Visualizing intracellular glycine with two-dye and single-dye ratiometric RNA-based sensors

Abstract

Glycine is an important metabolite and cell signal in diverse organisms, yet tools to visualize intracellular glycine dynamics have not been developed. In this study, diverse and bright RNA-based glycine biosensors were developed by fusing the architecturally complex glycine riboswitch with Broccoli class fluorogenic aptamers. The brightest sensor with the highest activation, glyS, and its two-dye ratiometric counterpart, Pepper-glyS, allowed for visualization of a drug-induced accumulation of endogenous glycine in live Escherichia colicells. However, a general limitation of two-dye orthogonal aptamer pairs is that differences in dye properties may prevent accurate quantitation of cellular targets. Here, a novel Golden Broccoli aptamer was developed that is readily resolved from Red Broccoli by spectral unmixing methods, even though both aptamers bind the same dye, DFHO. This enabled generation of the first single-dye ratiometric sensor, which detects glycine both in vitro and in live E. coli cells.

Graphical Abstract

Graphical Abstract.

Introduction

Glycine, the simplest amino acid, is a vital metabolite and cell signal in many organisms. In mammals, it serves as a neurotransmitter, influencing reflexes, memory, and brain development [1, 2]. In plants, glycine is a root exudate, facilitating symbiosis and communication with soil bacteria [3, 4]. Additionally, glycine holds several roles in bacteria. Glycine promotes virulence and spore formation in the pathogen Clostridioides difficile[5], while also potentiating kanamycin’s effectiveness against the fish pathogen Edwardsiella piscicida, demonstrating dual roles in infection [6]. Furthermore, glycine catabolism activity is reduced in serum-resistant Escherichia coli, while the ability of serum to eliminate this pathogen was restored by exogenous glycine [7]. Thus, a low abundance of glycine may be a biomarker for bacterial serum resistance. Beyond the implications for infection, both glycine and glycine riboswitches have demonstrated utility in bacterial metabolic engineering applications [8].

Given glycine’s diverse roles, detection methods are essential. Standard liquid chromatography-mass spectrometry is low-throughput and requires lysing bulk cells for intracellular detection [9]. A glycine-oxidase-based electrochemical sensor is high-throughput for extracellular or intradermal detection, though not intracellular measurement [10, 11]. Optical tools include riboswitch-based fluorescent turn-OFF sensors for in vitro detection [12] and a fluorescence resonance energy transfer-based sensor for monitoring extracellular glycine in brain slices and neuronal cultures [13]. However, none of these methods have demonstrated the ability to detect intracellular glycine dynamics in single cells. While riboswitch-based reporters that measure high glycine have been applied for metabolic engineering, they lack spatiotemporal resolution [8]. Thus, tools for visualizing single-cell glycine dynamics are still needed.

Here we have designed fluorescent turn-ON glycine biosensors by leveraging inter-aptamer interactions within glycine-binding riboswitches to stabilize a light-up RNA aptamer that then activates the fluorescence of a dye. These sensors displayed a broad range of affinities, high selectivity, and up to 23-fold turn-on. The best sensor, glyS, detects drug-induced endogenous glycine accumulation in live E. coli cells. Ratiometric imaging of glycine was enabled by Pepper-glyS, which combines aptamer-dye pairs Broccoli-DFHBI-1T [14] and Pepper-HBC620 [15]. These bright aptamer-dye pairs are considered “orthogonal” because each aptamer activates the fluorescence of a unique dye of a different color. However, a drawback of using orthogonal aptamer pairs is differences in their ion dependency [14–18], bio-orthogonality [19, 20], multimerization [21], dye solubility, and cellular dye permeability [22, 23]. While variations in these properties can be accounted for with careful experimental design or calibration, systems employing orthogonal aptamers will be more tricky to transfer and employ across different experimental conditions or cell types. Thus, these differences may confound the reliability of analyte quantitation.

In a novel approach, we reveal that a single dye can be spectrally tuned by binding to different, though highly structurally similar, RNAs. Inspired by the development of fluorescent protein YFP from GFP through a mutation that alters π-stacking to the chromophore [24], we created the Golden Broccoli aptamer by mutating a base triple that π-stacks with the dyes DFHO and OBI in Orange Broccoli [25]. This allowed spectral resolution from Red Broccoli [25] both in vitro and in live cells to generate the first single-dye ratiometric (SDR) sensor.

Materials and methods

Reagents

All DNA oligonucleotides for RNA constructs or cloning were purchased from the University of Utah HSC Core Facility or Integrated DNA Technologies (Chicago, IL). DFHBI was synthesized following previously described protocols [26, 27] and stored as an 18 mM stock at −20°C in DMSO. DFHBI-1T was purchased from Sigma–Aldrich (St. Louis, MO) and stored as a 50 mM stock at −20°C in DMSO. DFHO and OBI were purchased from Lucerna Technologies (Brooklyn, NY) and stored as 35 and 25 mM stocks at −20°C in DMSO, respectively.

Biological resources

Bacterial strains studied include BL21(DE3) Star E. coli(UC Berkeley MacroLab) and wild type 1012 Bacillus subtilis(MoBiTec Molecular Biotechnology). XL1-Blue E. coli (UC Berkeley MacroLab) were used for cloning. The pET31b plasmid was used for expression in BL21(DE3) Star cells, while the pHT253 plasmid was used for expression in B. subtilis.

Statistical analyses

For evaluating statistical significance, unpaired Student’s two-tailed t-tests were used with at least three technical or biological replicates, as indicated in figure legends. All data plotted are an average with error bars indicating the standard deviation.

In vitro transcription

For in vitro transcription, DNA templates were amplified with Phusion DNA Polymerase (UC Berkeley MacroLab) from Ultramer oligonucleotides for phylogenetic screening or sequence-confirmed plasmids for preliminary sensor screening or characterization experiments. The T7 promoter was contained within the template or appended via an overhang on 5′ end of the forward primer. The PCR product was purified and concentrated using the QIAquick PCR Purification Kit (Qiagen).

RNA was transcribed from the purified PCR product as previously described [28]. Briefly, ∼1 μg of DNA was incubated at 37°C for 4 h with homemade T7 RNA polymerase in 40 mM Tris–HCl, pH 8.0, 6 mM MgCl₂, 2 mM spermidine, 10 mM DTT,1 U of inorganic pyrophosphate, and 2 mM rNTPs. The RNAs for phylogenetic screening were purified using a spin-column format Zymo RNA Clean & Concentrator-25 kit, while RNAs for preliminary sensor screening or characterization were purified by denaturing (7.5 M urea) 6% polyacrylamide gel electrophoresis (PAGE). PAGE-purified RNA was visualized by UV shadowing and extracted from gel pieces using Crush Soak buffer (10 mM Tris–HCl, pH 7.5, 200 mM NaCl, and 1 mM ethylenediaminetetraacetic acid, pH 8.0). Extracted RNAs were ethanol precipitated, dried, and resuspended in ddH₂O. To accurately quantify the RNA, a neutral pH thermal hydrolysis assay was used to remove the hypochromic effect from RNA secondary structure [29, 30].

In vitro fluorescence assays

General protocol

In vitro fluorescence assays were performed in binding buffer containing 40 mM HEPES (pH adjusted to 7.5 with KOH), 125 mM KCl, and varying concentrations of MgCl₂, dye, RNA, and ligand (when applicable). For initial glycine sensor screening and characterization, a final concentration of 10 μM DFHBI (with ex/em: 448/506) was consistently used. For single dye ratiometric (SDR) sensing experiments and Broccoli variant analysis, final concentrations of 0.5, 2, or 5 μM DFHO or OBI were used (variable ex/em as indicated), except for binding affinity analysis (specified below).

Before all assays, RNA (10×) was renatured by heating at 70°C for 3 min followed by 15 min of slowly cooling to room temperature in binding buffer. For the fluorescence assays, the dye was added to the reaction mixture in a Greiner Bio-One 384-well black plate containing binding buffer, RNA, and ligand (as indicated) for a total reaction volume of 30 μl. After incubation in the plate for the duration and temperature described in the figure legends, the fluorescence emission was measured by the SpectraMax i3x plate reader (Molecular Devices) and averaged across technical replicates. For initial glycine sensor screening and characterization, the fluorescence turn-on of glycine sensors was calculated by dividing fluorescence in the presence of glycine by fluorescence in the absence of glycine. For SDR glycine sensing experiments, fold activation was measured by dividing the RBr/GBr fluorescence ratio in the presence of glycine by the RBr/GBr ratio in the absence of glycine (after spillover correction to fully resolve aptamer signals as indicated, detailed under “In vitro aptamer spillover correction”).

Binding affinity analyses

The binding affinities of glycine biosensors were measured with 100 nM RNA in binding buffer containing 10 μM DFHBI, 40 mM HEPES (pH 7.5), 125 mM KCl, and 3 mM MgCl₂. The dissociation constant (K_d) for each biosensor was calculated from the concentration-dependent fluorescence curves by fitting the normalized fluorescence intensity (FN) versus log of glycine concentration plot to a nonlinear regression [log(agonist) versus response (three parameter)] using GraphPad Prism 9 software. F_N was calculated as (F_i – F_min)/(F_max – F_min), where F_i is fluorescence intensity at each ligand concentration, F_min is fluorescence intensity with the lowest amount of ligand, and F_max is fluorescence intensity at the saturation point.

The binding affinities of Broccoli aptamer variants were measured with 100 nM RNA in the same binding buffer described above but instead of DFHBI, with the following DFHO concentrations: 2, 10, 20, 100, and 200 nM; 1, 2, 10, and 20 μM. Dye background fluorescence was subtracted at each dye concentration before normalization of fluorescence intensity to ensure fluorescence was due solely to dye binding to RNA. The dissociation constant for each aptamer was then calculated in an analogous way as described above, except with DFHO replacing glycine as the ligand.

In vitro fluorescence turn-on kinetics

In vitro fluorescence turn-on assays were adapted for measuring glycine biosensor kinetics as described previously [28]. Briefly, the glycine biosensor (100 nM) was renatured in binding buffer [40 mM HEPES (pH 7.5), 125 mM KCl, 3 mM MgCl₂] and pre-equilibrated with a final concentration of 10 μM DFHBI for 15 min at 37°C. Next, a final concentration of 10 mM glycine was added using the automated injector module of the SpectraMax i3x plate reader (Molecular Devices) at the 90 sec mark of the 30–60 min measurement period. Measurements were taken every 1.5 sec. Due to bottom reads being used with the plate reader in injection mode, these 100 μl reactions were performed in CORNING Costar 96-well black with clear flat bottom plates.

Excitation and emission spectra

The general protocol was followed for this method, except with an excess final concentration of RNA (10 μM) relative to the dye (0.5 μM) to ensure fluorescence was due to bound dye. Buffer background was subtracted from each measured point before normalization.

In vitro aptamer spillover correction

To fully separate Red and Golden Broccoli fluorescence signals in SDR constructs in vitro, tRNA-scaffolded Red Broccoli and Golden Broccoli single-color controls were measured under the same conditions for which SDR constructs were measured: 300 nM RNA in binding buffer containing 40 mM HEPES (pH 7.5), 125 mM KCl, 3 mM MgCl₂, and 5 μM DFHO or OBI. Spillover correction was used to assign Red Broccoli fluorescence alone to a red TxRed channel (ex/em: 561/610, bandpass: 25) and Golden Broccoli fluorescence alone to a green FITC channel (ex/em: 488/530, bandpass: 25) as described below.

For DFHO: under our conditions, only Red Broccoli showed spillover (68%) into the opposite aptamer’s fluorescence channel. Therefore, the TxRed channel represented solely Red Broccoli’s fluorescence, and only a one-way spillover correction was necessary for the determination of FITC channel fluorescence [31] due solely to Golden Broccoli. This correction to raw FITC channel fluorescence was performed as indicated with the following equation:

where F_FITC and F_TxRed are the raw fluorescence intensity values in each channel, S is the spillover coefficient (in our case, S = 0.68 due to 68% spillover), and ^GBrF_FITC is the spillover-corrected FITC channel fluorescence intensity that corresponds to Golden Broccoli’s signal alone.

For OBI: both Red Broccoli and Golden Broccoli single-color controls showed spillover into the opposite aptamer’s fluorescence channel. Under our conditions, 4.2% of GBr’s green fluorescence bled into the TxRed channel, while 22% of RBr’s fluorescence bled into the FITC channel. Therefore, a two-way spillover correction was necessary for the determination of FITC channel fluorescence due solely to Golden Broccoli (^GBrF_FITC), and TxRed channel fluorescence due solely to Red Broccoli (^RBrF_TxRed). This correction was performed as indicated with the following equations, adapted from Equations 1.14.11 and 1.14.12 derived in a guide to mathematical two-way compensation [31]:

where F_FITC and F_TxRed are the raw fluorescence intensity values in each channel, ^RBrS_FITC is the spillover coefficient from Red Broccoli into the FITC channel (in our case, 0.22), and ^GBrS_TxRed is the spillover coefficient from Golden Broccoli into the TxRed channel (in our case, 0.042).

RNA expression in E. coli

E. coli BL21 (DE3) Star cells were transformed with 10 ng of pET31b plasmid encoding tRNA-scaffolded aptamer or biosensor construct. The transformed cells were plated on LB/carbenicillin plates (Carb: 50 μg/ml). At least three single colonies representing at least three bio-replicates for each construct were separately inoculated in 3 ml of non-inducing (NI) media containing 50 μg/ml carbenicillin and grown for 24 h with shaking at 250 rpm at 37°C. Next, 30 μl of NI cultures were added to 3 ml of ZYP-5052 autoinduction media (AI) containing 50 μg/ml carbenicillin and grown for 17 h at 37°C in an incubator with shaking at 250 rpm to express the RNA constructs.

Live E. coli conventional flow cytometry

General flow cytometry analysis

After assay setup detailed in below sections, cellular fluorescence of at least 30 000 cells was measured on the Cytoflex S in the green FITC (ex/em: 488/530, bandpass: 30) and/or the red TxRed (ex/em: 561/610, bandpass: 15) fluorescence channels as specified in the figure legends. Using FlowJo software, forward and side scatter gating was used to select single bacterial cells. At this point, for SDR constructs, compensation was performed to correct spillover for full resolution of aptamers as described in further detail under “Compensation to resolve SDR aptamers in conventional flow cytometry.” The median fluorescence intensity (MFI) of the gated population for each bio-replicate was determined.

Intensiometric sensor (glyS) overnight treatment assays

The specified final concentration of glycine or cysteamine in the figure was added to the AI cultures for the 17 h RNA expression period. Cells were then diluted 100× into 1× phosphate buffered saline (PBS) with a final concentration of 50 μM of DFHBI-1T and incubated for 30 min before flow cytometry analysis.

Intensiometric sensor (glyS) glycine accumulation time-course assays

The final concentration of glycine or cysteamine specified in the figure was added to the AI cultures after RNA expression, for the specified amount of time on the benchtop at room temperature. After this initial incubation, cells from the AI culture were then diluted 100× into 1× PBS with final concentration of 50 μM of DFHBI-1T and incubated for an additional 30 min before flow cytometry analysis. Timepoint t = 0 represents dilution into PBS with dye immediately after glycine or cysteamine addition to the AI culture, followed by the dye incubation period. After flow cytometry analysis, ΔMFI/MFI was calculated at each timepoint as (MFI_t – MFI₀)/MFI₀, where MFI_t represents the MFI at variable time t, and MFI₀ represents the MFI at t = 0.

Two-dye ratiometric sensor (pepper-glyS) overnight treatment assays

The specified final concentration of glycine or cysteamine in the figure was added to the AI cultures for the 17 h RNA expression period. Cells were then diluted 100× into 1× PBS with final concentrations of 50 μM of DFHBI-1T and 200 nM HBC620 and incubated for 1 h before flow cytometry analysis. After flow cytometry analysis, G/R was calculated as the ratio of green fluorescence due to Broccoli divided by the red fluorescence due to Pepper.

Two-dye ratiometric sensor (pepper-glyS) glycine accumulation time-course assays

The final concentration of glycine or cysteamine specified in the figure was added to the AI cultures after RNA expression for the specified amount of time on the benchtop at room temperature. After this initial incubation, cells from the AI culture were then diluted 100× into 1× PBS with final concentrations of 50 μM of DFHBI-1T and 200 nM HBC620 and incubated for an additional 1 h before flow cytometry analysis. Timepoint t = 0 represents dilution into PBS with dye immediately after glycine or cysteamine addition to the AI culture, followed by the dye incubation period. After flow cytometry analysis, G/R was calculated as the ratio of green fluorescence due to Broccoli divided by the red fluorescence due to Pepper.

Cell viability assay

AI cultures were diluted into 100× into 1× PBS after tRNA-Broccoli expression with final concentrations of 50 μM of DFHBI-1T and 2.5 μg/ml propidium iodide (PI) for 1 h before flow cytometry analysis.

Time-course assays for Broccoli aptamer variants and Pepper aptamer

AI cultures were diluted into 100× into 1× PBS after RNA expression. Final concentrations of either 50 μM of DFHBI-1T, 200 nM of HBC620, or 200 μM of DFHO were used, as indicated in the figure legend. These cell solutions were analyzed on the flow cytometer after the amount of incubation time in PBS and dye specified within the figure, with t = −5 min representing MFI without dye addition, and t = 0 representing fluorescence measured immediately after dye addition.

Endpoint assays for Broccoli aptamer variants and Corn aptamer

AI cultures were diluted 100× into 1× PBS after RNA expression, with final concentrations of 50 or 200 μM DFHO or OBI, and incubated for 45 min or 1 h (as indicated) before flow cytometry analysis, as specified in the figure legend.

Single dye ratiometric (SDR glyS) sensor overnight treatment assays

The specified final concentration of glycine or cysteamine in the figure was added to the AI cultures for the 17 h RNA expression period. Cells were then diluted 100× into 1× PBS with a final concentration of 200 μM DFHO and incubated for 45 min or 1 h (as indicated) before flow cytometry analysis. After flow cytometry analysis, RBr/GBr was calculated as the ratio of spillover-corrected fluorescence due to Red Broccoli divided by the spillover-corrected green fluorescence due to Golden Broccoli.

Compensation to resolve SDR aptamers in conventional flow cytometry

In FlowJo, after gating to select single bacterial cells, gated cell-only negative controls as well as fluorescence-positive gated Red Broccoli-DFHO and Golden Broccoli-DFHO single-color controls were moved to the Compensation Group. In the toolbar, the Compensation function was selected. Only FITC and TxRed channels were included as Parameters, with Golden Broccoli selected as the Sample for the FITC channel and Red Broccoli selected as the Sample for the TxRed channel. For the Negative Sample, cell only controls (no dye added) were selected. For the Positive Sample, the fluorescence-positive gate previously made was selected. To generate the compensation matrix, the Calculate SSM function was used. To apply the compensation matrix to samples containing both Red and Golden Broccolis, the Apply to Group function was selected, followed by “All Samples.” At this point, the spillover-corrected median fluorescence intensities from each aptamer in each sample were determined by selecting Comp-FITC or Comp-TxRed channels for analysis.

Live E. coli spectral flow cytometry

The specified final concentration of glycine in the figure was added to the AI cultures for the 17 h RNA expression period. Cells were then diluted 100× into 1× PBS with a final concentration of 200 μM DFHO and incubated for 1 h before flow cytometry analysis. First, at least 50 000 cells in an unstained cell control sample and single-color control RBr and GBr samples were measured across all 5 lasers and 64 detectors on the Cytek Aurora. Next, the Unmix function in the instrument software was used to select these control samples for downstream unmixing of SDR construct fluorescence. Finally, the fluorescence of at least 50 000 cells expressing the SDR constructs was measured, and the automatically generated unmixed data were exported for analysis in FlowJo software. Using FlowJo, forward and side scatter gating was used to select single bacterial cells. After obtaining the unmixed MFI values from gated events, RBr/GBr was calculated as the ratio of unmixed fluorescence due to Red Broccoli divided by the unmixed fluorescence due to Golden Broccoli.

Live B. subtilis RNA expression and flow cytometry

Wild-type 1012 B. subtilis cells from MoBiTec Molecular Biotechnology were transformed with 1 μg of pHT253 plasmid encoding tRNA-scaffolded Red Broccoli. This was performed according to manufacturer instructions for competent B. subtilis transformation (https://www.mobitec.com/products/vector-systems/bacillus-expression-systems/pbs013/bacillus-subtilis-expression-vector-pht253-pgrac100-type-his-tag), with a few modifications. In step two, 1 ml of competent cells was inoculated into 3 ml LS media instead of 20 ml. In step eight, bacteria were plated on LB/chloramphenicol (Chlor: 5 μg/ml) rather than 2× YT medium.

Four single colonies were separately inoculated into 3 ml of LB media containing 5 μg/ml chloramphenicol and grown for 24 h with shaking at 250 rpm at 37°C. Cultures were diluted into 3 ml of fresh LB media containing 5 μg/ml chloramphenicol to an OD600 of 0.15. These diluted cultures were grown for about 2 h with shaking at 250 rpm at 37°C, until the OD600 reached 0.6–0.8. At this point, IPTG was added to a final concentration of 1 mM to induce expression of Red Broccoli, and the cultures were incubated for 2 h with shaking at 250 rpm at 37°C.

For the endpoint assay comparing Red Broccoli fluorescence with DFHO or OBI, these cultures were diluted 100× into 1× PBS after the expression period with final concentrations of 0 or 200 μM of DFHO or OBI, and were incubated for 1 h before flow cytometry analysis, as specified in the figure legend. During flow cytometry analysis, cellular fluorescence of at least 30 000 cells was measured on the Cytoflex S in the TxRed fluorescence channel (ex/em: 561/610, bandpass:15). Using FlowJo software, forward and side scatter gating was used to select single bacterial cells. The MFI of the gated population for each bio-replicate was determined.

Fluorescence microscopy

General protocol

For microscopy studies, RNA construct expression was induced in E. coli BL21 (DE3) Star cells with the same NI/AI protocol described earlier. Once AI cultures had incubated for 17 h, 60 μl aliquots of AI culture were centrifuged at 4000 rpm for 3 min in 1.5 ml tubes. Once the AI media was decanted, the pellets were resuspended in 200 μl M9 media and again centrifuged at 4000 rpm for 3 min. The M9 media was decanted, and the pellets were then resuspended with 200 μl of M9 media containing final dye concentrations as described below. The cells were allowed to incubate with the dye(s) for 1 h. Next, 4 μl aliquots of the cell solutions were added to agarose pads (1%, in M9 media) on a glass microscope slide. The cells were evenly spread on the agarose pads by rotating the glass slide in a circular motion until the cell solutions were fully absorbed and dried (about 3 min). Glass coverslips were placed on top of the agarose pads before microscopy analysis described below.

Two-dye ratiometric sensor (Pepper-glyS) cysteamine treatment assay

A final concentration of 2.5 mM cysteamine was added to the AI cultures as indicated for the 17 h RNA expression period. When preparing cells for microscopy as described above, final concentrations of 50 μM of DFHBI-1T and 200 nM HBC620 were incubated with the cell solution for 1 h. After immobilizing cells on the agarose pads and adding coverslips, images were acquired on the Olympus IX71 microscope with a 100× oil immersion objective at 5% laser power, with 150 ms exposure time at 494 nm for green fluorescence measurements with the FITC filter (em: 518 for Broccoli imaging) and 150 ms exposure time at 550 nm for red fluorescence measurements with the mCherry filter (em: 565 for Pepper imaging).

In ImageJ, to enlarge a region of cells for the images shown, 50 × 50 μm squares were selected on the red Pepper fluorescence image. This same selected area was applied to the green Broccoli fluorescence image using the Restore Selection function, and the two images were cropped. Merged images were obtained using the Merge Channels function.

For fluorescence quantitation, the ImageJ plugin MicrobeJ was used. In MicrobeJ, the red Pepper fluorescence image was defined as Channel 1, and the green Broccoli fluorescence image was defined as Channel 2. MicrobeJ was then used to count at least 150 single cells in an overlay based on their constitutive fluorescence in the red Channel 1 image, and simultaneously, this exact overlay was also applied to the green Channel 2 image. The fluorescence of each cell from each channel was then measured using the Measure Overlay function in ImageJ, and the fluorescence values from each cell were processed as a green/red ratio.

Red and Golden Broccoli images

When preparing cells for microscopy as described above, a final concentration of 200 μM of DFHO was incubated with the cell solution for 1 h. After immobilizing cells on the agarose pads and adding coverslips, images were acquired on the Leica TCS SP8 X White Light Laser (WLL) Confocal Microscope with a 60× oil immersion objective. For yellow channel measurements, the argon laser ran at 25% power, with 2% of that power sent to the sample with excitation at 488 nm, and emission was collected between 540 and 580 nm. For red channel measurements, the WLL ran at 70% power, with 15% of that power sent to the sample with excitation at 530 nm, and emission was collected between 595 and 645 nm. A pixel dwell time of 1.2 μs was used with 4× line averaging.

In ImageJ, to enlarge a region of cells for the images shown, 50 × 50 μm squares were selected on the yellow fluorescence image. This same selected area was applied to the red fluorescence image using the Restore Selection function, and the images were cropped.

For fluorescence quantitation of the raw images, the ImageJ plugin MicrobeJ was used. In MicrobeJ, a saturated brightness and contrast-enhanced yellow fluorescence image was defined as Channel 1 to detect even dim cells by cell shape, the yellow fluorescence image with brightness and contrast as shown in Supplementary Fig. S24A was defined as Channel 2, and the red fluorescence image with brightness and contrast as shown in Supplementary Fig. S24A was defined as Channel 3. MicrobeJ was then used to count at least 750 single cells in an overlay based on their constitutive fluorescence in the Channel 1 image, and simultaneously, this exact overlay was also applied to Channels 2 and 3 images. The fluorescence of each cell from Channels 2 and 3 (yellow and red channels) was then measured using the Measure Overlay function in ImageJ. Unmixing of GBr and RBr signals, thus assignment of their signals into respective yellow and red channels, was then carried out when indicated, described below under “Unmixing of SDR Microscopy Images.”

Single-dye ratiometric sensor (SDR glyS) glycine treatment assay

A final concentration of 10 mM glycine was added to the AI cultures as indicated for the 17 h RNA expression period. When preparing cells for microscopy as described above, a final concentration of 200 μM of DFHO was incubated with the cell solution for 1 h. After immobilizing cells on the agarose pads and adding coverslips, images were acquired on the Leica TCS SP8 X WLL Confocal Microscope with a 60× oil immersion objective. For yellow channel measurements, the argon laser ran at 25% power, with 2% of that power sent to the sample with excitation at 488 nm, and emission was collected between 540 and 580 nm. For red channel measurements, the WLL ran at 70% power, with 15% of that power sent to the sample with excitation at 530 nm, and emission was collected between 595 and 645 nm. A pixel dwell time of 1.2 μs was used with 4× line averaging.

In ImageJ, to enlarge a region of cells for the images shown, 50 × 50 μm squares were selected on the DIC image. This same selected area was applied to the yellow and red fluorescence images using the Restore Selection function, and the images were cropped. Unmixing of GBr and RBr signals, thus assignment of their signals into respective yellow and red channels, was carried out as described below under “Unmixing of SDR Microscopy Images.” The Subtract Background function was used to subtract background from fluorescence images using a rolling ball radius of 50 pixels. Merged images were obtained using the Merge Channels function.

For fluorescence quantitation of the unmixed images, the ImageJ plugin MicrobeJ was used. In MicrobeJ, a saturated brightness and contrast-enhanced yellow GBr fluorescence image was defined as Channel 1 (for optimal cell shape detection of even dim cells), the yellow GBr fluorescence image with brightness and contrast as shown in Fig. 6D was defined as Channel 2, and the red RBr fluorescence image with brightness and contrast as shown in Fig. 6D was defined as Channel 3. MicrobeJ was then used to count at least 350 single cells in an overlay based on their constitutive fluorescence in the Channel 1 image, and simultaneously, this exact overlay was also applied to Channels 2 and 3 images. The fluorescence of each cell from Channels 2 and 3 was then measured using the Measure Overlay function in ImageJ, and the fluorescence from each cell was processed as a GBr/RBr fluorescence ratio.

Figure 6.

SDR sensing of glycine. (A) Design of the SDR glyS construct. (B) Unmixed in vitro fluorescence of SDR glyS with different dyes; shown are averages with standard deviation from three technical replicates with different dyes at 5 μM concentration. Reaction components were incubated for 2 h before measurement. (C) Unmixed spectral flow cytometry data of SDR constructs with different overnight treatments. MFI values in RBr (middle) and GBr (bottom) channels were analyzed from 50 000 cells by flow cytometry and processed as a ratio (top), with averages and standard deviations shown from five biological replicates. (D) Fluorescence microscopy of SDR glyS with different overnight treatments in GBr (yellow) and RBr (red) channels. Scale bars, 10 μm. (E) Violin plot of microscopy quantitation representing the distribution of RBr/GBr fluorescence ratios from at least 350 cells. The solid line represents the median and the dashed lines represent quartiles. For these live cell experiments, cells were incubated with 200 μM DFHO for 1 h before measurement. P-values are ** < .01 and **** < .0001 using an unpaired student’s two-tailed t-test.

Unmixing of microscopy images for SDR glyS sensor

Along with the samples to be unmixed, same-day single-color control microscopy images of RBr and GBr alone were measured and quantified as described above under “Red and Golden Broccoli Images.” To create an unmixing matrix, first a matrix composed of the average fluorescence of GBr and RBr in both yellow and red fluorescence channels was created:

where RBr_Y and RBr_R are the average cellular fluorescence intensities of RBr in the yellow and red channels, respectively, and GBr_Y and GBr_R are the average cellular fluorescence intensities of GBr in the yellow and red channels, respectively. To normalize the matrix, these average fluorescence values of each aptamer in each channel were divided by the total fluorescence of each aptamer between both channels:

Finally, using the Inverse Matrix function in Wolfram Alpha, this normalized matrix was inverted to yield the unmixing matrix:

To obtain an unmixed red channel image of RBr fluorescence with GBr fluorescence removed, raw yellow and red channel fluorescence images for a given sample were opened in ImageJ. Using the ImageJ plugin CLIJ2, the Add Two Weighted Images on GPU function was used. In general, this function generates a new image by adding weighted images using the following equation:

In our case, the raw yellow channel image was selected as Summand1, and the raw red channel image was selected as Summand2. Based on the unmixing matrix, Factor1 was set to U₁, Factor2 was set to U₂, and the unmixed red channel image was obtained.

To obtain an unmixed yellow channel image of GBr fluorescence with RBr fluorescence removed, the above process in CLIJ2 was repeated, except Factor1 was set to U₃ and Factor2 was set to U₄. In general, linear unmixing from an inverted matrix based on single-color control images has been described previously and used to resolve fluorophores [32, 33].

Results

Design and screening of glycine biosensors

The complex inter-aptamer interactions found in glycine riboswitches informed our biosensor design [34–36]. Tandem glycine riboswitches have two glycine-binding aptamers that interact upon binding glycine via stacking and stabilization of the P1 and P3 stems of opposite aptamers (Supplementary Fig. S1A) [37, 38). In isolation, the second glycine-binding aptamer from a tandem riboswitch dimerizes in a way that mimics the full-length tandem riboswitch [38–40]. Type I and II singlet glycine riboswitches have only a single glycine-binding aptamer that is followed or preceded by a vestigial “ghost aptamer” (Supplementary Fig. S1B) thought to form similar stabilizing interaction with the aptamer P3 stem [34, 35, 41].

To ensure the glycine riboswitches used in preliminary designs were binding-competent, the initial library was based solely upon riboswitches previously validated for glycine binding in literature. Tandem riboswitches were sourced from Vibrio cholera (Vc), Fusobacterium nucleatum (Fn), and Bacillus subtilis (Bs) because of demonstrated glycine binding in an x-ray crystal structure for Fn [37], by in-line probing for Vc [36], or by transcription termination assays for Bs [36, 42]. Monomeric variants made of only the second glycine binding aptamer were sourced from Vc as well, as the monomer could still bind glycine as shown in x-ray crystallography [39], equilibrium dialysis [38], and in-line probing assays [40]. A natural type I singlet riboswitch from Listeria monocytogenes (Lm) and a type II singlet riboswitch from Desulfitobacterium hafniense (Dh) were shown to bind glycine in both equilibrium dialysis [34] and transcription termination assays [35].

To harness the inter-aptamer interactions of these riboswitches for novel glycine biosensors, a circularly permuted Spinach2 aptamer [43] was connected to the glycine riboswitch at the P3b stem for tandem-P3 (T-P3), aptamer 2-only-P3 (A2-P3), type I singlet-P3 (S1-P3), and type II singlet-P3 (S2-P3) designs (Fig. 1) [37, 42]. Alternatively, circularly permuted Spinach2 replaced segments of the ghost aptamers for type I singlet-ghost aptamer (S1-G) and type II singlet-ghost aptamer (S2-G) designs, using either natural singlet riboswitches or artificial singlets derived from tandem riboswitches in the case of Bs-c-f, Vc-n-s, and Fn-c-h sensors (Fig. 1). These designs are the first to our knowledge to employ inter-aptamer stacking interactions to stabilize the fluorogenic aptamer. At least two truncations of the riboswitch stem were tested per design.

Figure 1.

Design and screening of glycine biosensors. In vitro fluorescence of 42 initial glycine biosensors, with mean and standard deviation from three technical replicates. Reaction components were incubated for 30 min with 10 μM DFHBI before measurement. The highest fluorescence turn-on is indicated in red. Glycine biosensor architectures are indicated above the graph with color labels corresponding to designs. All RNA sequences from biosensor screening in vitro are indicated in Supplementary Table S1.

A preliminary screen of the 42 total biosensor candidates using a fluorescence plate reader revealed that 7 constructs exhibited fluorescence turn-on in response to glycine (Fig. 1). The most successful design class, S2-P3 with 2/2 hits, was based on a riboswitch from Desulfitobacterium hafniense[34]. This design included the Dh-b construct, which showed a 5.2-fold turn-on response to glycine. The second most successful design, S1-G, yielded 3/7 hits with up to a 1.7-fold turn-on, despite being based on a completely different riboswitch topology and Spinach2 linkage. For the Dh-b sensor, replacing Spinach2 (Sp2) with two truncations of Broccoli (Tr1 and Tr2) [44] led to improved brightness and fold turn-on at lower Mg²⁺ conditions for Tr1 (Fig. 2A), which was renamed “Dh” moving forward.

Figure 2.

Optimization and characterization of glycine biosensors. (A) In vitro fluorescence of Dh glycine biosensor containing Spinach2 or truncations of Broccoli after 1 h incubation. (B) Apparent dissociation constants (K_d) of glycine for biosensors after 2 h incubation. (C) In vitro fluorescence of Spn-b (glyS) with no ligand, glycine, or glycine analogs after 2 h incubation. (D) In vitro kinetic plots of biosensors after glycine addition to measure half-maximal response time (t_1/2). Above experiments were measured with 10 μM DFHBI; data shown are the mean of three technical replicates with error as standard deviation, except for Dh-b-Br-Tr2 + 1 mM glycine, which had two replicates. (E) Secondary structure model of Spn-b (glyS) and the binding pocket from the Fusobacterium nucleatum glycine riboswitch x-ray crystal structure (PDB: 3P49). The riboswitch sequence is shown in purple and Broccoli in green, and glyS mut denotes the point mutation that inactivates glycine binding. In the binding pocket, riboswitch is shown in purple, glycine in white with polar contacts as yellow dashed lines, and magnesium ions as white spheres. Sequences are in Supplementary Tables S1 and S2.

To identify improved biosensors, phylogenetic screening was performed with semi-randomly chosen candidate riboswitch sequences, ensuring a large degree of sequence diversity was represented. These candidate riboswitches were sourced from structural sequence alignment data from many additional bacterial species [34, 35] for S1-G and S2-P3 designs fused to Broccoli-Tr1. Of the S1-G sensors, Ho-b, from Halothermothrix orenii, met our screening criteria for further characterization of ≥1.9-fold activation (Supplementary Fig. S2). Meanwhile, 4 of the S2-P3 sensors showed ≥1.9-fold activation (Supplementary Fig. S3). These constructs (Spn-b, Spa-b, Ac-b, and Sc-b) originated from Streptococcus pneumoniae, Streptococcus parasanguinis, Anaerostipes caccae, and Streptococcus cristatus, respectively.

Characterization of glycine biosensors

Sequence-confirmed sensors displayed a wide range of affinities spanning nearly three orders of magnitude (Fig. 2B), allowing broad applicability. For sensors with similar affinities, Spn-b showed greater fold turn-on than Ac-b, while Sc-b showed greater fold turn-on than Ho-b (Supplementary Fig. S4). Therefore, along with Dh and Spa-b, we chose Spn-b and Sc-b for further characterization.

All sensors were highly selective for glycine (Fig. 2C and Supplementary Fig. S5). Spa-b and Dh showed no activation with other tested ligands. Spn-b and Sc-b showed ≤1.5-fold fluorescence turn-on to glycine analogs, compared to 23- or 9.0-fold turn-on to glycine, respectively. Our results agree with previous reports that glycine riboswitches are selective for glycine [36]. Kinetic measurements revealed half-maximal activation of sensors just minutes after ligand addition (Fig. 2D and Supplementary Fig. S6), consistent with previous RNA-based sensors [45, 46].

For Sc-b, Spn-b, and Dh sensors, brightness, apparent affinity, and fold turn-on increased with Mg²⁺ (Supplementary Fig. S7). In contrast, Spa-b exhibited increased fold turn-on when Mg²⁺ levels decreased from 5 to 1 mM due to reduced background. The requirement of ≥1 mM Mg²⁺ for optimal function aligns with the role of two Mg²⁺ ions in folding the ligand binding pocket (Fig. 2E) [37, 39].

Detection of glycine in live cells

Spn-b, renamed “glyS” for glycine sensor, was selected for live-cell studies for its superior brightness and fold turn-on. GlyS, glyS mut (Fig. 2E), which harbors a point mutation that abrogates glycine binding in vitro (Supplementary Fig. S8) [34], and the constitutive control Broccoli were expressed within the tRNA^Lys scaffold [14, 22, 47] for cellular studies. GlyS and glyS mut were spaced from the scaffold by 10 bp to promote optimal folding, which was expected to prevent riboswitch and scaffold structural elements from clashing by the length of a full helical turn, with additional linkers at riboswitch termini to form a three-way junction (Fig. 3A).

Figure 3.

Detection of intracellular glycine accumulation in E. coli with glyS. (A) Design of the glyS construct for cellular experiments. (B) Flow cytometry of glyS and controls in E. coli with different overnight treatments. (C) Representative flow cytometry histograms of E. coli expressing glyS at time points after 50 mM glycine addition. (D) Plot of ΔMFI/MFI values after glycine addition. (E) Schematic of cysteamine inhibition of glycine cleavage leading to buildup of endogenous glycine. The pyridoxal phosphate coenzyme that forms a Schiff base with glycine or cysteamine is abbreviated as PLP. (F) ΔMFI/MFI values after cysteamine addition over time. After cysteamine treatment for the time shown, live cells were incubated with 50 μM DFHBI-1T for 30 min before measurement. All MFI values were analyzed from at least 30 000 cells by flow cytometry, with average and standard deviation shown for three biological replicates. Sequences are in Supplementary Table S3.

Flow cytometry of E. coli cells expressing glyS showed a 3.6-fold fluorescence enhancement after overnight treatment with 10 mM glycine, while controls showed little to no changes (Fig. 3B). Also, untreated cells expressing glyS had 19-fold greater MFI than those expressing glyS mut. Since the background fluorescence of glyS is the same as glyS mut in vitro (Supplementary Fig. S8), this result is consistent with glyS sensing endogenous glycine levels. Overnight glycine treatment above 10 mM caused nonspecific fluorescence activation of controls, likely due to increased RNA expression, as glycine serves as a carbon source for nucleotide synthesis (Supplementary Fig. S9) [48].

Glycine was added to sensor-expressing cells for the indicated amount of time, then the cells were diluted into PBS with DFHBI-1T and incubated for 30 min, and the fluorescence response was quantified as ΔMFI/MFI, which normalizes to the baseline cellular fluorescence of different constructs and enables straightforward comparison of changes. GlyS exhibited a strong fluorescence turn-on response during the time course compared to controls (Fig. 3C and D), detecting glycine within 20 min. ΔMFI/MFI continued to rise until plateauing at 6 h, indicating full equilibrium. To our knowledge, these results demonstrate the first sensor to detect live single-cell glycine dynamics.

Visualization of drug-induced endogenous glycine accumulation

Since glyS could sense endogenous glycine levels, we challenged it to measure endogenous glycine dynamics by employing cysteamine, a treatment for cystinosis that also inhibits human glycine cleavage protein P, preventing glycine cleavage and leading to its accumulation (Fig. 3E) [49]. Recently, cysteamine was shown to also disrupt glycine utilization in Pseudomonas aeruginosa, resulting in increased glycine levels in cell lysates and reduced virulence of this human pathogen [50].

GlyS-expressing E. coli showed 3.7-fold fluorescence enhancement after cysteamine treatment overnight, while Broccoli showed none and glyS mut only 1.2-fold (Supplementary Fig. S10). When cysteamine was added to sensor-expressing cells, followed by the dye incubation step, a significantly greater increase in ΔMFI/MFI was observed for glyS compared to controls during the time course (Fig. 3F). Since neither glyS nor glyS mut displayed fluorescence enhancement in response to this cysteamine concentration in vitro (Supplementary Fig. S11), we again expect the small increase in glyS mut signal (ΔMFI/MFI ∼0.2) to be attributable to increased RNA expression with glycine accumulation.

To normalize fluorescence signals against changes in sensor levels, we developed a ratiometric sensor using the tRNA-DF30 scaffold [22] to display glyS (or glyS mut) separately from the orthogonal, red-emitting Pepper-HBC620 aptamer-dye pair as an internal control (Fig. 4A). From the green/red (G/R) fluorescence ratio, flow cytometry showed Pepper-glyS was similarly as sensitive to glycine and cysteamine overnight treatments as its intensiometric counterpart, while Pepper-glyS mut showed no responses (Fig. 4B). Repeating the cysteamine time course assay confirmed that Pepper-glyS, but not Pepper-glyS mut, exhibited increasing G/R ratios over time (Supplementary Fig. S12C and D) due to glycine accumulation, no longer confounded by changes in sensor levels. Ratiometric glycine imaging enabled fluorescence microscopy analysis after overnight cysteamine treatment, which revealed a 2.3-fold increase in G/R fluorescence ratio for Pepper-glyS (Fig. 4C and D).

Figure 4.

Two-dye ratiometric glycine sensing in E. coli with Pepper-glyS. (A) Design of Pepper-glyS for ratiometric glycine detection in E. coli. (B) Flow cytometry of Pepper-glyS and Pepper-glyS mut with different overnight treatments. MFI values in green (Broccoli) and red (Pepper) channels were analyzed from 30 000 cells and processed as a ratio, shown as the average with standard deviation for five biological replicates. (C) Fluorescence microscopy of Pepper-glyS and Pepper-glyS mut with different overnight treatments. Scale bars, 10 μm. (D) Violin plot of microscopy quantitation representing the distribution of green/red (G/R) fluorescence ratios from at least 150 cells expressing Pepper-glyS or Pepper-glyS mut. The solid line represents the median and the dashed lines represent quartiles. For the above experiments, live cells were incubated with 50 μM DFHBI-1T and 200 nM HBC620 for 1 h before measurement. P-value **** < .0001 is from using an unpaired student’s two-tailed t-test.

Interestingly, cellular dye permeability differences between DFHBI-1T and HBC620 were observed at their common working concentrations (50 μM and 200 nM, respectively; Supplementary Fig. S12E and F). While Broccoli fluorescence equilibrates within 10 min of DFHBI-1T addition to the cells, Pepper fluorescence equilibrates only after ∼75 min of HBC620 addition. Cell viability remains very high (96%) even after 1 h in PBS with dye (Supplementary Fig. S12G and H). These results highlight how differences in dye properties must be carefully considered when designing cellular assays with multiple dyes.

Expansion of the spectral range of Broccoli aptamers

In addition to developing a two-dye ratiometric sensor with orthogonal aptamers, we aimed to create spectrally resolvable aptamers for a ratiometric sensor that requires only a single dye. Altering the fluorophore’s chemical environment has been demonstrated to make spectrally resolved fluorescent proteins [24], and like Spinach, Broccoli binds fluorophores between a G-quadruplex and base triple (Fig. 5A and Supplementary Fig. S13A) [25, 51, 52]. Previously, mutating the base triple in Spinach-DFHBI blue-shifted the excitation and emission maxima [51]. Red Broccoli (RBr) and Orange Broccoli (OBr) changed a single nucleotide linker in the G-quadruplex, which may hydrogen bond with the oxime moiety of DFHO to tune its fluorescence (Supplementary Fig. S13B) [25]. Thus we hypothesized that the spectral range of Broccoli-DFHO could be expanded by combining base triple and G-quadruplex linker mutations.

Figure 5.

Design and characterization of the Golden Broccoli aptamer. (A) (Left) The binding pocket in the Broccoli RNA x-ray crystal structure (PDB: 8K7W) with DFHBI-1T (green), the G-quadruplex linker is shown in maroon and base triple nucleotides in red, black, and orange. Interactions of the base triple with the dye are shown as dashed red lines (pi-stacking interactions) and dashed black lines (hydrogen bonding). (Right) Schematic of WT and M1 base triples. (B) Table of Broccoli mutants with spectral properties and dissociation constants after 15 min incubation with DFHO at 37°C. NF indicates non-fluorescent. Excitation and emission spectra of (C) Broccoli-DFHO and blue-shifted mutants, or (D) Golden and Red Broccoli aptamers with DFHO, averaged from three technical replicates. FITC and TxRed channel excitation and emission are indicated for flow cytometry experiments. Sequences are in Supplementary Table S2.

Five base triple mutants were tested to tune Broccoli-DFHO excitation and emission, including M1, which recapitulates the blue-shifting mutation for Spinach-DFHBI [51], and M2–M5, which incorporate common base triples that maintain the same sugar and hydrogen bonding geometry as wild-type Broccoli (Fig. 5A and B) [53]. M1, M3, and M4 constructs activated DFHO fluorescence in vitro (Supplementary Fig. S14A and B), and exhibited blue-shifts in excitation (up to 21 nm) and emission (up to 10 nm) (Fig. 5B). By combining M1, the brightest of these mutants (Supplementary Fig. S14B), and the linker variant from OBr, we created “Golden Broccoli” (GBr), which exhibits the largest blue-shift in excitation and emission maxima (Figs. 5B and C).

GBr showed greater brightness than M1 and matched that of wild-type Broccoli upon incubation with DFHO (Supplementary Fig. S14C). This enhancement aligns with findings that OBr and RBr mutants are brighter than Broccoli with DFHO [54], supporting that mutation of adenine to a pyrimidine in the linker better accommodates the oxime moiety of DFHO. Further, GBr has an improved binding affinity compared to Broccoli and equal to OBr (K_D ∼1.31 μM, Fig. 5B, Supplementary Fig. S15A), indicating that a pyrimidine linker is more critical for DFHO binding than maintaining the wild-type base triple sequence. With RBr being the only red-shifted variant relative to Broccoli, GBr-RBr became the most spectrally distinct aptamer pair (Fig. 5B and D).

Broccoli-DFHO variants were analyzed using flow cytometry in FITC (ex/em: 488/530, bandpass: 30) and TxRed (ex/em: 561/610, bandpass: 15) channels. Though DFHO fluorescence plateaus within 5 min of addition to the cells (Supplementary Fig. S15B), longer incubation time points were still analyzed moving forward for the sake of consistency with assays presented earlier. As expected, the brightest aptamer in the TxRed channel is RBr, which exhibits some unwanted fluorescence spillover into the FITC channel. Despite identical brightness in vitro, Broccoli was brighter in cells than GBr in the FITC channel (Supplementary Table S4). This may be due to Broccoli, which was originally selected for live-cell fluorescence via FACS [14], expressing or folding better in live cells than GBr, though it is only three nucleotides different. However, GBr, as the most blue-shifted variant, did not exhibit any unwanted fluorescence spillover into the TxRed channel, unlike Broccoli and OBr (Supplementary Table S4).

Due to the unexpected discrepancy between in vitro and cellular results, we also investigated the dye OBI, which contains the oxime group important for spectral tuning, an additional benzimidazole group (Supplementary Fig. S13), and is brighter with the RBr aptamer in vitro and in mammalian cells [55]. In vitro, OBI provided improved resolution relative to DFHO, with a difference in excitation and emission maxima of 48 and 20 nm, respectively, between GBr and RBr (Supplementary Fig. S16). The total resolution of 68 nm is the largest to date between RNA aptamers that bind the same dye and is greater than the resolution between GFP and YFP [24].

In flow cytometry, however, OBI exhibited much lower MFI in FITC and TxRed channels compared to DFHO in E. coli (Supplementary Fig. S17A and B). This result is likely due to OBI having poorer permeability and accumulation in E. coli, a Gram-negative bacterium, as the opposite trend was observed in Bacillus subtilis, a Gram-positive bacterium (Supplementary Fig. S17C). Thus, dye permeability can be a strong confounding factor in analyzing orthogonal aptamer-dye pairs in varying cell types.

Taken together, we found that there is a tradeoff between resolution and live-cell brightness for different HBI-related dyes in E. coli, with DFHO providing the optimal balance. Interestingly, GBr with DFHO is brighter in the FITC channel in flow cytometry than Corn, a yellow dimeric aptamer selected for DFHO binding (Supplementary Fig. S18) [21, 54]. Thus, GBr provides a viable alternative to Corn when yellow fluorescence is desired without dimerization. Our development of the GBr aptamer also demonstrates that RNA structure need not differ substantially to achieve spectral resolution by tuning the properties of a single dye.

Generation of a single-dye ratiometric sensor

To create a SDR sensor, a similar tRNA-DF30 scaffold [22] was used as in Pepper-glyS, with Pepper on arm 1 replaced by GBr, and Broccoli within the glycine biosensor on arm 2 replaced by RBr (Fig. 6A). A spacer was added to extend arm 2 to promote optimal folding of the full RNA, generating SDR glyS.

SDR glyS and single-color controls were evaluated in vitro on a plate reader. First, the spillover of single-color controls RBr and GBr was quantified using plate reader conditions that mimic FITC and TxRed channels from the flow cytometer (Supplementary Fig. S19). With DFHO, RBr’s red fluorescence showed 68% spillover into the FITC channel, while GBr fluorescence was exclusive to the FITC channel. With OBI, RBr showed 22% spillover into the FITC channel, and GBr showed 4.2% spillover into the TxRed channel. The improved brightness and spectral resolution of these aptamers with OBI compared to DFHO suggests that while OBI may not be suited for studies in Gram-negative bacteria, it may be more useful for other applications.

These data on single-color controls allowed the individual aptamer signals of SDR GlyS to be corrected for spillover as described in the Materials and methods section. With added glycine, GBr fluorescence remained constant, while RBr fluorescence increased, regardless of whether DFHO or OBI was used (Supplementary Fig. S20A and B). This indicates that spillover correction effectively resolved signals from RBr and GBr. Consequently, the RBr/GBr fluorescence ratio increased with added glycine (Fig. 6B). Even without spillover correction, SDR glyS showed an increased RBr/GBr ratio with glycine, although less than with correction (Supplementary Fig. S21). Thus, RBr and GBr can be spectrally resolved in vitro for ratiometric sensing.

SDR glyS and controls were then analyzed in live E. coli using spectral flow cytometry (Fig. 6C). This instrument can resolve chromophores with spectral maxima only a few nanometers apart by measuring the emission spectra of single-color controls (RBr and GBr) across multiple channels and applying automated unmixing in software [56]. When processed as RBr/GBr fluorescence, SDR glyS showed 1.2-fold activation upon glycine treatment, whereas SDR glyS mut and the constitutive control GBr-RBr showed no change. SDR glyS exhibited 15-fold enhancement over glyS mut from sensing endogenous glycine even without glycine treatment.

For broader applicability, we also showed that SDR sensing can be achieved on a conventional flow cytometer. Using FlowJo to unmix signals via compensation with single-color controls RBr and GBr during post-experiment data processing (Supplementary Fig. S22), SDR glyS and controls exhibit the same trends observed on the spectral cytometer (Supplementary Fig. S23). Parallel to how GFP can be resolved from YFP using violet and blue lasers, with only GFP being activated by the violet laser [57], RBr can be resolved from GBr using yellow-green and blue lasers, with only RBr being activated by the yellow-green laser in the TxRed channel.

Since two-channel compensation effectively resolved GBr and RBr for ratiometric glycine sensing, we investigated whether similar mathematical unmixing in ImageJ could visualize SDR glycine sensing via live cell microscopy. First, the fluorescence of GBr and RBr controls in yellow and red channels was quantified (Supplementary Fig. S24A and B) then used to create an unmixing matrix for assigning GBr fluorescence to the yellow channel and RBr fluorescence to the red channel in ImageJ, which enhanced the resolution of the two signals (Supplementary Fig. S24C). Applying this matrix to images of cells expressing SDR glyS, we observed a modest but significant 1.1-fold increase in RBr/GBr fluorescence after overnight glycine treatment, which aligned with same-day flow cytometry results (Figs. 6D and E; Supplementary Fig. S25). Taken together, these findings demonstrate the success of the first RNA-based SDR sensor.

Discussion

This study demonstrates the first fluorogenic biosensors for intracellular detection of glycine, an important bacterial metabolite, neurotransmitter, and root exudate. We show that a complex riboswitch topology relying on inter-aptamer stacking interactions can be leveraged for sensor development, even when the fluorogenic aptamer is placed on a distal non-ligand-binding aptamer (S1-G design). This stacking modality contrasts with previously developed sensors, which mostly rely on either placement of the fluorogenic aptamer near the ligand binding pocket [43, 45, 46, 58, 59] or strand switching [60].

Sampling diverse bacterial riboswitch phylogeny enabled the creation of a suite of sensors that respond across a diverse range of glycine concentrations. This versatility will allow many potential applications both in vitro and across different cell types. The sensor glyS was applied to monitor intracellular glycine accumulation due to the drug cysteamine, making this the first tool to track endogenous glycine dynamics. Using the orthogonal aptamer-dye pair Broccoli-DFHBI-1T and Pepper-HBC620, we generated Pepper-glyS, which matches the brightness and sensitivity of glyS in live cells and enabled ratiometric imaging of endogenous glycine accumulation.

Despite their utility, RNA-based biosensors including glyS have limitations. Although these sensors are much faster than reporters [45], the reliance on the rate-limiting riboswitch conformation change for dye binding means that full sensor equilibration takes at least a few minutes [61]. The response time may be improved upon by incorporating transpose, transition, or unnatural orthogonal base pair mutations into the riboswitch [61]. However, as shown here, the additional requirement of cellular dye permeability for live-cell measurements can still introduce a delay to dynamics measurements, depending on the dye.

While the studies here focus on bacterial imaging, RNA-based sensors, including those using Broccoli-DFHBI-1T [62, 63] or Red Broccoli-OBI [55], have been successfully employed in mammalian cells. When expressed as circular RNA via autocatalytic transcripts, fluorogenic aptamers can accumulate in mammalian cells at similar levels as in bacteria for robust imaging applications, apparently without cytotoxicity or activation of the RIG-I innate immune response that recognizes 5′ triphosphorylated RNAs [62]. However, other studies have shown that endogenously produced circular RNAs can repress PKR activation through 16–26 bp dsRNA regions [64], whereas linear riboswitches that have elaborate and stable tertiary structures, including double-stranded regions, were shown to potently activate PKR [65]. Additionally, glyS is magnesium-dependent, requiring at least 1 mM of Mg²⁺ to fold the glycine binding pocket of the riboswitch [37, 39]. This means that either supplemental magnesium or using a less magnesium-dependent glycine sensor like Spa-b, which requires only 0.5 mM of Mg²⁺, may be needed to image glycine in mammalian cells. Thus, the sensors presented here may be useful for mammalian applications, upon evaluation of immunogenicity and functionality.

To our knowledge, this study is the first to achieve ratiometric measurements using an RNA-based biosensor with a single dye. By spectrally tuning DFHO or OBI through mutation of the ligand binding pocket, the aptamer Golden Broccoli was developed that is more spectrally resolved from Red Broccoli than Orange Broccoli, which was identified using a directed evolution approach from a randomized library of Broccoli sequences [54], and brighter than Corn, which binds the dye as a dimeric aptamer. We showed that SDR glyS, which incorporates Golden Broccoli and Red Broccoli, works for ratiometric sensing of glycine levels both in vitro and in live E. coli, and can be analyzed using a plate reader, spectral and conventional flow cytometers, and fluorescence microscope.

The single-dye approach to make a ratiometric RNA-based sensor contrasts with all previous studies, which used dissimilar aptamer-dye pairs to achieve spectral resolution [46, 66, 67]. This study advances the paradigm that using a single dye circumvents a layer of biological complexity in fluorogenic aptamer analysis that arises when different dyes are applied because of cellular permeability differences as we have observed. A limitation of the current sensor is that DFHO is not the brightest dye when used in E. coli, and the Golden and Red Broccoli aptamers require unmixing for full spectral resolution. However, DFHO also exhibits faster cellular equilibration than the other dyes tested, so is capable of near real-time sensing with a time delay of ∼5 min. Our findings here reveal that combining structure-based mutations can lead to additive spectral tuning effects and will motivate future efforts to generate brighter and more spectrally resolved single-dye aptamer pairs for multicolor RNA-based imaging applications. Even beyond RNA-based sensors and cellular imaging experiments, a recent study used DFHO with multi-channel ratio measurements of Red Broccoli, Orange Broccoli, and Corn to label and identify separate, orthogonal RNA condensates [68]. We expect that Golden Broccoli will also be useful for these and other types of engineered multi-RNA systems in cell-free and artificial cell systems [69, 70].

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by the National Science Foundation [1815508 to M.C.H. and GRFP 2139322 to M.R.B.] and the National Institutes of Health [R01 GM124589 to M.C.H. and T32 GM122740 to M.R.B.]. The Flow Cytometry Core is funded by the National Institutes of Health [S10OD026959 and P30CA042014-24]. Funding to pay the Open Access publication charges for this article was provided by the University of Utah through a Read and Publish deal with Oxford University Press.

Data availability

Sequences of RNA constructs are provided in Supplementary Information and deposited in Genbank (accession numbers PV197732–PV197869). Numerical data from plate reader, flow cytometry, and microscopy experiments are provided as Excel spreadsheet files. Flow cytometry data are publicly available through FlowRepository (ID: FR-FCM-Z92V).