Spatial Genomic Analysis – a multiplexed transcriptional profiling method that reveals subpopulations of cells within intact tissues

Abstract

Here we present Spatial Genomic Analysis (SGA), a quantitative single-cell transcriptional profiling method that takes advantage of single molecule imaging of individual transcripts for up to a hundred genes. SGA relies on a machine learning based image analysis pipeline that performs cell segmentation and transcript counting in a robust way, and is suitable for various in situ applications. The neural crest is a transient embryonic stem cell population important for formation of various vertebrate body structures. After being specified as multipotent neural crest stem cells in the dorsal neural tube, they go through an epithelial to mesenchymal transition in order to migrate to different destinations around the body, and gradually turn from stem cells to progenitors prior to final commitment. The molecular details of this process remain largely unknown, and upon their emergence, the neural crest cells have been considered as a single homogeneous population. Technical limitations have restricted the possibility to parse the neural crest cell pool into subgroups according to multiplex gene expression properties. By using SGA we were able to identify subgroups inside the neural crest niche in the dorsal neural tube. High sensitivity of the method allows detection of low expression levels and enabled detection of factors not previously shown to be present in neural crest stem cells, such as pluripotency or lineage markers. Finally, SGA analysis also provides prediction of gene relationships within individual cells.

1. Introduction

Recent development in single-cell transcriptional profiling largely relies on single-cell RNA sequencing based methods [1]. Despite the transcriptome level throughput of these applications, they lack spatial orientation of cells in host tissues. On the other hand, single molecule fluorescent in situ hybridization (smFISH) [2] is a useful method for quantitative analysis of individual transcripts in single cells, but low signal and small throughput has so far limited its use mainly for cell culture approaches with simultaneous detection of only a handful of genes. We have established a Spatial Genomic Analysis (SGA) pipeline [3] that overcomes these limitations by using hybridization chain reaction (HCR) signal amplification [4,5] coupled with sequential hybridization rounds [6], as shown in figure 1. In order to achieve single cell resolution in complex tissue samples, SGA contains a machine learning algorithm based cell segmentation as well as a dot counting routine of the individual transcripts [3]. Cells are divided into subgroups by using unbiased hierarchical clustering based on their multiplexed gene expression profile, and then mapped back to their original spatial context in the tissue. SGA is a powerful method that can be applied to identify transcriptionally distinct subpopulations in stem-cell niches or other heterogeneous cell populations according to expression profiles of tens or even up to a hundred genes.

Figure 1.

Signal amplification by using Hybridization Chain Reaction and the sequential hybridization steps in SGA. (a) Each SGA probe consists of a 20 nucleotide long CDS recognition sequence followed by a four nucleotide linker sequence before the initiator sequence in the 3’ end required for the HCR amplification. (b) Each transcript is hybridized with up to 24 individual DNA probes tagged with the same initiator sequence. CDS of the mRNA is marked with green color. A pair of metastable hairpins tagged with fluorophores is used to amplify the signal for each individual probe. The chain reaction that leads to an amplified signal begins as the single stranded initiator sequence tagged to the hybridized probe opens one hairpin, hybridizes into it, and then opens a second hairpin, which again reveals the original initiator sequence and so forth. (c) A cartoon of the multiplexing scheme where a set of five genes are hybridized and imaged followed by probe stripping with DNase I enzyme. This routine is repeated until the desired number of genes is reached. (Figure 1C is reprinted from Lignell A, Kerosuo L, Streichan SJ, Cai L, Bronner ME (2017) Identification of a neural crest stem cell niche by Spatial Genomic Analysis. Nature Communications 8 (1):1830. doi:10.1038/s41467-017-01561-w with permission from SpringerNature).

We used SGA to address heterogeneity in the developing dorsal neural tube of chicken embryos where the neural crest cells are known to develop. Unbiased hierarchical clustering revealed five subpopulations with spatially distinct localization (figures 2A,B) providing novel insight into the mechanisms of neural crest development in its transient stem cell niche. Additionally, SGA allowed us to determine a core set of neural crest genes and draw conclusions on gene relationships based on their synexpression patterns within individual cells (Figure 2C). Here we describe the SGA method in full detail and provide suggestions for usage in other tissues.

Figure 2.

(a) Unbiased hierarchical clustering analysis of 1190 cells in the cranial dorsal neural tube of a HH Stage 9 chicken embryo reveals distinct subpopulations according to transcriptional similarity of the 35 genes analyzed by using SGA. The NCstem (yellow) and NC (red) cells refer to premigratory neural crest cells with or without expression of pluripotency factors, respectively, and the green cells represent newly migrating neural crest cells (NCmig). Cells destined to become parts of the central nervous system are also divided to stem cell (Nstem) and “neural only” (N) subgroups. (b) Spatial orientation of the clustered cells in the neural tube reveals that the pluripotent neural crest stem cells are located around the midline. The colors refer to the cell clusters shown in the heatmap above. (c) Synexpression analysis showing the similarity of gene expression patterns in individual cells. A “core neural crest” cluster of genes is revealed showing the set of neural crest genes that are most likely to be expressed together in individual cells at this developmental time point, and which is further divided into two separate subgroups. This approach may provide a valuable addition for studies on relationships between genes in gene regulatory circuitries. (The figures 2A-C are reprinted from Lignell A, Kerosuo L, Streichan SJ, Cai L, Bronner ME (2017) Identification of a neural crest stem cell niche by Spatial Genomic Analysis. Nature Communications 8 (1):1830. doi:10.1038/s41467-017-01561-w with permission from SpringerNature)

2. Materials

Since RNA integrity is critical for the SGA method, all the buffers and chemicals should be purchased as RNase free quality (when available) and with highest purity. The glassware, tubes, bench space, and gloves should be made RNase free and filtered pipette tips should be used. Appropriate personal protective equipment (safety goggles, lab jacket, nitrile gloves, and fume hood) should be used when working with hazardous and toxic chemicals. Use RNase free water for all solutions.

2.1. Buffers

1. Standard Buffers: 10X PBS and 20X SSC buffers should be purchased directly in RNase free quality and aliquoted to 50ml conical tubes and stored in room temperature. The working solutions should be made fresh from these stocks right before use.

2. Fixation solution: 4% Paraformaldehyde in 1X PBS. The solution can be stored in −20°C.

3. Anti-bleach buffer (ABB): The base (ABBB) is 20nM Tris-HCl, and 50nM NaCl saturated with trolox. This can be made in 50ml conical tube by adding 120mg Tris-HCl, 146mg NaCl, and 200mg of trolox into RNase free water and vortexed thoroughly (Note A). All the components can be made in advance and stored in aliquots in −20°C. First, make a 8% d-glucose solution in water and aliquot to microcentrifuge tubes. Second, make a pyranose oxidase solution by diluting the solid protein in its container with water and measure the concentration with a spectrometer until OD (405nm) is 0.5, and then aliquot to microcentrifuge tubes. Finally, dilute the catalase enzyme into a 10mg/ml concentration in water and aliquot to microcentrifuge tubes. All three components should be thawed in room temperature before use.

4. Incubation/hybridization buffers: Use high molecular weight (>500.000) dextran sulfate (DS) to make a 10% solution (100mg of DS mixed with 100μl 20x SSC, 0-300ml of formamide (FA) for 0%, 10%, and 30% solutions, respectively, filled to 1ml total volume with water in microcentrifuge tubes). Hybridization buffers that are used for overnight incubations should contain 0.02% thiomersal to prevent fungal and bacterial growth. These solutions should be made right before use.

5. Pre-hybridization buffers: Make 1% PFA in 1X PBS buffer from a PFA stock and store aliquotes in −20°C. Additionally, make 0.5% SDS in 1X PBS, and 1% NaBH4 solution in 1X PBS (weigh 10mg of NaBH4 and mix with 1ml of 1X PBS on a weight boat right before use, see note B for safety instructions).

6. Hairpin solutions: The hairpins labeled with fluorescent dye can be ordered from Molecular Technologies (www.moleculartechnologies.org), please find a detailed description of their structure in this reference [5].

7. Immunostaining and blocking buffers: Immunostaining is performed in 1X PBS/0.2% Triton (PBT) with 5% bovine serum albumin (BSA) as a blocking and 1% DMSO as a permeabilization agent.

2.2. Cover glasses

1. Treat #1.5 cover glasses with (3-Aminopropyl)triethoxysilane (AS) to provide strong sample adherence to the glass surface as follows: first, sonicate cover glasses in a container filled with acetylene for 1h in an ultrasound bath, and then dip them into 2% AS solution in acetone followed by two additional washes with acetone. Finally, rapidly air dry the cover glasses under compressed air (Note C).

2. After collecting the samples as cryo-sections on the cover glasses (see sample preparation in 3.1), cover the cryosections with hybridization chambers (e.g. 8mm in diameter and 50μl volume) that stick to the glass around the sample. The chambers maintain the operational volume small to assure efficient usage of the expensive reagents, and also prevent the samples form drying and protect them from detaching. Imaging is performed from the bottom through the cover glass with an inverted microscope.

2.3. Probe design

1. Coding sequences (CDS) of each mRNA can be acquired from your database of choice (e.q. www.ncbi.nlm.nih.gov) and, if possible, a probe set of up to 24 single stranded DNA probes per CDS should be designed (the minimum number of probes per CDS we have successfully used is 13). Blast the sequences to ensure unique binding. Each probe consists of a 20 nucleotide reverse compliment DNA consensus sequence. Importantly, leave a separation gap of at least 5 nucleotides between the individual probes. GC-content of the probes should be between 40-60% and melting temperatures around 52-58°C. If the CDS is short, 18 or 19 nucleotide long probe sequences can be used while the melting temperature is maintained at the same range.

2. In addition to the CDS binding sequence, a four nucleotide linker sequence as well as a (B1-B5) HCR initiator sequence [5] is added to the 3’ end of each probe, resulting in a 60 nucleotides long probe (Figure 1A). The sequences are listed below, the initiators are shown in capital and linkers in lower case letters. The HCR amplification scheme is presented in Figure 1B.

B1: tataGCATTCTTTCTTGAGGAGGGCAGCAAACGGGAAGAG

B2: aaaaAGCTCAGTCCATCCTCGTAAATCCTCATCAATCATC

B3: taaaAAAGTCTAATCCGTCCCTGCCTCTATATCTCCACTC

B4: atttCACATTTACAGACCTCAACCTACCTCCAACTCTCAC

B5: atttCACTTCATATCACTCACTCCCAATCTCTATCTACCC

The probes can be ordered from an oligo manufacturer (e.g. www.idtdna.com) in a 96-well plate diluted to a 100μM concentration. The probe sets for each gene are then combined and diluted into a 100nM per probe concentration (100X) that is directly used for experiments. Both the 100μM probe stocks and the 100nM/probe working solutions are stored in −20°C.

3. Methods

All the steps are performed in room temperature unless otherwise noted.

3.1. Sample preparation and cryosectioning.

1. Fix samples overnight in 4°C with 4% paraformaldehyde in 1X PBS. After fixation, wash 3x with 1X PBS/0.2% triton (PBT) followed by gradual dehydration into 100% ethanol (0/25%/50%/75%/100% steps 5-15 minutes each depending on the size of the tissue), and keep in −80°C for at least two days before starting the next step, since this is part of the permeabilization process. If needed, samples can be stored in −80°C for months if not years without significant RNA degradation.

2. Rehydrate the samples by gradually bringing them back to 1X PBS from 100% ethanol (100%/75%/50%/25%/0% with 5–15 min/step), and finally wash 2x with 1X PBT. Bring the samples gradually into 20% sucrose in 1X PBT on a nutator at 4°C for 3-4h (note D).

3. Remove all sucrose carefully and replace it with the O.C.T. embedding compound. Rotate the tubes a few times by hand and let the samples equilibrate at room temperature for 10min before transferring the samples into cryomolds under a dissection microscope by using a sterile plastic transfer pipette (cut the tip if sample size is big). The O.C.T. compound is very viscous; operate with a relatively large amount (1ml per sample) to enable the transfer and to avoid bubbles. Once optimally positioned, stiffen the samples first in a flat position by using dry ice before snap freezing them in liquid nitrogen until they start to turn white (submerge the samples ~3-5x for 3 seconds each). Place the molds in −20°C for a few minutes (to soften them to enable transfer by using forceps), and collect the samples to individually labeled microcentrifuge tubes for storage at −80°C.

4. Section the samples into 12-20μm thick slices by using a cryomicrotome and collect them on AS-coated cover glasses (see 2.2). Depending on the sample size, you may want to try to collect the samples in pairs that will fit into the same hybridization chamber. Store the samples in 4°C, they can be used for further experiments the next day or up to a few weeks.

3.2. Hybridization and immunostaining

3.2.1. Hybridization

All the following steps are performed to samples encapsulated inside hybridization chambers. Once the experiment has been started, the samples should be protected from drying. The volume pipetted into each chamber can vary depending on the chamber size (~50 μl for the 8mm chambers). Several samples can be operated simultaneously, and with the possibility of sample damage or detachment from the cover slip during the long protocol, it may be wise to start the experiment with at least double the amount of samples than what you wish to use for the analysis.

1. Wash the samples three times with 1% PFA 1X PBS solution by slowly rotating them for 5 minutes. This step removes the O.C.T. compound matrix covering the sample and mounts them tightly onto a cover glass.

2. Permeabilize the samples with 0.5% sodium dodecyl sulfate (SDS) in 1X PBS solution by slowly rotating them for 5 minutes.

3. Post-fix the samples with 1% PFA in 1X PBS for 5 minutes.

4. Treat the samples with 1% NaBH4 in 1X PBS solution for 5 minutes to minimize the background autofluorescence (see Note B), and then carefully wash three times with 2X SSC by slowly rotating for 5 minutes.

5. Block the samples with 1μM random 60-mer oligonucleotide in 2X SSC for 1h.

6. Make the hybridization buffer: 10% DS / 30% FA / 2X SSC with 0.02% thiomersal with 1μM random 60-mer oligonucleotide. Prepare the probe sets of the current hybridization round by making a mix in the hybridization buffer that contains each probe in a final concentration of 1nM. Depending on how many channels are used for the imaging, up to five different initiators can be used simultaneously in the mix.

   1. Gene A - B1
   2. Gene B - B2
   3. Gene C - B3
   4. Gene D - B4
   5. Gene E - B5

7. Pipette the hybridization solution into the hybridization chambers and place the cover glasses into a humidified chamber (Note E) that is covered with laboratory parafilm. Incubate the samples in 100% humidity at 37°C for overnight (minimum of 8h).

3.2.2. Amplification and probe stripping (see Figure 1)

1. Wash the samples three times with 30% FA 2x SSC followed by three times with 2X SSC after the overnight hybridization (Note F).

2. Snap heat the hairpins that will be used for signal amplification (B1H1, B1H2, B2H1, B2H2, etc.) for 90 seconds in separate microcentrifuge tubes at 95°C by using a heat block followed by a cool down in room temperature for 30 minutes. The “standard dye-hairpin set” we have used in our experiments is the following: Cy7-B1, Alexa647-B2, Alexa594-B3, Cy3B-B4, and Alexa488-B5.

3. Mix the hairpins to a final concentration of 120nM per hairpin in 10% DS/2X SSC solution, pipette it to the hybridization chambers and incubate by slowly rotating them for 1.5h in the dark (Note G). After this step, the samples should be protected from light until they are imaged.

4. Wash three times with 30% FA / 2x SSC and three times with 2X SSC (Note F). The samples are now ready for imaging. If nuclear staining is desired, a 10 minute 1X DAPI /2X SSC staining can be done between the three sets of washes.

5. Since individual transcripts are imaged as diffraction limited dots, the use of anti-bleach buffer is highly recommended especially when photo-unstable fluorescent dyes are used. Mix the components of the anti-bleach buffer immediately before imaging (the solution is effective only for a couple of hours) with the following ratios: 7 volumes of ABBB, 1 volume of 8% d-glucose, 1 volume of pyranose oxidase, and 1 volume catalase enzyme. Pipette the mixed solution into hybridization chambers, cover them with another cover glass to prevent evaporation, and start imaging.

6. After imaging, wash the samples twice with 2x SSC. Prepare the DNase I solution according the manufacturer’s specifications with the protein concentration of 500 units/ml. Add DNase to the chambers and incubate for 1h in order to strip the probes.

7. Wash the samples three times with 30% FA / 2x SSC and three times with 2X SSC (Note H).

8. Now the samples are ready for the next hybridization round, and the protocol can be repeated from step 5 in section 3.2.1 (1μM random 60-mer oligonucleotide blocking in 2X SSC for 1h) with a new set of genes (Note I). RNA integrity is maintained for weeks, and if necessary, one can take brakes in between the long measurement routine. In that case the samples should be stored at 4°C with 0.2% thiomersal and by making sure they do not dry.

3.2.3. Antibody staining of plasma membranes

Achieving single-cell resolution is critical for SGA. For a reliable cell segmentation, immunostaining of proteins localized on the plasma membrane should be used. In order to achieve a strong and uniform signal it may be necessary to use two or even more antibodies targeting different membrane proteins that are then visualized by using secondary antibodies conjugated with the same fluorophore (e.g. β-catenin together with E-cadherin were used for the dorsal neural tube). Most importantly for successful data-analysis, the target proteins should be chosen based on their expression in the entire tissue of interest to define all cell boundaries. The antibodies should be chosen and tested in the respective tissue after treatment with the SGA sample fixation and hybridization conditions beforehand prior to starting the hybridizations with the actual samples.

1. After the last hybridization and DNA stripping round, wash samples three times with 30% FA / 2XSSC and three times with 2x SSC before changing the buffer to 1X PBS. Block the samples for 1h with the immunostaining blocking solution (5% BSA, and 1% DMSO in 1X PBT).

2. Incubate the samples with primary antibodies in blocking solution overnight at 4°C while rotating slowly. This step should be performed in a humidified box and the hybridization chambers should be covered with parafilm to prevent evaporation (note E).

3. The next day, wash the samples 5X 30min with 1X PBS and then incubate with the secondary antibodies in blocking solution for 3h at room temperature followed by a wash 5x 30min with 1X PBS. Finally, apply the anti-bleach buffer before imaging.

3.3. Imaging and data analysis

A spinning disc confocal microscope is the most useful microscope for SGA, as it provides the advantage of high speed and optical sectioning when multiple samples are imaged on multiple channels daily, which can become time consuming. In addition, the resolution and low background of a spinning disc confocal microscope with relatively small photo bleaching of fluorophores enables detection of diffraction limited dots (Note J). It is important to use high quality oil or water objectives with a magnification that produces ~ 120-180nm pixel resolution combined with a high quantum efficiency CCD or sCMOS camera. Six orthogonal channels (Cy7, Alexa647, Alexa594, Cy3B, Alexa488, and DAPI) that are well separated from each other with high quality emission and dichroic filter sets are routinely used. The imaging routine is the following:

3.3.1. Imaging

1. Begin by checking the RNA integrity of your sample as follows; choose a highly expressed gene in your tissue of interest for which you have been able to design a set of 24 probes. Design every other probe with one initiator/fluorophore, and the other half with another initiator/fluorophore combination (e.g. odds with Cy7-B1 and evens with Alexa647-B2). Perform the hybridization protocol, image with both channels and check to what extent the signal from two images with different channels overlaps by counting the percentage of co-localization of the diffraction limited dots. If the co-localization is >85%, the RNA integrity of the transcripts is acceptable and the experiment can be continued.

2. The N number of hybridization sets with S orthogonal channels are imaged and the DNA probes are stripped between the hybridizations. The number of genes that are measured during an experiment will scale linearly to N×S.

3. To check the RNA integrity after the final hybridization rounds, repeat the probe set used in the first hybridization set and check that the amount of dots in each cell correlate with the images from the first set. You should expect to see a >80% recovery of the signal, and no less than a >60% level is acceptable.

4. As the last imaging step after completion of all the hybridization routines, image the plasma membrane antibody staining that will be used for cell segmentation purposes during the data analysis.

3.3.2. Data-analysis

Data analysis is performed by using Ilastik toolkit [7] (ilastik.org) and Matlab scripts developed specifically for SGA which can be found at www.singlecellanalysis.org (this website also has a discussion forum and contact information in case help is needed during any step of the SGA protocol or analysis).

1. The Ilastik software is trained to detect plasma membranes and cell interiors, and this information is used to export a probability density map in a .h5 format to present these values in a 2D matrix.

2. Cell interiors from the .h5 file are used in Matlab as a seed for a watershed algorithm to detect cell boundaries and to convert each cell into 3D volumetric objects in space.

3. Ilastik is trained to recognize the diffraction limited dots from the images and this data is then exported in .h5 format.

4. 3D cells are aligned with dot images by using a semi-automated alignment routine in the Matlab program that is based on recognition of easily detectable unique morphological features of each sample.

5. The Matlab program counts dots (e.g. number of transcripts of each gene in each cell volume), and stores that data as cell volume corrected values for further analysis. This way, each cell ends up having volume corrected values of transcripts for each gene, and that data is stored in a matrix format.

6. The cells are hierarchically clustered based on their gene-expression profile and presented in a heatmap format. The clusters are mapped back to spatial context by visualizing each cell in a defined cluster with the same color (Figure 2).