Sixteen AD patients and an equal number of aMCI subjects and aged healthy subjects (NE) were enrolled and performed an fMRI study and a neuropsychological assessment.

Table 1 summarizes the demographic data of the three groups. Age, education level, and gender was not significantly different (ANOVA). Probable AD was diagnosed according to NIA-AA criteria (McKhann et al., 2011). aMCI subjects were diagnosed according to Petersen's criteria (Petersen et al., 1999, 2001). Only MCI subjects with significant atrophy of at least one hippocampus were included in the study (see later). The following is the exclusion criteria: patients unable to understand and/or follow instructions; severe attentional deficits, untreated psychiatric disorders, joint deformity of arthritic origin, visual or hearing deficits, previous stroke or other neurological disorders, contraindications to MRI. Persons with MCI or dementia were consecutively recruited from the Don Carlo Gnocchi Foundation Service for Cognitive Disorders and Dementia. Patients were diagnosed after taking their clinical history and carrying out medical and neurological examinations, routine blood tests and neuropsychological assessment. All patients also underwent either brain CTs or MRIs to evaluate the vascular lesion load and to exclude rarer causes of dementia (e.g., tumors, hematomas, etc.). The control group consisted of spouses or hospital volunteers.

This study was carried out in accordance with the Declaration of Helsinki. All subjects gave written informed consent. The protocol was approved by the Don Gnocchi Foundation ethical hospital committee.

During the fMRI scanning, participants were asked to perform 2 block-designed runs (A-B structure): in one run, they were instructed to observe (observational run−s) and in the remaining run to execute hand grasping movements (motor run), according to a paradigm described in Cabinio et al. (2010). Briefly, in the observation run, all subjects viewed movies of hand grasping different objects (see Figure 1A). In the motor run, subjects observed visual stimuli consisting of objects oriented in order to be grasped with the right hand and they were asked to perform a grasping movement appropriate to the shape of the object (see Figure 1B; subjects were asked to perform movements as if they had the observed object actually in their hand, therefore the execution condition was a mime of the action). The motor act was performed continuously during the observation of the image of each object, with a frequency of about 1 Hz and a total number of 27 objects. Examples of objects were a cup, scissors, a screwdriver, and a torch. Figure 1 shows some visual stimuli presented to the subjects in each experimental condition. The decision to use different types of grasp (both observed and executed) was driven by the need to recruit the highest percentage of mirror neurons and thus increment the BOLD signal (Ehrsson et al., 2000; Iacoboni et al., 2005; Cabinio et al., 2010; Marino and Ricciardelli, 2017). Before the fMRI experiment, the participants were verbally instructed to execute the movements as if the object was close to their hand, and to carry out the task only with the hand and the wrist: repeating this action, until the appearance of the next picture.

Participants were trained outside the scanner for 15 min before data acquisition. During the training session, subjects were instructed to keep their gaze on the fixation point for the entire duration of the experiment, and to execute grasping movements about once every 2 s. If the training task (that was conducted with different parallel versions), was not properly executed by the subjects, they were not included in the study.

We used an MR-compatible visual system to present the stimuli (VisuaStim Digital system, Resonance Technology Inc.). The use of E-Prime software (E-Prime 2.0 Psychology Software tool, http://www.pstnet.com) ensured exact timing of prompts during MR acquisition. The performance was visually checked by the examiner during the execution of the task. The examiner maintained accuracy by reporting all the tasks correctly performed. The design was fully randomized (both blocks and runs). All the stimuli were projected onto a screen at the front side of the magnet bore and subjects viewed them through a mirror attached to the head coil.

The neuropsychological assessment was centered on functions considered to be linked to the MN system, such as language (particularly action naming; Kemmerer et al., 2012) and empathy; we also tested memory (as memory deficits characterize AD and aMCI) and attention/executive functions (abilities which are associated to frontal lobe functioning) (see Table 2 for test list). Subjects were all right-handed in according to the Edinburgh Handedness Inventory (Oldfield, 1971).

The ANCOVA analysis showed statistically significant differences in bilateral hippocampal volume between the NE and the MCI groups, as well as between the Control and the AD groups. No volumetric differences were observed between the MCI and the AD groups. Differences were found in both Left and Right Hippocampi (see Table 3).

All subjects correctly performed the fMRI runs. Given the premise that regions included in the MN network must be activated both in movement execution and in movement observation, to identify cortical areas involved in this network we performed conjunction analysis between activation of the two conditions M and O. This analysis was carried out separately for the three groups of NE, MCI, and AD subjects. All regions of significantly increased activation (pFWE < 0.05 at cluster level with K = 50 contiguous voxels) are summarized in Table 4 and illustrated in Figure 2, each group separately.

In NE subjects the conjunction analysis showed a statistical significant [t contrast 3.41; k = 50; pcluster (FWE-corr) < 0.05] activation of the bilateral fronto-parietal network (left>righ side) formed by the frontal cortex (Precentral Gyrus, PrCG, BA6) and the parietal cortex (Inferior Parietal Lobule, IPL, BA40) along with activation of the right STG (BA22/42) and the right Fusiform Gyrus (FuG).

In MCI subjects the same conjunction analysis [t contrast 3.41; k = 50; pcluster(FWE-corr) < 0.05] revealed the activation of the frontal-parietal network (right < left side) including also the left Inferior Frontal Gyrus (IFG, BA44), but activation of BA40 was not as strong as in NE participants. A bilateral cerebellar activation was also detected (See Table 4).

Finally, in AD patients the conjunction analysis [t contrast 3.41; k = 50; pcluster (FWE-corr) < 0.05] showed a statistical significance recruitment in the bilateral FuG and in the left cerebellum. It found no activations in the fronto-parietal network.

An additional analysis to investigate potential differences in brain activation between the three groups for the M or O condition was carried out separately, and did not reveal any significant group differences. The Figure 3 and Table 5 report the statistical results of the main effect in the O condition [ANOVA, F contrast 14.74; K = 50; p(FWE-corr) < 0.05] and in the M condition [ANOVA, F contrast 13.48; k = 50; p(FWE-corr) < .05].

The results of AD patients were worse than NE subjects in all neuropsychological tests. These tests comprised tests of action naming and empathy, functions attributed to MN network. The only exception was the object naming test (probably due to a ceiling effect). aMCI subjects were significantly different from NE participants only in the episodic memory test, the FCRST, and in semantic fluency (see Table 6 for results).

In the NE group, the conjunction analysis (fMRI activation recorded during observation of hand grasping compared to overt execution of grasping movements) showed the activation of a bilateral, though strongly left-lateralized, fronto-parietal network (PrCG, BA6; IPL, BA40). The fronto-parietal network activated in our study is considered part of the human MN network (Rizzolatti and Craighero, 2004), given the established homology between the BA6 (human) and the ventral premotor cortex (monkey), the BA40 (human), and PF-PFG (monkey) parietal cortex (Cook et al., 2014). We also found activation of the STG (BA22/42) considered an area which is strongly associated with activation of MN in humans (Aziz-Zadeh et al., 2006).

fMRI and neurophysiological investigations aimed to study the MN network in human were mostly performed on young or adult subjects. We found that the MN network is largely preserved in aging. Our results are in accordance with those of Léonard and Tremblay (2007), who analyzed corticomotor facilitation associated with observation, imagery and imitation of hand actions in younger and older adults by monitoring changes of motor evoked potentials elicited in hand muscles by TMS. They found that corticomotor facilitation in association with covert action execution was largely preserved with aging, although with a loss in selectivity for activated muscles. Unlike the original study of Cabinio et al. (2010), in this study (where we exactly replicated their procedure), we noted a loss of selectivity, as the activation of mirror areas was more bi-lateralized than in the younger subjects. This could be explained by a specific mode of brain activation in aging. In fact, fMRI recording in younger, middle-aged and older participants performing the same unilateral hand movements (Fang et al., 2005) showed stronger premotor/motor cortex activity in the contralateral hemisphere in the “older” group when compared to the younger and middle-aged groups. Therefore, together with other authors, we concluded that the older brain requires larger areas to achieve the same task (Fang et al., 2005; Chow et al., 2017), a hypothesis that could also explain the results obtained during a task inducing a motor resonance mechanism (Rizzolatti, 2005). However, in order to exclude any bias, due to differences in the scanners, it would be worthwhile recruiting a young subject pool to definitively confirm the data.

Concerning the MCI subjects, we found a different pattern at the conjunction analysis in comparison with the NE subjects. The MCI and the NE groups showed similar activation of frontal and parietal areas. However, in the MCI group the left IPL (parietal BA40) was activated at a lesser extent than in the NE group and neither the right IPL nor the STG (BA22/42) were activated. Moreover, only the MCI participants exhibited activation of the left frontal area IFG-BA44. Therefore, even if the MN network is preserved in MCI, it is mainly recorded toward frontal areas. This suggests a greater resilience to the aging process of the MN areas located in the anterior part of the brain vs. posterior areas, and maybe consecutively the capacity of these areas to take on a part of the “work” done by the posterior MN regions. Interestingly, we did not observe any activation of the fronto-parietal network in AD participants. Altogether, our results suggest that the MN network is progressively affected in neurodegenerative cognitive decline following a posterior-anterior gradient. The hypothesis of a posterior-anterior decay in Alzheimer-type degeneration agrees with recent data obtained in MCI using EEG and MRI recording (Moretti, 2016). In this investigation, alpha3/alpha2 frequency power ratio, considered as a predictor of conversion into AD (Moretti et al., 2011) as well as cortical thickness at MRI were computed. Three MCI groups were obtained considering increasing tertile values of alpha3/alpha2 ratio. From this morphological MRI and EEG parameter data the author inferred that MN are impaired in prodromal AD. Conversely, we directly measured the activation of human MN with a specific fMRI task, thus giving a functional support to the hypothesis for the MCI and the AD group.

While unexpected, the present results raised several questions. How to explain the cerebellar activation recorded in AD patients and to a lesser extent in MCI subjects? Based on models incorporating the MN system in social learning, the STG-IPL-IFG circuit would create the motor representation available for imitation, starting with the visual input (Iacoboni et al., 1999; Oh et al., 2012). During the imitation phase, the overall processes are equivalent to those of the observation phase, the only difference is that, in parallel with the IFG to the IPL pathway, neural drive is sent to the musculoskeletal system through M1 to perform the action (Oh et al., 2012; Bassolino et al., 2013). It has been suggested that the cerebellum provides the prediction error for the IFG and the IPL to adjust internal models (Miall, 2003; Oh et al., 2012). In this context, the AD patient activation of cerebellar neurons involved in the MN network would correspond to an attempt to compensate for damage of prefrontal and parietal areas. This mechanism would already be present in aMCI. The activation of the cerebellum as a substitute of cerebral cortex alteration has also been suggested to enhance motor control and motor learning in functional recovery from a stroke following mirror therapy (Arya et al., 2017). In conclusion, cerebellar activation in AD might simply reflect the response of this neural structure to visual inputs during guided limb movements (Liu et al., 2003).

The alteration of the MN motor circuitry in AD patients raises an interesting question related to the link between action and cognition and supports an interesting direction of research. Precisely, it allows the interpretation of AD—as other neurodegenerative diseases (Bak, 2013; Eisen et al., 2014) in the light of the action-perception coupling hypothesis. In contrast to traditional views (considering motor mechanisms as a slave system of cognition), it is recurrently postulated that a part of cognition results from the coupling between action and perception representations, and corresponds to implicit action simulation (or “motor resonance”) instead of explicit recall of abstract symbols (Jeannerod, 1994, 2001, 2006; Jeannerod et al., 1995; Boulenger et al., 2008; Pulvermüller and Fadiga, 2010). Neuropsychological tests exploring cognitive functions that considered to be tightly linked to the MN system (language and empathy) were normal in aMCI subjects. The preservation and maybe hyperactivation of the anterior MN network presently seen in aMCI, might supplement the initial decay of the posterior part of the neural circuit, thus preserving cognitive performance. The only “language” test impaired in aMCI was semantic fluency, but this impairment could be due to semantic memory damage, rather, than true language impairment. As an alternative explanation, the preservation/hyperactivation of the anterior MN network could not be sufficient to support performance in this kind of task because it is highly dependent on parietal areas (Seghier, 2013). In contrast, in the AD group the MN network appears to be deficient as task performance in both language and empathy tests deteriorates.

The present study is limited by several factors: the number of subjects tested and the spatial and temporal resolution of our measurements (see Oosterhof et al., 2011, 2013; Möhring et al., 2014; Thibault and Raz, 2016); the fact that in the execution condition the hand movements were only visually controlled by the examiner. In particular, our results should now be replicated on the basis of data collected with different methods dedicated to record MN activity, as motor evoked potentials induced by TMS (Gangitano et al., 2004; Fadiga et al., 2005), or the reduction of magnitude of the mu rhythm at EEG (Fox et al., 2016 for a review). Furthermore, we have studied task-evoked neural activity during action observation and execution. Schilbach et al. (2016), in a recent study, have demonstrated differential patterns of dysconnectivity in MN and mentalizing networks in schizophrenia. Further research, therefore, needs to investigate functional organization of self-related brain networks during a “resting state.” This clinically available measure of functional connectivity could provide further knowledge of alterations of neurofunctional systems in AD.

The results of our study have possible implications in rehabilitation. Recently, the use of action observation as a supplementary therapeutic tool for patients who have had strokes, in order to stimulate brain plasticity and obtain positive functional results, has been reported (see Carvalho et al., 2013 for a review on this topic). This has been extended to motor deficits of children with Cerebral Palsy (Sgandurra et al., 2011; Buccino et al., 2012; Bassolino et al., 2015) and patients with Parkinson's disease (Buccino et al., 2011). The concept of motor cognition, if verified, would provide a clinical support to cognitive stimulation based on the motor resonance mechanism. However, the efficiency of such training is only conceivable if the MN system remains partly achievable. So far, the only published research of this kind of cognitive training via the stimulation of the MN in people with AD, showed improvements in attention and facial recognition (Eggermont et al., 2009) whereas another research showed negative results (Caffarra, 2016, unpublished, oral communication at XI Sindem National Congress, Italy). However, this last study did not test MN system integrity, that is supposed to be stimulated in the training program. As our data showed a malfunction in the MN network in the case of AD, our study explains why that intervention obtained negative results. A rehabilitative intervention based on the MN system would be better implemented at the MCI phase.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.