OUP user menu

Disrupted temporal lobe connections in semantic dementia

C. J. Mummery, K. Patterson, R. J. S. Wise, R. Vandenbergh, C. J. Price, J. R. Hodges
DOI: http://dx.doi.org/10.1093/brain/122.1.61 61-73 First published online: 1 January 1999

This article has a correction. Please see:


Semantic dementia refers to the variant of frontotemporal dementia in which there is progressive semantic deterioration and anomia in the face of relative preservation of other language and cognitive functions. Structural imaging and SPECT studies of such patients have suggested that the site of damage, and by inference the region critical to semantic processing, is the anterolateral temporal lobe, especially on the left. Recent functional imaging studies of normal participants have revealed a network of areas involved in semantic tasks. The present study used PET to examine the consequences of focal damage to the anterolateral temporal cortex for the operation of this semantic network. We measured PET activation associated with a semantic decision task relative to a visual decision task in four patients with semantic dementia compared with six age-matched normal controls. Normals activated a network of regions consistent with previous studies. The patients activated some areas consistently with the normals, including some regions of significant atrophy, but showed substantially reduced activity particularly in the left posterior inferior temporal gyrus (iTG) (Brodmann area 37/19). Voxel-based morphometry, used to identify the regions of structural deficit, revealed significant anterolateral temporal atrophy (especially on the left), but no significant structural damage to the posterior inferior temporal lobe. Other evidence suggests that the left posterior iTG is critically involved in lexical–phonological retrieval: the lack of activation here is consistent with the observation that these patients are all anomic. We conclude that changes in activity in regions distant from the patients' structural damage support the argument that their prominent anomia is due to disrupted temporal lobe connections.

  • semantic dementia
  • semantic processing
  • anomia
  • PET
  • temporal lobe
  • iFG = inferior frontal gyrus
  • iTG = inferior temporal gyrus
  • mTG = middle temporal gyrus
  • PPT = Pyramids and Palm Trees
  • sOG = superior occipital gyrus
  • T-O-P = temporo-occipito-parietal


The selective impairment of semantic memory first came to prominence with Warrington's (1975) study of three patients with progressive anomia and impaired word comprehension, although a similar pattern had in fact been described in Japan several decades earlier under the label of `gogi' (word-meaning) aphasia (Sasanuma and Monoi, 1975). The term `semantic dementia' subsequently was introduced to convey the pattern of profound semantic deterioration which disrupts factual knowledge and object recognition/comprehension as well as semantic aspects of language (Snowden et al., 1989; Hodges et al., 1992). In this degenerative disorder, patients present with a progressive loss of expressive and receptive vocabulary; they typically complain of difficulty in `remembering' the names of people, places and things. The language impairment appears strikingly restricted to lexicosemantic processing: at least until late in the course of the disease, syntactic and phonological processes are largely uncompromised (Breedin et al., 1994; Hodges et al., 1994). Furthermore, in contrast to more common and more global dementing conditions, particularly Alzheimer's disease, patients with semantic dementia have well-preserved episodic memory (at least for recent events: Graham and Hodges, 1997; Hodges and Graham, 1998), and achieve average or even superior scores on tests of visuospatial skills, frontal executive functions and problem solving which do not require comprehension of specific semantic concepts.

Semantic dementia is a variant of frontotemporal dementia in which the brunt of pathology falls upon the temporal lobes (Snowden et al., 1996; Neary, 1997). Structural brain imaging (MRI) in patients with semantic dementia demonstrates focal temporal lobe atrophy, typically bilateral, though sometimes markedly asymmetrical. In a few cases, abnormalities have apparently been confined to the left hemisphere (Tyrrell et al., 1990; Patterson et al., 1994a). The regions of most prominent atrophy are in anterolateral temporal cortex, especially the temporal pole and inferior and middle temporal gyri (Hodges and Patterson, 1996). The syndrome thus presents with cognitive deficits restricted to a specific domain and with reasonably restricted neuroanatomical abnormality on structural imaging. As is well known, however, the location and extent of structural abnormality may be a misleading guide to the brain systems specifically responsible for a cognitive impairment: structurally undamaged areas may fail to operate normally if they receive inadequate activation from regions to which they are connected anatomically.

Functional activation studies of patients with this disorder seem valuable in our quest to understand the neural basis of both normal and disrupted semantic memory. Experiments with normal subjects have documented a reasonably consistent network of brain regions active during semantic processing (Wise et al., 1991; Demonet et al., 1992; Mummery et al., 1996; Pugh et al., 1996; Vandenberghe et al., 1996; Price et al., 1997), despite employing several different semantic tasks (e.g. semantic categorization, associative semantic judgements, semantic category fluency) and stimuli presented in several different modalities (pictures of objects, spoken words, written words). This network involves predominantly left-sided areas, including the left anterior temporal region that is atrophied in semantic dementia.

By comparing activation patterns for patients with semantic dementia with those of normal subjects in a semantic activation study, we can ask at least two important questions. (i) Do the behavioural deficits in semantic dementia result directly, and solely, from malfunction of atrophied anterior temporal regions, or are they also attributable to underactivation of relatively intact, but disconnected brain structures? The evidence most germane to this question, from functional imaging studies, is currently extremely meagre. Cardebat et al. (1996) using SPECT and Patterson et al. (1994a) resting PET both reported significant hypometabolism in the left inferolateral temporal lobe, which is scarcely surprising. To our knowledge, there have as yet been no activation studies performed on patients with semantic dementia. (ii) Might there be areas of significant activation in patients that are not observed when normal subjects perform the same task? A positive answer here would suggest that such areas might be recruited in partial compensation for damage to the normal semantic network. For example, if homologous right-hemisphere regions (which do not seem, from existing PET evidence, to be essential for semantic tasks) can contribute to semantic processing when left-sided structures are compromised, we might expect increased activation for patients in these regions, and decreased activation relative to normals in the left hemisphere. Alternatively, if the patients' residual semantic abilities are based on partial functioning of the standard network, then activation for both patients and normals should be seen primarily in the left hemisphere.

We used PET to examine changes caused by focal disease to the operation of the network of regions activated by a semantic task. Subjects diagnosed as patients with semantic dementia on the basis of a battery of semantic tests designed by Hodges and colleagues (Hodges et al., 1992; Hodges and Patterson, 1995) were scanned while performing both a semantic task and a non-semantic control task using similar stimulus materials. The differences in regional activation patterns for the two tasks in the patients were compared with those in six age- and education-matched normal control subjects. The semantic task, a modified version of the Pyramids and Palm Trees (PPT) test (Howard and Patterson, 1992), was selected for two reasons. First, results of a previous PET study by Vandenberghe et al. (1996) employing this same task allowed us to predict the loci of activation in normal participants. These included inferior frontal regions, anterior and posterior temporal regions, and temporo-occipito-parietal (T-O-P) junctions, all mainly left lateralized. Secondly, all of the patients had been tested on the standard PPT test and were able to cope with its requirements, though with performance accuracy outside the normal range. Analyses first established regions of consistent activation in the six control subjects. Within this delineated network, we examined areas where the normals and patients showed activation in common, and areas where there were consistent differences between the two groups, thus allowing assessment of regions with normal and abnormal function.


Patient details

Six patients were selected for the study, all of whom had been diagnosed as having semantic dementia based on the criteria used by Hodges et al. (1992). To be suitable for the study, each patient had to be able to perform the PPT test (Howard and Patterson, 1992) on which the semantic PET task was modelled. The PPT test is an assessment of semantic associative knowledge which can be administered with either pictures of the concepts or their names (either spoken or written); the PET version used here is described in greater detail below, under `Psychological tasks'. Given the semantic deficit which is the hallmark of semantic dementia, it is not surprising that—with the exception of one patient (D.M., who is at an early/mild stage of progression)—all of the patients scored below normal limits when they were tested on the published version of the PPT, with pictures or written words or both. Nonetheless, in standard hospital or home testing, each of the six patients was able to follow the instructions of the task and to cope with its requirements, as indicated by an above-chance score. Under these usual test conditions, there is no time pressure; however, for the PET version, in order to standardize experimental conditions across subjects, it was necessary to impose a time deadline for each trial. Either because of this additional time pressure, or because the general atmosphere of being tested in a PET scanner is more anxiety-provoking than the usual over-the-desk condition, two of the six patients did not manage to score above chance on the associative task performed during scanning. These two subjects were therefore removed from analysis, and will not be considered further here. Of the remaining four patients, three were female, one male; age range 58–60 years (mean 58.75 years). Table 1 provides a summary of their performance on a range of neuropsychological tests.

Control group

The control group consisted of six normal, right-handed volunteers, age and education matched to the patient group. All control subjects (three males, three females; age range 52–64 years, mean 57 years) were fit, healthy, on no medication and free from any history of neurological or psychiatric illness. They were all strongly right-handed on the Edinburgh Handedness Inventory. The study was approved by the local hospital ethics committee and the Administration of Radioactive Substances Advisory Committee (UK) (ARSAC).

Data acquisition

Each subject underwent 12 PET scans of the distribution of brain activity over a 2 h period. Scans were obtained using an ECAT EXACT HR+ PET Scanner (CTI, Knoxville, Tenn., USA) with collimating septa retracted. Volunteers received a 20 s intravenous bolus of H215O at a concentration of 55 Mbq/ml and a flow rate of 10 ml/min through a forearm cannula for each scan.

Psychological tasks

Each subject was scanned three times in each of four conditions composed of two modalities (words or pictures) crossed with two tasks (semantic and visual judgement). Each trial consisted of a triad of words or pictures arranged in triangular format (as in the PPT test and the PET version designed by Vandenberghe et al., 1996), with the reference stimulus at the top of the triangle and the two response choices side-by-side below (Fig. 1). In the semantic task, on each trial, subjects pressed a left or right key-press button to indicate the response choice that was more closely semantically associated with the reference item; e.g. COW: horse, bear; SWITCH: light bulb, candle; COAT: glove, sock; CUCUMBER: tomato, corn (note that in these examples, the correct response is the one on the left; in the test, the correct response occurred in left/right position on half of the trials in random order). The visual control task consisted of a matching to sample task for physical size on the screen. Each trial contained the identical concept (word or picture) in all positions of the triangle; the correct response choice was the one closest in size to the reference stimulus. The correct response differed by 15% in size from the reference item, and the distractor was a further 15% different in size. Once again, subjects pressed a left or right key-press button to indicate their choices.

Each subject saw a particular concept triad only once, in either word or picture modality, and mode of presentation was counterbalanced across subjects such that each triad occurred equally often in the two modalities. Each run consisted of 12 within-condition trials, with a new triadic stimulus presented every 6 s. Note that this is a slower task pace than the 4.5 s inter-stimulus interval used in Vandenberghe et al. (1996). Presentation of tasks was also counterbalanced. Subjects were practised on each of the conditions prior to scanning.

Data analysis

The data were analysed with statistical parametric mapping (using SPM96 software from the Wellcome Department of Cognitive Neurology, London, UK; http//www.fil.ion.ucl.ac.uk/spm) implemented in Matlab (Mathworks Inc., Sherborn, Mass., USA). Scans from each subject were realigned using the first as a reference. A T1-weighted MRI was co-registered to the mean PET image for each subject and then transformed stereotaxically to a standard MRI template in the Talairach and Tournoux space (1988), and the same transformation matrix subsequently applied to the PET images (Friston et al., 1995b). As a final preprocessing step, images were smoothed with a Gaussian filter of 16 mm. The search volume went from z = –48 mm to z = +60 mm, with a final image resolution (full-width half-maximum) x = 10.7 mm, y = 11.0 mm, z = 11.4 mm. The condition and subject effects were estimated according to the general linear model at each voxel (Friston et al., 1995a). To test hypotheses about regionally specific condition effects, the estimates were compared using linear compounds or contrasts. The resulting set of voxel values for each contrast constitutes an SPM of the t statistic (SPM{t}). The SPM{t} values were transformed to the unit normal distribution (SPM{Z}) and thresholded at P = 0.001 uncorrected for multiple comparisons.

Data were analysed using a randomized block design with global brain activity as a (subject-specific) covariate of no interest. Analysis was performed using a multistudy design, allowing assessment of the contribution of all individual subjects to the group effects. This approach distinguishes areas of consistent activation across subjects from areas yielding significant intersubject variability (see Price et al., 1998). To identify group effects, we summed over the effects from the six normal subjects or the four patients. To identify differences between normal and patient effects, we report and interpret only those areas where each patient showed a significant difference from the normal group, i.e. the activation difference was consistent within each group (either patients or normals). This was achieved using a conjunction analysis on the four interaction terms (see Price and Friston, 1997a). Contrasts were used firstly to identify areas in the normals that were significantly more active (i) for the semantic task than the visual task; (ii) for semantic activation for words than for pictures, and vice versa (i.e. the interaction between stimulus type and task). Areas were then identified that were activated consistently across the two groups (normals and patients) for the above contrasts. Finally, areas that were significantly more active in normals than in all patients, or vice versa, were identified as an interaction of task and group. For the purposes of this study, we do not report the control activations, as we are interested specifically in the contrast of semantic versus visual control task.

Usually, with normal subjects, the grey matter threshold with PET images is set at 80% of whole brain activity to exclude voxels from analysis in extracranial and low flow white matter regions. However, as these patients had significant focal atrophy, we lowered the grey matter threshold to 50% to test for activation within and around atrophied regions. Altering the grey matter threshold does not alter uncorrected P values, but because there are more voxels included in an analysis, inference becomes more conservative when the correction for number of comparisons is applied.

We expected normal control subjects to show activity in a network of regions similar to that previously reported for this task (Vandenberghe et al., 1996): the left inferior frontal gyrus (iFG) [Brodmann area (BA) 45, 11, 47]; left inferolateral temporal region, involving areas BA 21, BA 20, BA 21/37; left T-O-P junction (BA 19/39); left superior occipital gyrus (sOG) (BA 19); left hippocampus (BA 34) and right cerebellum. We used these and homologous regions in the right hemisphere as regions of interest, interpreting activity significant at an uncorrected level of P < 0.001 (Z > 3.09). For areas outside these regions, we discuss areas surviving a corrected threshold of P < 0.05 (Z > 4.3), though, in the interests of communication, tables include all areas significant at P < 0.001 (uncorrected).

Structural analysis

For each subject, a T1-weighted high-resolution MRI scan was obtained, realigned, normalized into standard Talairach and Tourneaux space (1988), and segmented into grey matter, white matter, CSF and scalp. The images were then smoothed with a filter of 12 mm. After smoothing, grey matter density at each voxel for each patient was compared with the mean density for the group of six control subjects. The technique used (voxel-based morphometry) was implemented with Statistical Parametric Mapping (SPM96) as previously described by Wright et al. (1995). Figure 2 shows areas surviving a threshold of P < 0.05, corrected.



The most consistent and significant neuropsychological deficit shown by these patients with semantic dementia was in naming (see Table 1). It has been shown previously that the anomia in semantic dementia can be explained on the basis of progressive dissolution of the lexicosemantic system (Patterson et al., 1994b). The patients were also deficient on semantic tasks such as PPT and word–picture matching, though the group was heterogeneous in terms of the degree of loss of semantic knowledge. For example, D.M., who is at a relatively early stage of semantic dementia, appears normal on these relatively easy semantic tasks but is below normal limits on more stringent semantic assessments (Hodges and Graham, 1998). This variability between individual patients underlines the importance of constraining interpretation of patient activations to those that were common to all individuals in the group. In contrast to the significant deficit on tests of naming and semantic knowledge, the patients were, as predicted, relatively well preserved on tests of cognitive function outside the semantic domain.

For the semantic task conducted during the PET study, the patients were reliably less accurate than the controls in performance for both modalities (words P = 0.006; pictures P = 0.03), although each of the four patients achieved a score above the chance level of 50% (see performance in Table 2). The two subject groups responded with remarkably similar reaction times, with the one exception that patients were faster than controls in semantic judgements on pictures (P = 0.004). There may have been a hint of speed/accuracy trade-off between the two modalities in the semantic condition for patients, although their (slight) accuracy advantage for words > pictures did not approach statistical significance. Within the semantic task, the two groups showed differing reaction time patterns across modality: the controls made faster semantic decisions for words than for pictures (P = 0.04), whereas the patients were quicker to respond to pictures than to words (P = 0.04). Although normals showed a consistent tendency to be faster with words, none of the individual effects were significant. The patients' response times were heterogeneous: one was significantly faster on semantic decisions for pictures than for words; three showed no significant differences (one showing a reverse trend, i.e. words faster than pictures). There were no consistent effects of modality on accuracy in either direction in either group. Since this study is concerned with common differences between patients and normals, we do not place much emphasis on the interaction between subject group and stimulus modality on response times.

The visual control task was easier than the semantic task, as reflected in both significantly faster reaction times (P < 0.0001 for both groups) and higher accuracy on the size judgements than on the associative semantic judgements (control group P = 0.05; patient group P < 0.0001). The visual tasks showed no significant differences in reaction time or accuracy either within a group or between the two groups.

Structural analysis

Voxel-based structural morphometry was used to obtain a measure of neuronal loss in patients. All four patients showed marked atrophy of the left temporal pole and left anterolateral temporal region. In one case (D.M.), this measured structural abnormality was unilateral (left); in the other three cases, it also affected homologous areas on the right, but much less severely (Fig. 2). Two other important points emerge from this analysis. First, there were no significant structural differences observed in regions showing the most marked activation differences (see below). Secondly, none of the four patients had significant atrophy of the hippocampal region, which underlines the substantial neuroanatomical (and resulting behavioural) difference between semantic dementia and dementia of Alzheimer's type. The principal result of this analysis is statistically significant atrophy in regions consistent with previous structural studies of this syndrome (e.g. Hodges and Patterson, 1996).

PET results

Control activations

As noted above, we expected semantic judgements for both words and pictures to activate a network of regions for the control group similar to that seen with previous use of this task in normals (Vandenberghe et al., 1996), despite somewhat different stimulus parameters. This expectation was fulfilled (see Table 3). Regions activated in the main effect of semantic tasks versus visual size discrimination tasks (for both words and pictures) included the left iFG (both BA 47 and 45), left posterior iTG (BA 37/19, bilateral T-O-P junction (BA 19/39), left sOG (BA 19) and anterior cingulate cortex. The left posterior middle temporal gyrus (mTG) (BA 21/37) was more active with words than pictures in the semantic condition, in common with previous findings (Vandenberghe et al., 1996) (Table 3). In this study, further areas produced significantly higher activation with semantic judgements for words than for pictures: the left iTG (BA 20), left temporal pole (BA 38) and left iFG (BA 47), which previously were reported as common to both modalities (although with higher Z-scores for words than for pictures). Comparison of the size discrimination task in the two modalities suggests that this differential activation for the two semantic conditions may have resulted in part from activation of the semantic network by the control (size discrimination) task for pictures. Pictures may have provoked automatic semantic processing even when subjects were only asked to judge their relative size. This could in part be a consequence of the longer exposure duration (6 s) used. No areas were found to be significantly more active for semantic judgements for pictures than for words.

Patient activations

In common with the control group.

The contrast between semantic and visual control tasks across modalities yielded significant common activation across the two subject groups in left inferior frontal regions (BA 47 and 45), the left posterior mTG (BA 21/37), left T-O-P junction (BA 19/39), left sOG (BA 19), anterior cingulate and right cerebellum (Fig. 3, Table 3). Four regions were activated by both subject groups, but showed differential activation dependent on stimulus modality: the left iFG (BA 47), left iTG (BA 20), left temporal pole and posterior cingulate were more active for semantic judgements on pictures relative to words in the patient group, whereas the same region was the site of a modality effect in the opposite direction in the normal group (words > pictures). While this differential activation was relatively consistent for all normals, the patient group showed considerable variability in activation of the temporal regions. For example, two patients activated the iTG (BA 20) more for semantic judgements on words, but the other two relatively more for semantic judgements on pictures. As with the interaction between modality and group on the behavioural measure of response times, we are reluctant to offer any strong interpretation of the interaction between modality and task on regions of significant change in blood flow, due to inter-patient variability.

Reduced for patients relative to normals.

The left posterior iTG (BA 37/19) and right T-O-P junction (BA 39) were significantly less active in each patient relative to normal subjects regardless of stimulus modality (Fig. 4). Neither of these regions was the site of significant structural damage (Fig. 2). Because the patients were reliably less accurate than normal subjects on the semantic processing task, it is important to try to assess the possible relevance of this difference in performance to the activation abnormalities observed for the patients. Two additional analyses were therefore performed. First, a correlational analysis examined variation of PET activation with performance on the semantic task, and found no significant correlation between success of semantic judgements and level of activity in the left posterior iTG (BA 37/19) and right T-O-P junction (BA 39). In addition, there was no significant correlation seen between performance and activity in the left anterior temporal lobe. A second analysis used only performance-matched scans for the two groups. The accuracy range for control subjects on the semantic judgement task (across the 36 separate scans: six subjects × six semantic scans each) was 56–100% (mean 84.8%). We therefore discounted all patient scans (five of 24) on the same task that fell outside this range, giving an accuracy range of 60–100% (mean 78.6%) for the patients. Analysis using these scans showed the same pattern of activation differences between normal participants and patients as the initial analysis, notably reduced activity in the left posterior inferior temporal lobe (and enhanced activity in the anterior temporal lobes). These results demonstrate that significant alteration in regional activation for the patients relative to normals did not result simply from the patients' less competent perfomance on the semantic task.

Enhanced for patients relative to normals.

The semantic dementia patients in turn activated certain areas consistently more than normal subjects during semantic tasks. Specifically, the left premotor region (BA 44/6), left anterior superior temporal gyrus and right anterior temporal lobe all showed enhanced activity for both words and pictures in patients relative to normals (Fig. 4).


We have shown differences between a normal group and four patients with a focal semantic deficit in the network of areas activated by a semantic task. The control group showed activations consistent with those reported by Vandenberghe et al. (1996), thus successfully replicating—for an older group of participants and with different presentation parameters—some previous findings on the network of brain regions critically involved in this kind of semantic processing. A considerable part of this same network was activated by semantic judgements in the patient group as well, specifically the iFG (BA 44,45,47), left mTG (BA 21/37), left T-O-P junction (BA 19/39), left sOG (BA 19), anterior cingulate cortex and right cerebellum. Taken in conjunction with the behavioural data, this finding suggests that, despite the significant (and expected) difference in performance accuracy between patients and normals, the two groups were performing the tasks in a largely similar manner. It was of considerable importance, however, that there were also significant differences in regional activation between the two subject groups in the contrast between semantic and visual control tasks, and we will now discuss these differences with reference to the appropriate lesion data and previous functional imaging findings. Changes in the temporal/parietal lobes and frontal lobes will be dealt with separately.

Temporal/parietal lobes

Posterior regions—no lesion, functional deficit

Discrete areas activated in the controls— the left posterior iTG (37/19) and right T-O-P junction (19/39)—were consistently absent in all patients across both modalities of stimulus input. The left posterior iTG, first referred to by Nielson as a language formulation area (1946), and since then as the basal temporal language area (Burnstine et al., 1990), has direct connections to Wernicke's area (Di Virgilio and Clarke, 1997). It has been shown that stimulation of this region may cause specific naming deficits (Luders et al., 1991); more recently, damage in the posterior inferior portion of BA 37 has been linked to anomia without major associated semantic deficit, and a role proposed for this region in `allowing the semantic system access to stored lexical information' (Foundas et al., 1998).

Supporting evidence for this hypothesis comes from previous functional imaging studies, which have shown activation in the left BA 37 using several different language tasks, including naming (of objects and words: Bookheimer et al., 1995; of letter/colours/objects: Price and Friston, 1997a) and semantic word generation to auditory cues (Wise et al., 1991; Warburton et al., 1996). The posterior iTG also appears to be activated even when no explicit naming of a viewed word/object is required (Price et al., 1996). These studies suggest a role for the left basal posterior temporal area (BA 37/19) in retrieval of the name of a concept (Price and Friston, 1997a), independent of modality of the stimulus prompting retrieval, or whether naming is implicit or explicit. This putative role is consistent with our finding that the posterior iTG was consistently not activated in these four patients who had notable deficits in naming (Table 1). The fact that the patients had a variable degree of semantic associative deficit is not incompatible with the above conclusion. As previously noted, other regions are activated commonly by semantic tasks, crucially including the anterior temporal lobe. We are not claiming that the lack of activation of the posterior inferior part of BA 37 is responsible for the patients' core semantic deficit, merely that this physiological deficit is consistent with the patients' most severe and consistent behavioural abnormality, i.e. anomia.

The patients showed no visible posterior temporal atrophy on MRI, and the morphometric analysis performed revealed no differences in the posterior iTG (BA 37/19) compared with the six age-matched normals. Recent work has shown in at least one case of progressive fluent aphasia/semantic dementia (Harasty et al., 1996) that there can be pathological (post-mortem) evidence of disease in BA 37; the present study, however, provided no structural counterpart in these cases. Although we cannot exclude the possibility that the lack of activation is due to pathological involvement at the microscopic level, a more plausible explanation seems to be that this region is structurally intact but failing to function normally because of reduced input from the anterior temporal lobe. This hypothesis predicts relatively normal activation of BA 37/19 in tasks that do not require the anterior temporal lobe. Further imaging studies are therefore required to adjudicate between alternative explanations of this striking result.

The posterior mTG was activated in both patients and controls. This area is contiguous with the basal temporal lobe though lies more anterior and dorsal, including part of BA 37 and extending into BA 21. The posterior mTG is known to be activated by semantic tasks (e.g. Martin et al., 1995; Mummery et al., 1996; Vandenberghe et al., 1996), and is commonly affected in patients with aphasia plus a semantic deficit (Cappa et al., 1981; Chertkow et al., 1997). Both lesion and functional imaging data are accumulating to suggest that the region designated BA 37 is functionally `divided', though its differing roles are intimately connected. Both parts seem involved in word retrieval, but the mTG may be more concerned with some aspect of semantic knowledge, whereas the more inferoposterior temporal region may be more critical for lexical–phonological retrieval.

The other posterior temporal/parietal region to show decreased activation in patients relative to controls was the right T-O-P junction. This was activated significantly more for both modalities in normals relative to patients, whereas the left T-O-P junction activated consistently in both groups. Lesion studies suggest that damage to the left T-O-P junction also causes comprehension deficits (Dejerine, 1892; Hart and Gordon, 1990), and this left area has been activated in previous studies of semantic processing (e.g. Wise et al., 1991; Martin et al., 1995; Vandenberghe et al., 1996). The preservation of activation of the left T-O-P junction suggests that it may be involved in maintaining the patients' residual ability to perform the semantic task; the decrease in the right T-O-P junction could reflect a number of task-related differences between the two groups, including performance and degree of attention. For example, it has been shown that activity in this region induced by reading words is modulated by stimulus duration (Price and Friston, 1997b).

Anterior regions—structural deficit, continued activation

These patients have a (variable degree of) semantic deficit as well as anomia (Table 1). Previous imaging and lesion data concur that the anterior temporal lobes are involved in semantic processing (e.g. Hodges et al., 1992; Damasio et al., 1996; Vandenberghe et al., 1996). Structural analysis in our patients revealed significant anterior temporal lobe atrophy in all cases (left more than right) consistent with previous findings (Hodges, 1994). However, some parts of the anterior temporal lobes were activated in the patients, despite focal atrophy. In particular, both the left anterior superior temporal gyrus and right anterior temporal lobe (BA 38) actually showed enhanced relative activity in patients compared with controls for both task modalities. It must be emphasized that these effects are due to relative increases in blood flow in the semantic task as compared with the visual task: regional cerebral blood flow in the anterior temporal lobes (regions of significant structural change) was unsurprisingly lower in patients than in normals. However, examining the relative difference in blood flow between the semantic and control tasks, this relative change was greater for patients than normals. Patients therefore showed peri-damage activation close to significant structural changes. In addition, the right-sided activation suggests that the patients may be relying on these homologous right hemisphere structures more than normals do for semantic processing, although the patients' significant semantic deficit clearly demonstrates that right temporal structures are not sufficient for adequate semantic performance.

Summary of temporal lobe findings

Most of the currently available evidence suggests that, while some posterior cerebral regions are involved in semantic processing, the posterior iTG is more critically involved in lexical–phonological retrieval rather than semantic retrieval (Raymer et al., 1997; Foundas et al., 1998). Extrapolating from these results to our data, it is feasible that in normals the posterior inferior temporal region (BA 37/19) is involved in retrieving the name of a concept, and in semantic dementia this function is significantly compromised due to disrupted interactions between anterior and posterior temporal structures. Although no explicit naming is required in the PPT semantic task, implicit activation of regions involved in naming may occur for many semantic tasks (see Price et al., 1997).

Frontal regions

A large swathe of activity in the left frontal lobe was found in both groups for both modalities in the semantic task, extending from the iFG (BA47) through BA 44/45 to premotor and motor cortex. Although the role of the inferior frontal gyrus (BA 44/45/47) in semantic processing is under considerable debate, we suggest [along with Fiez (1997) and Thompson-Schill et al. (1997)] that these frontal areas are more likely to be involved in the strategic control or selection of semantic material than with semantic processing proper. The two main reasons for this, in our view, are the complementary facts (i) that there is no evidence for semantic deficits in patients with damage restricted to frontal regions and (ii) that patients with substantial semantic impairment, like M.J. and J.H. from the present study (see Table 1), often have no damage to frontal structures (see Fig. 2).

Within the large left frontal region activated for the semantic task in both control and patient groups (Fig. 3), several sub-regions were in fact significantly more active for patients than normals (Fig. 4B). If, as suggested above, these frontal structures are important for strategic control or selection in semantic judgement tasks, then it is possible that the patients' semantic impairment enhances the requirement for frontal involvement.


In summary, this study has shown activation of a network of regions for a semantic associative task in a group of patients with a focal cognitive deficit affecting naming and semantic knowledge. A number of regions in the network were activated alike for normal controls and patients (explaining why the patients are able to perform the task to some degree); however, consistent differences between subject groups were also found. The behavioural deficits in patients with semantic dementia apparently result both from malfunction of atrophied anterior temporal regions and underactivation of additional temporal regions, specifically left BA 37/19. The lack of activation of the posterior inferior temporal lobe is consistent with the observation that such patients all have anomia. We conclude that changes in activity in regions distant from the patients' structural damage support the argument that their prominent anomia is due to disrupted temporal lobe connections. While further research is required to establish a definitive account of this result, it provides a powerful argument for the importance of functional imaging studies in neuropsychology.

View this table:

Summary of neuropsychological test results

D.M. July '96M.J. April '97J.H. Sept. '96G.C.B. July '96
Summary of the patients' psychological profile for both semantic and general tasks prior to scanning. PPT = Pyramids and Palm Trees; GNT = Graded Naming Test; TROG = Test for the Reception of Grammar; VOSP = Visual Object Space Perception battery; NT = not tested; F = forwards,B = backwards. *Tests are from the Hodges and Patterson semantic battery.
Semantic tests
Naming*/4840 6 627
GNT/3012 0 5 0
Word–picture matching*/4848332637
category living(4 categs)41 5 535
category man-made(4 categs)37 1 617
letter (FAS)61151228
General tests
Rey copy/3633313136
VOSP cube analysis/10 9101010
Digit span 8F, 7B 6F, 4B 7F, 4B 6F, 6B
View this table:

Behavioural data for the patient and control groups

Mean RTSEAccuracyMean RTSEAccuracy
(mean %)(mean %)
Mean reaction times (RT), standard error (SE) and accuracy (as mean % correct for each condition) for controls and patients during the scanning tasks. Vis words = visual judgements on words; Vis pict = visual judgements on pictures; Sem words = semantic judgements on words; Sem pict = semantic judgements on pictures.
Vis words2453(71)93.52563(94)95.8
Vis pict2446(79)91.22345(65)97.2
Sem words3376(109)90.23621(159)71.8
Sem pict3691(109)85.23224(113)68.8
View this table:

Regions activated in both patients and controls, comparing semantic task with visual task

Main effectSimple main effects
Both groupsNormalsPatients N(W–V)N(P–V)P(W–V)P(P–V)
Regions of significant activation in the main effect of semantic minus visual tasks activated by both patients and normals: (A) regions activated for both words and pictures; (B) regions activated significantly more for words than pictures; (C) regions showing differential activation across modalities, i.e. greater activation for patients (relative to controls) in the semantic task on pictures, and greater activation for controls (relative to patients) in the semantic task on words. Coordinates (in millimetres) refer to the location in the stereotaxic space of Talairach and Tournoux of the voxels with the peak Z score within a particular activated region [Z > 3.1 (P < 0.001)]. N(W–V) = semantic–visual for words in normals; N(P–V) = semantic–visual for pictures in normals; P(W–V) = semantic–visual for words in patients; P(P–V) = semantic–visual for pictures in patients. *Activation differences: larger Z score for interaction indicates activation for words and deactivation for pictures.
(A) For both modalities
Left iFGBA 47–4430 –46.1–4228 –64.7–4030 –44.7
Left iFGBA 44/45–4224185.5–4420244.1–4224184.6
Left TP junctionBA 19/39–30–76523.3–32–76464.6–30–62363.3
Left sOGBA 19–24–94323.9–24–90384.2–24–94322.7
Left ant cingulate –814423.9–822503.5 –814423.2
Right iFGBA 474230–223.83832–263.54624–243.6
(B) Greater activation for words than pictures
Left mTGBA 21–60–30 –43.5–68–28 –24.5–60–30 –
Right cerebellum18–68–326.414–66–324.918–68–325.
Interaction*Normals W > PPatients P > WN(W–V)N(P–V)P(W–V)P(P–V)
(C) Differential activation for words and pictures
Left post cingulate 6–50125.1 6–48143.5 6–44124.53.23.2
Left iFGBA 47–4040–125.0–3842–122.7–3642–
Left iTGBA 20–64–24–244.4–64–24–244.3–54–26–304.
Left temporal poleBA 38–2816–343.5–2814–364.0–3416–362.03.33.0
View this table:

Regions showing significant differences in activation for the patients and controls, comparing semantic task with visual task

Main effectsSimple main effects
Regions of significant differential activation for the patient and control groups in the contrast of semantic–visual tasks. The group by task interaction is reported followed by the simple main effects for each contrast. (A) Regions with significantly reduced activation for patients relative to normals for both modalities; (B) Regions with significantly increased activation for patients relative to normals for both modalities. N(W–V) = semantic–visual for words in normals; N(P–V) = semantic–visual for pictures in normals; P(W–V) = semantic–visual for words in patients; P(P–V) = semantic–visual for pictures in patients.
Normals > PatientsNormalsPatientsN(W–V)N(P–V)P(W–V)P(P–V)
(A) Normals > patients for both modalities
Left iTGBA 37–54–52–103.9–54–46–
Right TP junctionBA 39/4062–60244.062–62223.93.12.2
Patients > NormalsNormalsPatients N(W–V)N(P–V)P(W–V)P(P–V)
(B) Patients > normals for both modalities
Left premr/motor BA 44/6–68–10263.7–68–10263.
Left ant sTGBA 21–66 –2 83.2–66 –2
Right temporal poleBA 385016–303.65018–323.
Fig. 1

Illustration of tasks performed by subjects during scanning. The four conditions are depicted as a 2 × 2 factorial design, i.e. semantic judgements on (i) objects (ii) words; visual judgements on (i) objects (ii) words.

Fig. 2

Structural analysis: regions of significant difference in grey matter density in each patient relative to the control group. The figures are thresholded at P < 0.05 (corrected). The sections illustrate left and right hemisphere regions of difference for each patient. The differences are displayed on a rendering of the standard brain MRI on this and all other contrasts. (For all contrasts: left = L; right = R.)

Fig. 3

Regions of significant regional cerebral blood flow change common both to patients and normals for the main effect of semantic judgements minus visual judgements. The threshold was set at Z > 3.1 (P < 0.001) in this and all other contrasts.

Fig. 4

Differential activation for patients and controls in the contrast of semantic minus visual judgements for both modalities: (A) regions showing significantly reduced regional cerebral blood flow changes for patients relative to controls; (B) regions showing significantly increased regional cerebral blood flow changes for patients relative to controls.


We would like to thank Richard Frackowiak for his helpful comments. C.J.M. is funded by an MRC Clinical Training Fellowship; C.J.P. is funded by the Wellcome Trust.


View Abstract