Right occipital cortex activations.

Using the tip of the tool to perceive vibrotactile stimuli on the left of fixation (as compared to on the right) increased the BOLD response resulting from a left visual stimulus, in a portion of the right lingual gyrus (peak Z-statistic = 3.55, (20, −68, −2), Figure 2B), in an area that probably lies close to the border of retinotopic visual areas VP, V2, and V4v, representing the horizontal meridian in the left visual field, where the visual stimuli were presented [24], [28]. This region has previously been activated in tasks involving visual spatial attention [25], and by the selection of visual and tactile targets for eye movements [29]. This cluster of right hemisphere lingual gyrus voxels also showed a significant decrease in BOLD response for right visual stimuli when the tip of the tool was used on the right side of fixation (as compared to when used on the left, see below). Additional tool tip position-dependent increases in BOLD signal following right visual stimuli were observed in the right posterior intraparietal sulcus and the superior (Z = 2.50, (32, −80, 40)), and inferior divisions of lateral occipital cortex (Z = 2.34, (50, −78, 16)). A nearby site in the superior division of the lateral occipital cortex (24, −82, 38) was previously shown to be activated in a task-independent fashion by spatially-congruent visual and vibrotactile stimulation [30]. The superior lateral occipital region activated in the present study is close to a region activated more strongly by spatial than by orientation visual discriminations [26].

Left occipital cortex activations.

When the tool tip was used to perceive vibrotactile stimuli on the right side of fixation, with a concurrent right visual distractor (as compared to on the left side, away from the same stimulus), the BOLD response in left occipital cortex was enhanced, with the peak voxel lying in primary visual cortex (Z = 2.87, (−14, −86, 14), Figure 2A), according to probabilistic cytoarchitecture [31]. This portion of visual cortex has previously been activated during saccades to visual and tactile targets [32], and has been argued to show supra-additive summation of visual and auditory inputs [33], [though see 27], suggesting a significant role for primary and secondary visual cortices in a number of multisensory integration tasks.

Right occipital cortex deactivations.

Tool tip position-dependent decreases in BOLD response were also observed in the right occipital cortex. Activity related to visual stimuli presented on the right of fixation, in the right superior and inferior divisions of lateral occipital cortex (Z = −3.57, (20, −86, 18), Figure 2C; Z = −2.66, (48, −88, −6), Figure 2D), probably the dorsal portion of V3 [28], [34], and in the right occipital pole (Z = −2.34, (14, −92, −8)), probably comprising part of visual area V2 [34], was significantly lower when the tool tip was used on the right side of fixation as compared to on the left.

Tool-position-dependent increases and decreases in BOLD response within the same voxels.

We noticed that a region of the right hemisphere lingual gyrus, as well as a number of other areas, displayed both increases and decreases in BOLD response relative to baseline, depending on tool position. We therefore performed an additional contrast, which assessed tool-position-dependent increases and decreases simultaneously. This contrast searched for areas in which the effect of visual stimulus position (either left>right or right>left) was larger when the tip of the tool was present next to the visual stimuli as compared to the effect of visual stimulus position when the tool was on the opposite side (i.e., for left>right visual stimuli: [TLVL>TRVR]>[TRVL>TLVR], and similarly for right>left visual stimuli). This contrast was assessed (voxelwise, p<.05, Bonferroni corrected) within a search volume restricted by a number of functionally-relevant criteria. We restricted the search volume to those voxels which: a) Showed a significant main effect of visual stimulus side (voxelwise Z≥2.33, p<.01, whole-brain cluster corrected to p<.05); b) Showed an increased response (Z>0, p<.5, uncorrected) when the tool was next to the contralateral visual distractor as compared to when positioned ipsilaterally (e.g., [TLVL>TRVL]), and; c) Showed a decreased response (Z<0, p<.5, uncorrected) when the tool was next to the ipsilateral visual distractor as compared to when positioned contralaterally (e.g., [TRVL>TRVR]). This multi-stage masking procedure resulted in a search volume of 6,409 voxels in the right hemisphere occipital cortex, and 3,118 voxels in the left hemisphere occipital cortex. The only brain region to survive these stringent multiple statistical criteria was a cluster of 13 voxels in the right hemisphere lingual gyrus, with peak Z-statistic of 3.93 and MNI coordinates: (18, −66, −2).

In summary, the analysis of the effects of tool position on the spatial processing of visual stimuli revealed clusters of activation contralateral to the visual stimuli, and clusters of deactivation ipsilateral to the visual stimuli. Of most prominence, a region in the ventral occipital cortex, most likely at the border of areas V2 and VP, showed both contralateral activations, and ipsilateral deactivations, depending on the relative positions of the tip of the tool and the visual distractors. Activity in this area of the right lingual gyrus showed clear modulations of visual processing during tool use. In analyses reported below, this area will also be shown to modulate its activity significantly as a function of the behavioural measures of multisensory integration, both in RT and percentage error measurements.

Since the BOLD response in the right lingual gyrus was sensitive to the position of the tool relative to the visual distractor on both sides of space, and also varied as a function of the behavioural measures of multisensory integration, we further analysed the ‘functional-’ and ‘effective-connectivity’ of this area. The raw signal, and the raw signal multiplied by +1 for task-, and −1 for rest-related epochs respectively, was entered as a regressor for each participant and run separately. Group analyses (voxelwise threshold of Z≥3.09, p≤.001, whole-brain cluster corrected, p≤.05) revealed that BOLD signal in widespread regions of bilateral dorsal, ventral, and lateral occipital cortex, bilateral superior and middle temporal gyri, and bilateral insula, thalamus, and putamen, covaried with signal in the right lingual gyrus (‘functional connectivity’). This large cortical territory may in part reflect vascular or other processes of little interest, so we further assessed the ‘effective connectivity’ of the right lingual gyrus. Signal covariation that was stronger during task periods than rest periods was restricted to bilateral portions of the occipital lingual and fusiform gyri, bilateral intracalcarine sulcus, and lateral occipital cortex in both hemispheres. These analyses suggest that, during the task periods, the right lingual gyrus functioned together with numerous regions in bilateral occipital cortex.

In an additional analysis, we searched for brain regions which showed a significant interaction between the positions of the visual distractor and the tip of the tool, both for the two hands (experiments) together, for each hand (experiment) separately, and for the three-way interaction including the factor of which hand held the tool (experiment). Such regions might be predicted to exist based on the idea of hand-centred or tool-centred processing of multisensory stimuli during tool use [10]. No such regions were found in the present datasets, either when searching across the whole brain, or restricting the search volume to just frontal and parietal sensory-motor areas (defined operationally as the approximate borders of Brodmann's areas 5, 6, 7, 8, 39, 40, 44, and 45, by creating a mask in MRICro, using canonical Brodmann area maps, overlaid in standard MNI space and smoothed with a 3D isotropic 4 mm FWHM filter to accommodate anatomical variability and imprecision). For comparison with the principal analyses reported in the present manuscript, uncorrected peak voxel coordinates and statistics for the interaction effects are provided in Table S2. Possible reasons for this failure to find significant hand- or tool-centred activations are discussed below. All other potentially relevant contrasts were examined within the factorial design of our experiments, for example, the simple effects of hand used, and the interaction between hand used and tool position. None of these additional contrasts resulted in any significant clusters of activation, and were not, in any case, of interest in the present report, so are not detailed further.

What is tool use?

This simple-sounding question is in fact extraordinarily difficult to answer. The interpretation of the present results and their relevance to other studies of ‘tool use’ clearly requires an answer to this question. Benjamin Beck is a renowned authority on the use of tools in the non-human animal kingdom, and in his 1980 book [46] on the topic he concluded: “After 15 years of trying [to provide a definition of tool use], I'm unhappy to report that I have not been totally successful.” [46, p. 4]. We concur with the thoughtful conclusions and the caveats raised by Beck, and believe that attempts to define tool use simply, non-arbitrarily, and non-circularly are fraught with difficulties. One option could be, for example, to resort to a dictionary definition, with the Oxford English Dictionary (OED) offering: “any instrument of manual operation”, where instrument is defined as an ”object used for a given purpose”. But this definition seems, to us at least, somewhat vague to be useful in a scientific context. A computer keyboard, for example, is a manually-operated object used for the purpose of data entry, but the sensory-motor interactions required in the prototypical use of a keyboard are rather different from those required to use a screwdriver, a rake, a pointer, or the canes used to navigate by the blind. According to the OED definition, the participants in our study were clearly manually operating an instrument for a specific purpose – without holding, correctly orienting, exerting downwards pressure, and moving the stick at the correct times, the participants in our task could not have performed correctly. From the fact that they did indeed perform the tasks well above chance, we can conclude that tool use was occurring during the fMRI sessions.

As compared to dictionary definitions, in the academic community studying the effects of a variety of tool use tasks on sensory-motor and multisensory interactions, tool use has rarely been defined explicitly. We have recently provided [47] one possible definition, extrapolating from Beck's definition in order to apply it both to humans in general, and to the scientific literature on tool use and peripersonal space in particular. In that definition, we aimed to draw a line (albeit an arbitrary one) between a number of tasks that either did or did not constitute tool use. Such a definition can of course never be the final word, particularly given the rate at which technology develops, especially that of tele-surgical and brain-computer interface devices. We believe, however, that our definition of tool use is both sufficiently liberal and sufficiently conservative to allow clear hypothesis-driven questions concerning multisensory attention (and multisensory peripersonal space) to be answered in a pragmatic way. Furthermore, we are extremely wary of the possibility that one might choose to define tools rather flexibly, on the basis of the results of particular studies of tool use – i.e., whether the results are consistent or inconsistent with one's beliefs about the effects of tool use on, for example, representations of peripersonal space or on multisensory attentional processes [48]. Since the key researchers working in the scientific field addressed by the present study must certainly agree that the task we used clearly constituted tool use (e.g., compare the wide variety of tool use tasks reported by such researchers [3]–[17]), we feel that this question over definitions is tangential to the relevance and importance of the present findings. In any case, if one were to decide that our task was not really ‘tool use’, we have still succeeded in showing that the position, relative to visual distractors, of the functional part of a manually-held object used to perceive distant vibrotactile stimuli significantly modulates the BOLD response in the occipital cortices, and that these modulations are most likely due to spatial shifts of attention towards the functional tip of that object. We see no reason why other, genuine, or more complex, forms of tool use would not also involve such shifts of attention to the functional part of the tool. This possibility, however, needs to be assessed in future research.

Finally, concerning definitions of tool use, we note the recent paper by St. Amant & Horton [49]. In their detailed theoretical discussion of how to define tool use in the animal and human behavioural literature, St. Amant & Horton also revised Beck's now classic definition of tool use to include those instances of tool use which “mediate the flow of information” between the tool user and the environment, including both direct physical interactions with objects, and communicative gestures. This ‘mediation of information flow’ through direct physical interactions with objects fits very well with the working definition of tool use that we proposed [47].

Visual attention at the tip of the tool, or a non-linear supra-additive interaction between multiple visual stimuli?

One of the reviewers of this manuscript pointed out that the increased BOLD responses in (contralateral) occipital cortex when the tool was held and used next to the visual stimulus could be due either, as we suggested, to the effects of spatial attention to the tip of the tool (and thus to the visual distractor presented at the same location), or else to an interaction between two separate visual stimuli (the visual distractor and the tip of the tool respectively), arising, simply and trivially, from the presence of more visual stimulation in that portion of the visual field. The interaction between the two visual stimuli could arise in one or both of two ways:

1) If the BOLD response in occipital cortex summed up separate, independent visual inputs in a non-linear and supra-additive fashion (the effects would have to be supra-additive because: a) The tip of the tool was present in the visual field for 12 or 36 s before the visual distractors were illuminated, likely giving sufficient time for the BOLD response to habituate to any simple visual effect of the tip of the tool alone, and; b) The critical contrasts were performed between conditions with identical visual distractors on the same side of space);

2) If the visual distractor stimulus increased the illumination of the tip of the tool relative to its background, thus increasing the overall illumination present in that portion of space. We can make four arguments against these possibilities.

First, regarding non-linear summation. Probably due to the habituation and saturation of the BOLD response, the large size of functional imaging voxels, and the presence of heterogeneous cell populations within a given voxel, studies of multisensory interactions in neuroimaging very rarely find any evidence for supra-additive summation, despite numerous clear examples of multisensory supra-additivity in single and multi-unit recordings, local field potentials, and in behavioural multisensory interactions [27], [33]. More commonly, additive or sub-additive effects are found in the BOLD response to independent stimuli, even when those stimuli are congruent with each other and presented close to each other in space and time. Similarly, and more directly relevant to the present concern, in one detailed and intensive study of the summation of responses to separate visual stimuli within a portion of retinotopic primary visual cortex, the BOLD response was found to sum linearly in two human subjects [50]. Given these findings, and assuming that no simple or physical visual interaction effects occurred (see the next paragraph), the possibility of non-linear and supra-additive spatial summation of separate visual inputs can be disregarded.

Second, it is possible that the visual distractors illuminated the tip of the tool, and that this increased illumination on the same side as the visual distractor resulted in additional BOLD response in contralateral occipital cortex. Note that such an explanation would not require any non-linear or supra-additive summation of BOLD responses, so is independent from the first possibility discussed above. Our experiments were performed in a dimly-lit scanner room. We did not make any measurement of the total illumination present at different positions in the visual field during the eight experimental conditions studied. We can only note here that the visual distractor stimuli were positioned at the rear of an unpainted aluminium box, mounted on a transparent acrylic table over the participant's legs. Both of these pieces of apparatus would have reflected some of the light of the distractor into the scanner bore. During the experimental set-up, we asked participants whilst inside the scanner bore to inform us if the visual distractor stimuli were being reflected off either surface, and if they were, we adjusted the distractor positions to remove these secondary visual inputs. By contrast with the smooth, reflective surfaces of the aluminium box and the acrylic table, the tip of the tool was wooden, rough, and less reflective. Thus, it is unlikely that any simple visual effects of reflection or illumination could explain the modulations of BOLD response in occipital cortex, since the less-reflective surface of the distal tip of the tool was occluding the more-reflective surfaces of the aluminium and acrylic apparatus when the tip of the tool was on the same side as the distractor.

A third point to note is that we observed both contralateral increases in BOLD response as a function of tool position (i.e., a tool used on the left side increased the BOLD response in right occipital cortex to left visual distractors as compared with when the tool was used on the right), and ipsilateral decreases in activation (i.e., a tool used on the left side decreased the BOLD response in left occipital cortex to left visual distractors as compared with when the tool was used on the right side). In order to explain our findings, any simple effects of two visual stimuli would therefore have to show both non-linear supra-additive positive BOLD in contralateral occipital cortex, and non-linear supra-additive negative BOLD in ipsilateral occipital cortex. This double and hemispherically-symmetrical non-linearity in the BOLD response is thus doubly unlikely as an explanation for our reported effects. Furthermore, in the presence of an attentionally-demanding RSVP task performed at central fixation, the BOLD responses elicited by large, high-contrast peripheral visual distractors in areas V1 to V4 in one hemisphere were unaffected by the presence of ipsilateral visual distractors, suggesting that ‘surround suppression’ does not operate inter-hemispherically [22]. This implies that the deactivations that we observed cannot be due to the mere presence of visual stimulation in the ipsilateral hemispace, but rather require an attentional explanation [21].

Fourth and finally, as detailed in the Results section, the regions of cortex in which the BOLD response was modulated significantly as a function of tool tip position, concur well with those activated in previous studies of visual or multisensory spatial attention, which is known to operate in a ‘push-pull’ manner, with both increases and decreases of attention-related activation in extrastriate cortex [21].

To conclude, we note that additional experiments in which participants performed both a passive and an active tool use task, under similar experimental conditions as reported here would be required in order definitively and finally to rule out the possibility of bilateral non-linear supra-additive spatial interactions due simply to the visual presence of the tip of the tool. However, given: 1) The generally linear summation of BOLD responses; 2) The fact that the tip of the tool was not highly visible and was in fact less reflective than the background; 3) The presence of both contralateral increases and ipsilateral decreases in BOLD response as a function of tool position, and; 4) The locations of the reported BOLD modulations relative to other studies of spatial attention, we believe that the attentional explanation is the more parsimonious one, since spatial attention has repeatedly been shown to lead to both increases in BOLD response for stimuli presented at the attended, and decreases in BOLD response for stimuli presented at unattended locations.

Absence of significant hand-centred or tool-centred multisensory interactions or BOLD responses.

The absence of evidence for a particular process can never, of course, be taken as evidence of the absence of that process. We did not find significant clusters of activation, or any strong indications that hand-centred multisensory processes were operating during our tool use task, under the conditions we studied, despite the presence of strong visual-vibrotactile behavioural interactions, and the indications from previous studies that such interactions may be hand- or tool-specific [9], [10], [12]. This is surprising if one believes that multisensory interactions during tool use occur in hand-centred reference frames, and that such interactions occur in or result from processing in parietal and or premotor cortices [10]. We do not hold this belief. Rather, we believe that many, if not all, of the reported effects of tool use on multisensory integration may in fact be due predominantly to eye-centred mechanisms of multisensory spatial attention, and that the locus of these effects may be in relatively ‘low-level’ or ‘early’ sensory (visual) cortices. We believe that the present data, our previous behavioural results, and many of the published results on the multisensory consequences of tool use from other laboratories, are most clearly and parsimoniously accounted for by such eye-centred effects of multisensory spatial attention, rather than by, for example, hand-centred mechanisms of peripersonal space. However, since the majority of studies in this field have not been able experimentally to distinguish between these alternative hypotheses, future studies are clearly required in order to resolve any apparent contradictions that remain in the literature.

Several other factors may account for the absence of hand-centred effects in the present dataset. First, the design of our study, or the fMRI protocol we used, may have been insufficiently powerful or sensitive to detect the very subtle, sparsely-distributed, or spatially very limited effects that have been proposed to generate the significant and apparently hand-centred multisensory interactions reported elsewhere. This possibility can only be confirmed with positive evidence showing that such processes do indeed exist and are measurable, while also providing evidence to rule out alternative possibilities such as the attentional hypothesis discussed here. Such evidence is not yet available, and we must therefore await future studies. If such effects do exist, yet are very weak or subtle, it raises the questions as to whether they can indeed explain all of the many reported multisensory consequences of tool use behaviours, and why the more powerful and more easily-detected effects of spatial attention are behaviourally and neurally less important or effective in this regard.

Second, it may be possible that the visual stimuli we used (static, flashing LEDs) are simply not able to activate the regions of parietal and premotor cortex that are thought to be involved in mediating sensory processing during or following tool use. This is an important possibility, which needs to be tested in future research. For now, we simply highlight the fact that, in the neuropsychological literature on the effects of tool use on cross-modal extinction, static flashing LEDs and three-dimensional rapid movements of an experimenter's finger have been shown, in one study, to result in comparable levels of visual-tactile interactions [14]. The possible implications from this study are either: a) That the stimuli used (both the rapidly moving fingers and flashing LEDs) were also insufficient to activate parietal and premotor cortex (since static LEDs and moving fingers produced comparable results) and that the reported effects must therefore have depended upon mechanisms of spatial attention and modulation of activity in occipital cortex, or; b) That LEDs are indeed sufficient to activate parietal and premotor cortex in an equivalent manner to rapidly moving fingers. The results of our study support the former, attentional interpretation.

Finally, it is possible that the specific kinds of tool used, or the specific tool use task performed, have a direct bearing on the kinds of multisensory interactions that occur, and the neural processes that are involved. This is almost certainly true in some respects, for example, relating to the well-known differences in the neural control of precision vs. power grasping movements, or between the reaching/transport and grasping/hand-shaping components of target-directed movements, which undoubtedly will differ between different tools. However, we only wish to note here that, with regard to the published evidence concerning the effects of a variety of tool use tasks on sensory-motor and multisensory processes, Maravita and Iriki [10] concluded:

“Intriguingly, whilst in some studies on humans the reported behavioural effects of tool-use occurred without any specific training … in other studies substantial tool-use training was required to elicit these effects … It might be that simple acts, like pointing or reaching with a stick will show behavioural effects without training, whereas more complex tasks involving dexterous use of a tool, such as retrieving objects with a rake require some training before any behavioural effects will emerge.” [10, p. 84].

We agree in general with these sentiments, however we remain cautious about the possibility that successful tool use is being defined here based upon the results (the ‘behavioural effects’) observed in the reported experimental settings, rather than upon the tool used and/or the tool use task being performed. The question therefore arises: Given a particular or novel tool use task, such as the one used in the present report, for how long should one continue the tool use training in order to test the hypothesis, for example, that tool use changes multisensory processing? If the answer is either: a) ‘Until multisensory processing changes’ or; b) ‘Until the well-known, prototypical effects of tool use emerge’, then it seems, at least to us, that such a hypothesis would be impossible to refute.

Piezoelectric vibrotactile stimulators.

Two custom-built MR-compatible piezoelectric-ceramic vibrotactile stimulators driven by a custom-built waveform generator were used to deliver vibrotactile stimuli. Each consisted of an aluminium box (5.5×2.2×8.0 cm) containing a 2 cm piezoelectric-ceramic element, vertically displacing a plastic rod ∼1 mm. The vibrating surface of the stimulus was 19.6 mm². Stimuli were presented at ∼200 Hz [52]–[53]. A small rubber semi-circular ‘guide’ was positioned on top of the vibrotactile stimulus to facilitate tool positioning during the experiment.

Additional apparatus.

The vibrotactile stimulator boxes were attached by Velcro™, 15 cm either side of the middle of an acrylic table (15×75×45 cm). One red LED (8 mm diameter, 660 nm, 550 mcd, 60° viewing angle) was positioned 1 cm above the vibrotactile stimulator on each side. The tool was a cylindrical wooden dowel (8 mm diameter×750 mm length). A rear-projection screen was positioned over the legs of the participants, ∼1.5 m from their eyes (Figure 1E). Responses were collected with a MRI-compatible button box, held by the participant in the hand opposite to the one holding the tool. The stimuli were controlled and the responses were collected using Presentation software (Neurobehavioural Systems Inc., Albany, USA). The experimental apparatus caused no detectable artefacts in the fMRI data.

Task blocks.

In the task blocks, participants performed a vibrotactile discrimination task and a visual fixation-monitoring task concurrently.

Vibrotactile discrimination task.

The participants were instructed to hold the distal tip of the tool in contact with the active vibrotactile stimulator on one side (left or right) throughout a block of trials, while maintaining fixation on a white cross (3×3 cm) presented centrally at the bottom of the rear-projection screen. Participants were instructed to respond as quickly and as accurately as possible to the vibrotactile target stimuli by pressing one of the two buttons: The left button for continuous, and the right button for pulsed stimuli, while maintaining central fixation and trying to ignore the visual distractor stimuli. The visual distractor stimuli were irrelevant to the task and were non-predictive of the vibrotactile target type.

Fixation-monitoring task.

The visual fixation cross dimmed for 250 ms, from white (100% screen brightness) to grey (70%) once during each task block. Following a fixation dimming event, the participants were required to withhold their response on the subsequent trial in the current block.

Rest blocks.

At the beginning of each rest block, the fixation cross was replaced by a white chevron (< or >) indicating the position (left or right, respectively) of the active vibrotactile target stimulator for the next task block. For half of the blocks, this required no change of the position of the tip of the tool, and for the other half, the participants were required to move the tip of the tool to the side indicated by the chevron as quickly as possible, while making the minimum of body movements (i.e., by moving only their wrist and fingers). After 5 s, the chevron was replaced with the fixation cross, which remained in place for a further 11 s, was extinguished for 500 ms, then re-illuminated for the last 1500 ms of the rest block as a cue for the participant to prepare for the impending task block. The participants were instructed to remain as still as possible, and to maintain visual fixation on the fixation cross throughout the rest block. It is very important to note here that the tool tip was in position and remained static for at least 12 s (and for half of the blocks, 36 s) before each and every block of experimental trials. This ensured that any changes in BOLD activity recorded in the task blocks relative to the baseline ‘rest’ blocks could not simply be due to the position of the tool or to the tip of the tool acting as an additional ‘visual’ stimulus (see also the Discussion). Task-related changes in BOLD response could therefore only be due to the main effects of visual distractor position, vibrotactile target/tool use location, the interaction between these variables, or to a main effect of task performance, which is of little theoretical interest and not reported here.

One or two days before the scanning session, participants trained on the tool use and fixation monitoring tasks for 10–20 minutes in a simulated scanner environment, lying supine, holding the same tool and discriminating the same target vibrations from the same apparatus. Recorded scanner noise was played in the background or over headphones. Such periods of training on tool use tasks has often been argued to result in significant changes in multisensory integration in peripersonal space [10].

Pre-processing.

All fMRI analyses were performed with FEAT (fMRI Expert Analysis Tool) Versions 5.63 or later, part of FSL (FMRIB's Software Library, http://www.fmrib.ox.ac.uk/fsl). The following analysis was applied to each functional run; Slice-timing correction using Fourier-space time-series phase-shifting; Motion correction using MCFLIRT [54]; Non-brain removal using BET [55]; 3D spatial smoothing using an isotropic Gaussian kernel of 8 mm full width at half maximum (FWHM); Global (volumetric) multiplicative mean intensity renormalization; High-pass temporal filtering (Gaussian-weighted LSF straight line fitting, with sigma = 54 s, corresponding to a low-pass cut-off of 1/108 Hz). Time-series statistical analysis was carried out using FILM with local autocorrelation correction [56]. Registration of each participant's functional T2*-weighted to high resolution T1-weighted scans and subsequently to the MNI52 standard brain template was carried out using FLIRT, with 7, and 12 degree-of-freedom linear transforms, respectively [54], [57].

Within-participant analysis.

The two time-series of functional data for each participant were modelled with a boxcar block design (24 s ON, 18 s OFF) convolved with a canonical (double-gamma) haemodynamic response function, and delayed by 6 s. Four regressors of interest corresponding to the four conditions (TLVL, TLVR, TRVL, & TRVR), and seven regressors of no interest (the temporal derivatives of the four main regressors, plus leftward tool movements, rightward tool movements, and fixation dimming events) were included in the model for each functional run. The two sets of contrast images obtained from each run for each participant were submitted to a higher analysis using a fixed effects model, by forcing the random effects variance to zero in FLAME [58]–[59]. The results of this within-participant analysis were passed-up to a higher-level between-participants (group) analysis.

Between-participants analysis.

The final group analysis was carried out using mixed effects (in which participant was a random effect) FLAME, in MNI152 template space, with final voxel dimensions of 2×2×2 mm. Statistically significant responses were determined by applying an initial Z-value (i.e., Gaussianised-t) cut-off as a cluster creation threshold, and assessing the size of contiguous clusters of voxels against Gaussian random field theory, correcting for multiple comparisons across the search volume to p≤.05, corrected. Several analyses also used Bonferroni corrections for multiple comparisons, as detailed in the Results section. Three different analyses were performed to assess visual, tool-dependent, and multisensory effects of interest.

In the first analysis, data from all eight experimental conditions (i.e., the four main conditions, for both the left and the right hand tool use experiments) were entered into a 3-way analysis of variance (ANOVA) with the variables hand (left, right), tool side (left, right), and visual distractor side (left, right). Weighted contrasts for all effects of interest were assessed with an initial Z-value threshold of Z = 2.33, p≤.01, and whole brain cluster corrected to p≤.05. The following contrasts were performed (Figure 1B–D): 1) The simple effects of visual distractor position were tested with two contrasts, one for the left [(TLVL+TRVL)>(TLVR+TRVR)], and one for the right visual distractors [(TLVR+TRVR)>(TLVL+TRVL)] (Figure 1B); 2) The effects of tool position were tested with two contrasts, restricted to the regions within the clusters defined by the simple effects of visual distractor position: [TLVL>TRVL], and [TRVR>TLVR]. The simple main effects of tool position (Figure 1C) were also tested; 3) Finally, the presence of hand-centred or tool tip-centred activations was assessed with the interaction between visual distractor position and tool tip position (Figure 1D). This was performed both separately for each experiment (i.e., for each hand used), collapsing across the two experiments, and with experiment as an additional variable in a three-way analysis.

In the second analysis, the mean of each of the eight experimental conditions was modelled out as a regressor of no interest, and two sets of eight behavioural measures (crossmodal congruency effect derived from RTs and percentage errors, with the across-participant mean subtracted per condition) were entered as predictors in separate analyses. Areas with BOLD responses that covaried significantly (either positively or negatively) with the behavioural measures of multisensory integration, were assessed across the whole brain, separately for RT and error score predictors.

In a third analysis, the first set of analyses (i.e., the contrasts involving the experimental variables of visual distractor position, tool position, and hand used) were repeated, but restricting the search volume to those regions in which the BOLD response varied significantly as a function of the behavioural measures of multisensory integration (i.e., the inclusive sum of significant clusters with either positive and negative covariation between BOLD response and RT or error measures of multisensory integration).

For all analyses, significant activation peaks of interest in the group statistical maps were interrogated using FEATQUERY: A mask was created manually in MNI template standard space, incorporating several voxels surrounding the peak voxel in order to create a mask, typically of 5–20 voxels (0.14–0.54 cm³) in volume after transformation back to each participant's native space. The number of voxels in the mask was determined by decreasing the Z-statistic threshold of the group activation map until at least five contiguous voxels around the peak were above threshold (this process was performed solely to ensure that sufficient voxels remained after transformation of the mask from MNI template standard space sampled at 2×2×2 mm, to the participants' native space sampled at 3×3×3 mm). The percentage signal change was averaged over those voxels for each of the eight experimental conditions separately, and further analysed with four-way repeated measures ANOVA (including the additional variable, task sequence order) in order simply to confirm the directions of particular effects following the whole-brain analyses. Since these analyses of signal change were biased by the prior selection of peak voxels in a given contrast, these exploratory signal change analyses were used predominantly: a) To exclude groups of voxels from subsequent interpretation which showed artefactual effects (for example, those peak voxels lying near the edges of the brain or ventricles); b) To exclude groups of voxels from subsequent interpretation which showed any significant effects or interactions with the order in which the tasks were performed (p≤.01), and; c) To assess whether significant contrasts resulted from activations, deactivations, or both directions of modulation relative to the baseline. Data from peak voxels in which artefactual effects of, or interactions with, block order were not included or discussed in the analysis or report. This process was performed after the main statistical analyses reported in the text. No other significant clusters were observed.

Two additional post-hoc analyses were performed based on the raw timeseries data of a cluster of ‘seed’ voxels in the right hemisphere lingual gyrus, centred on the MNI coordinate (20, −68, −2). The first analysis assessed the ‘functional connectivity’ between signal in this area and other brain areas by searching for voxels whose signal covaried with signal in the seed voxels. The second analysis assessed the ‘effective connectivity’ as a function of the task versus rest: The raw timeseries data were multiplied by +1 for signal reflecting task periods (i.e., delayed by 6 s), and −1 for signal related to rest periods. In each case, the demeaned timeseries was entered as a regressor in a first-level analysis without temporal filtering, convolution with the double-gamma HRF, or including temporal derivative regressors. Higher-level analyses were performed as described above.

All voxel coordinates in the text and figures refer to the MNI152 standard brain template in MNI152 template space.