Tracing DNA paths and RNA profiles in cultured cells and tissues with ORCA

Leslie J Mateo; Nasa Sinnott-Armstrong; Alistair N Boettiger

doi:10.1038/s41596-020-00478-x

. Author manuscript; available in PMC: 2021 Oct 19.

Published in final edited form as: Nat Protoc. 2021 Feb 22;16(3):1647–1713. doi: 10.1038/s41596-020-00478-x

Tracing DNA paths and RNA profiles in cultured cells and tissues with ORCA

Leslie J Mateo ¹, Nasa Sinnott-Armstrong ², Alistair N Boettiger ^1,^✉

PMCID: PMC8525907 NIHMSID: NIHMS1743237 PMID: 33619390

Abstract

Chromatin conformation capture (3C) methods and fluorescent in situ hybridization (FISH) microscopy have been used to investigate the spatial organization of the genome. Although powerful, both techniques have limitations. Hi-C is challenging for low cell numbers and requires very deep sequencing to achieve its high resolution. In contrast, FISH can be done on small cell numbers and capture rare cell populations, but typically targets pairs of loci at a lower resolution. Here we detail a protocol for optical reconstruction of chromatin architecture (ORCA), a microscopy approach to trace the 3D DNA path within the nuclei of fixed tissues and cultured cells with a genomic resolution as fine as 2 kb and a throughput of ~10,000 cells per experiment. ORCA can identify structural features with comparable resolution to Hi-C while providing single-cell resolution and multimodal measurements characteristic of microscopy. We describe how to use this DNA labeling in parallel with multiplexed labeling of dozens of RNAs to relate chromatin structure and gene expression in the same cells. Oligopaint probe design, primary probe making, sample collection, cryosectioning and RNA/DNA primary probe hybridization can be completed in 1.5 weeks, while automated RNA/DNA barcode hybridization and RNA/DNA imaging typically takes 2–6 d for data collection and 2–7 d for the automated steps of image analysis.

Introduction

Technical challenges have limited the application of Hi-C methods in complex tissues, where key biological processes are believed to be regulated by chromatin structure. The challenges arise because achieving high resolution with these sequencing-based assays requires: (i) relatively large numbers of homogenous cells, limiting application to rare-cell populations; (ii) dissociation of cells, leading to loss of spatial context; and (iii) nuclei extraction, leading to loss of RNA transcripts, which are the readouts of enhancer function. Microscopy studies using fluorescent in situ hybridization (FISH) have revealed various cell type–specific enhancer-promoter interactions in tissues^1–4, but largely achieve lower resolution than sequencing methods⁵. In addition, FISH is limited by the number of fluorescent channels available and, thus, allows only for observation of a targeted handful of loci. Super-resolution imaging using stochastic optical reconstruction microscopy (STORM)^6,7 provides further detail on the spatial structure of domains ranging from several kilobases to several mega-bases^2,8–11. However, the globular structures resolved via STORM cannot be used to distinguish sequence features, such as enhancer and promoter, or genomic positions.

Both sequencing and imaging approaches typically sacrifice analysis of RNA transcripts when measuring DNA contacts, which limits the ability to study how chromatin structure and transcription correlate at the cellular scale. Several studies have relied on correlating these properties at the bulk level by using separate assays, such as Hi-C for mapping cis-interactions and RNA-seq for characterizing the transcriptional state^12–23. However, at the single-cell level, both structure and expression vary substantially from cell to cell²⁴. Thus, we developed a method that reveals high-resolution single-cell chromatin structure for thousands of individual cells within cryosectioned tissue, which we have termed ‘optical reconstruction of chromatin architecture’ (ORCA)²⁵ (Fig. 1). A key feature of our approach is its compatibility with multiplexed RNA labeling of both mature and nascent transcripts^25–32, allowing the interaction of transcription and genome structure to be measured at the single-cell level. The protocols we describe here for DNA and multiplexed RNA labeling can be performed separately or sequentially, providing both measurements within the same individual cells.

Fig. 1 | — This figure was adapted from our previously published work¹³. a, In a cryosectioned *Drosophila* embryo, the bithorax complex (BX-C) region is tiled with primary probes. These probes partition the region of interest into 52 3-kb genomic segments and each segment contains a unique barcode. b, ORCA relies on sequential hybridization rounds to label and image every barcode through the process of adding a complementary readout oligo bound to a fluorophore, removing the readout probe after imaging it, and repeating this process for all 52 barcodes. The top represents a schematic of this process and the bottom illustrates example data from imaged barcodes. The 3D centroid of each labeled barcode, seen as a white spot, is shown as ‘+’ symbols on the spots. c, 3D positions of the labeled barcodes are seen as spheres and are pseudocoloured. Arrows are pointing at the first (‘1’) and last (‘52’) labeled barcodes. The spheres are linked together to generate a 3D polymer for a single cell.

Overview of the multiplex RNA labeling and DNA tracing procedure

Both multiplex RNA labeling and DNA tracing with ORCA begin with designing an Oligopaint library^33–35 (Fig. 2, Step 1A). For DNA probes, first select a genomic region of interest and an intended step size (e.g., 10 or 2 kb). Using the corresponding reference genome (e.g., dm3, mm10) and a list of abundant repetitive sequences to exclude, select ≥20 primary probes with a length of 40 bp per genomic step to hybridize onto the genome. Although prior multiplex single-molecule RNA FISH protocols have relied on shorter hybridization regions^{26–28,30–32}, in our hands, longer sequences have provided more robust detection. When selecting primary probes, make sure to avoid repetitive regions and oligonucleotide properties known to reduce hybridization specificity. Examples of such off-target lists are provided at https://github.com/BoettigerLab/ORCA-public/tree/master/probe-building. Selection of oligos targeting a genomic region while respecting filter conditions is a routine filtering task, which can be facilitated for rapid probe construction through recently described algorithms^{9,25,34,36,37}.

For DNA probes, if the target genomic region overlaps with transcribed regions, we recommend designing probes that match the sense strand, to avoid hybridization to RNA (Step 1B). Although all RNA probes should be anti-sense to the RNA transcript, for nascent RNA probes, we recommend targeting 5′ ends of the introns, because these are the first to be transcribed and will appear closer to the promoter firing event, which is especially important for long genes. After selecting the 40mer targeting sequences, these need to be concatenated with a 20mer barcode sequence and a common 20mer ‘fiducial-binding’ sequence. Whereas the barcode sequence is unique for each RNA species or, for DNA, each genomic step size, the fiducial-binding sequence is the same for all probes. Finally, unique primer sequences added to either end allow for amplification and multiplexing of different DNA/RNA probe templates in the same commercial oligonucleotide pool. RNA and DNA probesets should be given distinct primers so that they may be synthesized separately. Combining probesets into the same oligonucleotide pool is useful for reducing costs (most vendors sell fixed increments of unique oligos in a complex pool and have decreasing costs for larger pools). Example primer sequences, 384 unique barcode sequences and scripts to aid in design of the oligonucleotide pools are provided at https://github.com/BoettigerLab/ORCA-public/tree/master/probe-building.

Because the unique barcodes are sequentially imaged for a given genomic region of interest, the following types of oligos must be generated: (i) oligos that bind to the barcode for imaging and (ii) oligos that remove any previous barcode signal. These oligos are referred to as the ‘readout’ probe and the ‘strand-displacement’ oligo, respectively (see https://github.com/BoettigerLab/ORCA-public/tree/master/probe-building for readout and strand-displacement sequences). The 30mer strand-displacement oligos are designed first, and the 20mer readout sequences are designed afterward to be complementary to these sequences. When a strand-displacement oligo is introduced, it will bind to its complementary readout sequence, which will result in the removal of the readout probe from the barcode. In addition, strand-displacement sequences should not bind to the genome and should not form hairpin structures.

Primary probes are synthesized (Fig. 2, Steps 2–30) from the oligopool of probe templates by using non-saturating PCR to isolate and amplify the sequences, and are further amplified to high concentration probes through in vitro transcription. A reverse transcription (RT) reaction converts the RNA product to single-stranded DNA (ssDNA). After a final cleanup, DNA can be stored at 4 °C or frozen until needed for hybridization. This protocol produces a high concentration of ssDNA products while requiring only small-volume reactions³⁸, which lowers costs (Supplementary Data 1).

Because thin samples allow for better detection of fluorescent oligos, which bind to the readout probes, fixed Drosophila embryos (Steps 31–49) should first be cryosectioned at 6–10 μm. These tissue sections are adhered to positively charged cover glasses before labeling (Fig. 2, Steps 50–75). Cultured cells may be grown on cover glasses or plated on cover glasses from suspended cells and fixed (Steps 76–78).

For combining RNA with DNA measurements, RNA labeling and imaging (Steps 79–143) are conducted before DNA labeling (Steps 144–166) because the latter involves genomic DNA denaturation (Fig. 2). Performing RNA FISH first prevents cross-talk of RNA probes binding their genomic DNA complements. RNA labeling begins with a brief membrane permeabilization followed by probe and target RNA denaturation on a hot plate. Heat denaturing facilitates probe binding and alleviates the need for protease treatments to remove proteins that can mask RNA binding³⁹. The sample is hybridized overnight in an enclosed humidified chamber. Before imaging, a 96-well plate containing fluorescently labeled readout probes that bind to unique barcodes of the primary probe is placed into our homebuilt robotic fluidics system (Steps 95–103 and 167). The first well contains a Cy3-labeled fiducial oligo, which binds to the fiducial sequence in all primary probes. In addition, all wells—with the exception of the first one—contain the strand-displacement oligo that will remove the previous readout oligo. The sample is mounted in a flow chamber (Steps 104–113 and 168) connected to a robotic fluidics system for sequential readout probe hybridizations (see Supplementary Data 2 for information regarding the cost of each round of readout probe hybridization). Once the flow chamber is secured to a motorized stage on the microscope, the fields of view (FOVs) are manually chosen for imaging. Both sequential readout hybridization and barcode imaging are performed using automated software (Steps 114–143; Supplementary Data 3 has details of imaging time). See the Troubleshooting table (for Step 142) for our recommendations on validating RNA labeling.

After imaging of RNA species, the sample is removed from the microscope, and genomic DNA is hybridized with DNA probes (Fig. 2, Steps 144–166). The sample may be treated with RNase after RNA imaging and before DNA hybridization to avoid cross-hybridization; however, we find this is not essential if only sense-strand probes are made. Although DNA labeling is similar to the procedure for RNA labeling, it has an additional and crucial genomic DNA denaturation step, which is induced by addition of HCl after permeabilization and before heating. In addition, after overnight incubation, the sample is post-fixed in glutaraldehyde solution for further stabilization of 3D genome structure. The FOVs selected for imaging are the same as those that were previously imaged for RNA species. Sequential readout hybridization and barcode imaging are also performed using the automated software (Steps 169–198). See Supplementary Data 3 for details on how long it takes to complete an imaging experiment.

Additional counterstains, such as DAPI, nuclear lamina labeling or other antibody staining, may be added at any point after adhering the tissue/cells onto the coverslip, as long as the associated fluorescent probes do not overlap spectrally with those being used for imaging the readout probes. The efficiency of antibody labeling after RNA and/or DNA sample preparation and labeling is antigen specific; some antigens denature during the hybridization process. In general, adding and imaging counterstains before RNA/DNA labeling will provide the best signal, whereas imaging in parallel with the fiducial label (Steps 143 and 198) will provide the best registration.

The raw images are analyzed using our freely available graphical user interface (GUI)–based tools written in MATLAB, which use an ‘Experimental Table’ to distinguish between fiducial and readout data images (Fig. 2, Steps 199–233). Images are first corrected for x-y stage drift, which occurs between imaging rounds because of microscope stage movement, using Cy3 fiducial oligo data. The fiducial signal is used for registration purposes. Individual nuclei are cropped using supervised, automated segmentation and then further corrected for sub-pixel drift in the 3D image volume using fiducial data. Then, the images of each DNA barcode are fitted in x, y and z using a 3D-Gaussian spot-fitting algorithm. The nanoscale 3D positions of each barcode are recorded in a data table. This process occurs for every labeled nucleus throughout all FOVs. The degree of unspecific background signal is sample and probe set dependent. It is recommended that the user manually inspect the data and the quality of fits for a randomly selected subset of labeled chromosomes to evaluate the potential for fitting errors. We provide a GUI system with instructions to facilitate this process (see ChrTracer3—image analysis of DNA structure in the Procedure). See Supplementary Data 3 for details about the time it takes to complete the analysis of the raw images.

Mosaics are created for both RNA and DNA data where FOV-to-FOV drift is corrected using fiducial-labeled nuclei that overlap between adjacent FOVs (Fig. 2, Steps 235–255). The RNA and DNA mosaics are aligned on top of one another using the Cy3 fiducial oligo signal and the autofluorescence of the sample (Fig. 2, Steps 256–275). Lastly, once RNA and DNA information has been compiled into a single mosaic, regions of interest can be manually selected to analyze chromatin structure for a subset of cells, and RNA transcripts can be counted (Fig. 2, Steps 276–301). The user may also wish to consult various detailed protocols on RNA transcript counting^{28,32,40–45}.

We recommend that first-time users start with a downscaled experiment when learning the approach. For example, we recommend the use of adherent cells before using cryosectioned tissue samples. In addition, users can first start with a smaller number of barcodes when targeting RNA species and/or DNA—such as 5 for RNAs and 20 for DNA—when adopting the method, and then scale up to more complex experiments (e.g., 50–100 barcodes for DNA). For practical time considerations, we typically aim for ≤100 total barcodes in a DNA labeling experiment (Supplementary Data 3). For DNA labeling, the number of barcodes informs the choice of resolution—e.g., a 1-Mb domain can be imaged at 10-kb resolution with 100 barcodes, or a 200-kb domain at 2-kb resolution. To simplify the experiment, the user should image only 5–10 FOVs. With these recommendations, the user should take 1.5 d to synthesize primary probes, 3.5 d to hybridize and image RNA and DNA and <3 d to analyze the images. Thus, this starting experiment should take ~1 week. It is also advisable to try RNA and DNA labeling as separate experiments when learning the method before combining both on the same sample. Lastly, we recommend that users attempt their first experiment from our validated primary probe list found here: https://bit.ly/3fdwBx0. We have included fasta files of primary probes targeting RNA and DNA that have worked previously in Drosophila embryos and mammalian cultured cells^25,46,47.

Comparison to other methods

Bulk methods for analysis of chromatin structure, including chromosome conformation capture methods such as Hi-C (Fig. 3a) or ligation-free alternatives such as split-pool recognition of interactions by tag extension⁴⁸ and genome architecture mapping⁴⁹, have the ability to provide a genome-wide view of chromosome organization⁵⁰. However, these methods do not capture variation at the single-cell level. In contrast, ORCA provides access to high-resolution structural information from single cells for targeted genomic regions (Fig. 3b). In addition, because ORCA is compatible with RNA labeling, we have the ability to measure important co-localized features, such as transcript levels and cell position within a tissue, as well as the ability to directly connect these cell identity features with genome structure. Although single-cell Hi-C (scHi-C) combines whole genome coverage with cellular resolution^51–56, it loses access to these important cellular features that may drive variation in structure (Fig. 3b). In addition, scHi-C provides significantly lower genomic resolution of chromatin features compared to Hi-C or ORCA (Fig. 3a,b). This limitation arises not only from limited sample recovery, which might be improved, but also from inherent limitations from ligation-based proximity assays. In the dense nuclear environment, multiple sequences are probably within ligation distance of each restriction site, yet only one pair may ligate, intrinsically losing multi-way contact information and reducing the resolution of the data. Restraint-based modeling has been used to augment the resolution of scHi-C^52,55 but requires additional assumptions about the typical physical extent of the chromatin, which ORCA does not.

Compared to using traditional super-resolution approaches to image Oligopaint-labeled chromatin, such as sequential STORM^6,46,57 or DNA-PAINT^8,58,59 (Fig. 3c), ORCA provides a 3D polymer path rather than a ‘cloud’ of points arising from single-fluorophore localization events (Fig. 3d)³⁸. The steps in this polymer path represent the centroid of multiple fluorophores, all concentrated within a small region of the nucleus (e.g., a specific 2-kb genomic segment), rather than the centroid of individual fluorophores detected through STORM. Adjacent ‘steps’ along the genome are resolved in physical space at better than diffraction limit because they are sequentially detected, much like the sequential detection of adjacent fluorophores in STORM and DNA-PAINT. Moreover, because detection events are not stochastic, each detection event contains information about genomic sequence identity as well as the physical location (Fig. 3c,d).

Because sequential hybridization is much slower than stochastic switching of fluorophores, it takes much longer to reconstruct the image of a domain (days compared to minutes). However, ORCA has the capability to image an abundant number of cells because of its compatibility with large illumination volumes and large detectors. Thus, ORCA allows for the processing of chromosome paths from many cells in parallel, enabling us to access much larger cell numbers than we have with STORM^9,25. Data from both ORCA and STORM/DNA-PAINT can resolve ‘shape-properties’, such as density/compaction, eccentricity and the overlap of domains (Fig. 3c)^{2,9–11,20,60}. Mapping sequence-specific features, such as topologically associated domains (TADs) or enhancer-promoter loops, requires the sequence specificity provided by the sequential labeling of barcodes. Notably, sequential labeling of barcodes has also been used in conjunction with STORM^46,57, providing access to shape-properties within each barcode and genomic resolution at the scale of the barcoding (typically, 30–5,000 kb). Individual barcodes can be imaged much faster in ORCA because movies of blinking fluorophores are not required. In contrast, STORM images require ~100-fold more images to ensure sparse localization^9,35,46,57, which increases the challenges and costs for data storage and analysis.

In ORCA, the resolution is limited by the genomic step size and precision with which the barcode label can be imaged and aligned with the rest of the domain. We have successfully imaged tiles with step sizes as fine as 2 kb and as early as blastoderm stage Drosophila nuclei. Because of our strand-displacement strategy to remove the signal from the previously imaged barcode, rather than purely photobleaching, we can re-label the same barcode. By re-imaging a subset of barcodes at the end of the ORCA experiment, it is also possible to measure the localization precision and resolution of the experiment. We find the resolution typically to be ~20–35 nm in x/y and ~40–60 nm in z²⁵, similar to our results from the centroids of features measured with STORM⁴⁶. Overall, ORCA allows for faster imaging per genomic step but measures only the centroid of this step. The approach achieves super-resolution only when the barcodes are adjacently tiling short genomic regions (e.g., kilobases, not 100s of kilobases), and when drift and any chromatic aberrations have also been corrected to the nanoscale. Alternatively, larger probes spanning larger gaps of DNA have been effectively used to map chromosome structure at large length scales, with correspondingly reduced resolution, including entire chromosomes with sequential labeling⁶¹ or in separate multicolor experiments spanning a significant portion of the genome⁶².

Many of these advantages of ORCA are captured by a conceptually similar approach developed in parallel called ‘high- throughput, high-resolution, high-coverage microscopy-based technology’ (Hi-M)^63,64. This approach covered larger genomic distances with fewer probes (targeting ~21 genomic loci) by using variable-sized gaps between the probes. Avoiding these variable gaps in ORCA achieves a higher resolution view of the chromatin path that is more robust to the stochastic missed detection events. Regular genomic spacing between barcodes rather than variable spacing also facilitates interpretation and direct comparison with Hi-C data. In addition, the use of strand-displacement probes in ORCA, in place of bleaching or fluorophore-cleavage in other sequential FISH studies^{27,28,34,61,63,65,66}, allows individual signals to be repeatedly extinguished and then relabeled.

The ability to relabel the same genomic region at different times throughout the experiment (or repeat the entire experiment on the same cell if desired) provides a valuable control for whether sample chromatin structure changes during imaging and allows for accurate quantification of the resolution. Note that an initial photobleaching step before fluorescent labeling is still helpful to reduce auto-fluorescent background, and a rapid photobleaching between imaging different barcodes helps suppress the return of autofluorescence without extinguishing readout signal. The use of readout sequences instead of directly labeled barcodes^25,46 dramatically reduces the experimental cost for ORCA compared to Hi-M. Similar readouts or ‘bridges’ have also been used in STORM⁵⁷. In contrast to the approach described here using 6-μm cryosections, the Hi-M protocol demonstrated robust detection in the epithelial nuclei of whole Drosophila embryos (~10-μm layer around an ~100-μm embryo), overcoming the substantial background and signal aberrations associated with thicker samples.

Simultaneous labeling of DNA and RNA is historically difficult because of different hybridization conditions required for labeling^67,68 and the potential for RNA probes to also bind the complementary DNA sequences from which they were transcribed. Some prior protocols have addressed these challenges by compromising hybridization conditions to find a good balance to allow labeling of abundant RNAs (e.g., XIST, a highly expressed noncoding RNA involved in X-silencing) and repetitive DNA^33,67,69. Other groups cross-link RNA probes with amines or antibodies before DNA hybridization or imaging, which also works best with abundant transcripts^63,64,68.

Our approach circumvents the compromise in simultaneous RNA-DNA labeling through sequential rounds of labeling and imaging, similar to early multi-label methods⁷⁰. RNA is labeled first under hybridization conditions optimal for RNA labeling that do not denature the DNA helix and expose it to hybridization. Multiple RNA species can be labeled at once and distinguished through unique barcodes²⁵ or combinations of barcodes²⁸. After the imaging is complete, the sample is then removed from the microscope, and probes are hybridized to the DNA in an optimal DNA hybridization condition. Distinct barcodes are used with DNA and RNA labeling. The use of fiducial markers and automated alignment software minimizes the registration problem that made earlier implementations of a sequential imaging approach challenging⁷⁰. Successful avoidance of cross-talk is most easily assessed with intronic RNA probes targeting nascent transcripts and when the sample contains distinct cell populations that do and do not express the target gene. Bright puncta should be detected only in a subset of the cells that express the target, and none of the cells that do not (Extended Data Fig. 1). DNA probes target the sense strand to avoid potential hybridization to RNA targets, which can be validated with RNase treatment.

Limitations

Labeling of DNA with primary probes requires denaturation of the DNA double helix. To achieve robust probe hybridization, DNA must be sufficiently denatured through separate treatments with acid, formamide and heat. These conditions also denature nuclear proteins. Consequently, at the molecular scale, chemical structures have been disrupted. It is not well understood to what extent structural features at the super-resolution scales measured by ORCA (tens to hundreds of nanometers) are also disrupted by these denaturing conditions, because we cannot yet directly image and measure the structural features seen in ORCA without denaturation. Notably, some features that can be measured by other methods, such as the boundaries of TADs and the relative pairwise contact frequencies among domains, agree quantitatively with non-denaturing methods such as Hi-C^25,46. Importantly, denaturation is carried out in a densely cross-linked protein gel created by fixation before perturbation. Fixation will limit the mobility of the denatured proteins and DNA strands, and as long as they remain confined within the resolution limit of imaging, this perturbation will not significantly affect the nanoscale measurement of chromatin structure. Emerging alternatives to acid, formamide and heat, such as Cas9-mediated DNA targeting⁷¹ or the use of exonucleases to remove one strand¹⁶, may reduce the degree of perturbation.

We look forward to further improvements in this area. Ideally, such methods should be tested for sequences and chromatin state bias^9,25. Furthermore, existing and emerging alternative methods should be tested for their effect on the kilobase-scale chromatin path by measuring it before and after treatment. This requires the ability for reliable super-resolution imaging of the path of DNA before and after denaturation, which to our knowledge has not been achieved, although some creative proxies have been used⁷². Improvements in correlative light and electron microscopy and in super-resolution live imaging will probably improve our understanding of these perturbations and the ways they can be minimized in the future.

Denaturation creates several additional limitations to ORCA. If DAPI and/or fluorescent dyes are introduced to the sample before ORCA, the signal intensity of these additional labels is notably reduced after ORCA. Because our protocol involves acid and heat treatment of the sample, many antigens may unfold and reduce or disrupt recognition by antibodies (although some chromatin-associated proteins are compatible with DNA FISH). Examples include geminin, a cell cycle marker⁴⁶; fibrillarin, a nucleolar protein; and SC35, a splicing factor associated with nuclear speckles⁷³.

Failure to accurately correct drift, contamination of readout probes or improper dilution and mixing of readout oligos or strand-displacement oligos can lead to artifacts in the data. With proper controls and experience, such artifacts can be readily detected, as shown in the Troubleshooting table.

Raw data require considerable storage. For example, a single experiment with ~100 FOVs (each ~200 × 200 μm) using 100 barcodes and three barcode replicates requires >16 TB of storage (Supplementary Data 4). Large storage systems can be a challenge to maintain, and storage costs are easily in the range of two orders of magnitude greater than the cost of the chemical reagents. As with the analysis of the raw imaging data from Illumina sequencing, processed data result in significant compression, providing an avenue by which storage costs might be reduced in the future by limiting retention of image data after processing. Read/write operations on large datasets are also the primary limitation to the speed of data analysis. Consequently, processing data from external hard drives connected by USB to the analysis computer should be avoided, because USB achieves a slower read-write speed than internal serial ATA (SATA) or external high-speed network attached storage systems.

Another important limitation is the detection efficiency for individual barcodes, which affects both RNA and DNA labeling. Because of the stochastic nature of probe binding and photon emission, among other effects, individual genomic steps along the ORCA trajectory are not detected in some cells. For the barcodes provided in our GitHub repository (https://github.com/BoettigerLab/ORCA-public/tree/master/probe-building), we find that the efficiency is largely uniform across the sequences used, and shows no systematic variation with chromatin state or cell type in the domains that we examined (Extended Data Fig. 2). Typically, detection efficiency per barcode is in the 60–90% range, with the highest efficiency observed for larger step sizes and lower efficiency for smaller step sizes, as expected. This dropout rate has a minor effect on the number of cells one needs to analyze in a population to compute cell type–specific contact frequencies and identify cell type–specific boundaries, but does not otherwise impede these measurements. Similarly, the three-way or four-way interaction frequencies can be reliably measured with this barcode efficiency range⁴⁶. Care must be taken in the analysis that the missing data are excluded and not erroneously treated as a non-contact event. The analysis routines discussed below in the Procedure (see also Code availability) implement handling of missing data. An exciting future direction will be the development of improved approaches to impute some of the missed barcode positions from the collected data. Particularly for small genomic step sizes, there is a high correlation in the 3D position of adjacent sequences, which could be used to resolve otherwise ambiguous detection events that are discarded and counted as missed in the current approach.

Applications

The current ORCA protocol is suitable for both cryosectioned tissues and cultured cells, including adherent and suspension cell lines and mouse embryonic stem cell spheroids^25,46.

General considerations

This protocol details the approach originally described in our earlier report²⁵, in which we use ORCA to image the 3D DNA structure in nuclei of cryosectioned Drosophila embryos in combination with multiplexed RNA labeling. We include several updates reflecting recent improvements. These include adjustments for reading multiple barcodes per hybridization with spectrally distinct fluorophores and a wholly rebuilt pipeline for processing mosaic data made from overlays of multiple FOVs. We provide resources to guide users to build a robotic fluidics system to assist with sequential labeling of barcodes and a list of parts to replicate the optics system that we used in these experiments (Tables 1 and 2 and Supplementary Data 5). Manual pipetting can substitute for robotic pipetting and may be preferable for troubleshooting or for users with a small number of barcodes. We provide step-by-step instructions and recommendations for both wet laboratory and computational procedures. Our protocol is written for researchers with experience in microscopy and minimal computational skills. We provide MATLAB scripts that will be necessary to process all raw data (https://github.com/BoettigerLab/ORCA-public) along with a basic walkthrough of how to use this software. Sample data for testing the software are provided at https://bit.ly/2S6eCjk.

Table 1 |.

Details of the components used for the homebuilt fluidics system

Supplier	Part number	Quantity	Unit	Cost ($)	Description
Hamilton	7750–16	1	6/pk	71.00	25-gauge, Kel-F Hub needle, custom length (5 inches), point style 3, 6/pk
Gilson	F155006	1	Each	2,438.00	Peristaltic pump
Gilson	361832	1	Each	436.00	361832 361832: 508 interface module, 110–220 V
Gilson	F117933	1	Pk-12	41.00	PVC manifold tubing, 0.38-mm i.d. (0.015 inches), orange/green collars
Bioptechs	03060319–2-NH	1	Set	2,686.86	FCS2 chamber flow cell
Bioptechs	060319-2-2611	1	Each	970.26	Zeiss (K), Marzhauser, ASI, Ludl & prior adapter
Bioptechs	1907–750	0	Set-5	37.32	Blank gasket cut using die# 443792, 0.75-mm thick—5/pack—1907–250
Bioptechs	40-1313-03192	1	500 pk	522.00	40-mm coverslips
Amazon	N/A	1	Each	22.95	Keyspan by Tripp Lite USA-19HS high-speed USB serial adapter, PC, MAC, supports Cisco Break Sequence
Amazon	N/A	1	Each	9.99	Anker ultra slim 4-port USB 3.0 data hub
Amazon	N/A	1	Each	3.45	StarTech 6-feet straight through serial cable - DB9 F/F (MXT100FF)
Amazon	N/A	3	Each	5.79	AmazonBasics USB 3.0 extension cable—a-male to a-female—9.8 feet
Amazon	N/A	1	2-pack	8.99	CableCreation (2-pack) 3 feet 18 AWG universal power cord for NEMA 5–15P to IEC320C13 cable, 0.915M/black
McMaster	5583k52	0	Each	212.00	Abrasion-resistant white ETFE tubing, 0.020 inch i.d., 1/16 inch o.d., 50 feet
eBay	CNC3040T	1	Each	445.01	3 aXIS USB router engraver engraving drilling milling machine, 300 × 400 mm

Item	Cost ($)
Optics table and setup supplies	14,163.80
Lasers (405, 488, 560, 647 and 750 nm)	82,796.99
Laser optics for combing beams	10,534.50
CMOS camera	19,475.00
Microscope body and objectives	20,737.20
Microscope stage	30,852.20
AOTF	5,911.46
Focus lock	7,063.08
Laser-proof enclosure	691.27

Reagent	Quantity (μl)	Final concentration
ddH20	15	-
Phusion High Fidelity PCR Master Mix with HF buffer (2×)	25	1×
Forward primer (10 μM)	2.5	500 nM
Reverse primer with T7 promoter sequence (10 μM)	2.5	500 nM
Diluted Oligopaint library (−1 ng/μl)	2.5	−50 pg/μl
EvaGreen dye (20×)	2.5	1×
Total	50	-

Cycle no.	Denature	Anneal	Extend no. 1	Imagea	Extend no. 2
1	98 °C, 3 min	-	-	-	-
2–32	98 °C, 5 s	65 °C, 10 s	72 °C, 10 s	Image	72 °C, 8 s

Reagent	Quantity (μl)
PCR product (Step 13)	10
NTP buffer mix (20 mM each NTP)	10
T7 RNA polymerase mixture	2
RNAse inhibitor	1
Total	23

Reagent	Quantity (μl)
T7 reaction	20
dNTP (10 mM)	10
RT buffer (5×)	2
Labeled or unlabeled RT-primer (200 (μM)	3
Maxima H Minus RT enzyme	1
RNasin RNase inhibitor	4
Total	40

	Directory path input
Ex. (Experiment) Table	Type in the directory containing the ExperimentLayout.xIsx (e.g. E:\2017-11-30_WT_10-12hrs_(HoxRNA)_(BXC3kb)\DNA_Expt\ExperimentlLayout.xlsx)
Save Folder	Type in the directory targeting the Analysis folder (e.g. E:\2017-11-30_WT_10-12hrs_(HoxRNA)_(BXC3kb)\DNA_Expt\Analysis\)

Parameter name	Default value	Function
verbose	true	Results will be printed to the command terminal
recordValidation	true	Results of your selection will be saved and printed to the command terminal at the end
selectFOVs	inf	If you do not want to Validate Drift Fix for all FOVs, you can specify particular FOVs here (e.g. 1, 2, 3, 10, 11, 12)

Parameter name	Default value	Function
selectFOVs	1	If you do not want to select a fiducial spot in FOV 1, which is the default, you can specify particular FOVs here
imcontrastLow	0	Corresponds to the contrast to use when rendering an image. This fraction of the dimmest pixels will be set to zero/black
imcontrastHigh	0.999	Pixels brighter than this fraction will be saturated

ChrTracer3_SpotSelector Tools	Function
Add by Click	A cursor tool will allow you to manually select fiducial spots that might have been missed
Remove by Click	A cursor tool will allow you to manually select fiducial spots that are not real spots. This could be background noise or in the case of Drosophila embryos, e.g. signal from the embryonic yolk which does not represent real fiducial spots
Remove in Area	This is a useful bounding polygon feature that will allow you to select a region of interest, such as embryonic yolk, that might not have any real fiducial spots
Remove IDs	Each spot has a given ID and you can type the spot ID below Remove ID and then click this button to remove desired spots
Source Data	By default, spot selection is based on the fiducial from readout 1. You can choose to change spot selection to the data channel instead and can choose which readout to visualize. We typically do not change this parameter, but include it here in case there is too much background in the fiducial channel

Parameter name	Default value	Function
numParallel	10	Number of parallel pool of workers
overwrite	No	By default, the output files (e.g Fov001_AllFits.csv) will not get overwritten. Change to ‘Yes’ if you want to overwrite these files
checkOffset	false	Request software to align images along the z-dimension (axial) using the offset files
saveData	false	Saves a compressed version of the data that contains only the cropped regions of the image. See function Readlmage5D for details about this file format

Parameter name	Default value	Function
selectFOVs	inf	You can choose to correct stage drift for specific FOVs by specifying them under SelectFOVs. If inf (infinite) all FOVs will be used
saveFits	true	Will save the coarse alignments
alignContrastLow	0.7	Auto contrast to enhance feature matching between images. Pixels dimmer than this fraction of the brightness histogram will be set to zero
alignContrastHigh	0.9995	Auto contrast to enhance feature matching between images. Pixels brighter than this fraction of the brightness histogram will be saturated
refReadout	1	As default, all fiducials are aligned with the Fiducial imaged simultaneously with Readout 1. If you want, you can choose to align all Fiducial images with any readout. For example, if you rather align Fiducials based on the Fiducial image taken simultaneously with Readout 2, change refReadout from ‘1’ to ‘2’
maxSize	400	All images will be downsampled to this size in terms of pixels for the coarse registration, prior to a second round of alignment on a subset of the image, at higher resolution. Set equal to image size or larger (‘inf’ is allowed) to do single step alignment
fineBox	100	Corresponds to the 100×100 pixel box size from the coarse alignment that will be used for fine alignment between fiducials from all FOVs. The fine alignment will use the brightest area of your image for this step
fineUpsample	1	The fine scale alignment can be performed to sub-pixel accuracy by upsampling each image. Values between ‘1’ (no upsampling) to ‘6’ (1/6 pixel accuracy) are recommended
maxShift	inf	This allows for shifts between fiducial images to occur during the coarse alignment without any shift restrictions
gradMax	true	Align where the gradient of the cross-correlation is largest (i.e. where the correlation strength changes the fastest, rather than where it is maximum). This typically performs better than maximizing the cross-correlation when aligning punctile data with varying brightness
minGrad	-inf	Optionally, select a minimum positive value for the gradient difference to accept the fine scale alignment. If this minimum improvement is not met, the maximum of the 2D cross-correlation function will be used instead of the gradient
angles	0	This refers to the angle of rotation potentially needed to drift correction. No angles need to be corrected here as the sample has not moved in a rotational manner during the experiment
scales	1	Isotropic expansion or contraction of image around the centroid. Typically not used here
fineMaxShift	10	Refers to the max shift in pixels that can be used for the fine alignment
fineAngles	0	If upsampling was used, the image can be corrected to sub-pixel resolution by scanning angles here. Typically not used here
fineScales	1	If upsampling was used, isotropic expansion can be corrected to sub-pixel accuracy by changing the scale here. Typically not used here
fineCenter	[0 0]	The computer will choose the best x-y position to do the fine alignment based on the pixel intensity of the fiducial spots. Optionally, this can be set to focus the fine-alignment on a particular pixel
showplot	true	This allows you to see Fig. 10b. It is not recommended to change this as visual inspection is always useful. However, if you decide to change this to ‘false’, you will be able to use MATLAB freely without the figure continuously appearing on your screen
fastDisplay	true	Downsample the overlay images displayed on the screen to the size specified by ‘displayWidth’. If the ‘maxSize’ is larger than the ‘displayWidth’, fastDisplay as ‘true’ will increase the run speed
displayWidth	500	If you wish to increase or decrease the display resolution of Fig. 10b, you can increase the size here and vice versa
showExtraPlot	false	Display the 2D correlation and gradient of the 2D correlation in addition to showing the alignment. Only recommended for troubleshooting
minFinelmprovement	inf	Optionally, require the absolute improvement in the correlation-coefficient of the fine alignment to be at least this much better than the correlation-coefficient of the coarse alignment. If this condition is not met, the coarse alignment values will be used and a warning message ‘fineAlign failed’ will be printed
showCorrAlign	true	This allows you to visualize the overlay of the corrected Fiducial for the latest readout corrected FOV

Parameter name	Default value	Function
refHybe	1	Determines which hybridization round’s fiducial data is used as the reference for a final fine-scale 3D drift correction
keepsBrightest	1	Corresponds to the number of data spot(s) per readout that the computer will try to fit using the 3D Gaussian algorithm
maxXYstep	8	This is the number of pixels away from the center of the fiducial that can be considered a valid localization for the data
maxZstep	10	This is the number of z-sections away from the center of the fiducial signal that can be considered a valid localization for the data
datMinSep	3	If keepBrightest is more than 1, only readout spots within this number of pixels from one another will be considered to be in the same cell
datMinPeakHeight	500	This sets the minimum readout signal intensity value which is considered a valid localization
fidMinPeakHeight	200	This sets the minimum fiducial signal intensity value that is considered a valid image for alignment
maxXYdrift	4	The maximum number of pixels in XY that the algorithm is allowed to shift the image during the registration step
maxZdrift	6	Same as above, but for the z-drift
upsample	4	The correlation based alignment will upsample the fiducial data by this amount prior to registration. So ‘4’ corresponds to a registration accuracy of ¼ a voxel width
fidMinSep	5	If fidKeepBrightest is more than 1, Fiducial spots within this number of pixels from one another will be considered to be in the same cell
fidKeepsBrightest	1	This determines the maximum number of distinct fiducial spots the algorithm will try to detect per crop area. It will further attempt to find a data spot within the search range of each fiducial spot
fidMaxFitWidth	8	The Gaussian fitting will crop to a box of this width in pixels in x, y around the brightest pixel prior to fitting. This prevents stray background signals from dragging the fit off peak. This is used only for determining the centroid of the fiducial signal
fidMaxFitZdepth	12	This parameter is like the previous, but for the z-pixels (z-steps)
datMaxFitWidth	8	The Gaussian fitting will crop to a box of this width in pixels in x, y around the brightest pixel prior to fitting. This prevents stray background signals from dragging the fit off peak
datMaxFitZdepth	12	This parameter is like the previous, but for the z-pixels (z-steps). Separate control is allowed since z-step size may be different than x-y pixel size
datMinHBratio	1.2	The minimum ratio between the peak brightness and the background. This parameter affects how ‘spot like’ the signal needs to be counted. Decrease it to make fitting less stringent and increase the number to make fitting more stringent
datMinAHratio	0.25	The minimum ratio between the fitted amplitude of the Gaussian and the peak brightness. This parameter affects how ‘spot like’ the signal needs to be counted. Decrease it to make fitting less stringent and increase the number to make fitting more stringent
datMaxUncert	2	The maximum uncertainty in pixels for a spot to be recorded. Increase it to make fitting less stringent and decrease the number to make fitting more stringent
nmXYpix	154	This is the x-y pixel size in nanometers
nmZpix	100	This is the z-step between frames in the z-stack in nanometers (the z pixel size)
showPlots	true	The SpotFinder will plot the results of the fit as a tile of multicolor spots when finished

Parameter name	Default value	Function
autoSelectThreshold	0.997	If you want more fiducial spots, lower the threshold number and if there are too many spots then increase the threshold
autoSelectDownsample	3	Centroids are called on a downsampled image. If the spots are large, try increasing this value. You may need to lower autoSelectThreshold to compensate
coarseBackgroundSize	50	Image will be downsampled to this pixel size to estimate the illumination profile. “0” deactivates coarseBackground subtraction
densityFilterNeibs	0	Sets a threshold for the number of neighbors a spot may have within the cutoff distance set by densityFilterMinDist. Extra spots will be removed. A value 0 decativates this filter
densityFilterMinDist	30	Distance in pixels for the filter above
laplaceFilter	false	Uses a laplacian filter to correct brightness differences set to ‘true’
edgeThresh	0	Uses an edge detection filter to correct brightness differences with a threshold set here. Selection of 0 deactivates this filter
edgeSigma	1.2	Controls the width of the edge detection filter above
imContrastHigh	0.9995	Pixels brighter than this fraction will be saturated
imContrastLow	0.5	Corresponds to the contrast to use when rendering an image. This fraction of the dimmest pixels will be set to zero/black
displayContrastHigh	0.999	Changes the contrast in the display image only, does not saturate the image used to call spots
displayContrastLow	0.5	Changes the contrast in the display image only, does not saturate the image used to call spots, pixels below this fraction will be set to zero/black
backgroundCorrect	none	Choose between ‘none’, ‘file’, or ‘median’. Decide if illumination should be flattened. If you recorded dax files of a blank illumination, select ‘file’. If you have many fields of view with randomly positioned data, you may use ‘median’ to estimate the illumination bias from the data. To skip background correction, select 'none'
border	2	Spots within this number of pixels from the edge will be removed
removeEdgeStack	20	If this many spots are observed in the same pixel row or column, they will all be removed. This helps correct edge effects in some images
showExtraPlots	false	If true, the software will display the results of any of the filters selected above, and wait for any key-press before proceeding to show the result of the next filter. For troubleshooting only

Parameter name	Default value	Function
saveName	Mosaic	The Tif file will be saved in the ‘Analysis’ folder, along with the tile positions and tile filenames needed to reconstruct the mosaic from the max projected image files. Channel names are also saved as a txt file for easy reference. The default file name is ‘Mosaic’, but you can change it to any other name if you want
saveFolder		By default, the Tif file will be saved in the folder that was specified in the first step of BuildMosaicsGUI. However, you can change it to a new path directory here

Parameter name	Default value	Change	Function value to
chnTableFile			Enter the path to your experiment layout file here (e.g. E:\2017-11-30_WT_10-12hrs_(HoxRNA)_(BXC3kb)\DNA\ExperimentLayout.xlsx)
analysisFolder	.\Analysis\		Default value “.\Analysis\” is recommended. Saved alignment results and mosaics will be placed in the Analysis folder. “.\” notation saves data in whichever folder the chnTableFile is located, inside the newly created folder ‘Analysis’. There should already be an ‘Analysis’ folder created if ChrTracer3 ran before this step. No changes need to be made here
dataType	all	fiducial	This specifies which data to create max projections of, fiducial, readout data, or all. The default is ‘all’, which corresponds to both the fiducial and readout channel. For creating a DNA/ORCA mosaic, type in 'fiducial' since you will only need to know the nanoscale positions of fiducial spots and do not need to create a mosaic for individual DNA readouts. If ChrTracer3 was already used in this dataset, it would have already created max-projections and this step will go by quickly as BuildMosaicsGUI will only need to load already created max projected fiducial data
hybNumber	inf	1	To create a DNA mosaic, only the fiducial data from readout 1 is needed. Thus, change the default ‘inf’ to ‘1’ to specify to use data only from readout 1
fov	inf		The default is ‘inf’ to process data from all FOVs in the folder. This default value is recommended to use. It is optional to select a subset of FOVs (e.g. “[2:10,15:35]”)

Parameter name	Default value	Function
alignContrastLow	0.7	This fraction of the dimmest pixels will be ignored for the alignment
alignContrastHigh	0.9995	Pixels brighter than this fraction of the brightest pixels will be saturated for the alignment
refHybe	1	The fiducial data from this hybridization round (e.g. ‘1’) will be used as the alignment reference
maxSize	400	For the coarse alignment step, the images larger than this size will be downsampled so the average dimension is this size
fineBox	200	Fine scale alignment extends this far in pixels around the center point
fineUpsample	1	Data will be upsampled by this factor to improve accuracy of fine alignment
maxShift	inf	The maximum number of pixels the coarse alignment may shift the image
gradMax	true	Boolean indicates whether or not to use the gradient of the correlation plot instead of the correlation values
minGrad	-inf	Minimum slope observed in the correlation map to accept a shift. To use any shift, keep default of “-inf”
fineMaxShift	10	Maximum number of pixels the fine alignment may shift the image
fineCenter	[0 0]	Fine scale alignment is centered at this point. If left at [0,0], the brightest region (not brightest pixel) of the image will be selected
showplot	true	Show plots during imaging
fastdisplay	true	Boolean, indicating whether display should be accelerated by downsampling. Typically limited by screen resolution; if you aren’t pausing and zooming in on the data, this should not affect quality
displayWidth	500	Image size to downsample to for fast display
showExtraPlot	false	As the default, correlation maps will not be displayed. Can be changed to ‘true’ to show the correlation maps, which can be useful for troubleshooting
minFineImprovement	0	If the correlation coefficient does not improve by this amount upon fine-scale alignment, reject the fine scale alignment
showCorrAlign	true	Shows images of before alignment, after coarse alignment, and after fine alignment. It is not recommended to change this parameter as keeping it ‘true’ allows you to visually check if the alignment is happening correctly

Parameter name	Default value	Function
backgroundCorrect	none	Parameter options are ‘none’, ‘file’, and ‘median’. Decide if illumination should be flattened. If you recorded dax files of a blank illumination, select ‘file’. If you have many fields of view with randomly positioned data, you may use ‘median’ to estimate the illumination bias from the data. To skip background correction, keep the default ‘none’
method	edgeBlur	Keeping the default ‘edgeBlur’ is recommended. This setting uses a linear gradient of overlapping images regions to stitch borders. This is typically best for avoiding artificial edges. Other options include ‘mean’ - average the overlap, ‘sum’ - add the overlap, ‘last’ - use the last image and cover over what was imaged before, ‘first’ - use the first image and ignore later data
mosaicContrastLow	0.1	Contrast to use when rendering images. This fraction of the dimmest pixels will be set to zero/black
mosaicContrastHigh	0.9999	Contrast to use when rendering an image. This fraction of the brightest pixels in the image will be saturated
fovs	inf	“inf” default value is recommended to use all FOV in the folder. Optionally, select a subset (e.g. “[2:10,15:35]”)
hybs	1	Hybridization data that will be displayed in the mosaic. By default, we show data from readout ‘1’
dataType	fiducial	Data to show for alignment mosaic
transpose	true	The default is ‘true’, which gives you the option to transpose tiles in creating mosaic
scope	autoDetect	Depending on the camera orientation, individual tiles may be flipped across the left-right and up-down axis to properly render the mosaic. In our case, the camera orientation is automatically detected based on the pixel size of the raw images. Outside users will have to configure this setting in the MATLAB code to make sure that the tiles are oriented correctly. Our code within the Github link (https://github.com/Boettigerlab/ORCA-public) is commented in BuildMosaicsGUI.m to make sure the user knows where to make the appropriate changes based on their camera orientation
trimBorder	10	Remove the number of specified pixels around the border - sometimes our cameras have a bright line of pixels at the edge of the image, for reasons currently unclear. This removes them

Parameter name	Default value	Function
interactive	true	Boolean, set to ‘true’ to run the alignment tool in interactive mode or not. We recommend using the default setting. Set to false to skip alignment
corrAlign	true	Boolean, set to ‘true’ perform correlation based alignment. Set to false to skip alignment
showplot	false	Set to ‘true’ to show additional alignment plots; recommended only for troubleshooting alignments
method	edgeBlur	Uses a linear gradient of overlapping images to stitch borders. This is typically best for avoiding artificial edges
corrAlignLow	0.8	Controls the lower bound for image contrasting used during automated alignment. This fraction of pixels will be set to 0/black, so the alignment is driven by the data and not the background
corrAlignHigh	0.999	Controls the upper bound for image contrasting used during automated alignment. Pixels brighter than this fraction of all pixels will be saturated
mosaicContrastLow	0.1	Contrast to use when rendering image. This fraction of the dimmest pixels will be set to zero/black
mosaicContrastHigh	0.9999	Contrast to use when rendering an image. This fraction of the brightest pixels in the image will be saturated
fovs	inf	Use all FOVs in the folder. Optionally, select a subset (e.g. “[2:10,15:35]”)

Parameter name	Default value	Function
redContrastLow	0.1	For the red (DNA) channel, this fraction of the dimmest pixels will be set to zero/black
redContrastHigh	0.9999	For the red (DNA) channel, pixels brighter than this fraction will be saturated
cyanContrastLow	0.1	For the cyan (RNA) channel, this fraction of the dimmest pixels will be set to zero/black
cyanContrastHigh	0.999	For the cyan (RNA) channel, pixels brighter than this fraction will be saturated

Parameter name	Default value	Change value to	Function
saveName		ComboMosaic	By default, ‘ComboMosaic’ will be a common name given to all Tif files for every RNA readout channel. You can change it to any other name that you want
saveFolder			Value does not need to be changed as this will automatically save your files in the RNA ‘Analysis’ folder. You can choose to change the directory path here

PERMALINK

Tracing DNA paths and RNA profiles in cultured cells and tissues with ORCA

Leslie J Mateo

Nasa Sinnott-Armstrong

Alistair N Boettiger

Abstract

Introduction

Fig. 1 |. Optical reconstruction of chromatin architecture (ORCA) approach.

Overview of the multiplex RNA labeling and DNA tracing procedure

Fig. 2 |. Schematic overview of ORCA and RNA labeling experimental process and computational analysis.

Comparison to other methods

Fig. 3 |. Comparison between ORCA and other methods.

Limitations

Applications

General considerations

Table 1 |.

Table 2 |.

Fluidics setup

Fig. 4 |. Fluidics system layout.

Microscope setup

Fig. 5 |. Homebuilt microscope.

Lasers

Camera

Motorized stages enable automation

Focus lock assembly and calibration

Microscope software

Considerations about the laboratory facilities

Materials

Biological materials

Reagents

Probe synthesis

Urea PAGE gel

Apple juice–agar plate

Drosophila egg laying and collection

Drosophila embryo embedding and cryosectioning

Cell culture fixation preparation

Hybridization of RNA or DNA probe

Preparing a 96-well plate for hybridization

Imaging buffer

Bleach buffer

Wash buffer

Equipment

Primers and probes

Example dataset

Software

Reagent setup

Apple juice–agar plate preparation

Yeast paste

Borax solution

2% (wt/vol) agarose borax gel

NaOH-EDTA solution (1:1 (vol/vol) 0.5 M EDTA and 1 M NaOH)

Gel for urea PAGE (15% (wt/vol) urea TBE gel) (two 0.75-mm gels)

1x TBE

1x GelGreen in 1x TBE

1x PBS

2x SSC (bleaching buffer)

8% (vol/vol) formaldehyde + 50 mM EGTA in 1x PBS

50% (vol/vol) methanol in 1x PBS

30% sucrose (wt/vol) in 1x PBS

Chromium gelatin

4% (vol/vol) formaldehyde in 1x PBS

RNase A solution (10 μg/ml) in 1x PBS

Hybridization no. 1 (0.1% (vol/vol) Tween-20, 50% (vol/vol) formamide and 2x SSC)

Hybridization no. 2 (50% (vol/vol) formamide, 2x SSC, 0.1% (vol/vol) Tween-20 and 10% (vol/vol) dextran sulfate)

Post-fix buffer (8% (vol/vol) formaldehyde and 2% (vol/vol) GA in 1x PBS)

Imaging buffer (0.5 mg/ml glucose oxidase, 40 μg/ml catalase and 10% (vol/vol) glucose in 2x SSC)

25% (vol/vol) EC in 2x SSC

Wash buffer (30% (vol/vol) formamide in 2x SSC)

1x TE buffer

Readout probes and strand-displacement oligo preparation for a single well in a 96-well plate

Procedure

Oligopaint library design ● Timing 1 d

Fig. 6 |. Primary probe synthesis.

Primary probe synthesis from the oligopool ● Timing 2 d

Drosophila egg laying in collection cage ● Timing 1.5 d (Timing is based on the time that it takes to collect 10–12-h embryos. This process can be shorter or longer depending on the desired developmental stage.)

Fig. 7 |. Sample preparation for primary probe hybridization.

Drosophila embryo collection ● Timing 40 min

Embedding Drosophila embryos for cryosectioning ● Timing 2 h 35 min

Preparing slides for cryosectioning ● Timing 1 h

Embryo cryosectioning ● Timing 1 h

Parameter name	Default value	Function
chnTableFile	Enter Path To DNA Experimental Table: C:\	Enter the path to your experiment layout file here (e.g. E:\2017-11-30_WT_10-12hrs_(HoxRNA)_(BXC3kb)\DNA\ExperimentLayout.xlsx)
analysisFolder	.\Analysis\	The default value “.\Analysis\” is recommended. Saved alignment results and mosaics from previous ChrTracerGUI and BuildMosaicsGUI were placed in this folder
backgroundCorrect	none	Parameter options are ‘none’, ‘file’, and ‘median’. Decide if illumination should be flattened. If you recorded dax files of a blank illumination, select ‘file’. If you have many fields of view with randomly positioned data, you may use ‘median’ to estimate the illumination bias from the data. To skip background correction, keep the default ‘none’
method	edgeBlur	Keeping the default ‘edgeBlur’ is recommended. This setting uses a linear gradient of overlapping images regions to stitch borders. This is typically best for avoiding artificial edges. Other options include ‘mean’ - average the overlap, ‘sum’ - add the overlap, ‘last’ - use the last image and cover over what was imaged before, ‘first’ - use the first image and ignore later data
mosaicContrastLow	0.1	Contrast to use when rendering images. This fraction of the dimmest pixels will be set to zero/black
mosaicContrastHigh	0.9999	Contrast to use when rendering an image. This fraction of the brightest pixels in the image will be saturated
fovs	inf	“inf” default value is recommended to use all FOV in the folder. Optionally, select a subset (e.g. “[2:10,15:35]”)
hybs	1	Hybridization data that will be displayed in the mosaic. By default, we show data from readout ‘1’
data Type	fiducial	Show fiducial data for alignment mosaic
transpose	true	The default is ‘true’, which gives you the option to transpose tiles in creating mosaic
scope	autoDetect	Depending on the camera orientation, individual tiles may be flipped across the left-right and up-down axis to properly render the mosaic. In our case, the camera orientation is auto detected based on the pixel size of the raw images. Outside users will have to configure this setting in the MATLAB code to make sure that the tiles are oriented correctly. Our code within the Github link (https://github.com/BoettigerLab/ORCA-public) is commented in BuildMosaicsGUI.m to make sure the user knows where to make the appropriate changes based on their camera orientation
trimBorder	10	Remove the number of specified pixels around the border - sometimes our cameras have a bright line of pixels at the edge of the image, for reasons currently unclear. This removes them

Parameter name	Default value	Function
chnTableFile	Enter Path To RNA Experimental Table: C:\	Enter the path to your experiment layout file here (e.g. E:\2017-11-30_WT_10-12hrs_(HoxRNA)_(BXC3kb)\RNA\rnaTable.xlsx)
analysisFolder	.\Analysis\	The default value “.\Analysis\” is recommended. Saved alignment results and mosaics from previous ChrTracerGUI and BuildMosaicsGUI were placed in this folder
backgroundCorrect	none	Parameter options are ‘none’, ‘file’, and ‘median’. Decide if illumination should be flattened. If you recorded dax files of a blank illumination, select ‘file’. If you have many fields of view with randomly positioned data, you may use ‘median’ to estimate the illumination bias from the data. To skip background correction, keep the default ‘none’
method	edgeBlur	Keeping the default ‘edgeBlur’ is recommended. This setting uses a linear gradient of overlapping images regions to stitch borders. This is typically best for avoiding artificial edges. Other options include ‘mean’ - average the overlap, ‘sum’ - add the overlap, ‘last’ - use the last image and cover over what was imaged before, ‘first’ - use the first image and ignore later data
mosaicContrastLow	0.1	Contrast to use when rendering images. This fraction of the dimmest pixels will be set to zero/black
mosaicContrastHigh	0.9999	Contrast to use when rendering an image. This fraction of the brightest pixels in the image will be saturated
fovs	inf	“inf” default value is recommended to use all FOV in the folder. Optionally, select a subset (e.g. “[2:10,15:35]”)
hybs	inf	Hybridization data that will be displayed in the mosaic. All data from readouts taken throughout the RNA experiment will be shown
dataType	inf	Show both fiducial and readout data
transpose	true	The default is ‘true’, which gives you the option to transpose tiles in creating mosaic
scope	autoDetect	Depending on the camera orientation, individual tiles may be flipped across the left-right and up-down axis to properly render the mosaic. In our case, the camera orientation is auto detected based on the pixel size of the raw images. Outside users will have to configure this setting in the MATLAB code to make sure that the tiles are oriented correctly. Our code within the Github link (https://github.com/BoettigerLab/ORCA-public) is commented in BuildMosaicsGUI.m to make sure the user knows where to make the appropriate changes based on their camera orientation
trimBorder	10	Remove the number of specified pixels around the border - sometimes our cameras have a bright line of pixels at the edge of the image, for reasons currently unclear. This removes them

Parameter name	Default value	Function
angles	−12:0.5:12	List of rotation angles to explore for mosaic alignment
downsample	10	Downsample the image this many fold for coarse alignment
maxSize	400	This is for correlation based analysis, the coarse alignment on the whole image is still done in a two step process. This is the maxSize for the first step
alignContrastHigh	0.999	Auto contrast to enhance feature matching between images. Pixels brighter than this fraction of the brightness histogram will be saturated
alignContrastLow	0.9	Auto contrast to enhance feature matching between images. Pixels dimmer than this fraction of the brightness histogram will be set to zero
maxShift	300	The maximum number of pixels the coarse alignment may shift the image
showPlot	true	Boolean, when set to ‘true’ will show progress image of the multistep alignment. The final result will always be shown
verbose	true	Boolean, when set to ‘true’ will show text in the terminal. The text will include information/instructions of the GUI’s current step

Parameter name	Default value	Function
Interactive	true	Boolean, use the interactive GUI to guide adjust alignment
corrAlign	true	Boolean, use correlation based analysis
padAlign	150	Number of pixels with which to pad image prior to computing alignment
corrAlignHigh	0.998	Auto contrast to enhance feature matching between images. Pixels brighter than this fraction of the brightness histogram will be saturated
corrAlignLow	0.65	Auto contrast to enhance feature matching between images. Pixels dimmer than this fraction of the brightness histogram will be set to zero
angles	0	Rotations to try in docking already coarse rotated RNA images. Coarse correction in the previous step already took into account rotational changes and that is why the default is ‘0’
scales	1	Isotropic expansion contraction of image around centroid
showplot	true	Show alignment plots
maxSize	400	All images will be downsampled to this size in terms of pixels for the coarse registration, prior to a second round of alignment on a subset of the image, at higher resolution. Set equal to image size or larger (‘inf’ is allowed) to do single step alignment
showExtraPlot	false	Change to ‘true’ to show extra plots for troubleshooting alignment
fineBox	250	Size of fineBox for the correlation based alignment
minFinelmprovement	0.1	Minimum improvement for the correlation based alignment

Parameter name	Default value	Function
ROIs	all	By default, a distance map for all ROIs that have been saved will be populated. You can choose to change this for a specific ROI (e.g. “A1”). If you choose to view multiple ROIs, make sure to not insert a space between each ROI (e.g. “A1,A2,A3”)
figHandle	200	All figures are identified by a unique handle. To avoid overwriting currently active figures, select a unique number here
badHybes		After generating the distance map, you may notice that some barcodes have poor labeling (e.g. prominent stripes that are seen throughout the distance map). If so, enter the barcode number(s) that did not label well (e.g.^16,26,39). Additionally, if you have previously generated distance maps for the total population of cells after running ChrTracer3, you might already know which barcode (s) labeled poorly. Enter those barcode numbers here

Parameter name	Default value	Function
ROIs	all	By default, a distance map for all ROIs that have been saved will be populated. You can choose to change this for a specific ROI (e.g. “A1”). If you choose to view multiple ROIs, make sure to not insert a space between each ROI (e.g. “A1,A2,A3”)
threshold	150	This will display the fraction of cells in which the pairwise interactions are within 150 nm
figHandle	200	‘200’ is the default figure handle of the distance map (see the bottom-right corner of Fig. 13c). Having figures with unique figure handles allows you to access a specific figure, such as saving a specific figure. To avoid clearing a previously generated distance map, change the figHandle number to one that has not been previously used (e.g. 307)
badHybes		After generating the distance map, you may notice that some barcodes have poor labeling (e.g. prominent stripes that are seen throughout the distance map). If so, enter the barcode number(s) that did not label well (e.g. ^16,26,39). Additionally, if you have previously generated distance maps for the total population of cells after running ChrTracer3, you might already know which barcode(s) labeled poorly. Enter those barcode numbers here

Parameter name	Default value	Function
minSep	10	Spots within this number of pixels from one another will be considered to be in the same cell
cellSize	25	This is the maximum radius which may be assigned to a cell during the segmentation

Step(s)	Problem	Possible reason	Solution
3	Primary probes do not amplify during the run	Insufficient template DNA and/or primer	Add more Oligopaint library and/or primer to the reaction. Alternatively, dilute fresh primer
19	Substantial primer band is observed at the bottom of the PAGE gel	Primer dimer has formed	Use less forward and reverse primers next time
30	Concentration is below 500 ng/μl.	Incomplete elution	Concentrate the DNA using a DNA SpeedVac for 1–2 hrs. If the tube is completely dried after concentrating in the DNA SpeedVac, resuspend it in 25 μL ultrapure water
36	Flies have laid low numbers of eggs	Female flies could be old	Add virgin/young females to the collection cage as they lay more than older females
46	Embryos are floating between heptane-methanol layers after vortexing	Non-devitellined embryos can remain in the interphase	If you choose to collect all embryos, including non-devitellined embryos, add more methanol and continue vortexing until embryos sink to the methanol layer
68	Embryos cryosectioned onto glass coverslips coated with chromium-gelatin older than a month might peel away either during the primary probe hybridization and/or during imaging due to the constant flow of solutions within the flow chamber	Older chromium-gelatin may be less adhesive	Make fresh chromium-gelatin once a month
142	When testing a new RNA primary probe, the user is not sure if the probes labeled RNA or DNA	We recommend designing the RNA probes as antisense to avoid targeting DNA. However, some users might want to design probes differently and this might lead to cross-hybridizing to genomic DNA	We suggest the following 3 controls to make sure that probes are labeling RNA and not DNA: 1) Compare distinguishable populations of cells where one population should express the target RNAs and other should not, or at least should express it at a clearly distinct level. For example, Abd-B should never be expressed in the most anterior cells in the Drosophila embryo. 2) Compare distinguishable populations of cells labeled with primary probes targeting mature mRNAs. This should yield a clear difference in average fluorescence intensity of labeling seen at low magnification and in total spot counts seen at higher magnification. 3) Primary probes targeting mature mRNAs should provide multiple distinct puncta in the cytoplasm, whereas intronic-RNA and DNA primary probes should only label spots in the nuclei
197	High background during DNA imaging acquisition	Primary probes are antisense and could be binding to some RNA transcripts in the locus	Incubate the sample for 1 hr at 37°C in RNase solution in Step 152
209	Fiducials do not align well in ‘Fix Global Drift’	Larger than average stage drifts might have occurred during the imaging	Increase the size of fineBox. Keep in mind that increasing fineBox will slow down this step. You can also specify a particular region of the image to use instead of the default brightest area - see fineCenter
209	Fiducials do not align well in ‘Fix Global Drift’	Sample has peeled over the course of the experiment, which makes it difficult for the computer to align fiducials at the edge of the sample	Specify that fine alignment would occur on an x-y position that has remained flat throughout the experiment
212	Drift correction failed for one or more FOVs as seen with multiple colors for a single spot	See possible troubleshooting reasons for step 209	To correct the Bad FOVs, click ‘Back’ on the ChrTracer3 GUI. You will be back on the ‘Fix Global Drift’ step. Click on ‘Step Pars’ and type in the bad FOVs under SelectFOVs. Repeat Steps 208–212 for the Bad FOVs
222	General failure to detect the centroid of fiducial and readout signal	Fiducial data for the readout ‘1’ did not label well. By default ChrTracer uses this fiducial for the fine-scale drift correction when calculating for the 3D centroid of our spots	In ‘Step Pars’ change the refHybe to another readout, such as ‘2’, that has a fiducial signal that labeled well
222	Fit Spot’ is not detecting the readout spots	The readout spot might be further away, in terms of pixels, from the center of the fiducial. This could happen if the primary probe is labeling a large region of interest (~1 Mb)	In ‘Step Pars’, change the maxXYstep to increase the default search size in terms of X-Y pixels to fit the readout spot
222	Fit Spot’ is not detecting the readout spots	The readout spot might be further away, in terms of pixels, from the center of the fiducial. This could happen if the primary probe is labeling a large region of interest (~1 Mb)	In ‘Step Pars’, change the maxZstep to increase the default search size in terms of Z pixels to fit the readout spot
222	Fit Spot’ is not detecting the readout spots	The background level is very high, which can make it hard to detect readout spots	We find it is more robust to select spots based on the quality of the Gaussian fit, rather than a threshold, so this default value of datMinPeakHeight is intentionally chosen below the typical background. However, if you can visually see the readout probe against the high background, then you can choose to lower the default number. Don’t lower it too much or you might start picking up background noise
222	Fit Spot’ is not detecting the correct fiducial spots which will result in detecting erroneous readout spots	If the data has a lot of background in the fiducial channel, ‘Fit Spot’ might be choosing a fiducial spot that is actually background noise	In ‘Step Pars’, increase the threshold for fidMinPeakHeight. Remember that the fiducial signal is typically strong and, thus, you can easily increase the fidMinPeakHeight value to 750–1000
222	Fit Spot’ is not detecting the readout spots	During the registration step, the algorithm allows the image to shift based on the default maxXYdrift. Frames might require larger shifts	The amount of shift is reported in the command terminal. If any frames are shifted over the max drift allowed, it is an indication that the alignment is not good. This may be acceptable for individual rounds but is generally an indication that more accurate drift correction was required during the Fix Global Drift step
222	Fit Spot’ is not detecting the readout spots	Same as above. Shifts between images might require a larger shift in Z	In ‘Step Pars’, increase the search size for maxZdrift in terms of Z pixels
234	Prominent stripes are seen across the distance map may indicate poor barcode labeling or noisy background	If the stripe suggests that the pairwise distance between a barcode is far away in nanoscale distance from all labeled barcodes, it may indicate that the readout probe failed to hybridize to the barcode	To further verify failure of labeling, you can calculate the detection efficiency for that particular barcode by using the FovX_AllFits.csv files generated from ChrTracer. You may need to re-dilute a fresh readout probe if failure of labeling occurred for a single readout. If it seems that many barcodes were not labeled throughout the experiment, this may suggest that something is wrong with your fluidics system. Perhaps there is salt build up from the 25% EC in 2× SSC or a tear in the tube that is causing the flow to be delayed. To determine if the stripes are real and not an artifact, perform a replicate of the experiment using newly diluted readout probes at the proper concentration
234	Prominent stripes are seen across the distance map may indicate poor barcode labeling or noisy background	If a stripe suggests that a single barcode is close in nanoscale distance to all labeled barcodes, it may indicate that the background noise was high when imaging the barcode	Repeat the experiment with a fresh sample and new buffers. You may need to re-dilute a fresh readout probe if the stripe occurred for a single readout. Alternatively, we have found that certain readout probes cause unspecific staining in our sample - perhaps due to the readout sequence complementing some other part of the organism’s genome - and these readout probes can create a stripe across the distance map. If this is the case, then we suggest dropping these readout probes when analyzing the distance maps. To determine if the stripes are real, perform a replicate of the experiment using newly diluted readout probes at the proper concentration
234	Erroneous stripe-like features in the distance maps	May be indicative of the strand displacement oligo not working properly	If you see a partial stripe (e.g. a stripe that starts at readout ‘25’ and continues) this means that the strand displacement oligo failed to remove a particular readout signal. To fix this, you can either re-dilute the strand-displacement oligo from the stock plate, or re-optimize the strand-displacement sequence
234	Erroneous loop-like features in the distance maps	May be indicative of contamination of the readout sequence by introducing two readout probes at the same time, due to pipetting error or cross-contamination between readouts. This might result in an unexpected loop between two barcodes, which would incorrectly suggest that these two barcodes are in close 3D proximity to one another	To confirm that a loop is real or not, we suggest using newly diluted readout sequences and repeating the experiment with a new sample