Site search   
  Home    Publications

Browse Publications
List of Contents

Quantitative Neurogenetics & QTL Mapping

Genetics of Myopia

Control of Neuron Number and Stereology

Growth Cones and Dying Axons

Retina Development and Visual System Mutants

Grant Application

U.S. Patent


Need Help?
Help with Publications
Help with Nervenet
Contact Us

Note to the Reader Trends in Neuroscience, Vol. 15: 368-373.

Print Friendly
Lineage versus environment in embryonic retina: a revisionist perspective

Robert W. Williams and Dan Goldowitz

The idea that microenvironmental cues acting alone late in development determine a cell's phenotype has dominated recent discussion of retinal development and has successfully displaced any role for cell lineage in the process of cell determination. We argue that there is, in fact, evidence favoring a degree of lineage restriction during the development of the vertebrate retina. We propose that environmental factors modulate a process of progressive lineage restriction. In this model, progenitor cells are viewed as having unequal potential and their progeny are viewed as being committed to one of the major retinal cell classes before the stage at which they become postmitotic.


Jump to Letters to the Editor

Much progress has been made in the last ten years in developing ways to mark and analyze descendants of single progenitor cells—cellular clones—in the vertebrate CNS. A compelling reason to study these families of cells is that a careful analysis of their size, placement, and cell composition gives us insight into genetic and developmental mechanisms that generate fully differentiated cell and tissue types from progenitors that are initially undifferentiated (1-5). In this article we focus on the retina, an accessible part of the CNS with a comparatively simple layout that makes it a particularly favorable tissue in which to explore the relationship between a cell's lineage, its environment, and its phenotype.

In a set of recent studies, progenitor cells in retina have been marked at different stages of development, using a variety of methods (6-11). Without exception, these studies have shown that the resulting clones of retinal cells are tightly interknit clusters of cells that are aligned radially across the retinal layers (Fig. 1). Whether by design or happenstance, cells in a clone work together in adult retina, processing information from the same region of visual space.

A menagerie of clone types—dependence on time of labeling

When progenitor cells are marked early in development by combining genetically distinguishable 4- to 8-cell mouse blastocysts into a single embryo (12), the resulting clones contain a balanced representation of the major retinal cell types (10). In chimeric mice these clones can be visualized as beautifully discrete and complete retinal building blocks (Figure 1a). With minor exceptions, each clone contains the same ratio of cell types as the retina itself. This result has been confirmed and extended in a second vertebrate—the African clawed toad, Xenopus laevis. Huang and Moody (11) injected a fluorescent dye into individual cells at a stage when the entire embryo is made up of just 32 cells. They found that 10 of these 32 progenitors contribute to the pool of retinal progenitors, and no matter which of these 10 cells they marked, the resulting clones in retina contained nearly the same ratio of cell types as the retina as a whole. These complementary results in mouse and Xenopus demonstrate that early in development retinal progenitors have equivalent capacity to produce all major retinal cell types. In this key respect, the pool of progenitors is uniform.

In marked contrast, the structure of clones differs greatly when progenitor cells are labelled at later stages of retinal development in these same two species (6, 7, 9). Clones are now extremely variable in their cellular composition. One dramatic example of the range of variation is illustrated by clones generated by progenitor cells infected with a retroviral marker on day 14 of gestation (E14) in the mouse (7). One progenitor gave rise to a cluster of 33 rods exclusively, whereas another progenitor in the periphery of the same embryo gave rise to a large clone of 198 rods, 1 cone, 26 bipolar cells, 7 amacrine cells, and 2 Muller glial cells. A similar profusion of different types and sizes of clones (see Fig. 1B) has been found in Xenopus retinas following injections of heritable tracers—either horseradish peroxidase or fluorescent dye—during the middle stage of retinal development (6, 9).

Competing hypotheses: Lineage restriction and environmental regulation

There are two different explanations for the transformation from uniform clones generated by progenitors early in development to highly variable clones generated by progenitors labelled later in retinal development (Figure 2). The first explanation is that at some point in retinal development, an initially uniform pool of progenitors splits up into a variety of subtypes—each with differing proliferative potential and differing capabilities to make different types of retinal cells (Fig. 2A). From this perspective, the clone of 33 rods was produced by an E14 progenitor that was only capable of generating rods. In this cell a genetic switch instrumental in deciding the fate of all its progeny was presumably stuck in the rod-only position. This is a possible instance of simple and complete lineage restriction. The lineage restriction hypothesis views the variation among clones generated at later stages of retinal development as being a direct reflection of underlying variation in gene expression among retinal progenitors—some become totally restricted, others become only partially restricted, and some may retain their original pluripotence; some produce large clones via frequent symmetrical divisions, and others produce small clones via asymmetrical or differential divisions (Fig. 2A).

The second and more widely accepted explanation for the bewildering diversity among retinal clones at later stages of retinal development is that this diversity is a direct reflection of underlying microenvironmental heterogeneity (Fig. 2B). This idea was initially based on an exquisitely detailed analysis of electron microscopic images of cells in embryonic mouse retina undertaken by Hinds and Hinds (13). They found that the pool of progenitors appeared homogeneous at the ultrastructural level, even quite late in development. From this beginning the idea has evolved that homogeneous progenitors produce postmitotic "blank slate" progeny. These postmitotic, still uncommitted cells are then assigned a phenotype by interacting with neighbors that have already been committed and that have already begun to differentiate (6, 14). In essence, a spatio-temporal cascade of inductive and inhibitory interactions among cells, both local and global, is thought to be the key arbitrator of a cell's destiny (14-16). Environmental differences give rise to the great variety of mixture of cell types seen in neighboring clones. A corollary is that the large, modular clones generated by progenitors at early stages of development are composed of sets of these smaller and highly variable subclones.

Lineage and environmental hypotheses lead to different predictions about clone structure

The wealth of data on retinal clones in mouse and frog now makes it possible to assess the strengths and weaknesses of these two models. If the potential of progenitors to make different cell types is progressively restricted, then clones marked at later stages of development should contain fewer combinations of cell types than either those generated early in development or those generated by a process that randomly assigns a phenotype to each member of the clone. Thus, the frequency of clones containing only a few cell types should be much higher than expected by chance (Fig. 3). An extreme example of restriction is the clone of 33 rods—a combination which a random process would generate in the mouse with a frequency of only 1 in 13,000.

The opposite prediction follows from the environmental hypothesis—progenitor cells are thought to remain fully pluripotent. Consequently, sets of cells that these progenitors generate (members of a clone) should include many different cell phenotypes. However, these clones should not be random sets of different cell phenotypes. The retina is after all, a highly regular structure (17), and radial arrays contain balanced ratios of different retinal cell types. The environmental hypothesis holds that this regularity is achieved by stereotypic patterns of interactions among neighboring cells, a great many of which are inevitably members of the same clone (Fig. 1). For example, if one of the first cells in a clone differentiates as a ganglion cell, this cell will generate environmental signals that lower the probability that its neighbors—including other members of its clone—also become ganglion cells. Instead, this young ganglion cell should signal its neighbors to differentiate as bipolar, amacrine, or photoreceptor cells. Feedback interactions of this type will generate a greater diversity of cell types within clones than would a random process. There are several concrete experimental examples of the ways in which such mechanisms reestablish a more nearly balanced representation among cell types in retina following the selective ablation of specific phenotypes (15, 18). Perhaps the best current example of the numerical and phenotypic regularity that can be achieved by environmental interactions is the complex of photoreceptor types in the ommatidium of the fly (19). Here, a spatio-temporal gradient of cell production coupled to a series of ligand-receptor interactions between neighboring cells triggers an invariant mosaic of three different receptor types in ommatidia across the entire eye.

A critical assessment using a random model of cell determination

Given these two predictions, a way to test the relative importance of lineage and environment is to determine whether combinations of cell types in clones are less diverse than predicted by chance (favoring lineage restriction) or more diverse than predicted by chance (favoring environmental regulation).

We have tested these predictions using as a starting point the set of clones published in the landmark study by Turner, Synder, and Cepko (7). We performed a Monte Carlo simulation in which many thousands of computer-generated clones were compared to the real data set. To run this simulation, retinal cells of different types were assigned a selection probability based on their proportion in the adult mouse retina (Fig. 3). Adjustments were made to eliminate from consideration cells produced before the stage at which Turner and colleagues made their retroviral injections (E13 and E14). Cells were then randomly and repeatedly 'pulled out of a hat' thereby generating sets of simulated clones that contained precisely the same numbers of cells as observed by Turner and colleagues. The variety of cell types in these simulated clones was categorized and compared to the variety in the set of real clones.

The cellular composition of simulated clones is strikingly different from that of the set of 219 retrovirus-labelled clones (Fig. 3). An average of 20% of the Monte Carlo clones (44 of 219) contain all six of the most numerous retinal cell types produced after E14. In contrast, only a single retrovirus-labelled clone contained representatives of each of these six common cell types. Similarly, 60 Monte Carlo clones contain five different cell types, whereas fewer than a third as many of the retrovirus-labelled clones contained five cell types. Yet, one would expect that if the microenvironment modulates clone structure, then these canonical clones containing representatives in each cell layer, should be even more common than predicted by a merely random process. Complementing this first finding, the simulation also reveals that as the mixture of cell types within clones is restricted (lower part of Fig. 3), observed numbers of retrovirus-labelled clones rise substantially above numbers predicted by the simulation. For example, only 45 of the Monte Carlo clones contain just two or three different cell types, whereas 126 of these more restricted clones were observed in the real data set. In addition, many exceedingly improbable clone types with only one or two cell types, such as all-cone clones, were found following retroviral injections. Summing this work up, real clones with low cellular diversity are more common, whereas clones with the high cellular diversity are much less common than predicted either by a random process.

A similar analysis of retinal clones has also been performed in Xenopus by Holt and colleagues (6). In this amphibian, 24% of all retinal cells are photoreceptors, 54% are inner nuclear layer interneurons (predominantly amacrine and bipolar cells), and the remaining 22% are ganglion cells. They injected progenitors with HRP at early stages (before the production of an appreciable number of postmitotic cells), and compared the observed and expected mixtures of cell types in clones containing two or three cells. Their chi-square analysis reveals that the observed combinations often resemble those expected by chance—a finding consistent with neither lineage restriction nor environmental regulation. However, in scanning their Table 3, one cannot help noticing that the most restricted clone types—in particular, all-rod clones—are found more frequently than predicted by chance. Perhaps even more intriguing, only 1 of 23 three-cell clones contained a member in each of the three cell layers. Yet if inductive and inhibitory interactions among cells in these tightly intertwined clones (Fig. 1B) are influential—as they are, for instance, in ommatidia of the fly—one would expect a much larger percentage of clones with a balanced representation of cell types across all cell layers.

A synthesis

From the set of studies in mouse and frog we conclude that lineage restriction does occur during retinal development, just as it does to a certain degree both in cortex (see Box 1) and in the optic nerve (4). Several lines of evidence suggest that decisions are made and biases are introduced among members of the progenitor pool throughout development. For instance, in vitro studies by Reh and colleagues (20) have demonstrated that dividing cells taken from fetal retinas produce an abundance of ganglion cells, whereas those taken from neonatal retinas and put in an identical in vitro environment produce an abundance of rods. This work provides compelling evidence that the average internal state of progenitors shifts over time, possibly under the influence of changes in retinal environment, or possibly due to internal changes associated with cell division. Along with these temporal shifts, Drager and colleagues (21) have recently shown molecular heterogeneity among progenitors in the mouse at a very early stage. As early as E9, long before the production of any postmitotic cells, progenitors in dorsal retina, but not ventral retina, express high levels of alcohol dehydrogenase activity.

We have highlighted findings that suggest lineage restriction plays a role in retinal development. We have done this to counterbalance a growing perception that cell phenotype in the CNS is almost entirely under environmental control acting late in development. In a recent New York Times article (22), Cepko summarized her group's work as showing that "once the neurons have settled into a particular neighborhood, they learn what they are meant to do from signals that surround them, rather than from an innate genetic program." In the same vein, the title of the paper by Turner, Synder, and Cepko on retinal clones (7) reads "Lineage-independent determination of cell type in the embryonic mouse retina." Yet as we have shown in Figure 3, the paper by Turner and colleagues points to a surprising degree of lineage restriction. How can such a stark difference of interpretation arise, and how can it be resolved? Is there a middle ground in which both factors can be shown to play a role?

One problem may be what is meant by "lineage restriction." If restriction means that each progenitor gives rise to only a single cell type, then, yes, the data rule out such a process. But as Jacobson and Moody (23) have suggested, restriction is not necessarily all-or-nothing. Allowance should be made for the possibility that there are developmental shifts in probabilities that progenitors will make certain types of cells—restriction may be lax.

What role do we leave the environment? There can be no doubt that inductive interactions have the most profound influence on cell potential and phenotype—starting with the earliest interactions between germ layers. These interactions, examined in depth by Leo Buss in The Evolution of Individuality (24), are the key to creating multicellular organisms. At this point, the relevant questions include, Which cells are influenced by their environment, how, and in what sequence? We think that the environment targets progenitor cells directly and then, either in a stepwise or graded manner, restricts their potential. The diversification of retinal progenitors may begin just after the first contact between the eye vesicle and overlying ectoderm (21) and may cease only at the last cell division (25, 26). Based on a quantitative comparison between our clones and polyclones in chimeric mice and clones and subclones labeled with retrovirus by Turner et al., we think it is likely that restriction of these progenitors begins between E11 and E12 (10), roughly concurrent with the production of the first postmitotic retinal cells (27). The progressive restriction of progenitors may be fine-tuned by environmental signals.

Although one may disagree with our reading of the results, it is certainly premature to rule out a role for lineage. To critically assess the role of cell lineage and environment will require in vivo transplantation of single progenitor cells into host retinas at different stages of development (5, 28). In vitro experiments, in which progenitors are placed in well-defined environments, will also continue to help in determining the rigidity of commitment and in discovering ways to control the state and output of CNS progenitor cells.

Acknowledgments: We thank S. Moody and S. Huang for sharing a draft of their paper on clones in Xenopus retina before publication. We thank R. Wetts and M. Luskin and the reviewers for their thoughtful comments. Supported by the NEI.

Selected references


  1. Jacobson, M. (1985) Trends Neurosci. 8, 151-155
  2. Doe, C.Q. and Scott, M.P (1988) Trends Neurosci. 3, 101-106
  3. Sanes, J.R. (1989) Trends Neurosci. 12, 21-28
  4. Raff, M.C. (1989) Science 243, 1450-1455
  5. McConnell, S. (1991) Ann. Rev. Neurosci. 14, 269-300
  6. Holt, C.E., Bertsch, T.W., Ellis, H.M. and Harris, W.A. (1988) Neuron 1, 15-26
  7. Turner, D.L., Synder, E.Y., and Cepko C.L. (1990) Neuron 4, 833-845
  8. Price, J., Turner, D., and Cepko, C. (1987) Proc. Natl. Acad. Sci. USA 84, 154-160
  9. Wetts, R. and Fraser, S.E. (1988) Science 239, 1142-1145
  10. Williams R.W. and Goldowitz, D. (1992) Proc. Natl. Acad. Sci. USA 89, 1184-1188
  11. Huang, S., and Moody, S.A. (1992) J. Neuroscience (in submission)
  12. Goldowitz, D. (1989) Neuron 3, 705-713
  13. Hinds, J.W. and Hinds, P.L. (1979) J. Comp. Neurol. 187, 495-512
  14. Jessell, T.M. (1991) in Principles of Neuroscience. Kandel, E.R., Schwartz, J.H., and Jessell, T.M., eds. Elsevier, Amsterdam
  15. Reh, T.A and Tully, T. (1986) Dev. Biol. 114, 463-469
  16. Adler, R. and Hatlee, M. (1989) Science 243, 391-393
  17. Wassle, H. and Reimann, R.J. (1978) Proc. R. Soc. Lond. B 200, 441-461
  18. Negishi, K., Teranishi, T., and Kato, S (1982) Science 216, 747-749
  19. Hafen, E. and Basler, K. (1991) Development, Supplement 1, 123-130
  20. Anchan, R.M., Reh, T.A., Angello, J., Balliet, A. and Walker, M. (1991) Neuron 6, 923-936
  21. Drager, U.C., McCaffery, P., Tempst, P. (1991) Soc. Neurosci. Abst. 17, 186
  22. Angier, N. (1992) New York Times January 28, C1, C8
  23. Jacobson, M., and Moody, S.A. (1984) J. Neurosci. 4, 1361-1369
  24. Buss, L.W. (1987) The Evolution of Individuality. Princeton Univ. Press
  25. Turner, D.L. and Cepko, C. (1987) Nature 328, 131-136
  26. La Vail, M.M., Rapaport, D.H. and Rakic, P. (1991) J. Comp. Neurol. 309, 86-114
  27. Stent, G.S. (1985) Phil. Trans. Roy Soc. Lond B. 312, 3-19.

Fig. 1. Clones of cells in vertebrate retina. (A) is a cross-section through the retina of an adult chimeric mouse. These chimeric mice are made by combining genetically distinct mouse embryos in vitro. The resulting double-genotype embryos are implanted into psudeo-pregnant mothers and born normally at term (Ref. 12). The three narrow columns of unlabelled cells are of Mus caroli genotype, whereas the more extensive heavily labeled regions are made up of cells that have been labelled with a biotinylated DNA probe that hybridized with a Mus musculus satellite DNA sequence (Ref. 12). Cells within the radially-oriented arrays are in many cases derived from single retinal ancestors. These clonal columns are often sectioned obliquely, giving rise to clones that appear to be limited to one layer in single sections (right-most clone in A). In Ref. 10 we discuss the characterization of clones and polyclones in chimeric tissue. Combinations of cells in these relatively large and uniformly shaped clones reflect the underlying structure of the retina itself. (B, C, and D) Clones of retinal cells in larval Xenopus frog at stage 41 taken from Ref. 6. The three clones in B, C, and D were marked by Holt and colleagues by injecting single progenitor cells with HRP at stages 22 to 27. These small clones of cells contain a wide variety of combinations of different cell types. For example, (B), left-most clone, contains two amacrine cells and one ganglion cell; (C) the middle clone contains one cell in each layer; and (D) contains 3 cells, all of which are photoreceptors In all photographs the photoreceptor layer or outer nuclear layer (ONL) is at the top, the inner nuclear layer (INL), is in the middle, and the ganglion cell layer (GCL) is at the bottom. Calibration bar equals 25 um. (Fig. 1B,C, and D are taken with permission from Ref. 6.)


Box 1. Environment and lineage in the cerebral cortex: Both play a role.

Recent work on the development of the mammalian cortex has, as in the retina, focused on the role of cell environment and cell lineage in determining neuronal features that range from phenotypes of single cells to areal projection patterns (1). Finlay and Slattery (2) initially suggested that a uniform embryonic cortex differentiates into numerous cytoarchitectonic divisions via differential cell death that is itself regulated by ingrowing afferents. By transplanting small pieces of embryonic rat neocortex to ectopic cortical sites, O'Leary and colleagues (3) have demonstrated that projection phenotype of cortical cells are influenced by the local cortical environment. Frost, Sur and their coworkers (4,5) have shown that cortex is functionally pluripotent—both auditory and somatosensory cortex can process visual information following early alterations in cortical environment. Collectively, these studies have led to the idea that the entire cortex is initially a uniform sheet. The highly differentiated functional and structural state of adult cortex is thought to arise gradually under the control of developing neuronal connections.

Evidence for intrinsic, lineage-related determination of other cortical properties, such as laminar destination and cell type, have come from transplantation and retroviral lineage studies. Barbe and Levitt (6) have found that neurons from embryonic limbic cortex are committed to expressing a limbic system antigen even when transplanted into non-limbic neonatal cortex. McConnell and Kaznowski (7) have found that laminar destiny is determined during or before the final round of cell division. Finally, Luskin, Parnavelas and coworkers (8,9) have shown that cortical progenitors give rise to clones of a single phenotype, i.e., containing pyramidal cells or interneurons, oligodendrocytes or astroglial. Their work provides support for the idea that cell lineages are restricted along these phenotypic axes at least 2-3 cycles before mitotic exhaustion.

Cortical neurons are undoubtedly influenced by numerous environmental cues, but the fact that apparently undifferentiated cells can be altered in certain respects does not necessarily mean that those cells are unspecified. Cells and cytoarchitectonic regions have many phenotypic traits, some of which may be under relatively tight genetic control, while others depend on environmental cues. Interpretations of results in cortex depend upon the level of analysis, and to some degree on the willingness of observers to entertain the idea that cortical development is more complex than we would like it to be.



  1. Rakic, P. (1988) Science 241, 170-176
  2. Finlay, B.L. and Slattery M. (1983) Science 219, 1349-1351
  3. O'Leary, D.D.M. (1989) Trends Neurosci 12, 400-406
  4. Frost, D. O. and Metin, C. (1987) Nature 317, 162-164
  5. Sur, M. Pallas, S.L., and Roe, A.W. (1990) Trends Neurosci 13, 227-233
  6. Barbe, M. F. and Levitt, P. (1991) J Neurosci 11, 519-533
  7. McConnell, S.K. and Kaznowski, C.E. (1990) Science 254, 282-285
  8. Barfield, J.A., Parnavelas, J.G., and Luskin M.B. (1990) Neurosci. Abstr. 16, 1272
  9. Parnavelas, J.G., Barfield, J.A., Franke, E., and Luskin, M.B. (1991) Cerebral Cortex 1, 463-468


In the pursuit of trying to find out how a cell in the vertebrate retina decides what type of cell it's going to be, we and others (1-3) found that sister cells had fates that could not be predicted by any simple or strict lineage model. Williams and Goldowitz (4) in their recent TINS perspective article do not argue this point, but they suggest that in retinal histogenesis, lineage restrictions might nevertheless play a role, and might better predict the numercial data on clone constitution than a model based on cellular or microenvironmental interactions. While they present no numerical model of their own, and their Monte Carlo simulation-based challenge of the existing lineage data may be flawed (see the following letter by C. Cepko), they could still be right. They start by reminding us that late clones in the mouse naturally comprise primarily only a few types of cell (rod, bipolars and Muller glial cells), i.e. the ones that are born later. It seems natural, therefore, to suspect some kind of lineage restriction takes place. If by lineage restriction we mean that a cell carries a predisposition (in the form of nuclear changes, cytoplasmic determinants or cell surface receptors) that enables it to choose a more limited number of fates than its mother, this certainly could be consistent with the data. Clearly, however, even this hypothesis implies that a newly born retinal cell has at least a repertoire of possibilities (the various late-born types of cell) open to it, and the choice it eventually makes among these is ungoverned by lineage.

There is, however, another possibility that is consistent with the predominance of rods, bipolars and Muller cells in late clones of the mouse, a possibility that Williams and Goldowitz seem not to consider as clearly as they might. This is that the environment changes with time! The addition of certain differentiated cells, the release of trophic factors and the maturation of the extracellular matrices may all influence the possible fates of a fully pluripotent retinoblast. A decision between two possibilities: a lineage-related restriction in fate on the one hand, and a changing microenvironment influencing pluripotential cells on the other, must therefore be resolved experimentally.

A number of recent papers have suggested that the phenotype of a newly born retinal cell can be influenced by the environment. Embryonic cells from the rat retina, if removed and co-cultured with an excess of older retinal cells, give rise to many more rods than they do when cultured alone (5). Postmitotic chick retinal cells that would have become photoreceptors if cultured at one time, become mutlipolar neurons if they are allowed to stay in the retina for another day (6). In Xenopus, cell-cell interactions seem to be required for photoreceptor determination (7). In the mouse, a transiently expressed soluble factor can make cells choose a rod fate rather than what appears to be a bipolar fate (8,9). Single mouse retinal cells will become rods if they are cultured next to rat rods, but will grow multipolar neurites and turn on ganglion cell markers if they are plated next to cortical cells (10). These many examples of demonstrable flexibility of cell fate cannot go down too well with a model of lineage restrication as define above. Hartenstein and I (Ref. 11) also showed that various classes of cell types in all three retinal layers can arise in embryos in which cell division had been blocked from before the time when retinal histogenesis normally begins. In these embryos, retinoblasts chose particular fates without spinning off any of their normal postmitotic progeny. That cells 'deprived of lineage' can choose these various fates is another difficulty with the lineage restriction idea.

I don't want to suggest the Williams-Goldowitz idea has no validity, it is just that the data that exist now would seem to argue against it, at least in the retina. Things are clearly somewhat different in the cortex, where cells seem to be restricted at least to particular laminae in the S-phase prior to their final mitosis (12). I think we all agree that cellularly inherited determinants or induced states are clearly an important part of embryology. The questions facing us now concern the cellular and molecular bases of the inductions that restrict neuron fate. When these occur in devleopment may vary from tissue to tissue. In the cortex, the central glia (13) and the neural crest (14), some restrictions happen while the cells are still dividing. In the vertebrate retina, it seems that many of these cell-type decisions happen postmitotically.

William A Harris

Dept of Biology, University of California at San Diego, La Jolla, CA 92093, USA


  1. Holt, C.E., Bertsch, T.W., Ellis, H.M. and Harris, W.A. (1988) Nature 333, 737-741
  2. Wetts, R. and Fraser, S.E. (1988) Science 239, 1142-1145
  3. Turner, D.L., Snyder, E.Y. and Cepko, C.L. (1990) Neuron 4:833-845
  4. Williams, R.W. and Goldowitz, D. (1992) Trends Neurosci. 15, 368-373
  5. Watanbe, T. and Raff, C. (1990) Neuron 2, 461-467
  6. Adler, R. and Hatlee, M. (1989) Science 243, 391-393
  7. Harris, W. and Messersmith, S. (1992) Neuron 9, 357-373
  8. Altshuler, D. and Cepko C. (1992) Development 114, 947-957
  9. Watanabe, T. and Raff, M.C. (1992) Development 114, 899-906
  10. Reh, T.A. (1992) J. Neurobiol. 123, 1067-1083
  11. Harris, W.A. and Hartenstein, W. (1991) Neuron 6, 499-515
  12. McConnel, S.K. and Kaznowski, C.E. (1991) Science 254, 282-285
  13. Raff, M.C. (1989) Science 243, 1450-1455
  14. Anderson, D.J. (1989) Neuron 3, 1-12



Many hypotheses concerning retinal cell fate dtermination have been based upon the data generated by lineage analyses. When evaluating these hypotheses, it is important to bear in mind the limitations in the type of conclusions that can be drawn from studies of lineage. Lineage analysis is a descriptive technique in which the fate of cells left in situ, rather than the full potency of cells, is observed. What studies of lineage have not, and cannot, resolve are the mechanisms underlying the observed clonal compositions. Only by manipulating the environment can one resolve the extent to which the autonomous programs of cells and environmental interactions contribute to development.

Every lineage study conducted to date has indicated that the vast majority of early retinal progenitors are multipotent (Footnote 1). The existence of clones comprising multiple cell types rules out a model in which each cell type is generated from a mitotic progenitor committed to making only one cell type. While the finding of multipotency rules this model out, it is consistent with several other models. As we hypothesized in our paper on mouse retinal lineage (1), multipotency could reflect cells responding to environmental cues to become or produce the many types of daughters that are observed in retrovirally marked clones. Moreover, complex clones could result from progenitors that are totipotent, and thus equivalent, throughout development of the retina. Alternatively, progenitors could change in potency during development. Changes in potency could be due to environmental influences, autonomous 'programming', or both. More extensive discussions of these possibilities have already been presentd (Ref. 1, p. 843, and Ref. 2).

The scenarios presented above were offered as hypotheses consistent with the data, rather than as conclusions, as one cannot distinguish among them using data generated by lineage analysis. Nonetheless, Williams and Goldowitz (3) attempted to distinguish among these models on the basis of lineage data using a Monte Carlo simulation to predict the composistion of clones marked by retroviral infections. They believe that their analysis ruled out a model in which environmental interactions direct the choice of cell fate. However, what their simulation really tested was whether clones comprised random assortments of cells that could be statistically predicted by the frequency of each cell type in the adult retina. The failure of clonal composition to be predicted by their model is due to the fact that the assumptions of their model are inconsistent with the biology of retinal development, as discussed below. However, even had their assumptions been correct, many interpretations of their findings regarding mechanisms would have been possible, such as variations in environmental influences, differences in progenitors, or both.

An examination of the model used for the Monte Carlo analysis shows that a critical underlying assumption is inconsistent with what we know about retina cell generation. By using the frequencies of cell types in the adult retina to predict probabilities, Williams and Goldowitz made the assumption that there is no temporal variation in the probabilities that a particular cell type would be generated. However, retinal cell types are born in a temporal sequence. This means that the strongest predictor for the fate of a given cell, regardless of the mechanism by which that fate is achieved, is the birthday of the cell, not the frequency of cells in the adult retina. An analysis based upon the frequencies of cell types in the adult retina would quite predictably lead to clonal compositions that would be more complex than predicted for a system in which there is birth order. Thus, the result of the Monte Carlo simulation had to differ from the results of the retrovirus study regardless of whether commitment was environmentally influenced.

The problem created by the phenomenon of birthdate order is illustrated by the example of the composition of two-cell clones. According to the assumptions of Williams and Goldowitz, 70% of the cells in two-cell clones should be rods, as this is the frequency of rods in the adult retina. However, the birthdate data would predict that two-cell clones would most likely be composed of cell types born shortly after infection with a retrovirus. As predicted by the birthdate data, two-cell clones marked by retroviral infection at embryonic day 13 (E13) or E14 have no rods or Muller cells (cell types generated late in development), but do have the predicted early-generated cell types, all of which are rare in the adult retina.

Since lineage data cannot be manipulated in order to distinquish among modesl of mechanisms, several labroratories, including our own, have undertaken experiments designed to explore the potency of retinal progenitors and define mechanisms of determination. The approaches are varied, including heterochronic transplantation, assays of commitment and differentiation in vitro, sutdies of the effects of peptide growth factors, and identification, cloning and functional studies of genes proposed to act in development. Data generated by such studies will ultimately reveal what lineage analysis cannot, that is, the ptency of progenitors at differnt times in development and the mechanisms by which fate is achieved. Until such data are obtained, arguments about models regarding mechanisms are not really very productive, and obscure the interersting story that is currently unfolding regarding teh very resolvable problem of fate determinaation in the vertebrate retina.

Footnote 1. There are clones that comprise rods only. Interpretation of such clones is difficult as the absence of other cell types in a clone can be explained by a failure to express the marker used to identify the cell type or cell death. These problems are particularly vexing in this case as rods have a very low rate of cell death, non-photoreceptors neurons have an appreciably higher rate of cell death, and rods are both extremely common and late born.

Constance Cepko

Dept of Genetics, Harvard Medical School, 200 Longwood Av, Boston, MA 02115, USA.



  1. Turner, D.L., Snyder, E.Y. and Cepko, C.L. (1990) Neuron 4:833-845
  2. Cepko, C.L. in Progress in Retinal Research (Vol. 12) (Osborne, N.N. and Chader, G.J., eds), Pergamon Press (in press)
  3. Williams, R.W. and Goldowitz, D. (1992) Trends Neurosci. 15, 368-373


REPLY by RW and DG

We agree with Dr. Harris—environment plays an important role in developmental decisions made by retinal cells. But the work he cites, while demonstrating a role for one process (environmental modulation), does not rule out a role for another process (lineage restriction or bias). As we concluded in our review (1), this is not an "either-or" situation. Drs. Harris (2) and Cepko (3) have left us and other readers with the conclusion that cell type determination is entirely under environmental control and that the environment targets a homogeneous pool of postmitotic retinal cells. While many, if not most, retinal progenitors generated a variety of cell types (5), the fact that clone composition is so variable late in development (3) suggests to us that the progenitor pool is heterogeneous. In other words, the potential of progenitors is non-equivalent—these cells appear to be limited in the range of cell types they generate.

There are a few points in Constance Cepko's letter that we wish to comment on. First she notes that the temporal sequence of cell generation needs to be taken into account in modeling clone structure. We recognized this at the outset and therefore excluded from the simulation those cells known to be generated prior to the retroviral injections (see the legend to Fig. 3 in Ref. 1). Even considering later born cells, our conclusions remain valid— there are rather obvious signs of lineage bias in the original retroviral data. How else can one explain a clone made up of 33 rods? The progenitor of this clone must have undergone five or more cycles of cell division and all of the final progenitors deivided to produced nothing but rods. Can this result really be written off in a foonote as a retroviral artifact or as the result of highly selective cell death? If so, there are unpleasant implications for the remainder of the retroviral data set. We emphasize this clone because it is such a blatant example. However, there are other signs of lineage bias. For instance, of a total of 70 multicellular clones containing an average of 57.6 cells that were labeled at embryonic day 13 (E13), only three contain even a single retinal ganglion cell. Yet ganglion cell production peaks at this stage and continues until birth (6). This also indicates some form of restriction.

Second, we agree with Cepko that small clones, which are probably generated within days of the injection, would best be modeled separately using data on cell genesis just after the time of the injection. However, she draws an overly sharp distinction between early- and late-generated cell types. Rods are actually generated as early as E13, and by E14 two to three times as many rods are being generated as cones (Figs 4 and 6 or Ref. 7). Averaged over the E13-E14 interval, 30% of all postmitotic cells being produced differentiate as rods, 25% as amacrine cells, 20% as cones, and 20% as ganglion cells (7,8). Yet of 105 one- and two-cell clones labeled by the retrovirus, 61 are made up exclusively of cone photoreceptors (3). This observation strikes us as being consistent with the exhaustion of a discrete progenitor pool that undergoes its final round of cell division during this period.

We are uncomfortable with Cepko's statement that 'arguments about models regarding mechanisms are not productive and obscure the interesting story that is currently unfolding...' The ain of our review was to preserve a bit of intellectual space for arguments in favor of lineage. Discussion and dialectics should continue to influence the design of experiments on mechanisms of cell determination. It would be more productive if these studies could be interpreted in an environment in which more than one hypothesis is considered.

Finally, Cepko writes in her letter that she and her colleagues presented 'hypotheses consistent with the data, rather than conclusions'. However, the title of her paper with Turner and Snyder (3), 'Lineage-independent determination of cell type in the embryonic mouse retina' strikes us as a conclusion, one that could have heritable effects on the way in which future cycles of research are carried out.


  1. Williams, R.W. and Goldowitz, D. (1992) Trends Neurosci. 15, 368-373
  2. Holt, C.E., Bertsch, T.W., Ellis, H.M. and Harris, W.A. (1988) Nature 333, 737-741
  3. Turner, D.L., Snyder, E.Y. and Cepko, C.L. (1990) Neuron 4:833-845
  4. Jessell, T.M and Schacher, S. (1991) in Principles of Neuroscience, 3rd edn (Kandel, E.R., Schwartz, J.H., and Jessell, T.M., eds), pp. 887-907, Elsevier
  5. Williams, R.W. and Goldowitz, D. (1992) Proc. Natl. Acad. Sci. USA 89:1184-1188
  6. Drager, U.C. (1985) Proc. R. Soc. London Ser B. 224, 57-77
  7. Carter-Dawson, L.D. and LaVail, M.M. (1979) J. Comp. Neurol. 188, 263-272
  8. Young, R.W. (1985) Anat. Rec. 212, 199-205

Since 11 August 98


Neurogenetics at University of Tennessee Health Science Center

Print Friendly Top of Page

Home Page  |  Genome DBs  |  Phenome DBs  |  Publications  |  People & Associates
Mouse Brain Library  |  Related Sites  |  |   MBL.ORG

Robert W. Williams | Alex Williams © 2002, modify this page