide to determine the number of dead and live cells via a Luna II™ cell counter. Cell proliferation was measured using a CyQUANT™ XTT Cell Viability Assay (Invitrogen). Here, after sequential imaging or photobleaching, HEK293T cells were counted and seeded in a 96-well plate at a density of 1 × 10⁴ cells in 100 μl complete medium per well. The cells were then incubated in a 37°C incubator for 24 h. 6 ml of XTT reagent was mixed with 1 ml of the electron coupling reagent, and 70 μl of the mixture was added into each well. After incubating at 37°C for 4 h, absorbance was read at 450 and 660 nm to measure cell proliferation.

Quantitative reverse transcription PCR (qRT-PCR) determination

The expression level information of ACTB, HER2, STAT3 and TK1 mRNAs in SKBR3 and HEK293T cells were first obtained from Expression Atlas (www.ebi.ac.uk/) and then experimentally determined by qRT-PCR. Here, after counting and harvesting cells grown in a T-25 flask, total cellular RNAs were extracted with a GeneJET RNA Purification Kit (Thermo Fisher Scientific) according to the manufacturer's instructions. The purified RNAs was then reverse-transcribed to the cDNAs using an oligo-dT primer and SuperScript™ III First-Strand Synthesis System (Invitrogen). qPCR analysis of cDNA was performed using a Bio-Rad CFX96 Touch Real-Time PCR instrument. Each reaction mixture was composed of 2 μl cDNA, 10 μl Applied Biosystems™ PowerUp™ SYBR™ Green Master Mix, 2 μl upstream primer (5 μM), 2 μl downstream primer (5 μM) and 4 μl nuclease-free water. The qPCR conditions were as follows: staying at 50°C for 2 min and then 95°C for 2 min, followed by 40 cycles of 95°C for 15 s, 56°C for 15 s and 72°C for 1 min. The primer sequence for ACTB was from OriGene (Rockville, MD), while primers for HER2, STAT3 and TK1 mRNAs were obtained from PrimerBank (https://pga.mgh.harvard.edu/primerbank/). The data was collected using the Bio-Rad CFX manager software. Relative expression levels of HER2, STAT3, and TK1 were quantified by using ACTB mRNA as the reference, based on a calibration curve of Ct values versus ACTB amplicon RNA contents in serial dilution ratios.

siRNA knockdown

Predesigned siRNAs were purchased from MilliporeSigma that target ACTB (SASI_Hs01_00204238) or HER2 (SASI_Hs01_00077799) mRNA. These siRNAs were transfected into SKBR3 cells by lipofectamine™ 3000 (Invitrogen) according to the manufacturer's instructions. Briefly, in tube 1, 25 μl of Opti-MEM medium and 1.5 μl of Lipofectamine 3000 reagent were mixed. In tube 2, 25 μl of Opti-MEM medium and 15 pmol of siRNA were mixed. Afterwards, tube 2 solution was added to the tube 1, mixed well and incubate at room temperature for 10–15 min. 20 μl of the complex was then added to each well of cells and swirled gently to distribute it to the entire well. The medium was removed after 6 h, and the sensor plasmids were then transfected for fluorescence imaging.

Results

Identification of orthogonal FR/dye pairs

We first tested the orthogonality among several fluorogenic RNA aptamers and their cognate dyes, including Broccoli/DFHBI-1T, Chili/DMHBI⁺, Corn/DFHO, DNB/TMR-DN, Mango II/YO3-biotin, and Pepper/HBC620 (Supplementary Figure S1 and Table S1) (20–29). Under optimized RNA and dye concentrations, the DFHBI-1T, DFHO, DMHBI⁺, HBC620, and TMR-DN dyes exhibited great orthogonality: they can only be activated by corresponding fluorogenic RNAs, with 10–600-fold higher fluorescence signals than other tested RNA aptamers (Figure 1C and Supplementary Figure S1). In contrast, YO3-biotin can be nonspecifically activated by almost all the attempted FR aptamers, so it was excluded in the following studies.

A consistent linear increase in fluorescence signals was observed with increasing RNA concentrations for each orthogonal fluorogenic RNA/dye pair. This linear trend remained in the mixture of RNA aptamers (Supplementary Figure S2), which feature is critical for potential quantitative detection (30). Mixing with other aptamers did not affect Broccoli/DFHBI-1T and Pepper/HBC620 fluorescence intensities but caused a slight decrease (∼28–46%) in the DNB/TMR-DN and Chili/DMHBI⁺ fluorescence signals; interestingly, Corn/DFHO signals were even enhanced by ∼30% in the mixture (Supplementary Figure S2). These data indicated that some RNA–RNA interactions may exist in these mixtures, but still Broccoli/DFHBI-1T, Chili/DMHBI⁺, Corn/DFHO, DNB/TMR-DN and Pepper/HBC620 can function orthogonally in vitro and were identified as potential candidates for developing seqFRIES.

We next tried to image these five FR/dye pairs inside BL21 Star (DE3) E. coli cells. After expressing Broccoli, Corn, DNB, Pepper, or Chili RNAs, respectively, strong cellular fluorescence signals were observed in the presence of DFHBI-1T, DFHO, TMR-DN, or HBC620, but not for DMHBI⁺ (Supplementary Figure S3A). We speculate that the DMHBI⁺ dye has poor cell membrane permeability, and in fact, the Chili/DMHBI⁺ pair has indeed never been applied for cellular imaging. As a validation of the cellular orthogonality of the rest four FR/dye pairs, our imaging results clearly revealed that each kind of dye molecule could only be specifically activated by the cognate RNA aptamer (Figure 1D, E). Indeed, Broccoli/DFHBI-1T, Corn/DFHO, DNB/TMR-DN, and Pepper/HBC620 maintained their great orthogonality inside bacterial cells for potential multiplexed detection.

Cellular fluorescence activation and deactivation kinetics of the FR/dye pairs

Considering the importance of fast imaging-and-stripping in developing seqFRIES, we next assessed the cellular fluorescence activation/deactivation kinetics of each FR/dye pair. As shown in Supplementary Figure S3B, C, after adding DFHO, DFHBI-1T, and HBC620, very fast fluorescence enhancement was observed inside the corresponding Corn-, Broccoli- or Pepper-expressing E. coli cells. ∼80% of maximal cellular signals were shown within ∼7, 8 and 13 min, respectively. In contrast, the DNB/TMR-DN pair exhibited a much slower fluorescence activation, reaching ∼45% of its maximum at ∼30 min. While it is worth noting that because of the high brightness of DNB/TMR-DN, their cellular fluorescence can still be clearly visualized within ∼10 min.

We also determined how quickly the cellular fluorescence signals of FR/dye pairs can be stripped. For this purpose, FR-expressing E. coli cells were first incubated with their cognate dye molecule for 50 min, and then the medium was swapped into a dye-free Dulbecco's phosphate-buffered saline (DPBS) for two rounds of 1-min incubation. Cellular fluorescence of Broccoli/DFHBI-1T, Corn/DFHO, and Pepper/HBC620 could be easily removed: after a total of 2-min washing, only ∼17–20% signals remained in these bacterial cells (Supplementary Figure S4A, B). In the case of DNB/TMR-DN, the same 2-min washing protocol is insufficient. Even with increased incubation time of 6 min, ∼45% and 25% of DNB/TMR-DN signals remained after 2–5 rounds of washing (Supplementary Figure S4C, D).

Besides washing, we demonstrated two other stripping approaches for DNB/TMR-DN: photobleaching and competitive binding. Our results indicated that 1-min laser (561 nm, 1 mW/cm² power) irradiation could reduce cellular DNB/TMR-DN fluorescence to ∼1% (Supplementary Figure S5). Alternatively, by adding a non-fluorescent small-molecule ligand, DN-PEG₃-amine, to compete with TMR-DN for the dinitroaniline binding site in the DNB aptamer, cellular DNB/TMR-DN signals could be reduced to ∼15% in ∼5 min (Supplementary Figure S5). All these data support that fast imaging-and-stripping are feasible for the four orthogonal FR/dye pairs.

Sequential cellular imaging of orthogonal FR/dye pairs

We next prepared a pETDuet plasmid construct that can express all four fluorogenic RNAs. DNB and Broccoli were fused via a three-way junction F30 RNA scaffold and inserted into one expression cassette of the vector (Supplementary Figure S6A). Similarly, Pepper and Corn on another F30 scaffold were placed into the other cassette of the same vector. The resulting plasmid is named FDB-FPC. The F30 scaffold can promote proper folding of RNA aptamers and reduce the interference among RNA sequences (31). Indeed, the fluorescence intensities of Broccoli, Corn, and DNB within these F30 scaffolds increased by ∼0.6-, 0.7- and 1.8-fold compared to those without the scaffold (Supplementary Figure S6B). After cloning FDB-FPC into BL21 Star (DE3) cells and incubated with DFHBI-1T, DFHO, TMR-DN or HBC620 dyes for 30 min, bright cellular fluorescence signals were observed for all four FR/dye pairs (Supplementary Figure S6C, D). The cellular fluorescence of FDB-FPC/DFHO was even 1.7-fold higher than that of Corn/DFHO.

Using FDB-FPC-expressing E. coli cells, our next goal was to investigate if each FR/dye pair can be repeatedly imaged and stripped. During a total of six rounds of imaging-and-stripping, >70% of original cellular fluorescence signals can still be observed for Broccoli/DFHBI-1T, Corn/DFHO, and Pepper/HBC620 (Figure 2A and Supplementary Figure S7). The slightly decreased fluorescence is likely due to the cellular degradation of fluorogenic RNAs during these over 3-h-long experiments. In the case of DNB/TMR-DN, considering its relatively slow fluorescence activation and deactivation kinetics, we only tested three rounds of sequential imaging. Still, robust fluorescence activation and deactivation cycles were observed during the repeated imaging-and-stripping process (Figure 2A and Supplementary Figure S7).

Figure 2.

Sequential multiplexed imaging in living BL21 Star (DE3) cells that express all four FRs (FDB-FPC). (A) Fluorescence images after each round of 30-min incubation with 200 μM DFHBI-1T, 1 μM TMR-DN, 10 μM DFHO, or 1 μM HBC620. Three times of 1-min DPBS washing for DFHBI-1T, DFHO, and HBC620, or five times of 6-min washing for TMR-DN, were performed after each round of imaging to strip cellular fluorescence. Scale bar, 5 μm. (B) Fluorescence images after a sequential 20-min incubation with 1 μM HBC620, 10 μM DFHO, 200 μM DFHBI-1T, or 1 μM TMR-DN. Three times of DPBS washing for HBC620 (3 min), DFHO (1 min), and DFHBI-1T (1 min) were used to strip cellular fluorescence. Scale bar, 5 μm. (C) Correlations of cellular Broccoli, Corn, DNB, and Pepper fluorescence as measured in 100 individual cells following the 4-round imaging protocol. Pearson's correlation coefficient r was also shown. (D) Fluorescence images after a sequential 5-min incubation with 1 μM HBC620 and 10 μM DFHO, or 200 μM DFHBI-1T and 1 μM TMR-DN. Three times of 3-min DPBS washing were performed to strip cellular fluorescence. Scale bar, 5 μm. (E) Cytotoxicity measurement by incubating BL21 Star (DE3) cells with 1 μM SYTOX Blue for 5 min before (control) and after a 4-round or 2-round sequential imaging process. Cells pre-treated with 1 mM tetracycline for 30 min were used as the positive control. SYTOX Blue fluorescence was measured in ∼50 individual cells in each case upon a 405 nm laser irradiation. Shown are the mean and SD values from cellular images taken from at least three independent replicates.

We then set out to sequentially image the four FR/dye pairs, one at a time. The dyes were added and washed in the order of HBC620, DFHO, DFHBI-1T and TMR-DN. Almost 100% of FDB-FPC-expressing E. coli cells exhibited bright fluorescence signals in all four rounds of imaging cycles, and >90% of cellular fluorescence were stripped after each washing cycle (Figure 2B). Nicely correlated cellular fluorescence intensities were shown among all four FR/dye pairs (Pearson's correlation coefficient r, 0.79–0.91) (Figure 2C), suggesting these FR/dye pairs are well suited for the proposed seqFRIES platform.

Dual-channel rapid sequential imaging

We further studied if these four FR/dye pairs can be imaged within a shorter time frame via simultaneous dual-channel detection. Our previous work demonstrated that Broccoli/DFHBI-1T and DNB/TMR-DN possess a large difference in their excitation/emission spectra and can be imaged together (32). After validating this observation in FDB-FPC-expressing BL21 Star (DE3) cells (Supplementary Figure S8A), we also found that Corn/DFHO and Pepper/HBC620 exhibit minimal spectral overlap with almost no fluorescence interference between them. As a result, for a more rapid version of seqFRIES, we decided to image Corn/DFHO and Pepper/HBC620 together in one round, and Broccoli/DFHBI-1T and DNB/TMR-DN in another.

The dye incubation time was also shortened to ∼5 min based on the above-tested cellular fluorescence activation kinetics (Supplementary Figure S3). Three times of 3-min washing were performed between two imaging rounds, so the whole seqFRIES process was completed within ∼20 min. Following this protocol, the fluorescence signals from each of four FR/dye pairs can be clearly visualized within ∼100% of FDB-FPC-expressing cells (Figure 2D), also with good correlations in all four-channel cellular fluorescence intensities (Pearson's correlation coefficient r, 0.68–0.90) (Supplementary Figure S8B).

We further compared the single-channel four-round protocol (Figure 2B) versus dual-channel two-round protocol in terms of cellular fluorescence intensities of each FR/dye pair (Figure 2D). Even though the incubation time was shortened from 20 min to 5 min following the rapid two-round approach, similar fluorescence signals of each FR/dye pair (∼73–94%) were showed as compared to those from the four-round imaging protocol (Supplementary Figure S8C). Meanwhile, minimal cytotoxicity was detected following both four-round and two-round sequential imaging processes (Figure 2E). All these results demonstrated that seqFRIES is well adapted for rapid multiplexed imaging in living bacterial cells.

Multiplexed sequential imaging in mammalian cells

Our next goal was to test if seqFRIES can also function inside mammalian cells. To increase the cellular expression level of fluorogenic RNAs, a pAVU6 + 27-Tornado vector (33) was used to separately encode a circularized F30-Broccoli (abbreviated as cBroccoli), circular F30-Corn (cCorn), circular F30-DNB (cDNB) or circularized F30-Pepper (cPepper). We confirmed that cBroccoli/DFHBI-1T, cCorn/DFHO, cDNB/TMR-DN, and cPepper/HBC620 can still be imaged orthogonally in HEK293T cells (Figure 3A, B). As an additional control, we tried to localize these fluorogenic RNAs to specific subcellular locations—cBroccoli to nuclear speckles by tagging it with a 7SK small nuclear RNA (cBroccoli-7SK), and cCorn to nucleus as assemblies on a 5 × F30-tCorn (5 × Corn) construct (60). When co-transfecting cBroccoli-7SK and cDNB, we observed cBroccoli/DFHBI-1T and cDNB/TMR-DN signals distinctively in nuclear speckles and cytosol, respectively (Supplementary Figure S9). cPepper and 5 × Corn signals were also localized to the cytosol and nucleus, respectively, in the same cells with little crosstalk. We then investigated the cellular fluorescence activation and deactivation kinetics of cBroccoli, cCorn, cDNB and cPepper. As shown in Figure 3C, D, immediately after adding the DFHBI-1T or DFHO dye, very fast fluorescence activation was observed in cells that express cBroccoli or cCorn, with half maximal fluorescence signals reached within ∼1 min. Similarly, cPepper/HBC620 exhibited a rapid fluorescence activation (half maximal level at ∼9 min). In the case of cDNB/TMR-DN, similar to that shown in E. coli cells, a gradual increase in cellular fluorescence was observed, with ∼50% of the maximal signals detected at ∼18 min.

Figure 3.

Orthogonality and fluorescence activation kinetics of FR/dye pairs in HEK293T cells that express cBroccoli, cDNB, cCorn or cPepper. (A) Fluorescence images of HEK293T cells after a 30-min incubation with 40 μM DFHBI-1T, 1 μM TMR-DN, 40 μM DFHO, or 5 μM HBC620. Scale bar, 20 μm. (B) Cellular fluorescence intensities as quantified in ∼20 individual cells in each case. Shown are the mean and SD values from cellular images taken from at least three independent replicates. (C) Cellular fluorescence signals were monitored after adding 40 μM DFHBI-1T, 1 μM TMR-DN, 40 μM DFHO or 5 μM HBC620 at 0 min. Scale bar, 20 μm. (D) Cellular fluorescence activation kinetics as measured in ∼20 individual HEK293T cells in each case. Shown are the mean and the standard error of the mean (SEM) values from cellular images taken from at least three independent replicates.

As for the deactivation kinetics measurement, after five rounds of 1-min washing, cBroccoli/DFHBI-1T, cCorn/DFHO, and cDNB/TMR-DN signals were rapidly reduced to ∼9%, 20% and 23% of the original (Figure 4A, B). In contrast, cPepper/HBC620 fluorescence was more difficult to strip under this washing condition: ∼40% cellular signals remained. We further tried to apply the above-mentioned photobleaching-based stripping approach for the cPepper/HBC620 pair. Indeed, cellular cPepper/HBC620 fluorescence signals were rapidly reduced to ∼10% of the original after photo-stripping (2–4 min 561 nm laser irradiation at 1 mW/cm²) (Figure 4C). These washing and photobleaching steps have minimal influence on the cell viability and proliferation (Supplementary Figure S10A, B). All these results supported that fast fluorescence activation and deactivation of these four orthogonal FR/dye pairs can still be achieved inside mammalian cells.

Figure 4.

Sequential multiplexed imaging in HEK293T cells. (A) After incubating HEK293T cells that express cBroccoli, cDNB, cCorn or cPepper for 30 min with 40 μM DFHBI-1T, 1 μM TMR-DN, 40 μM DFHO, or 5 μM HBC620, fluorescence images were taken during and after five rounds of 1-min DPBS washing. Scale bar, 20 μm. (B) Cellular fluorescence deactivation kinetics as measured in ∼20 individual HEK293T cells in each case. Shown are the mean and SEM values from cellular images taken from at least three independent replicates. (C) Stripping of cellular fluorescence signals via brief photobleaching: 2–4 min of irradiation with a 1 mW/cm² power 561 nm laser. Scale bar, 10 μm. (D) Fluorescence images of HEK293T cells that express all four FRs (cFDB-cFPC) after a sequential 20-min incubation with 40 μM DFHBI-1T and 1 μM TMR-DN, or 40 μM DFHO and 5 μM HBC620. Three times of 2-min DPBS washing were performed to strip cellular fluorescence. Scale bar, 15 μm. (E) Percentage of HEK293T cells that exhibit 1–4 channel activated fluorescence signals as measured in ∼100 cells after 2-round sequential imaging from at least three independent replicates. (F) Heatmap of four FR/dye pair fluorescence signals from 100 individual HEK293T cells. The pseudo-color values indicated the relative brightness, calculated by subtracting threshold fluorescence from cellular fluorescence values, and then divided by the standard deviation of cellular fluorescence. Positive values indicated the ‘on’ fluorescence.

To prepare the vector for seqFRIES imaging in mammalian cells, we cloned both circularized F30-DNB/Broccoli (abbreviated as cFDB) and circular F30-Pepper/Corn (cFPC) into an AIO-Puro-based plasmid. After transfecting the cFDB-cFPC plasmid into HEK293T cells, we sequentially imaged cBroccoli/DFHBI-1T, cDNB/TMR-DN and cCorn/DFHO, cPepper/HBC620 pairs via the dual-channel two-round protocol (Figure 4D). As expected, the majority (∼85%) of transfected HEK293T cells exhibited bright fluorescence signals from four or three FR/dye pairs (Figure 4E). Good correlations among cPepper/HBC620, cDNB/TMR-DN, and cBroccoli/DFHBI-1T fluorescence were shown in these HEK293T cells (Pearson's r, 0.48–0.71) (Supplementary Figure S10C). The cBroccoli/DFHBI-1T signals could be observed in almost 100% of cells that showed at least one set of FR/dye fluorescence (Figure 4F), while in contrast, cCorn/DFHO was more difficult to image, which was activated in only ∼55% of the HEK293T cells with relatively poor correlations (Pearson's r < 0.27) with the other three FR/dye channels (Supplementary Figure S10C). It is worth noting that nucleus punctate features were shown in some cells after expressing FRs, via colocalization analysis with nucleus and nucleoli staining dyes (Supplementary Figure S11), we found that while the majority of small puncta (<1 μm) are within nucleoplasm, the larger ones (∼2–3 μm) are localized in nucleoli. To mitigate the influence of these nucleus punctate features on the fluorescence quantification, we will later only measure the cytosol signals of the seqFRIES system. Meanwhile, the cellular stability of these circular RNAs were also tested. Using cBroccoli/DFHBI-1T and cPepper/HBC620 as the examples, cellular fluorescence signals could be clearly visualized with minimal decay at > 84 h after transfection (Supplementary Figure S12A, C). When the cells were treated with a transcription inhibitor, actinomycin D, almost constant fluorescence signals were observed after 6 h (Supplementary Figure S12B, C), proving the superior stability of these circular RNAs. These results indicated the successful performance of seqFRIES for multi-color sequential imaging in living mammalian cells (Supplementary Table S2).

seqFRIES-mediated cellular detection of signaling molecules

After validating the seqFRIES system, we wanted to test if this sequential imaging platform can be used for detecting different target analytes. For this purpose, two bacterial second messengers, cyclic diguanylate (c-di-GMP) and guanosine tetraphosphate (ppGpp), were chosen because of their critical regulatory functions in bacterial physiology and stress responses (34–36). Using our previously engineered Broccoli-based ppGpp sensor (B-P₄G) (37) and DNB-based c-di-GMP sensor (D-CDG) (30), we first tested their orthogonality performance in a mixture together with another two Pepper- and Corn-based sensors (Supplementary Figure S13A) (38). As shown in Figure 5A, B, both the target binding affinity and selectivity of B-P₄G and D-CDG were well maintained. Physiological concentrations of ppGpp and c-di-GMP can be potentially detected based on the dose-response curves. Broccoli-, DNB-, Pepper- and Corn-based sensors can function independently in the same solution.

Figure 5.

seqFRIES-based sensors for detecting cellular signaling molecules. (A) Dose-response curves for the fluorescence detection of ppGpp and c-di-GMP by the optimal B-P₄G and D-CDG sensors or in a mixture of four sensors (including P-SAM and C-SAH) at 1 μM concentration. Shown are the mean and SEM values from at least three independent replicates. (B) Selectivity of B-P₄G and D-CDG as measured in a solution containing 1 μM sensor, 20 μM DFHBI-1T or 0.5 μM TMR-DN, and 100 μM indicated compounds. Shown are the mean and SEM values from at least three independent replicates. (C) Fluorescence images of BL21 Star (DE3) cells that co-express B-P₄G and FPD-CDG in the presence or absence of 100 μM tetracycline. 200 μM DFHBI-1T and 1 μM TMR-DN were added for 30 min before the first round of imaging. After three times of 5-min washing and a 30-min incubation with 1 μM HBC620, other images were taken. Scale bar, 5 μm. (D) Using Pepper/HBC620 signals as the reference, the relative fluorescence ratios of B-P₄G/DFHBI-1T (I_B-P4G/I_Pepper) and D-CDG/TMR-DN (I_D-CDG/I_Pepper) were measured in ∼100 individual cells in each case. Shown are the mean and SD values from cellular images taken from at least three independent replicates. Two-tailed Student's t-test: ****P< 0.0001; ns, not significant, P> 0.05.

To further study the cellular performance of B-P₄G and D-CDG for sequential multiplexed detection, we encoded F30-scaffolded D-CDG and Pepper (FPD-CDG) and B-P₄G into two T7 expression cassettes of a pETDuet plasmid (Supplementary Figure S13B). Pepper was used to generate reference signals for RNA expression level normalization in individual E. coli cells. While the cellular performance of D-CDG and B-P₄G have been validated in our previous studies (30,37), our goal here is to enable the first direct multiplexed live-cell measurement of both c-di-GMP and ppGpp, two critical signaling molecules shown to collaboratively facilitate bacterial survival under stress conditions, e.g. antibiotic treatment (39–43).

After co-expressing FPD-CDG and B-P₄G inside BL21 Star (DE3) cells, we first incubated the cells with TMR-DN and DFHBI-1T to simultaneously detect cellular c-di-GMP and ppGpp levels. Afterwards, three times of 5-min washing were used to strip cellular fluorescence before the next round of imaging for Pepper/HBC620 reference signals (Figure 5C). The same seqFRIES procedure was also performed in the presence of 100 μM tetracycline for studying the effect of antibiotic treatment. To normalize RNA expression level variations in each individual cell, ratiometric fluorescence intensities, I_B-P4G/I_Pepper and I_D-CDG/I_Pepper, were used to detect tetracycline-induced cellular changes in ppGpp and c-di-GMP levels. An ∼70% lower cellular I_B-P4G/I_Pepper signals were shown upon adding 100 μM tetracycline, while c-di-GMP levels were not significantly affected in the same cells (Figure 5D). To rule out the possibility that tetracycline may directly influence the cellular fluorescence of Broccoli/DFHBI-1T and DNB/TMR-DN, a control experiment was performed using FDB-FPC-expressing cells. Following the same seqFRIES procedure, none of the Broccoli/DFHBI-1T, DNB/TMR-DN, and Pepper/HBC620 fluorescence were affected by the addition of 100 μM tetracycline (Supplementary Figure S13E, F). Our results suggested that these E. coli cells tend to reduce the accumulation of intracellular ppGpp levels to survive under antibiotic (tetracycline) stress, while c-di-GMP may not play a direct role in this process. The trend was also confirmed by HPLC analysis (Supplementary Figure S14). Some more detailed mechanism studies will be followed in the future.

seqFRIES-mediated cellular detection of endogenous mRNA targets

Lastly, we wanted to apply seqFRIES for imaging different cellular mRNA targets including HER2, STAT3, TK1 and ACTB. Housekeeping β-actin-encoding ACTB was chosen as a reference (44), while HER2 (human epidermal growth factor receptor 2), STAT3 (signal transducer and activator of transcription 3), and TK1 (thymidine kinase 1) were selected as important cancer biomarkers (45–47). We first accessed the sensing ability of each FR/dye combination based on their RNA level-dependent cellular fluorescence signal changes (Supplementary Figure S15A–C). The detection limit of cBroccoli, cPepper, cCorn and cDNB (defined based on signal greater than the average plus three standard deviations of the background) was found to be equivalent of respectively ∼1.3%, 1.3%, 0.8% and 5.9% ACTB mRNA levels in HEK293T cells (Supplementary Figure S15D). To engineer seqFRIES sensors for different mRNA targets, we applied a structure-switching strategy (48,49) where each sensor is composed of four domains: blocker, target-binding region, linker and fluorogenic RNA (Figure 6A). In the absence of target mRNAs, the blocker hybridizes with part of fluorogenic RNA and results in minimal fluorescence. Meanwhile, the linker and part of target-binding region further elongates this hybridization region and together forms a hairpin structure. Once the mRNA analyte hybridizes to the target-binding region, the hairpin unfolds and allows the refolding of fluorogenic RNA to activate fluorescence signals.

Figure 6.

seqFRIES-based sensors for detecting different cellular mRNAs. (A) Schematic of the structural switching sensor design using P-ACTB as an example. Corresponding sequences of the blocker (black), target-binding region (blue), linker (purple), and fluorogenic RNA Pepper (red) were shown respectively. NUPACK and mFold software were used for the sequence design. The specificity of the target-binding region was checked by BLAST. (B) Identification of FR-based mRNA sensors as measured respectively with 1 μM Pepper-based ACTB sensor + 1 μM HBC620, 5 μM Corn-based HER2 sensor + 2 μM DFHO, 1 μM Broccoli-based STAT3 sensor + 20 μM DFHBI-1T, or 2.5 μM DNB-based TK1 sensor + 0.5 μM TMR-DN. A 1:3 RNA sensor-to-target ratio was applied, and the optimal sensor was marked with a black arrow. (C) Dose-response curves for the optimal sensors as measured respectively with 100 nM P-ACTB + 100 nM HBC620, 1 μM B-ST3 + 20 μM DFHBI-1T, 5 μM C-HR2 + 2 μM DFHO, or 2.5 μM D-TK1 + 0.5 μM TMR. (D) Fluorescence images of SKBR3 cells that were co-transfected with B-ST3-D-TK1 and C-HR2-P-ACTB at 1:1 ratio after a sequential 30-min incubation with 40 μM DFHBI-1T and 1 μM TMR-DN, or 20 μM DFHO and 1 μM HBC620. Three times of 5-min washing were performed to strip cellular fluorescence. Cells were treated with either 250 nM TSA or 0.1% DMSO (control) for 24 h before imaging. Scale bar, 10 μm. (E) Cellular fluorescence intensities as quantified in ∼10 individual cells in each case. Shown are the mean and SD values from at least three independent replicates. Two-tailed Student's t-test: ****P< 0.0001; ns, not significant, P> 0.05.

After optimizing the lengths and sequences of the blocker and linker domains, we were able to obtain a Pepper-based sensor for ACTB (P-ACTB), a Corn-based HER2 sensor (C-HR2), a Broccoli-based STAT3 sensor (B-ST3) and a DNB-based TK1 sensor (D-TK1), with ∼3–9-fold fluorescence enhancement (Figure 6B), high selectivity (Supplementary Figure S16A), and ∼10–100 nM target detection limit (Figure 6C). To study the potential impact of RNA circularization on the sensor performance, we also tested these sensors within a rigid tRNA scaffold (Supplementary Figure S16B) (48). Still, all the tRNA-scaffolded sensors could respond to their RNA targets with high sensitivity (Supplementary Figure S16C).

We next tried to use these sensors to image cellular mRNA levels in SKBR3 (breast cancer cell line that overexpresses HER2, STAT3 and TK1) and HEK293T cells (negative control with low levels of HER2, STAT3 and TK1 mRNAs). After transfecting these cells with a pAVU6 + 27-Tornado vector that separately encodes a circularized P-ACTB, C-HR2, B-ST3 or D-TK1 sensor, and also applied cPepper-, cCorn-, cBroccoli- or cDNB-expressing vector as the positive control and an empty F30 scaffold vector as the negative control, high fluorescence signals of C-HR2, B-ST3 and D-TK1 were indeed only observed in SKBR3, but not in HEK293T cells (Supplementary Figure S17 and Supplementary Table S3). While as expected, housekeeping ACTB fluorescence was shown in both cell lines. Our qRT-PCR results also validated the expression level differences of these mRNA targets, which were nicely correlated with cellular fluorescence signals (Supplementary Figure S18). To further confirm that these sensors can indeed capture the changes in target mRNA levels, we also performed siRNA knockdown of ACTB and HER2 mRNAs. As expected, cellular P-ACTB and C-HR2 signals significantly decreased corresponding to the diminished target levels, while the knockdown did not change cPepper or cCorn signals (Supplementary Figure S19).

Trichostatin A (TSA) is a potent inhibitor of histone deacetylase and potential anticancer drug (50). TSA can alter the expression profile of many genes (51) and was found to suppress HER2 transcription in SKBR3 cells (52,53). Studying the effect of TSA on several mRNA biomarkers will aid in understanding its molecular mechanism and pathways in gene regulation. To enable single-cell profiling of HER2, STAT3 and TK1 in response to the TSA treatment, we co-transfected circular P-ACTB, C-HR2, B-ST3 and D-TK1 into the SKBR3 cells using two pAVU6 + 27-Tornado-based vectors. Following a dual-channel two-round sequential imaging protocol, we could clearly observe the sensor signals from individual cells (Figure 6D). TSA-treated SKBR3 cells showed a decreased C-HR2/DFHO and D-TK1/TMR-DN signals compared to those from the untreated group. In contrast, B-ST3/DFHBI-1T and P-ACTB/HBC620 fluorescence were not affected by the TSA treatment (Figure 6D, E). As a control, cBroccoli/DFHBI-1T, cDNB/TMR-DN, cCorn/DFHO, and cPepper/HBC620 signals in SKBR3 cells were not affected by TSA (Supplementary Figure S20A, B). Using the P-ACTB/HBC620 signal as a reference, I_B-ST3/I_P-ACTB, I_D-TK1/I_P-ACTB, and I_C-HR2/I_P-ACTB ratiometric fluorescence intensities can respectively reflect the relative changes in cellular STAT3, TK1, and HER2 mRNA levels. Again, our ratiometric data suggested a decrease in TK1 and HER2 concentrations, while STAT3 remained unaffected (Supplementary Figure S20C). This trend was also confirmed by the qRT-PCR results (Supplementary Figure S20D). All these results demonstrated that seqFRIES can be applied for detecting multiple endogenous mRNA targets in living systems.

Discussion

We herein introduced a seqFRIES system for live-cell multiplexed fluorescence imaging. Using multiple rounds of sequential imaging-and-stripping, seqFRIES overcomes the spectral overlap issues in traditional fluorescence techniques. One unique feature of this fluorogenic RNA-based platform is that the dye molecules can be reversibly separated from the genetically encoded RNA molecules, and as a result, the chemical dyes can be added externally and afterward easily washed away. In this study, we identified four pairs of orthogonal fluorogenic RNA/dye conjugates, including Broccoli/DFHBI-1T, Corn/DFHO, DNB/TMR-DN, and Pepper/HBC620, and showed that by adding only one or two dyes at a time, all these fluorescent pairs can be imaged inside the same individual cells without spectrally influencing each other. It is worth noting that current cellular fluorescence signal-to-noise ratio of Corn/DFHO is relatively low, which may potentially be improved by adding more Mg²⁺ in the imaging buffer to promote the folding of Corn RNA or optimizing the emission filter.

We have also demonstrated that this novel multiplexed imaging technique can function in living bacterial and mammalian cells. The seqFRIES strategy does not require cell fixation or permeabilization. The dye molecules are permeable to the living cells membranes and exhibit fast cellular fluorescence activation and deactivation kinetics with their cognate RNA aptamers. The optimized high efficiency labeling-and-stripping procedure allowed us to image four fluorogenic RNA/dye pairs in ∼20 minutes. After repeated fluorescence imaging in the same individual cells, our results also validated the ability of seqFRIES in the single-cell analysis of different target analytes. These single-cell analysis can potentially provide valuable information regarding the cell-to-cell variations in their molecular signatures, which are usually masked in bulk analysis.

Genetically encoded fluorogenic RNA-based biosensors have become popular tools for the detection and monitoring of various cellular targets. Compared to small molecule- or fluorescent protein-based sensors, the unique advantages of these fluorogenic RNA-based sensing platforms are their high programmability and generalizability. Fluorogenic RNAs can function as imaging tags or via sequence-specific hybridizations for the cellular tracking of different mRNA and non-coding RNA analytes (4,5,19). RNA aptamers can also be selected for a wide range of cellular targets, including a variety of proteins, metabolites, ions and signaling molecules (54,55). These specific aptamers can be further modularly engineered into targeted fluorescent sensors and incorporated in the seqFRIES system. The circularized RNA design allows multiple RNA sensors, flanked by self-cleaving ribozymes, to be cloned under one promoter in a single plasmid (Supplementary Figure S21 and Table S4). It is worth noting that the exceptional stability of circular RNAs provides long-lasting cellular fluorescence signals, but also may raise potential concerns of target sequestering by the RNA sensors. To alleviate this concern, the RNA sensor level may need to be tuned down for low-abundance analytes by adding less transfection complex or using a weaker promoter for the RNA transcription. The versatility of seqFRIES makes it easily applied for the future detection of a large number of targets at the single-cell level.

Compared with other genetically encoded systems for multiplexed imaging in living cells (Supplementary Table S2), seqFRIES is a unique sequential imaging approach to solve the current spectral overlap problem. Meanwhile, the insert size of seqFRIES is much smaller, making the molecular cloning and delivery quite easier. The seqFRIES platform is also highly programmable, covering a wide range of cellular target analytes. Even though the current imaging protocol requires several operation steps, after further optimization, the sequential imaging and stripping processes can likely be fully automated. On the other hand, seqFRIES may not beat all other approaches in terms of its current multiplexing capability. To potentially allow this seqFRIES system to detect more than four targets, the identification of additional orthogonal fluorogenic RNA/dye pairs are still required to expand the current multiplexing platform. Indeed, many fluorogenic RNA/dye pairs have been discovered recently, such as SiRA/SiR and o-Coral/Gemini-561 (56,57), which can be potentially orthogonal to our current FR/dye pairs.

Furthermore, by using an internal reference FR/dye pair to normalize cell-to-cell variations in RNA expression levels, we have partially validated the possibility of applying seqFRIES in quantitative cellular analysis. However, it is worth noting that even with an internal reference and ratiometric fluorescence analysis, the heterogeneity of RNA sensor concentrations across individual cells can still affect the sensitivity and dynamic range of seqFRIES and influence its usage for target quantification. To potentially mitigate this effect, the absolute RNA sensor concentration in each cell may need to be first identified based on the fluorescence intensities of the reference FR/dye pair or a FISH probe (53). Meanwhile, corresponding calibration curves should be generated under different RNA sensor levels in order to determine the cellular target concentrations. With further optimization and expansion of the orthogonal FR/dye pairs, we expect that the seqFRIES platform can be broadly applied for multiplexed live-cell imaging and quantitative molecular profiling to reveal molecular correlations and to provide a better understanding of mysterious cellular processes.

Data availability

The data supporting the findings of this study are available upon request to the corresponding author.

Supplementary data

Supplementary Data are available at NAR Online.

Funding

Chan Zuckerberg Initiative Dynamic Imaging program [2023-321170 to M.Y.]; NSF CAREER award [1846152 to M.Y.]; Alfred P. Sloan Research Fellowship (to M.Y.); Camille Dreyfus Teacher-Scholar Award (to M.Y.); UMass Amherst start-up grant (to M.Y.); NIH Traineeship [T32GM139789 to R.Z.]; SLAS Graduate Education Fellowship (to L.M.); Paul Hatheway Terry Scholarship (to Z.S. and Y.B.). Funding for open access charge: Camille Dreyfus Teacher-Scholar Award.

Conflict of interest statement. None declared.