Critical brain regions for tool-related and imitative actions: a componential analysis

Using voxel-based lesion–symptom mapping in 71 patients with left hemisphere stroke, Buxbaum et al. assess neuroanatomical substrates of gestures to viewed tools, and imitation of either tool-specific or meaningless gestures. Temporal and parietal regions coding visual posture information and kinematic capacities were differentially required for tool-related and imitative gestures.

Abstract

Numerous functional neuroimaging studies suggest that widespread bilateral parietal, temporal, and frontal regions are involved in tool-related and pantomimed gesture performance, but the role of these regions in specific aspects of gestural tasks remains unclear. In the largest prospective study of apraxia-related lesions to date, we performed voxel-based lesion–symptom mapping with data from 71 left hemisphere stroke participants to assess the critical neural substrates of three types of actions: gestures produced in response to viewed tools, imitation of tool-specific gestures demonstrated by the examiner, and imitation of meaningless gestures. Thus, two of the three gesture types were tool-related, and two of the three were imitative, enabling pairwise comparisons designed to highlight commonalities and differences. Gestures were scored separately for postural (hand/arm positioning) and kinematic (amplitude/timing) accuracy. Lesioned voxels in the left posterior temporal gyrus were significantly associated with lower scores on the posture component for both of the tool-related gesture tasks. Poor performance on the kinematic component of all three gesture tasks was significantly associated with lesions in left inferior parietal and frontal regions. These data enable us to propose a componential neuroanatomic model of action that delineates the specific components required for different gestural action tasks. Thus, visual posture information and kinematic capacities are differentially critical to the three types of actions studied here: the kinematic aspect is particularly critical for imitation of meaningless movement, capacity for tool-action posture representations are particularly necessary for pantomimed gestures to the sight of tools, and both capacities inform imitation of tool-related movements. These distinctions enable us to advance traditional accounts of apraxia.

Introduction

The ability to perform complex tool-related actions and to imitate the actions of others is a hallmark of human motor performance. Deficits in both of these skills are central to the syndrome of limb apraxia, a heterogeneous and complex disorder of action planning and execution that cannot be attributed to weakness or sensory loss. Patients with apraxia perform gestural movements with errors in timing, sequencing, spatial organization, and less commonly, content. Despite over 140 years of investigation (Steinthal, 1871; Liepmann, 1900) and its frequent presence in left hemisphere stroke, Alzheimer’s disease, and corticobasal syndrome, limb apraxia remains relatively poorly understood. Nevertheless, recent research has demonstrated that apraxia has important implications for understanding the organization of object and action knowledge, motor simulation and planning, and the neuroanatomic substrates of these functions (Buxbaum and Kalenine, 2010; Binkofski and Buxbaum, 2013).

Although theoretical models of the praxis system differ in their details, many include both representational and production-related or kinematic action components (Roy and Square, 1985; Rothi et al., 1991; Heilman and Gonzalez Rothi, 1993; Cubelli et al., 2000; Buxbaum, 2001; and see Petreska et al., 2007 for a review). The representational component provides a processing advantage to the performance of familiar as compared to novel object-related actions (Buxbaum et al., 2005a), and has been aligned with conceptual action knowledge, also termed action semantics (Roy and Square, 1985). We have previously suggested that these action representations are likely to be patterns of co-occurrence of posture and movement features abstracted across multiple instances of spatiomotor and visual experience with a given class of similar actions (e.g. the arm and hand postures associated with using all hammers, regardless of their size and weight or the characteristics of the surface being hammered) (Buxbaum, 2001). In terms of the tests of gesture production historically associated with praxis testing, a strong measure of the representational component of praxis is ‘pantomimed gesture production in response to viewed tools’. Pantomime is generally held to be more sensitive than actual tool use because the shape and weight of tools in the hand is likely to provide feedback that alters the kinematics of movement, which in turn may augment deficient gesture representations (Goldenberg et al., 2004).

The kinematic component of praxis may be conceptualized as the basis for the planning of movement trajectories in terms of extent, direction and timing. Gesture kinematics may be particularly sensitive to current task constraints (e.g. the precise size and orientation of the hand and arm movements required to use a particularly-sized hammer on a surface of particular orientation to drive a particular type of nail). With reference to traditional praxis assessment, the kinematic component is perhaps most sensitively assessed with imitation of novel, meaningless gestures. Imitation of meaningless gestures (unlike imitation of known gestures) is thought to depend entirely on a so-called ‘direct’ route to action in which visual input must be transcoded into parameters for movement of the actor’s body, without the benefit of input from learned gesture representations (Rothi et al., 1992).

To this point, however, the neuroanatomic loci of these components of the praxis system are poorly understood. Apraxic production deficits in left hemisphere stroke have been observed after inferior parietal lobe and prefrontal lesions (Buxbaum et al., 2007; Goldenberg et al., 2007b), but also in patients with temporal and subcortical damage (Goldenberg, 1995; Tessari et al., 2007). Possible differences in patterns of performance attributable to lesions in these different loci have not been systematically explored. In this context, numerous recent functional neuroimaging studies have assessed the production of gestures using imitation, and to a lesser degree, pantomime tasks in neurologically-intact participants. A recent meta-analysis of gesture imitation (Caspers et al., 2010) revealed a large bilateral network of brain regions including inferior parietal lobe, temporo-occipital, premotor, primary somatosensory, and intraparietal areas that was activated irrespective of the hand used. Similarly, gesture pantomime has been associated with activation in numerous bilateral areas, including motor, premotor, inferior and superior parietal, and temporal regions (Hermsdoerfer et al., 2007).

In contrast, apraxia in stroke populations is also almost invariably a left hemisphere syndrome in right-handed and some left-handed individuals (Goldenberg, 2013). This pattern suggests that the right hemisphere activations associated with praxis in many neuroimaging studies may be epiphenomenal. Alternatively, it is possible that right hemisphere regions play a supporting, perhaps synergistic role in the facilitation of spatiomotor processing. Similarly, it is unclear which of the observed left hemisphere activations may be epiphenomenal or supportive rather than critical. Thus, lesion studies play an important role in our understanding of the importance and relative roles of differing brain regions in the components of praxis.

Previous literature helps in the identification of several candidate left hemisphere regions. Recent evidence suggests that the left posterior middle temporal gyrus is critical for representing semantic action knowledge. Lesions to this region impair recognition of tool use actions (Kalenine et al., 2010) and the ability to evaluate pictured actions (Tranel et al., 2003). Consistent with these data, left posterior middle temporal gyrus is consistently activated in action knowledge tasks with healthy subjects (Watson et al., 2013) with both picture and word stimuli (Kable et al., 2002, 2005; Assmus et al., 2007; Vingerhoets et al., 2009; Wallentin et al., 2011).

A second left hemisphere region of probable importance is the left inferior parietal lobe, long recognized as a substrate for skilled gesture production (Heilman et al., 1983; Heilman and Gonzalez Rothi, 1993). Patients with left inferior parietal lobe lesions produce spatial and temporal errors during imitation of tool use pantomimes (Halsband et al., 2001). Moreover, inferior parietal lobe-lesioned patients tend to be more impaired in imitating meaningless as compared to meaningful gestures (Kolb and Milner, 1981; Goldenberg and Hagmann, 1997; Haaland et al., 2000; Weiss et al., 2001; Tessari et al., 2007). As noted, the former may be disproportionately reliant on kinematic processing. Consistent with this reasoning, we demonstrated in a previous study using voxel-based lesion–symptom mapping (VLSM) that parietal lesions disrupt detection of errors in spatiotemporal (kinematic) gestural information (Kalenine et al., 2010; but see Goldenberg, 2009 for an alternative account focused on apprehension of categorical spatial relationships).

Additional regions of probable importance in praxis production are left motor, premotor, and prefrontal cortices, particularly the middle and inferior frontal gyri. Theta burst stimulation of the left inferior frontal gyrus interferes with gesture production (Bohlhalter et al., 2011). In patients with corticobasal syndrome, gesture production deficits are associated with left frontal cortical and subcortical volume loss (Borroni et al., 2008; Huey et al., 2009). Moreover, left inferior frontal gyrus and sensorimotor cortex are activated in functional neuroimaging studies of gesture imitation and production (Muhlau et al., 2005; Hamilton and Grafton, 2009), and left middle frontal or inferior frontal gyrus stroke may result in deficits in gesture imitation (Haaland et al., 2000; Goldenberg et al., 2007a).

Based on these data, we hypothesize that the representational component of praxis (which we and others align with semantic action knowledge) (Kalenine et al., 2010) is largely subserved by the left posterior middle temporal gyrus, and the kinematic component primarily by the left inferior parietal lobe. To assess these hypotheses, we used VLSM to assess the cortical brain regions associated with impairments in three types of gestures performed with the ipsilesional (left) hand of left hemisphere stroke patients: gestures produced in response to viewed tools (GestTool), imitation of tool-specific gestures demonstrated by the examiner (ImTool), and imitation of novel, meaningless gestures (ImNov). Thus, two of the three gesture types were tool-related, and two of the three were imitative, enabling pairwise comparisons designed to highlight conjunctions and disparities between tasks. Additionally, gestures were scored separately for postural (hand/arm positioning) and kinematic (amplitude/timing) accuracy to assess the hypothesis that representational and kinematic capacities would be differentially critical to the three types of actions. Specifically, we hypothesized that the kinematic aspect should be particularly critical for imitation of meaningless movement, capacity for tool-action representation should be necessary for pantomimed gestures to the sight of tools, and both capacities should inform imitation of tool-related movements.

Materials and methods

Participants

Seventy-three individuals who suffered a left-hemisphere stroke including the cortex were recruited for the study. Sixty-four of the strokes were ischaemic and nine were haemorrhagic. Participants were recruited from a large registry at the Moss Rehabilitation Research Institute. Patients over the age of 80 years or with histories of comorbid neurological disorders, alcohol or drug abuse, psychosis, or severe language comprehension deficits were excluded. All patients gave informed consent to participate in the behavioural testing in accordance with the guidelines of the Institutional Review Board of Einstein Healthcare Network and the Declaration of Helsinki and were paid for their participation. Fifty-three patients also provided informed consent to participate in an MRI or CT imaging protocol at the University of Pennsylvania School of Medicine. All participants were paid for their participation and reimbursed for travel expenses. Two participants were excluded after brain imaging revealed bilateral ischaemic infarction.

Demographic and experimental behavioural data for the remaining 71 participants (33 females; mean age 58 years, age range 35–80) are reported in Table 1.

Table 1.

Demographics and characteristics of the 71 patients participating in the study

Subject	GestTool (%)	ImTool (%)	ImNov (%)	Lesion Volume (voxels)	Gender	Hand	Education (y)	Age	Months post
01	45.0	75.0	72.5	6572	F	R	12	79	69
02	82.5	75.0	72.5	12252	M	R	16	63	7
03	87.5	100.0	100.0	60218	F	R	18	50	5
04	63.9	57.5	60.0	196024	F	R	18	54	165
05	90.0	100.0	92.5	50218	M	R	14	68	9
06	50.0	55.0	77.5	27095	F	R	12	41	12
07	85.0	100.0	85.0	118909	F	R	14	39	50
08	80.0	82.5	65.0	204861	M	L	8	71	152
09	63.9	53.3	55.0	16670	M	R	8	74	20
10	97.5	88.3	72.5	40102	M	R	11	62	45
11	75.0	75.8	72.5	6342	F	R	16	56	59
12	87.5	74.1	75.0	20369	M	R	11	57	62
13	85.0	75.9	77.5	34186	M	R	8	78	39
14	97.5	92.5	95.0	109589	F	R	13	47	82
15	75.0	55.0	57.5	181834	F	R	17	48	23
16	85.0	100.0	90.0	60271	F	R	12	51	16
17	62.5	70.0	62.5	155492	M	R	12	64	82
18	87.5	86.7	82.5	1913	F	L	12	52	9
19	65.0	55.0	57.5	135867	M	R	13	60	142
20	95.0	100.0	95.0	7507	M	L	10	51	13
21	57.5	75.0	65.0	83076	M	R	12	40	26
22	55.0	50.0	25.0	138731	M	R	16	58	135
23	92.5	64.2	72.5	44506	M	R	11	51	82
24	82.5	100.0	90.0	189818	F	R	16	72	170
25	87.5	95.0	92.5	88958	M	R	12	68	319
26	50.0	77.6	72.5	54830	F	L	12	49	14
27	92.5	100.0	92.5	90534	M	R	12	54	89
28	92.5	100.0	87.5	40759	M	R	11	42	22
29	92.5	100.0	92.5	41049	F	R	14	54	8
30	90.0	100.0	92.5	14749	F	R	15	44	9
31	90.0	100.0	92.5	48288	M	R	19	64	65
32	60.0	37.5	40.0	56061	F	R	16	80	10
33	37.5	47.5	50.0	188649	F	R	12	63	143
34	90.0	100.0	95.0	10583	F	R	13	77	8
35	62.5	47.5	57.5	41363	M	R	16	56	9
36	67.5	70.0	62.5	27223	F	R	12	67	91
37	87.5	100.0	95.0	33904	F	R	11	66	61
38	60.0	70.0	50.0	31436	M	R	11	79	27
39	87.5	100.0	92.5	13995	F	R	16	55	79
40	65.0	80.0	50.0	36717	F	R	12	58	6
41	77.5	100.0	95.0	72680	M	R	20	55	74
42	75.0	85.0	75.0	49432	M	R	12	50	17
43	97.5	90.0	90.0	7916	M	R	13	61	17
44	100.0	90.0	90.0	94098	M	R	14	69	20
45	92.5	82.5	82.5	3913	F	R	12	66	77
46	80.0	77.5	77.5	23437	F	R	12	50	145
47	97.5	77.5	77.5	114398	M	R	16	66	14
48	50.0	72.5	52.5	202634	F	R	10	41	127
49	85.0	76.7	70.0	8557	M	R	16	59	29
50	97.2	100.0	85.0	165844	M	R	16	55	22
51	75.0	60.0	57.5	185280	M	R	13	64	42
52	70.0	67.5	67.5	68474	M	R	16	49	35
53	85.0	100.0	92.5	105131	M	R	21	57	114
54	55.0	85.0	90.0	124897	F	R	16	40	2
55	87.5	100.0	90.0	15956	F	R	12	44	10
56	90.0	100.0	90.0	46576	F	R	12	35	7
57	85.0	75.0	70.0	70434	M	R	12	78	41
58	87.5	64.2	65.0	75994	M	R	11	61	27
59	77.5	71.7	75.0	264697	M	R	19	59	55
60	87.5	66.1	67.5	114029	F	R	17	50	23
61	85.0	84.2	87.5	43844	M	R	16	64	20
62	80.6	65.8	67.5	57332	F	R	14	74	21
63	82.5	80.0	67.5	19811	F	R	12	60	8
64	56.3	52.7	60.0	266719	F	R	16	55	12
65	97.5	94.2	95.0	45215	M	R	18	55	14
66	97.2	87.5	82.5	68902	M	R	14	42	9
67	97.5	86.7	77.5	10133	M	R	12	60	10
68	80.6	61.7	50.0	41726	F	R	12	56	22
69	90.0	81.7	75.0	27377	F	R	12	54	20
70	100.0	67.5	67.5	71050	M	R	12	71	14
71	87.5	100.0	95.0	20132	M	R	12	64	14

Behavioural tasks

Gesture to tool

Subjects were presented serially with 10 household tools (e.g. scissors, fork, comb) and asked to demonstrate precisely without touching the tool how they would if they were holding it with the (less impaired) left hand. Gestures were videotaped and later coded as correct or incorrect on each of five dimensions: content, hand posture, arm posture, amplitude, and timing. Scoring was performed according to a detailed error taxonomy by trained coders who demonstrated reliability with previous coders in our laboratory (Buxbaum et al., 2005a) as defined by Cohen’s Kappa > 0.85 (‘very good’ agreement, Altman, 1991). The Supplementary material provides details of the scoring criteria.

Postural scores were calculated by averaging the hand and arm posture component scores, which were moderately strongly correlated (P < 0.001) for each gesture type [GestTool: r(69) = 0.46, ImTool: r(69) = 0.75, ImNov: r(69) = 0.79]. Kinematic scores were calculated by averaging the amplitude and timing component scores, also moderately strongly correlated (P < 0.001) [GestTool: r(69) = 0.64, ImTool: r(69) = 0.70, ImNov: r(69) = 0.54]. Total gesture scores for each item were calculated by averaging all four component scores.

Imitation of tool-related gesture

Participants were shown videoclips of an experimenter performing 10 transitive pantomimes associated with household tools (e.g. scissors, screwdriver, toothbrush). Participants were asked to imitate the experimenter’s pantomime as precisely as possible. In this task, as well as in the novel imitation task described below, the experimenter used the right hand and faced the camera in second person perspective. Participants mirrored the gestures with the left hand. Each gesture was shown twice in succession, and the subject was permitted to begin while observing the target gesture. Gestures were videotaped and later coded by the same coders using the same scoring criteria applied to GestTool performance.

Imitation of novel gesture

Participants were shown videos of an experimenter performing 10 novel gestures with the left hand. The novel gesture task was developed with reference to the gestures assessed in the ImTool condition. Specifically, for each tool-related gesture from the ImTool task, the plane of movement (vertical/horizontal), joints moved (shoulder/elbow/wrist/fingers), grip type (hand open/clenched/partially open), and oscillations (present/absent) were tabulated. The items of the ImNov task preserved the characteristics of the meaningful gestures with respect to these attributes (see Buxbaum, 2001 for details of stimulus development). Each gesture was shown twice in succession, and subjects were permitted to begin imitating while observing. Participants’ gestures were videotaped and scored as described above.

Imaging, lesion segmentation and warping to template

Details of imaging, segmentation, and warping methods are provided in the Supplementary material. High-resolution structural brain images were collected from 53 participants. For 18 participants for whom MRI or CT research scans could not be obtained, a recent clinical CT or MRI was judged by the project neurologist (H.B.C.) to be of sufficient quality and resolution for lesion segmentation.

A research team member manually segmented the brain lesions of 19 of the participants who underwent research MRI scans. The patient structural scans and lesion masks were then warped to a common 1 × 1 × 1 mm template [Montreal Neurological Institute (MNI) space ‘Colin27’]; (Holmes et al., 1998). For the rest of the participants who received research scans (n = 34; CT = 21, MRI = 13) and those who had clinical scans (n = 18), H.B.C. drew a lesion mask directly onto the Colin27 volume. The Colin27 template was rotated to match the pitch of the imaged patient brain. All lesion masks were thresholded and quantized to produce a 0/1 map: voxels containing a mask value >0.5 were assigned a value of 1, and all others were assigned a value of 0 (Fig. 1).

Figure 1.

Illustration of the 71 left-hemisphere lesions displayed on a template brain. Lesions are represented on the surface of the brain but display both cortical and subcortical damage to an 8-voxel search depth.

Voxel-based lesion–symptom mapping

VLSM analyses were conducted using the VoxBo brain imaging package (Kimberg and Aguirre, 2001). At each voxel, VoxBo performs a one-tailed, independent sample t-test on the behavioural scores of patients with and without lesions. To correct for multiple comparisons, voxels with values exceeding a false discovery rate (FDR) threshold of q = 0.05 were considered significant (Genovese et al., 2002). Only voxels damaged in at least five participants were included in the analyses. The number of qualifying voxels was 384 896, or 52% of the 738 535 voxels in the left hemisphere based on the Automatic Anatomic Labelling (AAL) atlas (Tzourio-Mazoyer et al., 2002).

In addition to raw behavioural measures, residualized scores were computed by regressing one score against another to remove the shared variance between scores. Residualized scores enabled assessment of voxels associated with performance that was disproportionately impaired on one task or component given performance on another.

For all VLSM analyses, neuroanatomic labels for significant regions were generated both with Brodmann and AAL atlases as implemented in MRIcron.

Results

Behavioural results

As shown in Table 1, scores on the three gesture tasks were well distributed. Cut-off scores for normal performance (two standard deviations below the mean of age-matched healthy controls) were 79.8, 80.6 and 84.8 on GestTool, ImTool and ImNov, respectively, based on normative data published previously (Buxbaum et al., 2005a, 2007). On these criteria, 34, 41 and 32 patients performed abnormally on GestTool, ImTool, and ImNov, respectively. Patients’ scores were equivalent on the GestTool and ImNov tasks [mean 80.1 and 79.8, respectively, t(70) = 0.19, P = 0.85], and more impaired on the ImTool task [mean 75.6, t(70) > 2.5, P < 0.02 for both pairwise comparisons]. Scores on all three tasks were moderately strongly correlated [r(69) > 0.62, P < 0.001 for all three pairwise correlations]. Figure 2 shows examples of errors in the GestTool condition.

Figure 2.

Examples of hand and arm posture errors made by study participants on the pantomime to sight of tool task (GestTool). (A) Hand and arm posture errors. In pantomiming eating with a fork, the participant shakes fist with thumb out (hand posture error) and maintains arm in a fixed position lateral to the body throughout (arm posture error). (B) Hand posture error. In pantomiming winding a watch, the participant forms a static precision grip (‘pinch’) and rotates the entire hand in a circle parallel to the watch-face.

Lesion analyses

Total scores

Figure 3 shows an overlap of the number of patients with lesions in each voxel and suggests the relative power of each voxel for detecting an association, if one exists. The map shows good coverage of the regions of interest in the posterior temporal, inferior parietal, and frontal lobes, with 32 patients having lesions in the regions of maximum overlap.

Figure 3.

Map depicting lesion overlap of the 71 participants. Only voxels lesioned in at least five subjects were included. The regions rendered in purple and blue correspond to an overlap of 5–15 participants. The regions rendered in aqua and green correspond to an overlap of 16–23 participants. Regions rendered in warm colours (yellow to dark red) were lesioned in at least one-third of the sample (overlap of ≥ 24 participants).

Table 2 and Fig. 4 provide VLSM results for the GestTool, ImTool and ImNov tasks. For the GestTool task, a large region in the posterior temporal cortex encroaching on extrastriate visual cortex (Brodmann area 19) exceeded the FDR statistical threshold. Smaller significant regions were observed in the inferior parietal lobe and primary sensory area (S1), middle and inferior frontal gyri, thalamus, and angular gyrus. For the ImTool task, a large region exceeding the statistical threshold was again observed in the posterior temporal lobe. In addition, a second region was observed in primary somatosensory area (S1), primary motor area (M1), and supramarginal gyrus. For the ImNov task, a large region was seen in the posterior temporal lobe, along with additional sizeable regions of significance in supramarginal gyrus, S1, M1 and angular gyrus, as well as other smaller non-contiguous posterior temporal areas.

Table 2.

Brodmann’s and AAL regions of significant voxels in the GestTool, ImTool and ImNov analyses, intersection of significant tool-related voxels, and intersection of significant imitation-related voxels

				Peak
Gesture condition	AAL regions*	Brodmann areas*	Voxels	z-value	x	y	z
GestTool	Temporal_mid, temporal_inf, occipital_mid, occipital_inf, fusiform, temporal_pole_sup, temporal_sup, insula, hippocampus	37, 21, 20, 19, 39, 48, 22, 38, 36	26221	5.18	−51	−43	−6
	Parietal_inf, postcentral	48, 40, 3	1417	4.16	−23	−28	34
	Frontal_mid, frontal_inf_tri	44, 46, 45, 48	656	3.81	−39	26	33
	Thalamus	48	199	4.32	−27	−17	−4
	Angular, occipital_mid, temporal_mid	39	106	3.39	−39	−60	25
ImTool	Temporal_mid, temporal_inf, occipital_mid, occipital_inf, fusiform, temporal_sup	37, 21, 20, 19, 22, 39, 18, 42	22407	5.22	−41	−65	−7
	Postcentral, parietal_inf, supramarginal, precentral	3, 4, 40, 48, 2, 6, 43	3276	4.39	−43	−12	37
ImNov	Temporal_inf, temporal_mid, occipital_inf, fusiform	37, 20, 21	5844	4.59	−52	−35	−15
	Temporal_sup, temporal_mid, supramarginal, angular	22, 42, 21, 39, 48, 37, 41	1199	3.85	−56	−49	13
	Parietal_inf, supramarginal	40, 2, 3	1153	4.02	−51	−35	38
	Postcentral, precentral	4, 3, 6	746	4.32	−43	−12	37
	Angular	39	198	3.62	−53	−58	33
	Temporal_mid, occipital_mid, occipital_inf	37	192	4.26	−53	−71	−3
Tool-related conjunction	Temporal_mid, temporal_inf, occipital_inf, occipital_mid, fusiform, temporal_sup	37, 21, 20, 19, 39, 22	18514	−	−55	−49	−23
Imitation conjunction	Temporal_inf, temporal_mid, occipital_inf, fusiform	37, 20, 21	5582	−	−52	−56	−19
	Postcentral, precentral	4, 3, 6	717	−	−42	−16	35
	Parietal_inf, supramarginal	40, 2, 3	564	−	−51	−40	37
	Temporal_mid, temporal_sup	22, 21, 37, 39, 42	534	−	−63	−52	10
	Temporal_mid, occipital_mid, occipital_inf	37	192	−	−51	−71	−3

*Brodmann and AAL regions listed in descending order of number of voxels involved within that area. Brodmann areas with <10 significant voxels not shown.

Inf = inferior; sup = superior.

Figure 4.

Maps of the reliability (Z-scores) of the difference between patients with and without lesions in each voxel in (A) gesture to the sight of tools (GestTool), (B) imitation of tool-related gestures (ImTool), and (C) imitation of novel gestures (rendered on the MNI space ch2bet volume). For each of the three maps, voxels rendered in warm/hot colours correspond to Z-scores > 3.02, 2.7, and 2.7, respectively, that reached the FDR-corrected threshold of q < 0.05. Voxels rendered in blue correspond to voxels associated with FDR-corrected thresholds between q = 0.05 and q = 0.1.

As a supplementary analysis, we identified for each subject the proportion of voxels lesioned in the regions that were identified as significant in the group VLSM analysis. As expected, these proportions for each task were inversely correlated with magnitude of behavioural deficits for each task (GestTool: r = −0.59, P < 0.0001; ImTool: r = −0.61, P < 0.0001; ImNov: r = −0.62, P < 0.0001).

Conjunction analyses

To characterize regions common to the two tool-related gesture tasks (GestTool and ImTool), we performed a conjunction analysis of voxels surpassing the FDR-corrected threshold in both analyses (Nichols et al., 2005) by overlaying significant voxels from each analysis on the same template using MRIcron. As Fig. 5 and Table 2 show, there was a large common region in the posterior temporal lobe, encroaching on extrastriate visual cortex and angular gyrus.

Figure 5.

Maps of the conjunction of the voxels reaching the FDR-corrected threshold of P < 0.05 in two of the three gesture conditions. (A) Conjunction of voxels meeting the threshold for the two tool-related tasks (gesture to the sight of tools and imitation of tool-related gestures). (B) Conjunction of voxels meeting the threshold for the two imitation tasks (imitation of tool-related gestures and imitation of novel gestures).

Similarly, to characterize the regions common to the two imitation tasks (ImTool and ImNov), we performed a conjunction analysis of voxels that were significant in both tasks. Figure 5 and Table 2 show that there were several common brain regions, including a large region in the posterior temporal lobe, and smaller regions including M1 and S1, supramarginal gyrus, and angular gyrus.

Posture and kinematic subscores

Next, we examined whether impairments on the three gesture tasks were characterized by differential deficits in the postural or kinematic components of actions. Table 3 shows all significant results of these VLSM analyses.

Table 3.

Brodmann and AAL regions of significant voxels in the posture subcomponent scores of the GestTool and ImTool tasks, in the kinematic subcomponent scores of the ImTool and ImNov tasks, in the posture scores of ImTool controlling for ImNov, and the kinematics scores of ImTool controlling for GestTool

				Peak
Gesture condition	AAL regions*	Brodmann areas	Voxels	z-value	x	y	z
GestTool posture	Temporal_mid, temporal_inf, occipital_mid, occipital_inf, fusiform, temporal_sup, temporal_pole_sup, hippocampus	37, 21, 20, 19, 39, 22, 48, 36, 38	22835	5.83	−56	−45	−5
	Temporal_mid, temporal_inf	20, 21	361	3.08	−64	−25	−16
	Frontal_mid	46, 45	126	3.69	−41	36	39
ImTool posture	Temporal_mid, temporal_inf, occipital_mid, occipital_inf, fusiform, temporal_sup	37, 21, 20, 19, 22, 39, 18, 42	23565	5.42	−41	−65	−7
	Angular	39	140	3.29	−55	−62	34
	Parietal_inf, supramarginal	2, 40, 3	112	3.33	−57	−28	41
ImTool kinematics	Temporal_mid, temporal_inf, occipital_inf, occipital_mid, fusiform, temporal_sup	37, 20, 21, 19, 39, 22, 48	15425	4.91	−56	−38	−16
	Postcentral, parietal_inf, precentral, supramarginal, rolandic_oper, temporal_sup	3, 48, 4, 43, 6, 2, 40, 1, 22	13868	5.12	−41	−13	38
	Temporal_pole_sup, insula	38, 48, 34	191	3.69	−26	6	−20
	Temporal_sup, temporal_inf	22, 21	176	3.74	−64	−53	11
	Frontal_inf_orb, insula	47, 38	122	3.78	−27	25	−14
ImNov kinematics	Temporal_inf, temporal_mid, occipital_inf, fusiform	20, 37, 21	6196	5.35	−42	−43	−2
	Postcentral, precentral, parietal_inf	4, 3, 6, 43, 48	1213	4.45	−41	−13	38
	Parietal_Inf, supramarginal	40, 2, 3	445	3.68	−57	−28	41
	Postcentral, rolandic_oper	1, 2, 3	247	3.68	−61	−12	22
	Temporal_mid, occipital_mid, occipital_inf	37	164	4.5	−50	−71	−2
	Temporal_sup, temporal_mid	22	101	3.39	−56	−49	13
Posture: ImTool controlling for ImNov	Temporal_mid, temporal_inf, occipital_inf, occipital_mid, fusiform	37, 21, 20, 19, 39, 22	14143	4.98	−51	−67	10
Kinematics: ImTool controlling for GestTool	Postcentral, precentral, supramarginal, parietal_Inf	4, 3, 43, 6, 2, 48, 1	2467	4.87	−61	−14	41

*Brodmann and AAL regions listed in descending order of number of voxels involved within that area. Brodmann areas with <10 significant voxels not shown.

Inf = inferior; sup = superior.

Residual analyses

The ImTool task is unique among the three tasks we assessed in that it is both imitative and tool-related. To disentangle the contributions of those two aspects of ImTool, we performed residual analyses in which we controlled for the other two tasks. First, we examined voxels associated with residualized scores for the posture and kinematic gesture components of ImTool controlling for ImNov. In common to both imitation tasks is the requirement to process the visual input of the body movements to be imitated and to transform this input into an analogous plan of action to be performed with one’s own limb. Removing this common variance by way of a residual analysis enables examination of the uniquely tool-related component of the ImTool task. As shown in Fig. 6 and Table 3, the posture component of ImTool controlling for ImNov was associated with a large cluster in the posterior temporal lobe, extrastriate visual cortex, and angular gyrus. The kinematic component of ImTool controlling for ImNov was associated with no significant voxels, even at a relaxed threshold of q = 0.1. This null effect is likely to result from the substantial overlap in voxels associated with the kinematic component in these two tasks (Table 3).

Figure 6.

Maps of the reliability (Z-scores) of the difference between patients with and without lesions in each voxel in (A) posture scores on ImTool controlling for posture scores on ImNov, (B) kinematic scores on ImTool controlling for kinematic scores on GestTool (rendered on the MNI space ch2bet volume). For each of the two maps, voxels rendered in warm/hot colours correspond to Z-scores > 3.4 and 2.9, respectively, that reached the FDR-corrected threshold of q < 0.05. Voxels rendered in blue correspond to voxels associated with FDR-corrected thresholds between q = 0.05 and q = 0.1.

Finally, we assessed the voxels associated with impaired performance on ImTool controlling for GestTool. By removing the variance associated with access to the actions associated with tools, irrespective of whether the input is gestural or a tool, this analysis enabled assessment of the aspects of the ImTool task that are specifically relevant to imitation. There were no significant voxels associated with the posture component of ImTool controlling for GestTool, even at a relaxed threshold of q = 0.1. This null effect is likely to derive from the substantial overlap in voxels associated with the posture component for these two tasks (Table 3). For the kinematic component of ImTool controlling for GestTool, there was a large significant region that included M1, S1 and S2.

Discussion

We performed VLSM with 71 left-hemisphere stroke patients to assess the critical substrates of producing tool-related and imitative gestures. To our knowledge, this is the largest sample of stroke patients with lesion data to have been prospectively tested with reliably-scored gesture tasks (see also Manuel et al., 2013 for a retrospective study using clinical data). While past functional neuroimaging studies have revealed a broadly distributed bilateral network of regions activated in gesture tasks, the present data permitted us to precisely delineate the regions which, when lesioned, result in deficits in tool-related and imitative gesture production. Moreover, separate scoring of the postural and kinematic aspects of these tasks enabled us to assess the role of different brain regions in supporting these components of praxis.

Consistent with predictions derived from previous studies and our own model of the praxis system (Buxbaum, 2001), impairments in all three of the gesture tasks we studied were associated with lesions in left posterior temporal, inferior parietal, motor, and premotor regions. Perhaps more interestingly, the data revealed that tool-related actions, whether imitative or in response to viewed tools, are critically dependent on a large region of left posterior middle and inferior temporal lobe, as well as bordering regions of the occipital lobe. Both imitation tasks, in contrast, whether in response to familiar or novel actions, are dependent on a smaller posterior temporal region, as well as primary and secondary motor and sensory cortices and inferior parietal lobe.

Additionally, we hypothesized that tool-related and imitative actions would depend differentially on the integrity of postural and kinematic subcomponents of actions, respectively. Analyses with raw data as well as more refined residual analyses demonstrated that the postural components of tool-related actions are associated with posterior temporal regions, whereas the kinematic components of imitative actions rely upon regions specialized for motor, tactile, spatial, and body-related processing. There is thus a nearly complete double dissociation in the brain regions specialized for postural versus kinematic components of gestural action. In contrast, the posture component for novel imitation and the kinematic component for gesture to the sight of tool were not associated with significant voxels. We may speculate that the former null effect lies with the novelty of the gestures; there is no ‘stored’ component to the actions being imitated, and it is the stored component that is largely postural and mediated by the posterior middle temporal gyrus. We may speculate, conversely, that the latter null effect was obtained because gesture to the sight of tool relies relatively more reliably on the stored, postural component.

Posterior temporal lobe and extrastriate cortex

The conjunction of the two tool-related tasks (ImTool and GestTool) showed a large area of significance in the posterior temporal lobe, extending into extrastriate cortex and marginally into the angular gyrus. Moreover, the residual analysis of ImTool posture scores controlling for ImNov posture scores indicated the same region of significance, suggesting that deficits in the postural element drive the significance of the tool-related conjunction results. The posterior temporal lobe is viewed as a major component of the ventral ‘what’ pathway, specialized for object recognition (Ungerleider and Mishkin, 1982). Thus, a novel aspect of the present data is evidence that the posterior temporal lobe is critical for tool-related gesture production, and in particular, its postural components. Importantly, evidence for posterior temporal (as well as extrastriate visual cortex) involvement was observed not only in the GestTool condition, but even when no tool was present in the ImTool condition, indicating that the effects are not attributable to demands on tool recognition.

Instead, the data are consistent with an increasingly rich description of the computations that are carried out in a temporal-occipital region that includes area MT+, extrastriate body area, and posterior temporal lobe. Area MT+, also known as area V5 (Watson et al., 1993), is a region that seems to encode the speed and direction of rigid, unarticulated motion, as is characteristic of tools (see also Beauchamp, 2005; Beauchamp and Martin, 2007). Inferior to area MT+ in the posterior inferior temporal sulcus is the extrastriate body area (Downing et al., 2001, 2006), a region playing a role in both the perception and production of hand and arm actions (Astafiev et al., 2004; Orlov et al., 2010; Gallivan et al., 2013). The present data thus add to a growing body of evidence that extrastriate visual cortex plays a role in action processing.

Anterior to these regions lies the posterior middle temporal gyrus, a region with known relevance for the representation of action knowledge. For example, posterior middle temporal gyrus lesions disrupt the ability to match action verbs to videos of gestural actions (Tranel et al., 2003; Kalenine et al., 2010), and posterior middle temporal gyrus is consistently activated in functional neuroimaging studies assessing semantic tool knowledge. Watson and Chatterjee (2012) (see also Watson et al., 2013) have argued that the proximity of posterior middle temporal gyrus to area MT+ facilitates the derivation of abstract representations from visual motion in posterior middle temporal gyrus (Kable et al., 2002, 2005).

The proposal that posterior middle temporal gyrus representations are abstractions of various instances of visual experience is partly consistent with our previous proposals (Buxbaum, 2001; Binkofski and Buxbaum, 2013), but more strongly emphasizes the putatively visual format of the posterior middle temporal gyrus component of distributed gesture representations. As a consequence of repeated experiences, a given skilled gesture can be recognized readily despite variations in viewing angle, gesture size and shape associated with the physical properties of the exemplar of the tool being handled, task goals, and environmental constraints.

The present finding of specialization of the posterior temporal lobe for the production of postural aspects of tool-related actions extends what is known of the role this region plays in mediating semantic concepts for manipulable objects (Kalenine et al., 2009, 2010; see also Almeida et al., 2013; Mahon et al., 2013 for related data). Moreover, the present data suggest that the postural aspects of gesture may be among the ‘features’ upon which a system of tool-related action knowledge is organized (Klatzky et al., 1987; Lee et al., 2013; Watson and Buxbaum, under revision). Thus, the visual similarity of the postural aspects of actions may shape relationships between tool representations in the posterior temporal lobe (see Bracci and Peelen, 2013 for related data).

Finally, there was also evidence for some involvement of the posterior temporal region in kinematic aspects of gesture (Fig. 6B). Parallel to what has been described in theoretical models of the stages of lexical and phonological access in the language domain (Goldrick and Rapp, 2007), the temporal lobe aspect of kinematic representations may contain some degree of abstraction over multiple instances, and thus be less specified in terms of spatiomotor parameters than the parietal lobe kinematic component.

Inferior parietal lobe

Although the inferior parietal lobe was critical to all three tasks, the largest swath of significant anterior and posterior parietal regions was revealed in the two imitation tasks (ImTool and ImNov) and, accordingly, their conjunction, including S1, angular gyrus, and supramarginal gyrus. These data are consistent with a number of functional imaging studies (Muhlau et al., 2005), including a recent activation likelihood estimation meta-analysis showing that inferior parietal lobe is commonly activated in imitation tasks (Molenberghs et al., 2010, 2012).

Extending these data, the present results showed that the kinematic components of the imitation tasks drive the significance of the results in the inferior parietal lobe. These findings are consistent with observed deficits in the spatiotemporal aspects of gesture imitation and recognition in patients with inferior parietal lobe lesions (Halsband et al., 2001; Goldenberg, 2009; Kalenine et al., 2010), and with data showing that lesions to this region may disproportionately affect meaningless imitation (Goldenberg, 2009) as compared with tool-related movements (Manuel et al., 2013; but see Vingerhoets et al., 2009; Peeters et al., 2013; and Sunderland et al., 2013 for evidence for supramarginal gyrus involvement in tool-related actions). Finally, the secondary somatosensory area (S2, also known as Brodmann area 43), a region in the parietal operculum, was also implicated in the kinematic aspects of the imitation tasks. The finding of S2 involvement may be interpreted in light of the role of this region in both input and output processing, i.e. ‘mirror properties’ (Molenberghs et al., 2012; see also Mengotti et al., 2013).

Frontal lobe

In addition to the double dissociation between temporal and parietal significance in postural versus kinematic aspects of movement, respectively, another striking aspect of the present data is a second double-dissociation in the regions of the frontal lobe critical for the three different gesture tasks assessed. Both imitation tasks were associated with significant voxels in M1 as well as premotor cortex (Brodmann area 6). Conversely, the gesture to the sight of tools task (GestTool) showed little to no Brodmann area 6 involvement. Instead, frontal involvement in the GestTool task was more anterior, in middle frontal gyrus and inferior frontal gyrus of prefrontal cortex (Table 2). Moreover, at least some of the significance in these regions was driven by the postural and not kinematic aspects of action (GestTool posture analysis, Table 3).

Left inferior and middle frontal gyrus have long been recognized as important contributors to the performance of gestural movements. Haaland et al. (2000) demonstrated that middle frontal gyrus was the sole region of overlap in patients characterized as having ideomotor apraxia on the basis of an imitation task. On the other hand, Goldenberg et al. (2007a) showed that pantomime of tool use from photographs depends critically on inferior frontal gyrus (Bohlhalter et al., 2011). The previously-noted meta-analysis of functional MRI data from Caspers et al. (2010) indicated that numerous areas in the frontal lobe, including Brodmann areas 44, 45 and 46, were activated in imitation tasks.

Thus, the precise role of different prefrontal regions in imitation and pantomime is unclear. One possibility is that relatively dorsal versus ventral prefrontal regions contribute to the response selection as compared to semantic selection requirements of gestural tasks. A number of investigations report that selection among competing motor responses mainly activates dorsolateral prefrontal cortex, a portion of middle frontal gyrus (Schumacher and D'Esposito, 2002; Schumacher et al., 2003; Ridderinkhof et al., 2004), whereas selection among competing semantic alternatives activates ventrolateral prefrontal cortex in the inferior frontal gyrus (Thompson-Schill et al., 1997; Crosson et al., 2001; Tremblay and Gracco, 2006, but see Thompson-Schill et al., 1998; Fletcher and Henson, 2001; Rowe and Passingham, 2001; Nagel et al., 2008). In the present data, we observed both middle frontal gyrus and inferior frontal gyrus involvement in the GestTool task, but not in the imitation tasks. This may reflect the possibility that although gesture to the sight of tools taxes both response selection (i.e. manipulable objects may be associated with more than one response) and semantic selection (i.e. selecting the tool features relevant for tool-use), the two imitation tasks do not include these selection requirements. On the other hand, the strong interconnectivity of motor, premotor, and prefrontal cortex with the parietal lobe suggests that frontal regions would not be expected to play a specific role only in tool-related movements (Ramayya et al., 2010; and see Groh-Bordin et al., 2009 for evidence from meaningless gesture imitation). Thus, the precise role of prefrontal cortex in gesture tasks requires additional investigation.

Absence of superior frontoparietal and intraparietal sulcus significance

As with most stroke samples, the vast majority of our patients suffered strokes of the middle cerebral artery, which perfuses much of the convexity of the hemisphere, but not the superior parietal or superior frontal lobes. Numerous functional neuroimaging studies have described activation in superior parietal and intraparietal sulcus regions during gesture tasks (Johnson-Frey et al., 2005b; Kroliczak and Frey, 2009; Menz et al., 2009; Agnew et al., 2012). Given that the subset of patients with lesions in these regions was relatively small (8–15 subjects), our analyses may have lacked sufficient power to detect significant effects in the relevant voxels. At a relaxed statistical threshold, we observed several small sub-threshold regions in the anterior intraparietal sulcus in the GestTool task, as well as in a region including superior M1 and superior Brodmann area 6 in the residual analysis for the kinematics of ImTool controlling for GestTool. These data are consistent with the putative role of the dorso-dorsal visual processing stream in the spatiomotor aspects of movement (Buxbaum, 2001; Rizzolatti and Matelli, 2003), and with the fact that superior frontoparietal regions are frequently affected in patients with corticobasal syndrome, who exhibit a form of limb apraxia that tends to affect meaningless more than meaningful gestures (Buxbaum et al., 2007). Future investigations with even larger samples will be required to confirm these subthreshold results.

The distributed visuokinaesthetic engram: implications for theoretical models of gesture production

Although several accounts of gesture production distinguish between representational (or semantic) and ‘on-line’ or ‘production-related’ aspects of action in the left hemisphere (Roy and Square, 1985; Gonzalez Rothi et al., 1992), the present data are the first, to our knowledge, to provide direct evidence that postural and kinematic components of gestural action are neuroanatomically distinct. We proposed that the kinematic component is most sensitively assessed with imitation of novel, meaningless gestures, and the semantic component with pantomimed gesture production in response to viewed tools. We observed a dorsal versus ventral distinction between these action components, with imitation of novel gestures relying strongly on the inferior parietal lobe, and gesture to the sight of tools dependent upon the posterior temporal lobe. Moreover, we also showed that it is the kinematic aspects of gestural action, in particular, that rely upon inferior parietal and frontal regions, as compared with the postural aspects of (meaningful) action, which are dependent upon the left posterior temporal lobe. These data permit us to refine the claim that there are ‘direct’ and ‘indirect’ routes to action (Rothi et al., 1992; Rumiati and Humphreys, 1998) with evidence that the two routes are in fact aspects of a distributed system supported by relatively more dorsal versus ventral regions, respectively. Beyond this, an important contribution of the present data is the demonstration that aspects of processing required for gesture production are subserved by regions traditionally conceived as part of the ‘ventral’ visual processing stream.

The proposal that gesture posture representations are based upon abstracted patterns of visual feature co-occurrence is similar to several accounts of the organization of the visual semantic system in the posterior temporal lobe (McNorgan et al., 2007; Dilkina and Lambon Ralph, 2012). By extension, findings that tool-related actions depend on the posterior temporal lobe, and that posture scores drive posterior temporal significance, suggest that abstraction over multiple instances of experience in seeing tool actions of the self and others may be derived at least in part from visual motion information (Watson and Chatterjee, 2011). In contrast, the current details of an action’s timing and trajectory, fleshed out on-line on the basis of environmental constraints, may be relatively more spatiomotor in format. This differentiation, in turn, suggests that an important process occurring between the temporal and parietal lobe is gradual transformation of information from one format to another. Thus, at least in the case of tool-related actions, kinematic processing in the parietal lobe may be conceptualized as a translational capacity that enables transformation of a visual-motion representation in the posterior temporal lobe to one closer to readiness for motor output as it passes through the parietal lobe (see Pisella et al., 2006 for a related proposal).

We suggest that the component that the inferior parietal lobe contributes is the computation of movement plans in terms of dynamic changes in the relative spatial positions of body parts needed to reach a goal configuration. Such movement planning in ‘intrinsic’ spatial coordinates (Rosenbaum et al., 2001; Jax et al., 2006; Orban de Xivry et al., 2011; Parmar et al., 2011; Brayanov et al., 2012) is viewed as a preparatory step toward specification of the precise muscles that must participate in the movement. Unlike the visual tool-use representations subserved by the posterior temporal lobe, the computation of intrinsic coordinate control relies upon somatosensory processing. Consequently, patients with deficits in intrinsic coordinate control are abnormally reliant upon visual feedback of their own limb position in posture imitation tasks (Jax et al., 2006; see also Haaland et al., 1999; Buxbaum et al., 2005b).

Role of left and right hemisphere regions in gesture production

Finally, as noted above, many functional neuroimaging studies have demonstrated that numerous bilateral regions are activated in gesture production tasks, regardless of the hand used. There are several explanations for the discrepancy between these findings and those from the patient literature, including those reported here, which suggest that imitation and pantomime may be severely disrupted by lesions to the left hemisphere alone. One possibility is that bilateral activation reflects the potential for acting with both hands despite the unimanual nature of most tasks performed in the scanner. A second, related possibility is that right hemisphere activations may reflect strong connectivity between homologous left and right hemisphere regions subserving action (see Culham et al., 2006 for a similar argument). The present data suggest that the observed right hemisphere activations may reflect a relatively subtle (perhaps modulatory) role in gesture production, and underscore the continued importance of detailed studies of lesioned patients in advancing knowledge of the functional architecture of the human brain.

Glossary

Abbreviations

AAL

Automatic Anatomic Labelling

GestTool

gesture to the sight of tool

ImNov

imitation of novel gestures

ImTool

imitation of tool-related gestures

VLSM

voxel-based lesion–symptom mapping

Funding

Supported by NIH R01-NS065049 to Laurel Buxbaum.

Supplementary material

Supplementary material is available at Brain online.