Fluorescent Activated Cell Sorting (FACS) Combined with Gene Expression Microarrays for Transcription Enrichment Profiling of Zebrafish Lateral Line Cells

Abstract

Transgenic lines carrying fluorescent reporter genes like GFP have been of great value in the elucidation of developmental features and physiological processes in various animal models, including zebrafish. The lateral line (LL), which is a fish specific superficial sensory organ, is an emerging organ model for studying complex cellular processes in the context of the whole living animal. Cell migration, mechanosensory cell development/differentiation and regeneration are some examples. This sensory system is made of superficial and sparse small sensory patches called neuromasts, with less than 50 cells in any given patch. The paucity of cells is a real problem in any effort to characterize those cells at the transcriptional level. We describe here a method which we applied to efficiently separate subpopulation of cells of the LL, using two distinct stable transgenic zebrafish lines, Tg(cldnb:gfp) and Tg(tnks1bp1:EGFP). In both cases, the GFP positive (GFP+) cells were separated from the remainder of the animal by using a Fluorescent Activated Cell Sorter (FACS). The transcripts of the GFP+ cells were subsequently analyzed on gene expression microarrays. The combination of FACS and microarrays is an efficient method to establish a transcriptional signature for discrete cell populations which would otherwise be masked in whole animal preparation.

I. Introduction

The lateral line (LL) is a superficially located sensory organ, which is increasingly used as an organ model to study complex cellular processes in the context of a whole living animal. The posterior LL (PLL) of early zebrafish embryos has provided better understanding of cell migration, proliferation and differentiation and has validated numerous data previously solely established in cell culture [1–3]. The PLL start to form around 20 hour post fertilization (hpf) when cranial placodal cells delaminate to become a coherent migratory primordium that traverses the length of the fish’s trunk and tail over the next 28 hours. As it migrates, the primordium deposits groups of cells which will rapidly differentiate into neuroepithelial patches, called neuromasts that contain the mechanosensory hair cells [4]. Regeneration of hair cells, which are in many ways similar to hair cells of the human inner ear, has been effectively studied in the PLL of zebrafish larva [5–10]. Both organs are comprised of discrete sensory patches which are composed of hair cells and accessory (support) cells providing other functions to those sensory epithelia. In most vertebrates (excluding mammals), accessory cells comprise a pool of progenitors that replace damaged or dead hair cells. These highly sophisticated cells are very prone to damage by ototoxic agents, mechanical chocks and noise. In most of the acquired hearing disorders described in humans, hair cells are missing, defective or fragile. The lack of regenerative capacity in mammals is preventing their replacement. Therefore the lateral line provides a valuable tool for the study of collective directional cell migration, differentiation of sensory cells from multipotent progenitor cells and hair cell regeneration. Thus, a full characterization of these events requires a description of the cellular histories (lineage) and knowledge about the molecules that regulate these processes.

Over the last years, zebrafish has become a powerful vertebrate model system for gene expression profiling of transcript levels on a genome-wide basis. A sequenced genome and large-scale mutagenesis screens have contributed to establish this model as a robust genetic system and have provided numerous examples of the structural and functional conservation of genes across all vertebrates [11–15].

Unfortunately, gene expression profiling in multicellular organisms is complicated by large differences in gene expression in different parts of the body. One major limitation to performing microarray studies is the complex numbers and intermixing of cell types generated during development. Using whole organs or embryos would result in a dilution and masking of many important genetic pathways, as the RNA from multiple cell types would be simultaneously hybridized to the arrays. Recently, approaches such as transgenic reporter lines, tissue microdissection and cell isolation through Fluorescence activated cell sorter (FACS) have been coupled with cDNA microarrays to circumvent this problem [2, 5, 16–18].

We describe here a robust method to isolate GFP+ cells from transgenic zebrafish embryo and larva that are expressing the reporter gene in a subpopulation of cells of the LL. A gene expression microarray-based comparison between sorted GFP+ and GFP− cells was subsequently performed for the transcriptome analysis. The application of this technology has provided insights into the complex molecular mechanisms underlying organogenesis and cell fate regulation during the formation of the PLL and regeneration of mechanosensory cells.

2. Methods

2.1 Embryo/larva preparation

This first step of the procedure differed slightly for the two lines. The Tg(cldnb:gfp) transgenic zebrafish line expresses GFP in the PLL primordium and neuromasts, but also in the olfactory system, the ear, the branchial arches, and the pronephros [19]. As we wished to isolate mostly cells from the primordium and the neuromasts and we wanted to avoid other labeled cell types, we dissected embryos’ tails by sectioning the animals at the end of the yolk extension (posterior to the cloaca) (Figure 1A). We used only the posterior third of the transgenic embryos at two different developmental stages, 36 and 48hpf. Prior to the dissection, the embryos were anaesthetized with Tricaine (MS-222, Sigma) and then, tails were transected with a sharp scalpel. Up to five hundred tails were recovered for each stage and transferred to Hanks medium. Whole 5 day old Tg(tnks1bp1:EGFP) larvae were processed without any dissection (Figure 1B). Up to two hundred larvae were anesthetized on ice and processed at the time.

Figure 1. Transgenic embryo and larva used for isolating cells from the lateral line by FACS.

A. A 36hpf transgenic Tg(cldnb:GFP) embryo is showing expression of GFP in the eye, the ear and in nascent posterior lateral line and migrating primordium. The anesthetized embryo was sectioned as the cloaca (red line) and the posterior part (blue arrow) was retained for processing by FACS. B. Six day old transgenic Tg(tnks1bp1:EGFP) larva expressing the reporter gene GFP in the accessory cells of the neuromasts (red arrows) in the lateral line. The eye and the guts are strongly reflecting the light but do not express the reporter (as verified by other means). Scale bars: in A and in B =200µm.

2.2 Embryo/larva digestion

This step has slight differences between the two transgenic lines mainly because of the different developmental stages that were used. To form a cell suspension, the tails or the whole larvae were transferred to an embryo glass dish (Electron Microscopy Sciences #70543-45) and all liquid was removed. We then added one ml of 0.25% trypsin, 1mM EDTA (Gibco, #25200) to the sectioned tails, or 3ml plus 200 µl of a digestion cocktail (10mg/ml collagenase (Sigma, #C9891-100MG), 5mg/ml hyaluronidase (Sigma #H3506-100MG), 20mg/ml proteinase K (Invitrogen, #25530-049)) to the whole larvae. We originally used dilutions of 0.25% trypsin, 1mM EDTA, but quickly started to use it undiluted. This improved the digestion significantly and was enough for a rapid digestion of younger embryo tails. Additional enzymes in the buffer were only necessary for the older whole larvae. This cocktail was built empirically based on the idea that older larva have significantly more conjunctive tissues which are mainly made of extracellular matrix (ECM) which would hamper the dissociation of the tissues. Furthermore, neuromasts are embedded in the superficial layer of the epidermis sitting on the ECM, which makes those points particularly relevant. Therefore, we decided to use collagenase, to digest collagen, which is the most abundant protein in the ECM and hyaluronidase to digest hyaluronan, which is one of the main components of the ECM. We added proteinase K, which is regularly used in in situ hybridization (ISH) for increasing the permeabilization of tissues, but also to inactivate DNAses and RNAses that might be present in the mix. The choice of concentrations was based on those used in other fish techniques (immuno-staining, ISH) and as it worked in our hands, were not modify further.

Constant mechanical dissociation was exerted by pipetting up and down the mixture at room temperature, in order to obtain as rapidly as possible a relatively homogenous solution. This step was critical for maintaining the cells, which had been isolated early in the process, healthy. Quality of the digestion was only assessed visually under the stereoscope looking for a homogenate and no extensive monitoring of the quality of RNAs was done at this early stage of the procedure.. Tissues from younger embryos were dissociating more rapidly (~15min) whereas those from older larva would take longer (~25min). Cell suspensions from tails were then filtered through 40 µm nylon mesh (Biologix # 15–1040) into a 15ml falcon tube and the digestion stopped by adding 1ml of L-15 Medium (Gibco, #11415) supplemented with 10% FBS. The isolated cells were pelleted by centrifugation at 600g for 2 minutes. Cells suspension from whole larvae, were filtered into a 50ml falcon tube and the digestion was stopped by adding 22 ml of L-15 medium. The isolated cells were pelleted by centrifugation at 1200g for 5 minutes. The pellets were resuspended in 0.5ml of L-15 medium, adjusted to 1ml when needed, as judged by the opacity of the suspension. The cell suspensions were then rapidly processed by FACS.

2.3 Isolation of GFP+ single cells using fluorescence activated cell sorting (FACS)

In the next step, we passed the cell suspension through a fluorescent activated cell sorter (FACS Aria, Becton Dickinson) to separate GFP positive (GFP+) and GFP negative (GFP−) fractions, which were collected. The sorting was performed at room temperature with the laser (Coherent Innova 70) set at 488nm wavelength and 200mW power. To minimize RNA degradation and loss of material the sorted cells were directly collected into Trizol (Invitrogen) and when necessary, stored at −80°C. A typical sort of ~500 tails would generate between 3000 and 5000 GFP+ cells, which when yielding less than 500ng of total RNA would get frozen away and 2–3 samples would get pooled for the next step of amplification. The whole larvae, which were carrying many more neuromasts with each of them containing more cells (as they were more mature), would typically yield between 17500 and 30000 GFP+ cells/sort. The settings for the sorting were carefully determined empirically and exactly reproduced in each subsequent experiment. An illustration of those settings is shown in (Figure 2A and B). The left plot (showing the P1 gate) was used to sort cells according to cell size (forward scatter, FSC-A) vs. granularity (side scatter, SSC-A) to exclude cellular debris as well as clumps of cells that may give erroneous fluorescent readings. Only cells that fell within the defined gate in the light-scatter plot were subsequently analyzed for fluorescent expression. The right plot discriminated cells according to the GFP fluorescence intensity of cells (GFP FITC-A) vs. Phycoerythrin (PE-A). Gates were demarcated to sort GFP− and GFP+ cells. The speed of the sorting (usually no more than 3–4) was adjusted as to obtain a purity that was higher than 95%. A typical sorted of a 0.5ml sample would be performed in 20–30 minutes maximum, therefore limiting the time during which transcriptional changes in the isolated cells could occur. As we were only sorting cells according to the absence/presence of a fluorescent signal and not according to their viability, we decided to perform tests in parallel to assess the quality of the sort and the health of the sorted cells. First, we sorted cells into PBS for observation under the microscope. Enrichment in GFP expressing cells and purity (whole cells vs. debris) was visually assessed in both fractions. A cells counter (Invitrogen, Countess® Automated Cell Counter) was used to distinguish dead from live cells. The number of live cells was similar to the number of events counted by the FACS, confirming that we were able to sort viable cells as opposed to debris and/or cellular fragments.

Figure 2. Illustration of the settings used on the FACS (BD-Aria).

A. Forward (FSC-A) and side scatter (SSC-A) plot is used to separate cells according to size and granularity respectively. The P1 gate is set as to exclude cellular debris (typically found in the lower left onside corner) and clumps of cells, imperfectly isolated (typically found in the upper right onside corner). B. Fluorescence scatter is used to separate cells according to the GFP fluorescence intensity of cells (GFP FITC-A) vs. Phycoerythrin (PE-A). The gates for GFP positive (GFP Pos) and GFP negative (GFPNeg) cell population have been chosen with a maximum of stringency as to avoid cross contamination of the respective cell population.

2.4 RNA extraction

Total RNA was isolated from GFP+ cells and control cells by extraction with Trizol Reagent, according to the manufacturer’s instructions. RNA pellets were resuspended in nuclease-free water (Ambion). RNA concentration and purity was read using a nanodrop. A minimum of 500ng of total RNA was necessary and only samples with 260/280 < 1.85 were processed further. Once the protocol was optimized we consistently had less than 5% of excluded samples. The integrity of the RNA was additionally confirmed by visualization of the ribosomal RNA on picochips for the Bioanalyzer (Agilent). Samples were loaded in parallel with a control clearly displaying the 40s and 45s bands expected of the ribosomal RNA. Only samples showing two strong distinctive bands were processed further. Using our optimized protocol we had on average more than 80% of samples that were of high quality. Interestingly, we did not observe major difference in yield, purity of integrity in the preparation form the different transgenic lines and preparation stages. A representative example of a high quality sample that was used in subsequent steps is shown in Figure 3 a and B. Approximately 500 ng to 800 ng RNA was linearly amplified by using the Amino Allyl MessageAmp II aRNA Amplification kit (Ambion) with yields ranging from 12 to 30 µg of aRNA. aRNA samples were split and labeled, half with Cy3 mono NHS ester and half with Cy5 mono NHS ester (CyDyes from GE Healthcare; post-labeling reagents from the MessageAmp II kit).

Figure 3. Assessment of the quality of the extracted total RNAs on picogels ran on a Bioanalyzer and their corresponding profiles of distribution.

A. Lane (L) is the ladder. Lane 11 is the control sample containing purified ribosomal RNA exhibiting the expected bands of the small and large ribosomal subunits. Lanes 4, 5, 6, and 7 are examples of high quality samples which were processed for amplification of cDNAs. Lanes 1, 2, 3, 8, 9 and 10 are examples of samples of bad or insufficient quality that were discarded. B. Quantification (in fluorescent units, FU on the y axis) of the different size of RNA (in [s] on the x axis) present in the samples. Only samples 4, 5, 6, and 7 show the two picks similar to the control around 40 and 45s.

2.5 Microarray labeling and hybridization

For isolating lateral line-specific transcripts, we reasoned that we could subtract the RNA obtained from the sorted GFP− cells from the RNA obtained from GFP+ cells by carrying out a co-hybridization experiment using microarrays. The sorted GFP− sample contains unlabeled cell types belonging to many different tissues of the embryos/larvae and should thus allow us to remove most housekeeping transcripts from the lateral line transcriptome. Genes specifically expressed in the GFP+ population would appear as up-regulated and would represent the enriched gene transcripts characteristic of this cell population.

We used "in house" 34 K zebrafish oligo microarray chips, which was derived from three sources: Compugen (16,512 × 60 mers), MWG (14,240 × 50 mers) and Operon (3,479 × 70 mers). Microarray slides were printed with 34,000 oligonucleotides that correspond to 19,000 zebrafish-expressed sequences. The whole set contains replicates of several positive (known housekeeping genes) and negative control oligos (random sequences) to assess the homogeneity and specificity of the hybridization.

At least 3 biological and 3 technical replicates for each sample of RNA extracted from GFP+ cells were co-hybridized with RNA from GFP− cells (Detailed numbers are available in [2] and in [5]). Hybridizations were performed overnight at 45°C in Maui Mixer FL hybridization chambers (BioMicro Systems). Microarray slide post-hybridization processing and scanning were done as previously described [20]. Data points with average quality values below 1.0 were eliminated and the datasets were normalized by Lowess (R-Bioconductor).

2.6 Data processing

Data analysis to identify differentially expressed genes was carried out using three different software packages. First, we did a pair-wise comparison with the normalized data using FileMaker Pro 9 (FileMaker, Inc.). Second we did the same with GeneSifter http://www.genesifter.net/. Then, normalized data were log-transformed and GFP+ and GFP− values were separately averaged over the different experiments. Fold differences were calculated from log averages and Student's t-test with Benjamini and Hochberg correction to generate p-values that were used to determined statistical significance. Third, we use the online NCI mAdb microarray data analysis tools http://madb.nci.nih.gov/. A simple t-test analysis was performed with a p-value of < 0.001 and a mean fold change of 2 as cut-off.

2.7 Validation of the microarray data

In order to estimate the degree of success of our strategy for identifying specifically expressed sequences, all of our data sets were validated in three ways: (1) database survey, (2) whole mount in situ hybridization (WISH) and (3) quantitative RT-PCR (qRT-PCR) [2] and [5]. First, we collected information from publicly available sources (ZFIN and PubMed) to identify all genes previously known to be expressed in the lateral line system, primordium and/or neuromasts and lateral line accessory cells. In this database survey, we found 372 genes in which specific expression in the PLL primordium or neuromasts had been convincingly described. Two hundred and sixty-seven genes where actually represented on our chips; and out of those, 148 genes where overrepresented in the two data sets (36 hpf and 48 hpf). This means that we were able to detect more than half of the previously described genes and had a 68% success rate if the genes were represented on the chip. An additional limitation of our approach was that any gene, which was also expressed in other tissues or organs, would be masked in our arrays, as we were subtracting the LL specific expression against the general expression of all other cell type of whole embryo/larva. When restricting the data search to accessory cells, which are still a very poorly documented cell population, only 10 genes were listed and among those most were also expressed in other tissues than the LL. Nevertheless, most of the previously known markers of accessory cells of the LL behaved as expected in our data set and therefore, we concluded that our approach was valid. Second, we randomly selected different cDNAs from sequences showing enrichment in the GFP+ cells and we performed a qualitative WISH. Third, we used a quantitative qRT-PCR approach to confirm the presence of these transcripts in the purified lateral line cells. This 3-step validation of our combined FACS-genes expression microarray approach confirmed and strengthened the reliability of the results obtained with different embryonic/larval stages and with both stable transgenic zebrafish lines. All data related to results described above are available in [2] and [5].

3. Results

3.1 Tg(cldnb:gfp) transgenic line

In the 36 hpf set we identified 4,449 sequences enriched in the primordium, that represent 3,505 known genes and 944 ESTs that do not map uniquely to a single gene. In turn, in the 48hpf set we identified 3,425 genes and 575 ESTs enriched in the neuromasts. Using ontology software (GeneGo) we could show enrichment in specific biological roles that include, cell motility, adhesion, cell communication and cellular developmental processes [2]. These results were in accordance with known biological processes occurring during migration of the lateral line primordium. More importantly, this data is providing new molecular players in LL development and is allowing studies of further parallels with diverse biological functions in normal and pathological situations.

3.2 Tg(tnksbp1:EGFP) transgenic line

We found more than 1,180 genes which were enriched by at least two fold in the GFP + fraction with a p < 0.0001. The 150 top ranked genes, which were enriched by at least four-fold are all excellent potential cell specific markers for accessory cells of the LL [5]. This heterogeneous cell population, which is loosely termed accessory or support(ing) cells, is still poorly characterized, but it is well known that it comprises the progenitor/stem cells indispensable for regeneration of hair cells [7, 21–23]. To be able to trace and identify those accessory cells will be critical to advance our understanding of hair cell regeneration.

4. Concluding remarks

We describe here a robust method for isolating single cells suspension from transgenic zebrafish embryos and/or whole larvae. The transgenic lines that we used in those studies [2] and [5], were expressing GFP in a subpopulation of cells of the sensory organ called the lateral line, which is related to the human inner ear sensory epithelia. More generally, we predict that this method will be adaptable to most transgenic lines expressing a fluorescent reporter gene. Putative adaptations of the method will need to take into account the developmental stage of the animals, the nature and localization of the tissue/organ of interest and the number and nature of the cells expressing the reporter genes. Finally we have used a combination of FACS and gene expression micro-array approach [2] and [5], but the FACS isolates cells could be used in combination with different downstream applications like cell culturing for example. This method is meant to be a starting place for exploring other possible combinatory approaches.