*2.3. Insula*

The insula has a cytoarchitectonically diverse organization based on posterior (*granular*), intermediate (*dysgranular*), and anterior (*agranular*) subdivisions. It is thought to encode sensation of the physiological condition for the entire body, which is known as interoception [31,57]. Craig et al. [31] showed the existence of 15 distinct cortical areas within the insula of long-tailed macaque monkeys (Macaca fascicularis) [31]. The architectonic map of the macaque insula is an important step towards understanding the connections and function of the insular cortex subareas. Recent models sugges<sup>t</sup> interoceptive a fferents are received in the posterior insula, processed by the intermediate insula, and then integrated for e fferent autonomic regulation in the anterior insula [58].

A recent functional magnetic resonance imaging (fMRI) study suggested that the anterior insula, skin conductance, and pupil size encode a predictive model of pain. The anterior insula reflected the summed pain expectation and prediction errors from unexpected pain, while the posterior insula encoded pain intensity [59]. In subjects estimating pain intensity, fMRI activity during pain magnitude ratings was matched to that during visual magnitude ratings, to examine how pain perception is an assessment of stimulus intensity [60]. Posterior insular activity correlated with both the magnitude of the visual stimulus used in the task and pain intensity. Modality-specific pain estimation was located further anteriorly. The authors suggested this means pain perception within the insula results from the transformation of nociceptive information into subjective intensity assessment.

The insula is consistently included in circuits whose activity is correlated with pain perception. Some lesions of the insula result in a syndrome called pain asymbolia. In this condition, first described by Schilder in the 1930s, pain perception is intact, ye<sup>t</sup> a patient's emotional response to pain is inappropriate [61]. Even though both the ACC and insula show elevated activity when viewing painful versus non-painful images [62], the symptoms resulting from insular lesions sugges<sup>t</sup> the insula is more specifically involved in pain empathy. The organization of the insula lends itself to having both a ffective and somatosensory components. The ACC (discussed next) is likely more involved with a ffective processing, while the somatosensory cortex is more involved with sensory processing. Widespread damage to the insula, obliterating its function, appears to increase experiential noxious pain perception and somatosensory activation ipsilateral to the lesion [63]. The insula may, therefore, help to suppress pain perception via specific inhibition of the somatosensory cortex.

Interestingly, the insula and the ACC both contain the highest concentrations of von Economo cells in the human brain. These are specialized cells with large cell bodies and axons that project to homeostatic regions in midbrain PAG and the parabrachial nucleus [64]. These cells are thought to mediate rapid social or behavioral inhibition, based on social or cognitive interoception. A recent study using patch clamp recordings of von Economo neurons demonstrated that they are regionally specific excitatory neurons [65]. The loss of these neurons is implicated in many neurological disorders and diseases. The insula and ACC are likely implicated in integrating general interoceptive information, among which pain information is distributed [32].

Direct electrical stimulation of the insula, as in epilepsy mapping, produces a variety of responses, including somatosensory, olfactory, thermal, auditory, and gustatory percepts [66]. Although early investigators, such as Penfield, failed to identify any evoked pain percepts by insular stimulation [33], more recent studies show pain responses from the insula at about 10% of sites. Systematic investigation may clarify the role of the insula in pain processing.

### *2.4. Anterior Cingulate Cortex*

The ACC shows activation with pain, anxiety, and cognitive control, suggesting a role in responding to insults that are either corporeal or cognitive [67]. The ACC generally acts as a monitor that signals needed behavioral adjustments in response to corporeal or cognitive challenges [67,68]. In the late 1940s, frontal lobotomies were used to treat intractable pain and addiction with some success [69–71]. In the 1960s, Foltz and White [34] reported their experience performing cingulotomies (typically bilaterally) on 16 patients, for whom the a ffective components of chronic pain were particularly pronounced. The results were poor in only two of the 16 patients, although this was an early study lacking objective pain metrics. Intriguingly, the authors report that signs of opiate withdrawal were also lessened in these patients (14 of whom were "addicted") after cingulotomy. In 1988, Smith et al. [72] performed surgical cingulotomies on Sprague-Dawley rats, to investigate the role of the cingulate in stress-induced plasma beta-endorphin and morphine withdrawal. Beta-endorphin levels did not significantly increase after cingulotomy or postoperative induced stress, when each was tested independently. However, rat cingulotomy with postoperative induced stress caused a significant increase in plasma beta-endorphin concentrations. These findings suggested cingulum involvement in the regulation of the stress hormone response of beta-endorphin. More so, cingulotomy might be the cause of opiate withdrawal. A series of case reports and case series in humans followed this promising early work. A total of 13 case reports on cingulotomy were published prior to 2008—the majority before 1980 [73]. Results showed variable pain and opiate withdrawal improvements [73–75].

In the 1950s, attempts were made to electrically stimulate the ACC to treat chronic pain. Few e ffects were seen, most of which were adverse—such as speech arrest, vomiting, and tonic muscle contractions [76]. More recent studies using DBS with contemporary electrodes showed improved e fficacy and reduced adverse e ffects [77,78]. For example, in the most extensive study to date using DBS to treat pain, Boccard et al. [79] found a significant 43% improvement on a numeric pain rating scale after six months. Improvements were also seen in other quality-of-life metrics, and on a general health scale. The cognitive and a ffective elements of chronic pain in these studies showed the most significant improvements, suggesting that ACC DBS may improve these aspects of chronic pain [80].

In comparison, a positive emission tomography (PET) study demonstrated that the ACC is activated during thalamic DBS in patients with chronic pain [55]. Together, these e ffects strongly implicate the dorsal ACC in processing a ffective components of pain; however, results about the directionality of cingulate e ffects are few and mixed. Recent studies have reported the essential nature of examining the temporal dynamics with which the ACC tracks internal cognitive state and induces

cognitive control [74,81]. Improved understanding of the temporal dynamics of cingulate function relative to pain may be an important missing piece in understanding the cerebral localization of pain networks.

### *2.5. Beyond the Thalamus, Insula, and ACC*

Other recent studies suggested that entirely di fferent brain networks, engaged by the ACC, are responsible for pain regulation. Pain is understood to be a complex experience with sensory, cognitive, and emotional components [82]. Woo et al. [75] indicated that self-regulation of pain and the brain areas responsible for painful experiences are controlled by another circuit. The circuit consists of the genual and subgenual cingulate cortexes, the projections of the nucleus accumbens (NAc, involved in aversion, motivation, and reward valuation), and projections to the ventromedial prefrontal cortex. The suggested involvement of the cingulate is in the valuation of pain or weighting the context of painful stimuli, which preferentially projects to the NAc. A recent rodent study appears to support this mechanism of pain regulation [83]. In this study, the authors suggested that pain relief is associated with learning and motivation to seek environments associated with a relieved state. Endogenous opioid signaling in the rodent cingulate alters dopamine release in the NAc and pain avoidance behavior. Baliki et al. [84] also suggested a role for the NAc in pain valuation and analgesic potential. In this study, NAc activity in response to acute noxious thermal stimuli was compared in control and chronic back pain patients. At the acute noxious stimulus, normal subjects had positive phasic NAc activation, while chronic back pain patients demonstrated negative polarity phasic activity. The authors suggested the onset of acute pain relieved the chronic back pain, which was confirmed psychophysically. Mallory et al. [85] reported on sustained pain relief in a 72-year-old woman with a large right hemisphere infarct, who developed refractory left hemibody pain. Neither the NAc nor the PVG was successful in relieving pain when stimulated in isolation. The combined stimulation reduced the patient's pain from 10 to 0 at 11 months after surgery. The patient's post-stroke depression also stayed in remission. These results sugges<sup>t</sup> the emotional aspects of pain can be treated quite well when stimulation involves the NAc.

Such a distributed organization of pain information makes large-scale electrical disruption of gray matter, such as that employed in cingulate DBS or cingulotomy, theoretically unlikely to provide the specificity required to disrupt pain without unintended side e ffects. These considerations motivate the use of white matter DBS of particular pain-associated tracts, such as the ventral internal capsule, along with the aggregates of grey matter involved in pain processing. It is also important to study the temporal dynamics of pain to determine whether there are temporal or frequency components unique to the development of chronic pain and various pain syndromes. This information may be used to determine when is the best time to modulate information processing in these areas to most e ffectively improve centralized chronic pain.

### **3. Temporal Dynamics of Pain**

Studying the dynamic properties of pain perception is important for disentangling the neural correlates from the described anatomical areas that display a myriad of functions [86]. Few imaging studies have correlated fMRI signals with the temporal properties of pain [86–88]. An important consideration in studying the temporal dynamics of pain is that the time constants and activation functions of the various types of peripheral pain fibers are well characterized. The aforementioned unmyelinated C fibers transmit nociceptive information more slowly than myelinated Aδ fibers. These di fferences in conduction may resonate throughout the system in meaningful ways. As a psychophysical example, the discrimination of di fferent painful sources (i.e., thermal, mechanical, chemical) becomes impossible with the loss of the rapidly conducting myelinated pain fibers. While these peripheral biophysics sugges<sup>t</sup> that temporal dynamics may be necessary for understanding pain perception, studies have ye<sup>t</sup> to use temporal properties of signals to study pain discrimination. It is not clear to what extent, if any, temporal dynamics derived from the biophysics of peripheral sensors apply to chronic pain.

Electroencephalography (EEG) and magnetoencephalography (MEG) are the most readily available noninvasive methodologies to study high-temporal-resolution activity in the brain. EEG has been used to study acute responses to painful stimuli, mostly thermal pain evoked by lasers [89–92]. Laser-evoked pain has become the dominant psychophysical method for studying pain, because of safety, spatial precision, and how rapidly lasers turn on and o ff. The source localization of these responses also implicated the dorsal ACC and the insula [93,94]. It is not clear, however, to what extent these studies address the mechanisms underlying chronic pain. Studying chronic pain, per se, is fraught with numerous di fficulties. Not only does chronic pain have various causes, but the expression of chronic pain in humans is heterogeneous. It is associated with varied bodily locations, triggers, and unpredictable responses to those triggers. Within chronic pain research, these complications are often simplified in animal models or very specific study populations. This limits the general applicability of results [95].

There have been few and mixed results of EEG studies of chronic pain. A recent meta-analysis of chronic pain EEG studies involved recording subjects during rest, sensory stimulation, and cognitive tasks and showed that subjects with chronic pain have elevated power in various regions across a range of frequencies [96]. Chronic pain was also associated with decreased evoked response amplitudes during cognitive tasks and sensory stimulation [96]. The dominant resting frequencies were typically lower than in healthy controls; however, the specific regions with power and the frequency changes were not consistent across studies. Increased resting frontal theta power was the most reproducible result in these studies. In one study, pathologically high theta power was localized to the insula, although this region is di fficult to record with EEG [97]. Theta oscillations are associated with thalamocortical loops, so studies have attributed these oscillations to pathological integration of painful experience into normal circuits. As described, such pathological integration could be due to the regulation of excitability via tonic and burst firing modes in the thalamus. Altered dynamic ranges of frontal theta levels could represent chronic changes in thalamocortical networks.

Several studies have attempted to simulate chronic pain in the laboratory by elongating the duration of painful stimuli. Rhythmic or oscillatory responses to painful stimuli lasting up to tens of minutes in duration are consistently characterized by suppression of alpha- and beta-range power and increases in gamma power [98]. Perturbations of both the bottom-up (i.e., sensory) context and the top-down (i.e., attentional or cognitive) contexts consistently alter pain perception [99,100]. Furthermore, the oscillatory context is important for pain perception, as somatosensory alpha is negatively correlated with pain perception [101,102].

In one notable EEG study, painful tonic stimuli (10-min exposure) correlated with persistent frontocentral gamma power. The source localized to the dorsal ACC [98]. However, the reduction in beta oscillations that other studies have found was more posterior and determined to be correlated with the judgement of stimulus intensity. Such studies using long-duration painful stimuli sugges<sup>t</sup> that the neural representation of pain changes with the duration of the painful experience. Apkarian et al. [103] showed that acute pain stimuli generally activate the somatosensory, insular, and cingulate cortical regions. Patients with chronic back pain had brain activity that localized to the medial prefrontal cortex and into the ACC, which was unique to the chronic pain patients. As the duration of the painful experience increases, fewer somatosensory regions are activated and there is more recruitment from limbic/affective/motivational regions. Their interpretation was that transient nociceptive activity, at some point, is converted into sustained emotional su ffering. This result highlights the dynamic nature of chronic pain in time. How these dynamics evolve from the seconds–minutes domain of acute pain to the months and years of chronic pain is unknown, and they are important factors for delineating and treatment.

Finally, intracranial electrophysiological responses to acute painful stimuli in patients implanted with grid electrodes reproduced the changes in delta through beta power described in EEG studies [92,104]. Causal interactions were determined from local field potentials recorded during the response to a painful cutaneous laser stimulus. Cognitive pain control was related to information transfer between the ACC and somatosensory cortices [56,92]. These studies were only carried out in three patients and examined responses to brief thermal pain. Recording broadband high-frequency (e.g., high-gamma; ~70–200 Hz) local field potentials, which correlate with neuronal activity [105,106], yields an intriguing possibility of correlating pain network activity to neuronal populations with high temporal precision. Furthermore, the possibility of altering pain perception with brain stimulation could be highly informative for the development of DBS for chronic pain.

### **4. Pain Illusion Suggests Distinct Roles for the ACC and Insula in Chronic Pain**

The "gate control theory" of pain was first described by Melzack and Wall in 1965 [21]; it implicated the relative activation and inhibition of areas in the ascending pain pathway [18]. This theory was sustained in part by studies using the thermal grill illusion (TGI). In the TGI, first discovered by Thunberg in 1896 [107], closely spaced alternating hot and cold stimuli—themselves non-painful—are perceived as painful when felt simultaneously. A three-dimensional rendering of a thermal grill interface constructed by our group is shown in Figure 3a. Each bar on the interface is programmable to a range of temperatures. Images captured with an infrared camera, shown in Figure 3b, demonstrate examples of the temperature arrangements acquired with our thermal grill interface. Since the 1890s, many explanations have been investigated to understand the physiological basis for perceived pain [107,108].

**Figure 3.** Rendering of a thermal grill interface and infrared images demonstrating patterns of warm and cool temperatures. Drawing (**a**) is representative of the interface that our group constructed. The top of the interface consists of six copper bars. Each bar is 1.0 cm in length and the bars are spaced 0.15 cm apart. The width of the six-bar interface is 9.8 cm. The electronic components fit inside the labeled compartment next to the fans, which allow for necessary air flow. Each bar is connected to a Peltier device, allowing for programmable temperature control (pink and blue coloring represents bars programmed to warm and cool temperatures, respectively). Infrared images acquired using the device are shown in (**b**). Left (**b**) shows all bars set to a cool temperature close to 20.0 ◦C. Middle (**b**) shows all bars set to a warm temperature close 40.0 ◦C. Right (**b**) shows one bar set at 20.0 ◦C and another bar near 40.0 ◦C. The alternating temperature setting is used to produce the pain illusion.

In 1994, Craig and Bushnell suggested that the integration of pain and temperature is the basis of cold-evoked, burning pain [107]. Although the mechanisms remain unclear, the underlying thermal stimuli are thought to inhibit each other, making way for summated nociceptive information without its accompanying somatosensory identity. A recent study in mice showed that the concurrent inhibition and excitation of polymodal channels provides the sensory code for warm perception [108].

Over time, the thermal grill has increased in popularity as a model for studying pain, because it induces neural activity that is perceived as burning pain without actually causing physical harm [109,110]. Clinical studies using the thermal grill have varied from evaluating perceived pain with surveys [111] to using fMRI [112] and PET [113] to evaluate structural involvement. Neuroimaging used in TGI studies suggests that hot or cold stimuli alone activate the insula and somatosensory cortex. Illusory pain from the TGI additionally activates the ACC [113]. The TGI produces a conscious perception of pain, perhaps unique to the ACC. Isolated thermal stimuli (either hot or cold) activate the insula and somatosensory cortex, demonstrating a dissociation between these regions and the cingulate. This further suggests separate roles for the areas involved in pain perception. Bouhassira et al. [114] and others [115,116] have reported that a subset of volunteers in thermal grill experiments are classified as poor or nonresponders. Bouhassira et al. defined these subjects based on not reporting at least one paradoxical painful sensation. There are a variety of reasons someone could be a nonresponder, such as anatomical di fferences due to calluses, prior injury, or variations in pain tolerance. Including poor-responding subjects in thermal grill experiments during intracranial recording will provide greater understanding of individual variations in pain processing.

Future studies using the thermal grill and neurophysiological data acquired with electroencephalography (ECoG) or stereoelectroencephalography (SEEG) will help to substantiate the temporal dynamics of pain perception. This experimental system may also provide data o ffering insight on the a ffective–motivational ("unpleasantness") and the discriminatory ("pain-intensity") aspects of pain [112]. Real-time acquisition from the thalamus, insula, and ACC may disentangle the complex relationship between these structures in pain processing. This groundwork will elucidate targets of electrical stimulation as a treatment for chronic pain.
