- Email: [email protected]
The problem of neuronal cell types: a physiological genomics approach
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
TINS special issue: The Neural Substrates of Cognition
The problem of neuronal cell types: a physiological genomics approach Sacha B. Nelson, Ken Sugino and Chris M. Hempel Department of Biology and National Center for Behavioral Genomics, Brandeis University, MS 008, 415 South Street, Waltham, MA 02454-9110, USA
Neural circuits within the vertebrate brain are composed of highly diverse cell types. The exact extent of this diversity is a matter of continuing debate. For example, do cortical interneurons comprise a few, dozens or O100 distinct cell types? Recently, several groups have used microarrays to measure genome-wide gene expression profiles for specific neuronal cell types. These methods can offer an objective basis for neuronal classification. In this review, we argue that this approach should now be carried out more broadly and that it should be coupled to large-scale efforts to generate mouse driver lines in which tools for genetic manipulation, such as the Cre recombinase, are expressed in identified cell types within the brain. This would enable neuroscientists to begin to investigate more systematically the roles of specific genes in establishing particular cellular phenotypes, and also the roles of particular cell types within brain circuits. This review is part of the TINS special issue on The Neural Substrates of Cognition. Introduction We share nearly all of our genes with mice and the great majority of our genes with much more distant relatives such as flies and worms. But the neural structures built during human development are much larger, more complex and, at least with respect to flies and worms, contain a very different complement of cell types and neural organs. The substrates of our unique cognitive abilities must lie in these diverse neural cell types, in the complex circuits into which they are organized, and in the patterns of activity that those circuits sustain. There are two experimental problems that, if solved, would greatly accelerate progress in the search for the cellular and circuit-level substrates of cognition within the mammalian brain. First is the problem of being able to identify easily, rapidly, and objectively the many cell types that comprise circuits within the neocortex and other brain regions. Second is the problem of being able to manipulate the function of those cell types selectively and flexibly. Recently, application of modern genetic and genomic tools to questions in cortical neurobiology has offered new opportunities to investigate these two problems. Here, we review recent advances that open the door to a systematic examination of the first problem: Corresponding author: Nelson, S.B. ([email protected]). Available online 22 May 2006
that of objectively identifying and classifying cortical cell types. We also discuss how these advances might in turn permit solution of problem number two: that of manipulating those cell types to understand their function better. The realization that cortical cell types are highly diverse was first made by early anatomists using Golgi stain to reveal axonal and dendritic morphology. Although neurons could be broadly classified as having local or longrange axonal projections, or as having smooth or spiny dendrites, the exact number of distinct cell types remains a matter of continual debate. Today there is broad acceptance that the two major classes of cortical neurons are pyramidal neurons, which make long-range and local connections, use glutamate as a neurotransmitter and have spiny dendrites, and inhibitory interneurons, which make primarily local connections, use GABA and various neuropeptides as neurotransmitters, and have smooth or sparsely spiny dendrites. But although it is also broadly accepted that within these classes neurons can be highly diverse, the proper way to parse this diversity into recognizable cell types is far from agreed upon. The question of how to classify GABAergic interneurons has received tremendous attention in the pages of this and other journals [1,2] but without consensus. The same is true for cortical pyramidal cells, despite clear demonstrations of the dramatic differences between pyramidal cell subtypes [3–5]. The story is somewhat different from that in the retina, and the analogy is perhaps helpful because it provides hope that the problem is not insoluble and suggests a scale for the total number of neural cell types. Using a combination of anatomical, physiological and gene expression properties, retinal anatomists and physiologists have identified 55–60 distinct neuronal cell types [6,7]. This number has inspired some to speculate that the total number of cell types within the mammalian cortex is w1000 [8,9]. Most retinal cell types can be recognized across vertebrate orders and considerable progress has been made in understanding their connectivity and function. Such success has been possible in the retina owing to its accessibility to direct study using physiological stimuli, and to its clearly defined laminar structure, cellular morphologies and cellular functions. The problem in the cerebral cortex arises primarily from the fact that
www.sciencedirect.com 0166-2236/$ - see front matter Q 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tins.2006.05.004
340
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
morphology, connectivity and circuit function of the constituent cell types are not as easily segregated. For this reason, definitive mapping between the various features of cellular phenotype has not been possible: a classification based purely on morphology does not adequately capture the diversity of intrinsic firing properties or of synaptic properties and, conversely, classifying cells purely on the basis of physiological properties does not yet fully predict their morphology or synaptic connectivity. It is perhaps valid to ask whether this process of identification and classification is really important [10]. We believe the answer is a resounding yes. The importance lies less in the biological insights implicit in the classification than in the enabling of multiple investigators to refer unambiguously to the same neurons. If we are to begin collectively to ‘circuitbreak’ the cortex and related structures, and to identify homologous cells and circuits across cortical regions and across mammalian species, we must first have a broadly accepted cellular ‘parts list’. Genome-wide expression profiling for neuronal classification The field of phylogenetic taxonomy was revolutionized by the use of genomic sequence comparison [11,12]. Similarly the classification of protein families, such as those of ion channels, has been greatly clarified by measures of DNA-sequence similarity [13]. Although different cell types within an individual organism are likely to be genetically identical, they differ in which genes they express. Therefore, it has been suggested that genome-wide gene expression profiling could provide a sound and unbiased method for neuronal classification [2,14]. This approach rests on the assumption that the overall function of a neuron within a circuit is highly correlated with its gene expression profile. DNA microarrays provide an ideal platform for such studies. To date, however, microarrays have not been effectively and systematically employed to address problems in neuronal taxonomy – the majority of neuronal microarray experiments have been carried out on tissue homogenates where distinctions between cell types are lost [15–19]. Neuronal classification, of course, requires analysis at the cellular level. Single-cell gene expression studies have been underway for over a decade [20,21] but the tiny amount of mRNA available from a single cell has limited such investigations to a tiny fraction of neuronally expressed genes. For replicable microarray-based genome-wide expression analysis, high yields of linearly amplified mRNA are required and it is not yet clear whether these can be reliably derived from single cells. Despite this technical difficulty, some investigators have been able to generate expression profiles from single cells. However, these profiles were of randomly selected neurons from mixed populations [22,23], in which too few neurons were compared to identify recognizable classes or distinguish technical from biological variability. Both homogenate and single-cell expression studies have been useful for suggesting functional gene classes involved in schizophrenia [15], identifying genes involved in maturation of olfactory neuron precursors [23], www.sciencedirect.com
identifying neuronal markers in the amygdala [19,24], and aiding in the classification of cortical neurons using small numbers of known markers [25,26]. But neither approach exploits the potential of expression profiling for the purposes of neuronal taxonomy. This requires generating genome-wide expression profiles from defined neuronal populations. How is this goal best achieved? The most straightforward strategy is as follows: (i) a functionally distinct cell type must be labeled with a marker; (ii) that population must somehow be purified from the rest of the tissue using that marker; and the mRNA must be (iii) extracted and (iv) amplified to yield a sample that can be probed using a microarray. A specific implementation [27] of such a strategy is outlined in Figure 1. However, each of these steps can be carried out using various approaches (Table 1). Optimizing and integrating all of these steps has required considerable technical and methodological innovation in many laboratories. These efforts, which we will review in the subsequent paragraphs, have finally over the last few years begun to bear fruit. (a)
Labeled mouse
(b) Slicing and protease digestion
(c)
(h)
(g)
Data analysis
Microarray hybridization and scanning
Microdissection
(f) Cell lysis, poly(A) RNA amplification and labeling B
B
B AAAAAA
(d)
Dissociation, trituration and plating
(e)
Manual sorting
TRENDS in Neurosciences
Figure 1. One method used to evaluate gene expression in genetically defined subpopulations of neurons. This is the strategy employed by Sugino et al. [27]. Once dissociated cells are obtained (a–d), this approach involves fluorescencebased manual sorting of live neurons (e) followed by expression analysis by microarray (f–h).
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
341
Table 1. Various methods for cell-type-specific expression profilinga Step 1: Labeling Methods Refs No [18,21,22] labeling
Description Cells are chosen randomly or based on position, morphology or electrophysiology
Signal NA
Ease ***
Cost $
Pros No labeling necessary
Antibody
[30]
Cells distinguished by known marker(s) are immunocytochemically labeled
Varies
**
$$
Identifies neuronal population based on known marker(s)
Tracer injection
[27,31,32]
Cells are labeled by stereotaxic injection of fluorescent tracers into projection target
**
**
$$
Cell population defined by projection
Transgenic
[27,35–39]
Defined subpopulations of neurons express a fluorescent protein under a specific promoter or enhancer
Varies
*
$$$
(i) Genetically defined population (ii) Population can be identified across animals for other studies (e.g. morphology, physiology)
Description Following patch-clamp recording, cytoplasmic contents are aspirated
Yield *
Purity **
Cost $$
Pros (i) Simultaneous electrophysiology and anatomy (ii) Not populationbased (i) Commercial support (ii) Medium yield
Step 2: Isolation Methods Refs Single-cell [20,21,25,26] aspiration
Cons (i) Identification of cell types is post-hoc, so rare types can be missed (ii) Repeatability difficult to assess (i) Requires antibody (ii) For live cells, epitope must be extracellular (iii) In fixed tissue, RNA can be degraded (i) Labels only projection neurons, not interneurons (ii) Multiple cell types can share projection targets (i) Generating useable transgenics is difficult and labor-intensive (ii) Few cell-type-specific promoters or enhancers are known
Cons A tiny sample (fraction of one cell) leads to high measurement noise and false-negative rate
LCM
[22,28–31]
A laser is used under microscopic control to cut out cell-sized regions from thin tissue sections
**
*
$$$
FACS
[32,35–39]
***
**
$$$
Manual sorting
[27]
*
***
$
MRNA tagging
[33,34]
Droplets containing dissociated cells are automatically optically analyzed and sorted A glass pipette is used to collect and purify dissociated cells under a fluorescence dissecting microscope Defined subpopulations of neurons express PABP; cells are fixed and PABP immunoprecipitated to isolate bound mRNA
***
?
?
(i) Commercially available (ii) Fast (iii) High yield (i) High purity (ii) Applicable to adult neurons (iii) Applicable to dimly labeled neurons No cell sorting required
Sensitivity ***
# Genes *
Cost
Pros
Cons
$
High sensitivity
Low # genes
**
**
$$$
**
***
$$
(i) Can reveal novel transcripts (ii) Absolute count of transcript copy number (i) High multiplexity (ii) Commercially available, reliable
(i) High sequencing costs (ii) Laborious cloning step Requires amplification if starting from small samples
Step 3: Detection Methods Refs PCR
[25,26]
SAGE
[55]
Microarray
[21–23,27–40]
Description Individual transcripts are assayed by gene specific primer sets 15-bp sequence ‘tags’ are cloned and sequenced Many spotted complementary sequences detect abundance of transcripts by hybridization
(i) Requires RNA amplification (ii) Contamination by surrounding cells (iii) Samples must be fixed or frozen (i) Cannot be used with adult neurons (ii) Requires brightly labeled cells Low yield (requires RNA amplification)
(i) Requires fixation (ii) Sensitivity not yet established
a Each of the three main components of cell-type-specific gene expression profiling (labeling, purifying and transcript detection) can be accomplished using the different methods listed and evaluated here. Abbreviations: FACS, fluorescent-activated cell sorter; LCM, laser capture microdissection; NA, not applicable; PABP, poly(A) binding protein; SAGE, serial analysis of gene expression. # Genes refers to the number of genes that can be assayed from each sample.
www.sciencedirect.com
342
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
Neuronal cell-type-specific expression profiling by LCM Some investigators employ laser capture microdissection (LCM) to purify neuronal subpopulations for microarray analysis [28–31]. Thin (w10 mm) brain cryosections are prepared, stained using antibodies against proteins that distinguish the cell type of interest, and then microdissected so that only stained tissue is recovered for further processing. Midbrain dopaminergic neurons have been a favored target for such investigations, both because of their clinical importance in neurodegenerative disorders such as Parkinson’s disease and because they are readily labeled by antibodies to tyrosine hydroxylase (TH), the enzyme responsible for the production of dopamine. Two groups [29,30] compared dopaminergic neurons from neighboring midbrain regions, the ventral tegmental area (VTA) and the substantia nigra, to identify genes that might contribute to the higher susceptibility of the latter population to neurodegeneration in Parkinson’s disease. Both studies identified numerous genes that had statistically significant differential expression. Yao et al. [31] compared VTA dopaminergic neurons with corticostriatal pyramidal cells retrogradely labeled by striatal injection of fluorogold. Several dozen genes were identified as differentially expressed using very rigid criteria. Out of 17 of these genes that were subjected to quantitative realtime PCR on independent samples, differential expression was confirmed for 11 and not confirmed for six. These six genes were known to be expressed in glia, indicating that their identification was the result of contamination of samples by surrounding tissue. Several genes enriched in corticostriatal neurons match those found in this cell type by other investigators using cell-type-specific expression analysis [27,32]. These studies demonstrate that LCM achieves sufficient, although not perfect, purity and RNA integrity, and thus promises to be a useful complement to the cell-sorting approaches we will go on to outline, especially because it can be more readily applied to the study of fixed human tissue. Neuronal cell-type-specific expression profiling by RNA-tagging RNA tagging provides a way to purify mRNA from a genetically labeled cell population without the need to sort cells. A FLAG-tagged poly(A)-binding protein (PABP) is expressed in the population of interest, where it binds to polyadenylated mRNA. The tissue is fixed, to cross-link the PABP covalently to the mRNA, and then homogenized. mRNA from the cell population of interest is purified by anti-FLAG affinity purification. Kunitomo et al. [33] used mRNA tagging to generate expression profiles of ciliated sensory neurons in Caenorhabditis elegans. Yang et al. used it to generate expression profiles of photoreceptors in Drosophila melanogaster [34]. In both cases, genes known to be expressed in the targeted cell types were identified along with many other genes whose expression pattern was not known. Neuronal cell-type-specific expression profiling by live-cell sorting Higher purity of neuronal populations can be achieved using live-cell sorting because dissociated cells can be www.sciencedirect.com
stripped completely bare of contaminating factors such as glia, neuronal fibers and the extracellular matrix. In addition, such samples provide RNA in a pristine form because they need not be subjected to freezing or fixation. The first successful execution of this strategy was carried out by Zhang et al. [35], who expressed green fluorescent protein (GFP) in the six touch-receptor neurons of the nematode C. elegans; they purified these fluorescent cells from thousands of dissociated embryos on an automated fluorescence-activated cell sorter (FACS) after culturing them overnight to increase GFP expression, and finally extracted and amplified mRNA for microarray analysis. Subsequent implementations of this method in C. elegans have been aimed at identifying genes specifically expressed in other populations of neurons, including olfactory and thermosensory neurons [36] and GABAergic motoneurons [37,38]. Similar progress has been made in cell-type-specific expression profiling in the mammalian nervous system. Buchstaller et al. [39] FACS-sorted and profiled GFPexpressing neural crest stem cells at several developmental time points as they differentiated into Schwann cells, revealing hundreds of genes likely to be involved in this process. Arlotta et al. [32] published the first genomewide expression profiles from specific neuronal subpopulations in the mammalian CNS using live-cell sorting. Cortical projection neurons were retrogradely labeled by injection of tracer into contralateral cortex or spinal cord. These were FACS-sorted and profiled at four different developmental time points: embryonic day 18, and postnatal day (P)3, P6 and P14. In addition, corticotectal neurons were profiled at P14. Hundreds of genes were found to be differentially expressed between these closely related cell types. These genes spanned a wide range of functional categories and most of these had not previously been associated with cortical function, demonstrating the ability of this approach to discover entirely novel gene functions. A subset of these genes was highlighted as having expression trends that suggest involvement in developmental specification of the profiled cells. One of these genes, encoding the transcription factor Ctip2, was found to be crucial for the development of corticospinal axons. A second transcription factor gene, Fezl, was found to be required for the specification of corticofugal motor axons [40,41]; these projections seem to be absent from FezlK/K mouse cortices. These studies demonstrate how the wealth of data produced by neuronal cell-type-specific microarray studies can be used to generate novel, specific and testable hypotheses about gene function in neurons. Sugino et al. [27] used transgenic fluorescent-proteinexpressing mice to profile particular cell types in the mammalian forebrain. These mouse lines were gathered from various laboratories [42–45]. Through a combination of promoter specificity in the transgene construct and positional effects determined by the transgene genomic insertion point, each line expresses fluorescent protein in a highly restricted subset of neurons. The manual sorting method employed to purify these labeled populations offers two key advantages over automated FACS sorting. First, it is a gentle procedure that enables purification of
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
343
types from different regions of the neocortex were found to be nearly identical, supporting the ‘canonical microcircuit’ hypothesis of cortical architecture. This study showed that genome-wide expression profiles can be used to classify neurons based on objective criteria that can be applied to any labeled cell type in the nervous system. This sets the stage for the profiling and unbiased classification of a much larger number of cell types in the cortex and throughout the brain.
adult neurons. Second, it does not require thousands of neurons, so limited populations can be readily purified. For example, a single, rare interneuron subtype from a single cortical region can be isolated. However, the flip side of this is that the yield of manual sorting is low (typically 30–100 neurons). Fortunately, the small sample amplification procedure is reliable in this range. In contrast to previous studies, which had compared expression in two or three neuronal cell types, Sugino et al. [27] profiled 12 distinct populations of mouse neurons chosen to cover a wide range of cell types from different forebrain regions, with different neurotransmitter phenotypes, morphologies and electrophysiological characteristics (Figure 2a). This screen provided a wealth of novel candidate molecular neuronal markers and, as we will discuss later, identified particular functional categories of genes over-represented in differentially expressed genes. More importantly, this was the first study to use microarrays to address the problem of neuronal classification in the mammalian nervous system. A taxonomic dendrogram was generated based on the expression data (Figure 2b). The position on the dendrogram reflects the relative distance in ‘expression space’ between cell types. The taxonomy captured the expected major relationships between the profiled populations, such as the distinction between GABAergic and glutamatergic neurons, and also unexpected relationships between some lower-order subtypes. Homologous cell
Expression modules Sugino et al. [27] identified sets of genes that distinguish cell types and then used the Gene Ontology database (http://www.geneontology.org/) [46] to identify biological functions known to be subserved by these genes. For example, different cell types expressed different sets of ion-channel subunits (presumably reflecting differences in their firing patterns) and different sets of cytoskeletal proteins (presumably reflecting differences in their morphologies). It seems likely that, just as cellular phenotypes (e.g. fast-spiking or stellate morphology) are repeated across cell types in disparate regions, so there might be expression modules (sets of co-regulated genes) that give rise to these phenotypes. Just as proteins are described as collections of unique and common domains or motifs, cell types might be best described as collections of expression modules, each module providing a particular function (e.g. transmitter phenotype, axon-targeting or
(a)
(b) Hippocampus
Cingulate cortex
GIN-HP
YFPH-CG
CT6-CG
YFPH-S1
YFPH-AM
YFPH-HP
YFPH-S1
YFPH-CG
CT6-CG
G30-AM
G30-S1
G30-CG
G42-CG
GIN-CG
G30-S1
GIN-HP
G42-LG
GIN-CG
YFPH-HP
Somatosensory cortex G30-CG
Amygdala YFPH-AM G42-CG
G30-AM
G42-LG
Thalamus Figure 2. Cell-type-specific expression profiles for neuronal taxonomy. Adapted from Ref. [27]. (a) Twelve neuronal populations profiled by Sugino et al. [27] are named according to transgenic strain (YFPH [45], GIN [44], G30 [43] or G42 [42]) or projection-neuron class retrogradely labeled by tracer injection (layer 6 corticothalamic neurons, CT6), and according to brain region [amygdala (AM), cingulate cortex (CG), hippocampus (HP), lateral geniculate nucleus (LG) or primary somatosensory cortex (S1)]. For example, YFPH-CG is from strain YFPH in cingulate cortex. A low-power micrograph taken at the locations marked on the atlas sections shows selective labeling of the populations by either green or yellow fluorescent protein or red fluorescent microspheres. A representative electrophysiological recording is also shown for each population. (b) Dendrogram reflecting distance in ‘gene expression space’ between the twelve populations. This was generated by hierarchical clustering according to relative differences in genome-wide expression profiles. Each point represents data from one microarray and each population was profiled in triplicate. www.sciencedirect.com
344
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
spiking properties). The partial independence of these modules might explain the observation that in cortical interneurons, for example, the same firing type can be associated with multiple morphologies, and vice versa [14]. Identifying the molecular fingerprint of a particular cell type will also enable investigators to ask comparative questions – how conserved are cell types between species? Cajal initially thought that the cortex of humans contained many cell types not found in ‘lower’ mammals, but his student Lorente de No argued forcefully that even the lowly mouse contained a comparable number and complexity of cell types [47]. Expression distance, or perhaps more refined versions of this measure based on similarity of expression modules, might provide an objective, quantitative metric on which to make such comparisons. After defining sets of markers for cell types in the mouse, it might be feasible to address this in other species such as monkeys and humans using a combination of retrograde labeling and live sorting, or of antibody labeling and LCM. Such comparisons are crucial if we are to relate the functional analyses of circuits in cats, monkeys and other species to cellular and molecular work in rodents. In addition, with the sequencing of multiple mammalian genomes, including those of mouse, rat, dog, ape, human and other related vertebrate species, a new field of comparative neurogenomics is likely to emerge, driven by the search for genomic differences that regulate species differences in brain structure and function. These genomic differences are likely to be not only in the coding sequences of individual genes but also in the extragenic regulatory sequences that control their expression.
A mammalian neurome project? Although the number of distinct neuronal cell types in the mammalian ‘neurome’ (103–104) is likely to be within an order of magnitude of the number of genes in the mammalian genome (2–4!104), it seems unlikely that single genes specify, or can serve as markers for, single cell types. For example, although parvalbumin expression is often considered the sine qua non of fast-spiking basket and chandelier cells, not all basket cells express parvalbumin [14], and parvalbumin is also expressed in the auditory brainstem [48], somatosensory layer 5 cortical pyramidal neurons [49] and many other structures. Ideal markers (i.e. that label one cell type and no others) are relatively abundant when comparing a small number of cell types, but are likely to be rare when comparing across all cell types. A more complete picture of the pattern of expression within the brain of all genes within the mouse genome is likely to emerge shortly from large-scale mapping efforts such as the Allen Institute Brain Atlas (http://www. brainatlas.org/) [50,51] and Gensat (http://www.gensat. org/index.html) [52]. But this will not, by itself, enable a complete map of brain neuronal cell types to be assembled. Cell types are likely to be defined by coexpression of many genes and in most cases it will be hard to discern precise patterns of coexpression with cellular resolution by www.sciencedirect.com
comparing hybridization or reporter expression patterns across multiple tissue sections. We suggest that a systematic attempt to map neuronal cell types is now feasible and represents an important next logical step in post-genomic neuroscience. Several genetic strategies are available to create a much larger library of fluorescently labeled cell types in mice. These include bacterial artificial chromosome (BAC) transgenics [52], BAC recombination to knock-in to endogenous loci [53], and enhancer-trap strategies using lentiviral vectors [54]. The loci to use for knock-in approaches might be identified based on restricted spatial expression patterns in the Allen Atlas or Gensat databases, or from profiling studies such as those we have already described. Enhancer traps use random insertion of a minimal promoter that drives celltype-specific reporter expression only when it inserts next to regulatory elements that direct expression in those cell types. None of these approaches is likely to label only a single cell type within the brain. Instead, careful physiological, anatomical and genomic characterization will be needed to distinguish the multiple cell types labeled in each structure within each line. The importance of the labeling is that it enables multiple investigators to query the same cell types repeatedly and therefore to share reliable data about their phenotypic and genotypic properties. To be most useful, given the large characterization effort required, these new genetic resources should also permit cell-type-specific manipulation of gene expression. Unlike the majority of transgenics produced so far, future efforts are likely to focus increasingly on expressing both fluorescent proteins for cell identification and, in the same cells, additional molecules such as Cre-recombinase to permit cell-type-specific manipulation of gene expression. Such manipulation will be crucial in elucidating the contribution both of the gene to the cell type and of the cell type to the circuit. Recognizing the importance of such approaches, the NIH has recently issued a call for proposals to generate transgenic ‘driver’ lines (RFA Number MH-06–007; http://grants.nih.gov/grants/guide/ rfa-files/RFA-MH-06-007.html). With the increasing availability of such tools the study of mammalian brain circuits enters a new phase, in which individual neuronal cell types can be easily and repeatedly identified and in which their function can be selectively and flexibly altered. References 1 Monyer, H. and Markram, H. (2004) Interneuron Diversity series: Molecular and genetic tools to study GABAergic interneuron diversity and function. Trends Neurosci. 27, 90–97 2 Mott, D.D. and Dingledine, R. (2003) Interneuron Diversity series: Interneuron research – challenges and strategies. Trends Neurosci. 26, 484–488 3 Koester, S.E. and O’Leary, D.D. (1992) Functional classes of cortical projection neurons develop dendritic distinctions by class-specific sculpting of an early common pattern. J. Neurosci. 12, 1382–1393 4 Zhang, Z.W. and Deschenes, M. (1997) Intracortical axonal projections of lamina VI cells of the primary somatosensory cortex in the rat: a single-cell labeling study. J. Neurosci. 17, 6365–6379 5 Kasper, E.M. et al. (1994) Pyramidal neurons in layer 5 of the rat visual cortex. I. Correlation among cell morphology, intrinsic electrophysiological properties, and axon targets. J. Comp. Neurol. 339, 459–474 6 Masland, R.H. (2001) Neuronal diversity in the retina. Curr. Opin. Neurobiol. 11, 431–436
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
7 Masland, R.H. (2004) Neuronal cell types. Curr. Biol. 14, R497–R500 8 Koch, C. (2004) The Quest for Consciousness: A Neurobiological Approach, Roberts and Company Publishers 9 Stevens, C.F. (1998) Neuronal diversity: too many cell types for comfort? Curr. Biol. 8, R708–R710 10 Parra, P. et al. (1998) How many subtypes of inhibitory cells in the hippocampus? Neuron 20, 983–993 11 Woese, C.R. and Fox, G.E. (1977) Phylogenetic structure of the prokaryotic domain: the primary kingdoms. Proc. Natl. Acad. Sci. U. S. A. 74, 5088–5090 12 Zuckerkandl, E. and Pauling, L. (1965) Molecules as documents of evolutionary history. J. Theor. Biol. 8, 357–366 13 Nelson, S. (2002) Cortical microcircuits: diverse or canonical? Neuron 36, 19–27 14 Markram, H. et al. (2004) Interneurons of the neocortical inhibitory system. Nat. Rev. Neurosci. 5, 793–807 15 Mirnics, K. et al. (2000) Molecular characterization of schizophrenia viewed by microarray analysis of gene expression in prefrontal cortex. Neuron 28, 53–67 16 Sandberg, R. et al. (2000) Regional and strain-specific gene expression mapping in the adult mouse brain. Proc. Natl. Acad. Sci. U. S. A. 97, 11038–11043 17 Xie, T. et al. (2002) Changes in gene expression linked to methamphetamine-induced dopaminergic neurotoxicity. J. Neurosci. 22, 274–283 18 Zhao, X. et al. (2001) Transcriptional profiling reveals strict boundaries between hippocampal subregions. J. Comp. Neurol. 441, 187–196 19 Zirlinger, M. and Anderson, D. (2003) Molecular dissection of the amygdala and its relevance to autism. Genes Brain Behav. 2, 282–294 20 Eberwine, J. et al. (1992) Analysis of gene expression in single live neurons. Proc. Natl. Acad. Sci. U. S. A. 89, 3010–3014 21 Hinkle, D. et al. (2004) Single neurons as experimental systems in molecular biology. Prog. Neurobiol. 72, 129–142 22 Kamme, F. et al. (2003) Single-cell microarray analysis in hippocampus CA1: demonstration and validation of cellular heterogeneity. J. Neurosci. 23, 3607–3615 23 Tietjen, I. et al. (2003) Single-cell transcriptional analysis of neuronal progenitors. Neuron 38, 161–175 24 Zirlinger, M. et al. (2001) Amygdala-enriched genes identified by microarray technology are restricted to specific amygdaloid subnuclei. Proc. Natl. Acad. Sci. U. S. A. 98, 5270–5275 25 Cauli, B. et al. (2000) Classification of fusiform neocortical interneurons based on unsupervised clustering. Proc. Natl. Acad. Sci. U. S. A. 97, 6144–6149 26 Toledo-Rodriguez, M. et al. (2004) Correlation maps allow neuronal electrical properties to be predicted from single-cell gene expression profiles in rat neocortex. Cereb. Cortex 14, 1310–1327 27 Sugino, K. et al. (2006) Molecular taxonomy of major neuronal classes in the adult mouse forebrain. Nat. Neurosci. 9, 99–107 28 Backes, E. and Hemby, S.E. (2003) Discrete cell gene profiling of ventral tegmental dopamine neurons after acute and chronic cocaine self-administration. J. Pharmacol. Exp. Ther. 307, 450–459 29 Chung, C.Y. et al. (2005) Cell type-specific gene expression of midbrain dopaminergic neurons reveals molecules involved in their vulnerability and protection. Hum. Mol. Genet. 14, 1709–1725 30 Greene, J.G. et al. (2005) Gene expression profiling of rat midbrain dopamine neurons: implications for selective vulnerability in parkinsonism. Neurobiol. Dis. 18, 19–31 31 Yao, F. et al. (2005) Microarray analysis of fluoro-gold labeled rat dopamine neurons harvested by laser capture microdissection. J. Neurosci. Methods 143, 95–106
www.sciencedirect.com
345
32 Arlotta, P. et al. (2005) Neuronal subtype-specific genes that control corticospinal motor neuron development in vivo. Neuron 45, 207–221 33 Kunitomo, H. et al. (2005) Identification of ciliated sensory neuronexpressed genes in Caenorhabditis elegans using targeted pull-down of poly(A) tails. Genome Biol. 6, R17 34 Yang, Z. et al. (2005) Isolation of mRNA from specific tissues of Drosophila by mRNA tagging. Nucleic Acids Res. 33, e148 35 Zhang, Y. et al. (2002) Identification of genes expressed in C. elegans touch receptor neurons. Nature 418, 331–335 36 Colosimo, M.E. et al. (2004) Identification of thermosensory and olfactory neuron-specific genes via expression profiling of single neuron types. Curr. Biol. 14, 2245–2251 37 Fox, R.M. et al. (2005) A gene expression fingerprint of C. elegans embryonic motor neurons. BMC Genomics 6, 42 38 Cinar, H. et al. (2005) Expression profiling of GABAergic motor neurons in Caenorhabditis elegans. Curr. Biol. 15, 340–346 39 Buchstaller, J. et al. (2004) Efficient isolation and gene expression profiling of small numbers of neural crest stem cells and developing Schwann cells. J. Neurosci. 24, 2357–2365 40 Molyneaux, B.J. et al. (2005) Fezl is required for the birth and specification of corticospinal motor neurons. Neuron 47, 817–831 41 Chen, B. et al. (2005) Fezl regulates the differentiation and axon targeting of layer 5 subcortical projection neurons in cerebral cortex. Proc. Natl. Acad. Sci. U. S. A. 102, 17184–17189 42 Ango, F. et al. (2004) Ankyrin-based subcellular gradient of neurofascin, an immunoglobulin family protein, directs GABAergic innervation at Purkinje axon initial segment. Cell 119, 257–272 43 Lopez-Bendito, G. et al. (2004) Preferential origin and layer destination of GAD65-GFP cortical interneurons. Cereb. Cortex 14, 1122–1133 44 Oliva, A.A., Jr. et al. (2000) Novel hippocampal interneuronal subtypes identified using transgenic mice that express green fluorescent protein in GABAergic interneurons. J. Neurosci. 20, 3354–3368 45 Feng, G. et al. (2000) Imaging neuronal subsets in transgenic mice expressing multiple spectral variants of GFP. Neuron 28, 41–51 46 Ashburner, M. et al. (2000) Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. 25, 25–29 47 Lorente de No, R. (1992) The cerebral cortex of the mouse (a first contribution – the ‘acoustic’ cortex). Somatosens. Mot. Res. 9, 3–36 48 Felmy, F. and Schneggenburger, R. (2004) Developmental expression of the Ca2C-binding proteins calretinin and parvalbumin at the calyx of held of rats and mice. Eur. J. Neurosci. 20, 1473–1482 49 Preuss, T.M. and Kaas, J.H. (1996) Parvalbumin-like immunoreactivity of layer V pyramidal cells in the motor and somatosensory cortex of adult primates. Brain Res. 712, 353–357 50 Gewin, V. (2005) A golden age of brain exploration. PLoS Biol. 3, e24 51 Boguski, M.S. and Jones, A.R. (2004) Neurogenomics: at the intersection of neurobiology and genome sciences. Nat. Neurosci. 7, 429–433 52 Gong, S. et al. (2003) A gene expression atlas of the central nervous system based on bacterial artificial chromosomes. Nature 425, 917–925 53 Cotta-de-Almeida, V. et al. (2003) A new method for rapidly generating gene-targeting vectors by engineering BACs through homologous recombination in bacteria. Genome Res. 13, 2190–2194 54 Lois, C. et al. (2002) Germline transmission and tissue-specific expression of transgenes delivered by lentiviral vectors. Science 295, 868–872 55 Blackshaw, S. et al. (2001) Comprehensive analysis of photoreceptor gene expression and the identification of candidate retinal disease genes. Cell 107, 579–589
TRENDS in Neurosciences Vol.29 No.6 June 2006
TINS special issue: The Neural Substrates of Cognition
The problem of neuronal cell types: a physiological genomics approach Sacha B. Nelson, Ken Sugino and Chris M. Hempel Department of Biology and National Center for Behavioral Genomics, Brandeis University, MS 008, 415 South Street, Waltham, MA 02454-9110, USA
Neural circuits within the vertebrate brain are composed of highly diverse cell types. The exact extent of this diversity is a matter of continuing debate. For example, do cortical interneurons comprise a few, dozens or O100 distinct cell types? Recently, several groups have used microarrays to measure genome-wide gene expression profiles for specific neuronal cell types. These methods can offer an objective basis for neuronal classification. In this review, we argue that this approach should now be carried out more broadly and that it should be coupled to large-scale efforts to generate mouse driver lines in which tools for genetic manipulation, such as the Cre recombinase, are expressed in identified cell types within the brain. This would enable neuroscientists to begin to investigate more systematically the roles of specific genes in establishing particular cellular phenotypes, and also the roles of particular cell types within brain circuits. This review is part of the TINS special issue on The Neural Substrates of Cognition. Introduction We share nearly all of our genes with mice and the great majority of our genes with much more distant relatives such as flies and worms. But the neural structures built during human development are much larger, more complex and, at least with respect to flies and worms, contain a very different complement of cell types and neural organs. The substrates of our unique cognitive abilities must lie in these diverse neural cell types, in the complex circuits into which they are organized, and in the patterns of activity that those circuits sustain. There are two experimental problems that, if solved, would greatly accelerate progress in the search for the cellular and circuit-level substrates of cognition within the mammalian brain. First is the problem of being able to identify easily, rapidly, and objectively the many cell types that comprise circuits within the neocortex and other brain regions. Second is the problem of being able to manipulate the function of those cell types selectively and flexibly. Recently, application of modern genetic and genomic tools to questions in cortical neurobiology has offered new opportunities to investigate these two problems. Here, we review recent advances that open the door to a systematic examination of the first problem: Corresponding author: Nelson, S.B. ([email protected]). Available online 22 May 2006
that of objectively identifying and classifying cortical cell types. We also discuss how these advances might in turn permit solution of problem number two: that of manipulating those cell types to understand their function better. The realization that cortical cell types are highly diverse was first made by early anatomists using Golgi stain to reveal axonal and dendritic morphology. Although neurons could be broadly classified as having local or longrange axonal projections, or as having smooth or spiny dendrites, the exact number of distinct cell types remains a matter of continual debate. Today there is broad acceptance that the two major classes of cortical neurons are pyramidal neurons, which make long-range and local connections, use glutamate as a neurotransmitter and have spiny dendrites, and inhibitory interneurons, which make primarily local connections, use GABA and various neuropeptides as neurotransmitters, and have smooth or sparsely spiny dendrites. But although it is also broadly accepted that within these classes neurons can be highly diverse, the proper way to parse this diversity into recognizable cell types is far from agreed upon. The question of how to classify GABAergic interneurons has received tremendous attention in the pages of this and other journals [1,2] but without consensus. The same is true for cortical pyramidal cells, despite clear demonstrations of the dramatic differences between pyramidal cell subtypes [3–5]. The story is somewhat different from that in the retina, and the analogy is perhaps helpful because it provides hope that the problem is not insoluble and suggests a scale for the total number of neural cell types. Using a combination of anatomical, physiological and gene expression properties, retinal anatomists and physiologists have identified 55–60 distinct neuronal cell types [6,7]. This number has inspired some to speculate that the total number of cell types within the mammalian cortex is w1000 [8,9]. Most retinal cell types can be recognized across vertebrate orders and considerable progress has been made in understanding their connectivity and function. Such success has been possible in the retina owing to its accessibility to direct study using physiological stimuli, and to its clearly defined laminar structure, cellular morphologies and cellular functions. The problem in the cerebral cortex arises primarily from the fact that
www.sciencedirect.com 0166-2236/$ - see front matter Q 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tins.2006.05.004
340
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
morphology, connectivity and circuit function of the constituent cell types are not as easily segregated. For this reason, definitive mapping between the various features of cellular phenotype has not been possible: a classification based purely on morphology does not adequately capture the diversity of intrinsic firing properties or of synaptic properties and, conversely, classifying cells purely on the basis of physiological properties does not yet fully predict their morphology or synaptic connectivity. It is perhaps valid to ask whether this process of identification and classification is really important [10]. We believe the answer is a resounding yes. The importance lies less in the biological insights implicit in the classification than in the enabling of multiple investigators to refer unambiguously to the same neurons. If we are to begin collectively to ‘circuitbreak’ the cortex and related structures, and to identify homologous cells and circuits across cortical regions and across mammalian species, we must first have a broadly accepted cellular ‘parts list’. Genome-wide expression profiling for neuronal classification The field of phylogenetic taxonomy was revolutionized by the use of genomic sequence comparison [11,12]. Similarly the classification of protein families, such as those of ion channels, has been greatly clarified by measures of DNA-sequence similarity [13]. Although different cell types within an individual organism are likely to be genetically identical, they differ in which genes they express. Therefore, it has been suggested that genome-wide gene expression profiling could provide a sound and unbiased method for neuronal classification [2,14]. This approach rests on the assumption that the overall function of a neuron within a circuit is highly correlated with its gene expression profile. DNA microarrays provide an ideal platform for such studies. To date, however, microarrays have not been effectively and systematically employed to address problems in neuronal taxonomy – the majority of neuronal microarray experiments have been carried out on tissue homogenates where distinctions between cell types are lost [15–19]. Neuronal classification, of course, requires analysis at the cellular level. Single-cell gene expression studies have been underway for over a decade [20,21] but the tiny amount of mRNA available from a single cell has limited such investigations to a tiny fraction of neuronally expressed genes. For replicable microarray-based genome-wide expression analysis, high yields of linearly amplified mRNA are required and it is not yet clear whether these can be reliably derived from single cells. Despite this technical difficulty, some investigators have been able to generate expression profiles from single cells. However, these profiles were of randomly selected neurons from mixed populations [22,23], in which too few neurons were compared to identify recognizable classes or distinguish technical from biological variability. Both homogenate and single-cell expression studies have been useful for suggesting functional gene classes involved in schizophrenia [15], identifying genes involved in maturation of olfactory neuron precursors [23], www.sciencedirect.com
identifying neuronal markers in the amygdala [19,24], and aiding in the classification of cortical neurons using small numbers of known markers [25,26]. But neither approach exploits the potential of expression profiling for the purposes of neuronal taxonomy. This requires generating genome-wide expression profiles from defined neuronal populations. How is this goal best achieved? The most straightforward strategy is as follows: (i) a functionally distinct cell type must be labeled with a marker; (ii) that population must somehow be purified from the rest of the tissue using that marker; and the mRNA must be (iii) extracted and (iv) amplified to yield a sample that can be probed using a microarray. A specific implementation [27] of such a strategy is outlined in Figure 1. However, each of these steps can be carried out using various approaches (Table 1). Optimizing and integrating all of these steps has required considerable technical and methodological innovation in many laboratories. These efforts, which we will review in the subsequent paragraphs, have finally over the last few years begun to bear fruit. (a)
Labeled mouse
(b) Slicing and protease digestion
(c)
(h)
(g)
Data analysis
Microarray hybridization and scanning
Microdissection
(f) Cell lysis, poly(A) RNA amplification and labeling B
B
B AAAAAA
(d)
Dissociation, trituration and plating
(e)
Manual sorting
TRENDS in Neurosciences
Figure 1. One method used to evaluate gene expression in genetically defined subpopulations of neurons. This is the strategy employed by Sugino et al. [27]. Once dissociated cells are obtained (a–d), this approach involves fluorescencebased manual sorting of live neurons (e) followed by expression analysis by microarray (f–h).
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
341
Table 1. Various methods for cell-type-specific expression profilinga Step 1: Labeling Methods Refs No [18,21,22] labeling
Description Cells are chosen randomly or based on position, morphology or electrophysiology
Signal NA
Ease ***
Cost $
Pros No labeling necessary
Antibody
[30]
Cells distinguished by known marker(s) are immunocytochemically labeled
Varies
**
$$
Identifies neuronal population based on known marker(s)
Tracer injection
[27,31,32]
Cells are labeled by stereotaxic injection of fluorescent tracers into projection target
**
**
$$
Cell population defined by projection
Transgenic
[27,35–39]
Defined subpopulations of neurons express a fluorescent protein under a specific promoter or enhancer
Varies
*
$$$
(i) Genetically defined population (ii) Population can be identified across animals for other studies (e.g. morphology, physiology)
Description Following patch-clamp recording, cytoplasmic contents are aspirated
Yield *
Purity **
Cost $$
Pros (i) Simultaneous electrophysiology and anatomy (ii) Not populationbased (i) Commercial support (ii) Medium yield
Step 2: Isolation Methods Refs Single-cell [20,21,25,26] aspiration
Cons (i) Identification of cell types is post-hoc, so rare types can be missed (ii) Repeatability difficult to assess (i) Requires antibody (ii) For live cells, epitope must be extracellular (iii) In fixed tissue, RNA can be degraded (i) Labels only projection neurons, not interneurons (ii) Multiple cell types can share projection targets (i) Generating useable transgenics is difficult and labor-intensive (ii) Few cell-type-specific promoters or enhancers are known
Cons A tiny sample (fraction of one cell) leads to high measurement noise and false-negative rate
LCM
[22,28–31]
A laser is used under microscopic control to cut out cell-sized regions from thin tissue sections
**
*
$$$
FACS
[32,35–39]
***
**
$$$
Manual sorting
[27]
*
***
$
MRNA tagging
[33,34]
Droplets containing dissociated cells are automatically optically analyzed and sorted A glass pipette is used to collect and purify dissociated cells under a fluorescence dissecting microscope Defined subpopulations of neurons express PABP; cells are fixed and PABP immunoprecipitated to isolate bound mRNA
***
?
?
(i) Commercially available (ii) Fast (iii) High yield (i) High purity (ii) Applicable to adult neurons (iii) Applicable to dimly labeled neurons No cell sorting required
Sensitivity ***
# Genes *
Cost
Pros
Cons
$
High sensitivity
Low # genes
**
**
$$$
**
***
$$
(i) Can reveal novel transcripts (ii) Absolute count of transcript copy number (i) High multiplexity (ii) Commercially available, reliable
(i) High sequencing costs (ii) Laborious cloning step Requires amplification if starting from small samples
Step 3: Detection Methods Refs PCR
[25,26]
SAGE
[55]
Microarray
[21–23,27–40]
Description Individual transcripts are assayed by gene specific primer sets 15-bp sequence ‘tags’ are cloned and sequenced Many spotted complementary sequences detect abundance of transcripts by hybridization
(i) Requires RNA amplification (ii) Contamination by surrounding cells (iii) Samples must be fixed or frozen (i) Cannot be used with adult neurons (ii) Requires brightly labeled cells Low yield (requires RNA amplification)
(i) Requires fixation (ii) Sensitivity not yet established
a Each of the three main components of cell-type-specific gene expression profiling (labeling, purifying and transcript detection) can be accomplished using the different methods listed and evaluated here. Abbreviations: FACS, fluorescent-activated cell sorter; LCM, laser capture microdissection; NA, not applicable; PABP, poly(A) binding protein; SAGE, serial analysis of gene expression. # Genes refers to the number of genes that can be assayed from each sample.
www.sciencedirect.com
342
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
Neuronal cell-type-specific expression profiling by LCM Some investigators employ laser capture microdissection (LCM) to purify neuronal subpopulations for microarray analysis [28–31]. Thin (w10 mm) brain cryosections are prepared, stained using antibodies against proteins that distinguish the cell type of interest, and then microdissected so that only stained tissue is recovered for further processing. Midbrain dopaminergic neurons have been a favored target for such investigations, both because of their clinical importance in neurodegenerative disorders such as Parkinson’s disease and because they are readily labeled by antibodies to tyrosine hydroxylase (TH), the enzyme responsible for the production of dopamine. Two groups [29,30] compared dopaminergic neurons from neighboring midbrain regions, the ventral tegmental area (VTA) and the substantia nigra, to identify genes that might contribute to the higher susceptibility of the latter population to neurodegeneration in Parkinson’s disease. Both studies identified numerous genes that had statistically significant differential expression. Yao et al. [31] compared VTA dopaminergic neurons with corticostriatal pyramidal cells retrogradely labeled by striatal injection of fluorogold. Several dozen genes were identified as differentially expressed using very rigid criteria. Out of 17 of these genes that were subjected to quantitative realtime PCR on independent samples, differential expression was confirmed for 11 and not confirmed for six. These six genes were known to be expressed in glia, indicating that their identification was the result of contamination of samples by surrounding tissue. Several genes enriched in corticostriatal neurons match those found in this cell type by other investigators using cell-type-specific expression analysis [27,32]. These studies demonstrate that LCM achieves sufficient, although not perfect, purity and RNA integrity, and thus promises to be a useful complement to the cell-sorting approaches we will go on to outline, especially because it can be more readily applied to the study of fixed human tissue. Neuronal cell-type-specific expression profiling by RNA-tagging RNA tagging provides a way to purify mRNA from a genetically labeled cell population without the need to sort cells. A FLAG-tagged poly(A)-binding protein (PABP) is expressed in the population of interest, where it binds to polyadenylated mRNA. The tissue is fixed, to cross-link the PABP covalently to the mRNA, and then homogenized. mRNA from the cell population of interest is purified by anti-FLAG affinity purification. Kunitomo et al. [33] used mRNA tagging to generate expression profiles of ciliated sensory neurons in Caenorhabditis elegans. Yang et al. used it to generate expression profiles of photoreceptors in Drosophila melanogaster [34]. In both cases, genes known to be expressed in the targeted cell types were identified along with many other genes whose expression pattern was not known. Neuronal cell-type-specific expression profiling by live-cell sorting Higher purity of neuronal populations can be achieved using live-cell sorting because dissociated cells can be www.sciencedirect.com
stripped completely bare of contaminating factors such as glia, neuronal fibers and the extracellular matrix. In addition, such samples provide RNA in a pristine form because they need not be subjected to freezing or fixation. The first successful execution of this strategy was carried out by Zhang et al. [35], who expressed green fluorescent protein (GFP) in the six touch-receptor neurons of the nematode C. elegans; they purified these fluorescent cells from thousands of dissociated embryos on an automated fluorescence-activated cell sorter (FACS) after culturing them overnight to increase GFP expression, and finally extracted and amplified mRNA for microarray analysis. Subsequent implementations of this method in C. elegans have been aimed at identifying genes specifically expressed in other populations of neurons, including olfactory and thermosensory neurons [36] and GABAergic motoneurons [37,38]. Similar progress has been made in cell-type-specific expression profiling in the mammalian nervous system. Buchstaller et al. [39] FACS-sorted and profiled GFPexpressing neural crest stem cells at several developmental time points as they differentiated into Schwann cells, revealing hundreds of genes likely to be involved in this process. Arlotta et al. [32] published the first genomewide expression profiles from specific neuronal subpopulations in the mammalian CNS using live-cell sorting. Cortical projection neurons were retrogradely labeled by injection of tracer into contralateral cortex or spinal cord. These were FACS-sorted and profiled at four different developmental time points: embryonic day 18, and postnatal day (P)3, P6 and P14. In addition, corticotectal neurons were profiled at P14. Hundreds of genes were found to be differentially expressed between these closely related cell types. These genes spanned a wide range of functional categories and most of these had not previously been associated with cortical function, demonstrating the ability of this approach to discover entirely novel gene functions. A subset of these genes was highlighted as having expression trends that suggest involvement in developmental specification of the profiled cells. One of these genes, encoding the transcription factor Ctip2, was found to be crucial for the development of corticospinal axons. A second transcription factor gene, Fezl, was found to be required for the specification of corticofugal motor axons [40,41]; these projections seem to be absent from FezlK/K mouse cortices. These studies demonstrate how the wealth of data produced by neuronal cell-type-specific microarray studies can be used to generate novel, specific and testable hypotheses about gene function in neurons. Sugino et al. [27] used transgenic fluorescent-proteinexpressing mice to profile particular cell types in the mammalian forebrain. These mouse lines were gathered from various laboratories [42–45]. Through a combination of promoter specificity in the transgene construct and positional effects determined by the transgene genomic insertion point, each line expresses fluorescent protein in a highly restricted subset of neurons. The manual sorting method employed to purify these labeled populations offers two key advantages over automated FACS sorting. First, it is a gentle procedure that enables purification of
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
343
types from different regions of the neocortex were found to be nearly identical, supporting the ‘canonical microcircuit’ hypothesis of cortical architecture. This study showed that genome-wide expression profiles can be used to classify neurons based on objective criteria that can be applied to any labeled cell type in the nervous system. This sets the stage for the profiling and unbiased classification of a much larger number of cell types in the cortex and throughout the brain.
adult neurons. Second, it does not require thousands of neurons, so limited populations can be readily purified. For example, a single, rare interneuron subtype from a single cortical region can be isolated. However, the flip side of this is that the yield of manual sorting is low (typically 30–100 neurons). Fortunately, the small sample amplification procedure is reliable in this range. In contrast to previous studies, which had compared expression in two or three neuronal cell types, Sugino et al. [27] profiled 12 distinct populations of mouse neurons chosen to cover a wide range of cell types from different forebrain regions, with different neurotransmitter phenotypes, morphologies and electrophysiological characteristics (Figure 2a). This screen provided a wealth of novel candidate molecular neuronal markers and, as we will discuss later, identified particular functional categories of genes over-represented in differentially expressed genes. More importantly, this was the first study to use microarrays to address the problem of neuronal classification in the mammalian nervous system. A taxonomic dendrogram was generated based on the expression data (Figure 2b). The position on the dendrogram reflects the relative distance in ‘expression space’ between cell types. The taxonomy captured the expected major relationships between the profiled populations, such as the distinction between GABAergic and glutamatergic neurons, and also unexpected relationships between some lower-order subtypes. Homologous cell
Expression modules Sugino et al. [27] identified sets of genes that distinguish cell types and then used the Gene Ontology database (http://www.geneontology.org/) [46] to identify biological functions known to be subserved by these genes. For example, different cell types expressed different sets of ion-channel subunits (presumably reflecting differences in their firing patterns) and different sets of cytoskeletal proteins (presumably reflecting differences in their morphologies). It seems likely that, just as cellular phenotypes (e.g. fast-spiking or stellate morphology) are repeated across cell types in disparate regions, so there might be expression modules (sets of co-regulated genes) that give rise to these phenotypes. Just as proteins are described as collections of unique and common domains or motifs, cell types might be best described as collections of expression modules, each module providing a particular function (e.g. transmitter phenotype, axon-targeting or
(a)
(b) Hippocampus
Cingulate cortex
GIN-HP
YFPH-CG
CT6-CG
YFPH-S1
YFPH-AM
YFPH-HP
YFPH-S1
YFPH-CG
CT6-CG
G30-AM
G30-S1
G30-CG
G42-CG
GIN-CG
G30-S1
GIN-HP
G42-LG
GIN-CG
YFPH-HP
Somatosensory cortex G30-CG
Amygdala YFPH-AM G42-CG
G30-AM
G42-LG
Thalamus Figure 2. Cell-type-specific expression profiles for neuronal taxonomy. Adapted from Ref. [27]. (a) Twelve neuronal populations profiled by Sugino et al. [27] are named according to transgenic strain (YFPH [45], GIN [44], G30 [43] or G42 [42]) or projection-neuron class retrogradely labeled by tracer injection (layer 6 corticothalamic neurons, CT6), and according to brain region [amygdala (AM), cingulate cortex (CG), hippocampus (HP), lateral geniculate nucleus (LG) or primary somatosensory cortex (S1)]. For example, YFPH-CG is from strain YFPH in cingulate cortex. A low-power micrograph taken at the locations marked on the atlas sections shows selective labeling of the populations by either green or yellow fluorescent protein or red fluorescent microspheres. A representative electrophysiological recording is also shown for each population. (b) Dendrogram reflecting distance in ‘gene expression space’ between the twelve populations. This was generated by hierarchical clustering according to relative differences in genome-wide expression profiles. Each point represents data from one microarray and each population was profiled in triplicate. www.sciencedirect.com
344
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
spiking properties). The partial independence of these modules might explain the observation that in cortical interneurons, for example, the same firing type can be associated with multiple morphologies, and vice versa [14]. Identifying the molecular fingerprint of a particular cell type will also enable investigators to ask comparative questions – how conserved are cell types between species? Cajal initially thought that the cortex of humans contained many cell types not found in ‘lower’ mammals, but his student Lorente de No argued forcefully that even the lowly mouse contained a comparable number and complexity of cell types [47]. Expression distance, or perhaps more refined versions of this measure based on similarity of expression modules, might provide an objective, quantitative metric on which to make such comparisons. After defining sets of markers for cell types in the mouse, it might be feasible to address this in other species such as monkeys and humans using a combination of retrograde labeling and live sorting, or of antibody labeling and LCM. Such comparisons are crucial if we are to relate the functional analyses of circuits in cats, monkeys and other species to cellular and molecular work in rodents. In addition, with the sequencing of multiple mammalian genomes, including those of mouse, rat, dog, ape, human and other related vertebrate species, a new field of comparative neurogenomics is likely to emerge, driven by the search for genomic differences that regulate species differences in brain structure and function. These genomic differences are likely to be not only in the coding sequences of individual genes but also in the extragenic regulatory sequences that control their expression.
A mammalian neurome project? Although the number of distinct neuronal cell types in the mammalian ‘neurome’ (103–104) is likely to be within an order of magnitude of the number of genes in the mammalian genome (2–4!104), it seems unlikely that single genes specify, or can serve as markers for, single cell types. For example, although parvalbumin expression is often considered the sine qua non of fast-spiking basket and chandelier cells, not all basket cells express parvalbumin [14], and parvalbumin is also expressed in the auditory brainstem [48], somatosensory layer 5 cortical pyramidal neurons [49] and many other structures. Ideal markers (i.e. that label one cell type and no others) are relatively abundant when comparing a small number of cell types, but are likely to be rare when comparing across all cell types. A more complete picture of the pattern of expression within the brain of all genes within the mouse genome is likely to emerge shortly from large-scale mapping efforts such as the Allen Institute Brain Atlas (http://www. brainatlas.org/) [50,51] and Gensat (http://www.gensat. org/index.html) [52]. But this will not, by itself, enable a complete map of brain neuronal cell types to be assembled. Cell types are likely to be defined by coexpression of many genes and in most cases it will be hard to discern precise patterns of coexpression with cellular resolution by www.sciencedirect.com
comparing hybridization or reporter expression patterns across multiple tissue sections. We suggest that a systematic attempt to map neuronal cell types is now feasible and represents an important next logical step in post-genomic neuroscience. Several genetic strategies are available to create a much larger library of fluorescently labeled cell types in mice. These include bacterial artificial chromosome (BAC) transgenics [52], BAC recombination to knock-in to endogenous loci [53], and enhancer-trap strategies using lentiviral vectors [54]. The loci to use for knock-in approaches might be identified based on restricted spatial expression patterns in the Allen Atlas or Gensat databases, or from profiling studies such as those we have already described. Enhancer traps use random insertion of a minimal promoter that drives celltype-specific reporter expression only when it inserts next to regulatory elements that direct expression in those cell types. None of these approaches is likely to label only a single cell type within the brain. Instead, careful physiological, anatomical and genomic characterization will be needed to distinguish the multiple cell types labeled in each structure within each line. The importance of the labeling is that it enables multiple investigators to query the same cell types repeatedly and therefore to share reliable data about their phenotypic and genotypic properties. To be most useful, given the large characterization effort required, these new genetic resources should also permit cell-type-specific manipulation of gene expression. Unlike the majority of transgenics produced so far, future efforts are likely to focus increasingly on expressing both fluorescent proteins for cell identification and, in the same cells, additional molecules such as Cre-recombinase to permit cell-type-specific manipulation of gene expression. Such manipulation will be crucial in elucidating the contribution both of the gene to the cell type and of the cell type to the circuit. Recognizing the importance of such approaches, the NIH has recently issued a call for proposals to generate transgenic ‘driver’ lines (RFA Number MH-06–007; http://grants.nih.gov/grants/guide/ rfa-files/RFA-MH-06-007.html). With the increasing availability of such tools the study of mammalian brain circuits enters a new phase, in which individual neuronal cell types can be easily and repeatedly identified and in which their function can be selectively and flexibly altered. References 1 Monyer, H. and Markram, H. (2004) Interneuron Diversity series: Molecular and genetic tools to study GABAergic interneuron diversity and function. Trends Neurosci. 27, 90–97 2 Mott, D.D. and Dingledine, R. (2003) Interneuron Diversity series: Interneuron research – challenges and strategies. Trends Neurosci. 26, 484–488 3 Koester, S.E. and O’Leary, D.D. (1992) Functional classes of cortical projection neurons develop dendritic distinctions by class-specific sculpting of an early common pattern. J. Neurosci. 12, 1382–1393 4 Zhang, Z.W. and Deschenes, M. (1997) Intracortical axonal projections of lamina VI cells of the primary somatosensory cortex in the rat: a single-cell labeling study. J. Neurosci. 17, 6365–6379 5 Kasper, E.M. et al. (1994) Pyramidal neurons in layer 5 of the rat visual cortex. I. Correlation among cell morphology, intrinsic electrophysiological properties, and axon targets. J. Comp. Neurol. 339, 459–474 6 Masland, R.H. (2001) Neuronal diversity in the retina. Curr. Opin. Neurobiol. 11, 431–436
Review
TRENDS in Neurosciences Vol.29 No.6 June 2006
7 Masland, R.H. (2004) Neuronal cell types. Curr. Biol. 14, R497–R500 8 Koch, C. (2004) The Quest for Consciousness: A Neurobiological Approach, Roberts and Company Publishers 9 Stevens, C.F. (1998) Neuronal diversity: too many cell types for comfort? Curr. Biol. 8, R708–R710 10 Parra, P. et al. (1998) How many subtypes of inhibitory cells in the hippocampus? Neuron 20, 983–993 11 Woese, C.R. and Fox, G.E. (1977) Phylogenetic structure of the prokaryotic domain: the primary kingdoms. Proc. Natl. Acad. Sci. U. S. A. 74, 5088–5090 12 Zuckerkandl, E. and Pauling, L. (1965) Molecules as documents of evolutionary history. J. Theor. Biol. 8, 357–366 13 Nelson, S. (2002) Cortical microcircuits: diverse or canonical? Neuron 36, 19–27 14 Markram, H. et al. (2004) Interneurons of the neocortical inhibitory system. Nat. Rev. Neurosci. 5, 793–807 15 Mirnics, K. et al. (2000) Molecular characterization of schizophrenia viewed by microarray analysis of gene expression in prefrontal cortex. Neuron 28, 53–67 16 Sandberg, R. et al. (2000) Regional and strain-specific gene expression mapping in the adult mouse brain. Proc. Natl. Acad. Sci. U. S. A. 97, 11038–11043 17 Xie, T. et al. (2002) Changes in gene expression linked to methamphetamine-induced dopaminergic neurotoxicity. J. Neurosci. 22, 274–283 18 Zhao, X. et al. (2001) Transcriptional profiling reveals strict boundaries between hippocampal subregions. J. Comp. Neurol. 441, 187–196 19 Zirlinger, M. and Anderson, D. (2003) Molecular dissection of the amygdala and its relevance to autism. Genes Brain Behav. 2, 282–294 20 Eberwine, J. et al. (1992) Analysis of gene expression in single live neurons. Proc. Natl. Acad. Sci. U. S. A. 89, 3010–3014 21 Hinkle, D. et al. (2004) Single neurons as experimental systems in molecular biology. Prog. Neurobiol. 72, 129–142 22 Kamme, F. et al. (2003) Single-cell microarray analysis in hippocampus CA1: demonstration and validation of cellular heterogeneity. J. Neurosci. 23, 3607–3615 23 Tietjen, I. et al. (2003) Single-cell transcriptional analysis of neuronal progenitors. Neuron 38, 161–175 24 Zirlinger, M. et al. (2001) Amygdala-enriched genes identified by microarray technology are restricted to specific amygdaloid subnuclei. Proc. Natl. Acad. Sci. U. S. A. 98, 5270–5275 25 Cauli, B. et al. (2000) Classification of fusiform neocortical interneurons based on unsupervised clustering. Proc. Natl. Acad. Sci. U. S. A. 97, 6144–6149 26 Toledo-Rodriguez, M. et al. (2004) Correlation maps allow neuronal electrical properties to be predicted from single-cell gene expression profiles in rat neocortex. Cereb. Cortex 14, 1310–1327 27 Sugino, K. et al. (2006) Molecular taxonomy of major neuronal classes in the adult mouse forebrain. Nat. Neurosci. 9, 99–107 28 Backes, E. and Hemby, S.E. (2003) Discrete cell gene profiling of ventral tegmental dopamine neurons after acute and chronic cocaine self-administration. J. Pharmacol. Exp. Ther. 307, 450–459 29 Chung, C.Y. et al. (2005) Cell type-specific gene expression of midbrain dopaminergic neurons reveals molecules involved in their vulnerability and protection. Hum. Mol. Genet. 14, 1709–1725 30 Greene, J.G. et al. (2005) Gene expression profiling of rat midbrain dopamine neurons: implications for selective vulnerability in parkinsonism. Neurobiol. Dis. 18, 19–31 31 Yao, F. et al. (2005) Microarray analysis of fluoro-gold labeled rat dopamine neurons harvested by laser capture microdissection. J. Neurosci. Methods 143, 95–106
www.sciencedirect.com
345
32 Arlotta, P. et al. (2005) Neuronal subtype-specific genes that control corticospinal motor neuron development in vivo. Neuron 45, 207–221 33 Kunitomo, H. et al. (2005) Identification of ciliated sensory neuronexpressed genes in Caenorhabditis elegans using targeted pull-down of poly(A) tails. Genome Biol. 6, R17 34 Yang, Z. et al. (2005) Isolation of mRNA from specific tissues of Drosophila by mRNA tagging. Nucleic Acids Res. 33, e148 35 Zhang, Y. et al. (2002) Identification of genes expressed in C. elegans touch receptor neurons. Nature 418, 331–335 36 Colosimo, M.E. et al. (2004) Identification of thermosensory and olfactory neuron-specific genes via expression profiling of single neuron types. Curr. Biol. 14, 2245–2251 37 Fox, R.M. et al. (2005) A gene expression fingerprint of C. elegans embryonic motor neurons. BMC Genomics 6, 42 38 Cinar, H. et al. (2005) Expression profiling of GABAergic motor neurons in Caenorhabditis elegans. Curr. Biol. 15, 340–346 39 Buchstaller, J. et al. (2004) Efficient isolation and gene expression profiling of small numbers of neural crest stem cells and developing Schwann cells. J. Neurosci. 24, 2357–2365 40 Molyneaux, B.J. et al. (2005) Fezl is required for the birth and specification of corticospinal motor neurons. Neuron 47, 817–831 41 Chen, B. et al. (2005) Fezl regulates the differentiation and axon targeting of layer 5 subcortical projection neurons in cerebral cortex. Proc. Natl. Acad. Sci. U. S. A. 102, 17184–17189 42 Ango, F. et al. (2004) Ankyrin-based subcellular gradient of neurofascin, an immunoglobulin family protein, directs GABAergic innervation at Purkinje axon initial segment. Cell 119, 257–272 43 Lopez-Bendito, G. et al. (2004) Preferential origin and layer destination of GAD65-GFP cortical interneurons. Cereb. Cortex 14, 1122–1133 44 Oliva, A.A., Jr. et al. (2000) Novel hippocampal interneuronal subtypes identified using transgenic mice that express green fluorescent protein in GABAergic interneurons. J. Neurosci. 20, 3354–3368 45 Feng, G. et al. (2000) Imaging neuronal subsets in transgenic mice expressing multiple spectral variants of GFP. Neuron 28, 41–51 46 Ashburner, M. et al. (2000) Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. 25, 25–29 47 Lorente de No, R. (1992) The cerebral cortex of the mouse (a first contribution – the ‘acoustic’ cortex). Somatosens. Mot. Res. 9, 3–36 48 Felmy, F. and Schneggenburger, R. (2004) Developmental expression of the Ca2C-binding proteins calretinin and parvalbumin at the calyx of held of rats and mice. Eur. J. Neurosci. 20, 1473–1482 49 Preuss, T.M. and Kaas, J.H. (1996) Parvalbumin-like immunoreactivity of layer V pyramidal cells in the motor and somatosensory cortex of adult primates. Brain Res. 712, 353–357 50 Gewin, V. (2005) A golden age of brain exploration. PLoS Biol. 3, e24 51 Boguski, M.S. and Jones, A.R. (2004) Neurogenomics: at the intersection of neurobiology and genome sciences. Nat. Neurosci. 7, 429–433 52 Gong, S. et al. (2003) A gene expression atlas of the central nervous system based on bacterial artificial chromosomes. Nature 425, 917–925 53 Cotta-de-Almeida, V. et al. (2003) A new method for rapidly generating gene-targeting vectors by engineering BACs through homologous recombination in bacteria. Genome Res. 13, 2190–2194 54 Lois, C. et al. (2002) Germline transmission and tissue-specific expression of transgenes delivered by lentiviral vectors. Science 295, 868–872 55 Blackshaw, S. et al. (2001) Comprehensive analysis of photoreceptor gene expression and the identification of candidate retinal disease genes. Cell 107, 579–589