Single Molecule Fluorescence In Situ Hybridization (smFISH) Analysis in Budding Yeast Vegetative Growth and Meiosis

To evaluate how well the smFISH protocol worked (outlined in Figure 1), we designed a set of 54 probes that tile the open reading frame of the NDC80 gene (Figure 2A, top). The probe located at the most 5' end is referred to as Probe 1; the next one, Probe 2; and the third one, Probe 3, etc. The 27 probes assigned an odd number (Probe 1, 3, 5…) are all conjugated to the CAL Fluor 590 (CF590) fluorescent dye; and the 27 probes assigned an even number (Probe 2, 4, 6…), conjugated to the Quasar 670 (Q670) fluorescent dye. Hence, these alternating probe sets are often referred to as "odd/even" probes. Upon hybridization, both probe sets should label the same transcript.

After spot detection, we used a few measurements to judge the quality of the smFISH data set. The first one was the degree and quality of colocalization for the odd/even probes. In our case, 88% of all the smFISH spots colocalized (Figure 2A and 2B), with greater than 95% of the spots paired within 2 pixels of each other (Figure 2C, paired), which is within the expected value given any chromatic and detection aberration between the two fluorescent channels. By comparison, fewer than 10% of unpaired spots had a nearest-neighbor distance of less than 2 pixels, showing that the probability of misidentifying a spot pair is low (Figure 2C). The 12% of unpaired spots were evenly divided between the two channels (Figure 2B, compare CF590-only with Q670-only); and thus, we concluded that for this 2.4 kb gene with a range of expression between zero to 45 transcripts per cell, ~94% of RNA molecules were accurately detected in each fluorescent channel. If the smFISH protocol were suboptimal, one would observe (1) a greater fraction of the smFISH spots with only one of the two fluorescent signals (non-colocalized), and/or (2) cells with a very low signal-to-noise ratio or no signal at all.

We next asked if the detection of the smFISH signal was biased with respect to the total number of RNA molecules per cell. In a population, the total number of a particular RNA in each cell lies in a distribution, with some cells harboring more RNA molecules than others. A good smFISH protocol should robustly detect the RNA regardless whether a high number or low number of RNA molecules is present in each cell. To test this, for each cell, we calculated the fraction of the smFISH spots with colocalized signals and the fraction with only one of the two fluorescent signals. After grouping the cells that had the same number of total spots per cell, we calculated the average fraction of colocalized (paired) or non-colocalized (CF590-only or Q670-only) spots in a given group, and graphed this average as a function of the total number of spots per cell (Figure 3). For each category of spots, the fractions were similar across the entire range of the total number of spots per cell. For instance, comparing the cells with a total of 20 fluorescent spots per cell versus those with a total of 30 spots, the average fractions of the colocalized spots were similar. This result suggested that our protocol could detect a range of RNA molecules in a cell (up to at least ~40 molecules per cell). If the protocol worked suboptimally, one might observe a bias. For example, the fraction of the spots with only one of the two fluorescent signals might increase as the total number of spots per cell increased.

Using this optimized protocol, we examined the expression of the NDC80 gene, which encodes a kinetochore protein, in different growth conditions. The NDC80 gene expresses two mRNA isoforms: The long undecoded transcript isoform, NDC80^luti, has a ~400 base pairs extension at the 5' end in comparison to the short NDC80^ORF isoform, but both transcripts share the coding region of the NDC80 gene (See schematic in Figure 4A). We designed two sets of probes: The CF 590 set binds to the unique 5' region of NDC80^luti; and the Q 670 set binds to the common region of NDC80^luti and NDC80^ORF. The NDC80^luti transcripts were detected as the smFISH spots where the signals from both probe sets colocalized, whereas the NDC80^ORF transcripts were those with the signal only from Q 670. Due to the size of the unique segment of NDC80^luti, we could only tile 20 oligonucleotide probes along this region, which is fewer than the recommended 30 to 48 probes. Thus, we first determined if this probe set could robustly detect RNA in our optimized protocol. For either fluorescent channel, we graphed the signal-to-noise ratio (SNR, defined as the variance of the pixels surrounding a spot relative to the spot intensity) against the signal detected for each smFISH spot, and generated a scatterplot for all the spots identified in the entire field of view (example images in Figure 4A and plots in Figure 4B). Two distinct populations were clearly identified for both the Q 670 and the CF 590 probe sets (Optimized, Figure 4B), suggesting that the true smFISH spots (inside the grey area) could be separated from the background signal. Note that in the suboptimal condition, such separation was less obvious (Suboptimal, Figure 4B).

We used these two probe sets to study the expression of NDC80^luti and NDC80^ORFduring vegetative growth and meiosis. In vegetative cells, robust signal from the Q 670 probe set was detected, but not from the CF 590 probe set (Figure 5A, vegetative), agreeing with the observation by northern blotting that only the short isoform was expressed during vegetative growth¹⁶. This result also suggested that the CF 590 probe set is specific, yielding low background in our optimized smFISH protocol. In contrast, robust signal from both probe sets was detected in meiotic prophase, and the majority of the spots had colocalized signal (Figure 5A, meiotic prophase). Together with northern blot analysis (Figure 5B), this observation confirmed that the long isoform NDC80^luti was expressed specifically in meiosis. These two types of mRNAs were detected in the cytoplasm (outside of the DAPI region), suggesting that both were exported from the nucleus, consistent with the ribosome profiling data¹⁸.

To determine if sufficient data were collected to accurately account for the biological variation intrinsic to our data set, we performed bootstrap analysis using the statistics of each cell, which included (1) the fraction of the colocalized spots per cell and (2) the fraction of the spots with one of the two fluorescent signals (CF 590 only and Q 670 only). In this analysis, a program randomly selected one cell from the 437 quantified cells for 500 iterations. Next, the mean and the variance of the respective statistics associated with these 500 selected cells were calculated. This process was then repeated for randomly selecting two cells from all the cells, without replacement; and then for randomly selecting three cells, etc. until one half of the total data set size was reached. The variance in the data set plateaued after ~40 cells, suggesting that after this point most of the variation in the mean is intrinsic to the data rather than an artifact of undersampling (Figure 6, inset). This effect became more obvious when we graphed the sample variance divided by the sample mean, as a function of the number of cells sampled in each iteration cycle (Figure 6). With the data shown, the change in variance became diminishingly small after ~60 cells. The number of cells quantified in an smFISH experiment should exceed the minimum number of cells required to achieve a stable mean and a plateau of the variance. In our case, we quantified over 95 cells per sample, per replicate. The total number of cells (> 400 cells) well exceeded the minimum number of cells (~60 cells) required to reflect the population mean.

With a custom-made Matlab program¹⁶, vegetative cells were found to have a median of 5 NDC80^ORF transcripts per cell, whereas in meiotic prophase cells, the median significantly dropped to 4 transcripts per cell (two-tailed Wilcoxon Rank Sum test, p = 0.026) (Figure 7). The median number of NDC80^luti transcripts per cell was 21 transcripts, and 100% of the cells expressed the NDC80^luti transcripts. We graphed the fraction of the cells with a given number of transcripts as a step-wise, relative frequency histogram because the number of mRNA molecules is a discrete quantity. The largest bin of each histogram was normalized to the same height.

Figure 1: Flow-chart for the smFISH protocol. Cells are fixed with formaldehyde at room temperature for 20 min or at 4 ˚C overnight, for vegetative growth samples and meiotic samples, respectively. After washing in Buffer B three times, cells are digested by zymolyase until ~70% – 90% of cells are digested. The digested samples are then washed one time with Buffer B to remove the zymolyase, and subsequently permeabilized in 70% ethanol (EtOH) for ~3.5 hours. To prepare for hybridization, samples are first incubated in 10% formamide wash buffer (FWB) for ~20 min. Next, the samples are resuspended in ~50 µL of Hybridization Solution, which contains the fluorescent probes, for overnight incubation in the dark at 30 ˚C. After hybridization, the samples are incubated in 10% FWB for 30 min to wash away the excess probes, and then incubated in 10% FWB with 4',6-diamidino-2-phenylindole (DAPI) to stain the DNA. For the sample imaged immediately after the FWB/DAPI incubation, the sample is resuspended in the GLOX buffer supplemented with catalase, Trolox, and glucose oxidase (GLOX + enz); whereas, the other samples are resuspended in the GLOX buffer without enzymes, and stored at 4 ˚C up to ~3 hours. Please click here to view a larger version of this figure.

Figure 2: Assessment of the smFISH quality using odd/even probes. (A) Top: Schematic for the odd/even probe sets. Fifty-four oligonucleotide probes tiling the NDC80 gene were designed. The odd-numbered probes were labeled with one fluorophore (CF590, shown in magenta), and the even-numbered probes, with another fluorophore (Q670, shown in green). Bottom: Representative smFISH images of meiotic prophase cells acquired using the odd/even probe sets. Samples were taken 6 hours after cells (UB8144) were transferred to sporulation medium, a time when these cells were arrested in meiotic prophase. DNA was stained with DAPI (shown in blue). Images are shown as the maximum-intensity projections of z-stacks. Scale bar: 5 µm. (B) Percentage of the paired or unpaired smFISH spots obtained using the odd/even probe sets. A total of 428 meiotic prophase cells were analyzed, pooling from two independent experiments. This figure is modified from Figure 2-figure supplement 4 of Chen et al.¹⁶ (C) A Cumulative Density Function (CDF) of the distance between each pair of paired spots and distance between the nearest neighbor of an unpaired spot. For each detected spot in one fluorescent channel, the "k nearest neighbor" algorithm was applied to identify the closest detected spot—and the distance to that spot—in the complementary channel. Localizations that were mutual nearest neighbors were considered to be paired, and a new list was generated recording the paired, CF590-only, and Q670-only detections. For each category of detections, a CDF histogram of the distances was plotted in Matlab, confirming that correctly paired spots indeed were much closer in distance than those without a corresponding spot in the other channel. Please click here to view a larger version of this figure.

Figure 3: Fractions of paired and unpaired spots, as a function of the total RNA number per cell. To test if the detection and pairing of the smFISH spots was biased at different expression levels, individual cells were grouped by total RNA expression. The mean fractions of paired, CF590-only, and Q670-only detections were calculated for each group of cells. Please click here to view a larger version of this figure.

Figure 4: Quality assessment of the smFISH data generated using the probes designed for NDC80^luti and NDC80^ORF mRNAs. The Q 670 probes (shown in green) hybridize to the common region shared between NDC80^luti and NDC80^ORF mRNAs, whereas the CF 590 probes (shown in magenta) hybridize to the unique 5' region of NDC80^luti. (A) Representative smFISH images of NDC80^luti and NDC80^ORF in meiotic prophase, acquired under optimized versus suboptimal conditions. Strains UB8144 (optimized condition) and UB1337 (suboptimal condition) were induced to undergo meiosis and fixed during meiotic prophase. These two strains harbor different synchronization systems to induce meiosis, but the use of either system does not affect smFISH quality (data not shown). In the suboptimal condition, cells were fixed at room temperature for 20 min (no overnight fixation), digested with zymolyase without VRC supplemented, and hybridized in a lower concentration of VRC. Images are shown as the maximum-intensity projections of z-stacks. DNA was stained with DAPI (shown in blue). Scale bar: 5 µm. (B) Scatterplots displaying the signal-to-noise ratio (SNR) and the signal of each smFISH spot detected in the field of view presented in Figure 4A, for either fluorescent channel, as well as for the optimized or suboptimal conditions. In the optimized condition, two populations of the smFISH spots were present. The true smFISH spots were located inside the grey area, separating from the background signal. In the suboptimal condition, individual mRNAs were hard to distinguish by eye, and the separation between the true spots and background after running the spot detection software was less obvious. Please click here to view a larger version of this figure.

Figure 5: Expression of NDC80^luti and NDC80^ORF transcripts are temporally controlled. (A) Representative smFISH images of NDC80^luti and NDC80^ORF during vegetative growth and meiosis. Vegetative samples were taken when cells (UB8144) were growing exponentially in nutrient rich medium. Meiotic prophase samples were taken 6 hours after cells (UB8144) were transferred to sporulation medium, a time when these cells were arrested in meiotic prophase. The Q 670 probes (shown in green) hybridize to the common region shared between NDC80^luti and NDC80^ORF mRNAs, whereas the CF 590 probes (shown in magenta) hybridize to the unique 5' region of NDC80^luti. DNA was stained with DAPI (shown in blue). Each cell was staged by its Zip1-GFP signal, a marker for meiotic prophase. Our smFISH protocol preserves strong GFP signal without further modification. Vegetative growth: Zip1-GFP negative. Meiotic prophase: Zip1-GFP positive. Images are shown as the maximum-intensity projections of z-stacks. Scale bar: 5 µm. This figure is modified from Figure 2C of Chen et al.¹⁶. (B) NDC80^ORF, NDC80^luti, and Ndc80 protein (Ndc80p) level during meiosis (UB4074). NDC80^luti and NDC80^ORF levels were determined by northern blot, and Ndc80p level was determined by anti-V5 immunoblot at the indicated time points. Hxk1p, loading control for immunoblot. As strain UB8144, this strain also harbors the pGAL-NDT80 GAL4-ER synchronization system^19,20. Cells were transferred to sporulation medium at 0 hour and released from pachytene arrest by β-estradiol addition 6 hours later. The NDC80^luti transcript was robustly detected in S/meiotic prophase, whereas the NDC80^ORF transcripts was predominantly present prior to meiotic entry (0 hour) and during the meiotic divisions (7-9 hours). This figure is modified from Figure 6J of Chen et al.¹⁶ Please click here to view a larger version of this figure.

Figure 6: Bootstrap analysis performed for the meiotic prophase samples shown in Figure 5. All of the quantified cells were pooled, and a given number (n) of cells were randomly sampled 500 times. The mean and 95% confidence interval were calculated for the fraction of paired and unpaired mRNA per cell. These data were plotted for each choice of the number n (Inset). The plateau in variance was visualized by plotting the sample variance divided by the mean, as a function of the number of cells (n) sampled. The total number of cells measured (437 cells) greatly exceeded the number at which the error plateaued (~60 cells), indicating that additional data would not improve the confidence in the measurements. Please click here to view a larger version of this figure.

Figure 7: Quantification of the smFISH data shown in Figure 5, graphed as the relative frequency histograms of the cells with a given number of NDC80^luti and NDC80^ORF transcripts per cell, using the data pooled from three independent experiments. The dashed line indicates the median number of NDC80^luti and NDC80^ORF transcripts per cell. Each histogram was normalized so that the maximum bin height was the same across all of the histograms. A total number of 637 cells were analyzed for vegetative growth and 437 for meiotic prophase. Two-tailed Wilcoxon Rank Sum test was performed for NDC80^luti and NDC80^ORF, respectively, comparing vegetative and meiotic prophase samples. This figure is modified from Figure 2D of Chen et al.¹⁶ Please click here to view a larger version of this figure.

Table 1. Buffers and media