What do brain lesions tell us about theories of embodied semantics and the human mirror neuron system?

1. Introduction

Groundbreaking work from the early 1990s reported that a distinct group of neurons in area F5 of the macaque premotor cortex fire both when the monkey performs an action (i.e., execution) and when it observes someone else performing it (i.e., observation) (di Pellegrino et al., 1992). Much work since then has attempted to discover a similar ‘mirror neuron system’ in humans (HMNS). According to one version of ‘embodiment theory’, the sensorimotor regions of cortex normally involved in action execution are also recruited during action observation, planning and mental imagery (e.g., Rizzolatti and Craighero, 2004; Rizzolatti and Luppino, 2001; Rizzolatti et al., 2001, 1996; Gallese et al., 1996). A related theory – embodied semantics – is specific to humans, and suggests that the conceptual representations accessed during linguistic processing may also recruit those same sensorimotor regions (e.g., Kemmerer and Gonzalez-Castillo, 2010; Gallese and Lakoff, 2005; Pulvermüller, 2005; MacWhinney, 1999).

In a strict interpretation of this view, the same regions of premotor cortex should be recruited in all three conditions: 1) executing actions, 2) observing actions, and 3) processing/comprehending words referring to actions. More lenient versions of the theory might predict partially overlapping – but not identical – regions comprising a general motor-language network used across a number of linguistic tasks. Support for these more ‘lenient’ or ‘inclusive’ interpretations comes from reports of MNS-related activity in regions outside the motor and premotor cortices, including Broca’s area, somatosensory cortex, supplementary motor cortex, middle cingulate, temporal cortex, cerebellum, and the inferior/superior parietal lobule (e.g., Gazzola and Keysers, 2009; Kemmerer and Gonzalez-Castillo, 2010; de Zubicaray et al., 2008; Postle et al., 2008; Pobric and Hamilton, 2006; Möttönen et al., 2004; Tremblay et al., 2004; Tranel et al., 2003; Nishitani and Hari, 2002, 2000; Buccino et al., 2001; Rizzolatti et al., 1996).

While the macaque and human research to date have granted considerable credibility to the mirror properties of the first two conditions (i.e., executing and observing), the third condition – linguistic comprehension, or embodied semantics – has been more controversial. This is understandable, since this relatively recent addition to the theory requires experimentation in humans using various functional imaging methods. This type of work is arguably more complex and open to interpretation than work done with non-human primates.

Several recent fMRI, ERP and TMS studies testing embodied semantics presented healthy participants with lexical stimuli (either single words or sentences) referring to different body parts (mostly the hand, mouth and foot) (e.g., Pulvermüller et al., 2009; Esopenko et al., 2008; Kemmerer et al., 2008; Aziz-Zadeh et al., 2006; Buccino et al., 2005; Tettamanti et al., 2005; Hauk et al., 2004; Shtyrov et al., 2004). Activations for these stimuli were reported in somatotopically-organized regions. In other words, the pattern of peak activations for each effector followed, to varying degrees, the somatotopic organization of the motor homunculus first described by Penfield and Boldrey (1958). That is, moving along the motor strip in a ventral to dorsal direction, the mouth region is located ventrally, followed by the hand region, and then the foot region at the most dorsal end (Catani and Stuss, 2012).

However, critics of these findings suggest that the somatotopically-distributed activations reported in many of these studies are less than exact, and that a good match for all three effectors across tasks is rarely reported (Fernandino and Iacoboni, 2010; Turella et al., 2009). Many have also argued that the neurobiological boundaries of the primary and premotor cortices, as well as these regions’ somatotopic organization, have yet to be sharply defined (Kemmerer and Gonzalez-Castillo, 2010; Schubotz and von Cramon, 2003; Sanes and Schieber, 2001). Another major criticism is the noticeable lack of studies testing all three conditions in the same set of participants (Kemmerer and Gonzalez-Castillo, 2010; Dinstein et al., 2008). In response to this criticism, two recent studies that indeed tested all three conditions on one set of participants found that activations elicited by action word meaning representations did not match the activations observed for execution and observation (de Zubicaray et al., 2008; Postle et al., 2008).

Another critical consideration is that the evidence thus far only suggests that the putative human MNS may participate in and enhance language comprehension, but it does not confirm whether this system is necessary or sufficient to support such processing (Fischer and Zwaan, 2008). In other words, fMRI, PET and TMS studies can only tell us which brain regions participate in carrying out a given task. Lesion data, on the other hand, can help us understand which regions are in fact necessary for the task to be completed (Kemmerer and Gonzalez-Castillo, 2010; Mahon and Caramazza, 2009).

To date, very few studies have used lesion data to explore this area of inquiry. One set of studies with apraxic patients found that being unable to appropriately manipulate objects is not accompanied by an inability to recognize those same objects (Mahon et al., 2007; Negri et al., 2007; Rosci et al., 2003). In other words, at least for one group of patients, comprehension and execution are not subserved by the same brain regions. In our work with aphasic patients, we have chosen to focus on the comprehension component of the putative HMNS. In a previous study from our group, patients and controls showed a double dissociation on the ability to name and repeat action/object stimuli associated with manipulation; namely, patients were less accurate on manipulation-associated relative to neutral stimuli, while controls were relatively more accurate on the manipulation-associated stimuli (Arévalo et al., 2007). In that study, 60% of the patients who showed this significant ‘manipulability effect’ had a lesion in motor cortex or the nearby white matter.

In the current study, we tested the notion that a simple lexical semantic comprehension task would recruit the semantic embodiment component of the HMNS. We predicted that lesions due to stroke in key sensorimotor regions would impact patients’ accuracy in matching pictures and words associated with body parts, and extended our earlier work on manipulability by including stimuli associated with the mouth and foot.

2. Methods

3. Results

First, we analyzed the behavioral data from the group of healthy, age-matched control participants (n = 10) to ensure our task was appropriate for older subjects and that accuracy rates did not differ for body-related versus neutral items. The control group did not differ in accuracy for any of the comparisons: Hand versus Hand Neutral, t(9) = 1.14 (100% vs 99%), Mouth versus Mouth Neutral, t(9) = 1.73 (96% vs 98%), and Foot versus Foot Neutral, t(9) = −.58 (100% vs 99%), all ps > .05. Please see Table 2 for each patient’s score on all 6 conditions.

4. Discussion

There has been much discussion in recent years over whether an adequate counterpart to the macaque MNS exists in humans. The most ‘human’ component of this putative system – embodied semantics – has been tested extensively with a myriad of linguistic tasks designed for several types of functional neuroimaging experiments. These experiments have reported activation in response to action-associated linguistic stimuli in several regions normally involved in the execution and observation of those actions (e.g., Pulvermüller et al., 2009; Esopenko et al., 2008; Kemmerer et al., 2008; Aziz-Zadeh et al., 2006; Buccino et al., 2005; Tettamanti et al., 2005; Hauk et al., 2004; Shtyrov et al., 2004). Critics have countered such conclusions by arguing that the reported overlap of such regions across tasks is inconsistent, within as well as across different studies (Fernandino and Iacoboni, 2010; Turella et al., 2009; de Zubicaray et al., 2008; Postle et al., 2008).

While fMRI, PET and TMS studies can help identify regions which may participate in such tasks, lesion data can help us determine whether certain brain regions are necessary for such processing to take place. In the current study, 27 patients with left-hemisphere lesions due to stroke were asked to match picture and word stimuli typically associated with the use of the hand, mouth, foot, or no body part at all. In Analysis 1, we grouped patients based on the presence or absence of lesions in four key sensorimotor regions (BA 4/6, 1/2/3, 21, and 44/45). For all four comparisons, patients with lesions in each of the regions were less accurate on Foot versus Foot Neutral items. In addition, patients with lesions in BA 1/2/3 showed a trend for reduced accuracy on Mouth versus Mouth Neutral items. Therefore, this analysis revealed that lesions in a range of key areas previously associated with the putative HMNS can lead to a deficit for understanding foot-related concepts.

It is unclear why the foot stimuli elicited a stronger effect than the other two body parts tested. One possibility is that ‘Foot’ items are less salient than items referring to other body parts, or perhaps not as easy to depict in 2D drawings. However, these arguments cannot be reconciled with the fact that the ‘Foot’ items in this study did not differ from items in the other categories with respect to imageability, frequency, objective visual complexity or item difficulty.

With respect to the mouth, there was a trend for lesions in BA 1/2/3 to result in lower accuracy for Mouth relative to Mouth Neutral stimuli, but the effect was not as striking as the Foot item comparison. Some authors have distinguished between types of mouth-related actions as being either communicative or ingestive, each of which may carry its own set of evolutionary implications (Ferrari et al., 2003; Möttönen et al., 2004). Our set of stimuli did not allow us to match items based on this distinction.

Finally, what about the hand? In a previous study testing only manipulability and using stimuli from the same picture corpus, left-hemisphere patients were less accurate at naming pictures and repeating words referring to manipulable stimuli, while control participants were more accurate at naming those same items (Arévalo et al., 2007). However, that study differed from the current study in that the task required the oral production of a word (not only comprehension) and no bar pressing was involved. Previous studies have suggested that responding to stimuli referring to a specific effector (e.g., hand) by using that effector (e.g., manually pressing a space bar) can influence performance, but results have been mixed. While some authors have found an interference effect (i.e., slower RTs to hand items relative to foot items when responding with the hand; Sato et al., 2008), others have found a facilitation effect (e.g., Pulvermüller et al., 2001; Scorolli and Borghi, 2006). Although our study focused on accuracy rather than RT, we cannot entirely rule out the possibility that manually pressing a space bar could have facilitated responses to the Hand items (relative to the neutral items), thus reducing a difference in performance on Hand versus Hand Neutral items.

One possible way to avoid this problem is to have participants use the congruent effector for each body-associated set of items (i.e., respond to hand items with the hand, to mouth items with the mouth, and to foot items with the foot). However, this design choice was not possible for the current study, given that some patients had limited mobility, in addition to deficits in attention and memory. It is plausible that such inherent differences between hand, mouth and foot stimuli could be driving some of the results obtained with the current task, and future studies will consider these issues.

Clearly, dividing patients according to presence or absence of a lesion has implicit limitations. For example, it cannot tell us whether lesions of different sizes or lesions confined to specific subregions of the target area would have a greater or lesser impact on performance. Therefore, in Analysis 2, we used VLSM (Bates et al., 2003) which analyzes data on a voxel-by-voxel basis rather than restricting comparisons to specific regions (BA or otherwise). Under a strict statistical correction procedure, only the Hand and Foot maps had significant voxels. These regions included left BA 6 (premotor cortex), BA 21/22 (middle and superior temporal cortex), BA 44/45 (posterior inferior frontal gyrus or Broca’s area), BA 42 (auditory association cortex) and BA 47 (inferior frontal gyrus). In addition, the Hand map included BA 38 (temporopolar cortex) and BA 41 (primary association cortex), while the Foot map included BA 4 (primary motor cortex). Therefore, two of the effector-associated maps revealed significant voxels in most regions tested in Analysis 1 (with the exception of BA 1/2/3), as well as in some additional left frontal and temporal regions. There was no evidence for any type of somatotopic organization of the effectors.

It is unclear why BA 1/2/3 did not appear as an area of significant involvement in the VLSM maps. It is possible that voxels in this region did not survive the strict statistical threshold we established for our analyses. The same threshold issue might be at play when considering the third body map, Mouth. Alternatively, the difference across the three maps might reflect the quality of the stimuli or the nature of the effectors themselves, as discussed above. It is important to point out that VLSM maps are limited by the power available in certain regions of the brain relative to others. This limited our ability to identify lesions in very dorsal ‘foot’ regions, relative to more ventral ‘hand’ and ‘mouth’ regions. However, as illustrated in the power map shown in Fig. 7, there was adequate power to detect differences relating to HMNS involvement in key regions previously implicated in this system.

In sum, our results suggest that there is interaction between motor networks and the language network in humans, and that these associations do not seem to be confined to a particular region in premotor/motor cortex. Rather, motor-language areas appear to be spread over a range of different cortical regions, which in this study can only be confirmed for the left hemisphere. Damage to certain regions of this language-motor network will not completely block patients’ ability to process motor-associated concepts, but can result in lower relative accuracy on some effector-associated stimuli. One issue to consider is that this task as it stands cannot tell us whether the degree of the engagement of motor regions reflects basic semantic processing in and of itself, or whether it is due to some post-comprehension cognitive operations, such as motor imagery (Boulenger et al., 2006; Kemmerer and Gonzalez-Castillo, 2010; Tomasino et al., 2008). Perhaps motor imagery is a strategy used by some people but not others, and this type of individual variation may yield different results and hence different interpretations, especially when relying on group results. Future work could focus on disentangling such possible processing strategies with specially designed tasks sensitive to such distinctions.

Our current results have implications for rehabilitation work as well. Motor imagery, along with linguistic and motor tasks, have been of interest to investigators who are working on developing rehabilitation therapies for stroke patients. These groups’ aim is to facilitate language through motor tasks (or vice versa), by taking advantage of the connections between the language and action systems in the brain (e.g., Sharma et al., 2009, 2006; Pulvermüller and Berthier, 2008; Buxbaum et al., 2005; Johnson-Frey, 2004; Catani et al., 2012).

Although there is abundant evidence for the existence of a MNS in the macaque and a similar counterpart in humans, we suggest that the third component proposed for the human MNS – embodied semantics – is not controlled by the same regions that subserve the execution and observation of actions. In agreement with several studies, our lesion data suggest that a number of regions in premotor and motor cortex, as well as additional regions in frontal and temporal cortex, play a complementary rather than central role in processing words referring to motor-related concepts.