DEVELOPMENTAL DYNAMICS 244:774–784, 2015 DOI: 10.1002/DVDY.24274

TECHNIQUES

Quantitative Single Cell Gene Expression Profiling in the Avian Embryo a

Jason A. Morrison,1 Andrew C. Box,1 Mary C. McKinney,1 Rebecca McLennan,1 and Paul M. Kulesa1,2* 1

Stowers Institute for Medical Research, Kansas City, Missouri Department of Anatomy and Cell Biology, University of Kansas School of Medicine, Kansas City, Kansas

Developmental Dynamics

2

Background: Single cell gene profiling has been successfully applied to cultured cells. However, isolation and preservation of a cell’s native gene expression state from an intact embryo remain problematic. Results: Here, we present a strategy for in vivo single cell profiling that optimizes cell identification, isolation and amplification of nucleic acids with nominal bias and sufficient material detection. We first tested several photoconvertible fluorescent proteins to selectively mark a cell(s) of interest in living chick embryos then accurately identify and isolate the same cell(s) in fixed tissue slices. We determined that the dual color mDendra2 provided the optimal signal/noise ratio for this purpose. We developed proper procedures to minimize cell death and preserve gene expression, and suggest nucleic acid amplification strategies for downstream analysis by microfluidic reverse transcriptase quantitative polymerase chain reaction or RNAseq. Lastly, we compared methods for single cell isolation and found that our fluorescence-activated cell sorting (FACS) protocol was able to preserve native transcripts and generate expression profiles with much higher efficiency than laser capture microdissection (LCM). Conclusions: Quantitative single cell gene expression profiling may be accurately applied to interrogate complex cell dynamics events during embryonic development by combining photoconversion cell labeling, FACS, proper handling of isolated cells, and amplification stratC 2015 Wiley Periodicals, Inc. egies. Developmental Dynamics 244:774–784, 2015. V Key words: single cell; RNAseq; RT-qPCR; gene expression; whole transcriptome amplification; expression analysis; molecular profiling; LCM; FACS; avian embryo; photoconversion Submitted 7 January 2015; First Decision 6 March 2015; Accepted 6 March 2015; Published online 24 March 2015

Introduction The dynamic nature of early development makes it challenging to understand the mechanistic underpinnings of complex cell behaviors that shape the embryo. Genomic profiling has rapidly emerged as a means to better link molecular information with particular morphogenetic events. However, traditional transcriptome approaches have focused on analyzing large cell populations that average the information of individual cells. Now, single cell studies are revealing molecular heterogeneities at the individual cell level that may offer clues to pattern formation within a cell population (Huang, 2009; Eberwine et al., 2014; Macaulay and Voet, 2014; Patel et al., 2014; Pollen et al., 2014; Shalek et al., 2014; Stegle et al., 2015). Unfortunately, there are inherent difficulties associated with cell isolation and robust quantification of gene expression from embryos which have limited the collection of molecular information (Shapiro et al., 2013; Macaulay and Voet, 2014). Thus, techniques aimed at more efficiently extracting a cell(s) from the embryo and interrogating its gene expression profile would address this roadblock and improve our understanding of embryonic development.

*Correspondence to: Paul M. Kulesa, 1000 E 50th Street, Kansas City, MO 64110. E-mail: [email protected]

774

Current techniques to isolate embryonic cells of interest often rely on manual tissue dissection of the embryo and subsequent dissociation of cells for FACS. Alternatively, a cell(s) of interest may be isolated from live tissue by micropipette or from fixed tissue sections by laser capture microdissection (LCM) (Morrison et al. 2012). FACS is typically used when large cell populations are readily accessible for dissection and isolation. In contrast, LCM is commonly used to isolate single and small cell numbers when cell numbers are limited or molecular markers to uniquely identify cells are unknown. In all of these scenarios, it remains difficult to time the cell isolation to occur during a particular dynamic process and limit changes in gene expression as a result of the manipulation. Photoconversion cell labeling has emerged as an efficient tool to selectively mark cells of interest within the living embryo and follow subsequent dynamic cell movements over time (Stark and Kulesa, 2005a, b, 2007; McKinney et al., 2011, 2013; Caneparo et al., 2011). Photoconvertible fluorescent proteins loaded into cells and excited by laser light may undergo an irreversible photoconversion from one fluorescent color to another. Thus, integrating in vivo photoconversion cell labeling with identification and isolation of the same cell(s) in fixed tissue slices by LCM or Article is online at: http://onlinelibrary.wiley.com/doi/10.1002/dvdy. 24274/abstract C 2015 Wiley Periodicals, Inc. V

AVIAN SINGLE CELL GENE EXPRESSION PROFILING 775

Developmental Dynamics

FACS would enhance our ability to correlate molecular information with specific morphogenetic events. In this report, we examine combinations of photoconversion cell labeling, LCM, FACS, immunohistochemistry, RNAseq and reverse transcriptase quantitative polymerase chain reaction (RTqPCR) to more precisely extract gene expression information from a cell(s) of interest in the developing embryo. We test several different photoconvertible fluorescent proteins in a strategy to selectively mark a cell(s) of interest in the living chick embryo then accurately identify and isolate the same cell(s) in fixed tissue slices. We compare LCM and FACS methods for single cell isolation and measure the similarity of profiles and the percentage of single cells sufficiently profiled from each method. Lastly, we develop procedures to minimize cell death and preserve native gene expression profiles. We suggest nucleic acid amplification strategies for downstream gene expression analysis by microfluidic RT-qPCR or RNAseq. Together, our results offer an efficient means to mark, identify and isolate a cell(s) from the embryo and more accurately assess molecular information during embryonic development.

Results In Vivo Photoconversion Using mDendra2 Allowed the Selective Marking and Accurate Identification of a Single Cell(s) for LCM Isolation in Fixed Tissue Slices There are several challenges to selectively mark a cell(s) of interest within the living embryo and later identify the same cell(s) in a fixed tissue section for LCM isolation and profiling. This is primarily due to: (1) the rapid flash freezing, cryosectioning, and tissue slice desiccation postphotoconversion that is necessary to prevent degradation by endogenous RNases and preserve native transcripts; (2) the difficulty in accurately identifying a single photoconverted cell(s) within a fixed, cryosectioned tissue due to tissue autofluorescence and lack of anatomical landmarks to correlate the in vivo cell position within the appropriate tissue slice. As such, we had to revise our previous photoconversion cell labeling approach (Stark and Kulesa, 2005a,b, 2007). To address the first challenge, we evaluated four different photoconvertible fluorescent proteins (PCFPs), including mKikGR (Habuchi et al., 2008), Kaede (Ando et al., 2002), PSCFP2 (Chudakov et al., 2007a), and mDendra2 (Chudakov et al., 2007b) for the best signal/noise ratio to identify an in vivo photoconverted cell(s) after rapid flash freezing, cryosectioning, and tissue slice desiccation (Fig. 1; Table 1). Each PCFP was individually tested by first microinjecting the PCFP into the lumen of the neural tube and electroporation delivery into premigratory neural crest cells. We chose the neural crest model system because these cells migrate into unlabeled tissue and are accessible to identification and single cell extraction. After a short (680 nm) and an optimized filter set (605/55 nm) makes photoconverted mDendra2 cells more distinguishable from background autofluorescence. R–U: Spectra of photoconverted cells and autofluorescence. T–U: The shaded gray regions indicate the unoptimized emission filter spectra from (T) 625–705 nm and (U) 577.5–632.5 nm. Scale bars ¼ 50 mm.

AVIAN SINGLE CELL GENE EXPRESSION PROFILING 777

Developmental Dynamics

TABLE 1. Comparison of Photoconvertible FPs for Intravital Marking and Post-Fixation Isolation of Single Cells by LCM Photoconvertible FP

Pros & Cons

KikGR Kaede monomeric Dendra2 psCFP2

Unable to distinguish converted KikGR from autofluorescence, even with spectral imaging Converted Kaede signal is dim compared to other photoconvertible constructs Converted mDendra2 is bright compared to other photoconvertible constructs Less autofluorescent background. Potential for converting when imaging

from neural tube cultures by LCM (Fig. 2G). These results suggest that our method to minimize cell death and preserve native gene expression profiles in FACS were very effective, nearly replicating the expression profiles of single cells flash frozen in tissue and isolated by LCM. We then compared the relative efficiencies of each isolation technique to generate single cell expression profiles based upon outliers identified using Fluidigm’s Singular outlier detection tool (Fig. 2H). We found that FACS had the highest rate of success, generating usable single cell profiling data for over 96% of the single cells harvested (Fig. 2H). In comparison, we determined that the LCM-based isolation methods (cryosectioned tissue and neural tube cultures) both produced successful gene expression profiles in approximately 60% of the single cells analyzed (Fig. 2H). These outcomes suggest that when sufficient cells are presented for collection and optimized protocols are used, FACS is able to generate single cell profiles with much higher efficiency and speed than LCM without significantly altering native single cell gene expression profiles.

An Integrated Method for Quantitative Single Cell Gene Profiling in the Embryo Involves Vital Cell Labeling, Rapid Tissue Isolation, and Robust Preamplification To cover the broadest application of single cell profiling, we sought to optimize an integrated protocol that allowed flexibility in: (1) vital labeling of a single cell(s) of interest; (2) single cell isolation techniques; and (3) robust gene expression analysis methods. We schematically describe critical aspects of quantitative single cell gene profiling in the embryo (Fig. 3). Important considerations include vital cell labeling and rapid removal of tissue from the embryo (Fig. 3). Depending on the isolation method used, this may include important considerations of laser power for accurate tissue cutting and extraction of a cell(s) of interest by LCM (Fig. 3) or optimization of nozzle size and system pressure during FACS (Fig. 3). Lastly, we note the importance of cDNA synthesis and nuclei acid amplification before single cell profiling (Fig. 3).

How Many Single Cells Should be Harvested? Recent studies have described the analysis of cell numbers ranging from the tens to hundreds of cells (Shalek et al., 2013; DurruthyDurruthy et al., 2014; Shalek et al., 2014; Wu et al., 2014). The number of cells required depends on the rarity of cell types to be confidently identified and compared, and the costs associated with profiling a large number of single cells. For example, we focused on migratory neural crest cells that travel through extracellular matrix and loosely connected mesoderm. Cranial neural crest cell streams in the chick embryo typically number in the hundreds of

cells and when fluorescently labeled may be distinctly identified from the unlabeled surrounding tissue through which cells travel. Photoconversion of migrating neural crest cells allowed us to interrogate single cells within distinct microenvironments along the migratory pathway. We typically analyze a minimum of 48 single neural crest cells per condition for single cell experiments. For this study, approximately 138 genes were interrogated by RT-qPCR, while RNAseq typically detected more than 2,000 genes per single neural crest cell (data not shown).

How Deep Should Single Cells be Sequenced? As single cell sequencing is still an emerging technology, appropriate sequencing depth is widely discussed. Shapiro and colleagues propose an RNAseq depth of 2 million reads per single mammalian cell (Shapiro et al., 2013). However, Shalek and colleagues found little benefit in sequencing beyond 1 million reads per cell (Shalek et al., 2013). Other groups report that substantially different cells can be distinguished with 20,000 or fewer reads (termed ultra-low depths) per single cell (Jaitin et al., 2014; Pollen et al., 2014). While economically advantageous and useful for specific applications, ultra-low depth enables detection but not accurate quantification of transcripts expressed a low levels (Pollen et al., 2014). Ultimately, the proper number of reads per single cell will depend on some combination of the comparison(s) to be made from the resulting data, the cell type(s) analyzed, costs and the sequencing platform used. In practice, we sequenced each single neural crest cell at a depth of approximately 5 million reads using paired-end RNAseq. We observed few additional transcripts detected when sequencing beyond 1 million reads per cell (Fig. 4).

Maintaining Cell Health During FACS A common concern during FACS is that cells appear unhealthy or not viable. There are several aspects that can affect cell health. First, experiments should be performed in a rapid and smooth manner to minimize cell death. This includes close coordination between the time of harvest, dissociation and sorting of tissue. Proper handling of cells and tissue is then critical to achieving good quality results. During tissue dissociation, we describe a combination of mechanical and enzymatic dissociation (Experimental Procedures). Delicate cell types may not respond well to these conditions and may require a specific dissociation protocol (Bourne, 1986). Lastly, FACS settings, particularly the nozzle size and system pressure, should be set based upon the cell type of interest (Fig. 3). We found that settings of 100 mm nozzle and 20 psi successfully produced gene expression profiles for more than 96 percent of the single chick neural crest cells analyzed (Fig. 2H). This setting may require optimization for other cell types/ sizes.

Developmental Dynamics

778 MORRISON ET AL.

Fig. 2. Single cell isolation by FACS or LCM. A: A typical chick embryo (cranial and post-otic regions, HH St 15) with YFP-labeled neural crest cell migratory streams. B: Representative region (box in (A)) showing a typical pre-otic neural crest cell migratory stream (YFP-labeled) manually isolated from the embryo. Distal branchial arch 2 tissue containing two YFP-labeled neural crest cells (arrows) is excised (dashed line) and segregated from the remaining proximal stream. Specific tissue of interest can be dissociated and cells sorted for single cell profiling. C: Typical transverse cryosection through the cranial region of a HH St 15 chick embryo containing the NT. D: YFP-labeled migratory neural crest cells are brighter than the surrounding unlabeled, autofluorescent tissue and have emigrated from the dorsal NT. D0 : Arrowheads point to single, YFPlabeled migratory neural crest cells that are (D0 –F) cleanly removed from the surrounding tissue by LCM. The scale bar in (D) is 150 mm. The scale bar in (D0 ) is 75 mm and applies to (D0 –F). G: ANOVA pairwise correlation comparing significantly different genes (P < 0.05) from 138-gene expression profiles from each single cell isolation method. (FACS: n ¼ 32 cells; LCM from cryosections: n ¼ 36 cells; LCM from neural tube cultures: n ¼ 24 cells) (H) Percentage of single cells successfully profiled by isolation method. NT ¼ neural tube.

Developmental Dynamics

AVIAN SINGLE CELL GENE EXPRESSION PROFILING 779

Fig. 3. Critical aspects of single cell isolation in the embryo. Key steps for single cell isolation by either LCM or FACS are sequentially displayed with an emphasis on those that are critical to successful single compared with multi-cell sample harvest. (1) Cells of interest are labeled and (2) tissue extracted from the embryo into chilled buffer by common techniques before the protocol diverges for tissue preparation. (3a) Tissue must be flash frozen and cryosectioned for single cell isolation by LCM. (4a) Precise laser cutting of single cells demands fine-tuning of both the laser power and focus. With proper settings, laser ablation widths can be diminished to 1 mm. (3b) FACS requires live tissue dissociation to single cells and viability staining. (4b) Nozzle size and system pressure are both important for maintaining cell health during the sorting procedure. Desired single cells must be angled away from the central waste dump toward the bottom of the collection wells during FACS for the most efficient single cell isolation. Regardless of isolation technique, (5) single cell lysates must be converted to cDNA and (6) the resulting nucleic acids amplified before being (7) robustly interrogated by RNAseq or RT-qPCR.

Optimizing FACS and LCM Cell Isolation During FACs isolation, the efficiency in generating single cell profiles can be low, increasing both the cost and time required for a successful experiment. When a large percentage of single

cell reactions fail to produce profiling results, this usually indicates inaccurate deposition of single cells into buffer during FACS sorting. We discuss optimizing techniques using a side stream approach (Fig. 3; Experimental Procedures section). We

780 MORRISON ET AL.

tion efficiency in a subset of wells. This may be the case when different reagents or equipment are used for different reactions. We recommend reviewing the experimental design to ensure that all reactions are processed under the same conditions.

Discussion

Developmental Dynamics

Conclusions on Single Cell Expression Analyses in the Embryo

Fig. 4. Single cell RNAseq beyond 1 million reads per cell detects few additional genes. Saturation plot showing the number of genes detected (FPKM > 1) when each of 38 single cells is sampled to different numbers of reads (millions). Each shade of blue represents a different single cell. Using bootstrapping results from 4 sub-samples per sum-sample sequencing depth, the upper and lower bounds of the error ribbon (mean 6 standard deviation) are represented with gray shading between. Each single cell was sequenced to a depth of approximately 5 million reads; however, the saturation plots were cut off at either 4 million reads or the point of oversampling.

find that it may also be beneficial to map the failing single cell reactions back onto the plate wells to identify any patterns of failure. For FACS, we suggest using the instrument’s single event sort mode, when available, and using fluorescently labeled microspheres (e.g., Life Tech A-16503) before the actual experiment to validate accurate single event sorting. Second, robotic instrumentation is commonly calibrated based upon the coordinates of a single well (for example, A1) in a 96well plate. As such, accuracy may deteriorate as the instrument progresses through subsequent wells. To resolve this issue, we typically register both the first and last wells of the plate. We also find that deposition of test puddles onto the lid of a 96-well plate can be helpful in determining accurate placement of samples into wells. Alternatively, sorting of fluorescent microspheres into a 96-well plate and imaging of individual wells may verify the well targeting before sorting experimental samples. When isolating single cells, especially by LCM, some samples may have globally elevated expression values compared with other single cells. Elevated expression values in some single cells compared with others is often caused by accidental collection of multiple, instead of single cells per well. These wells are often eliminated as outliers during analysis, but are a costly waste of time and reagents. To minimize the chances of accidental multicell harvest into the same well by LCM, we recommend to always cryosection tissue to a single cell thickness and use membranebound labeling to determine the perimeter of each cell. Alternatively, elevated expression values may result from increased reac-

Single cell analysis is dramatically changing our understanding of cellular states and function. To broaden the examination and understanding of the inherent heterogeneities of cells within intact tissues and embryos, techniques for small cell numbers and single cell extraction must be adapted to in vivo scenarios. Here, we detailed complete methods for gene analysis that maintain native expression profiles using fluorescent cell marking in the vertebrate chick embryo and single cell isolation by either LCM or FACS. We showed that photoconversion cell labeling with mDendra2 and an optimized filter set allowed for the selective marking and proper identification of a single cell(s) after embryo fixation and tissue sectioning, in comparison to other PCFPs tested. We also showed that with proper handling and precautions, these methods produce highly correlative results and recommend specific cDNA amplification techniques suitable for analysis of up to 300 genes by microfluidic RT-qPCR or whole transcriptome interrogation by means of RNAseq. We discussed the benefits and limitations of each method and suggested an ability to customize workflows without altering native gene expression profiles. New techniques such as multiplex fluorescent in situ hybridization chain reaction allow more accurate readout of multiple gene expression patterns in the same tissue with single cell resolution (Choi et al., 2010, 2014). In summary, quantitative single cell gene expression profiling methods are rapidly advancing and offer a powerful approach to interrogate the mechanistic basis of complex cell dynamics events during embryonic development.

Experimental Procedures Fluorescent Cell Labeling Using Photoconversion Chick embryos at HH St 9–10 (Hamburger and Hamilton, 1951) were microinjected and electroporated in ovo with a PCFP at a concentration of 5 mg/ml using a stereomicroscope (Leica Microsystems, SV1000) and platinum electrodes, as described in Stark and Kulesa 2005a, b. Electro-poration techniques for a wide range of chick developmental stages have been developed (Itasaki et al., 1999; Krull, 2004; Stark and Kulesa, 2005a,b; Cui et al., 2007). Embryos were re-incubated at 38 C until a desired developmental stage and placed onto a glass slide for photoconversion. Several different PCFPs Kaede (MBL International Corp. AMV0012-NP), KikGR (MBL International Corp. AM-V0150-NP), mDendra2 (Chudakov et al., 2007a), or PSCFP2 (Evrogen pPSCFP2-N) were tested for their photoefficiency and ability to withstand fixation and tissue sectioning processes to accurately identify photoconverted cells of interest. Labeled cells of interest in embryos were photoconverted using 405 nm laser light (LSM 780, Carl Zeiss), according to our previous methods described in Stark and Kulesa (Stark and Kulesa, 2007), and subsequently observed by collecting distinct emission spectra (mDendra2, 566– 674 nm; PSCFP2, 495–658 nm; Kaede, 566–674 nm). Manually

AVIAN SINGLE CELL GENE EXPRESSION PROFILING 781

dissected tissue was flash frozen, cryosectioned and desiccated as previously described (Morrison et al., 2012), except that tissue freezing medium (TFM; VWR, 15148-031) was not removed before LCM.

Developmental Dynamics

Single Cell Isolation by Laser Capture Microdissection We suggest specific additional precautions to isolate single cells from cryosections (Figs. 2C–F, 3) that are distinct from previous descriptions for multicell LCM isolation (Morrison et al., 2012). To ensure precise single cell isolation, it is critical that LCM laser power and focus be optimized to provide sufficient cutting power with minimal cutting width (Fig. 3). Tissue type, thickness, and batch effects from processing will all affect the settings required for proper LCM harvest of single cells. Thus, all parameters should be thoroughly tested on nonexperimental samples from within the experimental batch to establish correct parameters. We performed single cell LCM on the Zeiss LCM system (PALM MicroBeam, Carl Zeiss) using a 40 long distance objective with 0.6 numerical aperture. We started with the laser cutting power set at 39 and focus at 61 to obtain the smallest effective cutting width. Single cells require less catapulting laser energy compared with multicellular samples. Laser pressure catapult (LPC) energy set between 10 and 20 is commonly sufficient to remove single cells (Figs. 2D,E, 3). We have determined that TFM does not interfere with preparation of template for downstream expression analyses and thus need not be removed as previously described (Morrison et al., 2012). Immediately after sectioning, cryosections on slides should be sealed in 50-ml conical tubes and placed at 80 C until screening or LCM. Upon removal from the freezer, immediately dehydrate sections on slides by placing them directly into a desiccation chamber. An inherent limitation of single cell isolation by LCM is the inadvertent capture of overlapping portions of neighboring cells. Cryosections should be cut to roughly a single cell thickness (e.g., 10 mm), but protrusions from adjacent cells may overlap within this thickness. Therefore, the relationship of the single cells of interest to their neighbors should be considered when using this method. The benefits of LCM are maintenance of each cell’s microenvironment by means of flash freezing and visual identification of cells of interest by morphological characteristics as well as known markers (immunohistochemistry).

Single Cell Isolation by FACS To maintain cell health and preserve gene expression state, it is critical to work rapidly under chilled and RNase-free conditions. By combining these precautions with the sorting parameters listed below, it is possible to obtain a very high analysis of variance (ANOVA) pairwise correlation between single cells harvested by LCM from cryosections and FACS (Fig. 2G). Because tissue is immediately flash frozen and kept frozen or desiccated until analysis, samples harvested by LCM more closely represent native expression profiles. While the expression profiles generated by LCM and FACS are highly correlative (Fig. 2G), other distinctions should be considered when selecting a method of single cell harvest. In addition to availability of instrumentation, cost, and the efficiency with which each method produces single cell profiles should be considered (Fig. 2H). For example, when isolating single cells by

FACS, were are able to generate expression profiles for >96% of the cells harvested, but noted a reduced efficiency for cells harvested with LCM (Fig. 2H). FACS isolation, however, requires known markers or other cellular attributes to distinguish cells of interest and is generally less efficient with smaller amounts of input material.

Tissue Harvest for FACS Control samples should be harvested and processed before experimental samples when applicable and appropriate. This will ensure complete tissue dissociation for the cell type of interest and optimize the setting of FACS gates. Quickly isolate tissue containing the cells to be dissociated and sorted and place them in chilled 0.1% DEPC PBS, pH 7.4 on ice. Transfer the tissue to a DNaseand RNase-free 2-ml tube and remove as much liquid as possible from the samples.

Tissue Dissociation for FACS For the most rapid and complete, yet gentle, tissue disaggregation into single cells, we combined enzymatic and mechanical dissociation protocols. Rapidly draw samples through a 25 gauge needle (BD 305122) attached to a 1 ml syringe (Covidien 8881501400) a few times. Add 0.5 ml 0.25% trypsin/EDTA (Life Tech 25200-056) per sample. Volumes of liquid added may be adjusted for larger or smaller tissue samples. Incubate 3 min in a 37 C water bath. Stop the reaction by adding 0.25 ml fetal bovine serum (FBS) (HyClone SH30070.03). Dissociate each sample individually by drawing samples through a 25-gauge needle attached to a 1-ml syringe a few times. Pellet cells by centrifugation at 4,000 rpm for 3 min at ambient temperature. Remove the liquid and resuspend the cells in 1 ml 0.1% DEPC PBS, pH 7.4 with 2% FBS. Centrifuge at 4,000 rpm for 3 min at room temperature. Re-suspend cells in 250 ml of 0.1% DEPC PBS, pH 7.4, place samples on ice and sort immediately.

Cell Sorting The number of single cells to be harvested should be determined in order to prepare the required amount of lysis solution. Aliquot an appropriate amount of lysis solution into each well of a 96well plate and keep the plate on ice until sorting. Stain cell suspensions with 5 nM SYTOX (Life Tech S34859) red for 5 min on ice to label necrotic/dead cells. Sort single cell suspensions of disaggregated tissue on a cell sorter, such as the MoFlo Legacy (Beckman Coulter), using a 100 mm nozzle tip (Beckman Coulter ML04230) at 20 psi system pressure, 1 Leinco preservative free Clearflow sheath fluid (Leinco Tech S621), and single sort mode. Alternatively, when operating a sorter lacking a 633 nm or 647 nm laser line, propidium iodide (PI) (Life Tech P3566) using 488 nm or 561 nm excitation or DAPI (Life Tech D1306) excited with 405 nm or UV can be used to label dead cells in lieu of SYTOX red. If using PI or other dyes that exhibit spill over into other detectors of interest, take care to set color compensation using single-color controls. The dead cell label should be visualized on a log scale. Single and live cell (e.g., SYTOX red negative) gates should be used to select the desired cellular events from dead cells, debris, clumps and other undesired events using a scatter plot of forward-scattered light (FSC) vs. side-scattered light (SSC) signal intensities. The location of dead cells in a scatter plot can be

Developmental Dynamics

782 MORRISON ET AL.

verified by back-gating dead cells from an ungated plot of SYTOX red intensity onto the FSC vs. SSC scatter plot. Once the distribution of live cells is identified in light scatter, a single-cell gate that removes doublets and other higher order clumps should be generated using suitable FSC and/or SSC pulse parameters. When sorting GFP (or Alexa Fluor 488) positive events, it is beneficial to plot green vs. yellow fluorescence after gating on light scatter, single events and live cells (that is, plot fluorescein isothiocyanate [FITC] vs. R-phycoerythrin [PE] detector intensity). Events with detectable GFP will typically appear off the diagonal in the scatter plot, with higher signal in the green vs. yellow channel (autofluorescence is often equally bright in both channels). This type of pattern observed from a GFPpositive population in a scatter plot of FITC vs. PE fluorescence is highly characteristic and easily identifiable. Most cell sorter instrument vendors provide several pre-set sort modes such as enrich, purify and single event sorting. Using a purify mode is suitable for bulk sorting of large cell numbers but potentially allows sorting of >1 event per well in the case where two gated events occupy the same droplet. A single event sort mode will limit the events sorted to drops containing no more than one of the desired cells. To ensure that the sorting system will deposit the correct number of cells per well, use an initial sort with various numbers of fluorescent microspheres into individual wells of a plate and manually count the number of microspheres deposited into each well on an appropriate imaging platform. This counting can be automated with beads of diameter greater than 10 mm. Accurate targeting of side streams to wells of a multi-well plate is critical for successful single cell deposition. When setting up the sorter for plate sorting, it is worthwhile to consider the angle at which side streams approach the target receptacle. Side streams should be adjusted such that deflection of a stream away from the center waste dump is not at a steeper angle than necessary to prevent sorted events from hitting the waste (Fig. 3). This will maintain more vertical side streams than with higher deflection values and help accurately target the bottom of a small receptacle. A final point to be aware of is the effect of large particles on droplet breakoff and the resulting loss of accuracy in targeting sorted events consistently into wells. As a rule of thumb, most sorter operators expect that particles with a diameter greater than 1/5 of the nozzle diameter will interfere with clean droplet breakoff. This will limit the ability to generate tightly targeted side streams (Shapiro, 2003). Therefore, it is highly recommended to filter samples through a 20- or 30-mm nylon mesh filter before sorting to ensure side streams consistently land in the center of wells, when performing single event sorting. This has the added benefit of reducing the likelihood of a large particle clogging the nozzle tip during the sort. This type of event can make an entire plate of material unusable, due to spraying of sheath fluid and unsorted material.

cDNA Synthesis, Pre-amplification, and Microfluidic RTqPCR from Cell Lysates Linear, gene-specific pre-amplification of transcripts for RTqPCR has been previously published (Morrison et al., 2012; Van Peer et al., 2012). Ambion’s Taqman PreAmp Cells to Ct kit (Ambion 4387299) allows cDNA synthesis directly from cell lysates, eliminating the need for RNA isolation, during which precious RNA can be lost. In addition to the time savings, Van

Peer and colleagues reported increased gene profiling sensitivity when cDNA was synthesized from lysate compared with purified RNA (Van Peer et al., 2012). Unfortunately, lack of RNA isolation eliminates the possibility of directly analyzing the quality of the RNA under examination. We have, however, indirectly determined RNA quality by analyzing RNA in nonexperimental cells from sample batches. For LCM-harvested samples, this is performed by isolating and analyzing the RNA from tissue remaining on the slide following LCM. This generally yields RNA Integrity Numbers (RINs) above 6.5 on a Bioanalyzer 2100 (Agilent) (Morrison et al., 2012). RNA quality of FACS isolated cells can be inferred from the collection and interrogation of RNA in a nondesired fraction of sorted cells. We routinely record RINs above 9 for RNA in FACS-isolated cells. For single cell LCM, a few critical amendments to our previously published protocol (Morrison et al., 2012) were required. First, the cDNA synthesis reaction volume was reduced proportionately to 16 ml, which included the entire 11 ml of lysate as template. Second, the downstream linear, gene-specific preamplification reaction volume was likewise minimized to 20 ml with 5 ml of cDNA as template and 18 pre-amplification cycles. Products should be diluted 1:4 with 1 TE before interrogation by microfluidic RT-qPCR. This approach allows for analysis of up to 300 genes from a single cell by microfluidic RT-qPCR.

cDNA Synthesis, Whole Transcriptome Amplification (WTA), Library Construction for RNAseq From Cell Lysates Several strategies for amplifying transcripts from small amounts of starting material for sequencing have been published and reviewed (Phillips and Eberwine, 1996; Tang et al., 2009; Islam et al., 2012; Shapiro et al., 2013; Lee et al., 2014; Macaulay and Voet, 2014; Morris et al., 2014; Saliba et al., 2014). As noted in a recent comparison of commercially available kits (Shanker et al., 2015), we found Clontech’s SMARTer Ultra Low Input RNA kit (Clontech 634826) detected the highest percentage of transcripts that were present in unamplified bulk control samples (data not shown). The SMARTer kit also produced the best linear correlation between amplified and unamplified sample expression values (data not shown) and has been used in several single cell RNAseq studies (Ramskold et al., 2012; Adiconis et al., 2013; Picelli et al., 2013; Shalek et al., 2013; Marinov et al., 2014; Patel et al., 2014). SMARTer amplification, library construction using the Nextera XT DNA preparation reagents (Illumina FC-131– 1024) and sequencing on the Illumina HiSeq 2500 (Illumina) platform should all be performed according to the manufacturer’s instructions.

Single Cell Data Analysis Compared with traditional cell population profiling, there are many unique aspects to single cell expression analysis (Stegle et al., 2015). First, a level of detection (LoD) must be established to distinguish background noise from actual expression. In RTqPCR, LoD is calculated statistically (Livak et al., 2013), while true expression in RNAseq is determined as either fragments per kilobase of exon per million fragments mapped (FPKM) greater than 1 or with the use of external RNA control consortium (ERCC) spike-ins (Jiang et al., 2011). Second, in single cell RTqPCR, samples are not normalized for input value using the

Developmental Dynamics

AVIAN SINGLE CELL GENE EXPRESSION PROFILING 783

traditional delta Ct method. Instead, expression values are generated by subtracting the actual Ct value from the LoD. Third, the output from single cell expression analyses encompasses cellular variation as well as fold change differences and statistical significance. The variation is a combination of the percentage of single cells expressing the gene as well as the distribution of expression. A major advantage of single cell profiling is that data from as few as 10 single cells can be combined during analysis to represent traditional population profiles (Marinov et al., 2014; Pollen et al., 2014; Streets et al., 2014; Wu et al., 2014). All single cell RNAseq results were processed by means of the TUXEDO Suite bioinformatics pipeline with default settings as reported (Trapnell et al., 2013; Wu et al., 2014). When validating a single cell approach, single cell profiles should be compared with a “bulk” control profile generated from a population of cells that are as similar to the single cells harvested as possible. This comparison is one way to define the sensitivity of the single cell profiling method. Another method is to sequence ERCCs spike-in controls along with single cells. The varying concentrations of ERCCs allow plotting of expected to observed concentrations of RNA, enabling determination of a lower limit of concentration at which the results become less correlative with the expected results. Outliers were defined by either a reduced number of alignable reads or lower expression of genes common to all cells analyzed (Kumar et al., 2014).

Acknowledgements PMK would like to acknowledge partial support from NIH R21 NS092001 and the kind and generous funding from the Stowers Institute for Medical Research. We also thank members of the BioInformatics, Histology and Molecular Biology core facilities at the Stowers Institute for Medical Research. Fluidigm Biomark HD dynamic arrays were analyzed at the Children’s Hospital Boston IDDRC Molecular Genetics facility.

References Adiconis X, Borges-Rivera D, Satija R, DeLuca DS, Busby MA, Berlin AM, Sivachenko A, Thompson DA, Wysoker A, Fennell T, Gnirke A, Pochet N, Regev A, Levin JZ. 2013. Comparative analysis of RNA sequencing methods for degraded or low-input samples. Nat Methods 10:623–629. Ando R, Hama H, Yamamoto-Hino M, Mizuno H, Miyawaki A. 2002. An optical marker based on the UV-induced green-to-red photoconversion of a fluorescent protein. Proc Natl Acad Sci U S A 99:12651–12656. Bourne GH. 1986. International review of cytology: a survey in cell biology. Vol. 105. San Diego: Academic Press Inc. Choi HM, Beck VA, Pierce NA. 2014. Multiplexed in situ hybridization using hybridization chain reaction. Zebrafish 11:488–489. Choi HM, Chang JY, Trinh le A, Padilla JE, Fraser SE, Pierce NA. 2010. Programmable in situ amplification for multiplexed imaging of mRNA expression. Nat Biotechnol 28:1208–1212. Chudakov DM, Lukyanov S, Lukyanov KA. 2007a. Tracking intracellular protein movements using photoswitchable fluorescent proteins PS-CFP2 and Dendra2. Nat Protoc 2:2024–2032. Chudakov DM, Lukyanov S, Lukyanov KA. 2007b. Using photoactivatable fluorescent protein Dendra2 to track protein movement. Biotechniques 42:553, 555, 557 passim. Cui C, Rongish B, Little C, Lansford R. 2007. Ex Ovo Electroporation of DNA Vectors into Pre-gastrulation Avian Embryos. CSH Protoc 2007:pdb prot4894. Durruthy-Durruthy R, Gottlieb A, Hartman BH, Waldhaus J, Laske RD, Altman R, Heller S. 2014. Reconstruction of the mouse otocyst and early neuroblast lineage at single-cell resolution. Cell 157:964–978.

Eberwine J, Sul JY, Bartfai T, Kim J. 2014. The promise of singlecell sequencing. Nat Methods 11:25–27. Habuchi S, Tsutsui H, Kochaniak AB, Miyawaki A, van Oijen AM. 2008. mKikGR, a monomeric photoswitchable fluorescent protein. PLoS One 3:e3944. Hamburger V, Hamilton H. 1951. A series of normal stages in the development of the chick embryo. J Morphol 88:49–92. Huang S. 2009. Non-genetic heterogeneity of cells in development: more than just noise. Development 136:3853–3862. Islam S, Kjallquist U, Moliner A, Zajac P, Fan JB, Lonnerberg P, Linnarsson S. 2012. Highly multiplexed and strand-specific single-cell RNA 5’ end sequencing. Nat Protoc 7:813–828. Itasaki N, Bel-Vialar S, Krumlauf R. 1999. ’Shocking’ developments in chick embryology: electroporation and in ovo gene expression. Nat Cell Biol 1:E203–E207. Jaitin DA, Kenigsberg E, Keren-Shaul H, Elefant N, Paul F, Zaretsky I, Mildner A, Cohen N, Jung S, Tanay A, Amit I. 2014. Massively parallel single-cell RNA-seq for marker-free decomposition of tissues into cell types. Science 343:776–779. Jiang L, Schlesinger F, Davis CA, Zhang Y, Li R, Salit M, Gingeras TR, Oliver B. 2011. Synthetic spike-in standards for RNA-seq experiments. Genome Res 21:1543–1551. Krull CE. 2004. A primer on using in ovo electroporation to analyze gene function. Dev Dyn 229:433–439. Kumar RM, Cahan P, Shalek AK, Satija R, DaleyKeyser AJ, Li H, Zhang J, Pardee K, Gennert D, Trombetta JJ, Ferrante TC, Regev A, Daley GQ, Collins JJ. 2014. Deconstructing transcriptional heterogeneity in pluripotent stem cells. Nature 516:56–61. Lee JH, Daugharthy ER, Scheiman J, Kalhor R, Yang JL, Ferrante TC, Terry R, Jeanty SS, Li C, Amamoto R, Peters DT, Turczyk BM, Marblestone AH, Inverso SA, Bernard A, Mali P, Rios X, Aach J, Church GM. 2014. Highly multiplexed subcellular RNA sequencing in situ. Science 343:1360–1363. Livak KJ, Wills QF, Tipping AJ, Datta K, Mittal R, Goldson AJ, Sexton DW, Holmes CC. 2013. Methods for qPCR gene expression profiling applied to 1440 lymphoblastoid single cells. Methods 59:71–79. Macaulay IC, Voet T. 2014. Single cell genomics: advances and future perspectives. PLoS Genet 10:e1004126. Marinov GK, Williams BA, McCue K, Schroth GP, Gertz J, Myers RM, Wold BJ. 2014. From single-cell to cell-pool transcriptomes: stochasticity in gene expression and RNA splicing. Genome Res 24:496–510. McKinney MC, Fukatsu K, Morrison J, McLennan R, Bronner ME, Kulesa PM. 2013. Evidence for dynamic rearrangements by lack of fate or position restrictions in premigratory avian trunk neural crest. Development 140:820–830. McKinney MC, Stark DA, Teddy J, Kulesa PM. 2011. Neural crest cell communication involves an exchange of cytoplasmic material through cellular bridges revealed by photoconversion of Kik GR. Dev Dyn 240:1391–1401. Morris J, Bell TJ, Buckley PT, Eberwine JH. 2014. Antisense RNA amplification for target assessment of total mRNA from a single cell. Cold Spring Harb Protoc 2014:pdb prot072454. Morrison JA, Bailey CM, Kulesa PM. 2012. Gene profiling in the avian embryo using laser capture microdissection and RT-qPCR. Cold Spring Harb Protoc 2012. Patel AP, Tirosh I, Trombetta JJ, Shalek AK, Gillespie SM, Wakimoto H, Cahill DP, Nahed BV, Curry WT, Martuza RL, Louis DN, Rozenblatt-Rosen O, Suva ML, Regev A, Bernstein BE. 2014. Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344:1396–1401. Phillips J, Eberwine JH. 1996. Antisense RNA amplification: a linear amplification method for analyzing the mRNA population from single living cells. Methods 10:283–288. Picelli S, Bjorklund AK, Faridani OR, Sagasser S, Winberg G, Sandberg R. 2013. Smart-seq2 for sensitive full-length transcriptome profiling in single cells. Nat Methods 10:1096–1098. Pollen AA, Nowakowski TJ, Shuga J, Wang X, Leyrat AA, Lui JH, Li N, Szpankowski L, Fowler B, Chen P, Ramalingam N, Sun G, Thu M, Norris M, Lebofsky R, Toppani D, Kemp DW, 2nd, Wong M, Clerkson B, Jones BN, Wu S, Knutsson L, Alvarado B, Wang J, Weaver LS, May AP, Jones RC, Unger MA, Kriegstein AR, West JA. 2014. Low-coverage single-cell mRNA sequencing reveals cellular heterogeneity and activated signaling pathways in developing cerebral cortex. Nat Biotechnol 32:1053–1058.

Developmental Dynamics

784 MORRISON ET AL.

Ramskold D, Luo S, Wang YC, Li R, Deng Q, Faridani OR, Daniels GA, Khrebtukova I, Loring JF, Laurent LC, Schroth GP, Sandberg R. 2012. Full-length mRNA-Seq from single-cell levels of RNA and individual circulating tumor cells. Nat Biotechnol 30:777–782. Saliba AE, Westermann AJ, Gorski SA, Vogel J. 2014. Single-cell RNA-seq: advances and future challenges. Nucleic Acids Res 42:8845–8860. Shalek AK, Satija R, Adiconis X, Gertner RS, Gaublomme JT, Raychowdhury R, Schwartz S, Yosef N, Malboeuf C, Lu D, Trombetta JJ, Gennert D, Gnirke A, Goren A, Hacohen N, Levin JZ, Park H, Regev A. 2013. Single-cell transcriptomics reveals bimodality in expression and splicing in immune cells. Nature 498:236–240. Shalek AK, Satija R, Shuga J, Trombetta JJ, Gennert D, Lu D, Chen P, Gertner RS, Gaublomme JT, Yosef N, Schwartz S, Fowler B, Weaver S, Wang J, Wang X, Ding R, Raychowdhury R, Friedman N, Hacohen N, Park H, May AP, Regev A. 2014. Single-cell RNA-seq reveals dynamic paracrine control of cellular variation. Nature 510:363–369. Shanker S, Paulson A, Edenberg HJ, Peak A, Perera A, Alekseyev YO, Beckloff N, Bivens NJ, Donnelly R, Gillaspy AF, Grove D, Gu W, Jafari N, Kerley-Hamilton JS, Lyons RH, Tepper C, Nicolet CM. 2015. Evaluation of commercially available RNA amplification kits for RNA sequencing using very low input amounts of total RNA. J Biomol Tech [Epub ahead of print]. Shapiro E, Biezuner T, Linnarsson S. 2013. Single-cell sequencingbased technologies will revolutionize whole-organism science. Nat Rev Genet 14:618–630. Shapiro HM. 2003. Practical flow cytometry. New York: Wiley-Liss. 681 p.

Stark DA, Kulesa PM. 2005a. In vivo marking of single cells in chick embryos using photoactivation of GFP. Curr Protoc Cell Biol Chapter 12:Unit 12.8. Stark DA, Kulesa PM. 2005b. Photoactivatable green fluorescent protein as a single-cell marker in living embryos. Dev Dyn 233: 983–992. Stark DA, Kulesa PM. 2007. An in vivo comparison of photoactivatable fluorescent proteins in an avian embryo model. Dev Dyn 236:1583–1594. Stegle O, Teichmann SA, Marioni JC. 2015. Computational and analytical challenges in single-cell transcriptomics. Nat Rev Genet. Streets AM, Zhang X, Cao C, Pang Y, Wu X, Xiong L, Yang L, Fu Y, Zhao L, Tang F, Huang Y. 2014. Microfluidic single-cell wholetranscriptome sequencing. Proc Natl Acad Sci U S A 111:7048– 7053. Tang F, Barbacioru C, Wang Y, Nordman E, Lee C, Xu N, Wang X, Bodeau J, Tuch BB, Siddiqui A, Lao K, Surani MA. 2009. mRNASeq whole-transcriptome analysis of a single cell. Nat Methods 6:377–382. Trapnell C, Hendrickson DG, Sauvageau M, Goff L, Rinn JL, Pachter L. 2013. Differential analysis of gene regulation at transcript resolution with RNA-seq. Nat Biotechnol 31:46–53. Van Peer G, Mestdagh P, Vandesompele J. 2012. Accurate RTqPCR gene expression analysis on cell culture lysates. Sci Rep 2:222. Wu AR, Neff NF, Kalisky T, Dalerba P, Treutlein B, Rothenberg ME, Mburu FM, Mantalas GL, Sim S, Clarke MF, Quake SR. 2014. Quantitative assessment of single-cell RNA-sequencing methods. Nat Methods 11:41–46.

Quantitative single cell gene expression profiling in the avian embryo.

Single cell gene profiling has been successfully applied to cultured cells. However, isolation and preservation of a cell's native gene expression sta...
846KB Sizes 3 Downloads 9 Views