Glypican-1 controls brain size through regulation of fibroblast growth factor signaling in early neurogenesis

Background Cell surface heparan sulfate proteoglycans (HSPGs) act as co-receptors for multiple families of growth factors that regulate animal cell proliferation, differentiation and patterning. Elimination of heparan sulfate during brain development is known to produce severe structural abnormalities. Here we investigate the developmental role played by one particular HSPG, glypican-1 (Gpc1), which is especially abundant on neuronal cell membranes, and is the major HSPG of the adult rodent brain. Results Mice with a null mutation in Gpc1 were generated and found to be viable and fertile. The major phenotype associated with Gpc1 loss is a highly significant reduction in brain size, with only subtle effects on brain patterning (confined to the anterior cerebellum). The brain size difference emerges very early during neurogenesis (between embryonic days 8.5 and 9.5), and remains roughly constant throughout development and adulthood. By examining markers of different signaling pathways, and the differentiation behaviors of cells in the early embryonic brain, we infer that Gpc1-/- phenotypes most likely result from a transient reduction in fibroblast growth factor (FGF) signaling. Through the analysis of compound mutants, we provide strong evidence that Fgf17 is the FGF family member through which Gpc1 controls brain size. Conclusion These data add to a growing literature that implicates the glypican family of HSPGs in organ size control. They also argue that, among heparan sulfate-dependent signaling molecules, FGFs are disproportionately sensitive to loss of HSPGs. Finally, because heterozygous Gpc1 mutant mice were found to have brain sizes half-way between homozygous and wild type, the data imply that endogenous HSPG levels quantitatively control growth factor signaling, a finding that is both novel and relevant to the general question of how the activities of co-receptors are exploited during development.


Background
Cell surface heparan sulfate proteoglycans (HSPGs) have been implicated as key regulators of patterning and growth in animal development [1,2]. Participation in these events is generally thought to reflect their functions as co-receptors for diverse growth factor families, including fibroblast growth factors (FGFs), Wnts, Hedgehogs and bone morphogenetic proteins (BMPs) [3][4][5][6]. In both vertebrates and invertebrates, disruption of heparan sulfate biosynthesis leads to severe, pervasive developmental abnormalities. For example, mice completely deficient in heparan sulfate arrest in gastrulation [7].
In contrast, the elimination of the individual core proteins that carry cell surface heparan sulfate generally produces more subtle or tissue-restricted defects, particularly in mammals [8][9][10][11][12][13][14][15][16]. Most likely this reflects the relatively large number of cell surface HSPGs (six glypicans and four syndecans in mammals), their overlapping patterns of expression, and a likelihood of functional redundancy that is made particularly high by the fact that carbohydrate moieties mediate much of their function. Despite such complexity, the analysis of core protein mutants has provided novel insights into at least some of the developmental and physiological processes in which HSPGs participate.
The glypicans define a structurally conserved family of glycosylphosphatidylinositol-anchored HSPGs that have been extensively studied for their roles in both development and cancer [17][18][19][20][21][22][23][24][25][26][27]. Of the six glypicans in mammals, glypicans 1 and 2 (Gpc1 and Gpc2) were identified early on as major HSPGs of the developing brain [28][29][30][31]. Subsequently, Gpc4 and Gpc5 were shown to be regionally expressed in the developing brain as well [32][33][34]. Biochemical studies suggest that the most abundant glypican in the rodent brain, at least from late gestation onward, is Gpc1 [29,31]. During development, Gpc1 is expressed in both neuroepithelial cells and mature neurons; it is particularly enriched in axons and nerve terminals [35,36]. We reasoned, therefore, that loss of Gpc1 might produce defects in neurogenesis and/or axonal guidance, both of which are driven by growth factors regulated, in many cases, by HSPGs. As described below, the brains of mice rendered null for Gpc1 were morphologically normal, except for subtle mispatterning of the anterior cerebellum, but were abnormally small. Unexpectedly, we found that the reduction in brain size in mutant mice reflects a specific role for Gpc1 at the earliest stages of neurogenesis, before embryonic day (E)9.5.

Generation of glypican-1 null mice
Homologous recombination was used in embryonic stem (ES) cells to flank the first exon of Gpc1 -which includes the translational start site and signal sequence -with loxP sites, and successfully targeted cells were transiently transfected with Cre recombinase to induce excision ( Figure  1A). Several independently excised ES clones were injected into C57BL/6 blastocysts to generate chimeric mice, one of which was extensively outcrossed to outbred (CD1; ≥10 generations) and inbred (C57Bl/6; ≥7 generations) mice ( Figure 1B). This mutant allele (formally des- ignated Gpc1 tm1 Alan ) will be referred to here as Gpc1 -. A second allele -a gene trap insertion of LacZ into the fourth intron of Gpc1 -will be referred to as Gpc1 LacZ .
Gpc1 -/and Gpc1 LacZ/LacZ mice were viable and fertile on outbred and inbred backgrounds, and appeared grossly normal. Biochemical studies revealed a complete absence of Gpc1 core protein from the brains of homozygous Gpc1 -/animals, with intermediate levels in heterozygotes ( Figure 1C,D). The expression of other brain heparan sulfate and chondroitin sulfate proteoglycans was not significantly affected ( Figure 1D). Immunohistochemistry also demonstrated a loss of Gpc1 staining in both Gpc1 -/and Gpc1 LacZ/LacZ mice (not shown). In situ hybridization studies, using both whole embryos and adult brain sections, also revealed a lack of Gpc1 transcripts in Gpc1 -/animals ( Figure 1E-I), suggesting that these animals are essentially protein-and message-null.

Loss of glypican-1 leads to reduced brain size and subtle patterning abnormalities of the cerebellum
Gpc1 -/mice were indistinguishable from wild type in gross appearance, size, lifespan, and a variety of simple behaviors (data not shown). Their internal anatomy appeared normal, with the exception of the brain, which was noticeably small (Figure 2A). On the CD1 background, adult Gpc1 -/brains weighed 15.5 ± 1.0% less than wild type, and Gpc1 -/+ brains weighed 7.8 ± 1.0% less (Figure 2B). Both results were highly statistically significant, as variation in brain size among individuals within mouse strains is normally very small [37].
On a C57/Bl6 background, reductions in Gpc1 -/and Gpc1 -/+ brain weight were also evident, but less pronounced (11% for Gpc1 -/-; data not shown), possibly reflecting the fact that wild-type C57/Bl6 brains are approximately 5% smaller than wild-type CD1 brains to begin with. As shown in Figure 2C, Gpc1 LacZ/LacZ mice also displayed reduced brain size (the higher variance of these data likely reflects the mixed CD1/C57 background of these animals). The brain weight reduction did not correlate with sex or body weight in Gpc1or Gpc1 LacZ mutant animals ( Figure 2D).
To determine whether changes in brain size were due to the presence of fewer cells or smaller cells, DNA was extracted from whole brains and quantified using Hoechst 33258 fluorescence [38]. As shown in Figure 2E, Gpc1 -/mice had about 20% less DNA per brain than their wildtype littermates. Thus, loss of Gpc1 leads to a 20% decrease in the number of brain cells.
Despite this, the shapes of Gpc1 -/forebrains were morphometrically normal (Additional file 1), and most structures showed no apparent patterning defects. The major exception was found in the gyri of the cerebellar vermis, where the most anterior lobe (lobe I) either did not form or was rudimentary ( Figure 3A-F). In addition, the primary fissure was shortened by 12%, with no significant change in the lengths of the other fissures, or in the overall cerebellar perimeter (data not shown). Notably, structures that are disrupted in heparan sulfate-deficient conditional Ext1 mutant mice, such as the interhemispheric commissures and the forebrain septum [39], were present in Gpc1 mutants ( Figure 3G-J). In addition, no obvious changes in lamination of the cerebellar or cerebral cortices were observed (Figure 3K-N; and data not shown), and staining for markers of neuronal subpopulations, such as calbindin and parvalbumin, appeared normal (data not shown).
Targeted mutation of the Gpc1 locus creates a null allele Figure 1 Targeted mutation of the Gpc1 locus creates a null allele. (A) Targeting strategy. A construct was created in which loxP sites flank exon 1 of Gpc1 and FLP recognition target (FRT) sites flank negative (TK) and positive (neo) selection markers. After transfection into embryonic stem (ES) cells, Southern blotting, using a probe located outside the targeted region, permitted verification of targeted integration. Transient tranfection of targeted ES cells with a Cre expression plasmid was then used to remove Gpc1 exon 1 and selection markers, prior to generation of mice. (B) Identification of the Gpc1 wild-type (+) and mutant (-) allele by PCR in mice. (C) Immunoblotting of adult brain membrane fractions with an anti-Gpc1 antibody. The presence of a discrete band found only in samples pretreated with heparitinase (Hase) demonstrates that the immunoreactive molecule is a heparan sulfate proteoglycan (HSPG) core protein. Note the decreased band intensity in heterozygous animals, and complete absence of immunoreactivity in homozygotes. The small apparent difference in mobility and slight tilt of the wild-type band is an artifact of uneven electrophoresis. (D) Immunoblotting of adult brain membrane HSPG core proteins using 3G10 antibody. Notice that the loss of a band at the correct molecular weight for Gpc1 in mutant animals is not accompanied by a substantial, consistent change in the presence of other HSPG cores. Hase, Heparitinase; Case, Condroitinase ABC. (E-G) Whole-mount in situ hybridization for Gpc1 at E8.5. Genotypes are as indicated. Red arrowheads point to areas of high Gpc1 expression in the developing brain and branchial arches. Notice that Gpc1 expression is greatly reduced or absent in homozygous mutants. Levels in heterozygotes are intermediate.

Brain size reduction reflects an early embryonic requirement for Gpc1
Although Gpc1 has been reported to be the major HSPG of the adult brain [35], it is also expressed throughout development of the nervous system, and can be detected in neural tissue as early as head-fold stage (around E7) [35,40]. We therefore measured brain size in Gpc1 -/mice at various developmental stages. At birth, Gpc1 -/brains weighed 15 to 20% less than wild-type brains, that is, they were smaller to about the same degree as seen in adulthood (data not shown). Prior to birth, weights could not be measured accurately given the small size of embryonic brains and the need to dissect under liquid; instead, volumes were estimated from photographic images of fresh embryos (Additional file 2). As shown in Figure 4A-B the estimated volume of Gpc1 -/brains was 22% below that of wild-type littermates as early as E9.5 ( Figure 4B). In contrast, brain volumes measured one day earlier (E8.5) exhibited no difference between mutant and wild-type littermate embryos. Apparently, a phenotype of about the same magnitude as seen in adult animals emerges over the course of a single day of development.
To test whether the smaller brain size of E9.5 mutant embryos is a consequence of general developmental delay, we plotted brain size against somite number, a marker of developmental stage ( Figure 4C). The data showed a slight trend -which was not statistically significant -toward Gpc1 -/embryos being delayed by about 1 somite (at E8.5, wild type = 11.5 ± 2.8 and mutant = 10.3 ± 3.2; at E9.5, wild type = 22.0 ± 2.4 and mutant = 21.2 ± 2.4). Even taking such a trend into account, regression analysis showed that >70% of the brain size difference at E9.5 is independent of somite number (note the down-ward shift of the regression line at E9.5). These data suggest that, between E8.5 and E9.5, Gpc1 plays a specific and important role in nervous system growth.
To test whether that role involves regulation of cell proliferation, we used phospho-histone H3 immunohistochemistry to quantify the density of cells in M-phase ( Figure 4D-H). If one considers the early central nervous system as a mass of cycling cells undergoing exponential expansion, with a typical cell cycle time of about 8 hours [41,42], then in order to produce a 20% reduction in cell mass over the course of one day of development (and no further decrease thereafter), one would need to lengthen the cell cycle during that day by about 7%. Thus, in Gpc1 -/embryos, we initially expected to see up to an approximately 7% decrease in mitotic labeling index at E8.5, followed by a return to normal on the following day.
Interestingly, we observed a larger and longer-lasting change in labeling index. For example, at E9.5, when we expected proliferation to have returned to wild-type levels, the labeling index in mutant brain neuroepithelium was decreased by 30% ( Figure 4G; P < 0.001) The effect was specific to the neuroepithelial cells of these embryos (Additional file 3), and did eventually disappear (by E11.5; Figure 4H). We considered the possibility that the larger-than expected drop in proliferation in Gpc1 -/embryos was being offset by decreased cell death, but this was not supported by direct measurements. TUNEL staining showed that dying cells are not particularly abundant in the wild-type central nervous system at these stages, and if anything, displayed a trend (not statistically significant) toward being more abundant in the mutant, not less. A likely explanation for the delayed and larger-than- The Gpc1 mutant brain phenotype emerges early in neural development Figure 4 (see previous page) The Gpc1 mutant brain phenotype emerges early in neural development. (A) Lateral images of embryonic day (E)8.5, E9.5 and E11.5 wild-type (+/+) and Gpc1 -/-(-/-) embryos. Because the developmental stage of embryos is highly variable at E8.5 to E9.5, not every image is fully representative of the average for that genotype (for quantitative data, see (B, C) below). (B) Gpc1 -/embryos display smaller brains starting at E9.5. Brain size was estimated from photographic images, as described in Additional file 2. Averages of estimated cranial area (at E8.5 and E9.5) and cranial volume (E11.5) were obtained for wild-type and homozygous mutant embryos in multiple litters (N = 23 for +/+ and N = 15 for -/-at E8.5; N = 25 for +/+ and N = 22 for -/-at E9.5; N = 10 for +/+ and N = 18 for -/-at E11.5). Units are pixels 2 (× 10 4 ) and pixels 3 (× 10 6 ) for area and volume measurements, respectively. Note the 14.1% decrease in area (* and **P < 0.05) that emerges at E9.5 (consistent with an approximately 20.4% decrease in volume). (C) Comparison of cranial area with somite number in E8.5 and E9.5 wild-type (+/+) and expected drop in neural labeling index became apparent later on, after more information about the probable mechanism of GPC1 action was obtained (see below).

Redundancy versus compensation within the glypican family
Six structurally similar glypicans are encoded by the mammalian genome. A possible explanation for the relatively mild phenotype in Gpc1 -/mutants would be the participation of other glypicans in the regulation of early embryonic neural proliferation. In fact, Gpc2, Gpc3, Gpc4, Gpc5, and Gpc6 can all be detected in E8.5 and E9.5 embryos, although Gpc3 expression is excluded from most of the nervous system, and Gpc5 is not strongly expressed in brain until later stages ( Figure 5A-E) [30,36,40,43]. Interestingly, the expression levels of Glypicans 2 to 6, as judged by whole mount in situ hybridization, were not noticeably altered in Gpc1 -/mutants ( Figure 5A-J), suggesting that there is little or no compensatory up-regulation of other glypicans in response to loss of Gpc1. Levels of Gpc4 RNA were also measured by quantitative RT-PCR and were unchanged in Gpc1 -/mice (data not shown).

+/+ -/-
To test for redundancy between Gpc1 and the other glypicans that are expressed in early embryonic brain, we generated double mutants with Gpc2 and Gpc4 (mutants in Gpc6, the third glypican that is strongly expressed in early embryonic brain, have not yet been produced). For Gpc2 we used a targeted allele that, when homozygous, produces phenotypically normal mice with a complete loss of Gpc2 protein (S Saunders and ADL, unpublished data). As shown in Figure 5K, the brains of Gpc2 -/mice are not reduced in size, and the brains of Gpc1 -/-;Gpc2 -/double mutants are no smaller than those of Gpc1 -/mice.
In contrast, when compound mutants were made between Gpc1 and a gene-trap allele of Gpc4 (Gpc4 LacZ ), brain size appeared synergistically reduced ( Figure 5L). This experiment may underestimate the contribution of Gpc4, since the Gpc4 LacZ allele is very likely not null (unpublished observations). Thus, within the limits of what can be assessed using existing mutants, Gpc1 appears to act redundantly with at least Gpc4 in controlling brain size.

Evidence for impairment of FGF signaling in the early Gpc1 mutant embryo
Numerous growth factor and morphogen signaling pathways have been implicated in the control of brain growth and development [44][45][46][47]. Many of these pathwaysincluding those mediated by Hedgehogs, Wnts, BMPs, and FGFs -are influenced by HSPGs in at least some developmental contexts [1,5,48,49]. To screen for disruption of these pathways in Gpc1 -/mice, we performed in situ hybridization at E8.5 and E9.5 for known downstream markers or reporter genes.
Although we saw no significant differences between wildtype and mutant embryos at these stages in the expression patterns of marker genes for Sonic hedgehog (Shh), BMP and Wnt signaling (Additional file 4), the data suggested that FGF signaling was potentially attenuated. For one marker of FGF signaling -the mitogen-activated protein (MAP) kinase phosphatase Pyst1 (also known as Mkp3 and Dusp6) -we observed no significant change in the Gpc1 -/mouse ( Figure 6A-D), but for a second marker -Sprouty 2 (Spry2) -we noticed a consistent, yet transient, reduction in expression. Specifically, Spry2 in situ hybridization was consistently weaker in Gpc1 -/brains at E8.5 to E9.5 ( Figure 6E-H), but returned to a near-normal level by E10.25 (Additional file 5). To verify this result, we turned to quantitative RT-PCR. As shown in Figure 6I, there was a 40% reduction in Spry2 mRNA in Gpc1 -/brains at E9.5, with some reduction in Spry1 mRNA (also a FGF target gene) as well. In contrast, and in agreement with the data from in situ hybridization, Pyst1 levels were unchanged.
To examine FGF signaling more directly, we produced short-term E9.5 dorsal forebrain explant cultures from wild-type and Gpc1 -/embryos, and measured the enzymatic activity of extracellular regulated kinase (ERK) MAP kinase. As shown in Figure 6J, wild-type explants exhibited a high basal level of ERK activity, suggesting a high degree of endogenous growth factor signaling within the explant. In contrast, Gpc1 -/explants exhibited 41% less ERK activity under the same conditions. In response to exogenous FGF2, wild-type explants displayed only a small additional increase in ERK activity, possibly because endogenous signaling was so high. We consistently observed that the exogenous FGF response was even lower in Gpc1 -/explants, but due to the small size of the FGF effect we were unable to establish statistical significance Loss of Gpc1 results in weakened fibroblast growth factor (FGF) signaling and premature neuronal differentiation in brain for this conclusion (data not shown). Overall, the data strongly suggest that Gpc1 deficiency diminishes the response of cells of the early nervous system to FGFs.

Loss of glypican-1 results in premature differentiation of postmitotic neurons
It is widely reported that FGFs promote the proliferation of neural progenitor cells [50][51][52][53][54][55][56][57][58][59]. In most cases, FGFs appear not to act by affecting cell cycle kinetics, but by increasing the probability that the progeny of dividing cells remain in the cell cycle instead of differentiating (that is, FGFs suppress cell cycle exit). Such a mechanism of FGF action has been established, for example, in the cerebral cortex, the midbrain, the olfactory epithelium and the telencephalic subventricular zone [50,52,53,55,58]. Indeed, even in non-neural tissues such as muscle, preventing cell cycle exit seems to be the central mode of action of FGF [60,61].
If impairment of FGF signaling is the primary mechanism by which Gpc1 deficiency reduces brain size, we reasoned that Gpc1-deficient embryos should exhibit accelerated cell cycle exit and, as a consequence, premature neuronal differentiation. Indeed, this turned out to be the case. As shown in Figure 6K-N, higher than normal numbers of TuJ1+ neurons were detected in the brains of early Gpc1 -/embryos. This effect was especially pronounced at E9.5, when there were over twice as many differentiated neurons in the fore-and midbrains of Gpc1 mutants as in wild-type animals (P < 0.005). To confirm that surplus TuJ1+ neurons were the progeny of cells that had recently undergone cell division, we pulsed embryos at E8.5 with the S-phase label 5-bromo-2'-deoxyuridine (BrdU), and one day later (E9.5) counted the number of TuJ1+/BrdU+ cells in the developing brain. In wild-type embryos we observed an average of 50 ± 3 such cells per mm 2 , whereas in Gpc1 -/embryos we observed three times as many (150 ± 49; P < 0.05).
Not only do these results support the conclusion that FGF signaling is the main target of Gpc1 in the early central nervous system, they also explain the unexpectedly large, prolonged decreases in mitotic labeling index in Gpc1 mutant embryos ( Figure 4G). This is because the labeling index represents the ratio of cells in M phase to total cells. When cells leave the cell cycle, they no longer affect the numerator of this ratio, but still contribute to its denominator, thereby causing an alteration in the labeling index that outlasts, for a period of time, any actual disturbance in the proliferative or differentiative behavior of cells (for a quantitative treatment of this point, see the Supplemental Appendix in Additional file 6). In short, the labeling index data in Figure 4G,H correspond well with the expected consequences of a temporary diminution in activity of a factor, such as FGF, that suppresses the differentiation of neuronal progenitors.

Glypican-1 acts through Fgf17
Several FGF family members have been implicated in brain development, with the subfamily formed by Fgf8, Fgf17 and Fgf18 having been shown to be especially important at early embryonic stages [44,45,47,[62][63][64]. Fgf8, the most intensively studied of the group, plays critical roles in brain patterning and morphogenesis, and it has been suggested that some of the phenotypes in the Nes-Ext1 null brain are the result of impaired Fgf8 function [39]. However, there is substantial overlap in expression of all three of these Fgfs, which also share similar receptor binding properties, and are all thought to control neural proliferation [65][66][67].
In initial studies of compound mutants involving Gpc1and Fgf8 neo (a hypomorphic allele of Fgf8), we failed to observe significant genetic interaction (that is, a mutant allele of Fgf8 did not further reduce brain weight in a Gpc1 mutant; data not shown). In contrast, we were struck by similarities between the reported defects in anterior cerebellar patterning in Fgf17 -/mice [66] and those reported here for Gpc1 mutants ( Figure 3C-F). We therefore generated and analyzed a set of adult Gpc1 -;Fgf17compound mutants (Figure 7).
Representative midline cerebellar morphologies of animals of various compound genotypes are shown in Figure  7A-E. Figure 7F summarizes the distributions of brain weights among the entire collection, by genotype. The data clearly show that, as with Gpc1, loss of wild-type Fgf17 alleles leads to a progressive reduction in brain size and anterior cerebellar defects. Moreover, in animals that possessed either one or two functional Fgf17 alleles, the additional loss of one or two Gpc1 alleles led to a further reduction in brain size. However, in animals null for Fgf17, loss of Gpc1 had no significant phenotypic effect, either on brain size or cerebellar morphology. Complete dependence of the Gpc1 phenotype on the presence of a functional Fgf17 gene provides strong evidence that Fgf17 is, if not the only FGF through which Gpc1 acts, certainly among the most important, at least with respect to the control of brain size and cerebellar patterning.

Discussion
The data presented here establish that Gpc1 plays a role in controlling brain size by regulating the behavior of progenitor cells during early brain development. Although the quantitative effect of Gpc1 loss is modest -a 20% decrease in cell number -it is highly significant when compared to normal variation in brain size within genetically homogeneous mice (Figure 2). The effect appears to be due to a shift in the balance of progenitor cell prolifer-ation versus differentiation over the course of approximately one day of development -from E8.5 to E9.5 (Figures 4 and 6). Diminished signaling by FGFs, but not other growth factors, could be demonstrated at around this time period in Gpc1 -/mice ( Figure 6). Genetic epistasis experiments strongly suggested that reduced Fgf17 signaling accounts for most or all of the Gpc1 mutant phenotypes observed here.
The transient and modest effects of Gpc1 loss raised the possibility of either compensation by, or redundancy with, other HSPGs, and although no evidence for compensatory up-regulation of other core proteins was obtained, studies with Gpc1 -;Gpc4 LacZ double mutants suggested that Gpc1 and Gpc4 may have at least partially overlapping functions in controlling brain size ( Figure 5). This idea is lent further support by recent studies in Xenopus, in which morpholino-mediated knockdown of Gpc4 led to a reduction in size of dorsal forebrain structures [68].
It is noteworthy that Gpc1 -/mice failed to display any of the severe developmental phenotypes reported for the Nestin1-Cre mediated conditional inactivation of Ext1, which produces a mouse in which all heparan sulfate is eliminated from the brain from about E10 onward. The Gpc1 phenotypes require Fgf17 Figure 7 Gpc1 phenotypes require Fgf17. Compound Gpc1/Fgf17 genotypes were produced by interbreeding Gpc1 +/-;Fgf17 +/mice. (A-E) Nissl-stained mid-sagittal sections through the cerebella of adult compound mutants. Red arrowheads mark the anterior-most lobe (lobe I), which disappears in both Gpc1 -/and Fgf17 -/animals. Red arrows mark the fusion of lobes III and IV, a phenotype observed in Fgf17 -/mice. Note that the Gpc1 -/-;Fgf17 -/phenotype is no more severe than the single Fgf17 -/phenotype. (F) Fresh brain weights of adult compound mutants. Genotypes for each category are as indicated; 'N' refers to the number of animals obtained of each genotype. The data show that, on their own, the presence of either one or two copies of mutant alleles for either Gpc1 or Fgf17 progressively reduces brain size. When animals are null for Fgf17, the presence of mutant Gpc1 alleles has no significant effect (*P < 0.005; **P < 0.001; by Student's t-test).
phenotypes of the Nes1-Ext1 null mouse include specific hypoplasia of the cerebral hemispheres, absence of the cerebellum and olfactory bulbs, and loss of certain midline commissural tracts. It is possible that the mild deficits in the anterior cerebellum that we observed in the Gpc1 -/mouse are related to the cerebellar agenesis seen in the Nes1-Ext1 null mouse. Interestingly, even when double mutant mice were made between Gpc1and Gpc4 LacZ (Figure 5), Gpc2 -( Figure 5) or Gpc5 -(YHJ, ADL and S Saunders, unpublished observations), Nes1-Ext1 null phenotypes were not observed. It may be that the combined loss of function of multiple glypicans, and/or glypicans as well as syndecans (the other major family of HSPGs), is required to substantially eliminate heparan sulfate function in brain development.
A fascinating aspect of the Gpc1 mutant phenotype is that the brain weight of heterozygous animals falls halfway between that of wild type and homozygous mutants (Figure 2). This implies that the amount of Gpc1 expressed by cells influences, in a continuous fashion, the level of growth factor signaling that cells perceive. In other words, Gpc1 may not merely be necessary for growth factor signaling, but a quantitative regulator of the 'gain' of signaling. Accordingly, regulation of Gpc1 expression may be an important part of the control circuitry that keeps brain size so tightly regulated in mammals [37,84].
Recent work suggests a similar quantitative role -albeit in the opposite direction -for Gpc3 [85]. Mouse Gpc3 mutants are known to display marked pre-and postnatal overgrowth of many organs, phenocopying Simpson Golabi Behmel syndrome (a syndrome caused by mutations in human GPC3 [8,9,86]). Because Gpc3 is on the X chromosome, gene dosage effects are not observable with null alleles, but studies of naturally occurring polymorphisms in the Gpc3 regulatory region have recently implied that quantitative variation in the level of Gpc3 expression directly (negatively) controls body size [85]. Even though the molecular mechanism of somatic growth inhibition by Gpc3 is likely to differ from the mechanism of brain growth promotion by Gpc1, the implication of both glypicans in quantitative regulation of size is striking. Intriguingly, a variety of studies in Drosophila also link invertebrate glypicans to organ size control [87][88][89].
The present study leaves unresolved the role that glypicans play in axons and nerve terminals, where Gpc1 and Gpc2 are especially abundant [36]. The absence of obvious defects in axonal pathways in the Gpc1 -/mouse suggests that these molecules may play more of a role in synaptic physiology than in axonal growth or guidance. Certainly, evidence for the participation of syndecan-2 and syndecan-3 in synapse formation [14,[90][91][92], as well as recent work on the Drosophila nervous system [89,[93][94][95], suggests that HSPGs may play a variety of as yet unappreciated roles in basic neurophysiology. To this end, it is intriguing that recent genome-wide association studies in man have identified both GPC1 and FGFR2 (which encodes a major FGF receptor of the brain) as members of a small handful of genetic loci that correlate with risk of schizophrenia [96,97], a psychiatric disorder also associated with a small, but significant, reduction in brain volume [98,99]. Clearly, a detailed behavioral and neurophysiological examination of the Gpc1 mutant mouse seems warranted in the future.

Conclusion
Cell-surface HSPGs are critical for growth and patterning in numerous tissues and organ systems, presumably as a consequence of their actions as growth factor co-receptors. Here we show that Gpc1 controls the size of the mammalian brain through an unexpectedly specific mechanism: regulation of the proliferation/differentiation behavior of progenitor cells during a very early stage of neurogenesis. We provide evidence that this action is mediated through regulation of Fgf17 signaling, and further show that Gpc1's effects are gene-dosage dependent. The data support the view that glypicans, and possibly HSPGs in general, serve as quantitative regulators of the gain of growth factor signaling during neural development.
For production of staged embryos, timed matings were used and noon of the day of vaginal plug was considered as E0.5. At early embryonic stages, more precise staging was obtained from somite number. To obtain BrdUlabeled embryos, pregnant mice were injected intraperitoneally with 50 μg of BrdU (B5002; Sigma-Aldrich St. Louis, MO, USA)per gram body weight and embryos were collected 24 hours later.
Mouse colonies were maintained, and all animal experimentation conducted, in accordance with the policies and guidelines of the Institutional Animal Care and Use Committee (IACUC) of the University of California, Irvine. (IACUC protocol number 1998-1656).

Measurement of brain size and DNA content
Postnatal and adult brains were freshly dissected. After removal of olfactory bulbs and remaining spinal cord (at the level of the posterior margin of the cerebellum), brains were immediately weighed on a laboratory scale.
Images of fresh embryos were collected using a Leica MZFLIII stereomicroscope and a SPOT camera (Diagnostic Instruments, Inc. Sterling Heights, MI, USA). For embryos at E11.5 or older, brain height, depth and width were separately measured from lateral and frontal images (Additional file 2), and multiplied to produce a volume estimate. For E8.5 and E9.5 embryos, measurements of area were obtained from perimeter tracings of lateral views using Image J analysis software [109]. At these stages the central nervous system comprises the majority of head tissue, so such tracings included the entire head, stopping ventrally at the rostral border of the first branchial arch, and dorsally at the top of mesencephalon. Volume was then estimated as area 3/2 . In some cases, volume was also estimated by the procedures outlined above for older embryos, and qualitatively similar results were obtained.
DNA content in brain homogenates was measured by enhancement of bisbenzimid fluorescence at 458 nm, as described by Labarca and Paigen [38]. A linear standard curve (1 to 10 μg/ml) was obtained using salmon sperm DNA (Invitrogen).

Quantitative RT-PCR
Forebrain vesicles of E9.5 and E8.5 wild-type and mutant mice were dissected in ice-cold PBS, and RNA was isolated and column purified (Aurum Total RNA Mini Kit, Bio-Rad, Herculeus, CA, USA) according to the manufacturer's instructions. cDNA was generated by reverse transcription with a mixture of oligo dT and random hexamers (Superscript First-Strand Synthesis kit, Invitrogen). PCR quality controls, experimental runs and statistical methods were performed as described [110,111]. Quantification of total mRNA expression was performed with an Opticon System (MJ Systems CFD-3200, Calgary, Denver, USA) and SYBR-Green (Bio-Rad).
All measurements were normalized to values for 18S RNA in the same samples. All cDNA samples were validated for reverse transcription reaction efficiency and minimal genomic DNA contamination (cDNA/genomic target ratio >10 5 ) for 40 cycles in duplicates. Average of duplicated cycle threshold (Ct) values were normalized as ΔCt (Ct gene of interest -Ct reference(18S) ). Relative levels were converted using the 2 -ΔΔCt method: ΔΔCt = ΔCt mutant -ΔCt wildtype [112] Averages of duplicate Ct, normalized ΔCt, ΔΔCt and relative level 2 -ΔΔCt and standard errors were calculated using Microsoft Excel.

Subcellular fractionation and analysis of proteoglycan content
Adult brains were dissected in ice-cold PBS and immediately homogenized. to obtain membrane and soluble fractions as described [114]. For SDS-PAGE analysis, samples prepared in this way were digested for 30 minutes at 37°C with Heparinase III or with Heparinase III plus Chondroitinase ABC (all used at 1.5 U/mg of protein; both enzymes were purchased from Seikagaku Corp., Tokyo, Japan) along with a proteinase inhibitor mixture (10 μg/ml pepstatin A, 20 μg/ml leupeptin, 2.5 mg/ml NEM, and PMSF in 50 mM Tris-hydroxyaminomethane, 15 mM phosphoric acid, pH7.3). Digested samples were boiled for 10 minutes in SDS-PAGE sample buffer and loaded at 50 μg protein per lane onto 7.5% SDS-polyacrylamide gels, and subjected to electrophoresis. Gels were transferred to PVDF membrane (Millipore, Billerica, MA, USA) and probed with rabbit anti-glypican-1 (1:3,000) antibody or mouse 3G10 monoclonal antibody (1:2,000; USBiological, Swampscott, MA, USA). Samples without enzyme treatment, or subjected to single enzyme treatment, were used where indicated. Blots were incubated with horseradish peroxidase-conjugated goat anti-rabbit or donkey anti-mouse antibody, as appropriate, and visualized using enhanced chemiluminescence.