Cell culture

Human bronchial epithelial cells (HBEC3-KT), derived from normal human bronchial tissue, were immortalized through the stable introduction of expression vectors that carry the genes for human telomerase reverse transcriptase (hTERT) and cyclin-dependent kinase-4 (CDK4) as described (Ramirez et al. 2004). HBEC3-KT were cultured in keratinocyte serum-free medium (Thermo Fisher Scientific, cat. no. 17005042) supplemented with bovine pituitary extract as per manufacturer’s instructions (Thermo Fisher Scientific, cat. no. 13028014), human growth hormone (Thermo Fisher Scientific, cat. no. 1045013), and 50 U/mL penicillin/streptomycin (Thermo Fisher Scientific, cat. no. 15070063).

Human foreskin fibroblasts (HFF), immortalized with hTERT (Benanti and Galloway 2004), were cultured in DMEM (Thermo Fisher Scientific, cat. no. 10569010) supplemented with 10% fetal bovine serum (Thermo Fisher Scientific, cat. no. 10082147) and with 50 U/mL penicillin/streptomycin (Thermo Fisher Scientific, cat. no. 15070063).

All cell lines were maintained at 37 °C with 5% CO2 and were split twice a week at a ratio of 1:4. The cells were plated in 384-well imaging plates (PhenoPlate 384-well, Revvity, cat. no. 6057500) and allowed to grow overnight until reaching approximately 80% confluency for the experiments. The seeding density per well for each cell line was optimized based on established protocols (Finn and Misteli 2021). The cells were then fixed in 4% paraformaldehyde (PFA; Electron Microscopy Sciences, cat. no. 15710) in phosphate-buffered saline (PBS; Millipore Sigma, cat. no. D8537) for 10 min. Postfixation, the plates underwent three PBS washes and were subsequently stored in PBS at 4 °C for subsequent DNA FISH procedures.

FISH probes

Following established FISH protocols (Shachar et al. 2015; Hart et al. 2015; Finn and Misteli 2021), we employed BAC FISH probes RP11-717D13 and RP11-98C17 (BACPAC Resources Center) to target the downstream regions of the MYC or EGFR genes on human chromosomes 8 and 7, respectively (Supplementary Fig. 1). The generation of fluorescently labeled BAC probes involved nick translation optimized based on previously described protocols (Finn and Misteli 2021). Briefly, nick translation was performed at 14 °C for 80 min, utilizing a reaction mixture consisting of 40 ng/mL DNA, 0.05 M Tris–HCl pH 8.0 (Thermo Fisher Scientific, cat. no. 15568025), 5 mM MgCl2 (Quality Biological, cat. no. 351033721), 0.05 mg/mL BSA (Millipore Sigma, cat. no. A9418), 0.05 mM dNTPs (Thermo Fisher, dATP: cat. no. 10216018; dGTP: cat. no. 10218014; dCTP: cat. no. 10,217,016) including fluorescently tagged dUTP (Dyomics, DY488-dUTP: cat. no. 488–34), 1 mM β-mercaptoethanol (Bio-Rad, cat. no. 1610710), 0.5 U/mL Escherichia coli DNA polymerase I (Thermo Fisher Scientific, cat. no. EP0042), and 0.5 mg/mL DNase I (Roche, cat. no. 11284932001). The reaction was halted by adding 1 μL of 0.5 M EDTA (Thermo Fisher Scientific, cat. no. 15575020) per 50 μL reaction volume, followed by 10 min heat shock at 72 °C. The resulting nick-translated probe was run on a 2% agarose gel for quality control to verify successful nick translation, indicated by a smear smaller than 1 Kbp (Finn and Misteli 2021).

The nick-translated probe was then ethanol-precipitated to concentrate it and to remove any residual DNase and DNA polymerase activities. The probe was resuspended in a solution containing 38 ng/μL human Cot-1 DNA (Millipore Sigma, cat. no. 11581074001), 256 ng/μL yeast tRNA (Thermo Fisher Scientific, cat. no. AM7119), and 0.1 M Sodium Acetate (Thermo Fisher Scientific, cat. no. R1181) in prechilled (–20 °C) 70% ethanol. The mixture was vortexed, spun for 1 min at 20,000g, and then chilled for at least 60 min at –20 °C. Before hybridization, the mixture was spun again at 4 °C for 30 min at 20,000g, the supernatant was discarded, and the pellet was air-dried for 10 min. Finally, the probe was resuspended in the hybridization buffer, as described below.

As an alternative approach, we used commercially available fluorescently labelled BAC probes (Empire Genomics). For our experiment, we used the same RP11 BAC probes tagged with green 5-fluorescein conjugated dUTP (Empire Genomics).

For nascent mRNA FISH targeting intron 1 of either MYC or EGFR, we used Stellaris® RNA Probes (LGC Biosearch Technologies). These probes consist of 48 single oligonucleotides, each 20 nucleotides in length labeled with Atto647N (Supplementary Fig. 1).

Simultaneous DNA/RNA HiFISH in 384-well plates

After PFA fixation and optional storage at 4 °C in PBS, cells were washed twice with PBS and permeabilized at RT for 20 min using 0.5% w/v saponin (Sigma-Aldrich, cat. no. 47036), 0.5% v/v Triton X-100 (Sigma-Aldrich, cat. no. X100), and 1× RNAsecure™ (Thermo Fisher Scientific, cat. no. AM7006) in PBS. Following two additional PBS washes, the cells were deproteinated for 15 min in 0.1 N HCl and neutralized for 5 min in 2× saline sodium citrate buffer (2× SSC) (Sigma Aldrich, cat. no. S6639) at room temperature. The cells were then equilibrated overnight in 50% formamide/2× SSC at 4 °C.

For hybridization, we combined 4 μL of 0.4 μg of precipitated DNA probe (or commercial Empire Genomics probe) with 0.5 μL of a 12.5 μM stock Stellaris® RNA probes and 5.5 μL of hybridization buffer which is made up of 30% formamide (pH 7.0), 10% dextran sulfate, 0.5% Tween-20, 2× SSC, 0.5× RNAsecure™ RNAse inhibitor, and 3% THE RNA Storage Solution (Thermo Fisher Scientific, cat. no. AM7001) dissolved entirely in molecular H2O. The hybridization mixture was shaken at 37 °C for 5 min. Samples were washed with prewarmed wash buffer (10% formamide in 2× SSC), followed by incubation at 37 °C for 10 min. Subsequently, 10 μL of the probe mixture was added to each well. The plate was centrifuged to eliminate bubbles, sealed, and denatured at 85 °C for 7 min using a ThermoMixer® C-PCR 384 (Eppendorf). After denaturation, the plate was immediately transferred to a 37 °C incubator for a 48-h hybridization.

After hybridization, the plate was washed several times, first in wash buffer (10% formamide in 2× SSC) at 37 °C for 1 h, then once at room temperature with 2× SSC, and then with 45 °C prewarmed 1× SSC and 0.1× SSC, each washed thrice for 5 min. Finally, DNA was stained with 3 mg/mL 4′,6-diamidino-2-phenylindole (DAPI) for 10 min, rinsed three times with PBS, and stored in PBS until the imaging step of the protocol.

Sequential DNA/RNA HiFISH in 384-well plates

After PFA fixation, the cells were permeabilized overnight at 4 °C with prechilled (−20 °C) 70% ethanol. Following ethanol removal, the cells were washed once with wash buffer containing 10% formamide in 2× SSC at 37 °C for 10 min. RNA hybridization was performed by adding 10 μL/well of 0.63 μM final concentration of Stellaris® RNA Probes in RNA FISH hybridization buffer (10% formamide, pH 7.0, 10% dextran sulfate, and 2× SSC). After adding the probe mixture, the plate was centrifuged, sealed, and incubated overnight at 37 °C.

After RNA hybridization, the plate underwent a series of washes: wash buffer (10% formamide in 2× SSC) at 37 °C for 1 h, followed by two consecutive room temperature washes with 2× SSC. The cells were stained with 3 mg/mL DAPI for 15 min, rinsed three times, and mounted in PBS for subsequent imaging on a high-throughput confocal microscope.

After image acquisition, a second permeabilization step was conducted using Triton/Saponin, as detailed in the simultaneous protocol. Following overnight formamide equilibration, 4 μL of 0.4 μg of precipitated DNA probe (or specified commercial source) was resuspended in 6 μL of DNA FISH hybridization buffer (50% formamide pH 7.0, 10% dextran sulfate, 1% Tween-20, 2× SSC in molecular H2O). 10 μL/well of the probe mixture was added to the plate and then centrifuged and sealed. Denaturation at 85 °C for 7 min was performed, followed by immediate transfer to a 37 °C incubator for a 48 h hybridization.

After hybridization, plates were rinsed once with 2× SSC at room temperature, followed by three rinses with 1× SSC and 0.1× SSC, all prewarmed to 45 °C. The cells were stained with 3 mg/mL DAPI for 10 min, rinsed thrice, and mounted in PBS for subsequent imaging on a high-throughput confocal microscope.

High-throughput image acquisition

High-throughput imaging was conducted using a Yokogawa CV8000 high-throughput spinning disk confocal microscope equipped with 405 nm (DAPI Channel), 561 nm (DNA probe channel), or 640 nm (RNA probe channel) excitation lasers. A 405/488/561/640 nm excitation dichroic mirror, a 60× water objective (NA 1.2), and 445/45 nm (DAPI channel), 525/50 nm (DNA probe channel), or 676/29 nm (RNA probe channel) bandpass emission mirrors were employed in front of a 16-bit sCMOS camera (2048 × 2048 pixels, binning 1×1, pixel size 0.108 microns). Z-stacks spanning 7 microns were acquired at 1-micron intervals and then maximally projected in real-time.

For imaging of simultaneous DNA/RNA HiFISH, the acquisition of all three channels for DNA, RNA, and DAPI immediately followed the completion of the FISH procedure. Conversely, sequential DNA/RNA HiFISH required two separate acquisitions: the initial acquisition occurred post RNA FISH to detect the RNA FISH and DAPI signals, followed by a subsequent acquisition post DNA FISH to detect the DNA and DAPI signals (Fig. 1). Typically, between 500–5000 alleles were imaged per sample. Edge wells were not used for imaging and no effects of well location on the plate were detected.

Image preprocessing

For the detection of DNA and RNA signals by simultaneous hybridization, images were directly used for image analysis as described below.

For the detection of DNA and RNA signals by sequential hybridization, the separate images generated by DNA imaging and RNA imaging required registration to align the DNA and RNA signals using the DAPI patterns. Images were subjected to an image registration algorithm based on the computation of the translation vector using cross-correlation techniques to align DNA and RNA images. The registration algorithm utilizes cross-correlation, a standard technique in signal processing, to determine the spatial translation required to align two images. In our case, these are the RNA and DNA images.

For two grayscale images \(A\) (representing the DNA image) and \(B\) (representing the RNA image), the cross-correlation \(C\) at a displacement \((\Delta x, \Delta y)\) is calculated as:

$$C(\Delta x, \Delta y) = \sum_\left(x,y\right)A\left(x, y\right)\times B\left(x + \Delta x, y + \Delta y\right).$$

Here, \(x\) and \(y\) are the pixel coordinates in the images, and the sum is taken over all pixels where \(A\) and the shifted \(B\) overlap.

The peak of the cross-correlation function, \(C\), indicates the displacement at which the images are best aligned. The coordinates of this peak \((\Delta _, \Delta _)\) represent the translation vector.

Once the translation vector \((\Delta _, \Delta _)\) is determined, it is applied to the RNA image to achieve alignment with the DNA image.

$$_ \left(x,y\right)=B\left(x- \Delta _, y- \Delta _\right).$$

In this equation, \(_\) is the spatially shifted RNA image, and the operation ensures that every pixel in \(B\) is moved according to the calculated translation vector. Boundary pixels are set to zero.

The image registration code is publicly available on GitHub: https://github.com/CBIIT/DNA_RNA_registration.

High-throughput image analysis

After image acquisition of simultaneous DNA/RNA FISH or after registration of separate RNA and DNA images for sequential FISH, we employed high-throughput image processing software (HiTIPS) (Keikhosravi et al. 2023) to analyze the image dataset consisting of DNA, RNA, and DAPI channels. Nuclei were segmented using the HiTIPS’s GPU-accelerated implementation of the CellPose deep learning segmentation model (Stringer et al. 2021), along with the Laplacian of Gaussian method for detecting FISH signals as described in (Keikhosravi et al. 2023). Before starting the analysis, these parameters were adjusted using real-time visual feedback provided by overlaying the results of the segmentation on the original images to maximize segmentation accuracy. Each plate was analyzed using specifically chosen parameters such as average cell size, Laplacian of Gaussian kernel size, thresholding method, etc., to maximize spot detection accuracy.

To calculate the radial position of FISH signals within the nucleus, first, a distance transform was calculated from binary images of nuclei. The distance transform returns the closest distance of each pixel from the background, meaning the pixels at the boundary have a value of 0, and the pixels at the center of the nucleus have the maximum value. The distance transform of each nucleus is then normalized separately and subtracted from 1. This will return 1 for the pixels at the periphery of the nucleus, which is farthest from the nucleus center, and 0 for the pixels at the center.

Single-cell and single-spot results were separately saved as flat text files for downstream data analysis. All image processing was done on the NIH HPC BioWulf cluster (https://hpc.nih.gov/) to maximize spot detection accuracy.

Data analysis

Data analysis was performed using R (The R Core Team 2024) and these R packages: tidiverse (Wickham et al. 2019), SpatialTools (Joshua French 2023), fs (Hester et al. 2023),

reshape2 (Wickham 2007), data.table (Barrett et al. 2024), and ggthemes (Arnold et al. 2024).

Briefly, single-cell data were read from several flat text files (one file per well) output by HiTIPS and concatenated in a single data frame for each experiment. The cells with nuclei smaller than 10 microns and a circularity value < 0.95 were filtered out as segmentation errors. FISH spot-level data were read from several flat text files (one file per well and per channel) output by HiTIPS and concatenated in a single data frame per experiment. Spot level data contain information on which nucleus/cell each spot belongs to. Spots that did not overlap with any nucleus in the image were filtered out.

Based on the spot level data, we calculated the number of spots for each cell and each channel. These discrete numerical values were further binned in a new variable for each channel that assumed the “0,” “1,” “2,” and “ >  = 3" values. Only cells with two DNA FISH spots and two or less RNA FISH spots were analyzed for downstream DNA FISH/RNA FISH Euclidean distances and for DNA FISH radial distance calculations or plots.

We used the X and Y coordinates for each spot to first calculate all the possible distances between the two DNA FISH spots and RNA FISH spots, if any, on a per-cell basis. We then calculated the minimum DNA FISH/RNA FISH Euclidean distance on a per DNA FISH spot basis. DNA FISH spots in cells that did not contain an RNA FISH spot were automatically assigned an “NA” value for the minimum DNA FISH/RNA FISH Euclidean distance. DNA FISH spots with a value of “NA” were classified as “no transcription.” DNA FISH spots with a DNA FISH/RNA FISH Euclidean distance below 1 micron, indicating proximity to an RNA FISH spot, were classified as “active.” Otherwise, the remaining DNA FISH spots were classified as “inactive.”

Normalized radial distance distributions between the active vs. inactive in simultaneous and the sequential protocols were compared using a two-sided Kolmogorov–Smirnov test.

Plots and tables from single experiments were generated by R, compiled in Microsoft Excel, and then organized into figures using BioRender.com.