Development 114, 271-283 (1992)Printed in Great Britain © The Company of Biologists Limited 1992

271

Lineage of radial glia in the chicken optic tectum

GRACE E. GRAY and JOSHUA R. SANES*

Department of Anatomy and Neurobiology, Washington University' School of Medicine, St Louis, MO 63110, USA

•Author for correspondence

Summary

In many parts of the central nervous system, theelongated processes of radial glial cells are believed toguide immature neurons from the ventricular zone totheir sites of differentiation. To study the clonalrelationships of radial glia to other neural cell types, weused a recombinant retrovirus to label precursor cells inthe chick optic tectum with a heritable marker, the E.coli lacZ gene. The progeny of the infected cells weredetected at later stages of development with a hLsto-chemical stain for the lacZ gene product. Radial gliawere identified in a substantial fraction of clones, andthese were studied further. Our main results are thefollowing, (a) Clones containing radial glia frequentlycontained neurons and/or astrocytes, but usually notother radial glia. Thus, radial glia derive from amultipotential progenitor rather than from a committedradial glial precursor, (b) Production of radial gliacontinues until at least embryonic day (E) 8, after thepeak of neuronal birth is over (~E5) and after radialmigration of immature neurons has begun (E6-7).

Radial glial and neuronal lineages do not appear todiverge during this interval, and radial glia are amongthe last cells that their progenitors produce, (c) As theymigrate, many cells are closely apposed to the apicalprocess of their sibling radial glia. Thus, radial glia mayfrequently guide the migration of their clonal relatives.(d) The population of labelled radial glia declinesbetween E15 and E19-20 (just before hatching), concur-rent with a sharp increase in the number of labelledastrocytes. This result suggests that some tectal radialglia transform into astrocytes, as occurs in mammaliancerebral cortex, although others persist after hatching.To reconcile the observations that many radial glia arepresent early, that radial glia are among the lastoffspring of a multipotential stem cell, and that mostclones contain only a single radial glial cell, we suggestthat the stem cell is, or becomes, a radial glial cell.

Key words: Lineage, radial glia, retrovirus, optic tectum,differentiation.

Introduction

Radial glia are a prominent cell type in many parts ofthe vertebrate central nervous system (Ramon y Cajal,1911; Levitt and Rakic, 1980; Edwards et al., 1990).Their cell bodies are confined to the ventricular zonebut their processes span the width of the neural tube: ashort basal process runs from the cell body to theventricular surface, and a long apical process ascends tothe pia where it terminates in an endfoot. In cerebralcortex, where they have been studied most intensively,radial glia appear very early in development and thendisappear perinatally (Ramon y Cajal, 1911; Schmecheland Rakic, 1979a; Levitt and Rakic, 1980; Misson et al.,1988b; Edwards et al., 1990). The shape of the radialglia and their transient nature are consistent with theidea that one of their functions is to guide immatureneurons from their birthplace in the ventricular zone totheir ultimate destination in the cortical plate. Ultra-structural studies in vivo (Rakic, 1972; Gadisseux et al.,1990), and studies of neuronal-glial interactions in vitro

(reviewed in Hatten, 1990), have provided support forthis hypothesis.

Although considerable information is available aboutthe structure and fate of radial glia (reviewed inEdwards et al., 1990; Misson et al., 1991; Cameron andRakic, 1991), relatively little is known about theirorigins and early development. Among questions thatremain unanswered are: What is the nature of theprogenitor that gives rise to radial glia? Is it committedto the production of radial glia or is it multipotential? Isthere any clonal relationship between radial glia and theneurons that migrate along them? In addition, the

272 G. E. Gray and J. R. Sanes

et al., 1990; Takahashi et al., 1990) and eventuallydisappear; several lines of evidence suggest that thisdisappearance reflects a transformation of radial gliainto astrocytes (Ram6n y Cajal, 1911; Choi andLapham, 1978; Schmechel and Rakic, 1979a; Pixley anddeVellis, 1984; Voigt, 1989; Cullican et al., 1990).

Here, we have used retrovirus-mediated gene trans-fer (Sanes et al., 1986; Price et al., 1987) to analyze thelineage, differentiation, and fate of radial glia in thechicken optic tectum. The optic tectum is a usefulsystem for such studies because of its regular structuralorganization and because of its accessibility at earlyembryonic stages. In previous work, we used recombi-nant retroviruses as clonal markers to study thegenealogical relationships and migratory paths of tectalneurons and astrocytes (Gray et al., 1988; Galileo et al.,1990; Gray and Sanes, 1991). We showed that severaltypes of neurons and two types of astrocytes derivefrom a common precursor in the tectum, and that manyneurons migrate along bundles of radial glial processesto their laminar destinations. To extend this analysis,we have now used similar methods to demonstrateclonal relationships among neurons, astrocytes and theradial glia that guide their migration.

Materials and methods

Injection of virusFertile White Leghorn chicken eggs were obtained fromSPAFAS (Roanoke, IL), and incubated on their sides at37.5°C in a Humidaire hatcher (New Madison, OH). Forlineage analysis, a retroviral concentrate was pressure-injected into the right tectal ventricle of staged (Hamburgerand Hamilton, 1951) embryos. The virus used, LZ10, is arecombinant Rous sarcoma virus in which the structural genesfrom wild-type virus are replaced with the E. coli fi-galactosidase (lacZ) gene. Virus was produced, collected,concentrated, and injected as previously described (Gray etal., 1988; Galileo et al., 1990). Following injection, eggs weresealed with transparent tape and returned to the incubatoruntil they reached an appropriate developmental stage.

FixationEmbryos were killed at E10, 12, 15, or 19-20 (approximatestages 36, 38, 41, and 45, respectively; hatching is at E20-21).Whole brains were removed from E10-15 embryos, and fixedby immersion in 2% paraformaldehyde plus 0.03% gJutaral-dehyde in 0.1 M phosphate buffer, pH 7.4. E19-20 embryoswere anesthetized with ether and perfused intracardially withthe same fixative before dissection. Tecta to be stained forlacZ were fixed for 1-2 h at room temperature. Those used forimmunohistochemistry or Dil labeling were fixed for up to 1week at 4°C.

X-gal histochemistryFixed tecta were sectioned perpendicular to their long axes at100-200 /an on a Vibratome (Pelco, Redding, CA). Sectionswere placed directly into a solution of 5-bromo-4-chloro-3-indolyl-/S-D-galactoside (X-gal), potassium ferrocyanide, andpotassium ferricyanide, prepared as described previously(Gray et al., 1988). Sections were incubated overnight atroom temperature, washed with phosphate-buffered saline(PBS; 150 mM sodium chloride, 15 mM sodium phosphate,

pH 7.3), dried onto subbed slides and mounted with polyvinylalcohol (Mowiol, Hoechst Celanese Corp., Sommerville, NJ).

ImmunohistochemistryFixed tecta were sunk overnight in 30% sucrose in PBS, thenfrozen in 2-methylbutane that had been cooled by liquidnitrogen. Sections were cut on a cryostat at 15-20 /an,mounted on subbed slides, air-dried, and frozen at —20°Cuntil they were stained. Sections were incubated with primaryantibodies overnight at 4°C, washed with PBS and stained 1 hat room temperature with secondary antibodies. Slides weremounted in 90% glycerol, 10% PBS, supplemented withparaphenylenediamine. Some sections were counterstainedwith bisbenzamide (1 /ig/ml; Hoechst 33258; Sigma, St. Louis,MO) to reveal nuclei. Primary antibodies used were R5, amouse monoclonal IgM that selectively stains radial glia(Drager et al., 1984; generously provided by Ursula Drager,Harvard University); H5, a mouse monclonal IgG that reactswith the same antigen as R5 (Herman et al., 1991); and arabbit anti-lacZ antiserum, generated in our laboratory usingpurified E. coli /J-galactosidase (Sigma, St. Louis, MO) as theimmunogen. Secondary antibodies were rhodamine-conju-gated goat-anti-mouse IgG+IgM and fluorescein-conjugatedgoat-anti-rabbit IgG (Organon Teknika-Cappel, Malvern,PA, and Boehringer-Mannheim, Indianapolis, IN).

DilDil (Molecular Probes, Eugene, OR) was dissolved at 2.5mg/ml in absolute ethanol as suggested by Honig and Hume(1986), and injected into the ventricles of fixed E10 tecta usinga 1 ml tuberculin syringe and a 25 gauge needle. Tecta wereincubated at room temperature for ~1 week, then sectionedon a Vibratome at 50-100 /an and examined with rhodaminefluorescence optics. The fine crystals that formed in theventricle often stained only isolated radial glial fibers.

Determination of number of radial gliaA ~2 mm x 2 mm piece was cut from the center of six E15and seven E19 tecta that had been fixed in 2% paraformal-dehyde. The pieces were stained with R5 antibody asdescribed above, then mounted whole on slides. For eachtectum, 5 fields of glial endfeet (or terminal glial shafts) werephotographed, and the number of endfeet per unit area wasdetermined from micrographs. To normalize for surface area,three tecta each from E15 and E19 embryos were sectionedcompletely at 200 /an on a Vibratome (Pelco, Inc.), and thesurface area of the tecta was calculated from the lengths of thepial surface in each section.

Results

Identification of radial gliaTo study cell lineage in the optic tectum, we inject arecombinant Rous sarcoma virus that bears the lacZgene into the ventricle of the embryonic mesencepha-lon. The viral genome is incorporated into the DNA ofdividing infected precursor cells, and the lacZ gene istranscribed and translated. LacZ is readily detected inthe progeny of infected precursors using either ahistochemical stain or antibodies to lacZ.

When we injected virus into the tecta of embryonicday (E) 3-5 chicken embryos and examined them onE15, we found that some of the labelled cells wereelongate in shape: their cell bodies lay in the ventricular

Lineage of radial glia 273

B

Fig. 1. Morphology of radial glia in the optic tectum. Radial glia were labelled by X-gal following infection at E5 (A) orE7 (B), by Dil applied to the ventricle (C, D, E), or by staining with R5 (F). Tecta are from E10 (C,D) or E15 (A,B,E,F)animals. Bars, 50 urn (A-D); 100 /OTI (E,F).

zone, and a single long, varicose process extended fromeach cell body to the pia (Fig. 1A,B). Short lateralprocesses were occasionally present as well,, but theapical process was unbranched. These cells were similarin size and shape to radial glia revealed by either of twomethods (Vanselow et al., 1989; Gray and Sanes, 1991):uptake of dil from crystals applied to the ventricular

surface (Fig. 1 C-E), and immunohistochemical stainingwith the monoclonal antibodies R5 (Fig. IF) or H5 (Fig.2B), both of which recognize an intracellular antigen inradial glia (Herman et al., 1991). Importantly, theperiodic thickenings of the apical processes observed inX-gal stained cells (Fig. 1A and B) were also evident indil-stained cells (Fig. 1C and D), indicating that these

274 G. E. Gray and J. R. Sanes

Fig. 2. Sections double labelled with anti-lacZ and H5, amonoclonal antibody that labels radial glia. (A,C) LacZ-positive fibers, (B,D) same fields labelled with H5. ThelacZ-positive fibers are stained by H5, demonstrating thatthey are the apical processes of radial glia. The tectum wasinjected with retrovirus at E4 and stained at E15. Bars, 25

swellings were cytoplasmic varicosities rather thanseparate adherent cells. This point (which was pre-viously made by Ramon y Cajal, 1891) was confirmedby staining with bis-benzamide to reveal DNA, andsnowing that the varicosities were anucleate (data notshown).

To confirm the identification of the elongated lacZ-positive cells as radial glia, we prepared cryostatsections from virus-infected tecta, and double-labelledthem with antibodies to lacZ plus R5 or H5. Fig. 2shows that the lacZ-positive apical processes were alsoH5-positive. Together, these results demonstrate thatsome of the cells infected by retrovirus on E3-5 wereprogenitors of radial glia.

Clonal relatives of radial gliaFor lineage analysis, tecta were injected with LZ10 atE3-4, fixed at E15, and stained histochemically for lacZ.Ousters of lacZ-positive cells were identified as clonesby criteria that have been detailed elsewhere (Gray etal., 1988, 1990). Approximately one third of the clonesanalyzed (for example, 35/106 from one set of tectainjected on E3, and 72/218 from a set injected on E4)contained cells that were unambiguously classifiable asradial glia on morphological grounds: they had a somain the ventricular zone and an apical process thatextended at least one third of the distance to the pialsurface. These clones were studied further.

Fig. 3 illustrates several features of radial glia-containing clones. First, some clonal relatives of radialglia were clearly neurons, as determined by their roundnuclei, abundant cytoplasm, and the characteristicarrangements of their processes (Fig. 3D and E). Manyother cells were likely to be neurons as well, but theirsmall size and poorly stained processes precluded theirpositive identification. Nonetheless, it is likely thatmost progenitors that give rise to radial glia give rise toneurons as well. Second, many clones containedastrocytes. These could be identified on the basis oftheir small, irregularly shaped cell bodies, and highlybranched filamentous processes (Fig. 3C). (Immunohis-tochemical confirmation of the identity of lacZ-positivecells as neurons and astrocytes has been reportedpreviously; see Galileo et al., 1990.) Third, many clonescontained both neurons and astrocytes as well as radialglia (Fig. 3A,B). Together, these results demonstratethat individual tectal progenitors labelled on E3-4 cangive rise to neurons, astrocytes, and radial glia.

Finally, most radial glia did not have other radial gliaas clonal relatives (Fig. 3A and B). For example, in oneset of 56 radial glia-containing clones from tectainfected at E3, there were 44 (79%) with only a singleradial glial cell, 9 (16%) with two, and 3 (5%) with four.In animals infected at E4, 96% (54/56) of radial glia-containing clones bore one radial glia, and the remain-der had two each. We showed previously that clonesmarked on E3 are composed of one or a few radialstrands of cells, whereas nearly all clones marked at E4or later consist of only a single strand (Gray et al.,1988). Each strand contains -10-15 cells by E8, whencell division is nearly complete, and may represent theprogeny of a single stem cell (Gray and Sanes, 1991). Ofthe 14 clones that contained >1 radial glial cellfollowing injection at E3 or E4, only 2 contained >1radial glial cell within a single strand. Thus, our resultsare consistent with the possibility that most stem cellsproduce no more than one radial glial cell.

Time of radial glia productionHaving found that radial glia and neurons were bothpresent in clones marked by injections on E3-4, weasked how long radial glia continue to be produced andwhether the radial glial and neuronal lineages eventu-ally diverge. Animals were injected with virus at E3, 4,5, 6, 7, 8, 10, and 12, then sacrificed and analyzed at

Lineage of radial glia 275

Fig. 3. Clonal relatives of radial glia. (A,B) Camera lucida drawings of clones marked by infection on E3 (A) or E4 (B),and examined at E15. Cells with neuronal and astrocyte morphologies are found in association with the single radial glialcell in each clone. Arrows in A and B indicate cells shown in C-F. (C) A cluster of astrocytes. (D) Fusiform neurons. (E)A large pyriform neuron. (F) A radial glial cell body and its proximal process. Bars, 50 fjan (A,B); 20 /an (C-F). P, pia;TP, tectal plate; 13, lamina 13 (also called stratum griseum centrale); IZ, intermediate zone; VZ, ventricular zone; V,ventricular surface.

E15. Numerous radial glia were labeled after injectionat E3-8, but few lacZ-positive cells of any type wereseen in tecta injected on E10-12. Although this resultmay indicate that radial glia are not born after E8, it isalso possible that virus was unable to infect progenitors

efficiently at late stages, for example, because thedeveloping basal lamina may impede access of virus tocells in the ventricular zone. Thus, we do not knowwhen production of radial glia ceases, but we canconclude that it proceeds at least until E8, which is after

276 G. E. Gray and J. R. Sanes

the peak of neuronal generation is over (~E5; La Vailand Cowan, 1971) and after radial migration of neuronshas begun (~E6-7; Gray and Sanes, 1991).

Fig. 4 shows typical radial glia-containing clonesgenerated from E5 and E8 infections. As expected, theaverage number of cells per clone was smaller at E8than at E5 and smaller at E5 than at E3 (compare withFig. 3). As in tecta injected earlier, nearly all clones thatcontained radial glia included only a single radial glialcell (30/30=100%, 60/61=98%, and 25/26=96% follow-ing injection at E5, E7 and E8, respectively). Further-more, even in small clones, which consisted of only 3-4cells, some of the clonal relatives of radial glia wereidentifiable as neurons (Fig. 4A; see also Figs 4B and7A). Thus, we obtained no evidence that the neuronaland radial glial lineages diverge.

The presence of labeled radial glia in tecta injected atE7 and E8, late in the period of neuronal birth (LaVailand Cowan, 1971), raised questions about the relativetiming of radial glial and neuronal production. Ifproduction of radial glia peaked before that of neurons,radial glia should make up a smaller percentage of thetotal population of lacZ-positive cells with successivelylater ages of infection. If, however, most radial glialwere produced after most neurons were born, thepercentage of labeled radial glia in late-infected tectashould increase and the percentage of neurons shoulddecrease. To distinguish these alternatives, we classi-fied lacZ-positive cells in a set of tectal sections fromeach age of injection as "radial glia", "astrocytes", or"neurons and others". Neurons were not countedseparately because, as noted above, many cells thatwere likely to be neurons could not be distinguishedunambiguously; in fact, most of the cells in the thirdcategory are likely to be immature neurons. As shownin Fig. 5, successively later injections label populationsof cells that are increasingly enriched in radial glia. Thistrend is evident whether or not astrocytes are includedin the totals - i.e., whether the frequency of labeledradial glia is calculated with respect to all labeled cells,or to only the population of presumptive neurons.Thus, radial glial production continues even afterneuronal production has begun to wane. Takentogether with the lineage analysis described above,these results indicate that radial glia are likely to beamong the last cells produced by their multipotentialprogenitors.

Radial glia as guides for clonal relativesIn E8-12 tecta, when cells are actively migrating fromthe ventricular zone to the tectal plate, most of theradially migrating cells are closely associated withfascicles of radial glial processes (Gray and Sanes,1991), suggesting that radial glia act as migratory guidesin tectum, as they are thought to do in mammaliancortex (Rakic, 1972; Gadisseux et al., 1990). In tectaexamined at E15, when migration is largely complete,clonal relatives of a radial glial cell were often found inclose association with that cell's process, sometimesapparently contacting it (e.g., Fig. 3). Taken together,these two observations raised the possibility that tectal

AE5-I5

— P -E8-I5

TP

TP

1313

Fig. 4. Radial glial clones from tecta injected at E5 (A) orE8 (B) and stained for lacZ at E15: camera lucidadrawings. (A) A clone consisting of a single radial glial cellwith two associated cells, one of which is a large multipolarneuron. In many clones, multipolar neurons migTatetangentially from the radial array (Gray and Sanes, 1991),but this one has apparently not done so. (B) A clone thatconsists of one radial glial cell plus one associated cell(arrow) that is probably a small neuron. Fewer cells areassociated with the radial glia in clones marked at thesestages, compared with clones marked at E3 and E4 (Fig.3), but neuronal and glial lineages have not diverged. Bars,50 /an. Abbreviations, as Fig. 3.

Lineage of radial glia 277

CD

50

40

9 3O

t 20

10

1 1

3 4 5 6 7 8 9AGE OF INJECTIONEMBRYONIC DAY)

Fig. 5. Percentage of lacZ-positive cells that are radial gliafollowing infection at various ages. All labeled cells werecounted in 34-42 sections from each of 5-10 animals perstage, and the percentage of radial glia was calculated withrespect to presumptive neurons (open circles) or to alllabeled cells (closed circles). By either measure, radial gliaare preferentially labelled by later infections.

neuroblasts might use sibling radial glia as migratoryguides.

To test this idea, we analyzed a set of clones at E10,during the peak of radial migration. For this exper-iment, tecta were injected on E4 (rather than on E3, asin Gray and Sanes, 1991), to generate relatively smallclones (single radial strands; see above). In these tecta,we found numerous clones that contained radial glia.The majority of these clones consisted of tight radialarrays of cells in which a single radial glial cell wasaccompanied by radially migrating cells whose somataand processes were closely apposed to the radial glialprocess (Fig. 6A). In other cases, individual clonescontained several cells with somata in the ventricularzone and intertwined processes that extended apically(Fig. 6B); it is likely that at least one cell in each ofthese clones was a radial glial cell and that the fullextent of its apical process was obscured by the opposedprocesses of its migrating siblings. In clones of bothtypes (Fig. 6A and B) several cells were frequentlyarranged along the length of the same process, both inthe intermediate zone and in the tectal plate. The moreapical cells in clones were sometimes displaced tangen-tially from the radial glial process (Figs 6A and 7A)consistent with the possibility that they had reachedtheir laminar destinations and ceased migration. Inaddition, in some clones, cells within an individualradial strand appeared to be aligned along two or threeadjacent fascicles, only one of which contained alabeled radial glia; there was not, therefore, a completerestriction of migrants to a single fascicle (see also Grayand Sanes, 1991). Nonetheless, in the majority of clonesthat contained both radial glial cells and radiallymigrating cells, the majority of the migrants werealigned along the fascicle that contained the process ofthe labeled radial glia. These results raise the possibility

P -

Fig. 6. Guidance of radially migrating cells by sibling radialglia. Tecta were injected at E4 and stained at E10, duringthe peak of radial migration. (A) Typical clone, in whichone radial glial cell is clearly identifiable. The migratingcells (arrows) are tightly associated with the glial process,one in the intermediate zone and two in the tectal plate. Afourth sibling, the most superficial, is displaced laterally by—20 /an. (B) A larger clone with several cells in theventricular zone that have ascending processes; at least oneof these is likely to be a radial glial cell. Bars, 50 /zm.Abbreviations, as Fig. 3.

that radial glial cells guide their clonal relatives from theventricular zone to their destinations in the tectal plate.

Fate of radial gliaTo follow the fate of radial glia over time, we injected

278 G. E. Gray and J. R. Sanes

groups of tecta at the same stage, and then fixed andstained subgroups at various times between ElO andE20. Between ElO and E15, as noted above, radial gliawere abundant, and they were accompanied by differ-entiating cells that could sometimes be identified asimmature neurons (Figs 3, 4, 6 and 7A). At E19-20, onthe other hand, a few clones were encountered thatcontained radial glia (Fig. 7B), but the vast majority ofclones contained no cells whose somata lay in theventricular zone. To quantify the apparent loss of radialglia, we calculated the percentage of all lacZ-positivecells that were radial glia as a function of stage ofsacrifice for three experiments in which tecta wereinjected at E3, E7, or E8, respectively. As shown inTable 1, there is a sharp decline in the population oflabeled radial glia during the period between ElO andE19, and this decline does not depend greatly on theage of infection.

There are several possible reasons for the decline inlacZ-positive radial glia. One possibility is that lacZmay not be expressed well in radial glia at lateembryonic stages. Another is that radial glia may dieduring this period. While we cannot exclude either ofthese alternatives, three observations provide indirectevidence for a third possibility: that at least some radialglia transform into astrocytes. First, between E15 andE19-20, clusters of astrocytes appeared throughout thetectal layers; occasionally these clusters were arrangedradial to one another. The appearance of astrocytescorrelated with the decline in radial glia. Second, intecta examined before E19, cells were often presentthat were intermediate in form between radial glia andastrocytes. These cell types were especially abundant atE15, which is approximately when the first immunohis-tochemically identifiable astrocytes appear in thetectum (Linser and Perkins, 1987). Fig. 8 illustrates thedifferent radial-glia-form cell types seen in the tectumat E12, 15, and 19, and suggests a possible progression

Table 1. Percentage of radial glia at different stages ofsacrifice

Day ofinjection

E3

E7

E8

Day ofkilling (n)

E15 (6)E19 (4)

ElO (3)E15 (7)E20(5)

E12 (3)E15 (7)E19 (4)

All cells(%)

06±1.40.7±0.7

35±6.222±4.705±1.2

45±8.920±2.402±0.4

Radial glia

All butastrocytes (%)

06±1.40.7±0.7

35±6.230±6.406±1.7

45±8.944±9.713±4.4

Animals were injected with virus on the day of incubationshown in the first column and killed on the day shown in thesecond column. Radial glia and other cell types were counted from2-7 sections from each animal. Ousters of astrocytes were countedas one "cell", based on the assumption that clusters arise from asingle cell dividing in situ. Percentages were determined for eachanimal in the set and then averaged; values given are means ±s.e.m. Values in parentheses (n), number of animals in the set.

BE3-I9

P—

TP

Fig. 7. LacZ-positive clones from ElO and E19 tecta. (A)A single radial glial cell associated with an immatureneuron (arrow), marked by injection on E7 and examinedat ElO. The neuron appears to be parting from the radialglial process. (B) A single radial glial cell forms the core ofa large clone that was marked by injection on E3 andexamined at E19. Clones of this sort are rare, as few lacZ-positive radial glia can be found at this age. Bars, 50 um.Abbreviations, as Fig. 3.

AE8-I2

Lineage of radial glia 279

P —

Fig. 8. Intermediate cell types suggest that radial glia transform into astrocytes. Camera lucida composites of glia fromtecta analyzed at E12 (A), E15 (B) and E19 (C). (A) Radial glia are numerous and astrocytes rare at E12. (B) Several celltypes seen at E15 that may represent the progressive transformation of radial glia into astrocytes. They are arranged herein a hypothetical sequence of transitional forms, suggested by studies in mammalian cerebral cortex (e.g., Schmechel andRakic, 1979; Misson et al., 1988b). (C) Ousters of astrocytes that appear as the numbers of labelled radial glia decline.Bar, 100 /an. Abbreviations, as Fig. 3.

in the transformation of radial glia into astrocytes.Third, if radial glia transform into astrocytes, onewould expect the ratio of radial glia to astrocytes to beconstant at any single time of sacrifice, regardless of theage of injection. In fact, there was no systematicvariation of this ratio with age of injection. Forexample, in animals analyzed at E15, the ratio of radialglia to all glia (radial glia plus astrocytes) was 0.59,0.45,0.55, and 0.38 after infection at E4, E5, E7, and E8,respectively. Finally, it is important to note that thesefindings are in agreement with evidence that radial glia

in mammalian cortex transform into astrocytes (seeDiscussion).

If the decline in lacZ-positive radial glia is represen-tative of the general behavior of all radial glia, oneshould see a decline in the total population of radial gliabetween E15 and 19. However, Vanselow et al. (1989)saw abundant radial glia at the time of hatching. Wetherefore compared the numbers of radial glia presentat E15 and 19, by labeling pieces of fixed tecta with R5,and counting the glial endfeet that lie just beneath thepia, as described in Materials and methods. After

280 G. E. Gray and J. R. Sanes

normalizing for changes in surface area we found adecline in radial glial endfeet of 14% between E15 and19. This difference is less than would be predicted fromthe data in Table 1, although the discrepancy could beexplained if (a) subpial radial glial branching increasedwith age, as has been seen in rodents (Edwards et al.,1990), (b) the transformed radial glia were replaced byproduction of new radial glia from unmarked progeni-tors, (c) transitional forms (Fig. 8) maintained pialendfeet after their somata left the ventricular zone, or(d) a decline in lacZ expression at late ages exaggeratedthe numbers of radial glia that were lost. Despite thediscrepancy, the fact that there is a decrease in radialglial numbers between E15 and 19 is consistent with theidea that some radial glia are transforming intoastrocytes.

Discussion

We have used retrovirus-mediated gene transfer tolabel radial glial cells and their clonal relatives in thechicken optic tectum. Our results show that radial gliaarise from a progenitor that also generates neurons andastrocytes, but generally does not produce more thanone radial glial cell. Radial glia are produced until atleast E8, which is after many neurons are born and aftermigration has commenced. During migration manyneurons associate with clonally related radial glia.Subsequent to migration, at least some radial glia mayturn into astrocytes, although many others remain inthe tectum at least until hatching.

LineageIn previous studies of cell lineage in the optic tectum,we demonstrated that several types of neurons and twokinds of astrocytes can arise from a common precursor(Gray et al., 1988; Galileo et al., 1990; Gray and Sanes,1991). In those studies, in which clones were analyzed atE19 (just before hatching), we identified few lacZ-positive radial glia. We therefore suspected that mostradial glia were generated early in development, beforewe had injected virus to mark progenitors. Sub-sequently, we were struck by the abundance of lacZ-positive radial glia in a set of clones analyzed at E15.Eventually, we learned that many lacZ-positive radialglia disappear between E15 and E19, although a fewremain. By combining studies at E15 and E19, we wereable to compare radial glia-containing clones with thosewe already knew to contain neurons and astrocytes.Together with our previous work, the data presentedhere indicate that radial glia derive from the same typesof progenitors that give rise to several other tectal celltypes. We do not know whether these progenitors areall of a single type, or whether several distinct types ofprogenitors exist that generate overlapping sets of cells.However, our results suggest that neuronal and radialglial lineages do not diverge, even at late stages ofneurogenesis.

An additional feature of radial glial lineage is theirlikely transformation into astrocytes after their role as

migratory guides has ended. Such a transformation wasorginally postulated to occur in chicken spinal cord byRam6n y Cajal (1911), on the basis of transitionalcellular forms observed in Golgi stained material.Ram6n y Cajal (1911), Choi and Lapham (1978), andSchmechel and Rakic (1979a) later extended this idea tothe mammalian cortex. More recently, antibodies thatstain mammalian radial glia, the transitional forms, andastrocytes, have been used to demonstrate a relation-ship among these cells in vivo (Levitt and Rakic, 1980;Pixley and deVellis, 1984; Misson et al., 1988a; Voigt,1989; Takahashi et al., 1990) and in vitro (Cullican etal., 1990). Finally, Voigt (1989) labelled radial glial cellswith an intracellular dye at an early stage, and lateridentified astrocytes containing the label, thus provid-ing the most direct evidence of transformation to date.Together, these studies provide substantial support forthe idea of radial glial transformation.

Three lines of evidence presented here indicate that asimilar transformation occurs in tectum. First, thenumber of lacZ-positive radial glia declines during theperiod that lacZ-positive astrocytes appear. Second,cells whose morphology is intermediate between radialglia and astrocytes appear in the tectum as the numbersof radial glia decline. Finally, the ratio of radial glia toastrocytes remains fairly constant, regardless of age ofinfection. On the other hand, a substantial number ofradial glia remain in the tectum of hatchlings, indicatingthat many do not transform into astrocytes, or do soonly posthatching.

The results of our clonal analysis permit us toevaluate several alternative models of radial gliallineage (Fig. 9). Perhaps surprisingly, our resultsappear to exclude, for the tectum, the scheme that iscurrently favored for mammalian cerebral cortex - i.e.,that neuronal and radial glial lineages diverge early,with radial glia giving rise to many (if not all) astrocytes,and a separate class of neuroepithelial cells giving riseto neurons (Fig. 9A). The idea that radial glial andneuronal lineages diverge early in cortex is supportedby observations on primates by Levitt et al. (1980,1981,1983). They showed that two populations of mitoticallyactive cells can be distinguished in the cortical ventricu-lar zone on the basis of immunoreactivity for theastrocyte-specific intermediate filament, glial fibrillaryacidic protein (GFAP), and that the GFAP-positivesubpopulation corresponds in large part to radial glia.In light of evidence that radial glia transform into(GFAP-positive) astrocytes (see above), they arguedthat the GFAP-negative progenitors give rise toneurons (reviewed in Cameron and Rakic, 1991).Misson et al. (1988a, 1988b, 1991) have made a parallelset of observations in murine cortex, using the mono-clonal antibodies RC1 and RC2 to distinguish radial gliafrom other neuroepithelial cells, and to follow thetransformation of radial glial into astrocytes. Ourretrovirus-based clonal analysis in murine cortex (Lus-kin et al., 1988), and that of Price and Thurlow (1988) inrat, have provided direct evidence that neurons andastrocytes have distinct progenitors, and thus indirectlysupport the model of Rakic and others. However,

Lineage of radial glia 281

Fig. 9. Models of radial glial lineage. (A) Scheme currentlyfavored for mammalian cortex, in which radial glia (RG)give rise to astrocytes (A), and neurons (N) arise fromseparate progenitors (NB) (see Cameron and Rakic, 1991).(B) Scheme A modified to fit data that tectal progenitorsgive rise to both neurons and astrocytes (see Galileo et al.,1990). (C) Scheme in which radial glia are early-bornprogeny of a multipotential stem cell. (D) Schemesuggested by data in this paper, in which a radial glia cellis one of the last progeny of a multipotential stem cell. Seetext for details.

clonal analyses that include radial glia themselves havenot yet been reported.

In tectum, radial glia, neurons, and astrocytes allfrequently descend from a single precursor, and radialglia may be among the last cells to be produced. Thuswe can exclude not only cortex-like schemes (Fig. 9A)but also modified schemes in which the radial gliallineage diverges early from a neuron-glia lineage (Fig.9B), or in which radial glia are among the first-bornprogeny of a multipotential progenitor (Fig. 9C).Instead, our results favor schemes such as that shown inFig. 9D, in which: 1) a single multipotential progenitorproduces neurons, astrocytes, and radial glia; 2) radialglia are among the last-born progeny of their progeni-tor; 3) progenitors marked at E3 or later produce only asingle radial glia; and 4) some but not all radial gliatransform into astrocytes or astrocyte precursors late inembryogenesis.

The progenitor in Fig. 9D is drawn as a stem cell - acell which undergoes asymmetric divisions to produceanother mitotically active stem cell plus a postmitoticdaughter. This feature is based on indirect evidence thatproliferative and stem cell divisions occur sequentiallyin tectum. First, progenitors divide symmetrically(proliferatively) to produce equivalent, horizontallydisplaced daughters that populate the ventricular zone.Subsequently, these cells use a stem cell mode ofdivision to produce radial arrays of postmitotic neuronsand glioblasts (Gray et al., 1988,1990). If this scheme iscorrect, our new results raise the possibility that thestem cell is itself an immature radial glial cell. This ideawould provide a way to reconcile the notion that many

radial glia are present early with the observations thatmany tectal radial glia are generated late, and wouldaccount for the result that <5% of the clones markedon E4 or later contain >1 radial glia. In fact, mitoticdivisions of differentiated radial glia have been ob-served in adult avian telencephalon (Alvarez-Buylla etal., 1990), denervated adult amphibian optic tectum(Stevenson and Yoon, 1981), and embryonic mam-malian cerebral cortex (Schmechel and Rakic, 1979b;Levitt et al., 1983; Misson et al., 1988b). In cortex, ithas been supposed that mitoses of radial glia generateadditional radial glia and, ultimately, astrocytes. Inadult birds, radial glia have been suggested to beprogenitors of neurons, based on the coincidence ofsites of mitotic activity with sites at which neurons areknown to arise (Alvarez-Buylla et al., 1990). Finally,Fredricksen and McKay (1988) used immunohisto-chemical similarities between neuroepithelial cells andradial glia to suggest that radial glia are stem cells.Thus, it is possible that radial glia are mitotically activestem cells in many parts of the nervous system, but giverise to different types of progeny in different regions.

One implication of the notion that radial glia are stemcells (or that stem cells become radial glia once theyhave ceased to produce other cell types) is that eachclone should contain a radial glial cell. However, only~35% of the clones marked by injection at E3 or E4and analyzed at E15 contain unambiguously identifiableradial glia. Careful examination of the remaining clonesshows, however, that another ~40% contained cellsthat were likely to be radial glia, but did not meet thestrict criteria we used in classifying cells for clonalanalysis - for example, some of these contained a somain the ventricular zone with an apical process thatextended less than one third of the way to the pia, andothers contained "transitional forms" (see Fig. 8)presumably derived from radial glia. Interestingly,inclusion of these "likely" radial glia did not greatlyincrease the number of clones that contained >1 radialglial cell. Therefore, few clones contained >1 radialglia-like cell per strand, only about 25% contained nosuch cells, and some of the latter might be accounted forby incomplete expression or detection of lacZ. In short,it is possible that the majority of tectal stem cells are, orbecome, radial glial cells.

Radial glia as migratory guidesCurrent interest in radial glia centers in large part ontheir role in guiding the migration of neurons to theirlaminar destinations. Two of our results may beinformative in this regard. The first is that many radialglia are generated late in the period of neurogenesis.Neuronal birth starts around E3 and peaks at E5-6(LaVail and Cowan, 1971). Newly born cells stack up inthe ventricular zone until about E6, when a minority ofcells begins a tangential migration along axons in themarginal zone. Few radially migrating cells penetratethe marginal zone until ~E7, when the first radiallymigrating cells can be found in association with radialglia (Gray and Sanes, 1991). It is possible that the lategeneration of radial glia (or their differentiation from

282 G. E. Gray and J. R. Sanes

stem cells; see above) opens a radial migratory path,thus terminating the recruitment of new neuroblasts tothe tangential pathway and initiating radial migration toform the tectal plate.

The second relevant observation is that there is aregular association between sets of clonally relatedmigrants and their sibling radial glial cell. Mostmigrating tectal cells migrate in association withbundles of radial glial fibers (Gray and Sanes, 1991).Clonally related cells generally migrate along one or afew bundles of radial fibers, resulting in radiallyoriented clones (Gray et al., 1988; Gray and Sanes,1991). Here, we have shown that the bundle that guidesa clone of cells frequently includes the apical process ofa clonally related radial glial cell. In that a number ofradial processes may be available to migrating cellswithin the bundle, we cannot conclude that neuroblastsmigrate only along clonally related glial cells. Lackingultrastructural studies on this point, it is premature tospeculate on whether the association reflects onlyproximity or some form of chemical recognition.Regardless of the mechanism, however, the associationof clonally related radial glia and radially migrating cellsmay be partially responsible for generating the arrays ofcells that became functionally interconnected (Jassik-Gerschenfeld and Hardy, 1984; Rakic, 1988), asdevelopment proceeds.

We thank Jeanette Cunningham and Robin Morris-Valerofor assistance. This work was supported by grants from theMcKnight Foundation and the National Institutes of Health.

References

Alvarez-Buylla, A., Theelen, M. and Nottebohm, F. (1990).Proliferation "hot-spots" in adult avian ventricular zone revealradial cell division. Neuron 5, 101-109.

Cameron, R. S. and Rakic, P. (1991). Glial cell lineage in the cerebralcortex: a review and synthesis. Glia 4, 124-137.

Choi, B. H. and Lapham, L. W. (1978). Radial glia in the human fetalcerebrum: a combined Golgi, immunofluorescent and electronmicroscopic study. Brain Res. 148, 295-311.

Culllcan, S. M., Baumrlnd, N. L., Yamamoto, M. and Pearlman, A.L. (1990). Cortical radial glia: identification in tissue culture andevidence for their transformation into astrocytes. / . Neurosci. 10,684-692.

Drager, U. C , Edwards, D. L. and Barnstable, C. J. (1984).Antibodies against filamentous components in discrete cell types ofthe mouse retina. J. Neurosci. 4, 2025-2042.

Edwards, M. A., Yamamoto, M. and Caviness, V. S., J r (1990).Organization of radial glia and related cells in the developingmurine CNS. An analysis based upon a new monoclonal antibodymarker. Neuroscience 36, 121-144.

Frederiksen, K. and McKay, R. D. G. (1988). Proliferation anddifferentiation of rat neuroepithelial precursor cells in vivo. J.Neurosci. 8, 1144-1151.

Gadlssenx, J. F., Evrard, P., Misson, J. P. and Caviness, V. S. (1989).Dynamic structure of the radial glial fiber system of the developingmurine cerebral wall. An immunocvtochemical analysis. Dev.Brain Res. 50, 55-67.

Gadisseux, J.-F., Kadhim, H. J., van den Bosch de Aguilar, P.,Caviness, V. S. and Evrard, P. (1990). Neuron migration within theradial glial fiber system of the developing murine cerebrum: anelectron microscopic autoradiographic analysis. Dev. Brain Res.52, 39-56.

Galileo, D.S. , Gray, G. E., Owens, G. C , Majors, J. and Sanes, J. R.(1990). Neurons and glia arise from a common progenitor in

chicken optic tectum: demonstration with two retroviruses and celltype-specific antibodies. Proc. Nad. Acad. Sci. USA 87, 458-462.

Gray, G. E., Glover, J. C , Majors, J. and Sanes, J. R. (1988). Radialarrangement of clonally related cells in the chicken optic tectum:lineage analysis with a recombinant retrovirus. Proc. Nail. Acad.Sci. USA 85, 7356-7360.

Gray, G. E., Leber, S. M. and Sanes, J. R. (1990). Migratory patternsof clonally related cells in the develping central nervous system.Experientia 46, 929-940.

Gray, G. E. and Sanes, J. R. (1991). Migratory paths and phenotypicchoices of clonally related cells in the avian optic tectum. Neuron 6,211-225.

Hamburger, V. and Hamilton, H. L. (1951). A series of normal stagesin the development of the chick embryo. J. Morph. 88, 49-92.

Hatten, M. E. (1990). Riding the glial monorail: a commonmechanism for glial-guided neuronal migration in different regionsof the developing brain. Trends Neurosci. 13, 179-184.

Herman, J. P., Victor, J. C. and Sanes, J. R. (1991).Developmentally regulated and domain-specific antigens of radialglia in chick optic tectum. Soc. Neurosci. Abstr. 17, 923.

Honlg, M. G. and Hume, R. I. (1986). Fluorescent carbocyanine dyesallow living neurons of identified origin to be studied in long-termcultures. J. Cell Biol. 103, 171-187.

Jassik-Gerschenfeld, D. and Hardy, O. (1984). The avian optictectum: neurophysiology and behavioral correlations. InComparative Anatomy of the Optic Tectum (H. Vanegas, ed.).(Plenum Press: New York), pp. 649-686.

LaVail, J. H. and Cowan, W. M. (1971). The development of thechick optic tectum. II. Autoradiographic studies. Brain Res. 28,421-441.

Levitt, P., Cooper, M. L. and Rakic, P. (1981). Coexistence ofneuronal and glial precursor cells in the cerebral ventricular zone ofthe fetal monkey: an ultrastructural immunoperoxidase analysis. J.Neurosci. 1, 27-39.

Levitt, P., Cooper, M. L. and Rakic, P. (1983). Early divergence andchanging proportions of neuronal and glial precursor cells in theprimate cerebral ventricular zone. Dev. Biol. 96, 472-484.

Levitt, P. and Rakic, P. (1980). Immunoperoxidase localization ofglial fibnllary acidic protein in radial glial cells and astrocytes of thedeveloping rhesus monkey brain. J. Comp. Neurol. 193, 815-840.

Llnser, P. J. and Perkins, M. (1987). Gliogenesis in the avian optictectum: neuronal-glial interactions influence astroglial phenotypematuration. Dev. Brain Res. 31, 277-290.

Luskin, M. B., Pearlman, A. L. and Sanes, J. R. (1988). Cell lineagein the cerebral cortex of the mouse studied in vivo and in vitro witha recombinant retrovirus. Neuron 1, 635-647.

Misson, J.-P., Edwards, M. A., Yamamoto, M. and Caviness, V. S.,Jr (1988a). Mitotic cycling of radial glial cells of the fetal murinecerebral wall: a combined autoradiographic andimmunohistochemical study. Dev. Brain Res. 38, 183-190.

Misson, J.-P., Edwards, M. A., Yamamoto, M. and Caviness, V. S.,Jr (1988b). Identification of radial glial cells within the developingmurine central nervous system: studies based upon a newimmunohistochemical marker. Dev. Brain Res. 44, 95-108.

Misson, J.-P., Takahashi, T. and Caviness, V. S., Jr (1991).Ontogeny of radial and other astroglial cells in murine cerebralcortex. Glia 4, 138-148.

Pixley, S. K. R. and De Vellis, J. (1984). Transition between immatureradial glia and mature astrocytes studied with a monoclonalantibody to vimentin. Dev. Brain Res. 15, 201-209.

Price, J. and Thurlow, L. (1988). Cell lineage in the rat cerebralcortex: a study using retroviral-mediated gene transfer.Development 104, 473-482.

Price, J., Turner, D. and Cepko, C. (1987). Lineage analysis in thevertebrate nervous system by retrovirus-mediated gene transfer.Proc. Nad. Acad. Sci. USA 84, 156-160.

Rakic, P. (1972). Mode of cell migration to the superficial layers offetal monkey neocortex. J. Comp. Neurol. 141, 283-312.

Rakic, P. (1988). Specification of cerebral cortical areas. Science. 241,170-176.

Ramdn y Cajal, S. (1911). Histologiedu systeme Nerveux de L'hommeet des vertebres. Paris: A. Maloine (Reprinted, 1972, by Institute

Lineage of radial glia 283

Ramon y Cajal, Madrid). Vol. I, pp. 619-637, and Vol. n, pp. 216and 859.

Ram6n y Cajal, S. (1891). On the structure of the cerebral cortex ofcertain mammals. La Cellule 7, 125-176. Reprinted in Cajal on theCerebral Cortex, (J. DeFelipe and E.G. Jones, eds.), New York,Oxford University Press, 1988, pp. 23-54.

Sanes, J. R., Rubinstein, J. C. R. and Nicolas, J.-F. (1986). Use of arecombinant retrovirus to study post-implantation cell lineage inmouse embryos. EMBO J. 5, 3133-3142.

Schmechel, D. E. and Raklc, P. (1979a). A Golgi study of radial ghalcells in developing monkey telencephalon: morphogenesis andtransformation into astrocytes. Anat. Embryol. 156, 115-152.

Schmechel, D. E. and Raklc, P. (1979b). Arrested proliferation ofradial glial cells during midgestation in rhesus monkey. Nature 277,303-305. .

Stevenson, J. A. and Yoon, M. G. (1981). Mitosis or radial glial cells inthe optic tectum of adult goldfish. J. Neurosci. 1, 862-875.

Takahashl, T., Misson, J.-P. and Cavtness, V. S., Jr (1990). Glialprocess elongation and branching in the developing murineneocortex: a qualitative and quantitative immunohistochemicalanalysis. J. Comp. Neurol. 302, 15-28.

Vanselow, J., Thanos, S., Godement, P., Hanke-Fahle, S. andBonhoeffer, F. (1989). Spatial arrangement of radial glia andingrowing retinal axons in the chick optic tectum duringdevelopment. Dev. Brain Res. 45, 15-27.

Voigt, T. (1989). Development of glial cells in the cerebral wall offerrets: direct tracing of their transformation from radial glia intoastrocytes. J. Comp. Neurol. 289, 74-88.

(Accepted 11 October 1991)