We used live imaging to visualize the transcriptional dynamics of the Drosophila melanogaster even-skipped gene at single-cell and high temporal resolution as its seven stripe expression pattern forms, and developed tools to characterize and visualize how transcriptional bursting varies over time and space. We find that despite being created by the independent activity of five enhancers, even-skipped stripes are sculpted by the same kinetic phenomena: a coupled increase of burst frequency and amplitude. By tracking the position and activity of individual nuclei, we show that stripe movement is driven by the exchange of bursting nuclei from the posterior to anterior stripe flanks. Our work provides a conceptual, theoretical and computational framework for dissecting pattern formation in space and time, and reveals how the coordinated transcriptional activity of individual nuclei shape complex developmental patterns.

Generation of MS2 tagged eve BAC

We used bacterial recombineering (Warming et al. 2005) to modify a bacterial artificial chromosome (BAC) (Venken et al. 2006) containing the D. melanogaster eve gene and all of its enhancers and regulatory elements (BAC CH322-103K22) (Venken et al. 2009). We  replaced the coding region with an array of 24 MS2 stem loops fused to the D. melanogaster yellow gene (Figure 1B; (J. P. Bothma et al. 2014) as described below. We inserted our eve::MS2::yellow BAC-based construct in the D. melanogaster genome at chromosome 3L through ΦC31 integrase-mediated recombination (see Generation of fly lines), and generated a viable homozygous fly line (w-; +; eve::MS2::yellow) as detailed below. 

Reporter design

In principle the length of the reporter should not limit our ability to estimate burst parameters. However, in practice a reporter construct that is too short will have insufficient signal. Further, one that is too long will increase the dwell time of each RNA polymerase molecule on the gene and, as a result, our cpHMM inference will require too many computational resources. Our choice of reporter construct structure strikes a balance between these two limitations and is ideally suited for inferring bursting parameters in the time range where eve resides, as well as for boosting the signal-to-noise ratio. See  Lammers et al. (2020) for a more detailed discussion of reporter length-related tradeoffs. 

Specifics of recombineering 

We modified a CHORI BAC CH322-103K22 derived from (Venken et al. 2009), which contained the entire eve locus and a GFP reporter instead of the eve coding sequence (CH322-103K22-GFP). We replaced the GFP reporter with MS2::yellow (6665 bp) through a two step, scarless, galK cassette-mediated bacterial recombineering (Warming et al. 2005). Briefly, we transformed our starting CH322-103K22-GFP BAC into E.coli recombineering strain SW102. We then electroporated the strain with a galK cassette flanked by 50bp-long DNA homology arms homologous to the MS2::yellow (6665 bp) reporter. Upon electroporation, we selected transformants on M63 minimal media plates with galactose as a single carbon source. We achieved a correct replacement of GFP sequence by galK cassette in the BAC context (CH322-103K22-galK), validated by observing the digestion patterns produced by ApaLI restriction enzyme.

We next purified the CH322-103K22-galK BAC and transformed it into fresh E. coli SW102 cells. We electroporated these cells with the purified MS2::yellow insert and used M63 minimal media plates with 2-deoxy-galactose to select against bacteria with a functional galK gene. We used colony PCR to screen for colonies with a correct MS2::yellow insertion (CH322-103K22-MS2) replacing the galK cassette. We validated this insertion by observing ApaLI, XhoI, SmaI, and EcoRI restriction digestion patterns and through PCR and Sanger sequencing of the insertion junctions. We transformed our CH322-103K22-MS2 BAC in E.coli EPI300 cells to induce high copy numbers and purified it with a Qiagen plasmid Midiprep kit.

Generation of fly lines

We sent a sample of our purified CH322-103K22-MS2 BAC to Rainbow Transgenic Flies, Inc. for injection in D. melanogaster embryos bearing a ΦC31 AttP insertion site in chromosome 3L (Bloomington stock #24871; landing site VK00033; cytological location 65B2). We received the flies that resulted from that injection and used a balancer fly line (w- ; + ; +/TM3sb) to obtain a viable MS2 homozygous line (w- ; + ; MS2::yellow). We used line  (yw; His::RFP; MCP::GFP) as the maternal source of Histone-RFP and MCP-GFP (Garcia et al. 2013).

Embryo Collection and Mounting

Embryo collection and mounting was done as specified in (Garcia and Gregor 2018). In short, we set fly crosses between ~30 males (w-; +; eve::MS2::yellow) and ~80 females (yw; His::RFP; MCP::GFP) in a plastic cage capped with a grape juice agar plate. We collected embryos from cages two to ten days old by adding a fresh plate for 30 minutes and aging for 60 minutes to target embryos 90 min or younger.

Embryos were mounted on a gas-permeable Lumox Film (Sarstedt - Catalog # 94.6077.317) embedded on a microscope slide hollowed on the center. Then, we coated the hydrophobic side of the Lumox film with heptane glue and let it dry. The film allows oxygenation of embryos during the 2-3h long imaging sessions while heptane immobilizes them. 

We soaked an agar plate with Halocarbon 27 oil, picked embryos with forceps, and laid them down on a 3 x 3 cm piece of paper tissue. We dechorionated embryos by adding 2 drops of bleach diluted in water (5.25%) on the paper tissue and incubating for 1.5 minute. We removed bleach with a clean tissue and rinsed with ~4 drops of distilled water. We then placed the tissue paper with dechorionated embryos in water, and picked buoyant embryos with a brush.

We lined ~30 apparently healthy embryos on the Lumox film slide and added 2-3 drops of Halocarbon 27 oil to avoid desiccation, and covered the embryos with a cover slip (Corning® Cover Glass, No.1, 18 x 18mm) for live imaging.

Imaging and Optimization of Data Collection

Movies of embryonic development were recorded on a Zeiss-800 confocal laser scanning microscope in two channels, (EGFP: 488 nm; TagRFP: 561 nm). We imaged embryos on a wide field of view, along their anterior-posterior axis, of 1024 x 256 pixels (202.8µm x 50.7µm), encompassing 3-5 stripes per movie. We tuned laser power, scanning parameters, master gain, pinhole size and laser power to optimize signal to noise ratio without significant photobleaching and phototoxicity. 

For imaging, the following microscope settings were used: 63x oil-objective, scan mode ‘frame’, pixel size of 0.2µm, 16 bits per pixel, bidirectional scanning at a speed of 7, line step of 1, laser scanner dwelling per pixel of 1.03µs, laser scanner averaging of 2, averaging method Mean, averaging mode Line, 488 nm laser power of 30µW (EGFP), 561 nm laser power of 7.5µW (TagRFP) (both powers were measured with a 10x air-objective), Master Gain in EGFP detector of 550V, Master Gain in TagRFP detector of 650V, Digital Offset in both detectors of 0, Digital Gain in both detectors of 1.0, and a pinhole size of 1 airy unit under the imaging conditions mentioned above (44µm, 0.7µm/section), laser filters EGFP:SP545 and TagRFP:LBF640. This resulted in an imaging time of 633 ms per frame and a full Z-stack of 21 frames in intervals of 0.5µm every 16.8s. Following (J. P. Bothma et al. 2014, 2015, 2018; Lammers et al. 2020), the imaging conditions were determined not to affect normal development as reported by the timing of the nuclear cycles in early development. We stopped imaging after 50 min into nuclear cycle 14, and took mid-sagittal and surface pictures of the whole embryo for localization of the recorded field of view along the embryo’s AP axis.

Image processing

We used a Matlab computational pipeline based on (Garcia et al. 2013; Lammers et al. 2020) to segment and extract numeric data from our raw movies. Briefly, this software segments and processes the images from the two channels (channel 1: MCP::GFP, channel 2: Histone::RFP) on which we collected our data. For segmentation of channel 1, we used Fiji-Weka Segmentation 3D software; this machine-learning-based method relies on the manual segmentation of a variety of MCP::GFP labeled transcriptional foci in a given 21 frame Z-stack from a single dataset (EVE_D11) to produce a model for the segmentation of all datasets recorded under the same imaging conditions. Next, we segmented and tracked the Histone::RFP labeled nuclei on channel 2. Subsequently, we assigned MCP::GFP labeled transcriptional foci to their corresponding Histone::RFP labeled nuclei. Since we collected whole embryo pictures of each of our datasets, we were able to match and locate the recorded fields of view to their right position along the body of their corresponding embryos. Finally, we extracted position and fluorescence values over time of all transcriptional foci to generate data structures ready to use in further analyses.