Measuring fast gene dynamics in single cells with time-lapse luminescence microscopy

INTRODUCTION

Time-lapse fluorescence microscopy of genetically encoded fluorescent proteins is the gold standard for measuring in vivo dynamics of gene expression in single cells. However, fluorescence microscopy uses intense light to excite a fluorophore, which can result in high background autofluorescence and phototoxicity in cells. This problem becomes acute when measuring multiple genes (which requires exciting different fluorophores with different wavelengths) on fast time scales (which requires frequent exposure to intense light). For these reasons, time-lapse fluorescence microscopy is typically limited to one or two fluorescent reporters. A further limitation is the slow chromophore maturation of fluorescent proteins, which can introduce a significant time delay between gene expression and measurable change in fluorescence. Fluorescent protein reporters with long maturation times cannot faithfully track gene expression dynamics (Wang et al., 2008; Gedeon and Bokes, 2012)

An alternative is luciferase, an oxidative enzyme that produces photons. Luciferases fold cotranslationally and are enzymatically active upon folding (Frydman et al., 1999). Thus, in contrast to fluorescent proteins, they do not require intense excitation light and do not exhibit maturation delays. Luciferases are an example of convergent evolution because diverse organisms (e.g., beetles, marine invertebrates, bacteria) have independently evolved enzymes to produce light of different colors using unique substrates (e.g., d-luciferin, coelenterazine, and N-decanal; Wilson and Hastings, 1998). Despite this enzymatic diversity, all known luciferases require oxygen to catalyze their substrate into an excited-state product that decays and emits a single photon. The ability to maintain in vivo oxygen and substrate concentrations at high and consistent levels during live-cell time-lapse luminescence microscopy has been a challenge. Another limitation is the number of photons produced per luciferase per second. The maximum catalytic rates of beetle and marine luciferases range from 0.1 to 10 photons/s per luciferase (Branchini et al., 1999; Loening et al., 2006). This is substantially lower than the 1000 photons/s per green fluorescent protein (GFP) at an excitation intensity of 10 W/cm² (Garcia-Parajo et al., 2000; Chiu et al., 2001).

Time-lapse luminescence microscopy has been successfully used to measure in vivo gene dynamics in larger organisms (e.g., animals, plants, tissue culture) or for slower processes (e.g., circadian oscillation), for which more total photons per pixel can be collected during one camera exposure (Liu et al., 2007; Nakajima et al., 2010). The state of the art for microbial time-lapse luminescence microscopy measured circadian gene dynamics in single cyanobacteria with a 30-min exposure time (Mihalcescu et al., 2004). To our knowledge, no one has developed methods of microbial time-lapse luminescence microscopy that measure in vivo gene dynamics for extended periods of time with subminute resolution (e.g., cell cycle).

Commercial time-lapse microscopes and sensitive charge-coupled device (CCD) cameras have greatly improved over the last decade and are routinely used by biologists. Thus we sought to improve in vivo luciferase enzymatic yield (photons per second) rather than photon detection. We developed a method for live-cell time-lapse luminescence microscopy that can measure subminute gene dynamics in budding yeast at single-cell resolution. This was achieved by optimizing d-luciferin delivery across the lipid bilayer, using microfluidics to trap cells and maintain substrates at saturating concentrations during imaging, and integrating multiple copies of luciferase genes into the yeast genome.

RESULTS

We first designed and constructed green (GrLuc), yellow (YeLuc), and red (RdLuc) luciferase derived from beetles (Viviani et al., 1999a, b; Fujii et al., 2007). We fused an N-terminal nuclear-localization signal (NLS) to concentrate these luciferases into a smaller volume and spread the light signal across fewer pixels (Figure 1A). For comparison, we also tested commercial beetle luciferases (CBG99, CBR, FLuc) and a new marine luciferase (NLuc; Hall et al., 2012) from Promega (Madison, WI; Figure 1B). Each luciferase reporter was regulated by a methionine-repressible promoter (MET17) and integrated into the yeast genome either in single copy or multiple copies; see Materials and Methods. We measured the emission spectrum of our designed luciferases to confirm that they were consistent with their expected color; see Supplemental Figure S1.

We then used a 96-well plate bulk assay with a Wallac Victor 3 luminometer (PerkinElmer-Cetus, Waltham, MA) to optimize conditions (substrates, pH) for in vivo yeast luminescence. Beetle luciferases require d-luciferin, ATP, and oxygen as substrates, whereas NLuc only requires furimazine (a coelenterazine analogue) and oxygen. We expected log-phase cells to maintain their ATP at high concentrations. However, in vivo oxygen, d-luciferin, and furimazine could be limiting because they must diffuse across the lipid bilayer from the external medium (Wood and DeLuca, 1987; Vieites et al., 1994). We determined that 200 μM d-luciferin and 50 μM furimazine were saturating for in vivo luminescence (Figure 2A). We further boosted the in vivo luminescence of beetle luciferases ∼30-fold by lowering the pH of the growth medium (Figure 2B). The luminescence increased because of the low extracellular pH, which reduced the charge of d-luciferin (pK_a = 2.9) and improved its diffusion across the lipid bilayer (Wood and DeLuca, 1987; Vieites et al., 1994). We verified that our yeast strains grew well at lower pH (3.8) and that the luciferases were not cytotoxic at high copy number (Figure 2C). The only exception is NLuc, which was already cytotoxic as a single-copy integrant (Figure 2C).

We also compared the relative brightness of different luciferases. To this end, we integrated a single gene copy of each luciferase into the same target locus. Our beetle luciferases GrLuc and YeLuc were ∼10-fold brighter than RdLuc (Figure 3A). This same color relationship persists for Promega beetle luciferases CBG99 (green), FLuc (yellow), and CBR (red). All of the commercial luciferases were at least ∼10-fold brighter (Figure 3A) than our designed luciferases. NLuc was the brightest. Unfortunately, furimazine was insoluble in aqueous solution and precipitated out, which made NLuc incompatible with long-term, live-cell time-lapse luminescence microscopy.

We screened all multicopy transformants using a 96-well plate luminescence assay and selected the “brightest” strains (Figure 3B) for time-lapse luminescence microscopy. The variability in luminescence between transformants arises from unequal copy number integration in chromosomal loci due to homologous ends-in recombination. Time-lapse luminescence microscopy using standard, agarose pad methods initially exhibited bright luminescence signal. However, the signal disappeared within 15 min. We reasoned that luciferase substrates (d-luciferin, oxygen) were being depleted. Thus we combined a CellAsic microfluidic device (EMD Millipore, Billerica, MA) with time-lapse luminescence microscopy. The microfluidic device trapped yeast cells and maintained growth medium at constant levels through perfusion. The medium exchange time was <1 min, and cells were unable to deplete their substrates. With microfluidics, we consistently measured in vivo gene expression dynamics with subminute time resolution for >12 h (Figure 4). We only stopped time lapse because the yeast microcolony extended beyond the field of view. We further validated our method by benchmarking and quantifying gene induction and repression dynamics of many green, yellow, and red luciferase reporters for several metabolite-repressed yeast promoters (MET17, LEU1, ADE17, LYS9); see Supplemental Figures S2–S4 and Supplemental Table S1.

To compare luciferase and fluorescent protein reporters directly, we used both time-lapse luminescence and fluorescence microscopy to measure gene induction and repression dynamics of a FLuc-yEVenus-PEST fusion (Figure 1C) regulated by the LEU1 promoter (Figure 4). We chose yEVenus-PEST because it is a fast-maturing, destabilized yellow fluorescent protein (YFP) commonly used to measure yeast cell cycle dynamics (Skotheim et al., 2008). Our results showed that the fluorescence kinetics of FLuc-yEVenus-PEST during induction and repression are consistently slower and delayed compared with luminescence. The slow kinetics is worse if FLuc-yEVenus is not destabilized with a PEST degron; see Supplemental Figure S5 and Supplemental Table S1. We obtained identical results with FLuc-yEVenus and FLuc-yEVenus-PEST fusions regulated by the MET17 promoter; see Supplemental Figure S6 and Supplemental Table S1.

These results suggested that our luciferase should more faithfully track cell-cycle oscillations than fluorescence reporters. To this end, we expressed FLuc-yEVenus-PEST under the transcriptional control of two different yeast cell cycle promoters, SIC1 and RNR1. We successfully tracked cell cycle luciferase dynamics with subminute resolution (Figure 5). SIC1 transcripts are known to peak at the M/G1 border near cytokinesis, whereas RNR1 transcripts peak at the G1/S border (Spellman et al., 1998). Our data showed that the luminescence of SIC1pr-FLuc-yEVenus-PEST peaks on average 13 min before budding, whereas fluorescence peaks 6 min after budding. The luminescence of RNR1pr-FLuc-yEVenus-PEST peaks on average 7 min before budding, and fluorescence peaks 11 min after budding. Thus fluorescence lagged the luminescence signal by 15–20 min. This delay was identical to the measured in vivo chromophore maturation delay of yEVenus-PEST (Charvin et al., 2008). To verify that our protein fusion was not interfering with yEVenus folding and/or maturation, we built strains with either yEVenus-PEST or FLuc under the control of SIC1 or RNR1 promoter, respectively. Fluorescence of yEVenus-PEST alone continued to exhibit a 15–20 min delay when compared with luminescence of FLuc (Supplemental Figure S7). A two-sample Student's t test of our fluorescence peak and budding data shows no significant statistical difference between FLuc-yEVenus-PEST (Figure 5) and yEVenus-PEST (Supplemental Figure S7) with either SIC1 or RNR1 promoter; see Supplemental Table S2 for a complete statistical analysis. The same is true for luminescence peak and budding data between FLuc-yEVenus-PEST and FLuc. Supplemental Table S2 also shows that PEST has no significant effect on the timing of peak signal and budding. We conclude that FLuc-yEVenus-PEST fusion does not significantly affect the timing of either FLuc luminescence or yEVenus fluorescence.

DISCUSSION

Here we optimized luminescence microscopy to measure in vivo gene dynamics (e.g., cell cycle) in yeast cells with subminute resolution. The temporal resolution was ∼4 times faster and the cellular volumes were ∼20 times smaller than previous methods with beetle luciferases (Supplemental Table S3). Our protocol now adds time-lapse luminescence microscopy to the microbiologist's toolbox for measuring fast gene dynamics with a beetle luciferase. Time-lapse luminescence can be used alone (if phototoxicity and/or chromophore maturation is a problem) or in conjunction with time-lapse fluorescence microscopy (if more gene reporters are needed). Successful time-lapse luminescence only required a dark room, a sensitive electron-multiplying (EM) CCD camera, and a microfluidic device to prevent depletion of critical luciferase substrates (e.g., d-luciferin, oxygen). All of this equipment is commercially available and compatible with standard time-lapse fluorescence microscopes. We did not measure dynamics of several genes simultaneously using our multicolor luciferases, but luminescence multiplexing can be done in batch culture (Nakajima et al., 2005) and in single cells (Kwon et al., 2010) with filters and spectral unmixing.

Our method was developed to study fast transcriptional dynamics. However, luciferases can be fused to proteins to study localization and protein–protein interactions via split luciferases (Ozawa et al., 2001; Luker et al., 2004; Malleshaiah et al., 2010) or bioluminescence resonance energy transfer (Xu et al., 1999). If protein localization or transport is fast, the ∼1-min exposure time required with a beetle luciferase fusion cannot provide the same spatial resolution as fluorescent proteins. Directed evolution and rational design of blue marine luciferases (e.g., NanoLantern, NLuc) have improved the photon flux ∼10- to 100-fold (Hoshino et al., 2007; Hall et al., 2012). For example, NanoLantern was able to measure protein localization and transport in mammalian cells (Saito et al., 2012). Unfortunately, in our hands, NLuc (a blue marine luciferase) was cytotoxic, and furimazine (a derivative of coelenterazine substrate) was insoluble in the microfluidic device. As conditions continue to be improved for in vivo luminescence, luciferases should eventually exhibit all the benefits of fluorescence proteins without the drawbacks. This will also require co-optimizing substrates to be water soluble and diffusible across the lipid bilayer.

MATERIALS AND METHODS

Plasmid construction

Luciferase sequences from larval click beetle (Pyrearinus termitilluminans; GrLuc, green), firefly (Photinus pyralis; YeLuc, yellow), and railroad worm (Phrixothrix hirtus; RdLuc, red) were obtained from GenBank (AF116843, AB261988, AF139645; Viviani et al., 1999a, b; Fujii et al., 2007). Restriction enzyme cut sites, an N-terminal SV40 NLS, and an ssrA degron tag were added to the sequence. The designed DNA sequence was de novo synthesized by DNA 2.0 (Figure 1) into their pJ204 vector. Optimized luciferases from click beetle, Pyrophorus plagiophthalamus (pCBG99, green; and pCBR, red), and deep-sea shrimp, Oplophorus gracilirostris, were obtained from Promega. The plasmid pRS303d-KAN-GAL1pr-Luc (A4V) (Robertson et al., 2008), which contains the optimized firefly (P. pyralis) luciferase from pGL3-Basic vector (Promega), was a gift from the Carl Johnson laboratory at Vanderbilt University (Nashville, TN).

We first built the MET17 promoter (also known as MET25) driving each of the luciferase genes. Parental p406-MET17pr-yEVenus vector was digested with PacI-AscI, and the reporter gene (yEVenus) was replaced with a PacI-AscI–digested luciferase insert (GrLuc, YeLuc, or RdLuc from pJ204-NLS-luciferase-ssrA). In the case of commercial luciferases, the insert genes were amplified from the original vectors (see oligos in Tables 1 and 2). Most of the forward oligos contained a PacI site, whereas the reverse oligos had AscI and BsrGI or XhoI sites (Figure 1 and Table 1). The resulting PCR product was then digested and ligated to parental p406-MET17pr-yEVenus vector that had been digested with PacI-AscI (Table 2) or integrating using the Sequence and Ligation Independent Cloning (SLIC) method. The “disintegrator” plasmids were built by cutting luciferase inserts and pIS385-MET17pr parental vectors with PacI-AscI, followed by ligation to create pIS385-MET17pr-luciferases.

All remaining plasmids with LEU1, ADE17, LYS9, SIC1, and RNR1 promoters were created either by swapping promoters (using SacI-PacI) or by swapping luciferases (using PacI-AscI).

We created a luciferase and fluorescent protein fusion by inserting the yEVenus gene, amplified from p406-MET17pr-yEVenus (see oligos in Table 1), into digested XhoI-AscI p406-MET17pr-FLuc using the SLIC method. PEST degron was extracted from p406-MET17pr-FLuc-PEST by XhoI-AscI digest and then ligated to XhoI-AscI–digested p406-MET17pr-FLuc-yEVenus. The resulting cassette was then inserted after LEU1, SIC1, and RNR1 promoters using PacI-AscI cut sites. All plasmids were verified by analytical restriction digest and fragment analysis by 3730xl DNA Analyzer (Life Technology, Grand Island, NY). Most plasmids in Table 2 are available from Addgene (Cambridge, MA).

Strain construction

Standard methods were used for bacterial and yeast transformations. All yeast strains in Table 3 were derived from MMY116-2C, which is the W303 laboratory strain in which the ade2-1 mutation was restored to ADE2. Multicopy strains in Table 3 were created using homologous ends-in recombination at the URA3 or LEU2 locus after plasmid linearization by StuI or AflII digestion, respectively. A consequence of ends-in homologous recombination is that different transformants will have different numbers of plasmid copies integrated into their target locus (Orr-Weaver et al., 1981). Most single-copy strains in Table 3 were created using a URA3 pop-in/pop-out “disintegrator plasmid” (Sadowski et al., 2007). Briefly, the luciferase gene and URA3 are integrated via homologous ends-in recombination into the LYS2 locus after NruI digestion of plasmids. Subsequent selection for URA3 pop out results in a single gene integration of luciferase in disrupted LYS2 locus (lys2Δ). Single-copy integration was confirmed by PCR. All strains in Table 3 are available upon request.

Ninety-six–well plate luminescence assay

A Wallac Victor 3 plate reader was used to measure luminescence intensity (Figures 2 and 3). Yeast strains were grown overnight to reach OD₆₆₀ ≈ 0.2. Our growth medium was SCD–Met at pH 3.8 for beetle luciferases and pH 6.0 for NLuc. The samples were then diluted to OD₆₆₀ ≈ 0.1 and loaded in a 96-well black assay plate (nontreated) with flat, clear bottom. At the beginning of the time course, 100 μM beetle luciferin or 20 μM furimazine was added to the cell cultures. All cultures were grown in a 30°C incubator, and before each measurement, plates were vortexed for 30 s to resuspend and oxygenate the cell cultures. Luminescence intensity in each well was measured over a 0.5-s exposure time, and samples were spread along the plate in a chess pattern to minimize light leakage from well to well.

Time-lapse luminescence microscopy

The brightest multicopy strain of each luciferase (Figure 3B) was grown in culture tubes overnight in a 30°C incubator to reach OD₆₆₀ ≈ 0.1–0.2. Metabolite-repressed promoter strains were pregrown in dropout medium to induce (SCD-Metabolite) or in synthetic complete medium to repress (SCD) luciferases. The CellAsic microfluidic chamber has a ceiling height of 4 μm, which is the dimension of haploid yeast. However, yeast cells are naturally flocculent and clump during pregrowth. Thus we sonicated the yeast in a Bioruptor UCD-200 sonicator (Diagenode, Denville, NJ) for 30 s at medium intensity to obtain single-cell suspensions before loading them onto microfluidic plates (CellAsic). Media distribution was controlled by ONIX microfluidic perfusion platform, which provides a well-oxygenated environment. Bioluminescence imaging of yeast cells was performed with a DV Elite microscope equipped with UltimateFocus, an Evolve EMCCD camera (Photometrics, Tucson, AZ), and a 60×/1.25 numerical aperture phase-contrast oil objective lens. Cells were grown at 30°C in complete (SCD) or dropout medium (SCD–Met, SCD–Ade, SCD–Lys, SCD–Leu) at pH 3.8. The medium was supplemented with 200 μM beetle d-luciferin. We imaged cells every 4 min. To improve signal-to-noise ratios, we used 2 × 2 binning (256 × 256 pixels) and a camera gain of 55 with a transfer speed of 5000. The microscope was in an isolated, dark room with a floor-to-ceiling curtain. All light sources in the microscope, computer, and other equipment in the room were wrapped with foil paper and dark tape. We blocked all excitation light, and we removed the emission filters to maximize light transmission. We also modified our time-lapse macro to turn off the UltimateFocus laser every time we did a luminescence exposure.

Time-lapse files with raw luminescence data were preprocessed in Fiji (Schindelin et al., 2012) in the following order: 1) remove bright spot artifacts (i.e., hot pixels), with Remove Outliers function (threshold 1000; 20-pixel radius) for each z-slice, 2) use Sum Slices projection under Z Project menu, and 3) save integrated z-stack in tiff format. Phase-contrast images were imported into CellStat (a MATLAB plug-in; Kvarnström et al., 2008) to segment and track yeast cells over time. Each cell contour over time was verified manually and refined when needed. To retrieve the total luminescence and fluorescence for each cell, the preprocessed luminescence files were imported into CellStat and merged with phase-contrast image contours. Time-course data and statistics were retrieved using the Highest method from CellStat.

Fitting a simple model of gene expression to luminescence data

We developed a quantitative model of gene expression to objectively compare different luciferases as transcriptional reporters. The simplest is a binary model of transcription, which assumes that upon induction or repression, the gene expression output jumps between two synthesis rates. The approach to steady state is dictated by the degradation time scale (τ), which could be luciferase inactivation, degradation, and/or dilution through cell growth. To account for delays due to transcription, mRNA processing, translation, folding, or maturation of either luciferase or fluorescent protein, we included an explicit delay term (δ).

Repression model

where S_max and S_min are the steady-state concentrations of induced and repressed luminescence signal, respectively. We used nonlinear least squares (nlinfit in MATLAB) to fit our repression model to average the luminescence signal of single cells repressed by the addition of metabolite. The best-fit curves are shown in Supplemental Figures S2–S6, with the corresponding best-fit parameters (and confidence intervals) shown in Supplemental Table S1. The same repression model and fitting procedure was used for fluorescence signal.

Induction model

where S_max and S_min are the steady-state concentrations of induced and repressed luminescence signal, respectively. We used nonlinear least squares (nlinfit in MATLAB) to fit our induction model to average the luminescence signal of single cells induced by the removal of metabolite. The best-fit curves are shown in Supplemental Figures S2–S6. with the corresponding best–fit parameters (and confidence intervals) shown in Supplemental Table S1. The same induction model and fitting procedure was used for fluorescence signal.

Cell doubling time.

We measured the doubling time of single cells before switching the medium at ∼6 h by counting number of cells as a function of time. We used nonlinear least squares (“nlinfit” function in MATLAB) of a simple exponential exp[ln₂(t/λ)] to fit the data. The estimated doubling time (λ) before medium switch and the 95% confidence intervals are shown in Supplemental Table S1. On average, cell-doubling time in medium with missing metabolites (SCD–Met, SCD–Leu, SCD–Ade, SCD–Lys) was ∼93 min compared to ∼86 min in complete medium.

FOOTNOTES

ACKNOWLEDGMENTS

REFERENCES