Gene discovery in genetically labeled single dopaminergic neurons of the retina

Abstract

In the retina, dopamine plays a central role in neural adaptation to light. Progress in the study of dopaminergic amacrine (DA) cells has been limited because they are very few (450 in each mouse retina, 0.005% of retinal neurons). Here, we applied transgenic technology, single-cell global mRNA amplification, and cDNA microarray screening to identify transcripts present in DA cells. To profile gene expression in single neurons, we developed a method (SMART7) that combines a PCR-based initital step (switching mechanism at the 5′ end of the RNA transcript or SMART) with T7 RNA polymerase amplification. Single-cell targets were synthesized from genetically labeled DA cells to screen the RIKEN 19k mouse cDNA microarrays. Seven hundred ninety-five transcripts were identified in DA cells at a high level of confidence, and expression of the most interesting genes was confirmed by immunocytochemistry. Twenty-one previously undescribed proteins were found in DA cells, including a chloride channel, receptors and other membrane glycoproteins, kinases, transcription factors, and secreted neuroactive molecules. Thirty-eight percent of transcripts were ESTs or coding for hypothetical proteins, suggesting that a large portion of the DA cell proteome is still uncharacterized. Because cryptochrome-1 mRNA was found in DA cells, immunocytochemistry was extended to other components of the circadian clock machinery. This analysis showed that DA cells contain the most common clock-related proteins.

Theories of brain function as well as interpretations of physiological experiments are still hindered by incomplete knowledge of the microcircuitry of the nervous system. Insufficient data on the identity of the various cell types and precise composition of neuronal populations represent a major limitation (1, 2). A striking example of the challenges imposed by the complexity of neural networks is the mechanism underlying the ability of the visual system to adjust its performance to the ambient level of illumination. A central role in neural adaptation to light is played by dopamine. Its synthesis and release are regulated by light, and agonists and antagonists at dopamine receptors mimic retinal responses to background illumination (3). Dopamine is synthesized and released by the dopaminergic amacrine (DA) cells (3). Despite their physiological importance, progress in the study of these cells has been difficult because they are very rare: only 450 DA cells are present in the mouse retina, representing ≈0.005% of the total number of retinal cells (4, 5). cDNA microarray technology allows the systematic analysis of gene expression in cells and tissues (6, 7). Because gene expression profiles determine cellular phenotypes, different types of neurons could be defined by the portion of the transcriptome they express. In turn, this knowledge could be used to design appropriate physiological experiments to clarify their function in the computations carried out by the networks of which they are components. However, the heterogeneity of the neuronal populations of the brain has significantly impaired gene cloning and expression profiling in specific neuronal types (8, 9). Crucial neuronal transcripts expressed in a limited number of cells are too diluted in the tissue samples to generate adequate targets for a sensitive cDNA microarray experiment or to permit cDNA identification, limiting the representation of the brain transcriptome in public databases. To confront this complexity, analysis of gene expression must be carried out either on homogeneous cell populations or single cells. Only rarely, however, can different types of neurons be recognized from size or shape: the vast majority escape identification. As a strategy, the promoter of the gene for tyrosine hydroxylase (TH), a cell-type-specific molecule, was used to drive the expression of a visible marker in retinal DA cells (4).

Here, we applied transgenic technology, single-cell mRNA amplification, and cDNA microarray screening to identify transcripts that are components of the DA cell expression profile. To amplify mRNAs from individual neurons, we combined a PCR-based technique [switching mechanism at the 5′ end of the RNA transcript or SMART (10)] with the T7 RNA polymerase amplification (6, 11). With this method (SMART7), single-cell targets were synthesized from genetically labeled, adult DA cells and used to screen RIKEN 19k mouse cDNA microarrays (12). Seven hundred ninety-five transcripts were identified in DA cells at a high level of confidence. Then, we confirmed the presence of the most interesting proteins by immunocytochemistry.

Materials and Methods

Harvesting of DA Cells. A transgenic line of mice was used in which the catecholaminergic neurons of the retina express human placental alkaline phosphatase (PLAP) on the outer surface of the cell membrane under the control of the TH gene promoter (4). As described (4), the retina was dissociated, and DA cells were labeled with a monoclonal antibody to PLAP directly conjugated to Cy3 (E6-Cy3). Thirty minutes after plating, cells were washed several times and kept for the rest of the experiment in extracellular solution (containing 137 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl₂, 1 mM MgCl₂, 5 mM Hepes, and 20 mM glucose). Immediately after identification of a DA cell, the UV beam was turned off, and the rest of the procedure was carried out in visible light with Nomarski optics. When the patch pipette, filled with intracellular solution (containing 130 mM KCl and 0.5 mM EGTA in 10 mM Hepes, pH 7.4), was immersed into the bath, positive pressure was continuously applied to avoid entry of cellular debris. After gigaseal formation and disruption of the patch membrane, the cellular contents were aspirated into the pipette by gently applying negative pressure until the residual cell ghost became stuck to the pipette tip. The cell was then positioned in front of a U tube and washed for 1 min with fresh extracellular solution under constant visual control. The electrode was then lifted from the bath and its tip broken into an Eppendorf tube that contained 2 μl of RNase inhibitor solution [0.4 μl of 5× first-strand buffer (containing 375 mM KCl, 30 mM MgCl₂, and 250 mM Tris·HCl, pH 8.3), 0.2 μl of 20 mM DTT, 2.5 units of SUPERase (Ambion, Austin, TX), and H₂O to volume]. After brief centrifugation, the tube was kept on ice until the end of the harvesting procedure. Aliquots of the medium and unlabeled postreceptor neurons were processed simultaneously with DA cells.

Amplified RNA (aRNA) Target Preparation. First-strand cDNA synthesis. The following reagents were added to each tube: 0.5 μl of 10 μM SMART II oligonucleotide (Clontech), 0.5 μl of 2 μM SMART7T₂₄ primer [TGAAGCAGTGGTAACAACGCAGACTAATACGACTCACTATAGGGAGAAGC(T)₂₄VN], 0.6 μl of 5× first-strand buffer, 0.3 μl of 20 mM DTT, 0.5 μl of 10 mM dNTPs, and 0.5 μl of 5% Nonidet P-40. Tubes were incubated at 65°C for 2 min, kept for 2 min in ice, and transferred to 55°C in a preheated PCR machine. After 3 min, the temperature was allowed to slowly (0.15°C s^-1) drop to 37°C, and tubes were kept at this temperature for 2 min. They were then transferred to 42°C and, after 5 min, 0.3 μl of 200 units/μl SuperScript II (GIBCO) was added. The reaction was carried out for 1 h.

Double-stranded cDNA synthesis. Ninety-five microliters of PCR Advantage 2 mix was prepared as follows: 10 μl of 10× PCR Advantage buffer (Clontech), 2 μl of 10 mM dNTPs, 4 μl of 10 μM PCR primer (AAGCAGTGGTAACAACGCAGAGT), 2 μl of Polymerase Mix Advantage 2 (Clontech), H₂O to volume. This mix was added to the first-strand cDNA synthesis reaction product, and thermal cycling was carried out in the following conditions: 95°C for 1 min, followed by 19 cycles, each consisting of denaturation at 95°C for 5 s, annealing at 65°C for 5 s, and extension at 68°C for 6 min. One-fiftieth of the RT-PCR products was tested in a 30-cycle PCR for the presence of gapdh and ribosomal protein s17 (rps17) cDNAs by using intron-spanning primers.

cDNA purification. Samples were blunt-ended with 3 units of T4 DNA polymerase at 37°C for 15 min, and the single-stranded molecules were removed by treatment with 2.5 units of mung bean nuclease (Promega) at 37°C for 10 min after addition of 11 μl of 10× mung bean buffer (containing 500 mM Na acetate, 300 mM NaCl, and 10 mM ZnSO₄, pH 5.0). cDNAs were purified by proteinase K treatment at 55°C for 15 min [2 μg of proteinase K (Qiagen, Valencia, CA) in 200 μl of 0.2% SDS, 10 mM EDTA, and 50 mM Tris·HCl, pH 8], followed by phenol/Tris and phenol/chloroform extraction. The upper phase was transferred into an Amicon 30 column (Millipore), spun at 14,000 × g, reduced to a volume of ≈5 μl, rediluted with 500 μl of H₂O, and spun again. After three repetitions, cDNA was recovered in a final volume of ≈5 μl.

aRNA synthesis. cDNA was template for a 20-μl in vitro transcription reaction using the T7 Megascript kit (Ambion) according to the manufacturer's recommendations. Reactions were carried out at 37°C for 5 h. After addition of 180 μl of H₂O, aRNA was purified by phenol/H₂O and phenol/H₂O/chloroform steps. Upper phases were loaded on an Amicon 30 column and underwent three-step purification.

Double-stranded cDNA synthesis and aRNA transcription. aRNA was template for a 20-μl first-strand cDNA synthesis reaction. After addition of 100 ng of pd(N)₆ random hexamer (Amersham Pharmacia Biosciences), samples were denatured for 3 min at 70°C and chilled in ice. The following reagents were then added: 2 μl of 10 mM dNTPs, 4 μl of 5× first-strand cDNA buffer (containing 375 mM KCl, 15 mM MgCl₂, and 250 mM Tris·HCl, pH 8.3), 15 units of RNAGUARD (Amersham Pharmacia), and 2 μl of 0.1 M DTT. After a 3-min incubation at room temperature and 2-min incubation at 42°C, 300 units of SuperScript II was added, and samples were incubated for 60 min. Second-strand synthesis occurred in a final volume of 100 μl after addition of the following reagents: 10 μl of 10× PCR Advantage buffer, 2 μl of 10 mM dNTPs, 0.6 μl of 1 μM primer T7T₂₄ [TGTAATACGACTCACTATAGGGAGAAGC(T)₂₄VN], 0.5 units of RNase H (GIBCO), and 2 μl of Polymerase Mix Advantage 2. Reactions were carried out as follows: 37°C 10 min, denaturation at 95°C 1 min, 2 cycles at 95°C 5 s, slow cooling (0.3°C s^-1) to 42°C, annealing for 2 min, and extension at 68°C for 30 min. Double-stranded cDNA was then blunted, purified as described above, and used as a template for a second in vitro transcription step.

cDNA Microarray Hybridization. Array preparation. Probe cDNAs were obtained from the RIKEN collection of mouse cDNA libraries, constructed by using the CAP trapper method to enrich for full-length inserts (12). cDNAs were amplified by PCR by using M13 forward and reverse primers. The PCR products were precipitated with isopropyl alcohol and resuspended in 15 ml of 3× SSC. Microarrays were printed on poly-l-lysine-coated glass slides by using a DNA arrayer with 16 tips (SMP3, TeleChem, Sunnyvale, CA). The diameter of the spots was 100-150 μm. Arabidopsis cDNAs and bacterial genomic and plasmid DNAs were used as negative controls.

Target preparation and hybridization. One-third of the aRNA synthesized from each DA cell was labeled by incorporating Cy3 during random-primed reverse transcription. The labeling was carried out at 42°C for 1 h in 30 μl containing 6 μg of pd(N)₆ random hexamer, 6 μl of 5× first-strand buffer, 0.1 mM Cy3-dUTP, 0.5 mM dATP, dCTP, and dGTP, 0.2 mM dTTP, 10 mM DTT, and 400 units SuperScript II. To remove unincorporated nucleotides, labeled cDNA was mixed with 500 μl of binding buffer [containing 5 M guanidine-SCN, 10 mM Tris·HCl pH 7, 0.1 mM EDTA, 0.03% gelatin, and 2 ng/μl tRNA (Sigma)], and 50 μl of silica matrix buffer [containing 10% matrix, 3.5 M guanidine·HCl, 20% glycerol, 0.1 mM EDTA, and 200 mM Na acetate, pH 5 (Amersham Pharmacia)], transferred to a GFX column (Amersham Pharmacia), and centrifuged at 15,000 rpm in a Sorvall centrifuge (RC-3B plus, H6000A/HBB6 rotor) for 3 min. The flow-through was discarded, and the column was washed with 500 μl of washing buffer (Amersham Pharmacia). Adsorbed targets were eluted with 17 μl of H₂O. Labeled targets were mixed with blocking solution containing 3 μl of 10 mg/ml oligo(dA), 3 μl of 20 mg/ml yeast tRNA (Sigma), 1 μl of 20 mg/ml mouse Cot1 DNA (Sigma), 5.1 μl of 20× SSC, and 0.9 μl of 10% SDS. RIKEN 19k mouse cDNA microarrays consisted of three multiblocks, and each required 10 μl of hybridization solution. Before hybridization, targets were heated at 95°C for 1 min and cooled to room temperature. Coverslips were hybridized overnight at 65°C, washed briefly in 2× SSC, 0.1% SDS to remove the coverslips, transferred to 1× SSC for 2 min, and rinsed in 0.1× SSC for 2 min. After washing, slides were spun at 800 rpm in the Sorvall centrifuge and scanned on a ScanArray 5000 confocal laser scanner (Packard). Images were analyzed by scanalyze (13).

Analysis of the data. Three DA cells were analyzed, and two aRNA aliquots from each of them were labeled and hybridized in separate experiments. First, results from corrupted spots were deleted with “flags” added manually (filtering). Then data from the two hybridizations were normalized for each cell according to the mean of the Cy3 intensity calculated on each multiblock of 7,056 spots. The average intensity of each spot was then calculated for each cell. To identify candidate genes, the expression threshold was set by adding five times the standard deviation to the mean of the intensities of the control spots.

Immunocytochemistry. Retinas were fixed for 2 h in 2% formaldehyde in 0.15 M Sörensen phosphate buffer, cryoprotected in sucrose solution, frozen, and cut in a cryostat. Radial sections were double-immunolabeled with an antibody to TH (either a mouse monoclonal from ImmunoStar, Hudson, WI, 1:1,000, or a rabbit polyclonal from Chemicon, 1:500) and one of the following antibodies: rabbit polyclonals to COP9/signalosome subunit (Csn)5, 1:500; Janus kinase 1 (Jak1), 1:500; Tyk2, 1:500; insulin, 1:500; CREB-2, 1:1,000; E47, 1:1,000; monocyte chemoattractant protein 1 (MCP-1), 1:100; CD59, 1:200; CD81, 1:200 (Santa Cruz Biotechnology); IFN-α receptor chain 2 1:500 (PBL Biomedical Laboratories, New Brunswick, NJ); IFN-α receptor, 1:1,000 (BioSource International, Camarillo, CA); ClC2, 1:1,000; cocaine- and amphetamine-regulated transcript (CART), 1:500; Cry1, 1:1,000; Cry2, 1:1,000; Clock, 1:500; Per1, 1:100 (Alpha Diagnostic International, San Antonio, TX); ferritin, 1:1,000 (DAKO); PKA, 1:1,000 (Transduction Laboratories, Lexington, KY); TCRβ (Santa Cruz Biotechnology); Csn2, 1:1,000 (Biotrend, Cologne, Germany); Csn1, 1:500; Csn3, 1:500; Csn7, 1:500; and Csn8, 1:500 (Affiniti Research Products, Mamhead, U.K.); mouse monoclonals to growth-associated protein (GAP)-43, 1:2,000 (Zymed); CP2, 1:1,000 (Transduction Laboratories); and goat polyclonals to 14-3-3, 1:100 (Santa Cruz Biotechnology). Secondary antibodies were: Oregon green goat anti-rabbit, 1:4,000; Alexa Fluor 568 goat anti-mouse, 1:1,000; Alexa Fluor 568 goat anti-rabbit, 1:200 (Molecular Probes); FITC goat anti-mouse, 1:200 (Southern Biotechnology Associates); Texas red donkey anti-mouse, 1:100; and FITC donkey anti-goat, 1:400 (Jackson ImmunoResearch). Fluorescence was detected by using a Bio-Rad MRC-1024 confocal imaging system equipped with an argon-krypton laser and a Zeiss Axiophot microscope.

Results

aRNA Preparation. After enzymatic digestion and mechanical trituration of the retina, DA cells were identified in the living state by labeling of their membrane with the fluorescent monoclonal antibody to human placental alkaline phosphatase E6-Cy3 (Fig. 1 a and b). We have previously shown by single-cell nested RT-PCR that large neurons stained by E6-Cy3 were DA cells because they expressed TH mRNA, whereas unlabeled neurons were negative (14).

Fig. 1.

Harvesting of DA cells. (a) After enzymatic digestion and mechanical trituration of the retina, a short-term culture of solitary retinal neurons was established. (b) DA cells were identified by staining with the E6-Cy3 antibody. (c) Gigaseal was established on the membrane of a DA cell. (d) The neuron was finally lifted from the glass bottom of the recording chamber, positioned in front of a perfusion tube, and washed with extracellular solution. (Bar = 100 μm.)

DA cells were patch clamped (Fig. 1c) and individually harvested between 30 and 90 min from the moment the retina was dissociated. This time interval was chosen to minimize changes in gene expression induced by cell damage during trituration, loss of synaptic input, or exposure to culture medium. After gigaseal formation and disruption of the patch membrane, cellular contents were aspirated into the pipette by gently applying negative pressure until the residual ghost became stuck to the pipette tip. The cell was then lifted from the bottom of the recording chamber (Fig. 1d), positioned in front of a perfusion tube, and washed for 1 min with fresh medium. The tip of the pipette was finally broken into an Eppendorf tube containing RNase inhibitor solution.

Single-cell mRNAs and samples of the medium were then amplified by a combination of SMART and T7 amplification techniques (Fig. 2). After brief denaturation in the presence of Nonidet P-40, first-strand cDNA synthesis was primed with SMART7T₂₄ primer in the presence of SMART II, a template switching oligonucleotide. cDNA was then amplified for 19 cycles in a long-range PCR. PCR products were discarded when aliquots of the medium or unlabeled postreceptor neurons contained rhodopsin cDNA, an evidence of contamination. An aliquot (1/50th) of the PCR products was tested for the presence of the high-abundance transcripts gapdh and rps17. Targets from 8 of 16 DA cells contained cDNAs for gapdh and rps17, proving that the cells had been harvested and their mRNAs amplified.

Fig. 2.

Flow chart of the SMART7 technique.

After purification, single-cell PCR products were amplified by using T7 RNA polymerase. aRNA was then used as a template for double-stranded cDNA synthesis and a second T7 amplification. Final aRNA preparations were tested for the presence of DA cell markers: four were positive for TH, plasmalemmal dopamine transporter (DAT), and the γ2 subunit of the GABA_A receptor.

cDNA Microarray Screening and Data Analysis. Single-cell targets were synthesized from aliquots of aRNA by incorporating Cy3-dUTP during random-primed reverse transcription. Two separate targets were obtained for each of three DA neurons and hybridized to the cDNA probes in high-stringency conditions. After filtering and normalization (see Materials and Methods), the mean fluorescence intensity of each spot was calculated from the two values obtained for each cell. The RIKEN 19k mouse cDNA microarrays contained 18,816 test cDNAs, representing 13,600 nonredundant genes (12). Because the libraries were synthesized from a variety of mouse organs and the cDNAs were arrayed during the initial phase of the RIKEN project, clones mostly represented “housekeeping genes,” that is, genes that are highly expressed in most tissues.

A large proportion of an additional 2,352 spots were control DNAs (plant, bacterial genomic, and plasmid). These were used to measure background hybridization values and set a threshold to select transcripts that were expressed in DA cells at a high level of confidence. We discuss here only spots whose fluorescence was more intense than a threshold equal to five standard deviations above the mean of the intensities of the background signals.

For each DA cell, 2,202, 1,671, and 1,612 spots (11.7%, 8.9%, and 8.5% of the test spots, respectively) exceeded the threshold value. Five hundred twenty-nine spots were positive in all three cells: because of clone redundancy in the arrays, they corresponded to 271 genes. Seven hundred ninety-five genes, present in a total of 1,330 spots, surpassed the threshold value in at least two of the three cells. Fig. 3 illustrates the distribution of 795 transcripts according to their subcellular localization, as specified by the terms of the Gene Ontology (GO) Consortium (www.geneontology.org); 38% of them were ESTs or transcripts encoding hypothetical proteins.

Fig. 3.

Distribution of the candidate transcripts of DA cells according to their subcellular localization. The category “others” includes all remaining cellular components listed in the GO database.

Immunocytochemistry. We used immunocytochemistry to confirm the expression of 24 of the 795 genes whose transcripts were identified in DA cells at a high level of confidence. Proteins were chosen according to the following criteria: (i) they were not coded for by housekeeping genes, such as ribosomal or mitochondrial proteins; (ii) their transcripts were not identified by ESTs only; (iii) their function was known; (iv) antibodies were available. Of the 24 proteins tested, 23 were present in DA cells, and these are listed in Table 1 according to their subcellular localization. Twenty-one have never been described before in this cell type: they include a chloride channel, receptors, and other membrane glycoproteins, kinases, transcription factors, cytosolic proteins such as ferritin, and secreted molecules (Fig. 4).

Table 1. Gene expression in DA cells.

Membrane	Cytoplasm	Nucleus	Secreted
Chloride channel-2 (ClC2)*	Csn5*	Csn2*	CART*
Complement regulatory glycoprotein CD59*	Ferritin*	Csn7*	Insulin*
IFN receptor-2 (IFNR2)*	GAP-43	Transcription regulators:	IFNα*
T-cell receptor β-chain (TCRβ)*	Kinases:	ATF4/CREB-2*	MCP-I*
Tetraspanin glycoprotein CD81*	Ca2+/calmodulin-dependent protein	CP2*
	Kinase II (CaMKII)*	Cry1*
	JAK1*	E47*
	JAK TYK2*
	PKA
	Molecular chaperone 14-3-3*

^*Genes not previously known to be expressed in DA cells.

Fig. 4.

Retinal sections were double-immunolabeled with anti-TH (red) and antibodies (green) to proteins whose presence in DA cells was either deduced from or suggested by the results of the microarray screening. Note that GAP-43 appears confined to the dendrites of DA cells. Whereas Clock is expressed in most retinal neurons, Cry2 is highly expressed in DA cells and a subset of ganglion cells. Among the signalosome proteins, Csn2 is ubiquitous, whereas Csn1 and Csn7 are expressed in subsets of retinal neurons, including DA cells. (Bar = 15 μm.)

Immunocytochemistry also confirmed the expression of proteins that are part of functional pathways or large multimolecular complexes whose existence in DA cells was not known before. Because of the intricate relations between light adaptation and retinal circadian rhythm, we were intrigued that DA cells expressed the cryptochrome-1 (Cry1) gene. Cry1, together with Cry2 and period proteins, acts as a repressor of the Clock-Bmal-activated transcription and thus fulfills a central function in the molecular clock (15). We therefore extended our analysis to other components of the clock machinery that were not represented in the arrays and thus proved that DA cells also expressed Cry2, Clock, Bmal, and Period 1 (Per1) (Fig. 4 and Table 2). Whereas Clock was contained in most retinal neurons (16), high-level Cry2 expression was restricted to DA cells and to a subset of ganglion cells.

Table 2. Components of the molecular clock and COP9 signalosome in DA cells.

Clock	mRNA	Protein	COP9	mRNA	Protein
Cry1	+	+	Csn1	NC	+
Cry2	NC	+	Csn2	+	+
Clock	NC	+	Csn3	NC	+
Bmal	NC	+	Csn4	NC	NA
Per1	NC	+	Csn5	+	+
			Csn6	NC	NA
			Csn7	+	+
			Csn8	NC	+
			COP1	+	NA

+/−, expression/nonexpression in DA cells; NC, cDNA not present in microarray; NA, no antibody available.

Finally, the microarray screening identified three of the eight subunits that comprise the COP9/signalosome, a 500-kDa nuclear protein complex responsible for protein degradation and light-regulated transcriptional events in plants (17). We therefore investigated by immunocytochemistry the presence of all COP9 components in DA cells. Table 2 shows that we detected the expression of six subunits, with two of them, Csn1 and Csn7, restricted to subsets of retinal neurons that included DA cells (Fig. 4).

Discussion

Technical Considerations. Although fluorescence-based cell sorting (18), antibody-based panning (19), and laser capture micro-dissection have been used to isolate cells from a variety of tissues (20, 21), patch clamp was our technique of choice because it minimized manipulations of the cells, reduced the risk of contamination, and allowed selection of healthy, physiologically active neurons (22). Furthermore, the patched neurons were positioned in front of a perfusion tube and washed, while constantly monitoring the quality of the seal. Cells were then stored in ice until harvesting was complete. Because we were unable to obtain PCR products from cells stored at -80°C, cell lysis, first-strand cDNA synthesis, and PCR were carried out immediately.

To amplify mRNA from single DA cells, we combined SMART (10) and T7 RNA polymerase (6, 11) amplification techniques. In SMART, the reverse transcriptase terminal transferase activity adds a stretch of nucleotides to the 3′ end of the cDNA during first-strand cDNA synthesis and a template-switching oligonucleotide anneals to this stretch, creating an extended template. The reverse transcriptase then switches template and continues replication to the end of the oligonucleotide. The cDNA is then amplified by PCR. SMART appeared to be a sensitive and reproducible method when used from a starting material as low as 1 μg of total RNA or single preimplantation embryos (23, 24). With the RNA polymerase-based method, two successive rounds of amplification were required to profile 1,000 neurons or 2 ng of total RNA as starting material (21, 25, 26). However, template-independent amplification products became prominent when smaller amounts of RNA were used, and these strongly competed with the amplification of specific transcripts (25).

In SMART7, first-strand cDNA was synthesized in the presence of the SMART oligonucleotide and PCR-amplified for a limited number of cycles. In turn, the amplification product was the template for two rounds of T7 aRNA synthesis. The combination of two different techniques of amplification kept the number of PCR cycles low (<20) and avoided strong competition from template-independent amplification. Furthermore, SMART-SMART PCR products, formed by unspecific internal priming of the SMART oligonucleotide during cDNA synthesis, were not amplified by T7. The SMART7 technique could easily be adapted to profile varying amounts of starting material. We successfully profiled 1 ng of total retina RNA after amplification of the SMART-anchored cDNA by 18 cycles of PCR and one round of T7 amplification. We were also able to profile in a reproducible manner 50 pg of total retina RNA by adding a second T7 round (data not shown).

At a single-cell level, the reproducibility of this technique needs further improvement. Several genes were detected only in two of three cells. These differences can be due to several factors: biases can be introduced during reverse transcription and PCR amplification, gene expression may vary in different cells after tissue dissociation and plating, and dendritic RNAs may be lost during harvesting.

Only a small fraction of the genes expressed in DA cells have been identified in this study. One reason is that we considered only hybridization signals above a threshold that was arbitrarily set very high. As a result, our analysis was probably confined to medium and highly expressed genes, thus missing mRNA species that are in the low-abundance category (<100 molecules). Second, RIKEN 19k microarrays contained a limited repertory of mouse cDNAs specific for the nervous system, because they consisted of clones mostly derived from housekeeping genes (12). For instance, the arrays did not contain the cDNAs of the seven subunit subtypes of the GABA_A receptor that are expressed in DA cells. Only the α6 subunit was represented, and this is absent in DA cells (14).

Biological Considerations. Biogenic amines and amino acid transmitters coexist with peptides in both neuroendocrine cells and neurons. Here, we show that DA cells express at least four additional neuromodulators: a neuropeptide (CART), a hormone (insulin), a cytokine (IFN-α), and a chemokine (MCP-1). Little is known about the function of these molecules in the central nervous system or retina. CART expression was increased by acute administration of cocaine and amphetamine in the rat striatum (27). Although no receptors have been identified so far, CART appeared to inhibit voltage-dependent intracellular Ca²⁺ signaling and attenuated cocaine enhancement of depolarization-induced Ca²⁺ influx in rat hippocampal neurons (28). Application of insulin to the in vitro bovine retina preparation decreased the amplitudes of the a- and b-waves of the electroretinogram (29). Insulin receptors are widely expressed in the retina (30), and administration of this hormone modified the permeability of voltage-gated K⁺ and Ca²⁺ channels through receptor-mediated tyrosine phosphorylation, thus modulating the activity of photoreceptors and Müller cells. The finding, however, that insulin is secreted by neurons within the retina is completely unexpected. Binding of IFNs to their cell surface receptors elicits complex responses that are associated with changes in the expression of hundreds of genes. As a neuromodulator, IFN-α enhanced the spontaneous activity of neurons in cerebral and cerebellar cat cortices and suppressed LTP in rat hippocampal slices (31) and neuronal activity in glucose-sensitive hypothalamic neurons (32). MCP-1 is a potent CC-chemokine that plays a primary role in the recruitment of inflammatory cells. Constitutive expression of its receptor CCR2 was recently observed in neurons and glia throughout the central nervous system, and its activation induced a transient increase in intracellular Ca²⁺ (33). Thus, DA cells potentially release several unexpected modulators. These may exhibit synergistic or antagonistic effects and convey fast, moderate, and slow signals to a spectrum of potential targets. It will be important to assess whether these modulators participate in light adaptation or mediate uncovered functions of DA cells.

Three types of secretory organelles were previously identified in the perikaryon of DA cells: 50-nm diameter clear or agranular vesicles, 80- to 125-nm dense core or granular vesicles, and 0.3-μm granules (34). The distribution of organelles containing the vesicular monoamine transporter-2 suggests that dopamine-containing vesicles are of the agranular type. In turn, CART and insulin are likely components of granular vesicles or granules. In other regions of the brain, peptide release requires bursting or high-frequency activity. Furthermore, insulin secretion is controlled in pancreatic β-cells by ATP-sensitive K⁺ channels and voltage-dependent Ca²⁺ channels (35). Differences in the molecular machinery of the secretory organelles may ensure differential release of the various secretory products. In this respect, the highly restricted expression of GAP-43 is particularly interesting, for this molecule participates in catecholamine release by PC-12 and chromaffin cells (36).

Gene expression in the retina is influenced by the circadian rhythm. The retina has an internal clock that seems to depend on the level of melatonin, which is high in the night and lower in the day. A large body of observations has pointed to the involvement of dopamine in circadian control: both light and dopamine were able to phase-shift the retinal melatonin rhythm, and ablation of DA cells in zebrafish showed that dopamine mediates the effects of the circadian clock on visual sensitivity (37). Furthermore, circadian variations in synthesis and release of dopamine were observed in pigeon, quail, and rats (38). Our microarray experiments identified Cry1 mRNA in DA cells, and immunocytochemistry showed that DA cells contain the most common circadian clock-related proteins. Interestingly, Cry2 is localized only in DA cells and a subset of ganglion cells, supporting the idea that these amacrine cells have a role in the circadian control of retinal function (39).

The effect of light on transcriptional regulation has been extensively studied in plants. The COP9/signalosome is a 500-kDa nuclear protein complex that acts as a transcriptional repressor in the cascade of events regulated by light during seedling development (17). It consists of eight subunits that show structural homology with the lid of the 26S proteasome and are involved in protein degradation. The presence of this macromolecular complex in the nervous system is previously undescribed; its functional significance remains to be elucidated.

Finally, it is significant that in our analysis, 38% percent of candidate transcripts were ESTs or encoding for hypothetical proteins, suggesting that the identity of a large proportion of the genes expressed in DA cells is still unknown.

Conclusion

Profiling of gene expression in single neurons will considerably enrich our understanding of neural networks. On the one hand, gene expression profiles of identified types of neurons will suggest novel electrophysiological experiments. On the other hand, microarray data on uncharacterized neurons will identify new cell-type-specific proteins, whose promoters can be exploited to express visible markers in additional neuronal populations. This type of research will be greatly facilitated by the recent completion of the sequence of the mouse genome and the description of a large portion of the mouse transcriptome. The mouse cDNA encyclopedia now comprises the annotation of 60,770 full length mouse cDNA sequences clustered into 37,000 “transcriptional units,” containing the large majority of molecules involved in brain function. Additional rare cDNAs, “singletons,” have been cloned mostly from restricted areas of the brain, increasing mouse gene estimates to 70,000 (40, 41). The fine analysis of the expression of these genes in individual cells may provide a molecular brain atlas in which each neuronal type may be defined by its molecular identity.