Cerebral palsy (CP) is a term used to define a group of disorders [1] characterized by a non-progressive abnormality of posture and movement, resulting from defects in the developing nervous system [2]. Approximately 1 in 250 to 1000 live births presents with CP, making it one the commonest congenital disabilities [3]. Many different aetiological factors have been implicated. Among preterm infants, the incidence of CP generally increases with decreasing gestational age and the origin in most cases may be traced to post/peri-partum periventricular leukomalacia and intraventricular/periventricular haemorrhage [4]. Conversely in term infants perinatal causes can only confidently be attributed where there is documented perinatal hypoxia/acidosis and clinical encephalopathy in the early neonatal period [5]. Prenatal risk factors in the aetiology of CP include low birth-weight, intrauterine infection and exposure to teratogens during pregnancy [6, 7]. The cause in a large proportion of cases remains obscure.

Depending on the overall clinical picture, CP can be sub-classified into a number of phenotypic groups [8, 9]. Dyskinetic CP accounts for ~20% of all cases, which may be further divided into choreoathetotic (5%) and dystonic (15%) forms. Ataxic CP (~10% of all cases) can also be sub-divided into two forms, simple (congenital) ataxia (5%) and ataxic diplegia (5%). Spastic CP is the most prevalent sub-type (~70%) and was the phenotype of the probands in this study [10]. It is characterised by muscular hypertonicity and pronounced rigidity of the affected limbs. Spastic CP can be sub-classified according to the topography of the affected limbs as hemiplegic (20%), monoplegic (<1%), diplegic (40%), or quadriplegic (10%) [11].

Kuban and Leviton [3] suggested that CP could be genetic in origin, as well as the result of environmental insult at any point during CNS development. Most estimates place the proportion of CP cases with a genetic aetiology at between one and two percent of the total [12]. Among infants and children with spasticity, symmetry of neurological signs has been identified as a strong indicator of a probable genetic aetiology [13, 14]. The proportion of cases demonstrating Mendelian inheritance varies among the different sub-types of CP [2, 13, 7]. X-linked, autosomal dominant and recessive inheritance patterns have been described for non-progressive CP. An ataxic diplegic autosomal recessive trait (OMIM:605388) [15] has been mapped to chromosome 9p12-q12. Progressive spastic paraplegia (SPG) has a similar pathology to CP. SPG displays autosomal dominant (SPG3A at 14q11-q21 encoding the atlastin GTPase; SPG4 at 2p21-22 encoding spastin, an AAA family ATPase/chaperonin; SPG6 at 15q11.1; SPG8 at 8q23-q24; SPG9 at 10q23-q24; SPG10 at 12q13; SPG12 at 19q13; and SPG13 at 2q24 encoding the HSP60 mitochondrial chaperonin), recessive (SPG5A at 8cen; SPG7 at 16q24.3 encoding paraplegin, an AAA family ATPase/inner mitochondrial membrane chaperonin; SPG11 at 15q13-q15; SPG14 at 3q27-q28; SPG15 at 14q22-q24; and SPG17 at 11q12-q14), or X-linked inheritance patterns (SPG1 at Xq28 encoding the L1CAM adhesion molecule; SPG2 at Xq22 encoding proteolipid protein-1; and SPG16 at Xq11.2).

A non-progressive, autosomal recessive, symmetrical spastic CP locus has been mapped to a 5 cM region between D2S124 and D2S333, at 2q24-31.1 (LOD score of 5.75) in consanguineous families originating from the Mirpur region of Pakistan (OMIM:603513) [10]. Affected individuals had no identifiable perinatal cause of CP, or underlying diagnosis and presented with developmental delay, mental retardation and sometimes epilepsy as part of the phenotype. We initially performed detailed physical mapping of the 5 cM region, so as to accurately define the marker order and to refine the linkage interval. The positions of a large number of genes and ESTs were defined accordingly, allowing the rapid identification of candidate disease genes. The minimum region of homozygosity was reduced to 0.5 cM by typing large numbers of microsatellite markers in the families. Subsequently, portions of this region have been sequenced in the human genome project. Within the region we have concentrated on the positional candidate GAD1, which codes for the 67 kDa isoform of L-glutamate decarboxylase (GAD EC:4.1.1.15).

GAD requires the cofactor pyridoxal 5'-phosphate (PLP) and catalyses the production of Gamma-aminobutyric acid (GABA) from glutamate [16]. Two separate, independently-regulated genes, GAD1 and GAD2 (at chromosome 10p11, encoding a 65 kDa GAD isoform), have presumably arisen by duplication and been conserved during evolution, as indicated by their sequence homology [17] and the retention of common intron-exon boundary splice sites [18]. Their N-termini demonstrate ~23% homology, but their C-termini, which contain the catalytic site, have ~73% amino acid sequence identity between the isoforms [19]. GABA and glutamate are the most abundant amino acid neurotransmitters in the brain. GABA, an inhibitory neurotransmitter, and excitatory glutamate, both play important roles in synaptic plasticity and neuroendocrine function [20]. Both isoforms of GAD are also involved in intermediary metabolism, participating in the GABA shunt, which bypasses two steps of the TCA cycle [21]. GAD activity of both isoforms, is ubiquitous, but highest in the brain and pancreatic islets of Langerhans. We therefore performed detailed mutational screening of GAD1 in familial spastic CP probands and unaffected family members.

CP is a term used as a collective definition for a group of neurological disorders [3]. The pathophysiology in most cases is poorly understood, but includes genetic syndromes, congenital malformation, infective intra-uterine encephalitis, cerebral haemorrhage or infarction, ischemic damage, periventricular leukomalacia (PVL), and non-infarctive telencephalic leukomalacia [11]. The contribution of Mendelian inherited cases of CP accounts for approximately 2% of the total number [2, 22]. A non-progressive form of autosomal recessive spastic CP has been identified [13]. McHale et al. [10] succeeded in identifying a 5 cM region on chromosome 2q24-31.1, which segregated with disease in consanguineous families. Linkage analysis identified a locus between markers D2S124 and D2S333, which produced a LOD score of 5.75, sufficient to warrant the further investigation described herein.

To refine and confirm the genetic marker order across a region, which was incompletely sequenced at the time, we used YAC and PAC clones to construct a physical framework and performed PCR to map ESTs and microsatellite markers to clones within the partial contig (Figure 1). When this contig was integrated with the BAC sequence contigs, rearrangement of the BAC order was necessary. With each subsequent DNA sequence update, the degree of inconsistency was reduced and this led to revision of microsatellite order compared with that used previously [10]. Using the YAC sizes, the sizes of gaps in the BAC contig could be estimated. Having generated a detailed map spanning the region, we selected microsatellite markers at evenly spaced intervals across the locus. These markers were used to refine the minimum region homozygous by descent in linked families, to between the markers D2S2345 and D2S326. The physical size of the region between these markers is approximately 0.5 cM. There was no suggestion of a founder haplotype common to the two families (Figure 2).

The Goldenpath Human Genome Working Draft Assembly 2001, is an annotation of the Washington BAC contig, combining sequence data of BACs, ESTs, known genes and hypothetical genes. We mapped ESTs and known, uncharacterised or hypothetical genes on the basis of sequence homology (NCBI http://www.ncbi.nlm.nih.gov, Whitehead http://www.genome.wi.mit.edu and Goldenpath databases), onto our YAC/PAC contig. ESTs were then collated according to their expression profiles to produce a "binned" EST map, on which candidate gene selection could be based. This reduced the number of hypothetical genes in the region and allowed the combination of genetic, physical mapping and expression data, into a single comprehensive map.

One interesting candidate within the minimal region was GAD1, which encodes GAD₆₇. Expression of its transcript is ubiquitous, including the CNS. The main function of GAD₆₇ is to catalyze the conversion of the excitatory amino acid and neurotransmitter glutamate to GABA, the main inhibitory neurotransmitter in the CNS [23]. In the developing CNS, GABA has an important role in neuronal differentiation and the control of plasticity [21]. GABA has also been implicated in the pathogenesis of various seizure and movement disorders [20].

Vertebrates have two separate genes coding for GAD, which produce distinct forms of the enzyme. GAD1 and GAD2 have diverged relatively recently in evolution, as indicated by their degree of sequence homology and the retention of common intron-exon boundary splice sites [17] (Figure 5A). The variants of GAD differ in molecular weight, cellular and sub-cellular localisation, and their interaction with the cofactor PLP [18, 20, 24].

GAD2, located at 10p13-p11.2, is transcribed to produce a 5.6 kb mRNA in islets and brain, encoding a 65 kDa protein (585 AA residues). The 67 kDa (594 AA residues) form [17] is localised to 2q25-26 and encoded by a 3.7 kb transcript (GAD1) [23]. There is also evidence for a 25 kDa inactive protein (GAD₂₅) produced from an alternatively spliced GAD1 transcript of 2 kb that contains an in-frame stop codon. This GAD1 splice variant has only been found in human islets, testis and adrenal cortex, although the homologue is present in fetal mouse brain [25]. GAD₆₇ and GAD₆₅ consist of two major sequence domains (Figure 5B). The N-termini (AA residues 1–94 in GAD₆₅ and 1–101 in GAD₆₇) demonstrate ~23% homology. These N-terminal domains are thought to be responsible for sub-cellular targeting and the formation of GAD₆₅–GAD₆₇ heterodimers [26]. The C-terminal domains (AA residues 96–585 in GAD₆₅ and 102–594 in GAD₆₇) contain the catalytic site, with ~73% sequence identity between the isoforms [19] (Figure 5).

In the CNS, GAD₆₅ appears to be preferentially distributed in axon terminals and the associated synaptic vesicles, whereas GAD₆₇ is also located in the cell bodies and more uniformly distributed throughout the neuron [24]. This suggests that each GAD isoform is involved in the synthesis of GABA in different sub-cellular compartments [21]. This is supported by the discovery that GAD₆₅ is the main source of apoGAD (an inactive reservoir), which responds to short-term changes in neuronal activity and is more responsive to levels of PLP [18]. On the other hand, GAD₆₇ predominantly exists bound to the PLP cofactor (holoGAD), providing a constitutive level of GABA production [20]. Bond et al. [27] showed that GAD₂₅ is expressed in a temporally controlled manner, in the developing striatum and cortex in rodents, suggesting this may provide a mechanism of regulating GABA production in differentiating neurons.

Asada et al. [28] undertook the selective elimination of each GAD isoform in order to determine their respective roles. Gad2-/- mice are slightly more susceptible to seizures, consistent with an excitatory increase in the relative ratio of glutamate/GABA. However, they showed no obvious overall change in neuronal GABA content. Therefore GAD₆₇ alone appears to produce sufficient GABA for effective neurotransmission [21]. Gad1-/- mice demonstrated a decrease of ~20% in total glutamate decarboxylase activity at birth. This was assayed by the conversion of ¹⁴C-labelled glutamate to ¹⁴CO₂ in the presence of PLP. There was also a marked (7%) reduction in total GABA content in cerebral cortex homogenate measured by liquid chromatography [28]. Unfortunately, these mice died neonatally of severe cleft palate, masking any potential neurological dysfunction and also illustrating a role for Gad₆₇ in non-neural tissues [21]. It is of interest that mice with mutations in the β-3 GABA receptor (GABRB3) at the Angelman syndrome (OMIM:105830) locus, also display cleft palate, implying a key role for GABA signalling in normal palate development [29, 30].

Pyridoxine-dependent epilepsy (PDE) is a rare autosomal recessive disorder (OMIM:266100), characterized by generalized seizures during the first hours of life. The associated pathology may result from an alteration in the binding of the co-factor PLP to GAD. Interestingly epilepsy is commonly associated with CP and grand mal epilepsy developed at age six months in the two linked pedigrees described here [10]. GAD1 mutation was previously suspected of being the cause of PDE. Linkage of pyridoxine-dependent epilepsy has however been reported to 5q31.2-31.3, with GAD1 and GAD2 excluded [31]. Decreased levels of brain and CSF GABA, increased levels of CSF and cortical glutamate, and decreased levels of PLP in the frontal cortex, have been described in this condition.

GAD₆₅ and GAD₆₇ have been identified as auto-antigens in "Stiff Person Syndrome" (SPS, OMIM:184850), and in cerebellar ataxia [32–34]. GABA-mediated synaptic transmission is thought to be functionally impaired by the production of autoantibodies to GAD₆₅ and GAD₆₇ [35–37]. This results in a reduction in brain levels of GABA, prominent in the motor cortex, which can be demonstrated by Magnetic Resonance Imaging (MRI) in SPS patients. SPS is a disabling disorder characterised by muscle rigidity and episodic spasms of the musculature, thought to be due to autoimmune-mediated dysfunction of supraspinal GABAergic inhibitory neurons [38]. Hyperexcitability of the motor cortex in SPS has been demonstrated by transcranial magnetic stimulation [39].

Anti-GAD₆₅ auto-antibodies in the CSF of ataxic and SPS patients selectively suppress GABA-mediated transmission in cerebellar Purkinje cells, without affecting glutamate-mediated transmission [37, 40]. Low CSF levels of GABA have been reported in patients with Kok disease (OMIM:149400 also known as hyperexplexia/exaggerated startle reaction/startle disease) [41]. The exact mechanism by which autoantibodies target these intracellular GAD antigens is not clear. However, it is interesting that SPS may also arise in individuals with autoantibodies to gephyrin, a cytosolic protein concentrated at the postsynaptic membrane of inhibitory synapses where it is associated with GABA_A receptors [42]. This provides a further example of chronic rigidity and spasm possibly secondary to disruption of the inhibitory synapses.

Mutations in the CLN3 gene are thought to be responsible for the neurodegenerative disorder Batten disease (OMIM:204200). In cln3-knockout mice autoantibodies to GAD₆₅ have been reported to be associated with brain tissue and result in inhibition of GAD activity [43]. These mice also demonstrate elevated brain glutamate levels as compared with controls, which may have a causative role in the astrocytic hypertrophy evident in cln3-knockout mice and along with anti-GAD₆₅ autoantibodies in Batten disease patients may contribute to the associated preferential loss of GABAergic neurons.

Drugs, which potentiate the action of GABA, such as benzodiazepines and baclofen, ameliorate muscle rigidity and spasticity. These GABA agonists are thought to counter disinhibition of the velocity-dependent increase in skeletal muscle during stretch reflexes, observed in spasticity, which is the result of inadequate presynaptic inhibition of the muscle spindles [44, 35, 37]. γ-Vinyl-γ-aminobutyric acid (GVG) is used to treat neurological disorders including epilepsy, tardive dyskinesia and spasticity. It has been reported that it is the GABA-elevating effect of this compound that is responsible for its anti-convulsive properties [20].

We have identified a GAD1 sequence change G(36)C, which segregates with autosomal recessive spastic CP in 4 affected siblings. This nucleotide substitution causes a missense mutation, changing serine (12) to a cysteine in the N-terminal domain. This serine residue is conserved between all mammals (human/mouse/rabbit/pig) for which data are available. The association of GAD₆₇ with membranes requires formation of heteromeric links with GAD₆₅, which are mediated via their N-terminal domains. The N-terminus of GAD₆₅ is palmitoylated and binds to the cellular membrane. The first 27 amino acids appear to be essential in this function [32]. The palmitoylation of cysteines 30 and 45 of GAD₆₅, and the inability of residues 1–29 of GAD₆₇, to substitute for this region, highlights the potential impact on cellular localisation of a nucleotide substitution in this domain [20]. GAD₆₅ also undergoes phosphorylation of the first four serine residues in the N-terminal domain. These post-translational modifications highlight the importance of the flexibility and accessibility of this domain. N-terminal epitopes of GAD₆₅, in the region corresponding to the residue, which undergoes mutation in GAD₆₇, are particularly prominent autoantigens [45].

S(12)C amino acid substitution may thus produce subtle effects on cellular localisation, protein-protein interactions and/or protein processing, with a subsequent effect on GABA production. This is not inconsistent with the mouse Gad1 knockout where complete loss of Gad1 enzymatic function (~20% reduction of total Gad activity in the cerebral cortex) resulted in a cleft palate phenotype and neonatal death [28, 30]. There is redundancy of GABA production as a result of the presence of two GAD proteins, and the precise function of each isoform may differ between man and mouse. It is interesting to note that the GAD₂₅ splice variant of GAD₆₇, also contains the S(12)C amino acid substitution in affected CP patients. This truncated variant is identical to the first 213 amino acids of GAD₆₇, with the addition of an extra 11 C-terminal residues. It lacks the binding site for the cofactor PLP and is believed to lack any GAD activity [25]. The function of GAD₂₅ is not known, but it may compete with GAD₆₇ for incorporation into protein complexes. Therefore the presence of an N-terminal mutation would affect both GAD₂₅ and GAD₆₇, and may disrupt a complex regulatory mechanism for GAD₆₇.

This study illustrates the difficulty of gene cloning in rare autosomal recessive diseases mapped in small, consanguineous pedigrees. Identification of an ancestral haplotype allows refinement of the locus, but this has not been possible in this example. Within the minimal linkage region, any sequence change will segregate with the disease phenotype. Detection of a nonsense mutation leading to a protein truncation would provide compelling support for a mutation as causative. However, in the present example we have not seen such a mutation in the candidate gene so far examined. We are now expressing the variant forms of GAD₆₇ (S12C) and GAD₂₅ (S12C), as recombinant proteins, to assess catalytic activity and binding properties with respect to their normal counterparts. However, it may well prove difficult to detect subtle effects based on sub-cellular localisation changes in the mutant proteins, or changes in post-translational modification patterns.

Stability of the mRNA transcripts from these variants will be assessed by transfection studies in neuronal cells. Eventually, it would be of interest to attempt knock-in experiments with GAD1 (G36C), into the Gad1-/- mouse to see if this can rescue the cleft palate phenotype and reveal a CP-like picture. These experiments will be reported elsewhere. However, the possibility that reduced GAD₆₇ activity may cause CP in the patients studied herein, in a manner reminiscent of that seen in SPS, leads us to report our findings at this stage. The success reported in treating SPS with intravenous immunoglobulin [40, 46], suggests further evaluation of GABA agonists in the management of this difficult clinical problem.