Characterizing this functional variability, particularly when considering the genetic level, ideally requires acquiring functional imaging data from hundreds of subjects and organizing these data into a large-scale database, together with genetic, behavioral and biomorphological data. Databasing and analysis of structural magnetic resonance images has already resulted in probabilistic anatomical atlases 2223. However, a similar large scale description of functional networks is still lacking.

Given that we are in the early stages of exploration of the causes of inter-individual variability, it would be desirable for such a functional imaging study to cover a broad variety of cortical territories and to describe cerebral correlates associated with various level of cognitive processing, from simple perceptual processing to higher-level cognitive functions that require explicit learning and education. For instance, recent advances in the genetics of dysphasia, dyslexia, and dyscalculia have provided several candidate genes whose impact on inter-individual variability in the normal population remains unknown. Considering some cognitive tasks that have been extensively described in the neuroimaging literature, we chose to include: a mental calculation task to investigate superior fronto-parietal networks 24 and a language comprehension task which focuses on the inferior frontal and superior temporal lobes 25. Using auditory and visual stimulation allows us to isolate cortical pathways associated with perceptual processing (superior temporal sulcus and occipito-temporal cortex 2627) while the use of conjunction analysis across modalities also allows us to isolate correlates of amodal processing (associative cortices, for instance). Finally, evolutionary and developmental models suggest that some primitive mechanisms may be (partially) shared between hand/finger motor representation, speech language areas and correlates of mental arithmetic 282930, noticeably in frontal and parietal lobes. If such assumptions are verified, crossing analysis among these tasks may then help to dissect the task-related networks into a more subtle functional parcellation, and enlighten developmental issues of human brain organization. In brief, these considerations suggest that it would be particularly valuable to obtain images of the cerebral substrates of speech comprehension, reading, and calculation in a large number of subjects, associated with genetic, anatomical and behavioral data, in a highly standardized manner, and at a low cost.

As shown in Figure 1, we planed to acquire four types of data: functional images and a high-resolution anatomical scan for a fine description of sulci, grey and white matter, as well as (not described here) behavioral and personal data, aimed to create a rough cognitive profile of the subject, and DNA sampling (cheek swab). Recall that this data collection occurs within the constraints of this project: to be added to other running protocol, with a minimal cost of people and time.

In the present research, our goal was to define a simple fMRI test, less than 5 minutes long, that could delineate, in a subject-specific manner, those cerebral circuits. A functional sequence was added to each functional imaging session performed in our lab (Figure 1), taking advantage of the continuous flow of volunteers recruited for various protocols. Because we wanted to capture the maximal amount of functional information in the minimum amount of time, we designed the sequence according the following challenging constraints:

▪ the sequence had to be short, so as to disrupt as little as possible the main protocol. We choose 5 minutes for performing 100 trials.

▪ we aimed to obtain for each subject a description of different levels of functional architecture, from sensori-motor areas (perception and action) to more associative areas involved in reading, language processing and calculation.

▪ we aimed to capture in 5 minutes most of the individual networks related to each task.

▪ individual networks described in 5 min had to be reproducible over sessions and time.

The feasibility of using short stimulation designs (ranging from 10 to 25 min long) to reveal individual functional maps has been previously assessed for language mapping 31, for visual areas 32 and recently for a set of functional networks covering sensorimotor processes, working memory, executive functions and emotional processes 33. Beyond that point, the main goal of the current study was to use the data obtained with this fMRI protocol with individual subjects as the cornerstone of a large-scale hybrid database. Because the individual functional information that can be captured in such a short sequence should be considered with caution, we focus here on the design efficiency and within-subject. We then describe preliminary data obtained from 81 subjects scanned in a 3T scanner and address new methodological issues including statistical methods for analysis and visualization of inter-individual functional variability. Subsequent publications will exploit the potential of this database to focus on characterizing inter-individual variability.

Figure 2 shows the evolution of the statistics of four functional contrasts of two subjects with scanning time (T value, number of activated voxels, spatial location), allowing us to determine the minimal number of blocks required to reach a satisfactory description of the individual maps. Interestingly, all of the selected peaks exceeded the p < 0.001 threshold at voxel level after only one block and averaging t-values showed that one block was enough to reach about 60% of the final statistical significance. Locations of these peaks were on average 7 mm away from the final coordinates obtained after completion of 6 blocks. Finally, we reported that mean intra-subject and intra-session activation significance varies with block and across contrasts, suggesting that weaker activation peaks may be missing in some blocks when a fixed statistical threshold is applied. Importantly, this variability does not seem to reflect overall factors such as arousal or attention, which would have been predicted to yield similar profiles of activation for each contrast across sessions.

Reliability of individual maps obtained from one single block is assessed in Figure 3. Functional networks obtained from the first 5-minutes block appeared very similar to those obtained after a 30-minute session, except that the statistical significance increased with the number of blocks performed, thus allowing for a more stringent threshold. A first threshold-dependant quantitative approach, detailed in Table 1, showed that about the three-fourths of maxima reported for the two subjects after the completion of 6 blocks were already present as local or main maxima in the one-block analysis, with an averaged spatial shift of 6 mm (2 voxels), varying between 2 and 10 millimeters depending of subject and task. Conversely, about 40% of maxima isolated in the one block analysis did not appear as a local/main maxima in the six blocks analysis at the selected threshold. More interestingly, a second threshold-free approach (ROC analysis, Figure 3) revealed a high probability of a consistent classification of voxels into active and inactive categories across the 5-minutes and 30-minutes maps of a subject. For right hand action, auditory stimulation, calculation and reading task, respectively, the discriminative power (D_p) values were 1, 0.99, 0.98 and 0.87 for the first subject and 0.99, 0.99, 0.95 and 0.94 for the second one. We also reported a gradient of subject-specificity over contrasts: for right hand action, auditory stimulation, calculation and reading task, respectively, the averaged inter-subject D_p values were 0.95, 0.92, 0.78 and 0.63. This result confirms the sufficiency of the 5-minutes map and suggests high intra-subject consistency but lower cross-subject reliability for reading and calculation tasks.

Interestingly we directly compared individual activation maps obtained from the present protocol with those obtained from a more classic block-design paradigm completed during the same session. In the Additional File 1 we show six examples of within-subject reproducibility across experimental designs. These results suggest that the main individual foci of activation collected in our database truly reflect areas crucial for a cognitive task independently of the design (block vs. task shifting), number of trials and task details.

Examination of individual representative contrast images from subjects who participated in two fMRI sessions often indicated a high reliability of the activations (Figure 4, right column). Inter-subject variability is also illustrated and appears to vary with task: for the motor contrast, subjects displayed a similar pattern of activation around the left central sulcus. For the reading contrast, the inter-subject variability seems much higher, due to various combinations of activation, but mainly restricted to the left hemisphere (due to language left-lateralization). For calculation, spatial variability across individuals extended both in lateralization and in the relative amounts of frontal versus parietal activity. For instance, subject 5 strongly activated bilateral fronto-parietal regions, but only a subpart of this pattern was isolated for subject 2 and 3.

To quantify these observations, an inter-scans distance based on the calculation of a similarity coefficient were computed between activation patterns of the 13 subjects and 2 sessions and plotted in a reduced 3-dimensional space for three contrasts (Figure 4). A high degree of within-subject reliability was observed for reading and calculation tasks, as the two scans obtained in a given subject (dots with same color) were mostly tightly grouped together in this summary space. Considering the small size of the motor activations, we suspected that the distance computation was strongly influenced by voxels of non-interest. Indeed, when we masked the distance computation by the right-left motor activation from the RFX group analysis, all pairs of intra-subject scans were grouped together and the variance explained increased up to 66%. Overall, this suggests a high degree of similarity between intra-subject scans both in location and time course of activations. This observation was statistically assessed by a non-parametric Wilcoxon signed rank test computed on the 13 differences between inter-scan intra-subject distance (scan 1 – scan 2 distance) and averaged inter-scan inter-subject distances (scan 1 – 24 other subject's scans). P values for the difference between intra and inter-subject distances were 0.0215, 0.0398 and 0.0012 for motor, reading and calculation task, respectively.

A voxel-based random effect analysis (RFX) performed on the 81 subjects' contrast images allowed us to compare the efficiency of our fast protocol with results already described in the neuroimaging literature (see Figure 5 and Table 2 for a detailed listing of areas – maxima were labeled using the Anatomical Automatic Labeling (AAL) program package 34 except for visual areas where a functional labeling was preferred). Contrasts based on checkerboards, auditory and visual stimuli activate cortical areas related to perception and stimulus encoding: occipital lobe, several sites of the occipito-temporal pathway, auditory temporal cortex. Left and right hand action contrasts demonstrated activation of contralateral sensorimotor cortex, SMA and ipsilateral cerebellum, while the conjunction of audio and visual motor – sentences contrasts showed a large bilateral set of areas included SMA, rolandic operculum, cerebellum, thalami, postcentral and precentral areas. The main correlates of reading were a set of middle and superior temporal areas with a strong trend toward left lateralization, left frontal sites and inferotemporal areas and a set of subcortical sites. The conjunction of auditory and visual sentences mainly restricted the previous network to bilateral middle and superior temporal sites that surround auditory cortex and left frontal areas. The conjunction of audio and visual calculation minus sentences showed increased activation in bilateral intraparietal areas, putamen and left precentral gyrus. Because it is supposed that the cerebral mechanisms of mental numerical manipulation may partially overlap with visuospatial and language processing, we isolated areas common to calculation and reading: the conjunction of auditory, visual calculation and reading versus checkerboards isolated a small left lateralized network comprising SMA, precentral and temporal area. The conjunction of auditory, visual calculation and reading versus rest revealed additional left frontal and superior occipital sites, right cerebellum and right calcarine activation.

We compared different approaches for the functional group analysis to test methods that should both improve the statistical power of the group analysis and automatically collect individual functional information of a large fMRI database. A set of highly significant temporal, frontal and parietal activations related to motor, reading and calculation processing were isolated with the voxel-based RFX and plotted on a left sagital slice (column 6a). However, the overlapping of individual functional maps thresholded at p < 10^-2 at the voxel level shows that the activated voxels are rarely common to more than a third of the subjects (Figure 6b). Even in left motor cortex, the highest degree of overlap was about 66%. Conversely, some areas, like the left inferior temporal gyrus in the calculation contrast, were present in overlap maps but did not pass the RFX threshold. We enhanced the level of cross-subjects replication when using non-voxel based group analyses. The first one aimed to extract automatically for each contrast a list of individual maxima (called functional landmarks, or BFL) reliable across subjects but not located at exactly identical coordinates that conveys subject-specific information of activation magnitude, statistical significance and spatial location. Detection of single-subject peaks then reached 90% in left motor cortex and 93% in left intraparietal area (column 6c), that are supposed to be crucial for motor and calculation tasks, respectively. Interestingly, the average coordinates of the BFLs were close to the corresponding peak locations found in the overlap and RFX maps. The variances of BFLs location were comparable across all contrasts: the standard deviation error for Talairach coordinates was about 3.5–4 mm for most of the functional landmarks located in both sensori-motor and associative areas. A second procedure was performed to identify similar brain areas (parcel) across subjects mainly on the basis of their functional profile across conditions (Figure 6d). This functional parcellation isolated networks with a very similar topology to those observed with RFX analysis, but with an increase of sensitivity and more extended regions of activation functionally dissociated in subsets of areas: in Figure 7, we report in detail the functional profiles for some of these parcels. It appeared that the parietal activations reported in the voxel-based RFX correspond to a functional gradient, with an anterior site (area P1) strongly involved in all motor trials, the horizontal part of the IPS (P2) equally involved in motor action and calculation, while motor activation was essentially absent in the posterior intraparietal sulcus (P3) which seems more specifically activated by calculation trials, and shows a significant deactivation for language processing. Similar mosaics can be described in frontal lobe: descending from upper part of the precentral gyrus to inferior frontal sites, we observe an area strongly activated for all tasks except checkerboard flashing (parcel F1), then an area which shows preferential activation for motor and calculation tasks (F2). Below this (F3), preference for auditory sentence increases to reach its maximum specificity in (F4) close to Broca's area, but also in the anterior part of the superior temporal gyrus (T5). The posterior superior temporal areas show a high sensitivity to auditory material compared to visual ones (T2) whereas this tendency is inverted in a very close middle temporal area (T1) and reaches its maximal preference for visual word perception in the fusiform gyrus (T3).

Protocol duration was the main constraint in the current project. Adding a 5-minute functional scan to each fMRI acquisition performed in the lab could be done without difficulty. Our results indicate that this was sufficient to describe in a subject-specific and reliable way individual topologies of brain activations, with an average spatial accuracy of 2 voxels.

Furthermore, when several subjects were tested twice at an interval of several months, the topology of activations was largely reproducible, even for cognitive tasks such as mental calculation which might be expected to be subject to fluctuations due to learning or attention.

It is worth mentioning that for some subjects, similar patterns were obtained for the two sessions only when lowering the threshold of one session. As shown for subjects who performed 6 blocks in one session, the significance level of a given contrast may be subject to fluctuation even during a single session, and considering all subjects' contrasts images at a fixed threshold may obscure similarities and increases the risk of type I or type II errors in maxima detection. This observation is supported by the high discriminative power of individual statistical maps obtained after a 5 min session and strongly argues for a structural description of individual activations 3536, less dependent on anatomical location and statistical threshold but more consistent over time and sessions.

We emphasize that reproducibility should be assessed relative to our study's goals: as shown in Figures 3 and 4, some activation peaks can be missed by a given 5-minute scan, presumably due to statistical noise, but possibly also due to changes in strategies, attention, or task-related adaptation. In addition, experimental and/or analysis procedures are not infallible as illustrated by a few subjects for whom no BFL was extracted for motor tasks. However, in the present context, which is to create a database intended to perform behavioral-brain correlations at the scale of hundreds of individuals, this level of reliability is likely to suffice.

Group analysis performed on a first database of 81 subjects first allowed us to report several functional networks which fit with previous fMRI results in the fields of speech perception, reading, motor execution and number manipulation. Visual stimuli activated the ventral occipito-temporal pathway from basic retinotopic organization to a more anterior site of the left fusiform gyrus involved in visual word processing 37. Another gradient of functional specialization was observed in superior temporal gyri, from bilateral primary auditory cortices, easily detected with auditory stimulation, up to temporal areas located around the superior temporal sulcus (with a trend toward left hemisphere dominance) and recruited during multimodal sentence comprehension. They closely mirror the speech-processing areas detailed by Price et al. 38, with an anterior temporal activation associated to the processing of sentence structure, and a posterior middle temporal activation close to sites previously associated with sentence-level semantics 39. We also observed left frontal areas and SMA, which have been already reported for syntactic and semantic processing of sentences 4041. The calculation task involved the bilateral intraparietal and fronto-cingular network classically reported as active during simple number manipulation 4243444546, together with the bilateral putamen. These subcortical areas are not usually considered as a part of the core numerical system, but have been tentatively linked to sequential processing of multi-stage calculations or to the retrieval of arithmetical facts 44.

In addition to this coarse functional description of brain areas, due to the simplicity of tasks, the co-existence in our protocol of motor, language and calculation tasks allows examination of their respective neural correlates and gives new insights into some debated cognitive issues. For instance, a restricted subset of three left-lateralized cortical areas (precentral gyri, SMA, and posterior middle temporal gyrus) was isolated by examining the conjunction of sentence comprehension and calculation in both visual and auditory modalities. Although studies of patients have suggested a relative semantic and syntactic independence between language and arithmetic in the adult brain, such functional overlap may represent a core system on which language and calculation are articulated regardless of modality. In particular the left posterior temporal gyrus that has been described as a multimodal integration region 4748 could be a candidate for convergence of visual, verbal and non-verbal magnitude codes. The left superior occipital gyrus and right visual areas, involved in visual stimuli processing (reading task and checkerboard perception), were also equally recruited during calculation trials performed from visual or auditory inputs. This observation is compatible with the hypothesis that processes sustaining mental calculation may involve a top-down activation of a symbolic digital code 45 and may also share some cortical territories with visuo-spatial areas 46. It was possible to perform a complementary and improved functional dissection of cortical maps using a parcellation technique over experimental conditions and designed to compensate for inter-subjects anatomical differences. For instance, the two bilateral intraparietal clusters isolated in the calculation RFX, while very close to those reported literature, appeared spatially restricted compared to the individual activation maps where more anterior or posterior additional sites can be seen along the intraparietal sulcus. After individual parcellation of functional areas, the parietal areas related to calculation appeared more extended along the intraparietal sulcus, with an antero-posterior functional gradient that corroborates the geometrical layout reported by Simon et al. 49, as well as the hypothesis of distributed overlapping parietal representations proposed by 50. A detailed examination of the anterior parietal area indicates that a co-location of motor- and calculation-related activations exists at the single subject level, perhaps illustrating an extension of the numerical system to a sensori-motor representation of hands 29. Interestingly, a recent study of inter-individual variability of the infero-parietal cytoarchitecture showed a reproducible topography of areas that however vary in size and extent 51. This biological evidence supports the framework of our individual functional parcellation algorithm which assumed, as a methodological constraint, that parcels are connected in a similar spatial organization across brains. However, further investigations have to be done to see if functional parcellation matches is some respects cytoarchitectonic boundaries. This observation is also true for the mosaic of inferior frontal areas, which were underestimated, both in terms of spatial extent and statistical significance by the voxel-based RFX analysis; this lobe appears to have a complex functional topography, with a widespread intermingling of areas involved in motor response, language comprehension and mental calculation. Interestingly, the left inferior temporal area was absent at the selected threshold (Z > 4) for the calculation task voxel-based RFX analysis. Because the functional profile of this parcel shows a consistent involvement in both auditory and visual calculation tasks and because the overlap of activations across different individuals is only about 27%, an elevated degree of anatomical variability may be suspected in this area. In conclusion, the combination of tasks and modalities allows drawing a detailed functional continuum for middle and inferior frontal lobes, middle and inferior temporal lobes and superior parietal lobe, from more modality-dependant areas up to more abstract cortical regions.

Even if cerebral correlates of sensori-motor, reading, language comprehension and calculation processes are well identified and reproducible at the group level, subject level analyses revealed that individual activations were distributed in more complex patterns. Here, one of the most variable patterns across subjects, though associated with the highest intra-subject reliability, was the substrate of calculation. All subjects presented several sites of activation in the intraparietal sulci and frontal lobes but their extents, locations and combination varied. Because this individual patterns were highly reproducible over sessions, not only for large but also for small clusters, inter-subject variability should not be considered as noise but as a physiological signature of subject's brain activation during the numerical task. Whether such individual organizations depend on anatomical or functional constraints is a fundamental question. While we still presented our results in the MNI coordinate system and do not take into account the underlying anatomical structures, we believe that the automatic extraction of individual functional landmarks presented here was a first step to estimate for each functionally defined area both realistic spatial variability and frequency in the population and provides a promising context to address this issue. For instance, spatial variability of the left frontal functional landmarks described in our results fits well with the standard deviation associated to normalized left frontal anatomical landmarks location 8, which ranges from 2.5 to 5.7 mm. This underlies how structural changes may directly affect the location of functional sites and matches well with our estimate of their spatial variability. A recent paper reported a similar conclusion considering only visual cortical areas, where spatial variability of sulcal features was found to reflect those of the functional topography 52. As suggested by Juch et al. 8, automatically extracting and labeling activation and cerebral structures (sulci) jointly at the individual level will help to describe reliable cerebral organizations while separating anatomical and functional sources of inter-individual variability. Interestingly, a preliminary analysis of the spatial distribution of the functional landmarks in the left hemisphere did not show strong differences between tasks, suggesting that higher cognitive circuits are not associated with a greater spatial variability than sensori-motor ones. These results must be viewed with great caution, as this might be due to the algorithm of BFL extraction, which is based on a recursive method primarily derived from the RFX group analysis. However, the range of variability reported here is similar to that reported in another fast functional localizer study 33, at least for left-lateralized language areas (Broadmann area 44/45 and 22/42).

Concerning the functional features that should constrain individual circuits, it is particularly tempting to describe some frontal or parietal associative areas as functional 'nodes' because of their implication in various of tasks. We suspect that these sites are highly relevant to define individual functional reference frames for brain organization, similar to 'sulcal roots' in global brain gyrification 53 that appear early in the fetal brain and seem to constraint the adult global cortical folding. In a similar vein, detection of systematic spatial co-variation/co-lateralization of activation sites with these functionally defined 'nodes' (e.g. between calculation- and language-related areas) may be informative about the links that exist between different cognitive functions, and may help to specify the developmental constraints or evolutionary roots of the functional cerebral networks in the adult brain 54.

Further analysis and extension of the database to a greater number of subjects will be needed to disentangle the various sources of inter-subject variability listed in the introduction. We estimate that a minimum of 150–200 subjects will be necessary to begin to describe the variety of activation patterns in the population as well as to reach the minimal statistical power required to correctly isolate sub-groups that can be characterized by a combination of behavioral, anatomical, physiological and genetic features. We have no a priori assumptions about the rate of these possible 'variants' in the healthy population. For example, we may recall that 5% of right-handed subjects present an absence of usual left-hemispheric dominance in language correlates. Atypical organization of functional networks may also be suspected in subjects with self-reported developmental difficulties in arithmetic (~5% of our 'normal subjects' sample) or in adults who reported suffering of verbal difficulties during childhood (~10% of our sample). Identifying and characterizing some of these variants in various cognitive domains (verbal, visuo-motor, visuo-spatial, numerical...) may require hundreds of subjects. As demonstrated by a few studies, mostly from clinical populations, genetic variation may also contribute to the phenotype of cerebral activation in verbal 55 or numerical cognition 1456. A recent work has emphasized that genetic studies of human cognition might greatly benefit from considering a continuum of cognitive abilities in healthy populations 57. In the further, we plan to explore the impact of genotype on the structural and physiological features of non pathological functional networks, considering the normal range of variation in subjects' verbal, visuo-spatial and numerical abilities.

94 functional scans were acquired: 66 subjects performed one session, 13 subjects were recruited twice with an interval of several weeks, and 2 subjects performed 6 times the protocol in a thirty minutes session. In total, we obtained a database of 81 different functional scans and anatomy of French healthy subjects (35 males and 46 females; 24 ± 4.1 years old; 80 right handed and one left handed according to the Edinburgh inventory), 13 test/re-test situations and 2 multi-sessions to illustrate the evolution of results precision according to the number of scans acquired. All subjects gave their written informed consent. The study was approved by the regional ethical committee (Hopital de Bicêtre, France).

Ten types of trials were mixed together and presented successively in a fixed order, which appeared random to the subject (Figure 1), with the E-Prime software (Psychology Software Tool, Inc.): 1-passive viewing of flashing horizontal checkerboards (10 trials), 2-passive viewing of flashing vertical checkerboards (10 trials), 3-pressing three times the left button with the left thumb button according to visual instructions (5 trials), 4-pressing three times the right button according to visual instruction (5 trials), 5-pressing three times the left button according to auditory instruction (5 trials), 6-pressing three times the right button according to auditory instruction (5 trials), 7-silently reading short visual sentences (10 trials), 8-listening to short sentences (10 trials), 9-solving silently visual subtraction problems (10 trials), 10-solving silently auditory subtraction problems (10 trials). 20 rest periods (black screen) were inserted in the sequence and served as null event for a better hemodynamic deconvolution. Subjects were briefly instructed before starting acquisition with a visual sequence projected in the scanner during a non-functional acquisition. A post-acquisition debriefing served to ensure that the subject correctly understood and performed the tasks.

Visual stimuli were displayed as four successive screens (250 ms) separated by 100 ms interval, and each composed of group of one to three words, resulting in 1.3 sec of visual stimulation. Auditory stimuli were digitally recorded by a male speaker (resolution of 16 bits and sampling frequency of 22.05 kHz) and had a similar duration (about 1.6 sec for motor instruction, 1.2–1.7 sec for sentences and 1.2–1.3 sec for subtraction). Calculation consisted of a one-digit number (range from 4 to 9) subtracted from a two-digit number (range from 10 to 19). (See Additional File 2 for a detailed description of the stimuli).

The experimental protocol was organized as a fast event-related paradigm. The 100 trials were presented in a fixed sequence with a stochastic SOA (2400 ms, 2700 ms, 3000 ms, 3300 ms or 3600 ms; mean SOA = 3 sec). This sequence was optimized for both statistical detection and hemodynamic response estimation using a Matlab script developed in the lab and inspired by the genetic algorithm of Wager and Nichols 2003 58. One original feature of our procedure was to optimize the sequence according to more than one contrast. Eight contrasts of interest were selected: right vs. left hand action, vertical vs. horizontal checkerboards, auditory stimuli vs. rest, visual stimuli vs. rest, auditory calculations vs. auditory non-numerical stimuli, visual calculations vs. visual non-numerical stimuli, auditory stimuli vs. visual stimuli and visual stimuli vs. checkerboards.

Functional images were acquired on a 3T Brucker scanner using an EPI sequence (TR = 2400 ms, TE = 30 ms, matrix size = 64 × 64, FOV = 24 cm × 24 cm). Each volume consisted of 34 slices of 4 mm thickness. Anatomical T1 images were acquired with a spatial resolution of 1 × 1 × 1.2 mm. Data were pre-processed using SPM2 software 59 in Matlab7 environment according to the following procedure: slice timing, subject motion estimation and correction by realignment, coregistration of the anatomical image to the MNI template, spatial normalization of functional images (resampled voxel size = 3 × 3 × 3 mm) and smoothing (5 mm FWHM). Each voxel time series was fitted with a linear combination of the canonical hemodynamic response function and its temporal derivative. A temporal high pass filter was applied (cutoff 128 sec. and AR(1) whitening).

Individual contrast images were generated using SPM2. Individual conjunction maps were constructed to isolate voxels that were activated over a fixed threshold for multiple tasks: these maps were created by taking the min value of each of the contrasts considered for each voxel.