Finding a Partner in a Crowd: Neuronal Diversity and Synaptogenesis

Abstract

Over 125 years ago, Camillo Golgi published a technique for silver-staining neurons that retains his name and is still used to this day. Because only relatively few, widely separated neurons in a sample of cell-dense neural tissue react, the Golgi technique was important in enabling biologists to discover an incredible morphological diversity of neurons. Many types of neurons were described using the technique, including the sole output neurons of the cerebellar cortex, Purkinje cells. Santiago Ramón y Cajal, studying the neuronal input to these Purkinje cells at the turn of the century, discovered that early in development, these neurons were innervated primarily on their cell bodies by hindbrain-derived axons. He found that later in development, these axons break these synaptic connections and “climb” the dendritic tree to form new connections elsewhere on the same Purkinje cell. From this reconfiguration of connectivity, Ramón y Cajal concluded that synapses in the nervous system were not made at random; instead, synaptic connections must be specified if they could be unmade and remade in so regulated a fashion (9Ramón y Cajal S Histology of the Nervous System, Vol. 2, L.W. Swanson and N. Swanson, trans. Oxford University Press, New York1995Google Scholar). Two questions arise from Ramón y Cajal's observation that synaptic connections are specified. First, how many different types of neurons need to be specifically connected? Second, what is the molecular basis of this synaptic specificity? Recent work, described below, suggests that the number of neuronal types is very large and could be as high as several hundred in the cerebral cortex alone. Fortunately, as discussed below, the discovery that members of the cadherin family of cell-adhesion proteins are present at synapses has given us insight into a possible molecular basis for synaptic specificity among so many cell types. In particular, the protocadherin members of this family have been found to have a scope of potential variation and an immunoglobulin-like arrangement of their genes that makes them excellent candidate molecules for ensuring that neuronal diversity is utilized in specific, selective synaptic connections. How many different neuronal types exist? In an attempt to quantitate the diversity in a single class of neurons5MacNeil M.A Masland R.H Neuron. 1998; 20: 971-982Abstract Full Text Full Text PDF PubMed Scopus (374) Google Scholar undertook to describe the amacrine cell population of the mammalian retina. Amacrine cells are vital to the processing of visual information by the retina, as these cells directly influence retinal ganglion cells, which are the output cells of retina. Because the retina is a highly ordered laminar structure, the shape and extent of an amacrine cell's dendritic arbor determine its connections with other neurons, thus determining the manner in which it can influence ganglion cell responses. By selecting and detailing the dendritic shape and stratification of 261 randomly chosen amacrine cells (using a new method they developed, termed “photofilling”), MacNeil and Masland were able to describe at least 26 different types of this neuron, including the four major types previously described (Figure 1). They found that most of the cell types “tiled” the retina very efficiently; namely, the amacrine cells of a given class populated the retina at a density such that their dendritic arbors filled the retina with little overlap between adjacent arbors. This indicates that the authors very likely did not “invent” cell types and that the description they present is likely to be a very good estimate of the actual diversity present. Tiling of a brain region by a given neuronal type makes sense teleologically: One assumes that each particular type of neuron contributes to information processing in a distinct way, and therefore all other neuronal types should have similar access to at least one member of each distinct class. Tiling would accomplish this by filling a brain region with minimal redundancy. 13Stevens C.F Curr. Biol. 1998; 8: R708-R710Abstract Full Text Full Text PDF PubMed Google Scholar has argued that, assuming tiling by individual neuronal types is a general organizational principle, an upper limit on the possible number of neuronal types within a brain region such as the neocortex can be calculated. Given the density of neurons in the cortex (105 neurons/mm2), the average dendritic spread of each neuron (0.05 mm2), and a generous estimate of 10-fold redundancy in the neocortex relative to the streamlined retina (i.e., ten times as many neurons of a given type than is necessary for 1× coverage), the total number of individual neuronal types in the neocortex is calculated to be around 500. A recent study has indeed indicated that the degree of cortical neuronal diversity could be that high. 8Parra P Gulyás A.I Miles R Neuron. 1998; 20: 983-993Abstract Full Text Full Text PDF PubMed Scopus (284) Google Scholar attempted to quantitate the number of different inhibitory interneurons present in the CA1 area of the hippocampal cortex. Inhibitory interneurons comprise only 10%–20% of the total number of neurons in the cerebral cortex, but they had been previously shown to be very diverse. Parra et al. first classified these neurons on the basis of somatic location, dendritic orientation, and the precise innervation zones on target neurons chosen by their axons. Then, action potential firing patterns and sensitivity to modulatory neurotransmitters were measured in an attempt to provide additional criteria for classification. Most of these criteria are very germane to the issue of neuronal diversity, since such differences should be important in how a given neuron functions within brain circuitry. The authors were careful to focus on large or categorical differences among neurons as the basis of their classification. The analysis they performed had a striking outcome: Every neuron they looked at was different. They found sixteen different morphological types, three different firing modes, and 25 different combinations of neurotransmitter receptors, and every cell was unique across these criteria. Either the number of neuronal types is very high (at least 52 different types estimated from all of their data), or the authors conclude, every neuron is unique with respect to seemingly important functional differences. A key question raised by both of these studies is, of course, what exactly constitutes a neuronal “type”? Most would probably agree that a unique function or role within a circuit is a basis for discrimination. Because many differences in the role of a neuron (due, one would expect, to differences in morphology, axon targeting, or physiology) are likely to be the result of, or heavily influenced by, past or current variation in gene expression, one way to answer this question initially would be to use gene expression analysis to delineate different classes of neurons. The success of such an approach is amply illustrated in the work of Jessell and colleagues, who have been able to match differences in transcription factor gene expression (LIM and ETS families) to functional distinctions in the wiring of spinal motoneurons and their sensory afferents (18Tsuchida T Ensini M Morton S.B Baldassare M Edlund T Jessell T.M Pfaff S.L Cell. 1994; 79: 957-970Abstract Full Text PDF PubMed Scopus (857) Google Scholar, 4Lin J.H Saito T Anderson D.J Lance-Jones C Jessell T.M Arber S Cell. 1998; 95: 393-407Abstract Full Text Full Text PDF PubMed Scopus (276) Google Scholar). Attempts to analyze neuronal diversity by examining gene expression at a genome-wide scale are in their infancy, but they are likely eventually to provide a basis for the systematic investigation of neuronal cell type. That so much diversity exists begs the question of how different neuronal types recognize one another and link up. Different neuronal types need not synapse in different ways in order to remain useful to the nervous system (by different ways, one could mean, for example, different partners, different subcellular locations, different synaptic sizes, or different possibilities for activity-dependent modulation). However, sufficient heterogeneity in synaptogenesis has been found to suggest some differences among different cell types. Although one can easily envision mechanisms leading to postsynaptic partner or even subcellular specification that rely upon such phenomena as axon guidance or the relative timing of neurogenesis, the possible existence of molecular labels that lead to recognition and synaptogenesis between pre- and postsynaptic neurons has been an intriguing hypothesis for decades (the chemoaffinity hypothesis; 12Sperry R.W Proc. Natl. Acad. Sci. USA. 1963; 50: 703-710Crossref PubMed Scopus (1348) Google Scholar). Although both odorant and pheromone receptors and neurexins have the necessary molecular diversity, cadherin family proteins are possibly the best candidates to provide labels for neurons during synaptogenesis. The recent findings that cadherins are localized at synapses have provided an attractive model consistent with the chemoaffinity hypothesis (reviewed in 11Serafini T Trends Neurosci. 1997; 20: 322-323Abstract Full Text PDF PubMed Scopus (26) Google Scholar). Cadherins were discovered as the proteins on vertebrate cell surfaces that mediate Ca2+-dependent cell–cell adhesion and are the main adhesive protein associated with the actin-based zonula adherens junctions in epithelial cells (reviewed in 20Yap A.S Brieher W.M Gumbiner B.M Annu. Rev. Cell Dev. Biol. 1997; 13: 119-146Crossref PubMed Scopus (675) Google Scholar). While the finding that cadherins are at synapses suggests that these structures are specialized versions of adhesive junctions found elsewhere, the characteristics of cadherins also make these molecules especially compelling as mediators of synaptic specificity: They display (mostly) homophilic binding preferences, many different cadherins exist (roughly 15 “classic” cadherins), and different cadherins have been found to segregate into different synapses on the same neuron. In fact, recent work has shown that the same cadherins are expressed by neurons comprising multiple-neuron circuits (for example, see 14Suzuki S.C Inoue T Kimura Y Tanaka T Takeichi M Mol. Cell. Neurosci. 1997; 9: 433-447Crossref PubMed Scopus (242) Google Scholar). This said, however, no direct evidence has yet been published detailing a role for cadherins in synapse formation. The only evidence that cadherins play any role at synapses comes from studies in which interfering with cadherin function in vitro led to the elimination of long-term potentiation (LTP), a form of synaptic plasticity, in hippocampal slice cultures (16Tang L Hung C.P Schuman E.M Neuron. 1998; 20: 1165-1175Abstract Full Text Full Text PDF PubMed Scopus (291) Google Scholar). Cadherins are single-pass transmembrane proteins roughly 750 aa in length that are comprised of five similar extracellular (EC) domains (110 aa each), a transmembrane domain, and a conserved cytoplasmic domain that interacts with the catenins, proteins that mediate the association of cadherins with the actin cytoskeleton, and which are required for mediating cadherin-based cell–cell adhesion. The specificity-determining region has been localized to the most N-terminal EC (EC1) domain for at least two different cadherins. Cadherins are thought to exist as homodimers and perhaps monomers, with these different forms postulated to be responsible both for the observation that cadherins can support axon extension and for their hypothesized role in synaptic adhesion (15Tamura K Shan W.S Hendrickson W.A Colman D.R Shapiro L Neuron. 1998; 20: 1153-1163Abstract Full Text Full Text PDF PubMed Scopus (267) Google Scholar). They are also the founding members of a family of related proteins, which include such members as desmosomal cadherins, the product of the fat gene of Drosophila, and protocadherins. Different family members can be distinguished from one another by a variable number of EC repeats, landmarks within these repeats, and by distinct cytoplasmic portions. While the first protocadherins had been identified some time ago (10Sano K Tanihara H Heimark R.L Obata S Davidson M St John T Taketani S Suzuki S EMBO J. 1993; 12: 2249-2256Crossref PubMed Scopus (318) Google Scholar), recent work has not only greatly enlarged this family but also placed some protocadherins at synapses along with cadherins. A novel family of genes in the mouse, the cadherin-related neuronal receptors or CNRs (23% identity in the extracellular domain, on average, with the “classic” cadherins), was found to encode protocadherins that resided at synapses (3Kohmura N Senzaki K Hamada S Kai N Yasuda R Watanabe M Ishii H Yasuda M Mishina M Yagi T Neuron. 1998; 20: 1137-1151Abstract Full Text Full Text PDF PubMed Scopus (376) Google Scholar). These proteins possess six EC domains and a cytoplasmic C terminus distinct from that of the classic cadherins and were found through a two-hybrid screen for proteins that interact with the nonreceptor tyrosine kinase Fyn. Remarkably, the 3′ portions of the CNR cDNAs are identical at the nucleotide level, which suggested to Kohmura, et al., that a diversity of 5′ exons commonly combined with a smaller number of 3′ exons in the genome to generate CNR mRNAs. Suzuki and colleagues also noticed this as well in the sequences of other protocadherins (7Obata S Sago H Mori N Davidson M St John T Suzuki S.T Cell Adhes. Commun. 1998; 6: 323-333Crossref PubMed Scopus (36) Google Scholar). However, work reported in Cell by Wu and Maniatis is notable for uncovering the scope of variation and for presenting a unified view of the different protocadherins in humans as the products of three highly variable, similarly organized genes (Pcdhα [human CNRs], Pcdhβ, Pcdhγ) (19Wu Q Maniatis T Cell. 1999; 97: 779-790Abstract Full Text Full Text PDF PubMed Scopus (519) Google Scholar). A rush of adrenaline is sure to accompany the viewing of several figures from their paper, in which the relationships among the three different human protocadherin genes and their protein products are displayed. The numbers: three genes, each with nearly twenty 5′ exons (52 total), each combined uniquely with constant region exons to create a final mRNA. Such an organization is immediately reminiscent of that of the highly variable immunoglobulin and T cell receptor genes involved in immune system function. As has been observed for cadherins, a homophilic binding preference has been observed for at least one protocadherin (6Obata S Sago H Mori N Rochelle J.M Seldin M.F Davidson M St John T Taketani S Suzuki S.T J. Cell. Sci. 1995; 108: 3765-3773Crossref PubMed Google Scholar). However, nearly identical EC1 domain sequences among some protocadherins suggest that these proteins might be capable of heterophilic, and perhaps graded, interactions. Regardless of whether or not heterophilic binding exists, if multiple exons are selectively utilized in an individual neuron from among such a large number of possibilities, the protocadherins might, especially in combination with the classic cadherins, provide sufficient variation in binding capabilities such that large numbers of different neuronal types within discrete brain regions have the ability to link up selectively with other neurons. Indeed, different neurons within the olfactory bulb were observed to express different CNRs (3Kohmura N Senzaki K Hamada S Kai N Yasuda R Watanabe M Ishii H Yasuda M Mishina M Yagi T Neuron. 1998; 20: 1137-1151Abstract Full Text Full Text PDF PubMed Scopus (376) Google Scholar), suggesting that studying the utilization of the variable 5′ exons might provide a framework for addressing the larger issue of neuronal diversity. How these variable 5′ exons are actually matched up with the constant region exons is bound to tell us much about basic mechanisms of generating molecular and cellular diversity in the nervous system. While a range of possibilities for generating the diverse mRNAs was cited by Suzuki and colleagues and Wu and Maniatis (differential promoter utilization, differential splicing, etc.), the most intriguing possibility they raise is that of a somatic rearrangement in the DNA itself, as is undergone by immunoglobulin and T cell receptor genes. Particularly interesting in this light is the recent finding that mice lacking functional XCRR4 and LigIV genes, which encode proteins required to repair double-stranded DNA breaks (such as those introduced during gene rearrangements in the immune system) are inviable, largely due to massive apoptosis in newly minted, postmitotic neurons (2Gao Y Sun Y Frank K.M Dikkes P Fujiwara Y Seidl K.J Sekiguchi J.M Rathbun G.A Swat W Wang J et al.Cell. 1998; 95: 891-902Abstract Full Text Full Text PDF PubMed Scopus (562) Google Scholar). One can easily imagine a scenario in which neurons that fail to properly rearrange their protocadherin genes are eliminated, since these neurons might form synapses inappropriately, compromising the function of neural circuitry. However, unlike the situation in the immune system, where somatic rearrangement and allelic exclusion lead to a single type of immunoglobulin or T cell receptor being expressed in an individual cell, multiple versions of CNRs are expressed by individual neurons. This observation suggests that DNA rearrangement, if it exists, might not necessarily lead to single protocadherin expression in a particular neuron, although it still might limit possibilites. Of course, we know even less about the roles of the protocadherins at synapses than we do about cadherins (although an analysis of protocadherin function during embryogenesis has indicated a role distinct from that of cadherins; 1Bradley R.S Espeseth A Kintner C Curr. Biol. 1998; 8: 325-334Abstract Full Text Full Text PDF PubMed Scopus (81) Google Scholar). Some obvious questions include: Are they expressed at the right time to play a role in synapse formation? Are multiple types expressed by individual neurons in a manner that correlates with those of synaptic partners? Are they necessary for synapse formation? Are they necessary for directing synaptic specificity? Are these genes only capable of fostering adhesive interactions, or might they also be capable of mediating repulsive as well as attractive interactions? As a case in point, Eph tyrosine kinase receptors and their ligands, the ephrins, are normally thought to mediate repulsive interactions during axon guidance, but they have recently been found clustered at synapses (17Torres R Firestein B.L Dong H Staudinger J Olson E.N Huganir R.L Bredt D.S Gale N.W Yancopoulos G.D Neuron. 1998; 21: 1453-1463Abstract Full Text Full Text PDF PubMed Scopus (392) Google Scholar). With over a third of the genome expressed in the brain, perhaps a large number of different neuronal types shouldn't be too surprising. Yet, such cellular diversity could very well be matched by a molecular diversity at the cell surface involved in specifying connections. The party may be large, and the house may be crowded, but everyone just might be wearing a name tag.