Language dominance in neurologically normal and epilepsy subjects A functional MRI study

Transcription

1 Brain (1999), 122, Language dominance in neurologically normal and epilepsy subjects A functional MRI study Jane A. Springer, 1 Jeffrey R. Binder, 2,3 Thomas A. Hammeke, 2 Sara J. Swanson, 2 Julie A. Frost, 2 Patrick S. F. Bellgowan, 2 Cameron C. Brewer, 6 Holly M. Perry, 5 George L. Morris 2 and Wade M. Mueller 4 1 University of Iowa Hospital and Clinics, Iowa City, Correspondence to: Jeffrey R. Binder, MD, Department of Departments of 2 Neurology, 3 Cellular Biology and Anatomy Neurology, Medical College of Wisconsin, 9200 West and 4 Neurosurgery, Medical College of Wisconsin, Wisconsin Avenue, Milwaukee, WI 53226, USA 5 Department of Psychology, Marquette University, jbinder@post.its.mcw.edu Milwaukee, USA and 6 Department of Psychology, University of Scotland, Glasgow, UK Summary Language dominance and factors that influence language were considered left hemisphere dominant and 6% had lateralization were investigated in right-handed, neurologically bilateral, roughly symmetric language representation. normal subjects (n 100) and right-handed None of the normal subjects had rightward dominance. epilepsy patients (n 50) using functional MRI. Increases There was greater variability of language dominance in in blood oxygenation-dependent signal during a semantic the epilepsy group, with 78% showing left hemisphere language activation task relative to a non-linguistic, dominance, 16% showing a symmetric pattern and 6% auditory discrimination task provided an index of showing right hemisphere dominance. Atypical language language system lateralization. As expected, the majority dominance in the epilepsy group was associated with an of both groups showed left hemisphere dominance, earlier age of brain injury and with weaker right hand although a continuum of activation asymmetry was dominance. Language lateralization in the normal group evident, with nearly all subjects showing some degree was weakly related to age, but was not significantly related of right hemisphere activation. Using a categorical to sex, education, task performance or familial left- dominance classification, 94% of the normal subjects handedness. Keywords: cerebral dominance; language; functional MRI; lateralization Abbreviations: EHI Edinburgh Handedness Inventory; fmri functional MRI; IAP intracarotid amobarbital procedure; LI laterality index Introduction Hemispheric specialization for language functions has been is often assumed to carry out essentially all language a central topic in brain research for well over a century (Dax, functions, an alternative possibility is that there are varying 1865). Despite the vast literature on this topic, several basic degrees of dominance that determine vulnerability to language issues regarding language lateralization remain unresolved. deficits following injury to either the right or left hemisphere. Although it is generally accepted that most people have left Finally, there is evidence that several genetic, developmental, hemisphere language dominance, the actual incidence of environmental and pathological factors may influence atypical dominance in the general population is not known. language lateralization, but the power of these influences The risk of losing language function after right brain surgery remains poorly defined. for adult-onset disease is consequently estimated to be small, Several techniques have been used to examine these issues. but has not been determined precisely. For the majority of The intracarotid amobarbital procedure (IAP) (Wada and the population with left hemisphere dominance, it is not Rasmussen, 1960; Loring et al., 1992) enables testing of known to what extent the right hemisphere also contributes each cerebral hemisphere individually for language to language processing. Although the dominant hemisphere competence after temporary anaesthetization of the opposite Oxford University Press 1999

2 2034 J. A. Springer et al. hemisphere. It has been called the least ambiguous method (McGlone, 1977, 1980), though this finding may also be for determining language dominance (Snyder et al., 1990) explained by lesion size and location effects (Kertesz and and is routinely used in the presurgical evaluation of epilepsy Sheppard, 1981). patients (Risse et al., 1997). Findings from IAP studies Adult-onset lesion studies may more closely reflect normal provide estimates of language dominance patterns in the language dominance patterns than do studies of epilepsy general population (Hecaen and Albert, 1978; Kolb and patients, yet this method provides only an indirect and Whishaw, 1990) and suggest that handedness, age of seizure approximate measure of these patterns. Bilateral language onset and sex may all influence language lateralization. In representation cannot be determined in individual cases from one of the largest studies to date, Rasmussen and Milner a unilateral lesion. The likelihood of detecting aphasia after examined the effects of early brain injury and handedness a lesion depends not only on the side of injury, but also on on language dominance (Rasmussen and Milner, 1977). the extent and location of the lesion within the hemisphere Among 396 epilepsy patients who underwent the IAP, ~96% and the sensitivity of the language tests used. Similar lesions of right-handers and 70% of left-handers without early brain often cause different effects in different patients, and it injury showed left hemisphere dominance for simple speech may be difficult or impossible to determine whether these functions. The incidence of atypical speech representation differences reflect true variability in functional organization was higher in patients with early brain injury. Results of or simply small differences in lesion characteristics or other IAP studies (Branch et al., 1964; Mateer and Dodrill, assessment methodology. Most important, the method may 1983; Strauss and Wada, 1983; Rausch and Walsh, 1984; underestimate the incidence of right hemisphere participation Powell et al., 1987; Rey et al., 1988; Woods et al., 1988; in language, since the absence of aphasia after right Zatorre, 1989; Loring et al., 1990; Helmstaedter et al., 1997; hemisphere injury could be due to sparing of relatively small Risse et al., 1997) show that the incidence of left hemisphere language areas, insensitivity of the examination methods to language dominance ranges from approximately 63% to 96% subtle disturbances or rapid compensation for the injury by for right-handers and from 48% to 75% for left-handed the dominant hemisphere (Albert et al., 1981). and ambidextrous patients. Investigations of sex effects on Unlike IAP and lesion methods, which are based on language lateralization using IAP are less common, although inference of language localization from observation of one study found that women with a left hemisphere seizure language deficits, activation imaging techniques such as PET focus were more likely than men to show right hemisphere and functional MRI (fmri) provide direct observation of language functions (Helmstaedter et al., 1997). Despite the brain activity during language processes. Such studies significant contributions of these studies to our current demonstrate left-lateralized activation in normal subjects understanding of language laterality, use of the IAP is during a variety of language tasks (Petersen et al., 1988; precluded in normal subjects because the procedure is invasive Frith et al., 1991; Wise et al., 1991; Démonet et al., 1992; and has health risks (Rausch et al., 1993). Because epilepsy Howard et al., 1992; Sergent et al., 1992; Zatorre et al., is associated with varying degrees of brain dysfunction and 1992; Price et al., 1996b; Vandenberghe et al., 1996; Binder tissue damage, reorganization of language functions may et al., 1997). However, inferring language dominance from occur relatively often in such patients (Woods et al., 1988). such data requires consideration of two critical issues. First, Extrapolations about the incidence of atypical language the activation task used may engage a number of brain systems dominance in normals from the findings of IAP studies may not specifically related to language, including sensory, motor thus be inaccurate, presumably resulting in overestimations and attentional systems. Measurement of language (Risse et al., 1997). lateralization with such techniques thus requires that Investigations of previously normal patients with unilateral activation of these systems be controlled for by a contrast or lesions secondary to stroke show that the incidence of baseline task. Secondly, even when such controls are crossed aphasia (i.e. aphasia in right-handed patients after employed, it is conceivable that some areas engaged a right hemisphere injury) is 2% (Zangwill, 1967; Gloning specifically by the language task may not be critically et al., 1969; Hecaen and Albert, 1978). Compared with the necessary for normal language performance. Thus, the extent IAP studies, these data thus suggest an even lower incidence to which activation imaging data correlate with conventional of right hemisphere language dominance in right-handed measures of language dominance based on lesion methods individuals. Lesion studies also indicate greater incidence of must be determined empirically for each language atypical language organization in left-handed than in righthanded activation task. patients. Left-handed aphasic patients tend to have These issues are underscored by the fact that in virtually incomplete and unusual aphasic syndromes (Brown and all of the activation studies cited above, some right hemisphere Hecaen, 1976) and to recover more rapidly (Hecaen and activation was observed during the language tasks. How Sauguet, 1971), although these findings may not be robust should this non-dominant activation be interpreted? Most of when lesion location and size are controlled for (Gloning the reports involved averaged group data, so one possibility et al., 1969; Naeser and Borod, 1986; Borod et al., 1990). is that the right hemisphere activation reflects inclusion of a Several lesion studies also suggest that men have more few right-dominant subjects in the group. However, studies strongly lateralized language functions than women that included individual data show that subjects typically

3 Language dominance and fmri 2035 have variable amounts of right hemisphere activation even variable examined thus far using functional imaging methods when their overall pattern is left-lateralized (Hinke et al., in the normal population was subject gender, and these 1993; Binder et al., 1995, 1996; Desmond et al., 1995; Xiong studies produced disparate findings (Buckner et al., 1995; et al., 1998). Such findings suggest either that the controls Shaywitz et al., 1995; Price et al., 1996a; Pugh et al., 1996; for activation of non-linguistic systems were inadequate or Frost et al., 1999). The effects of early brain injury on that the right hemisphere typically participates in language language organization have not been studied with functional processing even in subjects with left hemisphere language imaging. Based on data from previous IAP studies, it was dominance. This last account also implies that language predicted that either the presence of a left hemisphere seizure dominance observed in lesion studies reflects a relative focus or an earlier age of seizure onset would be associated rather than a complete lateralization of function (Loring with a higher incidence of atypical lateralization in the et al., 1990) and that such relative lateralization may be epilepsy group (Rasmussen and Milner, 1977; Mateer and continuously variable across individuals in a population. Dodrill, 1983; Rey et al., 1988; Woods et al., 1988; Strauss Support for the idea of relative lateralization and its et al., 1990; Satz et al., 1994; Helmstaedter et al., 1997; relationship to conventional concepts of dominance would Risse et al., 1997). Additional variables examined in one or require not only a demonstration of variable involvement both subject groups included age, sex, degree of righthandedness, of the right hemisphere in language processing, but also presence of left-handed family members, demonstration that the degree of right hemisphere education, level of task performance and IQ. For each of the involvement in language predicts the degree of language three goals, parallel analyses were conducted using both a impairment following right hemisphere dysfunction. We continuous measure of activation asymmetry as an index of provided such evidence in a previous study comparing fmri language dominance and a more conventional categorical data with the IAP (Binder et al., 1996). The fmri tasks, measure of dominance derived from the activation based on those of Démonet and colleagues (Démonet et al., asymmetry data. 1992), consisted of a contrast between a semantic word categorization task and a perceptual discrimination task that controls for primary sensory, motor response, working memory and attentional requirements. Variable amounts of Method left and right hemisphere activation were observed using this Participants task contrast in 22 consecutive epilepsy patients. A continuous The normal group consisted of 100 healthy subjects (48 measure of relative lateralization computed from each women, 52 men) with no history of neurological or psychiatric subject s fmri activation map was strongly correlated with disease. The epilepsy group consisted of 50 consecutive relative interhemispheric asymmetry of language deficits patients (29 women, 21 men) who met the following selection observed during the IAP. Thus, patients with relatively more criteria. All patients were undergoing comprehensive activation in the right hemisphere during fmri had relatively evaluation for surgical treatment of epilepsy that was more language deficit during right hemisphere medically intractable, restricted their lifestyle and had a anaesthetization in the IAP. These results strongly support a probable focal onset. In addition to fmri, the presurgical model of language dominance based on variable degrees of evaluation for epilepsy subjects included long-term in-patient relative lateralization. They also demonstrate that activation video-eeg monitoring, neuropsychological testing, routine measured by this fmri procedure is of direct functional brain MRI and a standardized IAP. All had a full scale IQ significance and thus related to conventional (i.e. lesion 70 as measured by the Wechsler Adult Intelligence Scale based) measures of language dominance. Revised (Wechsler, 1981). All normal and epilepsy subjects In this report we present fmri data, obtained using the were right-handed as defined by a handedness quotient 50 previously described language activation protocol, from a on the Edinburgh Handedness Inventory (EHI) (Oldfield, consecutive series of 100 right-handed, healthy subjects and 1971). Subjects with inadequate performance on the activation from a consecutive series of 50 right-handed epilepsy patients. tasks were excluded. Inadequate performance was defined as Our first goal was to describe the degree of variability in below the 5th percentile of a normative group for the language lateralization that occurs in neurologically normal, tone decision task and below chance-level for the semantic right-handed subjects and, more specifically, to estimate the decision task. English was the primary language of all incidence of atypical language dominance in this group. subjects. Subjects were recruited on a voluntary basis. After Our second goal was to compare language lateralization in full explanation of the risks and purposes of this study, the normal and epilepsy groups. Many previous estimates of all subjects gave written informed consent according to the incidence of atypical language dominance were based on institutional guidelines. The study was approved by the studies of epilepsy patients. We assessed the validity of this Human Research Review Committee of the Medical College approach by testing the hypothesis that atypical dominance of Wisconsin. is more frequent in epilepsy patients than in normal subjects. Demographic information is summarized in Table 1. The A third goal was to identify subject variables associated with two groups did not differ in handedness, sex or familial the direction and extent of language lateralization. The only sinistrality defined as having at least one left-handed parent,

4 2036 J. A. Springer et al. Table 1 Group demographic information Normals (n 100) Epilepsy (n 50) P value Mean (SD) Mean (SD) Age (years) 23.2 (5.0) 35.8 (10.2) Education (years) 15.1 (2.7) 13.2 (2.0) Handedness (EHI score*) 84.4 (14.1) 87.9 (14.4) n.s. Familial sinistrality (%) n.s. Sex (% male/% female) 52/48 42/58 n.s. n.s. not significant. *EHI Edinburgh Handedness Inventory. Table 2 Descriptive information of epilepsy group Table 3 Cognitive systems involved in the semantic decision and tone decision tasks Epilepsy variable Mean (SD) Range Cognitive system Tone decision Semantic Full scale IQ* 93.6 (9.2) task decision task Age of onset (years) Probable brain injury 9.7 (9.9) 0 32 Lexical semantic processing Intractable seizures 14.7 (10.3) 0 43 Phonetic processing Lateralization of surgery Attention, working memory Left 23/41 (56%) Auditory processing Right 18/41 (44%) Motor response Localization Temporal 38/41 (93%) Extra-temporal 3/41 (7%) sound delivery apparatus have been described elsewhere *Based on Wechsler Adult Intelligence Scale Revised; see text (Binder et al., 1995). Each echo-planar image series consisted for definition and explanation. of multiple periods of linguistic activation, during which subjects performed a semantic decision task, alternating with grandparent or sibling. The normal group was significantly periods of a control task, during which subjects performed a younger (t 10.18, P 0.001) and had more years of tone decision task. The rationale behind these tasks and formal education (t 4.34, P 0.001) than the epilepsy details of the procedure were described previously (Binder group. Information on IQ and neurological history in the 50 et al., 1995, 1996, 1997); a brief summary follows. In the epilepsy patients is provided in Table 2. Seizure localization tone decision (control) task, subjects heard trains of 3 7 data are provided for the 41 patients who underwent surgery tones. Each tone had a duration of 150 ms and a frequency after fmri. of either 500 Hz or 750 Hz. Subjects were instructed to press a button for any train containing two high-pitch (750 Hz) tones. In the semantic decision task, subjects heard names Image acquisition of animals (e.g. turtle ) and were instructed to press a button Imaging was performed on a 1.5 Tesla GE Signa scanner for animals they considered to be both found in the United (GE Medical, Milwaukee, Wis., USA) using a three-axis States and used by humans. The two tasks were matched local gradient coil and insertable transmit/receive RF coil for stimulus intensity, average stimulus duration, average (Medical Advances, Milwaukee, Wis., USA). fmri used a trial duration and frequency of positive targets. Responses blipped, gradient echo, echo-planar sequence (TE 40 ms, TR consisted of a thumb press to a magnet-compatible response 4000 ms, field of view 24 cm, matrix pixel matrix). device held in the subject s left hand. Subjects were given Functional slices were acquired in the sagittal plane and instructions and a brief practice session before entering covered the lateral two-thirds of each hemisphere (six or the scanner. seven slices per hemisphere, slice thickness 7 mm, voxel Table 3 lists the important cognitive systems thought to size mm). High-resolution, T 1 -weighted be engaged by the tone and semantic decision tasks. The anatomical reference images were obtained as a set of 124 contiguous sagittal slices using a 3D spoiled-gradient-echo sequence (GE Medical). tone task has been shown to produce activation of superior temporal auditory processing systems as well as lateral frontal lobe attentional and working memory systems. The semantic decision-tone decision subtraction produces strongly lateralized activation in prefrontal and temporoparietal heteromodal cortex of the dominant hemisphere in normal Stimuli and activation tasks Stimuli were 16-bit, digitally synthesized tones and sampled subjects, which is thought to represent engagement of male speech sounds presented binaurally at precise intervals language processes that are relatively specific to the semantic using a computer playback system. Characteristics of the decision task (Binder et al., 1995, 1997).

5 Language dominance and fmri 2037 fmri data analysis with left and those with right seizure focus. Because side of Identification of event-related MRI signal changes used the seizure focus was defined conservatively using the side of correlation approach of Bandettini and colleagues (Bandettini surgery, analysis of this variable was restricted to the 41 et al., 1993). In brief, this method correlates the time series patients who underwent surgery. Simple correlational analysis echo-planar data (excluding signal equilibration images at was used to identify numerical variables associated with LI the beginning of the series) with a reference vector on a in each group. Variables examined in the normal group were voxel-by-voxel basis after removal of linear trends. A age, education, EHI quotient and task performance. The threshold value for the correlation coefficient is chosen to relationships between these variables and a functional identify those voxels significantly correlated with the neuroimaging index of language lateralization have not as activation procedure. Reference vectors were derived directly yet been systematically studied in a large sample of normals. from each data set using a two-step method that selects Variables examined in the epilepsy group were age, education, voxels strongly correlated (r 0.70) with a simple square EHI quotient, task performance, full scale IQ and ages of wave model of the activation procedure (Binder et al., 1995) onset of probable brain injury and intractable seizures and then computes a data set-specific reference vector as the (see below). In the event that more than one variable was mean vector of this selected group of voxels. The correlation significantly correlated with LI, stepwise regression analysis test was then performed again for each voxel using this (P in 0.05, P out 0.10) was used to investigate the empirically derived reference vector, with a correlation incremental variance accounted for by predictor variables threshold of r The probability that any given voxel within each group. will exceed this threshold by chance is approximately P After subjects were classified into left, symmetric or right (with symmetric and right being considered atypical ) or P 0.05 for each set of 500 voxels. Across all dominance patterns, group differences in prevalence rates subjects, the mean number of supratentorial brain voxels were studied with χ 2 analyses (or Fisher s Exact Test in included in these analyses was 7899 (SD 916). Details situations with cell counts 5). These included comparisons of the analysis method were described previously (Binder of epilepsy versus normal subjects, men versus women in et al., 1996). the normal group, men versus women in the epilepsy group Activation volumes were determined in each subject by and right versus left seizure focus in the epilepsy group counting the significantly activated voxels in the lateral (using side of surgery as the grouping criterion). These latter two-thirds of each hemisphere. Due to the contralateral comparisons were done in an attempt to replicate previous connections of the cerebral hemispheres with the cerebellar studies that have found males at greater risk than females hemispheres, only supratentorial structures were included for for atypical language lateralization (Strauss et al., 1992a, b) analysis. A laterality index (LI) was calculated for each and a higher prevalence of atypical language patterns in subject as the ratio [V L V R ]/[V L V R ] 100, where V L epilepsy patients with left than with right temporal seizure and V R are activation volumes for the left and right focus (Risse et al., 1997). hemispheres. This approach yields LIs ranging between 100 Similarly, because early brain injury has been associated (strong left hemisphere dominance) and 100 (strong right with a greater incidence of atypical language dominance, we hemisphere dominance). LIs were subsequently classified as wished to investigate differences in prevalence rates for left hemisphere language dominant (defined as LI 20), dominance patterns related to this variable. However, the symmetric ( 20 LI 20) or right hemisphere dominant criteria used by various investigators to determine early (LI 20). These cut-off points were derived statistically brain injury have varied considerably, likely reflecting the by determining the mean number of activated voxels across availability of evidence and the investigators wishes to be all subjects (mean 286) and testing different hypothetical conservative or liberal in their inclusion. Some investigators partitions of this total between hemispheres against a null relied predominantly on radiological evidence, e.g. CT hypothesis of equal activation in each hemisphere (i.e. 143 evidence of brain pathology that presumably occurred early activated voxels on each side) using χ 2 analysis (Binder in life (Mateer and Dodrill, 1983), while others used a et al., 1995). It was determined that for asymmetries greater combination of clinical and medical historical information, than 173 : 113 the null hypothesis of symmetry was rejected e.g. a neurological event early in life associated with an at a probability P This partition of active voxels objective sign such as hemiparesis (Rasmussen and Milner, corresponds to an LI cut-off point of ). In the context of limited access to early medical A combination of non-parametric, correlational and records, onset of seizures is a commonly used marker for multiple regression techniques was used to analyse the age of brain injury, although it is recognized that a seizure continuous variable LI. Because the distributions of LI in disorder may develop well after the precipitating neurological normals and epilepsy subjects were not normal, a non- event (Salazar et al., 1985). Many investigators who used parametric statistical procedure (Mann Whitney U test) was seizure onset to define the age of brain injury have not clearly used to evaluate group differences in LI between normal and specified whether the defining event was the first seizure of epilepsy subjects, males and females, those with positive and any type or rather the onset of recurrent, medically intractable those with negative familial sinistrality, and epilepsy subjects seizures. Moreover, many investigators are unclear whether

6 2038 J. A. Springer et al. febrile seizures were included as a legitimate first sign of a seizure disorder or brain injury. Lastly, investigators have varied with regard to the age cut-off used to define early, with cut-offs ranging between 1 and 6 years of age. In the present study, two classification criteria were used to designate onset of brain injury, one representing a relatively liberal (or overly inclusive) criterion and the other a relatively conservative (or stringent inclusion) criterion. The aim of the first criterion was to identify the earliest age of brain injury by identifying the earliest potential sign of a neurological problem. This age, referred to as the age of probable brain injury, was identified by the age of the earliest known neurological event thought to precipitate epilepsy (e.g. birth trauma, head trauma, bout of meningitis or encephalitis) or, if none was identified, then the first seizure of any type (including febrile seizures). The aim of the second criterion was to provide a more stringent definition of early brain injury and emphasized unquestionable evidence of brain dysfunction. For this criterion, the age of onset of medically intractable seizures was used. Onset at age 5 years or less was defined as early onset for both probable brain injury and intractable seizures. Fig. 1 Frequency distributions of language LI in normal and epilepsy subjects. Results Task performance All subjects learned the tone decision task easily and performed well above chance levels (normal group mean 97% correct, SD 3.9; epilepsy group mean 91.3% correct, SD 10.4). Performances were also adequate on the semantic decision task in the normal group (mean 90.7% correct, SD 6.2) and the epilepsy group (mean 86.3% correct, SD 7.9). However, there was a significant difference between group means on both the tone decision task [t(148) 4.85, P 0.001] and the semantic decision task [t(148) 3.68, P 0.001], indicating that the epilepsy patients had more difficulty on both tasks compared with the normals. Fig. 2 Scatterplot of language LI as a function of age in the normal group, including the regression line, LI (age). dominance ( 97) to symmetric language distribution ( 5). Patterns of language dominance The LIs of the epilepsy group showed greater variability, Activation was observed in all subjects in a group of ranging from exclusive left hemisphere representation ( 100) supratentorial regions previously reported to be activated to strong right hemisphere dominance ( 81). during this task contrast, including prefrontal cortex of Using dominance classification criteria, the majority of the the superior, middle and inferior frontal gyri; mid-anterior subjects in the normal group (94%) had left hemisphere cingulate gyrus; posterior cingulate/retrosplenial cortex; dominance and the remainder had symmetric language anterior superior temporal sulcus; middle temporal and representation (6%). Although most subjects showed clear posterior inferior temporal gyri; mid-fusiform and anterior left dominance, no normal subject had exclusive activation parahippocampal gyri; anterior hippocampus; angular gyrus; of the left hemisphere and the average normal subject had anterior thalamus; and head of caudate (Binder et al., 1995, 19% of activated voxels in the right hemisphere. The majority 1996, 1997; Frost et al., 1999). The median LI was 66.1 for of the epilepsy patients (78%) also had left hemisphere the normal group and 60.5 for the epilepsy group, a difference dominance. However, a significantly larger proportion (22%) that was not statistically significant (Z u 1.42, P 0.05). of the epilepsy group had atypical language lateralization, As illustrated in Fig. 1, the LIs of the normal subjects varied including 16% with symmetric and 6% with right hemisphere continuously over a range from very strong left hemisphere dominance (χ , P 0.01; see Table 4).

7 Language dominance and fmri 2039 Table 4 Incidence (percentage) rates of language (r s 0.20, P 0.05). The age versus LI scatterplot is dominance in right-handed subjects from IAP studies and the present fmri study shown in Fig. 2. Language lateralization* Epilepsy group LIs of men (median 67.0) and women (median 57.5) Left Bilateral Right in the epilepsy group were not significantly different (Z u Present fmri study 0.49, P 0.05). χ 2 analysis using categorized LIs also did Normals (n 100) not show differences between the sexes (χ , P Epilepsy 0.05). Similarly, LIs did not differ between those with and All without familial sinistrality (Z u 1.57, P 0.05). Early LI was correlated with age of probable brain injury (r Brain injury 5(n 25) Intractable seizures 5(n 12) , P 0.01), age of onset of intractable seizures (r Late 0.50, P 0.001) and the EHI quotient (r 0.50, P Brain injury 5(n 25) ), but did not correlate with age, education, full scale Intractable seizures 5(n 38) IQ or task performance. In the stepwise regression analysis, IAP studies age, EHI quotient, full scale IQ, age of intractable seizures Branch et al., 1964 (n 48) and task performance were used as predictor variables for Kurthen et al., 1994 (n 142) LI. The first variable entered into the equation was age of Loring et al., 1990 (forced relative intractable seizures (r , P 0.001), followed by EHI dominance; n 91) No early injury ( 6 years; n 66) quotient, accounting for an additional 15% of variance in LI Mateer and Dodrill, 1983 (n 66) (r , P 0.001). Partial correlations of the other Rasmussen and Milner, 1977 variables were not significant. When age of probable brain Early damage (n 42) injury was used in the predictor variable list in place of age No early damage (n 140) of intractable seizures, the first variable entered into the Rausch and Walsh, 1984 (n 56) Rey et al., 1988 (n 29) equation was EHI (r , P 0.001), and was followed Risse et al., 1997 (n 304) by age of probable brain injury, accounting for an additional Strauss and Wada, 1983 (n 66) % of variance in LI (r , P 0.001). Woods et al., 1988 (n 118) Epilepsy patients were also categorized according to age Zatorre, 1989 (n 38) of brain injury in order to compare language lateralization *To facilitate comparisons across studies, some values were patterns of patients with and without an early age of brain recombined from original articles with emphasis given to injury or onset of intractable seizures. Twenty-five patients presenting data that separates subgroups of early and late brain had an age of probable brain injury 5 years, and 25 had injury, and classifications using figures based on relative probable brain injury after 5 years of age. There was a dominance when possible; values include subjects with early seizure onset or brain injury; see text for definition. significant difference between the LIs of the early (median 53.0) and late (median 69.0) onset groups (Z u 2.60, Factors associated with language lateralization P 0.01). Likewise, a χ 2 test using categorized LIs demonstrated that a significantly greater proportion of the Normal group group with early brain injury had atypical (symmetric or There was no significant difference between the LIs of men right dominant) language dominance (χ , P 0.02; (median 65.4) and women (median 67.4) in the normal see Table 4). Similar group differences were found when group (Z u 0.56, P 0.05). χ 2 analysis of sex differences patients with early (n 12, median 25.0) and late (n using categorized LIs also did not reach significance (χ 2 38, median 68.0) onset of intractable seizures were 0.55, P 0.05). LIs were not different between those with compared using either the continuous variable LI (Z u (median 63.0) and those without (median 66.1) familial 3.06, P 0.01) or categorized dominance patterns (χ 2 sinistrality (Z u 1.02, P 0.05). 7.21, P 0.01). Figure 3 illustrates how differences in the Simple correlation analysis found that education, EHI frequency of atypical language dominance relate to three laterality quotient and task performance data were not intervals of age of onset of intractable seizures: 0 5 years, correlated with LI. Only age correlated with LI (r 0.23, 6 15 years and 16 years or greater. P 0.05) such that older subjects showed less strong LIs in subgroups of epilepsy patients with left (median language lateralization. Because visual inspection of the 53.0) and right (median 64.2) hemisphere seizure focus scatterplot relating age to LI revealed two older age outliers were not significantly different (Z u 0.09, P 0.05). who may have been contributing disproportionately to this Likewise, the χ 2 test comparing left and right hemisphere correlation, a non-parametric rank order correlation was groups using categorized LIs was not significant (χ , computed. This post hoc test also showed a significant but P 0.05). Among patients with both early brain injury and relatively small negative correlation between age and LI a left seizure focus, 6 out of 12 (50%) had atypical language,

8 2040 J. A. Springer et al. Fig. 3 Frequency of atypical language dominance in normal subjects and in epilepsy patients with onset of intractable seizures (IS) after age 15 years, between age 6 and 15 or before age 6. control for non-specific neural processes. The degree to which the target and control tasks accomplish their intended purpose is critical to understanding the activation patterns. The linguistic and contrast tasks in the current study were selected to activate the neural networks involved specifically in speech perception and lexical-semantic processing. The activation patterns produced by these tasks have been studied in relatively large samples of neurologically normal humans and, in general, identify regions of the left hemisphere that have been linked to language processing through studies of aphasic patients with brain lesions (Binder et al., 1997). In addition, LIs computed from these tasks were strongly correlated with a similar index generated from IAP testing (Binder et al., 1996). Considering that the incidence of atypical language dominance in our epilepsy patients falls closely in line with previous IAP studies (see Table 4), these findings support the current fmri results for tasks which on average yield language lateralization results comparable with IAP. Still, the IAP is multifaceted, and it is likely that the fmri LI does not correlate equally well with all aspects of the IAP examination. Rare patients have been reported while among patients with both early brain injury and a right hemisphere focus, 3 out of 8 (37.5%) had atypical language. This difference did not reach statistical significance (P with interhemispheric dissociation of component linguistic 0.67, Fisher s Exact Test). processes, e.g. left hemisphere comprehension and right hemisphere vocal speech (Kurthen et al., 1994; Risse et al., 1997). If such cases exist, it is our suspicion that the fmri Discussion These observations are the first direct measurements of language lateralization in a large sample of neurologically normal, right-handed subjects. Language LIs for these subjects ranged along a continuum from left hemisphere dominance to relatively symmetric representation, and there LI used here will correlate more closely with the language comprehension aspects of the IAP than with the vocal speech components. was no obvious dichotomous or bimodal clustering of scores Right hemisphere language representation in (Fig. 1). Atypical patterns were uncommon, with only 6% normal and epilepsy subjects of normal subjects showing relatively symmetric activation The results suggest that right hemisphere language dominance and none showing clear right hemisphere dominance. is less common in normal subjects than can be inferred from Although right hemisphere dominance was not observed in IAP studies in right-handed subjects without early brain the normals, all normal subjects had some activation in the injury. Estimates from such studies of the incidence of right right hemisphere by the linguistic task used here. Atypical hemisphere language dominance fall in the 3 5% range (see dominance patterns were more frequent among epilepsy Table 4), with the findings of Rasmussen and Milner of 4% patients, comprising 22% of the total in this group. A weak being the most widely quoted (Rasmussen and Milner, 1977). relationship between age and LI was observed in the normal Similarly, using fmri in our own epilepsy sample and group, with older subjects showing more symmetric LIs. applying the most stringent exclusion criteria for late brain None of the other variables examined, including sex, degree injury (brain injury 5 years), an estimate of 4% is obtained. of right-handedness, familial sinistrality, education and level Still, fmri in 100 right-handed, neurologically normal of task performance, were related to language lateralization individuals failed to identify a single person with clear right in the normal sample. Early age of brain injury and weaker hemisphere dominance. In this regard our results are more right hand dominance were associated with an atypical LI in consistent with adult-onset lesion studies, which suggest that the epilepsy group. right hemisphere language dominance is present in 2% of normal dextrals (Zangwill, 1967; Gloning et al., 1969; Mastronardi et al., 1994). The most obvious explanation for Lateralization and task used in fmri the discrepancy between normal and epilepsy group estimates It is likely that LI, as computed in this study, would change is that efforts to exclude all the factors that influence language significantly if either the target or contrast task were changed. lateralization in the epileptic patients (e.g. eliminating patients Identification of a cognitive system using functional with language reorganization after early brain injury) are neuroimaging typically requires a comparison between incompletely successful and, thus, an estimate of crossed activation produced by a target task and one designed to dominance from this population is inherently biased.

9 Language dominance and fmri 2041 One practical interpretation of these findings is that disruption. Such a categorization method has the effect of language mapping may not be necessary in right-handed increasing the estimate of bilateral language representation patients prior to right hemisphere surgery when the illness and decreasing the estimate of left dominance, an effect prompting surgery is acquired after childhood. However, 6% identified in other studies as well (Loring et al., 1990; Risse of normal subjects and up to 13% of late onset epilepsy et al., 1997). Similarly, in the current fmri study, incidence patients had 40% or more of their language activation in the rates for atypical dominance are a function of the cut-off right hemisphere (i.e. had LIs 20), which suggests a points used for classifying LI. If language lateralization is significant potential risk for language deficits following truly a continuous variable with a unimodal distribution, such right hemisphere surgery. One critical issue is whether categorization is inevitably arbitrary and leads to loss of bihemispheric representation of language implies that either information. This information could also be critically valuable side can mediate language functions independently, or rather in clinical applications that depend on highly precise estimates that both hemispheres are necessary (Risse et al., 1997). The of risk. Using the LI from an fmri test, for example, it may previously mentioned correlation between fmri and IAP eventually be possible to predict not just the risk associated laterality indices (Binder et al., 1996) indicates a direct linear with being in the left dominant category, but the much relationship between the relative extent of right hemisphere more precise risk associated with each particular LI. language activation measured by fmri and the severity of language deficits induced by right hemisphere anaesthetization. This relationship suggests that right hemisphere language-related activation observed by fmri Factors related to atypical language may have predictive value in determining the risk of lateralization in normal subjects postoperative language deficits from right hemisphere surgery. Variability of language lateralization in normal subjects was Given these findings, the safety of fmri mapping not strongly related to sex, degree of right-handedness, recommends its general use prior to brain surgery in sensitive presence of left-handed family members, education or task cortical regions, particularly in individuals with brain disease performance. The range of some of these variables was acquired early in life. Ultimately, however, the cost of the relatively restricted, which may have lessened the ability to procedure will need to be weighed against the likelihood of detect significant relationships. For example, all subjects in finding significant right hemisphere language activation in a this study were right-handed and had EHI laterality quotients given individual. While this study provides a reasonable in the range 50 to 100. Had left-handed and ambidextrous estimate of this likelihood in normal, young, right-handed subjects been included, a significant relationship between LI adults, future studies will need to be conducted on samples and EHI scores might have emerged. Because the normal of older, left-handed and ambidextrous normal subjects, as subject sample was drawn mainly from local universities and well as on larger samples of patients with brain disease the medical college campus, levels of education and task acquired at different ages. performance also tended to be relatively homogeneous. Thus, The presence of right hemisphere language-related while no significant effects of these variables were found activation in nearly all subjects, the graded variation in over the ranges studied, future investigations using less language lateralization demonstrated in Fig. 1 and the linear homogeneous samples would be useful to conclusively rule relationship between LI and IAP language lateralization all out the possibility that these variables have a small but argue for the view that language lateralization is a measurable relationship to language lateralization. continuously variable phenomenon. Use of categorical The lack of sex effects on language laterality observed designations such as left dominant, right dominant and here is consistent with recent PET studies of language bilateral has been ubiquitous and no doubt of value in processing (Buckner et al., 1995; Price et al., 1996a). In some clinical settings, but the utility of this approach for contrast, Shaywitz and colleagues found, using fmri, that understanding the neurobiology of language organization is men had greater left lateralization of activation in the inferior far less certain. Several authors have proposed, for example, frontal gyrus during orthographic, phonological and semantic that much of the variability in frequency of atypical language tasks (Shaywitz et al., 1995; Pugh et al., 1996). dominance observed in different IAP studies (Table 4) is due These discrepancies could conceivably be due to any of simply to differences in categorization methods (Loring et al., several differences in methodology among the different 1990; Snyder et al., 1990; Kurthen et al., 1994; Risse studies, which included differences in task demands, image et al., 1997). For example, while most IAP studies show an acquisition techniques, data analysis methods and statistical incidence rate of atypical dominance (combined right and thresholds. Because the LI used in the present study is based bilateral language) in the 7 13% range, the data obtained by on activation over the hemisphere as a whole, the current Zatorre (Zatorre, 1989) imply an incidence rate of 37%. This analysis may not detect gender differences affecting more appears to be due to a categorization rule that identified a focal regions such as the inferior frontal gyrus. For this patient as having bilateral language representation when reason, we recently performed several additional voxel-wise language was disrupted in any way on both hemispheric and region-of-interest analyses of our normal subject data, injections, without regard for relative severity of the including an analysis using regions of interest identical to

10 2042 J. A. Springer et al. those used by Shaywitz and colleagues (Shaywitz et al., a relatively strong correlation (r 0.54; P 0.001) between 1995). These additional analyses showed no statistically language lateralization and degree of right-handedness on significant differences between women and men in language- the EHI, despite the fact that the normal and epilepsy groups related activation at a voxel level, and no sex differences in had very similar mean EHI scores and EHI score variance. lateralization of language-related activation for any region of This finding concurs with previous studies demonstrating interest (Frost et al., 1999). Future studies will be needed to a close relationship between left-handedness and atypical explain the differing results, but the preponderance of current language dominance in epilepsy patients (Rasmussen and evidence suggests that sex differences in large-scale language Milner, 1977; Mateer and Dodrill, 1983; Rausch and Walsh, organization, if present, are probably small in comparison 1984; Rey et al., 1988; Woods et al., 1988; Loring et al., to the degree of similarity of this brain system in men 1990; Strauss et al., 1990; Satz et al., 1994; Helmstaedter and women. et al., 1997; Risse et al., 1997) and suggests that pathological An unexpected observation was that LI in the normal processes in epilepsy exert a common set of influences on group was negatively correlated with age, indicating that the development of both manual and language lateralization. older subjects tended to have more symmetric activation. In contrast, normal subjects do not show this relationship, Although relatively small, this effect also proved significant presumably because handedness variation within this range in a subsequent rank-order correlation test. While unexpected, in the normal population is due to non-pathological (e.g. this finding is not surprising in that other investigators have genetic) causes that do not affect language lateralization. shown age-related changes in the patterns of functional The other significant predictors of LI in the epilepsy group activity in visual perception (Grady et al., 1994) and memory were age of onset of intractable seizures (r 0.50) and age (Grady et al., 1995). These changes in activation patterns of suspected brain injury (r 0.39). Together with degree were interpreted as adaptations that compensate for age- of right-handedness, the intractable seizures variable related changes in primary memory and perceptual systems. accounted for 40% and the brain injury variable 39% of the Because the age range of our normal subjects was relatively variance in LI. Consistent with several previous studies limited, the magnitude of age-related effects may be (Rasmussen and Milner, 1977; Mateer and Dodrill, 1983; underestimated. Future studies of normal subjects will need Strauss and Wada, 1983), we found that patients with evidence to include a wider age range to obtain a more reliable estimate of earlier brain injury tended to have more right hemisphere of this effect. language representation. The relationship between LI and age of onset of intractable seizures was slightly stronger, probably reflecting the fact that recurrent seizures represent Factors related to atypical language a more definitive marker of permanent brain dysfunction than might a febrile seizure or an early neurological event lateralization in epilepsy that could have only transient sequelae. Visual inspection of Language lateralization was not strongly related to sex, the scatterplots relating these variables to LI did not show education or task performance in the epilepsy patients, any obvious non-linearity in the relationship, nor was the corroborating similar negative findings in the normal group. effect any stronger when age of brain injury and intractable Task performance was somewhat more variable in the epilepsy seizures were treated as categorical variables (P 0.05) than group, suggesting that any relationship that may exist between as a continuous variable in a linear regression analysis (P this variable and LI is weak and would only be observable 0.001). These results imply that there may not be an absolute over a wide range of variation or in a very large subject critical period during which brain injury can influence sample. Age and full scale IQ also showed no significant language organization and after which no such influence is relationship to LI over the ranges studied. The lack of a possible. Instead, the roughly linear relationship between detectable age effect in the epilepsy group could be explained these variables suggests that later brain injury can still affect by the fact that there were fewer subjects, allowing other language lateralization, although less profoundly. more potent variables, namely handedness and brain injury, Given prior findings that reorganization of language is to mask this weak relationship. In particular, there was a most marked after early left hemisphere injury (Strauss et al., weak positive correlation between age and age of onset of 1990; Satz et al., 1994), it was somewhat surprising that intractable seizures (r 0.27, P 0.06), indicating that there was no significant relationship between the side of patients with onset of intractable seizures at a younger age seizure focus and LI, especially since a conservative criterion tended to present to our programme at a younger age. This was used to define the seizure focus (i.e. side of surgical selection bias would have produced an effect on LI that was resection). There are several possible explanations for this opposite in direction to the age effect observed in normal finding. First, the sample size may have been too small to subjects. Because the effect on LI of age of intractable detect differences due to seizure side. The difference in seizure onset was much stronger than the age effect observed median LI between left and right groups, although not in normal subjects, this latter relationship may have been unobservable in the epilepsy group. In contrast to the normal group, the epilepsy group showed significant, did show a slight trend in the expected direction, i.e. the left seizure focus subgroup had slightly weaker left language dominance than the right seizure subgroup. A