Down syndrome (DS) is caused by trisomy 21 (Lejeune et al., 1959). It affects many systems, resulting in craniofacial abnormalities, impaired cognitive function, Alzheimer's-like histopathology in the brain, hypotonia, and increased risk of cardiac defects, Hirschsprung's disease, and childhood leukemia (Cohen, 1999). The complex developmental origins of many of these phenotypes constitute a barrier to understanding their pathogenesis. Most DS phenotypes are incompletely penetrant and variable in expressivity, and the mechanism(s) by which increased gene dosage causes any specific DS feature is not established.
In rare cases, DS occurs in individuals with segmental trisomy for only a portion of human chromosome 21 (HSA21). “Phenotype maps” (Delabar et al., 1993; Korenberg et al., 1994) have been constructed by determining the smallest region of overlap (SRO) among individuals with segmental trisomy 21 who display the same DS feature. The SRO for given feature is referred to as a “Down syndrome critical region” (DSCR). Many features of DS have been associated with a DSCR of approximately 5 Mb from D21S17 to MX1 (approximately 15% of HSA21) in band q22.3 (Hattori et al., 2000), which contains 38 genes.
A qualitative, or gene dosage, hypothesis of gene action in DS predicts that a DSCR contains a dosage-sensitive gene or genes, dosage imbalance for which is sufficient to cause specific phenotypes (Epstein, 1990; Pritchard and Kola, 1999). Alternatively, a quantitative or amplified developmental instability hypothesis asserts that DS phenotypes result from small nonspecific effects of hundreds of genes at dosage imbalance, with the expectation that a small region at dosage imbalance would not be sufficient to produce most DS features and any larger segment should have the same consequences regardless of what genes are triplicated (Shapiro, 1989, 1997). These concepts are not mutually exclusive. A sensitization model has been proposed in which critical dosage sensitive genes have little or no effect when triplicated by themselves, but exert a visible effect on phenotype when overexpressed in combination with small effects of many other genes at dosage imbalance (Reeves et al., 2001).
These hypotheses cannot be tested in humans, but mouse models with dosage imbalance for defined regions can be evaluated for phenotypes analogous to those in DS. The most informative phenotypes are those that occur in every individual with trisomy 21. Anomalies of the craniofacial skeleton and a disproportionate reduction in the volume of the cerebellum have been quantified in humans. We have established quantitative measures of these phenotypes in the Ts65Dn mouse, a genetic model for DS with segmental trisomy. We observed exact parallels with corresponding structures affected in DS (Reeves et al., 1995; Baxter et al., 2000; Richtsmeier et al., 2000). In fact, demonstration of reduced granule cell density in the Ts65Dn cerebellum correctly predicted a corresponding phenotype in DS (Baxter et al., 2000). If the qualitative hypothesis of DS is correct, then any mouse model with trisomy for a region containing the critical gene(s) should display the same phenotype.
The Ts1Cje mouse (Sago et al., 1998) and the Ms1Cje/Ts65Dn mouse (Sago et al., 2000) are trisomic for complementary subregions of the mouse chromosome 16 (MMU16) segment that is triplicated in Ts65Dn. In this study, we compared the cerebellar phenotypes of Ts65Dn with those of Ts1Cje and of the few viable Ms1Cje/Ts65Dn mice that could be obtained. All of these models have three copies of regions of MMU16 that are conserved in both gene content and order with HSA21 (Fig. 1). Overall, we find that the cerebellum is less affected in Ts1Cje, and even milder changes are seen in Ms1Cje/Ts65Dn mice. The specific patterns of phenotypic anomalies would not have been predicted by either the qualitative or the gene dosage effects hypotheses for gene action in DS.
Extent of Segmental Trisomy in Three Mouse Models of Down Syndrome
Analysis of MMU16 sequence that is conserved with HSA21 (Hattori et al., 2000; Pletcher et al., 2001; Mural et al., 2002) demonstrates the relative genetic content of the Ts65Dn, Ts1Cje, and Ms1Cje/Ts65Dn mouse models (Fig. 1). The Ts65Dn mouse is trisomic for the region between Mrpl39 and Znf295 (Fig. 1B; Gardiner et al., 2003; Kahlem et al., in press). This region spans 16.4 Mb in the human and 13.6 Mb in the mouse. Most of the difference in size between human and mouse appears to be in the content of repetitive sequences. A recent comparison found orthologs for 166 HSA21 genes in the MMU16 segment at dosage imbalance in Ts65Dn (Gardiner et al., 2003). The Ts1Cje and Ms1Cje/Ts65Dn mouse models subdivide this region into two parts. The Ts1Cje mouse is trisomic for the region between Sod1 and Znf295 (but not including Sod1), which contains 112 genes, or 67% of those at dosage imbalance in Ts65Dn. Both Ts1Cje and Ts65Dn include the variously defined DSCR (Delabar et al., 1993). The Ms1Cje/Ts65Dn mouse is trisomic for the region between Mrpl39 and Sod1 (but not including Sod1), including orthologs of 53 HSA21 genes.
Cerebellar Volume Is Reduced in Ts1Cje Mice
We measured Ts1Cje cerebella to make comparisons with the previously characterized Ts65Dn mouse (Baxter et al., 2000). High-resolution three-dimensional magnetic resonance imaging (MRI) images were obtained from six Ts1Cje and six euploid littermate brains (pixel dimensions 62.5 μm × 62.5 μm × 125 μm, images in the sagittal plane), in which the cerebellum could easily be distinguished from the rest of the brain (Fig. 2), and were evaluated by using the MRIcro Program. Relative cerebellar volume was calculated by summing the volumetric pixels measured in serial MRI images and dividing by total brain volume to normalize for variation in brain size (Table 1). The Ts1Cje cerebellum was reduced in volume to 88.8% of the euploid value (P = 0.007). The same results were obtained by comparing cross-sectional area of the cerebellum at the midline, which was shown to be correlated directly to cerebellar volume in Ts65Dn mice (Baxter et al., 2000). When Ts1Cje cerebellar area was measured at the midline in histological sections (Fig. 3) and normalized to total brain area, it was reduced to 86.3% of the euploid value (P = 0.001).
Table 1. Comparison of Cerebellar Phenotypes in Ts65Dn, Ts1Cje, and Ms1Cje/Ts65Dn Mice
Data for the Ts65Dn mouse are from Baxter et al. (2000).
Normalized cerebellar volume
88.1% of euploid
n = 10 Ts65Dn
n = 6 Ts1Cje
n = 3 Ms1Cje/Ts65Dn
n = 10 euploid
n = 6 euploid
n = 4 euploid
P = 0.0003
P = 0.007
P = 0.34
Granule cell density
76.0% of euploid
n = 8 Ts65Dn
n = 6 Ts1Cje
n = 3 Ms1Cje/Ts65Dn
n = 8 euploid
n = 6 euploid
n = 4 euploid
P = 0.0001
P = 0.09
P = 0.05
Purkinje cell density
89.5% of euploid
n = 6 Ts65Dn
n = 5 Ts1Cje
n = 3 Ms1Cje/Ts65Dn
n = 6 euploid
n = 5 euploid
n = 4 euploid
P = 0.03
P = 0.27
P = 0.46
Granule Cell Density Is Moderately Reduced in Ts1Cje
Ts65Dn mice show a significant reduction in cerebellar granule cell density in sagittal sections, to 76% of euploid, at the midline and in lateral sections. This phenotype correctly predicted a comparable change in DS brains (Baxter et al., 2000). Granule cell density in Ts1Cje mice was averaged from twelve 5,000 μm2 fields per cerebellum (Fig. 4). The granule cell density in the Ts1Cje cerebellum is 91.2% of that in euploid (Table 1), and this trend approaches statistical significance (P = 0.09).
Purkinje Cell Density Is Not Reduced in Ts1Cje Mice
Purkinje cells are found in a single-cell layer between the granule cell and molecular layers (Fig. 3). Purkinje cell density was determined by calculating the number of cells per unit length in a midline cerebellar section. No difference in Purkinje cell density was seen between the Ts1Cje and the euploid cerebellum (P = 0.27).
Ms1Cje/Ts65Dn Cerebellar Phenotypes
The balanced T(16;12)1Cje translocation, which gave rise to Ts1Cje mice, arose from a reciprocal translocation between mouse chromosomes 12 and 16 (Sago et al., 1998). Crossing mice with this balanced translocation to Ts65Dn can produce Ms1Cje/Ts65Dn mice, which contain two copies of genes that are over-represented in Ts1Cje and are trisomic for the segment of MMU16 at dosage imbalance in Ts65Dn but not Ts1Cje mice (see Fig. 1A). However, neither of the translocation chromosomes segregates at Mendelian ratios, and in two years of breeding, we recovered only three Ms1Cje/Ts65Dn mice. Despite this small number, these animals were typed for the same characters tested in Ts1Cje mice.
Nonparametric tests that are less sensitive to sample size were used in addition to parametric statistics in the assessment of Ms1Cje/Ts65Dn mice. When compared with euploid, the Ms1Cje/Ts65Dn cerebellum was not reduced (97.3% of euploid, P = 0.34) and Purkinje cell density was the same (P = 0.21). Granule cell density in the Ms1Cje/Ts65Dn cerebellum was reduced to 90.8% of the euploid value, similar to the reduction in Ts1Cje mice, and this reduction was significant (P = 0.05). Given the very small number of Ms1Cje/Ts65Dn mice, however, these results must be viewed as being only suggestive.
Ts65Dn, Ts1Cje, and Ms1Cje/Ts65Dn Show Different Patterns of Cerebellar Phenotypes
Cerebellar volume, granule cell density, and Purkinje cell density are all significantly reduced in Ts65Dn (Baxter et al., 2000), but different phenotypic profiles are observed in Ts1Cje and Ms1Cje/Ts65Dn (Table 2). Ts1Cje cerebellar volume is reduced to the same extent as in Ts65Dn mice, despite that 1/3 fewer genes are at dosage imbalance. Granule cell density is mildly affected in Ts1Cje mice. Purkinje cell linear density is reduced significantly in Ts65Dn but not affected in either Ts1Cje or Ms1Cje/Ts65Dn mice. It is of interest that Ms1Cje/Ts65Dn showed the same granule cell density reduction as Ts1Cje, even though these mice had entirely different sets of over-represented genes, and Ts1Cje has twice as many genes at dosage imbalance as Ms1Cje/Ts65Dn mice. This reduction was statistically less than that seen in Ts65Dn mice. The small segmental trisomy in Ms1Cje/Ts65Dn had no apparent effect on cerebellar volume.
Table 2. Comparison of Cerebellar Phenotypes Between the Ts65Dn, Ts1Cje mouse, and Ms1Cje/Ts65Dn Mouse Strains
Different from Ts65Dn
Different from Ts1Cje
Different from Ms1Cje/Ts65Dn
Analysis of variance with post hoc least significant difference comparisons, α = 0.1.
Kruskal–Wallis test with post hoc least significant difference comparisons, α = 0.1.
Humans with DS exhibit a characteristically reduced cerebellar volume (Jernigan and Bellugi, 1990; Raz et al., 1995; Aylward et al., 1997) and reduction in granule cell density (Baxter et al., 2000), but it is not known how this anatomical abnormality influences cerebellar function. The cerebellum is responsible for motor learning and event timing, and cerebellar lesions cause movement abnormalities that may be due to an inability to control precise timing (Ivry et al., 2002). Additional roles for the cerebellum remain an area of active investigation (Schmahmann, 1997). To date, there are few reports of cerebellar function in DS populations. Eye-blink conditioning is abnormal in individuals with DS who are over 35 years of age (Woodruff-Pak and Papka, 1996) but normal in adolescents (Stedron et al., 2002). DS also results in impaired performance on finger-tapping tasks (Stedron et al., 2002). Development of corresponding tests in mice is ongoing. Surprisingly, Ts65Dn mice perform as well as or better than euploid littermates in an accelerating rota-rod paradigm used to assess balance and coordination (Baxter et al., 2000; Hyde et al., 2001).
It is intriguing that, whereas cerebellar volume is reduced to the same degree in Ts65Dn and Ts1Cje mice, granule cell density is significantly reduced in the former but not the latter. In Ts65Dn mice, the combination of reduced granule cell density and reduced size of the IGL results in a substantial reduction in the number of granule cells as compared with euploid mice (Baxter et al., 2000). The estimated number of granule cells in Ts1Cje is higher than Ts65Dn, because the density is only slightly reduced, yet the reduction in volume is the same in Ts65Dn and Ts1Cje mice. The few Ms1Cje/Ts65Dn mice had a significant reduction in granule cell density, to the same level as that in Ts1Cje, but no reduction in volume. Thus, although cerebellar granule cells account for 90% of all cells in the cerebellum (Gao et al., 1991), the number of these cells does not correlate with the size of the cerebellum. Because the brain and skull arise in concert, it may be significant that the occiput, which overlies the back of the brain, including cerebellum, is flattened in DS and is also smaller in both Ts65Dn and Ts1Cje adult mice (Richtsmeier et al., 2002).
The Ts65Dn skull and craniofacial skeleton display several direct parallels with developmental anomalies that occur in DS (Richtsmeier et al., 2000). When the same craniofacial parameters are measured in Ts1Cje mice, the overall phenotype is somewhat attenuated relative to Ts65Dn, and a few features affected in Ts65Dn are not significantly affected in Ts1Cje (Richtsmeier et al., 2002). Like Ts65Dn, however, the inter-individual variability of Ts1Cje mice is higher than euploid. An analogous pattern is seen with behavioral analysis, in which Ts1Cje mice are somewhat less affected than Ts65Dn in the Morris water maze (Sago et al., 1998). No single gene at dosage imbalance has been shown to cause any of these phenotypes.
The question of how dosage imbalance for HSA21 genes causes features of DS remains a subject of debate (Pritchard and Kola, 1999; Shapiro, 1999; Reeves et al., 2001). Both the qualitative and the quantitative hypothesis could be construed to support the results obtained here, but neither predicts the outcome of these experiments. More than 20 years of investigations of transgenic mice have produced no examples of single or even groups of transgenes (e.g., YAC transgenic mice, Smith et al., 1997) that reproduce phenotypes seen in Ts65Dn and Ts1Cje mice with segmental trisomy. Clearly, many phenotypes of DS will arise from the interaction of multiple genes and also from cascade effects of multiple affected tissues and processes during development. In other words, it is not possible to understand the genetic causes of aneuploid phenotypes without a thorough understanding of the etiology of particular features.
Trisomy 21 is the most complex genetic insult compatible with human survival. We suggest that simplistic explanations of the mechanisms by which gene dosage affects development will not adequately explain most aspects of the multifarious phenotypes of DS. A complete gene list and significant progress toward a spatial and temporal gene expression catalog (Hattori et al., 2000; Gitton et al., 2002; Reymond et al., 2002) provide fundamental information that is necessary to approach this complex problem. Detailed, quantitative observations of phenotype in mouse models that not only replicate but predict DS phenotypes provide the biological substrate for direct experimentation that will be required to understand the mechanisms that underlie the complex effects of trisomy 21 on development and function (Baxter et al., 2000; Cooper et al., 2001; Richtsmeier et al., 2002).
Ts1Cje mice (Sago et al., 1998) on the C57BL/6JEi background were bred to C3H/HeJ animals (The Jackson Laboratory, Bar Harbor, ME). Offspring with the translocation were mated to (C57Bl/6JEi × C3H/HeJ)F1 mice to produce the equivalent genetic background to Ts65Dn. That is, Ts65Dn mice are an advanced B6 X C3H intercross, meaning that each individual contains on average 50% B6 alleles and 50% C3H alleles such that an average 25% of loci are homozygous B6, 25% are homozygous C3H, 50% are heterozygous, and the actual loci in each class differ among individuals. The same is true for F2 mice. Using trisomic and euploid littermates for relative comparisons further reduces genetic variation between trisomic and euploid mice.
To generate Ms1Cje/Ts65Dn animals, balanced translocation carrier T(12;16)1Cje heterozygous Sod1 male mice (Sago et al., 2000) on a C57BL/6JEi background were bred to Ts65Dn females on a (C57Bl/6JEi × C3H/HeJ)F2 background. The 1216 chromosome is marked by a neomycin resistance sequence and is identified by polymerase chain reaction (PCR). The 1612 chromosome lacks Sod1 and is identified by tracking the CuZnSOD protein polymorphism (AA, AC, or CC). The 1716 chromosome is identified as a small chromosome by cytogenetic analysis. Therefore, each genotype is identified by a combination of PCR, superoxide dismutase (SOD) gel staining, and cytogenetic analysis. Primers for PCR amplification of the neomycin resistance sequence were Cite 19UP (5′ ctcgccaaaggaatgcaaggtctgt 3′) and Cite 324L (5′ cccttgttgaatacgcttgaggaga 3′), and those for internal controls were Grik1F2 (5′ ccccttagcataacgaccag 3′) and Grik1R2 (5′ ggcacgagacagacactgag 3′). PCR was performed by using a hot start followed by 30 cycles of 94°C, 30 sec, 62°C, 30 sec, 72°C, 30 sec in a 25-μl reaction mixture containing DNA (50–100 ng), 10 mM Tris-HCl (pH 8.3), 50 mM KCl, 1.5 mM MgCl2, 0.001% gelatin, 0.2 mM dNTPs, 0.25 μM of each primer, and 0.7 units of AmpliTaq polymerase (Perkin-Elmer). The CuZnSOD a and c isoforms were detected by staining for SOD activity from blood samples as described (Luche et al., 1997). Conventional karyotyping was performed by using PHA-stimulated lymphocytes as described (Davisson and Akeson, 1987).
Both males and females were used, as there is no sexual dimorphism in adult mouse brain volume in these strains (Goffinet and Rakic, 2000). All mice were 10 to 16 weeks old when perfused. All procedures were approved by Institutional Animal Care and Use Committees.
Mice were deeply anesthetized with methoxyfluorane (Metofane, Medical Developments, Springvale, Australia) and perfused intracardially with 4% paraformaldehyde as described in (Baxter et al., 2000).
MRI images of the brains of six Ts1Cje and six euploid littermate controls were obtained as described in (Baxter et al., 2000) using a 400 MHz Omega NMR Spectrophotometer (General Electric) interfaced to a 9.4 T/89mm vertical bore magnet equipped with Accustar actively shielded gradients. The data matrix size was 256 × 256 × 64 pixels, and the corresponding field of view was 16 mm × 16 mm × 8 mm, yielding pixel resolution of 62.5 μm × 62.5 μm × 125 μm.
Brains of three Ms1Cje/Ts65Dn and four euploid littermate controls were suspended in fomblin for MRI imaging and vacuumed for 5 to 10 min to remove air bubbles. Images were obtained as in Baxter et al., (2000) but with adiabatic radiofreqency pulses of 2-msec duration, repetition time of 1.2 sec, and echo time of 80 msec. The sampling data points were 512 × 70 × 64 pixels, resulting in a data matrix of 512 × 140 × 128 after zero filling. The field of view was 17 mm × 11 mm × 10 mm. Images were postinterpolated to make the volumetric pixel size isotropic (33 μm3). The investigator was blinded to the sample genotypes for both MRI collection and data analysis.
MRI images were analyzed by using the MRIcro program (Rorden and Brett, 2000) available at http://www.psychology.nottingham.ac.uk/staff/cr1/mricro.html. Volumetric measurements were taken of cerebellum and total brain. All measurements excluded olfactory bulbs, brainstem posterior to the cerebellum, cerebellar peduncles, and ventricular space. Brain regions were defined manually by using atlases of the mouse and rat brain for reference (Slotnik and Leonard, 1975; Paxinos and Watson, 1986). Entire brain images were analyzed except in the case of tissue damage to one lateral half of the brain. In these cases, volumes were calculated from the sagittal midline to one lateral edge and then doubled for an estimate of the total brain.
Brains were embedded in paraffin, and serial sections of 5 μm were taken that spanned the midline and were stained with hematoxylin and eosin. For low-magnification imaging of total mouse brain sections to analyze midline cross-sectional area, digital images were made by using an Epson Perfection 1240U scanner at 1,200 dpi resolution. All measurements excluded the olfactory bulbs, brainstem posterior to the cerebellum, and ventricular spaces. For intermediate-magnification imaging of the cerebellum to measure Purkinje cell layer length, sections were viewed by using a Nikon SMZ-2T dissection microscope at ×15 magnification, which was coupled to an Alpha Imager 2000 video camera system (Alpha Innotech, San Leandro, CA). High-magnification imaging of the cerebellar cell layers was accomplished by viewing sections with a Zeiss Axiophot microscope at ×40 magnification and digitally capturing images by using a Photometrics CH250 CCD camera and the Metamorph 4.5.5 imaging program (Universal Imaging Corporation, Downingtown, PA). Images were analyzed by using the BrainImage program (A.L. Reiss, 1999, BrainImage version 2.2.4, Stanford University Psychiatry Neuroimaging Lab).
Granule cells were counted within square 5,000 μm2 fields of the internal granule cell layer. The granule cell count for each mouse brain was averaged from 12 independent, nonoverlapping, randomly selected fields from mid-sagittal cerebellar folia III, IV, and V. The linear density of Purkinje cells was calculated by dividing the total number of Purkinje cells in a section by the length of the Purkinje layer. The Purkinje cell density value calculated for each mouse brain was the average of the values for two midline sagittal sections. We showed previously that equivalent results for granule and Purkinje cell density are obtained when the analysis is performed on midline or lateral sections (Baxter et al., 2000).
Differences between either Ts1Cje or Ms1Cje/Ts65Dn and euploid littermates (within strain differences), for cerebellar volume, granule cell density, and Purkinje cell density were detected by using a one-tailed Student's t-test (PROC TTEST; SAS, Cary, NC). Results were verified by using a nonparametric Kruskal–Wallis test (PROC NPAR1WAY WILCOXON) in cases of small sample sizes and slight deviations from normality. To compare these phenotypes across the Ts65Dn, Ts1Cje, and Ms1Cje/Ts65Dn genotype classes (between strain differences), an index was established for each animal with a strain-specific relationship to euploid animals. The measured volume or density of an individual animal was divided by the average volume or cell density of the euploid animals of that strain, giving a percentage of the euploid value (Fig. 5). Analysis of variance was used to determine differences between strains (PROC GLM). When slight deviations of normality occurred due to small sample sizes of Ms1Cje/Ts65Dn mice, a nonparametric Kruskal–Wallis test (χ; PROC NPAR1WAY WILCOXON) was used. Least significant difference post hoc comparisons (contrasts) were used to determine differences between strains for individual phenotypes. A significance level of α = 0.10 was used in all multiple comparison tests.
We thank V.P. Chacko, S. Mori, and R. Xue for technical assistance with the MRI analysis. L.E.O. was supported on a Howard Hughes Predoctoral Fellowship. This work was supported by NICHD awards to R.H.R. (HD 24605) and to C.J.E.