Introduction

The frontal variant of frontotemporal dementia (fv-FTD) produces changes in personality and social behavior [48, 55]. The most common cognitive deficit in fv-FTD is an impairment of executive functions and/or working memory [7, 40]. Patients also present attentional deficits, difficulties in mental set shifting and perseverative tendencies [64]. Deficits in planning, organization and other aspects of executive function become universal as the disease progresses. In accordance with these neuropsychological features, structural and functional neuroimaging studies of fv-FTD have shown that the locus of pathology typically starts in the ventromedial portions of the prefrontal cortex [29, 63] and then extends to the dorsolateral prefrontal cortex. In the advanced stages of the disease, the lesions affect extensive cortical and subcortical regions [19, 66].

Patients with the temporal variant of frontotemporal dementia, referred to as semantic dementia (SD) [27, 65], have degraded knowledge about the meanings of words and objects. This lexical semantic impairment is in singular contrast to a relative preservation of the syntactic and phonological processes of language [28]. Executive functions, visuospatial abilities, perceptual abilities and at least some aspects of episodic memory are relatively preserved. Brain imaging reveals a characteristic pattern of atrophy and hypometabolism involving the anterior portion of the temporal lobes, almost always asymmetrical and frequently predominant in the left lobe [16, 20, 42], but inevitably becoming bilateral as the disease progresses.

Given the respective executive dysfunction and semantic disorders of fv-FTD and SD, a comparison of the verbal fluency performances of these patients can yield particularly interesting data. Tests of verbal fluency consist in generating words belonging to a particular semantic category (e.g., animals) or words beginning with a given letter (e.g., the letter ‘p’) within a specified time limit, such as 2 min. These tasks are amongst the measures most widely used in neuropsychological examinations to explore semantic memory and executive functions. Semantic fluency is regarded rather simplistically as being largely dependent upon the integrity of semantic memory, while according to Perret’s much-cited conclusions [47], phonemic fluency is supposedly more sensitive to executive dysfunction. By administering a dual-task procedure to normal subjects, Martin et al. [39] found that phonemic fluency was dissociated from semantic fluency. Whereas phonemic fluency was more negatively affected by the performance of a concurrent motor task than by that of a concurrent object decision task, the opposite was true for semantic fluency. This dissociation suggests that phonemic fluency is more dependent upon the frontal lobe, whereas semantic fluency depends more on the temporal lobes. However, converging lines of evidence suggest that such a dichotomy is somewhat of an oversimplification and that interpreting verbal fluency deficits is not quite that easy.

In a meta-analytic review, investigating verbal fluency task performances by patients with focal cortical lesions, Henry and Crawford [25] showed that while both types of fluency impose comparable demands on the frontal structures, semantic fluency is relatively more dependent upon the integrity of temporal structures. Executive processes are involved in the initiation and monitoring of both fluency tasks, but semantic fluency is relatively more dependent on the integrity of semantic memory [25, 61]. More specifically, both fluency tasks involve different cognitive processes, including immediate attention in order to initiate the generation of words, an available knowledge base from which to select relevant items, an ability to retrieve items from verbal declarative memory and executive abilities to coordinate these processes, including working memory to monitor performance and avoid breaking rules [62]. Even though some of these cognitive processes are common to both semantic and phonemic fluency tasks, additional processes are differentially involved in the two forms of fluency. Hence, although both fluency tasks involve access to semantic information, different types of strategic semantic retrieval are engaged according to the nature of the task. Whereas the semantic fluency task, based on a categorical criterion, requires a categorization process, efficient phonemic fluency production, limited by an orthographic criterion, involves set shifting processes to a greater extent. Because both measures place comparable demands on frontal structures, although semantic fluency is relatively more dependent on temporal structures, it has been suggested that comparing the relative magnitude of deficits on phonemic and semantic fluency could be used to draw inferences about the prominence of executive dysfunction and semantic memory dysfunction, respectively.

Fv-FTD patients score more poorly than controls on both semantic and phonemic tasks, but display the same disproportionate results as the latter, i.e., like the controls, they perform better on the semantic fluency task than on the phonemic one [29, 44]. This profile is regarded as reflecting an inefficient search process (e.g., difficulty in generating search strategies, or switching to a new subset when previous ones are exhausted). SD patients, on the other hand, perform equally poorly on both fluency tasks [36, 59]. These performances are thought to stem from the disruption of semantic knowledge caused by SD. Although a few studies have directly compared the verbal fluency performances of fv-FTD and SD patients [59], none of them has studied the underlying cognitive determinants of these patients’ performances. Furthermore, even though frontal and temporal dysfunctions are thought to be responsible for verbal fluency impairment in fv-FTD and SD, respectively, the precise location of such dysfunction has yet to be properly established.

The aim of this study was to provide a direct comparison of performances by fv-FTD and SD patients on semantic and phonemic fluency tasks, thus holding out the hope of improving the differential diagnosis between fv-FTD and SD patients. We also sought to investigate the processes and the main cortical regions involved in both semantic and phonemic fluency tasks in fv-FTD and SD. We used positron emission tomography (PET) and statistical parametric mapping (SPM) to establish the correlations between resting-state FDG uptake in the whole brain and the fluency scores of the fv-FTD and SD patient groups, separately. The sites of significant correlation can be considered as reflecting the structures indispensable to the cognitive process being assessed [68], while activation mapping essentially highlights the cerebral structures implicated in, but not necessarily required by the task [52]. Voxel-based mapping of the correlations between cognitive performances and resting-state FDG uptake is a sensitive approach to delineating the neural substrates of cognitive impairment (for a detailed discussion, see [17]). This method has successfully been used in Alzheimer’s disease [14, 15, 18, 26, 34, 57], but only rarely in fv-FTD [51] and never, to the best of our knowledge, in SD.

Materials and methods

Subjects

We studied a group of 36 patients suffering from fv-FTD (n = 18) and SD (n = 18) selected according to the criteria established by Neary et al. [43]. For each patient, the selection was made according to a codified procedure, by senior neurologists (VDLS and SB) whose main activity is the diagnosis and follow-up of patients suffering from neurodegenerative disorders. To perform reliable differential diagnosis, all patients underwent a neurological examination, an extensive neuropsychological assessment including a non-verbal episodic memory assessment and information concerning patients’ past and present behaviors was obtained with a caregiver. Diagnoses were also supported by structural and/or functional imaging (SPECT). Some of these patients have already been described in previous investigations [16, 40, 51]. We also studied a group of 18 healthy elderly subjects, selected to match the fv-FTD and SD patients in terms of age, sex and education level. The demographic data of the patient and control groups are presented in Table 1.

Table 1 Demographic, general clinical and neuropsychological data of control and patient groups
Full size table

All participants were French native speakers and had a minimum level of education equivalent to the ‘certificat d’études primaires,’ a diploma generally obtained at ~14 years, after 8 years of primary education. None of the participants had a history of alcoholism, head trauma or neurological or psychiatric illness. Duration of illness was the length of time between the testing session and the appearance of the first symptoms of the disease according to the patients and/or their caregivers. All subjects gave their written informed consent to the study, which was approved by the local ethics committee.

Verbal fluency tasks

Both semantic and phonemic verbal fluency tasks were administered using the standard procedure developed by a reflection group on executive functions evaluation (GREFEX) widely used by French-speaking clinicians and researchers. Instructions insist on the variety of items to produce by giving an example with another category, but no indication on subcategories to explore is given. Such a procedure does not provide clues to the patients and does not suggest strategy to improve their performance or to overcome their difficulties.

In the semantic task, participants were instructed to generate as many different names of animals as possible in the space of 2 min. In the phonemic task, they had to produce as many words as possible beginning with the letter ‘p’ in 2 min. They were told that proper names, repetitions or words with different suffixes would not count. For all participants, the semantic fluency task was proposed first. The total score was the total number of words generated, excluding perseverative and intrusive errors.

Neuropsychological assessment

In addition, all the patients and controls underwent the same neuropsychological assessment. Verbal fluency impairment appears to be due to a semantic memory deficit and/or to the poor initiation or inflexibility of search and retrieval processes [67]. We therefore investigated semantic memory, working memory and a number of executive functions (updating and set shifting processes).

Semantic memory assessment

In a semantic knowledge task [21], three components (naming of drawings, categorical knowledge and concept attribute knowledge) were assessed. The subjects had to name 30 drawings corresponding to 30 concepts. Correct naming was scored 2 points. If the subject failed, a recognition task of the noun was carried out and was scored 1 point. Then, the subject had to answer ‘yes’ or ‘no’ to a series of questions about each of the 30-item concepts: superordinate category (“Does it occur naturally or is it manmade?”), category membership (“Is it an animal, a plant, an object or a body part?”), subcategory (“Is it a domestic or a wild animal?”) and specific attribute—either functional (“Is it edible?”) or perceptual (“Does it have a mane?”). The score was the total number of correct answers, the maximum being 236. The percentage of correct responses was calculated.

Working memory and executive function assessment

The working memory assessment comprised four tasks used to explore the three components of Baddeley’s theoretical framework [4]. The phonological loop was tested by a forward digit span (FD-Span) and the visuospatial sketchpad by a forward visuospatial span (FVS-Span). To assess the central executive, we used a backward digit span (BD-Span) and backward visuospatial span (BVS-Span).

We used the trail making test [58] (TMT) to assess ‘shifting’ and the running span test [54] to assess ‘updating.’ The time taken to process part A of the TMT gives a measurement of processing speed and that taken to process part B a measurement of the ability to flexibly shift the course of an ongoing activity [2]. In the running span task, 16 strings of consonants of an undisclosed length (4, 6, 8 or 10) are given orally, and the patients are required to recall the four most recent consonants in the right order. An ‘updating’ score is calculated on the basis of the 12 trials with a sequence of more than four consonants.

PET methodology

All 18 fv-FTD patients and only 15 SD patients (because of medical contraindications) underwent a resting PET study using [18F]fluoro-2-deoxy-d-glucose. Data were collected using the high-resolution ECAT Exact HR + PET device, with an isotropic resolution of 4.6 mm × 4.2 mm × 4.2 mm (FOV = 158 mm). [18F]Fluoro-2-deoxy-d-glucose uptake was measured in the resting condition, with eyes closed, in a quiet and dark environment. A catheter was introduced in a vein of the arm for radiotracer administration. Following 68 Ga transmission scans, 3–5 mCi of [18F]fluoro-2-deoxy-d-glucose were injected as a bolus at time 0, and a 10 min PET data acquisition was begun at 50 min post-injection. Sixty-three planes were acquired with septa out (volume acquisition), using a voxel size of 2.2 mm × 2.2 mm × 2.43 mm (x, y, z). During PET data acquisition, head motion was monitored continuously with laser beams.

In addition, patients underwent a T1-weighted volume MRI scan (1.5-T Signa Advantage EchoSpeed; General Electric) in order to perform the partial volume effect (PVE) correction, using the optimal voxel-by-voxel method, described in detail in Quarantelli et al. [53], and already used in our laboratory [10, 16]. All image processing steps were carried out using ‘PVE-lab’ software. Using statistical parametric mapping (SPM2; Wellcome Department of Cognitive Neurology, London, UK), the PVE-corrected PET data were then subjected to coregistration onto their respective MRIs, to spatial normalization using parameters obtained from the normalization of the corresponding MRI data into a customized MRI template obtained from our patient sample, and to a reslicing with a voxel size of 2 mm × 2 mm × 2 mm. The spatially normalized sets were then smoothed with a 14-mm isotropic Gaussian filter to blur individual variations in gyral anatomy and to increase the signal-to-noise ratio. The “proportional scaling” routine was applied to the PVE-corrected PET data to control for individual variations in overall FDG uptake. In order to minimize “edge effects” without excluding hypometabolic tissue in our subjects, only those voxels with values >40% of the mean for the whole brain were selected for the statistical analysis.

Statistical analysis

The demographic and clinical data, naming scores, total semantic score, working memory scores and updating scores were all subjected to one-way analyses of variance (ANOVA), with group as the between-subjects factor. Two Group × Test ANOVAs were carried out on TMT and fluency performances, with, respectively, TMT part A versus TMT part B performances and total semantic versus phonemic fluency scores as between-subjects variables. These ANOVAs were followed by post-hoc comparisons using Tukey’s honestly significant difference (HSD) tests to compare group means. Age at onset and duration of illness were compared between the patient groups using the paired Student’s t test. Correlations between both fluency scores in each group were studied by means of the Pearson correlation coefficient. Non-parametric equivalent to parametric analyses done were also conduced (data not shown). Results and conclusions were quite the same.

To investigate the relationships between the fluency scores and the neuropsychological data, we used the Pearson’s correlation coefficient, and to determine the predictive value of the neuropsychological scores for fluency scores, we conducted regression analyses. All hypotheses were tested at alpha = 0.05.

Correlations between the verbal fluency scores and resting FDG uptake in the whole brain were performed using SPM2. Considering the within group variability for age and the influence of this factor on cerebral metabolism [31], it was set as a confounding variable in a linear regression. Only correlations in the neurobiologically expected direction were conducted, i.e., positive correlations. We used a statistical threshold of P < 0.005 (uncorrected for multiple tests) for the voxels and a cutoff of k (corresponding to the number of voxels in a particular cluster) >150 to limit the attendant risk of false positives. Anatomical localization was based on Talairach’s atlas, using Brett’s set of linear transformations (see http://www.mrc-cbu.cam.ac.uk/Imaging/mnispace.html) and the AAL toolbox [70].

Results

Demographic and general clinical data

There were no differences between the patient and control groups for age [F(2,51) = 0.61; P = 0.54] or educational level [F(2,51) = 2.25; P = 0.12]. There was a significant group effect for the MMSE score [F(2,51) = 26.4; P < 0.001], with significantly lower scores for both patient groups compared with controls, and no significant difference between the scores of the SD and fv-FTD patients. There was no difference between the patient groups regarding either age at onset of illness [t(34) = 1.08; P = 0.29] or duration of illness [t(34) = 1.14 P = 0.26] (Table 1).

Fluency task performances

Mean total scores (i.e., number of words produced without rule-breaking errors or perseverations) are shown in Fig. 1. A two-way ANOVA: 3 groups (fv-FTD, SD and controls) × 2 fluency tasks (semantic and phonemic) indicated a significant group effect [F(2,102) = 56.51; P < 0.0001] and a significant fluency task effect [F(1,102) = 15.22; p < 0.0002], but no significant interaction [F(2,102) = 2.36; P = 0.099]. Post-hoc analyses indicated that the performances of both patient groups were poorer than those of the controls on the semantic and phonemic fluency tasks. Fv-FTD patients produced more words in the semantic fluency task than SD patients, whereas there was no difference between the two patient groups on the phonemic task. In addition, like the control group, the fv-FTD patients performed better on the semantic fluency task, whereas there was no significant difference between the two tasks for the SD group.

Fig. 1
figure 1

Semantic and phonemic fluency scores (mean and standard deviation) for controls, fv-FTD and SD patients

Full size image

Neuropsychological background tests

Regarding the total score on the semantic memory test, a one-way ANOVA revealed a group effect [F(2,51) = 18.07; P < 0.001]. Post-hoc analyses indicated that, unlike the fv-FTD patients, the SD patients scored more poorly than the controls. One-way ANOVAs indicated a group effect for both the FD-Span [F(2,51) = 7.1; P = 0.002] and FVS-Span [F(2,51) = 9.2; P < 0.001]. The SD patients’ performances did not differ from those of the controls, whereas the fv-FTD patients performed significantly worse than the controls and SD patients on the FD-Span. In the backward conditions, one-way ANOVAs indicated significant group effects for the BD-Span [F(2,51) = 12.9; P < 0.001] and the BVS-Span [F(2,51) = 9.7; P < 0.001]. In each condition, the patient groups’ performances, while no different from each other, were significantly poorer than those of the controls. For the updating score, a one-way ANOVA revealed a group effect [F(2,51) = 25.66; P < 0.001]. There was no difference between the fv-FTD and SD groups, and both differed from the controls, as indicated by the post-hoc analyses. Regarding the TMT, a two-way ANOVA: group (controls, fv-FTD and SD) × part (A and B) indicated a group effect [F(2,102) = 14.97; P < 0.001], a part effect [F(1,102) = 54.16; P < 0.001] and a marginal interaction effect [F(2,102) = 2.80; P < 0.066]. Post-hoc analyses did not indicate any significant differences among the three groups of subjects for part A. For part B, however, both patient groups performed more slowly than the controls, but in spite of a longer processing time for fv-FTD patients compared with SD patients, the difference between the two groups was not significant.

Correlation analyses

Interestingly, we observed significant correlations between the semantic and phonemic fluency scores within each patient group (r = 0.72; P > 0.001 for fv-FTD and r = 0.59; P = 0.01 for the SD), whereas there were no significant correlations for the controls (r = 0.30; p = 0.23). Table 2 shows the correlations for each patient group among the semantic and phonemic fluency scores and the semantic memory, working memory and updating scores and the processing times for the TMT parts A and B. Broadly speaking, significant correlations were observed between the two fluency scores and the semantic memory score in the SD group and mainly between the two fluency scores and the working memory and executives process scores in the fv-FTD group.

Table 2 Correlations among the fluency scores (semantic and phonemic) and semantic memory, working memory and executive scores in fv-FTD and SD groups
Full size table

Regression analyses

Stepwise regression analyses were performed with the TMT part B processing time, semantic memory, FD-Span, BD-Span, BVS-Span and updating scores (i.e., scores correlated with the fluency performances of the fv-FTD and/or SD patients) in order to gauge their effectiveness as predictors of each fluency score in each patient group. In SD patients, the best predictor of both fluency scores was the semantic memory score, while the TMT part B processing time and working memory scores were the best predictors in fv-FTD. The full set of results is presented in Table 3.

Table 3 Results of the stepwise regression analysis, with TMT part B processing time, FD-Span, BD-Span, BVS-Span and semantic memory and updating scores as predictors of the scores for the semantic and phonemic fluency tasks
Full size table

Correlations between fluency scores and FDG uptake

There were significant positive correlations among the fv-FTD group’s semantic fluency score and the normalized resting-state FDG uptake of the left and right frontal cortices, and, to a lesser degree, the right angular and inferior parietal gyri and the posterior part of the right insular cortex. For the phonemic fluency score, significant correlations were restricted to the right frontal inferior gyrus, extending to the posterior part of the insular cortex (Table 4; Fig. 2).

Table 4 Main peaks of significant (P < 0.005, uncorrected) positive correlations between FDG uptake and the scores of the semantic and phonemic fluency tasks for SD and fv-FTD patients
Full size table
Fig. 2
figure 2

SPM2-T maps for positive correlations between resting FDG uptake and the scores of the semantic and phonemic fluency tasks for SD and fv-FTD patients

Full size image

In the SD group, significant correlations concerning the semantic fluency score involved the left and posterior part of the inferior temporal and fusiform gyri, the posterior portion of the superior fontal gyrus and the supplementary motor area. As far as the phonemic fluency score was concerned, there was once again a significant correlation between the SD group’s total score and the normalized resting-state FDG uptake of the left temporal lobe, but this time its anterior part. One cluster involved the middle and superior temporal gyri and another the inferior temporal and fusiform gyri and the amygdala. Significant correlations also involved the left thalamus and caudate and left cerebellum (Table 4; Fig. 2).

Discussion

Cognitive study

We studied the semantic and phonemic fluency of 36 patients suffering from the frontal or temporal variants of frontotemporal dementia. The patient and control groups were rigorously matched for sex, age and educational level. We reported impaired performances for both groups of patients on both fluency tasks. The interaction between groups and types of fluency task was not significant, which may be due to the sample sizes and to the important variability of data. However, we observed differences between performances for both semantic and phonemic tasks for the two groups of patients. While patients suffering from fv-FTD performed better than SD patients on the semantic fluency task, we failed to observe any difference between the two groups in the phonemic fluency task. Previous studies have already reported verbal fluency impairment in fv-FTD [44, 56] and SD patients [36] compared with Alzheimer’s disease patients. However, although one or two studies have directly compared verbal fluency in fv-FTD and SD patients [59], none has investigated the underlying cognitive mechanisms of reported impairments.

The fv-FTD patients in our study presented considerably impaired performances on semantic fluency, whereas they performed normally on the semantic memory test assessing 30 items in different kinds of tasks (picture naming, categorization and specific questions). SD patients presented a similar impairment in phonemic fluency, but displayed a less severe executive impairment. These results confirm the multi-component nature of these tasks [62], suggesting that executive dysfunction contributes to low semantic fluency scores and that phonemic fluency also depends on the integrity of the lexical semantic store. The correlation and regression analyses we carried out unravelled the underlying mechanisms of verbal fluency impairment in fv-FTD and SD.

In the fv-FTD group, regression analyses for semantic fluency performances highlighted the extensive involvement of executive processes, especially set shifting and verbal working memory. Significant correlations were also found with the updating score, reflecting the fact that, given the limited capacity of working memory, the content of the phonological store had to be updated while the task was being performed. In the same vein, phonemic fluency performances were mainly related to set shifting and to verbal and visuospatial working memory storage capacities, witness the results of the regression and correlation analyses. One strategy for efficiently performing the phonemic fluency task consists in exploring the different words that can be produced on the basis of an initial phoneme. In addition to the high demand on the updating process, this also requires the subject to shift to another phoneme, once the one that has been selected no longer allows words to be produced. This process clearly involves the phonological loop, as reflected by the significant correlation between the phonemic fluency score and the FD-Span.

In the SD group, regression analyses clearly showed that performances depended mainly on the integrity of semantic memory, whichever fluency task was being performed. Correlation analyses also indicated links with additional cognitive functions. For the semantic fluency task, the significant correlations involved the BVS-Span and TMT scores, which may reflect the strategies implemented to perform the task. In addition to semantic memory, searching for various exemplars could involve visual imaging, especially for concepts defined primarily by their visual, form-related features, such as animals. Moreover, in order to produce a large number of animal names, subjects have to switch between the different subcategories of animals. Regarding phonemic fluency performance, in addition to the semantic memory abilities highlighted by the regression analysis, executive processes were also involved, as revealed by the significant correlations with the updating and BVS-Span scores.

Thus, the results of the cognitive study clearly indicate that, regardless of the nature of the fluency task, the fv-FTD patients’ performances were essentially related to their executive dysfunction, particularly set shifting abilities, whereas the SD patients’ performances depended mainly on their semantic memory abilities. This dichotomy between the underlying mechanisms of verbal fluency impairment in fv-FTD and SD could also be observed in the relative magnitude of deficits on the phonemic and semantic fluency tasks. The highly restricted semantic store of SD patients gives rise to extremely limited production whatever task is being performed. In fv-FTD patients, however, difficulties in retrieving words affect both fluency tasks to an equal degree, with the result that they maintain the performance profile observed in healthy subjects (i.e., with higher scores on semantic fluency than on phonemic fluency). As already highlighted in the meta-analytic review conducted by Henry and Crawford for focal cortical lesions [25], this result challenges the classic view that, compared with semantic fluency, phonemic fluency is more sensitive to executive dysfunction [47].

Correlation between fluency scores and FDG uptake

The difference between the more or less executive or semantic origins of verbal fluency impairment in fv-FTD and SD was also borne out by the metabolic study. In the fv-FTD group, our results indicated that semantic fluency performance was correlated with the metabolism of much of the frontal lobes, including the superior and middle gyri, bilaterally and the right inferior frontal gyrus. Consistent with the semantic nature of this task, the involvement of the left frontal lobe in semantic fluency has already been reported in Alzheimer’s disease, using the correlative method [14]. It has also been observed in activation studies of healthy subjects during semantic memory retrieval, for a review see [8]. The correlation we observed with the frontal lobe could also result from the involvement of executive processes in strategic retrieval from semantic memory. As demonstrated by our cognitive study, semantic fluency mainly involved set shifting in our group of fv-FTD patients (see above). This is consistent with the results of a recent study, which reported that performances on a nonverbal fluency task were specifically related to the volume of the left and right frontal lobes in fv-FTD patients [33]. The authors concluded that both frontal lobes play an important role in set shifting. Consistent with this idea, several functional imaging studies have also reported left-sided dorsolateral and medial frontal activity during various set shifting tasks, including the TMT [41, 71, 73], while other studies have reported a contribution from both frontal lobes [3]. In addition, the involvement of the frontal lobes that we have reported here is consistent with the involvement of the updating and verbal working memory capacities that we reported in the cognitive section of the study. The updating process specifically involves the left frontopolar cortex and the right inferior and superior frontal sulcus [11]. The prefrontal cortex in general has frequently been assigned a major role in working memory [23, 50]. Imaging studies have also demonstrated that bilateral parietal regions are engaged when verbal information has to be retrieved from short-term memory [13, 35]. It has been claimed that the left parietal cortex is the verbal short-term store [30, 45], whereas the right parietal cortex has been found to be involved in task switching [6, 74]. Correlations with the fv-FTD patients’ semantic fluency performances also involved the right frontal cortex. When they tested semantic fluency in normal subjects, Cardebat et al. [9] reported right dorsolateral and medial frontal activation, and assumed that it reflected the implementation of a cognitive strategy. Our findings suggest that this strategy consists in shifting between subcategories in order to conduct an in-depth exploration of the category in question.

Contrasting with the activation of an extensive frontal bilateral network observed for the semantic fluency task, the correlation analysis for the phonemic fluency task mainly highlighted the involvement of the right inferior frontal gyrus. Aron et al. [3] have suggested that this region may be especially involved in the inhibitory processes subtending switching, particularly the suppression of irrelevant responses. This idea is consistent with the involvement of these processes in phonemic fluency, which essentially requires the selection of words respecting the letter criterion from all the words that spring to mind on the basis of phonemic similarities.

Lastly, in both fluency tasks, we observed significant correlations with the metabolism of the right insula. These results may reflect the role of this region in aspects of phonological processing, such as rhyming and phonological verbal short-term memory; for a review, see [5].

Concerning the SD group, we found significant correlations between the scores on both fluency tasks and the left temporal lobe (Fig. 2), which is thought to subtend semantic representations [37]; for a review, see [22]. Interestingly, the anterior part of the inferior and superior temporal gyri and the posterior part of the inferior temporal and fusiform gyri would appear to be differentially activated in phonemic and semantic tasks. We believe that this distinction stems from the nature of the semantic representations activated in the different fluency tasks. All kinds of semantic representations may be used in a phonemic fluency task, whereas in the semantic fluency task, they are restricted to a specific category (animals in our study). Interestingly, the anterior temporal lobe is thought to ensure the modality-neutral representation of concepts [60, 69, 72], whereas the left inferior temporal gyrus is specifically involved in the semantic representation of living things [38]. We also found that the fusiform gyrus was associated with the visual features of semantic representations [12]. This result fits in with the sensory/functional theory that visually based categories activate ventral temporal regions that are typically allocated to visual object recognition.

The correlation analysis also indicated a link between semantic fluency performance and the left supplementary motor area and superior frontal area. This could reflect the involvement of working memory processes, as suggested by findings in normal subjects, e.g., [24]. We also observed significant correlations between phonemic fluency performance and the metabolism of the left thalamus and caudate, consistent with previous results in healthy subjects [46], which may be due to their connection with the prefrontal cortex, via the frontostriatal-thalamic circuits described by Alexander et al. [1]. Lastly, correlations with the cerebellum are consistent with the activation of this structure during word-generation tasks [32, 49].

Conclusion

In this study, we observed semantic and phonemic fluency impairment in fv-FTD and SD patients. Our results indicate that executive dysfunctions and semantic memory deficits may impact on both fluency tasks, but differentially so, according to the type of task and the nature of the pathology. They clearly demonstrate that the fv-FTD patients’ performances were primarily linked to executive functions and to the bilateral or right-sided metabolism of the frontal lobe, whereas the SD patients’ impaired performances on both fluency tasks are attributable mainly to a semantic memory disorder and to the hypometabolism of the left temporal lobe. Thus, semantic fluency impairment may be due either to executive dysfunction, especially the ability to shift between subcategories in fv-FTD, or to the deterioration of the semantic store in SD. Similarly, although the phonemic fluency task is thought mainly to involve executive and working memory functions, a semantic memory deficit drastically affects phonemic fluency performances. More specifically, we suggest that the inhibitory aspect of set shifting and the updating function are essential for carrying out phonemic fluency tasks. It is important to note that the relative magnitude of performances on semantic and phonemic fluency tasks could be used to draw inferences about the nature of the underlying deficit in standard examinations.