System neuroscience: Past, present, and future

Summary

In this review, we discuss first the anatomical and lesion studies that allowed the localization of fundamental functions in the cerebral cortex of primates including humans. Subsequently, we argue that the years from the end of the Second World War until the end of the last century represented the “golden age” of system neuroscience. In this period, the mechanisms—not only the localization—underlying sensory, and in particular visual functions were described, followed by those underlying cognitive functions and housed in temporal, parietal, and premotor areas. At the end of the last century, brain imaging techniques were developed that allowed the assessment of the functions of different cortical areas in a more precise and sophisticated way. However, brain imaging tells little about the neural mechanisms underlying functions. Furthermore, the brain imaging suffers from 3 major problems: time is absent, the data are merely correlative and the testing is often not ecological. We conclude our review discussing the possibility that these pitfalls might be overcome by using intracortical recordings (eg stereo‐EEG), which have millisecond time resolution, allow direct electrical stimulation of specific sites, and finally enable to study patients while freely moving.

1. INTRODUCTION

Neuroscience today is an extremely diverse field. Broadly speaking, it consists of 2 major, at large independent, subfields: the cellular and molecular neuroscience, and the system neuroscience. System neuroscience, which is the topic of this review, studies how neurons work together in complex networks giving rise to integrative functions such as vision, audition, somatosensory sensation and actions, as well as to higher order functions such as the understanding of actions, intentions and emotions of others, and empathy.

In this review, after a historical prelude in which we will discuss the fundamental experiments forming the scaffold of system neuroscience, we will concentrate on the limitations of brain imaging techniques and their impact on system neuroscience. As a case of study, we will focus on the characterization of the mirror mechanisms through the different methodologies. Subsequently, we will discuss how intracranial recording techniques (like stereo‐EEG) may contribute the missing piece of information necessary to unravel the mechanisms underlying social and cognitive functions.

2. BASIC ANATOMICAL AND FUNCTIONAL DATA

Anatomy represents the foundation of system neuroscience, whose development has always been tightly linked to the mapping of cerebral cortex. There is a large number of anatomical brain maps, and worth mentioning those of Vogt and Vogt,1 Von Economo2 and Sarkissow.3 However, probably never in the history of neuroscience a single illustration has been as influential as the human cytoarchitectonic map published by Brodmann.4 This map presents the segregation of the cerebral cortex into 43 areas, as visible in cell body‐stained histological sections. Still now, Brodmann maps are extensively used in clinical neurology and in neuroimaging.

Anatomical atlases allow the localization of brain functions. This has been a successful joint enterprise of physiologists and neurologists of the first half of the twentieth century. The studies were mostly based on lesions in animals or on careful descriptions of clinical symptoms following cortical lesions, the so‐called experiments of nature. Among a series of fundamental discoveries, it is worth mentioning the pioneer study by Broca5 on language localization, and the description of the complex organization of visual pathways in humans made by Inouye,6 the Japanese physician who performed careful studies of the wounded soldiers during the Russian‐Japanese war.

At the beginning of the Second World War, the neurologists were able to localize the cortical areas related to visual, somatosensory, acoustical, and motor functions. In addition, some progresses were made in identifying the areas (or the networks) underlying complex psychological functions like estimation of distance and impaired depth perception,7, 8 space perception,9, 10 motor engrams for complex motor actions,11 and the network involved in perceptual and expressive language.12 There was, however, something lacking: No hint was provided on how the different cortical areas performed their function. The turning point occurred approximately after the Second World War.

3. THE GOLDEN AGE OF SYSTEM NEUROSCIENCE

For the first time, a series of medical and technical advances allowed neurologists and neuroscientists not only to localize areas involved in specific functions, but also to describe the mechanisms underlying these functions. This breakthrough was mainly due to the improvement in anesthetic procedures and the discovery of antibiotics on the medical side, and on the technical side to the development of techniques allowing single neurons recording first in anesthetized and then also in behaving animals. It was the beginning of the golden age of neuroscience.

A classical example of the difference between localizing a function and understanding the underlying mechanism is given by visual physiology. For example, although there is no doubt that the occipital lobe houses the visual cortex, most neurons of this area do not respond to the presentation of visual stimuli.13 The answer to this discrepancy, known as the riddle of the visual cortex, was given by Hubel and Wiesel.14 These authors, instead of using flashes of diffuse light covering the whole visual field (a stimulus typically used for exciting retina and the lateral geniculate body), employed restricted stimuli, which allowed them to map the tuning of the recorded neurons only to some parts of the visual field, defined receptive fields. They found that only small, restricted stimuli were able to excite area 17 neurons. Furthermore, the effective stimuli were those presented with a specific orientation. If the stimuli had a “wrong” orientation or exceeded the receptive field, the response was weak or absent.

The fundamental studies of Hubel and Wiesel were followed by an extensive investigation of the occipital region formed by Brodmann area 17, 18, 19. Neurons selectively responsive for observed motion, shape, and color were reported in different subsectors within area 19. It was subsequently found that the visual areas found in are 19 were organized in 2 major streams: the ventral and the dorsal visual streams.15, 16 The ventral stream conveys visual information to the infero‐temporal cortex, while the dorsal stream conveys information to the parietal lobe.

Vision research moved then outside the extrastriate visual areas to the so‐called “association areas” like the infero‐temporal cortex (IT) and the superior temporal sulcus region (STS). It was found that these areas contain neurons that represent the synthesis of the processing previously found in the visual occipital areas. The work of Gross and his coworkers17 revealed the existence of neurons that respond selectively to faces, while the experiments of Perrett et al18 showed even more complex neurons responding to the vision of faces presented from different perspectives, and (in another part of the same region) neurons encoding the vision of a hand grasping an object, providing seminal information on the visual processing of action observation.

A fundamental experiment of this golden era is that by Mountcastle and coworkers.19 Posterior parietal lobe belongs to what was classically defined as an “association cortex.” Historically, association cortex was terra incognita, and it was assumed that its functional role was to “put together” information coming from different sensory modalities. As a result, different types of percepts occurred that could be used for a multiplicity of purposes.

Mountcastle et al,19 and in the same period Hyvarinen,20 demonstrated a new view of the functions of the posterior parietal lobe. These authors recorded single neuron activity from the inferior parietal lobule (IPL) in behaving macaque monkeys and assessed their activity not only in response to sensory stimuli, but also during motor behavior. Their results showed that many parietal neurons responded to sensory stimuli and, in accord with the classical view, to sensory stimuli of different modalities (eg visual and somatic stimuli), but also that a large number of neurons discharged during specific motor behavior.

These pioneering findings were confirmed recently with more advanced techniques. Using arrays of electrodes, it was established that in IPL there is a somatotopic organization with the face located rostrally, the hand in the intermediate part and eye movements in the caudalmost part.21 Other sectors of the posterior parietal lobe such as, for example, the areas buried in the intraparietal sulcus (eg anterior intraparietal areas‐AIP), demonstrated a similar organization. In particular, Sakata and coworkers22 showed that area AIP transforms object affordances in the appropriate manipulative motor programs.

Taken together, these data radically modified the ideas on the functional role of the posterior parietal lobe: from an associative area to an area organized in motor terms, on which different types of sensory information impinge according to the motor function of different neurons. While these data showed that the parietal lobe contain a large number of motor neurons involved in cognitive functions, the next step was to move toward the motor/premotor areas to test whether they also play a role in cognition.

First evidence in favor of this role was the demonstration of the presence of motor neurons in the ventral premotor cortex (area F5), which responded to the presentation of tridimensional stimuli transforming object affordances into potential motor acts.23 These neurons were subsequently dubbed “canonical neurons”.24 More surprisingly other neurons were found in the same areas that, instead of responding to the presentation of 3‐D object, discharged in response to the observation of actions similar to that motorically encoded by the recorded neurons. These neurons are known as “mirror neurons”.25, 26

Although there is evidence that human ventral premotor cortex is involved in action imitation,27 the most intriguing propriety of the mirror system is the role played in action understanding. Of course one may wonder why motor neurons should be involved in action understanding when, as mentioned above, primates have a large numbers of visual areas. An answer to this question was given by Marc Jeannerod: “A mere visual perception, without involvement of the motor system would only provide a description of the visible aspects of the movements of the agent, but it would not give precise information about the intrinsic components of the observed action which are critical for understanding what the action is about, what is its goal, and how to reproduce it”.28

Almost immediately after the discovery of mirror neurons in primates, thanks the development of new techniques (PET and fMRI), the existence of the mirror system was also demonstrated in humans. Similarly to monkeys, this system is housed by a series of parietal and premotor areas.

The development of brain imaging techniques opened a new era in neuroscience, providing neuroscientists with a new set of tools for testing perceptual and cognitive functions noninvasively in humans. Given its unquestionable advantages, brain imaging spread out pervasively in system neuroscience, quickly becoming the benchmark for neuroscientists. However, these techniques suffer from some inherent limitations which impeded brain imaging to tackle the issue of the mechanisms at the basis of brain functioning.

4. THE BRAIN IMAGING ERA

At the end of the last century, a new set of techniques, collectively known as brain imaging, was developed: first, the positron emission tomography (PET) and, subsequently, the functional magnetic resonance (fMRI), which became progressively the prevalent research technique in system neuroscience.

The principle underlying both these techniques is rather simple and based on old experiments by Roy and Sherrington.29 They demonstrated that scratching the paw of a rabbit determines a reddening of the somatosensory cortical areas of the opposite hemisphere, interpreting this phenomenon as a regional blood flow increase due to the metabolic request of the underlying somatosensory neurons.

For many years, this discovery remained merely theoretical. After the Second World War, however, with the development of studies on radioactivity, it was argued that this classical observation could be used as a starting point for the study of the organization of neural networks. Sokoloff30 injected a radioactive molecule of deoxyglucose (a modified glucose molecule which is slowly metabolized) and examined in which regions radioactivity increased. The experiments were made in animals, which were subsequently sacrificed, and the radioactivity evaluated in brain slices. Later on, technical advancements allowed the detection of radioactive traces directly from the skull, in a noninvasive manner. The technique could be therefore employed also in humans.

With regard to the visual system, the first activation PET study was reported in 1984. In this study, Fox & Raichle31 demonstrated that the increase in regional cerebral blood flow within the visual cortex depended on the rate of the visual stimulus given. Few years later, PET studies showed how specialized areas of the visual cortex could be targeted directly by appropriate stimulus design (see for example,32 the first study localizing the human color area). Concerning the mirror mechanism, PET was the first recording technique used to demonstrate that, also in humans, motor areas are active during the observation of actions made by others. Rizzolatti et al33 showed that observing the experimenter grasping common objects significantly activates the cortex of the middle temporal gyrus and the adjacent superior temporal sulcus, as well as the ventral premotor cortex. These results paralleled those by Perrett et al18 concerning the temporal lobe, and by Parma neuroscientists25, 26 concerning premotor cortex. Intriguingly, a similar activation distribution was reported also during motor imagery.34, 35

In general, the introduction of PET opened to the exciting possibility to look at a single glance at a brain map indicating which areas are active during a specific task. However, the need to administer radioactivity in PET examinations prevented a massive diffusion of this technique, which was gradually replaced by fMRI.36

The notion of brain mapping through functional magnetic resonance (fMRI) was essentially introduced in 1991 by 2 presentations at the 10th annual meeting of the Society for Magnetic Resonance in Medicine. Using dynamic susceptibility contrast (DSC) MRI with a gadolinium‐based contrast agent, Belliveau mapped the changes in cerebral blood volume (CBV) in a subject responding to a simple visual stimulus.37 A few days later, Brady and Kwong showed a video demonstrating MRI detection of brain activation based on changes in deoxyhemoglobin concentration, thus anticipating the importance of blood oxygen level‐dependent (BOLD) contrast for functional imaging.38 In synergy with the work of other groups that published complementary findings and developed accessory software tools (eg SPM),39 the development of fMRI led to a revolution in the neuroimaging field. In 1993, the number of published articles citing fMRI was fewer than 20. In 2003, that number was nearly 600. In 2013, it exceeded 2500.40 These numbers witness how fMRI, and more generally brain mapping, has become the most popular language of system neuroscience.

Overall, the main strength of fMRI is its relatively high spatial resolution. In addition, it is readily available to both clinical and academic researchers, is noninvasive, and can provide high‐resolution anatomic scans in the same session to use for localization as well as identification of vessels or development of maps of white matter connectivity through the use of diffusion tensor imaging.41

There is no doubt that brain imaging allowed great advances in refining the localization of brain functions. Typically, the most convincing results were obtained capitalizing upon previous data from lesions in humans or physiological experiments in animals. Unfortunately, brain imaging also produced something close to a new phrenology. Indeed, the conclusions deduced from these studies are inherently localizationistic in nature,42 thus attributing specific cognitive functions to restricted parts of the brain. A recent example could be the technically complex and sophisticated parcellation of the brain carried out with the aim to better specify the localization of different functions, but leading to controversial claims about the existence of areas devoted to very specific and restricted functions, such as a gambling‐devoted region in premotor cortex.

5. BEYOND BRAIN IMAGING

In addition to the possibility of its phrenological misuse, fMRI has 3 major limitations intrinsic to it: (a) lack of a good temporal resolution depending on the impossibility to detect blood flow at the millisecond temporal scale (time problem); (b) lack of the possibility to establish the causal role of a given activated area in determining a specific function (correlation problem); (c) lack of possibility to study subjects in ecological situations, depending on the physical constraints of the scanner and the artifacts due to movements (ecological problem).

All these 3 pitfalls of brain imaging can be solved by complementing brain‐imaging techniques with the new emerging techniques coming from neurosurgery. In fact, the improvement of neurosurgery techniques allows now one to insert electrodes intracortically and electrically stimulate and/or to record intracranial electroencephalographic rhythms (Figure 1) and, in some case, even single neurons.

Figure 1.

Panel A shows 1 electrode for stereo‐EEG, with the zoomed region detailing the spacing between adjacent leads. Panel B shows how the pre‐ and postimplantation neuroradiological images are merged together to obtain a comprehensive picture of the cortical areas explored by each recording lead. Panel C shows a sample of stereo‐EEG recording, along with the synchronous videocamera recording

Thus, the high temporal resolution of intracranial EEG allows one to overcome the temporal problem, because it gives information about neural activity in the range of milliseconds, rather than in the range of seconds. Electrical stimulation enables to solve whether the correlations between fMRI activation of given brain regions and functions, reflect a causal role of these areas in the investigated functions, thus solving the “correlation problem.” Finally, the possibility of implanted patients to move freely and to execute a variety of cognitive and behavioral tasks, allows one to study the patient behavior from an ecological point of view, thus overcoming the ecological problem. In the next section, we will discuss, separately, these 3 advantages of intracranial EEG, which should lead to a new era in system neuroscience.

5.1. Solving the temporal problem

The pervasive usage of brain imaging created a sort of habituation in system neuroscientists: the answer to any question has to be mapped, with brain structures colored according to the degree of activation. Beyond the unquestionable advantages listed above, however, such an approach prevents from appreciating the temporal dynamics of the neural activity. Functional MRI informs about the regions whose oxygen consumption increases over a time period of 2‐3 seconds, without any hint about the local dynamics of the nodes, as well as the relative timing of their activity, information crucial to fully understand human brain functions.43, 44 In other words, fMRI is the ideal tool to localize the whole network involved in a given function, but with a timeless result, which cannot render the real neural mechanisms.

In the era of brain imaging, many attempts have been made to overcome this lack of temporal information. However, available noninvasive recording and imaging techniques seem to follow a “Heisenberg uncertainty”‐like principle whereby it is impossible to attain both high spatial and temporal resolution. Hence, a precise and comprehensive four‐dimensional cartography of human neural activity appears unattainable. Beyond metabolic imaging techniques, researchers tried to transform Electro‐Encephalography (EEG) and Magneto‐Encephalography (MEG) recordings in brain mapping tools.45, 46, 47, 48 However, even if their temporal resolution allowed one to observe the intra‐ and inter‐areal dynamics, to date the spatial resolution remains too poor in localization power (1‐2 cm).43, 49 The combination of EEG and fMRI has been suggested as a solution, using EEG to determine the temporal dynamics within and between the areas identified with fMRI.50 However, the disparate nature of the 2 signals recorded (hemodynamic for fMRI, electrical for EEG) creates discrepancies in the results that prevent precise matching of these methods.43

Invasive intracranial EEG offers a unique opportunity to overcome these limitations, thanks to unparalleled combination of spatial and temporal resolution. Indeed, anatomical and electrical data can be combined to generate highly resolved four‐dimensional maps of human cortical processing. The feasibility of this approach was recently demonstrated in a study by Avanzini et al,51 who showed how the somatosensory processing following electrical stimulation of the median nerve recruits a wide cortical network (Figure 2), with each area characterized by a peculiar time course, ranging from phasic short lasting activities to tonic ones. This information constitutes a key element to discuss the functional role of each area in the process. This time course characterization is not only valuable in subdividing the stages of a single neural process, but also to reveal differences and similarities between different experimental conditions. In a recent study,52 tonic activity was recorded from posterior insular and opercular cortices in response to any tactile stimulation, regardless the stimulated limb, thus revealing an involvement of these regions in high‐order functions like tactile memory and awareness.

Figure 2.

Panel A indicates the cortical regions activated following median nerve stimulation, regardless their timing. However, these territories do not follow a common temporal profile of activation. Panels B, C, and D report the maps of cortical territories active at 3 different latencies, namely 20, 60, and 110 ms following the peripheral stimulus delivery

The temporal information achievable by having access to a time‐resolved recording from many precisely localizable cortical points is not restricted to latencies. Indeed, one can wonder also whether the time course of activation is tuned to specific events, thus revealing a major role of an area in encoding them. Using this approach, Caruana et al53 asked patients to observe tool action videos, evaluating the distribution of recording leads whose activity best aligned with video‐onset, action onset, and hand‐object contact, respectively. Noteworthy, this approach allowed authors to decompose the neural activity during action observation in a 3‐stage hierarchy. First, visual regions became active at the video onset, in the absence of any observed movement. Second, action onset activated a set of occipito‐temporal, parietal, and premotor regions corresponding to the action observation/execution network,54, 55, 56 which is known to house mirror neurons in the monkey.57, 58 Finally, the contact between the tool and the target object triggered activity in 2 brain regions, the dorsal premotor cortex and the human SII, possibly reflecting a mirroring of the observed hand‐object contact. These responses to contact, which escaped the previous fMRI studies,59 demonstrate that action observation, similarly to action execution, is a dynamic process where different brain regions become sequentially active in time, hence requiring a time‐resolved technique to be studied and characterized.

This time‐resolved approach, applied to any of the system neuroscience open points, may contribute the missing piece of information necessary to switch from the localization of the involved network to the description of the neural mechanism subserving a given function.

5.2. Solving the correlation problem

Brain imaging studies investigate how a part of the brain reacts to perceptual or cognitive tasks performed by the experimental subject. The correlation between the presented stimuli and the activity of a given brain region is interpreted as the evidence of its fundamental role in encoding the investigated function. However results obtained by correlative studies are typically “underdetermined”,60 as there are several possible reasons accounting for the activation of an area even when it is not fundamentally involved in the examined function—including spurious reasons such as arousal or motor preparation.

In addition, the development of new statistical tools to analyze imaging studies, instead of solving the correlation problem, led to rather contradicting results. A good example is the case of face perception, where results obtained by multivoxel pattern analyses (MVPA) appear to contradict previous evidence about the existence of category‐selective regions in the fusiform face area (FFA; see 61). A similar problem rises in studies on the neural correlate of basic emotions, where different types of meta‐analyses (ALE, MKDA, BSPP) suggested both the existence of brain networks dedicated to emotions61, 62, 63, 64 and the opposite conclusion, that is a complete lack of functional specialization in the emotional brain.65, 66, 67

A complementary view is offered by stimulation studies, where the experimenter evaluates how stimulating the brain affects the overt behavior of a subject, or its subjective experience. Noninvasive stimulation techniques routinely employed in system neuroscience, such as transcranial magnetic stimulation (TMS) and transcranial direct current stimulation (tDCS), proved to be particularly effective in interfering with normal brain functions. Concerning the mirror mechanism, TMS studies played a major role in unraveling the mechanisms involved in the visuomotor transformation underlying action observation: first, by demonstrating that the observation of others’ actions is accompanied by motor facilitation of the observer central nervous system and, second, that such facilitation occurs during specific temporal windows.68, 69, 70, 71 In addition, the recent development of continuous theta‐burst stimulation (cTBS) and tDCS techniques also suggested that motor plays a causal role in action understanding, rather than a correlative one.72, 73

While technical limitations prevented these techniques from inducing complex behavioral responses, intracranial electrical stimulation proved to be the most effective technique in eliciting perceptual, mnemonic, or emotional experiences as well as complex behaviors.

Introduced in neurosurgery since the late XIX century, the acme of electrical stimulation was reached in the 50s, thanks to the work by Penfield, in humans, and Woolsey, in the monkey. In the early days of electrical stimulation, however, results were mostly circumstantial, also because of the paucity of its employment in humans. In addition, variables such as the possible spread of current in the brain because of intra and after discharges were largely uncontrolled. Accordingly, its main contribution to system neuroscience was restricted to the establishment of the homuncular arrangement of sensory and motor areas.

Nowadays, in contrast, electrical stimulation is routinely performed in several surgical centers, providing system neuroscience with a growing amount of data. In addition, technical developments in the field of stereo‐EEG and ECoG allow one to electrically stimulate specific brain sites while recording intracranial EEG from the nonstimulating depth electrodes, thus permitting to evaluate the presence of intra and after discharges in surrounding brain regions.

Intracranial electrical stimulation is performed using either low‐frequency (LF; typically 1 Hz) or high‐frequency (HF; typically 50 Hz) pulses. The aim of LF stimulation is to map primary motor and sensory (somatosensory, acoustic and visual) areas by evoking simple clinical signs (eg jerks or visual or auditory hallucinations), whereas its main usage in system neuroscience lies in the study of functional connectivity using cortico‐cortical evoked potentials (CCEPs).74, 75

HF stimulation, in contrast, is systematically employed to elicit subjective or objective responses, ranging from complex motor behaviors, to simple and complex sensory hallucinations and emotional states. It is worth noting that these responses cannot be studied using standard imaging techniques, also because some of these reactions, such as emotional ones, cannot be elicited “at command” as requested by standard neuroscientific laboratory settings.

HF stimulation is today in a position to provide system neuroscience with unique causal, rather than correlative, information, on which one may ground brain imaging correlative findings. The case of face perception, mentioned above, represents a clear example of how stimulation findings can disentangle between alternative interpretations provided by imaging techniques. The issue of whether FFA is devoted to face processing rather than to more general high‐order visual stimuli has been recently tackled in 2 studies by integrating correlative data with electrical stimulation.76, 77 The finding of both papers was that the same sites where intracranial recordings showed visual responses to faces, affected face processing only. Interestingly, Schalk and coworkers also reported that “when electrically stimulated in the fusiform face area while viewing objects, the patient reported illusory faces while the objects remained unchanged. When stimulated in nearby color‐preferring sites, he reported seeing rainbows”.76

Another interesting field where electrical stimulation can be crucial to interpret some brain imaging data is the case of pain localization. Several experiments showed that, besides activating somatosensory cortices, nociceptive stimulation also activates the cingulate cortex and, more specifically, a sector of area 24 (aMCC). This evidence led to the shared view that this region plays a role in pain processing.78 Interestingly, the evidence that the same cingulate neurons react to both painful stimulation and the observation of the same stimulation in others79 led to the hypothesis that aMCC hosts a mirror mechanism for pain.80 In contrast with this view are some imaging data showing that nociceptive stimulation triggers the same cingulate region found to be active following non‐nociceptive somatosensory, auditory, and visual stimuli, when presented with a similar attentional salience.81 These data suggested that the activation of the aMCC can be accounted for not by its specificity for pain but, rather, by its encoding salient stimuli possibly representing a potential threat.

The causal data provided by high‐frequency electrical stimulation in humans strongly support the latter interpretation, showing that the stimulation of aMCC appears triggering a variety of motor behaviors, but not painful sensations,82, 83 also when the stimulation is applied to the very same sites active during self or others’ painful stimulation.79 Taken together, electrical stimulation data on the aMCC lead to a motor interpretation of the data concerning the cingulate contribution of felt and observed pain, offering new insights for the interpretation of the aMCC mirror mechanism.

6. SOLVING THE ECOLOGICAL PROBLEM

Another important limitation of brain imaging studies concerns the physical constrains where experiments are conducted. During fMRI studies, in fact, subjects lie in the restricted space of the scanner, which prevents any naturalistic and ecological movement. These physical constraints limited the possibility to study the overlap between action observation and execution. Indeed, although grasping movements have been occasionally investigated by requiring finger displacements or small tool actions,84 the execution of complex hand actions, not to mention whole body movements, were impossible to study.

In addition, brain imaging techniques can hardly be useful for studying emotions. First, because of the constrained conditions and the lack of social interactions, preventing the emergence of spontaneous emotional expressions. Second, because of movement artifacts accompanying facial emotional displays. Accordingly, the majority of fMRI studies on emotions are indirect, as they are based on the recognition of facial and body emotional expressions.

A good example is the case of laughter. This behavior involves spontaneous facial grimaces, vocalizations, and postural adjustments due to coordinated movements of the diaphragm, inspiratory muscles, and larynx, which render impossible to study this expression within an fMRI scanner. Accordingly, the majority of our knowledge on this behavior is based on hypotheses derived from neuropsychological patients.85, 86 In contrast, this problem has been solved by stereo‐EEG. The stimulation of the pregenual sector of the anterior cingulate cortex (pACC) determines both the overt (facial expression) and, most frequently, subjective feeling of merriment.87 In addition, when coupled with the presentation of the same expression, a significant activation is observed, indicating the excitation of the same site which elicits laughter when stimulated.88 A similar approach was proved effective in studies about fear89, 90 and disgust,91, 92 opening new avenues for the exploration of human emotions and the search for mirror mechanisms outside the motor domain.

7. CONCLUSION

In summary, here we reviewed the major advancements which characterized system neuroscience since the golden age, through the brain imaging era and up to the modern era. In this latter, the use of multimodal approaches combining different recording techniques and advanced signal processing will hopefully lead to a better understanding of the neural mechanisms underlying neurological and cognitive functions. To this aim, the contribution offered by intracranial recordings will be fundamental, thanks to the possibility to detail with a millisecond‐based temporal resolution the time course of neural activity, and to investigate the causal relationship between the activation of cortical areas and the emergence of specific behavior. In addition, given that patients undergoing stereo‐EEG are free to move while being recorded, intracranial recordings should lead to a major step forward in the knowledge about the organization of the human motor system and the spatiotemporal dynamics ruling the motor contribution to cognitive functions.

Rizzolatti G, Fabbri‐Destro M, Caruana F, Avanzini P. System neuroscience: Past, present, and future. CNS Neurosci Ther. 2018;24:685–693. 10.1111/cns.12997