It is very difficult to avoid superlatives when describing recent progress in the study of the molecular mechanisms regulating cell differentiation in the nervous system in general and in the retina in particular (reviewed by Jean et al., 1998; Adler, 2000, Lupo et al., 2000; Galli-Resta, 2001; Livesey and Cepko, 2001, Vetter and Brown, 2001; Marquardt and Gruss, 2002; Zhang et al., 2002a; Boulton and Albon, 2004; Hatakeyama and Kageyama, 2004, Malicki, 2004; Mu and Klein, 2004). There can be little doubt, moreover, that this field will continue to advance at an extremely fast pace. Its remarkable success and bright future should not distract us, however, from recognizing and addressing the important challenges that still lie ahead of us, and particularly those that are unlikely to be solved without the development of new methodologies. The goal of this article is not to provide a comprehensive review of the literature but, rather, to describe some of these challenges using examples from studies of retinal cell differentiation, with which the author's laboratory has direct experience. Studies in invertebrates, although relevant and important (reviewed in Frankfort and Mardon, 2002; Wernet and Desplan, 2004; Yang, 2004; Mollereau and Domingos, 2005) will not be covered due to space limitations. Most of the issues that will be considered, however, are equally relevant to studies of neuronal differentiation in other regions of the vertebrate nervous system.
INVESTIGATING CELL COMMITMENT
Operational Definition of Cell Commitment
At early stages of embryonic development, the neural retina consists of a population of morphologically homogeneous and mitotically active neuroepithelial cells. Early neuroepithelial cells have been shown to be multipotential, i.e., to retain the capacity to give rise to two or more types of differentiated retinal cells (Holt et al., 1988; Wetts and Fraser, 1988; Cepko, 1993; Harris, 1997). Considerable (and increasing) heterogeneity arises within the retina as neuroepithelial cells become postmitotic and begin their migration toward their future position in one of the retinal layers. This process continues for a period of days to weeks, depending on the species, with extensive overlap in the generation of different cell types (Barnstable et al., 1988; Rapaport et al., 1996, 2004; Harris, 1997; Adler, 2000; Mey and Thanos, 2000; Malicki, 2004). The differentiation of the cells thus generated also starts, and advances to various degrees, during this period. Cells undergoing proliferation, terminal mitosis (“cell birth”), migration, and differentiation overlap with each other in time and space. As discussed below, the heterogeneous and dynamic nature of the embryonic populations thus generated creates many challenges for the design of experiments aimed at investigating the molecular regulation of these developmental mechanisms.
It is generally accepted that, at some stage during its developmental history, each multipotential progenitor becomes “committed” to a specific differentiated fate. Cell commitment still is defined operationally, through experiments in which the microenvironment of the cells is altered by pharmacological treatments and/or by cell and tissue transplantation, recombination, or culture. Multipotential progenitors are defined as those that change their differentiated fate in response to changes in their microenvironment, whereas committed progenitors are those that always follow the same developmental pathway, regardless of their microenvironment. These issues have been investigated in different laboratories with a variety of approaches, which have been the subject of excellent recent reviews (Cepko et al., 1996; Reh and Levine, 1998; Fuhrmann et al., 2000; Levine et al., 2000; Perron and Harris, 2000a, b; Cepko, 2001; Layer et al., 2001; Livesey and Cepko, 2001; Zhang et al., 2002b; Martinez-Morales et al., 2004). It would be beyond the scope of this article to provide an additional comprehensive overview of this literature; rather, our own experience using a cell culture approach will be used to illustrate the limits of phenomenological, operational definitions of cell commitment.
An example of a cell culture experiment testing cell commitment is illustrated in Figure 1. In this case, cells dissociated from the chick embryo retina before the onset of overt differentiation were grown in low density culture, in the absence of contact-mediated intercellular interactions. The goal of the experiment was to test whether, in this homogeneous but artificial microenvironment, the undifferentiated cells would: (1) differentiate, (2) follow divergent developmental pathway(s), and (3) express phenotypic properties similar to those of cells that differentiate within the retina in vivo. Figure 1A illustrates the morphological homogeneity observed at culture onset. As shown in Figure 1B, many of the cells did indeed differentiate after several days in vitro and did follow divergent developmental pathways as photoreceptors or nonphotoreceptor (predominantly amacrine) neurons (Adler et al., 1984). The cells that differentiated as photoreceptors were analyzed in more detail and were found to express a very complex phenotype, which resembled in many respects the phenotypes of photoreceptors that develop in vivo, while differing significantly from the phenotype of amacrine neurons developing within the same culture microenvironment (reviewed in Adler, 2000). Taken together, the data suggested that some progenitor cells were committed to a photoreceptor fate and others to a nonphotoreceptor neuronal fate.
This same experimental paradigm was used, with some modifications, to investigate whether progenitor cells become committed to a photoreceptor or nonphotoreceptor fate before or after their terminal mitosis. Given that the highly heterogeneous embryonic retina contains mixtures of cells born at different times during a period of several days (see above), it was necessary to identify cells born during narrow time intervals. This identification was accomplished with a method known as the “window-labeling” technique, shown in Figure 1C, which is based on the sequential administration of [3H]thymidine and bromodeoxyuridine (BrdU; Repka and Adler, 1992; Belecky-Adams et al., 1996). The study thus was focused on a cohort of cells born during a 5-hr interval on embryonic day (ED) 5 (“WL5” cells) and is summarized in Figure 1D. In vivo, as assessed on ED18, 75% of the WL5 cells became nonphotoreceptor neurons, and only 25% gave rise to photoreceptors (Belecky-Adams et al., 1996). When WL5 cells were isolated for culture on ED8 (i.e., after remaining within the retinal environment for approximately 3 days after their terminal mitosis), they mimicked their in vivo behavior, giving rise to similar neuron/photoreceptor ratios. The developmental fate of the WL5 cells, however, was quite different when they were isolated from the retina on ED6, after a shorter exposure to the retinal microenvironment. In this case, 80% of the WL5 cells gave rise to photoreceptors, and only 20% became nonphotoreceptor neurons (Belecky-Adams et al., 1996). Control experiments indicated that differences in cell proliferation or cell death could not explain these striking differences in cell fate, which, therefore, appeared to indicate a real change in the developmental potential of WL5 cells during the interval between ED6 and ED8. This change, in turn, suggested that many precursor cells remained plastic (i.e., uncommitted to specific differentiated fates) for some time after their terminal mitosis.
From Operational Definitions to a Molecular Description of Cell Commitment
Operational tests of cell commitment such as those summarized above have been very instructive but share two very important limitations. First, they evaluate cell commitment retrospectively given that, by the time the developmental fate of progenitor cells has been experimentally assessed, they do not exist as progenitors any longer. Second, they provide information at the population, rather than at the single cell level; these experiments, in other words, can show that a population contains progenitors committed to different fates, but do so without identifying the cells committed to each fate. The interpretation of such experiments is further complicated by the heterogeneity of cell populations in the developing central nervous system (CNS), including the retina (see Operational Definition of Cell Commitment section). For example, if a treatment increases the frequency of ganglion cells and decreases the frequency of photoreceptors that differentiate in a culture of retinal cells, the result could be due to the induction of uncommitted precursors to differentiate as ganglion cells rather than as photoreceptors, and/or to increased survival of ganglion cells, and/or to increased death of photoreceptors, and/or to increased proliferation of progenitors destined to become ganglion cells, and/or to decreased proliferation of progenitors destined to become photoreceptors. There are now a variety of methods for the quantitative assessment of cell death or cell proliferation in vivo or in vitro, but they do not always allow distinguishing between these different scenarios. A main reason for this limitation is the lack of accurate criteria for the identification of subpopulations of undifferentiated but otherwise “committed” progenitor cells (see below).
Further progress in the analysis of cell differentiation in the retina (as in other CNS regions) appears to require the development of methods that would allow evaluating the commitment of individual cells, prospectively, and at the molecular level. An important step in this direction has been the development of methods that allow the amplification of RNA from individual cells either linearly (Eberwine et al., 2001) or exponentially (Brady and Iscove, 1993; Dulac, 1998; Brady, 2000). These methods have found increasing applications for the analysis of differentiated cells (ibid), including retinal cells (Wahlin et al., 2004), but have not yet found similar applications in the case of undifferentiated progenitor cells. Differentiated chick embryo rod and cone photoreceptors, for example, can be identified by morphological criteria and captured individually after the retina is dissociated (Fig. 2A; their identity can be verified with immunocytochemical methods, as well as by polymerase chain reaction (PCR) amplification of cell-specific gene products (Fig. 2B–D). Methods such as suppression subtractive hybridization make it possible to identify genes, including transcription factors, that are differentially expressed in individual rod or cone photoreceptors (Huang and Adler, 2005). Could techniques of this type be used to compare, at the molecular level, multipotential and “committed” progenitor cells? Unfortunately, such comparisons are currently not feasible, because subpopulations of progenitor cells cannot be distinguished morphologically (Fig. 1A), and suitable molecular “markers” are not yet available for their characterization. This limitation creates quite a conundrum: multipotential and committed progenitors would have to be identified to be isolated for molecular analysis, but they cannot be identified without information about their molecular composition, which does not currently exist.
Alternatives to capturing individual cells can be considered, which could theoretically allow the isolation of homogeneous populations of progenitor cells, (e.g., fluorescence-activated cell sorting, FACS; Maric and Barker, 2004), or their generation by clonal cultures derived from individual stem or progenitor cells (Tropepe et al., 2000; Seaberg and van der Kooy, 2002; Irvin et al., 2003). Such methods, however, are not devoid of limitations. FACS, for example, depends on the availability of molecular “markers” for the cells to be sorted which, as discussed above, have yet to be identified for multipotential and committed neural progenitors. Similarly, the assumption that clonal cultures are homogeneous has been challenged by recent studies showing that, despite their origin, a high degree of heterogeneity develops spontaneously in the cultures (e.g., Tropepe et al., 2000; Maric and Barker, 2004). It appears, therefore, that the development of suitable methods and reagents for the objective identification and molecular characterization of progenitor cells remains a major and critical challenge for developmental neurobiology.
DISSECTING THE COMPLEXITY OF INTRACELLULAR SIGNALING SYSTEMS THROUGH GAIN- AND LOSS-OF-FUNCTION APPROACHES
Power of Gain- and Loss-of-Function Methods
A major contributor to the explosive rate of progress in developmental neurobiology during the past several years has been the introduction of powerful techniques and reagents for gain- and loss-of-function experiments, such as gene transfection, viral vectors, electroporation, transgenic and inducible transgenic animals, dominant-negative and constitutively activated molecular constructs, knockouts and conditional knockouts, morpholinos, RNAi, and random and targeted mutagenesis. These gain- and loss-of-function approaches have generated (and undoubtedly will continue to generate) many valuable insights into molecular aspects of neural development. At the same time, however, their use has also shed light on some of their own limitations, which should foster efforts to develop complementary (but qualitatively different) approaches.
The Challenge of Investigating Combinatorial Gene Effects
The origin of retinal cells has been experimentally traced as far back in development as to individual blastomeres (e.g., Huang and Moody, 1993). During this long developmental history, progenitor cells and their descendants undergo many sequential transitions before becoming mature, differentiated cells. Gain- and loss-of-function studies have led to two significant generalizations regarding these transitions: (1) most if not all of these developmental phenomena are regulated or otherwise influenced by several different extracellular signaling molecules and intracellular transcription factors, and (2) many signaling molecules and transcription factors participate in the regulation of two or more developmental transitions, exerting different (even opposite) effects in each case. The homeobox gene Pax6, for example, is necessary for the determination of the retinal anlage and for the initial establishment of a pigment epithelial domain and a neural retinal domain within the optic vesicle (Macdonald et al., 1995; Schedl et al., 1996; Oliver and Gruss, 1997; Mathers and Jamrich, 2000; Schwarz et al., 2000, Marquardt et al., 2001; Ziman et al., 2001; Toy et al., 2002). Cell differentiation within the neural retina is also influenced by Pax6 at somewhat later stages, but its effects are quite different in different cell types. Thus, Pax6 expression appears necessary for the differentiation of ganglion and amacrine neurons (Marquardt et al., 2001, Toy et al., 2002), but its overexpression into progenitor cells inhibits their differentiation as photoreceptors (Belecky-Adams et al., 1997; Toy et al., 2002). These and many similar observations in other tissues have shown that the functions of transcription factors depend on the molecular context in which they act (reviewed in Silver and Rebay, 2005). Although a single transcription factor can have by itself a profound effect on cell fate, as illustrated by the transformation of rods into blue cones in the NRL knockout mouse (Mears et al., 2001), even a “master gene” such as Pax6 can only induce the formation of an ectopic eye when it is expressed under the appropriate conditions (Chow et al., 1999; Gehring and Ikeo, 1999).
A major breakthrough in the investigation of combinatorial gene effects of this type was the introduction of microarray technology, which allows comparing the levels of expression of thousands of genes in two RNA samples. This powerful methodology has provided much useful information about the retina (Zareparsi et al., 2004) and can be applied not only to tissue extracts, but also to single cells (e.g., Brady, 2000; Goto et al., 2001; Chiang and Melton, 2003). The wealth of data that can be obtained using microarray technology could have been expected to allow the elucidation of combinatorial gene effects in the regulation of cell differentiation. As recently noted, however, microarray technology has so far had only a modest impact on our understanding of developmental mechanisms (Livesey, 2003; Smith and Greenfield, 2003). Rather than an intrinsic problem of microarray technology, this limited impact is likely to reflect, to a very large degree, a drawback shared by currently available methods for gain- and loss-of-function experiments: these methods only allow targeting one or two genes at a time and do so in an all-or-none (“black-or-white”) manner (Fig. 3). Therefore, while microarray data allow formulating hypotheses about regulatory networks involving the combinatorial effects of a series of genes expressed at particular levels, these hypotheses are likely to remain untested until methods allowing the modulation of the levels of expression of groups of genes become available (Fig. 3).
ROLE OF EXTRACELLULAR SIGNALS IN THE REGULATION OF NEURONAL DIFFERENTIATION
Impressive progress has been made in recent years in the analysis of the role of extracellular signaling molecules in the regulation of neuronal differentiation. Emphasis has been placed predominantly on studies of individual signaling molecules, largely because secreted signaling molecules were initially considered highly cell type-specific. In turn, this concept was strongly influenced by the “neurotrophic hypothesis,” derived from studies of developmental neuronal death and the nerve growth factor (NGF; Davies, 1996; Yuen et al., 1996). Methodologically, the studies were also influenced by the increasing availability of purified signaling molecules and of well-characterized neuronal cultures that provided suitable bioassays for their analysis. Many factors active on particular types of neurons were identified in this manner. In the case of embryonic photoreceptor cells, for example, the list includes ciliary neurotrophic factor (CNTF; Kirsch et al., 1996, 1998; Ezzeddine et al., 1997; Fuhrmann et al., 1998; Ogilvie et al., 2000; Xie and Adler, 2000), fibroblast growth factor (FGF; Hicks and Courtois, 1992; Fontaine et al., 1998), sonic hedgehog (SHH; Levine et al., 1997), leukemia inhibitory factor (Neophytou et al., 1997), retinoic acid (Stenkamp et al., 1993; Kelley et al., 1994, 1995, 1999; Hyatt et al., 1996; Wallace and Jensen, 1999; Soderpalm et al., 2000), thyroid hormone (Kelley et al., 1995; Yanagi et al., 2002), glial-derived neurotrophic factor (Politi et al., 2001), pigment epithelium-derived factor (Jablonski et al., 2000), docosahexaenoic acid (Rotstein et al., 1998; Politi et al., 2001), taurine (Altshuler et al., 1993; Wallace and Jensen, 1999; Young and Cepko, 2004), and activin (Belecky-Adams et al., 1999).
There is now growing awareness that the microenvironmental regulation of neuronal differentiation is likely to be much more complex than once thought. It has become well established, for example, that mono-specific signaling molecules are likely to be an exception rather than the rule (reviewed in Patterson, 1994; Snider, 1994; Davies, 1996; Levi-Montalcini et al., 1996; Merrill and Benveniste, 1996; Carter and Lewin, 1997; Murphy et al., 1997; Tolkovsky, 1997; Adler et al., 1999). Signaling molecules of this type generally appear to be quite pleiotropic, triggering different responses on different cells through a variety of receptors and transduction pathways (e.g., Goumans and Mummery, 2000; Rajan et al., 2003; Waite and Eng, 2003, Storkebaum et al., 2004; Velde and Cleveland, 2005). There is also a fairly extensive body of literature showing that signaling molecules are modulated by interactions with each other and with extracellular binding proteins and that they can exert combinatorial, complementary, and/or redundant effects on cells (e.g., Ip and Yancopoulos, 1996; Phillips, 2000; Butte, 2001; Balemans and Van Hul, 2002; Abe et al., 2004; Harrison et al., 2004; Rosenstein and Krum, 2004; Sebald et al., 2004; Vergara and Ramirez, 2004). The interactive and pleiotropic nature of signaling molecules is particularly significant in view of the abundance and diversity of such molecules in the retina (e.g., Campochiaro et al., 1996; Hallbook et al., 1996; Schoen and Chader, 1997; Hicks, 1998; Belecky-Adams et al., 1999; Frade et al., 1999; Belecky-Adams and Adler, 2001; Carri, 2003; Yang, 2004). Retinal cell differentiation, moreover, can be influenced not only by secreted molecules but also by contact-mediated interactions (Linser and Moscona et al., 1984; Moscona et al., 1988; Hausman et al., 1993; Henrique et al., 1997; Layer et al., 1998; Prada et al., 1998; Becker and Mobbs, 1999). Not surprisingly, therefore, a picture of high complexity has emerged since the development of techniques and reagents for loss- and gain-of-function experiments made it possible to investigate the role of extracellular signals not only in cell culture, but also within the complexity of the intact embryo. For example, very similar changes in the development of the ventral retina and pigment epithelium can be induced by overexpressing SHH (Nasrallah and Golden, 2001; Zhang and Yang, 2001) and by blocking bone morphogenetic protein (BMP) signaling with noggin or dominant-negative BMP receptors (Adler and Belecky-Adams, 2002). It appears likely that SHH acts upstream from BMP in those effects (Zhang and Yang, 2001). Some of the phenotypic abnormalities triggered by noggin overexpression, moreover, were likely mediated and/or modulated by other signaling molecules, including retinoic acid, netrin, R-cadherin, laminin, and FGF-8 (Adler and Belecky-Adams, 2002). These and similar observations (e.g., Hunter et al., 2004; Martinez-Morales et al., 2005; Murali et al., 2005) suggest that the microenvironment in which retinal cells differentiate is a highly homeostatic system, in which complex changes are likely to result from experimental manipulation of an individual signaling molecule. Therefore, while analytical approaches based on perturbations of individual factors will undoubtedly continue to provide important and useful information, it may be necessary to develop new strategies and approaches to generate a more integrated description of the microenvironmental influences that regulate neuronal differentiation.
CHARACTERIZATION OF DIFFERENTIATED PHENOTYPES
Contributions of “Cell Markers” to the Study of Neuronal Cell Differentiation
Cell differentiation research owes much of its recent progress to methods such as in situ hybridization and immunocytochemistry, which allow comparing the expression of particular genes in different cell types within heterogeneous tissues. The power of these methods has led to widespread use of “cell markers,” that is to say, molecules whose expression in a cell is considered unambiguous proof of the cell's identity and differentiated state. There is now growing evidence, however, that cell markers are not devoid of potential pitfalls, some of which will be discussed below.
Specificity of Neuronal Markers Is Frequently Not Absolute
Some “markers” are restricted to a single cell type early in development but become more broadly distributed at later stages. An example is the transcription factor islet1, once considered ganglion cell-specific throughout chick retinal development (Austin et al., 1995). Based on this premise, increases in islet1-positive cells resulting from Notch down-regulation were interpreted to represent a specific increase in ganglion cell differentiation (ibid). This interpretation must be revisited, however, because islet1 has now been shown to be ganglion cell-specific only at early stages of chick embryo development and to be subsequently expressed by other differentiating neurons (Henrique et al., 1997). There are also examples of the opposite type of change in patterns of “cell marker” expression, because molecules restricted to specific cell types in the mature retina sometimes have broader distribution at earlier stages; this temporal pattern has been reported for several transcription factors (e.g., Freund et al., 1996; Belecky-Adams et al., 1997; Oliver and Gruss, 1997; Mathers and Jamrich, 2000). An additional layer of complexity is that undifferentiated cells not only can express differentiated cell markers, but can also express markers corresponding to more than one lineage; multilineage gene expression, for example, has been reported to occur before cell commitment in the hematopoietic system (Hu et al., 1997). It has also been proposed that transcription of individual genes during cell differentiation may well occur in the “wrong cells,” as a probabilistic event (Paldi, 2003). Against this background, it appears reasonable to reconsider whether a “marker” normally restricted to a particular cell type in the mature retina, when detected in a newly generated cell, should be considered an unambiguous indication of its commitment to that lineage. This explanation has been suggested to be the case, for example, for chick embryo progenitors that, shortly after terminal mitosis, express a cytoskeletal protein recognized by the monoclonal antibody RA4, a ganglion cell “marker” (McLoon and Barnes, 1989; Waid and McLoon, 1995).
The use of molecules expressed during terminal differentiation as markers of progenitor cell commitment would only be justified in cases in which the entire process of cell differentiation is controlled cell-autonomously by intrinsic mechanisms set in motion at the time of progenitor cell commitment to a particular lineage (Fig. 4). On the other hand, the approach would not be warranted if sequential inductive events are necessary before a committed progenitor can reach terminal differentiation (Fig. 4). A distinction between these two scenarios has been difficult for many cell types, but photoreceptor cells have been amenable to their investigation because their differentiation can be analyzed with many structural, molecular, and functional criteria. An initial indication of the existence of different regulatory mechanisms for different aspects of photoreceptor differentiation was the observation, made in several laboratories, that there is considerable asynchrony in the onset of expression of photoreceptor-specific genes (Bruhn and Cepko, 1996; Stenkamp et al., 1997; Johnson et al., 2001; Bradford et al., 2005). Some of those genes are already expressed at, or shortly after, the time of photoreceptor birth, preceding by many days other important landmarks in photoreceptor differentiation, such as the formation of outer segments (Saha and Grainger, 1993; Bruhn and Cepko, 1996; Stenkamp et al., 1997; Johnson et al., 2001; Bradford et al., 2005). The spatial patterns of expression of these early genes suggest that they are controlled by intracellular determinants, rather than by diffusible signals (Johnson et al., 2001). Cell-autonomous mechanisms also appear to control other “early” aspects of photoreceptor differentiation in the chick embryo (Adler, 2000; see the Investigating Cell Commitment section) as well as in other species (Cook and Desplan, 2001).
Many other aspects of photoreceptor differentiation occur much later in development and appear to be regulated by different mechanisms. There are many genes, for example, that only become detectable many days or even several weeks after photoreceptors are born. The onset of expression of these late genes correlates approximately with the onset of outer segment formation (Bruhn and Cepko, 1996; Cepko, 1996; Bumsted et al., 1997; Johnson et al., 2001; Bradford et al., 2005). Although the mechanisms responsible for this prolonged lag between photoreceptor birth and terminal differentiation remain unknown, this temporal difference between early and late genes provides circumstantial evidence against the notion that photoreceptor-specific genes could be globally coregulated. We recently tested this issue more directly, using in situ hybridization, reverse transcriptase-polymerase chain reaction (RT-PCR), and real-time PCR to investigate the expression of 18 photoreceptor-specific genes in retinal cells treated with agents previously reported to modulate the expression of specific visual pigments (Bradford et al., 2005). These agents are activin, which inhibits the expression of the red cone pigment; iodopsin (Belecky-Adams et al., 1999); CNTF, which up-regulates the expression of the green cone pigment (Xie and Adler, 2000); and staurosporine, which induces rhodopsin expression and down-regulates expression of the red cone pigment (Xie and Adler, 2000). Rhodopsin, for example, is not detectable in control cultures, even by RT-PCR but becomes readily detectable by in situ hybridization in 80% of the photoreceptors in staurosporine-treated cultures. Importantly, this strong inductive effect is not accompanied by detectable changes in other rod-specific genes. Similarly, a variety of cone genes remains unchanged when the red and green cone pigment genes are inhibited by staurosporine, and/or when the green cone pigment gene is up-regulated by CNTF treatment (Bradford et al., 2005). These cell culture experiments, therefore, provide additional evidence for the independent regulation of different photoreceptor-specific genes.
A key issue in evaluating the expression of cell markers in the face of perturbation is how to determine whether altered marker expression reflects cell fate, rather than a change in regulation of the marker. For example, the apparent differences between regulatory mechanisms controlling early and late aspects of photoreceptor differentiation, together with the lack of coordinated regulation of many photoreceptor-specific genes, raise some concerns about the use of molecules expressed during terminal differentiation (a late event) as indicators of the commitment of retinal progenitor cells to the photoreceptor lineage (a much earlier event). Controversies regarding the specific role of signaling molecules in the control of photoreceptor development may perhaps be explained by such use. Increases in rhodopsin-immunoreactive cells in rat retinal cultures treated with 9-cis retinoic acid, for example, were initially interpreted to represent increases in progenitor cell commitment to the photoreceptor fate (Kelley et al., 1994). However, subsequent studies found no evidence of a fate switch and showed that retinoic acid acts by shortening the maximum time between terminal mitosis and detectable rhodopsin expression (Wallace and Jensen, 1999).
Similarly, although there is consensus that CNTF causes a decrease in the number of rhodopsin(+) cells in rat retinal cultures, the effects have been interpreted as a re-specification of cells destined to become rods (Ezzeddine et al., 1997), arrested differentiation of cells already committed to the rod fate (Neophytou et al., 1997), or a transient and reversible down-regulation of rhodopsin expression (Schulz-Key et al., 2002). The ambiguity between cell fate determination and modulation of the expression of markers of terminal differentiation also remains unresolved for other factors that regulate photoreceptor development, such as taurine (Altshuler et al., 1993), ligands of steroid/thyroid receptors (Kelley et al., 1995), and sonic hedgehog (Levine et al., 1997). It appears, in summary, that molecules expressed during terminal differentiation are not suitable indicators of progenitor cell commitment to the photoreceptor fate; whether similar limitations apply to other neuronal types remains an open question, which deserves to be investigated.
The impressive body of information on mechanisms of neuronal differentiation that has been generated to date provides a solid foundation for future studies, many of which will continue to make use of currently available methods and approaches. Equally necessary for further progress, however, is our awareness of the gaps that still exist in our present knowledge, of the questions that remain to be answered, and of the limitations of the methods and approaches currently available for their investigation. Overcoming these limitations will not be easy, but the effort is likely to have a broad impact, because, although they have been discussed in this article within the narrow confines of neuronal differentiation in the retina, they are similarly relevant to cell differentiation studies in other parts of the nervous system, and even in non-neural tissues.
The author thanks Valeria Canto Soler and Karl Wahlin for comments and suggestions on the article, and Betty Bandell for secretarial assistance. Research in the author's laboratory was supported by grants from the National Eye Institute and the Foundation Fighting Blindness. R.A. is the Arnall Patz Distinguished Professor, and a Senior Investigator of Research to Prevent Blindness, Inc.