Nociception, pain, negative moods and behavior selection

Abstract

Recent neuroimaging studies suggest that the brain adapts with pain, as well as imparts risk for developing chronic pain. Within this context we revisit the concepts for nociception, acute and chronic pain, and negative moods relative to behavior selection. We redefine nociception as the mechanism protecting the organism from injury; while acute pain as failure of avoidant behavior; and a mesolimbic threshold process that gates the transformation of nociceptive activity to conscious pain. Adaptations in this threshold process are envisioned to be critical for development of chronic pain. We deconstruct chronic pain into four distinct phases, each with specific mechanisms; and outline current state of knowledge regarding these mechanisms: The limbic brain imparting risk, while mesolimbic learning processes reorganizing the neocortex into a chronic pain state. Moreover, pain and negative moods are envisioned as a continuum of aversive behavioral learning, which enhance survival by protecting against threats.

INTRODUCTION

Classically, pain has been conceptualized from the narrow viewpoint of nociceptive processing. The field has generated extensive knowledge regarding the transduction, transmission, and spinal cord processing of nociceptive signals related to acute and chronic pain; similarly, animal studies have unraveled properties of primary afferents, their spinal cord circuitry, and related specialized pathways in the brain that mediate pain-like behavior. Post-nerve injury reorganization of nociceptive afferents and spinal cord circuitry, in particular, has been extensively characterized in rodent models, with the tacit assumption that acute and chronic pain is best understood through this circuitry. In parallel, human brain imaging studies have identified nociceptive brain circuits. However, recent human brain imaging studies examining a variety of pain conditions indicate that the brain plays an active role in acute and clinical pain perception (figure 1), leading to a heated debate regarding the respective importance of peripheral afferents versus the brain’s interpretation of afferent signals. Here we review implications of these competing concepts in light of emerging evidence.

Figure 1.

Descartes’s concept of sensation illustrates the pain system. In addition reorganization of its components are superimposed based on modern rodent model physiology and human brain imaging studies. The Cartesian illustration is explicit regarding an impinging stimulus being transduced and transmitted to a specific brain region where perception takes place. The additional panels emphasize the modern evidence that all components of this system undergo reorganization following an injury that gives rise to a persistent or chronic pain state. End organ injury gives rise to changes in the local milieu, inflammatory soup, and in afferent response properties; collectively described as peripheral sensitization (adapted from (Julius and Basbaum, 2001)). Additionally, spinal cord circuitry undergoes a large number of changes resulting in central sensitization (adapted from (Scholz and Woolf, 2002)), which includes enhanced glutamatergic signaling, changes in second order messenger processes and activation of microglia. At the level of the brain, human neuroimaging studies indicate anatomical and functional reorganization.

The standard definition of pain emphasizes its subjectivity. Subjectivity, in turn, implies a conscious experience. A central goal of our perspective is to revise the understanding of conscious pain perception by incorporating nociception, acute and chronic pain, and negative moods into the unifying framework of behavior selection, where behavioral selections encompass the full range of possible actions for stimuli, whether internal or external, conscious or subconscious, and voluntary or involuntary. This viewpoint calls for a re-examination of the definitions we have inherited, as their narrow meanings have limited the types of questions posed within the field. Furthermore, we will introduce a novel interpretation of supraspinal processing of pain distinguishing between the subjectively conscious pain state and sub-, un- or pre-conscious nociceptive processes. We propose a comprehensive mechanistic model that properly incorporates acute and chronic pain, where the emotional limbic brain plays a critical role in bridging nociception and pain perception, as well as in the transition from acute to chronic pain; leading to the generalization of the functional continuity between pain and negative moods. Given that the literature supporting our model remains recent and fragmentary, we highlight important gaps in knowledge, as well as fruitful directions of inquiry.

Nociception

Sherrington coined the term nociception (Sherrington, 1900), and outlined its underlying neural structures. He viewed nociceptive reflexes and pain perception as tightly linked processes, such that “were the brain intact, [nociceptive activity] would, we may presume, evoke ‘pain’” (Woodworth and Sherrington, 1904). Since these first observations more than a hundred years of research has produced incontrovertible evidence regarding the specialized neuronal/molecular properties that define and characterize the nociceptive machinery (Basbaum et al., 2009; Woolf and Salter, 2000). Activation of nociceptors and nociceptive pathways undoubtedly can give rise to pain, and the close correspondence between nociceptor properties and human pain perception has been confirmed using a variety of experimental approaches (Marks et al., 2006).

On the other hand, ample evidence indicates that nociceptors can be active in the absence of pain perception. For example, any pain psychophysicist would agree that applying a 50 kg weight on a 1 cm area of skin would evoke excruciating pain. Yet, experienced ballerinas dance with point shoes for many hours, reporting deep positive emotional satisfaction while their toes carry the weight of their body, an activity that must massively and continuously activate their toe nociceptors. Therefore, at least ballerinas with intense training are capable of dissociating pain from nociception. The primary reason I fidget in my chair while writing this article is because nociceptors innervating my skin, muscle, and bone command that my posture needs adjustment. Whereas proprioceptors provide (conscious, but usually habituated) information about the location and position of my body in space, nociceptors provide inputs that protect my body from injury. One (especially the pain scientist) commonly forgets the fact that most humans, and perhaps many other species as well, spend most of their lives free of pain and with no obvious tissue injuries. This must be ascribed to active nociceptors, as there are no other alternative neuronal mechanisms available to continuously protect the body and subvert the potential for injury and resulting pain perception.

The necessity of nociceptive activity in the absence of stark pain is perhaps best illustrated by patients with various pain insensitivities due to leprosy, where the simple act of walking a mile gives rise to severe soft tissue and bone damage due to the lack of nociceptive afferents which render these subjects unable to modulate their gait and incapable of avoiding tissue injury (Brand, 1993). The best related clinical evidence comes from painless Charcot joints of tabes dorsalis (Sanders, 2004), and painless channelopathies (Bennett and Woods, 2014). Dissociations between nociceptive stimuli and perceived pain are also demonstrable in the lab setting. For example, repetitive painful laser stimuli induce rate-dependent modulation of evoked potentials, presumably reflecting nociceptive barrage of varying intensity, for a stable perception of pain (Mouraux et al., 2013; Wang et al., 2010). Moreover, when pain perception is assessed dynamically for complex stimulus patterns (“offset analgesia”; (Yelle et al., 2008) (Cecchi et al., 2012)), the statistical power function relating stimulus intensity to perceived magnitude of pain degrades and becomes highly nonlinear (Price, 1988; Stevens, 1957).

The electrophysiology of nociceptors and of nociceptive pathways are fully consistent with this idea, as extensive evidence shows that the majority of peripheral nociceptors can be activated with stimuli that are sub-threshold for pain perception, and a large proportion of central nociceptive neurons respond convergently to non-nociceptive stimuli as well (Kenshalo et al., 1980; Meyer et al., 2006; Willis and Coggeshall, 1978). As a result, the nociceptive control of behavior routinely occurs in the absence of consciously perceived pain, rendering it “subconscious.” An important caveat to this position relates to microneurographic studies in humans that elegantly demonstrate the existence of certain peripheral nociceptors that, when individually activated (single action potentials), can give rise to pain perception within a perceptive field that closely matches the receptive field of the stimulated nociceptor (Weidner et al., 1999). The latter establishes existence of a class of nociceptors with direct access to pain, at least in the controlled lab setting, but does not discount the existence of other nociceptors activated with various bodily postures but that usually do not give rise to pain.

The active role of subconscious nociceptors extends directly to behavior, as even the most common behavioral repertoires require nociception to avoid injury. Daily motor movements could easily produce injury and tissue damage if one exceeds their natural range of motion, such as extreme extensions of fingers, elbows, shoulders, hips, knees or ankles, or chewing on or hitting hard objects, etc., which supports the conclusion that motor behaviors are collectively inhibited by nociceptors. Surprisingly, this idea has not been studied, apart from spinal reflex behaviors. Similarly, there is no physiological evidence describing the nociceptive circuitry inhibiting motor repertoires. It is unlikely that nociceptive modulation of motor programs are mediated through the motor cortex, as only excitatory nociceptive inputs have been described in the region (Chen et al., 2011b; Frot et al., 2013). More likely it is probably mediated through dorsal striatal circuitry, where nociceptive inputs have been reported (Braz et al., 2005; Chudler et al., 1993; Newman et al., 1996) and where a large portion of the output is inhibitory to thalamocortical motor circuits.

In summary, we argue that nociception continuously occurs in the absence of pain perception and it is a fundamental physiological process (in the language of Christof Koch: the zombie agent of the sense of pain (Koch, 2012)) that subconsciously provides more veridical and instantaneous information that protects the organism from tissue damage. From a mechanistic viewpoint, we presume that behaviors modulated by nociception, in the absence of pain, are contingent on already established habitual repertoires. In contrast, when pain is evoked it gives rise to new peripheral and spinal cord nociceptive learning/sensitization (Basbaum et al., 2009; Ikeda et al., 2003; Woolf and Salter, 2000), as well as emotional learning that is potentiated by the salience and perceived value of the aversive event. Yet by and large nociceptive functioning and its underlying mechanisms in the absence of pain surpass current understanding and require investigation in their own right.

Acute pain

We next re-examine the definition of pain in relationship to the “subconscious” concept of nociception. We emphasize that pain is a conscious subjective experience that is most commonly driven by nociceptive activity. Yet, its threshold and magnitude can be readily modulated by mood and attention (Bushnell et al., 2013), monetary reward (Vlaev et al., 2009), simple changes in instructions e.g. (Baliki et al., 2010), and through expectations (Wiech et al., 2014). A large brain imaging literature continues to examine the dependence of acute pain perception on various cognitive and emotional modulators, with a focus on the specific brain circuitry that mediate these relationships (Apkarian et al., 2005; Bushnell et al., 2013). Despite these efforts, the specificity of these cortical modulators of pain perception—as well as their shared characteristics across other sensory modalities—remains inadequately understood.

Consistent with and complementary to the variety of factors modulating pain and the related changes seen in brain circuitry, ample evidence demonstrates that conscious acute pain perception is highly malleable and a standardized nociceptive barrage does not translate to a fixed brain activity or to a prototypical perception. Pain perception can reflect moment-to-moment shifts in value judgments regarding the self and regarding the relationship between the self and the environment. Its value can change, as seen with rodents for which the availability of food trumps pain-related escape behavior when both are simultaneously present (Foo et al., 2009). Pain perception can be delayed for hours or days, when soldiers surviving the horrors of a battlefield do not report pain or suffering. We should also mention that Pavlov was able to train his dogs to salivate for painful stimuli (Pavlov, 1926, 2003). Thus the human verbal descriptions and behavioral experience of pain (as exhibited in motor, language, and facial and bodily expressions) - for a qualia that one would presume to be subjectively common - has a broad, culture- and context- dependent, repertoire. Pain therefore reflects an interaction amongst memory, attentional, and affective brain circuitry and afferent sensory inputs.

Given that nociceptors are continuously active in everyday behavior, one must posit the existence of a threshold phenomenon that transforms nociception to conscious pain (θ in figure 2). From a classical spinal cord physiology viewpoint, this threshold may be equated to the “gate control theory” (Melzack and Wall, 1965), where the balance between innocuous and noxious afferents at the level of the spinal cord determines the nociceptive signal traveling cephalad that, upon reaching the cortex, is interpreted as pain (the more modern alternative is some variation on “central sensitization,” especially for persistent pain conditions, where the gain of the spinal cord nociceptive synapse is amplified (Woolf and Salter, 2000)). In contrast, from a behavior selection viewpoint, the nociception-pain threshold is envisioned to be generated through reverberating circuitry between the ventral tegmentum/substantia nigra and ventral striatum/nucleus accumbens; modulated by limbic and cortical inputs that reflect past experiences, values, expectations, and salience relative to the self; where the output in turn modulates striatal-cortical loops to control behavioral repertoires. The threshold phenomenon emerges from a counterbalance between reward and aversion within the context of the learned history and the instantaneous state of the self that gates conversion from subconscious nociception to conscious pain (figure 2).

Figure 2.

Brain circuitry and temporal dynamics for the threshold phenomenon, θ, which determines conversion of nociception to conscious pain perception. a. The block diagram indicates θ is the output of the limbic brain. Internal states of the limbic brain, relative to neocortical memories determining current state of the organism (value, expectation, and salience), as well as the afferent nociceptive drive control θ. Other similar threshold processes in turn modulate the state of the organism through learning mechanisms, thus modifying values, expectations, and salience. b. More detailed circuit diagram emphasizing the interaction between limbic circuits, θ, and behavior selection. Diagram is adapted from studies of the reward/aversion circuitry, regarding striatal-cortical control loops (based on illustrations in (Luscher and Malenka, 2011; Nakanishi et al., 2014; Russo and Nestler, 2013)). Dense glutamatergic inputs from amygdala, hippocampus, and prefrontal cortex (mPFC) control the affective and motivational properties of accumbens (NAc) that responds to novel reward/aversion-related stimuli. Dopaminergic-GABAergic loops between accumbens and ventral tegmental area (VTA) provide the resultant value for θ, which through GP/SNr and thalamocortical circuits modulates behavior. Dopaminergic projections control synaptic properties and thus the affective state of the organism. c. Corticobasal ganglia-cortical loops conveying limbic, associative, and sensorimotor information. These loops are generally envisioned as a series of parallel projections. However the relay points, especially in the basal ganglia, provide opportunities for interactions between the loops. This organization enables the functional propagation of the limbic threshold phenomenon to influence goal-directed and habitual behaviors. (panle adapted from (Redgrave et al., 2010)). d. Conscious experience of acute painful events (P) depends on nociception (N) and the corticolimbic threshold, θ. e. Transitioning from subacute to chronic pain also depends on the individuals’ θ. Left panel depicts the classic viewpoint where nociceptive signal amplitude controls transition to chronic pain. Right panel is the view advanced here: For a similar injury, with equivalent nociception relayed to the brain, individuals with corticolimbic risk factors will persist to chronic pain while resilient ones will recover.

Once pain is present its salience draws attention, interferes with other thought processes, and imposes a state of negative mood. As such, the conscious perception of pain is the ultimate negative “cerebral celebrity” (Dennett, 1993) or negative “potential cerebral celebrity” (Chalmers, 1996), as it “‘perseveres’, ‘monopolizes resources long enough to achieve certain typical “symptomatic” effects – on memory, on control of behavior and so forth” (Dennett, 1993). If we view conscious perception as availability of information within the global workspace (Baars, 1988), then conscious properties of a painful state would affect even spinal nociceptive sensitivity and thus actively influence “gate control” and/or “central sensitization” spinal nociceptive processes, mediated through descending pathways as demonstrated by (Vera-Portocarrero et al., 2006).

Negative affect is an intrinsic aspect of the IASP standard definition of pain, and one commonly envisions unpleasantness as a necessary feature of this percept/qualia. It is classically explained as a consequence of transmission through medial spinocephalad pathways. If nociceptors are active in everyday behavior, and typically in the absence of pain, then transmission through both medial and lateral spinocephalad pathways must control subconscious nociceptor-driven behaviors, with or without the presence of pain. Human brain imaging studies tend to localize pain unpleasantness to specific cortical areas, most commonly either to rostral anterior cingulate (Rainville et al., 1997) or to some portion of the insula (Segerdahl et al., 2015). Yet these claims remain unconvincing. Instead, pain unpleasantness must be considered a part of the brain’s emotional repertoire that is tapped into by this particular qualia, which implies that the emotional limbic circuitry must be included in the neural network associated with pain. However, there is surprisingly little consistent evidence for activation of the amygdala, hippocampus, ventral striatum, or medial prefrontal cortex (the four most prominent nodes comprising the limbic brain) in acute pain studies (Segerdahl et al., 2015).

From the viewpoint of behavioral selection, we conclude that acute pain is not a warning signal but rather is the failure of the machinery (nociceptor activity) designed to avoid pain. As such once the conscious perception of pain occurs, aversion has failed or is imminent to fail. Thus the behavioral repertoire following pain is shifted into minimizing injury or retracting away from the environment that has the potential for injury, to protect the organism from further injury and promote healing. Therefore, here we extend the standard definition of pain to include the experiential shift as nociceptor activity breaks through from the subconscious to become a conscious unpleasant perception (figure 2d). It is unlikely that this phenomenon applies uniquely to the sense of pain (even for this sense the evidence remains minimal); rather, it likely generalizes across the sensory modalities. For example, there is extensive psychophysical evidence that large parts of the visual field do not enter into conscious visual perception, and complimentarily, visual perception is usually a coherent whole, the components of which may not even be physically present (Murray and Herrmann, 2013). Consistently there is now good evidence that activation of the primary visual cortex is not sufficient for conscious visual perception and that top down controls from the frontal cortex are necessary (Dehaene and Changeux, 2011). A similar top down control must also occur for pain. Moreover, we posit that the frontal cortical drives, across sensory modalities, are in fact embedded in corticostriatal circuits that actively control the threshold for incorporating sensory afferent inputs to cortical conscious states (figure 2a–c).

Patient H.M

There is a long and controversial history of the effects of brain injuries on pain perception (Nicholson, 2000). Unfortunately the data regarding pain insensitivity with brain lesions remains scant, localization of the site of injury is imprecise, and there is little documentation of the impact of such conditions on the everyday life of such patients.

Patient H.M. (Henry Gustav Molaison), arguably the most famous and most intensively studied patient in neuroscience, for over fifty years and by hundreds of scientists, did not feel acute thermal pain applied over diverse body parts (Hebben et al., 1985): “H.M. never gave reports of pain for the most intense stimulus applied either to his forearms or chest”. Thirty years prior to this formal testing for pain sensitivity, H.M. underwent bilateral resection of the uncus, amygdala, anterior hippocampus, and parahippocampal gyrus for severe and intractable epilepsy. The authors conclude that the amygdala injury is probably the main reason for lack of pain perception. What remains incontrovertible is the fact that H.M. had normal peripheral nociceptors and an intact spinothalamic pathway and, thus unperturbed sensory encoding of noxious stimuli. Therefore, the extensive bilateral limbic resection must have dramatically increased his striatal threshold for pain perception (θ in figure 2). Even more remarkable and further consistent with the latter explanation is that unlike patients with congenital insensitivity to pain, the loss of pain perception was not coupled with any obvious tissue injury, implying that subconscious nociception was intact and protecting his body. The contrast between H.M. and subjects lacking peripheral nociceptive afferents is informative, as the former seems to have led a life free of bodily injuries whereas subjects of the latter type are unable to protect the body from injuries even after extensive training.

Localization of brain activity for acute pain

Hundreds of PET and fMRI studies now describe a long list of brain regions activated during acute thermal, mechanical, and/or chemical stimuli in humans, see (Apkarian et al., 2005; Bushnell et al., 2013). Given this large literature, meta-analytic approaches can be used to summarize reproducible trends reliably found across laboratories and scanner modalities. A particularly elegant meta-analytic tool for neuroimaging data was developed by Yarkoni and Wager (Yarkoni et al., 2011) (www.neurosynth.org), which uses text-mining and machine-learning techniques to generate mappings between search terms and brain activity. The reverse inference map for the term “pain” illustrates the confidence with which pain-related brain activity is observed in 311 studies (Figure 3A). Assuming at least 10 subjects were used in each of the studies included, the obtained map represents neural activity associated with pain in more than 3000 subjects. The regions with most robust activity are highlighted and cover more than 15% of the neocortical mantle. Thus, in contrast to the single neuron electrophysiological results that describe sparse numbers of nociceptive responsive neurons in the neocortex (Chen et al., 2011b; Kenshalo and Isensee, 1983; Mazzola et al., 2009; Vierck et al., 2013), fMRI brain activity associated with pain identifies a much larger brain network. However, the latter map of pain-related activity, often referred to as “the pain matrix”, must be interpreted with caution, given numerous methodological limitations. These issues have been debated across many pain neuroimaging studies, and it remains unclear if they can be fully resolved at the level of local regional activity.

Figure 3.

Constructing the brain acute pain representation map from resting state brain activity. a. Brain regions identified for the reverse inference for the term “pain”, which identifies 311 PubMed studies in the Neurosynth meta-analysis tool (Yarkoni et al., 2011). The map is thresholded for z-values larger than 3.0. Highest confidence activations (z-values > 8.0) are localized to six brain regions: bilateral secondary somatosensory cortex (S2), anterior cingulate (ACC), bilateral anterior and posterior insula (aINS, pINS), thalamus (TH), and periaqueductal grey (PAG). b. Resting state functional connectivity networks for the six main nodes most robustly associated with the term “pain”. Functional connectivity is derived from resting state activity from 1000 subjects (Biswal et al., 2010), generated in Neurosynth (thresholded at correlation values > 0.3, approximately corresponding to > 3 standard deviations from baseline). Essentially the same network is identified when ACC, aINS, or S2 are used as seeds. The pINS seed identifies bilateral pINS as well as posterior cingulate/supplementary motor area. The TH network is limited to bilateral thalamus, and PAG seed only shows connectivity limited to itself. c. Overlap between the map for the term “pain” and sum of six resting state networks. Blue is the same map shown in panel a. Red is the sum of all functional connections identified in panel b. The overlap between red and blue maps is 72% of the blue map.

Even though we see some differentiation of properties for different brain regions activated with painful stimuli, especially when attentional and distraction-related variability is superimposed (Bushnell et al., 2013), the specificity of these regional properties remains far from clear. As a result, our ability to make inferences regarding their specific roles remains vague, see (Poldrack, 2011). In fact, the specificity of the nociceptive system or lack thereof has been the focus of a longstanding and contentious debate in pain research. Given that roughly 15% of the cortical mantle seems responsive to nociceptive stimuli, the chances that any of these regions are nociceptive specific is remote. Moreover, all brain regions shown in figure 3A also show responses to a multiplicity of other stimuli or tasks. Even at the level of the spinal cord, a large majority of nociceptive responsive neurons also respond to nonpainful tactile, proprioceptive, muscle and visceral manipulations, and this general principle seems maintained throughout the central nervous system. Furthermore, in a series of elegant studies, it has now been clearly demonstrated that almost all components of the nociceptive brain activity are not specific to nociception or to pain perception (Iannetti and Mouraux, 2010; Liang et al., 2013; Mouraux et al., 2011; Wang et al., 2010). This evidence overwhelmingly demonstrates that it is unlikely that the cortex contains neural tissue specifically responsive to nociceptive inputs or linked specifically to pain perception. Yet, the Cartesian view posits this exact hypothesis, and pain scientists continue to search for such specialized brain tissue.

Integration of brain activity for perception of acute pain

Since the work of Weber in the 18^th century (see English translation (Weber, 1978)), the primary quantitative outcome measure for pain has been its subjective perceived magnitude. It is therefore proper to study the brain activity for pain in relation to the stimulus intensity and/or perception magnitudes in order to functionally differentiate components of the underlying circuitry (Porro et al., 2004). In general, somatosensory cortical areas and anterior cingulate activity seem to track stimulus intensity, whereas lateral prefrontal and posterior parietal regions do not, e.g. (Buchel et al., 2002; Coghill et al., 1999; Davis et al., 1997; Peyron et al., 1999). Brain activity related to perceived magnitude of pain is studied less systematically (Baliki et al., 2009; Johnstone et al., 2012; Loggia et al., 2012; Moulton et al., 2012). Different temporal transformations are observed between stimulus and perception in various brain regions, and the region best related to perceived pain (anterior insula) seems to also reflect perceived magnitude for visual stimuli (Baliki et al., 2009). Thus, cortical tissue specifically dedicated to perceived magnitude of pain remains unidentified.

One can relax the Cartesian interpretation to include the notion that, despite a lack of cortical regional specificity, an appropriately weighted sum of the brain regional activity implicated in nociception/pain altogether could provide specific information regarding pain states. The latter was recently demonstrated in a rather comprehensive approach (Wager et al., 2013). The authors used brain activity for acute pain (based on the maps in figure 3A) to construct a machine learning-based regression model for pain perception. The model could differentiate between tasks and predict perceived magnitude of pain, accurately across several independent data samples. The strength of the study is the authors’ claim of having derived a generally applicable brain signature for identifying pain perception magnitude. The resultant tool can be readily tested to determine extent of validity, especially in tasks that activate very similar brain regions, for example in tactile magnitude rating (Apkarian, 2013) which gives rise to an activity pattern closely matching that seen in figure 3A.

An alternative approach is to examine the local pattern of brain activity and determine whether the across-voxel ensemble pattern can differentiate between conditions. This technique, multivariate pattern analysis pioneered by Haxby (Haxby, 2012), was used to differentiate between sensory modalities (Liang et al., 2013). The study was able to establish that within-subjects, across-voxel activity patterns in the somatosensory cortex best differentiated between pain and other sensory modalities. Yet, the study also demonstrated that many other brain regions can make a similar distinction, albeit at a lower confidence level. The authors further showed that the ability to distinguish between sensory modalities is not a unique trait of the sensory cortices. These results are somewhat consistent to the observations made by (Wager et al., 2013), in the sense that both studies identify a distributed signal in the brain that can differentiate pain from other states. Both studies (Liang et al., 2013; Wager et al., 2013) remain unclear as to the extent to which the identified signals are specifically linked to the perception, or if they reflect secondary or ancillary responses due to the presence of pain.

Brain resting state activity and acute pain

The discovery of the brain resting state activity (Biswal et al., 1995) has fundamentally changed modern notions of neural dynamics. In its most general form, resting state activity is a signature of neural oscillations synchronized across large scale networks that occur in the absence of external inputs (Fox and Raichle, 2007; Fox et al., 2005), persisting together with external inputs (Cabral et al., 2014), and reflecting local information integration as measured by frequency content (Baria et al., 2011; Baria et al., 2013). Properties of resting state activity are now thought to reflect an individual’s history of learning and memory that underlie perceptual variability (Lewis et al., 2009). Resting state activity is composed of a set of large-scale synchronized networks. Task related activity can be thought of either as a collection of networks that synchronize with each other (Deco and Corbetta, 2011), or components of these networks that break apart and link up with others. Figure 3B shows resting state networks identified from Neurosynth. The networks are derived for six voxels that correspond to the most pain-specific locations in figure 3A. Three of the six seeds essentially identify the same network; the thalamic seed only identifies bilateral thalamic connectivity, and the PAG seed identifies itself. The figure illustrates that any single voxel’s activity is in fact synchronized with, and thus embedded within, a large underlying brain network. Figure 3C illustrates that, with a few seed voxels, we can recapture most of the brain activity for pain. It is therefore not surprising that the properties of painful stimuli can be captured across many of the brain regions commonly identified for acute pain. More importantly, the identification of the pain-related network during (non-painful) resting state casts doubt as to the sufficiency of this network in pain perception.

Pain, vision, and their perception

There is far deeper understanding of the cortical encoding of vision compared to pain. Yet the fundamental mechanisms that transform sensory inputs to conscious perception remain equally mysterious for both senses. Here we briefly contrast the two in order to highlight their conceptual commonalities and differences.

Both electrophysiology and brain imaging studies indicate that about one-third of the primate neocortex is dedicated to visual information processing, which can be subdivided into more than thirty modules. The inter-relationship between these regions as well as their specialization has been and continues to be described in great detail (Felleman and Van Essen, 1991; Wandell and Winawer, 2011). In contrast, existence of cortical tissue specialized for processing nociceptive information remains contentious. Yet, the subjective intensity of qualia associated with each of these sensory modalities is at least equivalent, and the common saying “if you doubt reality, kick a rock with bare feet” implies that pain is often far more salient than vision. Thus, the intensity of saliency for specific qualia is not related to the amount of neocortical tissue dedicated to that sensory modality.

Additionally, although we have garnered detailed knowledge regarding the properties of visual information processed by individual neurons and information flow across the visual modules, we still lack the ability to construct the integrated visual percepts that one experiences holistically, for example see (Rokers et al., 2009). In similarity it is important to point out that existence of the meta-analytically derived map for acute pain (variations of figure 3a) with component regions preferentially encoding pain magnitude, affective characteristics, sensitivity to attention, emotion, arousal or to analgesics, is not sufficient to be concluded as the “pain matrix”/”signature”/”mechanism” for conscious perception of pain. It should be emphasized that even the scale (single molecules, single neurons, groups of neurons, networks of neurons, or the whole brain network) of the fundamental physical process of consciousness remains unresolved and mysterious. Still as our common subjective experience is comprised of qualia with access to language and influences behavior, “cerebral exuberance”, conscious experience of pain can only be properly understood within the full context of the nuances of consciousness, for which the minimal neural circuit must incorporate long-distance network interactions (Dehaene and Changeux, 2011).

Currently, the brain science of pain, in similarity to the science for all other sensations, remains correlative to psychophysics. In parallel with the recent dissociation of brain circuitry for conscious and unconscious visual stimuli (Dehaene et al., 2014), we suggest that important paradoxes regarding central pain representation can be resolved by unfolding unconscious nociception from conscious pain. Specifically, the seemingly irreconcilable evidence that pain is either localizable to specific brain sites (Garcia-Larrea and Peyron, 2013; Segerdahl et al., 2015) or it requires integrated representation across brain networks (Wager et al., 2013) may only be resolved once nociception and pain perception are delineated. Regarding an overall strategy to understand the neural signature for conscious pain we again quote Christof Koch: “We must resist the hypnotic appeal of hot spots in brain scans with their naïve phrenological interpretation: the perception of faces is computed over here, pain over there, and consciousness just yonder. Consciousness does not arise from regions but from highly networked neurons within and across regions.” (Koch, 2012).

Chronic pain

Chronic pain is an enormous health care issue with a massive price tag, yet its science has remained rudimentary. If pain persists and becomes chronic, it can lead to dramatically reduced quality of life, depression and suicide, insomnia, lowered immune function, changes in eating patterns, impaired cognitive function, maladaptive stress responses, and other long-term deleterious effects. Its prevalence has increased worldwide to affect more than 15% of the world population and 30% of the US population (Murray and Lopez, 2013). The associated health care cost in the US is a staggering $600 billion per year (Medicine, 2011). Over a third of individuals with chronic pain define their pain as severe and 40% of those suffering from chronic pain are not satisfied with their care (Breivik et al., 2006; Johannes et al., 2010). Despite considerable research on the topic, no consistently effective therapies have been identified.

Beecher first coined the term chronic pain in 1950’s. Additionally he emphasized that, in the clinical context, pain is far longer lasting and the relationship between pain and its inciting stimulus or injury remains imprecise and unpredictable, and thus that the science of pain and of analgesics must be centered on clinical trials performed in actual patients suffering from pain. The science of human neuroimaging of pain is also slowly converging to Beecher’s conclusion as accumulating evidence demonstrates that the brain in chronic pain is not equivalent to the brain experiencing prolonged acute pain.

The definition of chronic pain remains tautological, as it simply asserts that it is a long-lasting pain, or a pain persisting past the normal healing period. Over the last 30 years, studies in animal models of persistent pain have established that pain chronicity is associated with peripheral reorganization of afferent signaling, changing sensitivity for nociceptors, and perhaps also for tactile afferents (figure 1). At the level of the spinal cord, we now know of hundreds of molecular changes reorganizing neuronal circuitry and engaging glial processes, all of which give rise to heightened sensitivity of afferents and which generally conspire to give rise to “central sensitization” (Basbaum et al., 2009; Woolf and Salter, 2000). Accumulating brain imaging-based evidence now also shows that the human brain undergoes extensive reorganization in chronic pain conditions (Apkarian et al., 2011). This florid neural reorganization from the periphery to the neocortex (figure 1), which seems to uniquely manifest for different types of chronic pain, strongly agrees with Beecher’s viewpoint and affirms that, to understand chronic pain, one has to study patients suffering from its myriad clinical manifestations, comparing within and across types. The seemingly specific brain properties that are reliably linked with distinct chronic pain conditions, as well as the long-term and continued condition-specific reorganization of the brain across chronic pain diagnoses, justify the notion that chronic pain is a maladaptive neuropathological disease state (Davis and Moayedi, 2013; Tracey and Bushnell, 2009). Below we propose that chronic pain can be dissected into component phases and elaborate on the underlying mechanisms of each phase.

From a more general view, and given the model we have advanced regarding the role of striatal circuits in the conversion of nociception to acute pain, one can further expand our proposed model to chronic pain. Borrowing from the literature on mechanisms underlying addictive behavior for positive rewards (Robinson and Kolb, 2004; Schultz, 2000; Volkow et al., 2010; Willuhn et al., 2012), we affirm that long-term shifts in the threshold mechanisms that gate the conversion from nociception to pain also underlie the transition to chronic pain (figure 2e). We further propose that the threshold shift is dependent on limbic circuitry invoking synaptic learning-based reorganization (Apkarian, 2008; Apkarian et al., 2009). Taken together, these ideas can be simplified as a lowered mesolimbic threshold for the conscious perception of pain, which functionally renders the brain addicted to pain. The lowered striatal threshold is proposed to be mediated by learning mechanisms driven by limbic properties (Apkarian, 2008; Apkarian et al., 2009), which induce reorganization of neocortical memory traces (Johansen and Fields, 2004; Li et al., 2010; Xu et al., 2008).

The human brain in chronic pain – a general model

Ventral striatal circuitry links nociception, acute pain and chronic pain. This circuitry assesses salience of impending pain as well as expected reward value for relief of pain (Baliki et al., 2010). Given that its output controls motivated behavior (figure 2b,c), properties of this circuitry become critical in understanding the transition from acute to chronic pain. There is now evidence that some brain properties are candidate risk factors, while others reflect the transition to chronic pain, and the mesolimbic circuitry drives brain reorganization through synaptic learning mechanisms. These results suggest that chronification of pain may be subdivided into four temporally distinct and functional separate phases (figure 4). We presume that due to genetic and developmental forces different subjects are prone to developing chronic pain as a consequence of specific injuries. Simplistically, we assume these predisposing factors are captured by the limbic brain anatomy and physiology. Given such predispositions a specific injury initiating a large nociceptive barrage results in activating the corticostriatal circuitry into either a response that copes with the injury and in time recovers towards the healthy state, or a response that diminishes the corticostriatal threshold thus functionally amplifying the afferent signal, enhancing the gain for inducing learning, which in turn imprints novel neocortical anatomical and functional memory traces thus creating the chronic pain state (figure 2e). Recent human brain imaging data (figure 4a–h) and animal model evidence (figure 4:1–9) are consistent with this model. The model is also intentionally kept as simple as possible, because much detail remains to be uncovered. Note that it is fully consistent with observations of transient abolition of chronic pain by massive blockade of nociceptive afferents (Haroutounian et al., 2014; Vaso et al., 2014), since central amplification (by reducing mesolimbic θ) becomes immaterial in the complete absence of inputs. The model raises the more nuanced question: whether ongoing afferent nociceptive activity generated in chronic pain patients would by itself be perceived painful by healthy subjects.

Figure 4.

Transition to chronic pain may be deconstructed to four component phases: Predisposition, injury or inciting event, a transition period, and a maintenance phase. Brain circuitry and their interactions across the phases are illustrated in human brain imaging studies, a–h. a. Specific brain white matter regional properties (red) impart risk for developing chronic pain following an acute episode of back pain (Mansour et al., 2013). b. Limbic brain structural properties may also impart risk for pain chronification (e.g. shape and/or size of the hippocampus) (Mutso et al., 2012). c. In the transition phase, strength of information exchange between the prefrontal cortex and accumbens, after an end organ injury, determines long-term pain chronification (Baliki et al., 2012). d. The transition process is the influence of predisposing brain factors in combination with the injury-induced nociceptive signals that control mesolimbic learning mechanisms, altogether determining extent of prefrontal-accumbens information exchange (modulating θ in figure 2). Chronification of pain gives rise to: e. condition specific subjective pain-related brain activity patterns (Baliki et al., 2006; Hashmi et al., 2013; Parks et al., 2011), f. increased information exchange within the hippocampus and between the hippocampus and the cortex (Mutso et al., 2013), g. reorganization of brain grey matter regional similarity (Baliki et al., 2011), and h. distortions in information sharing in resting state brain activity, specifically brain activity phase relationship between the default mode network and the rest of the brain shows chronic pain type-specific patterns (Baliki et al., 2014b).

In rodent models for persistent pain, the four phases are better conceptualized as pre-injury manipulations that influence post-injury pain-like behavior, and early and late post-injury consequences. Supraspinal circuits implicated in the rodent four phases of pain persistence are highlighted in 1–9: 1. Bilateral lesion of the rat basolateral amygdala (BLA) diminishes post-injury tactile allodynia for 28 days after neuropathy (Li et al., 2013). 2. Lidocaine infusion within accumbens in the rat diminishes post-injury tactile allodynia for the duration of infusion (14 days), after a neuropathic injury (spared nerve injury, SNI) (Chang et al., 2014). 3. Hours following induction of an arthritis model in the rat, amygdala neurons become hyperexcitable (Neugebauer et al., 2003). 4. Five days after SNI neuropathy in the rat, accumbens covariance of receptor gene expression is upregulated (Chang et al., 2014), and 5. dendritic size and branchings of prefrontal pyramidal neurons are expanded (Metz et al., 2009). 6. Fifteen days after SNI neuropathy adult hippocampal neurogenesis is downregulated (Mutso et al., 2012). 7. Accumbens medium spiny neurons with dopamine D2 receptors show decreased AMPA/NMDA ratio in neuropathic injured rodents (Schwartz et al., 2014). 8. Resting state whole-brain functional network in the anesthetized rat shows increased (red) and decreased (blue) functional connections 28 days after SNI neuropathy relative to sham injury (Baliki et al., 2014a). 9. Six months following neuropathic injury prefrontal (PFC) cortical grey matter volume is decreased in the rat (Seminowicz et al., 2009).

Overall, the human data illustrates brain risk factors for, and brain reorganization with, chronification of pain. The rodent results show persistent painlike behavior is dependent on, and in turn reorganizes, limbic brain circuitry.

Our model posits that chronic pain is dependent on the interplay between the brain threshold phenomenon and the injury-related sensory input. The injury in most cases is envisioned to be a disturbance in nociceptive afferent input but in some conditions there may also be central drivers, such as in central pain or phantom pain conditions. The pain research community has long debated the relative contribution of the end organ (injured bodily structure) in relation to brain or gene predispositions (Robinson, 2009). The proposed model is a combination of both and it is envisioned that the relative weights of each component will be condition specific. For example, fibromyalgia seems to be mostly driven by central predispositions (Phillips and Clauw, 2011), even though hyper-excitable nociceptors were recently described in such patients (Serra et al., 2014). On the other hand, osteoarthritis may have a larger peripheral nociceptor contribution, as evidenced by the success rate of pain relief with joint replacement surgery (Buchbinder et al., 2014). The model also suggests that the rate of transition to chronic pain is condition specific and dependent on limbic brain properties. We foresee that unraveling the specifics of this circuitry for various types of chronic pain conditions will pave the way for the development of novel treatment and/or prevention therapies. As this model assumes that brain properties are the primary determinants of risk for chronic pain, chronic pain is more dominantly defined as a neurological disease and to a lesser extent a nociceptive abnormality.

The human brain in chronic pain – Anatomy

About ten years ago we discovered regional anatomical brain abnormalities that correlated with intensity and duration in chronic back pain (CBP) patients (Apkarian et al., 2004). This initial observation is now replicated across many clinical pain conditions that primarily show regional decreases in grey matter density, although there is also some evidence for increased density in a subset of pain populations, as well (May, 2008); see recent meta-analyses (Cauda et al., 2014; Smallwood et al., 2013). A direct comparison between different clinical pain conditions indicates partially overlapping maps of anatomical aberrations (Baliki et al., 2011). Mechanisms and processes underlying such changes remain unclear and speculative. Growing evidence confirms that these regional gray matter decreases can partially renormalize following successful amelioration of chronic pain (Ceko et al., 2015; Gwilym et al., 2010; Moayedi et al., 2011; Rodriguez-Raecke et al., 2009; Seminowicz et al., 2011). The simplest hypothesis regarding physiological processes that control these anatomical changes is the notion that voxel-wise gray matter properties (average profile of millions of neurons) reflects local variations in synaptic density, although neuronal atrophy may also be occurring as some of these changes persist over decades (Baliki et al., 2011).

There is also evidence that these regional gray matter changes are actually a reflection of a more pervasive reorganization of the inter-relationship of the anatomy of the neocortical mantle (based on whole-brain gray matter self-similarity analyses (Baliki et al., 2011)), which is highly distinguishable between clinical conditions and which suggests a reorganization of information shared across the neocortex. A meta-analysis of brain gray matter properties for multiple pain conditions when analyzed from the viewpoint of resting state networks, identifies a core set of networks (mainly the salience and attentional networks) as the main circuits commonly affected, while sensory regions seem to show more condition specific reorganization (Cauda et al., 2014). In our longitudinal study, where sub-acute back (SBP) patients were tracked over a year in their transition either to recovery (SBPr) or persistence to pain chronification (SPBp) indicated that gray matter decreases occurs only in the SBPp, with changes starting within the first few months, in proportion to functional connectivity changes, and in proportion to the intensity of back pain (Baliki et al., 2012). All of which implies strongly that this anatomical reorganization is part of the process of the transition to chronic pain.

Brain white matter abnormalities have also been observed in chronic pain conditions. The first such report suggested that white matter regional fractional anisotropy (reflecting myelin properties) may be linked to the grey matter changes (Geha et al., 2008), implying shared mechanisms between the two processes. A number of studies now show regional white matter abnormalities in diverse chronic pain conditions (Chen et al., 2011a; Ellingson et al., 2013; Khan et al., 2014). On the other hand, in our longitudinal study of SBP, it was observed that white matter fractional anisotropy differences between SBPp and SBPr were present in the earliest brain scan and persisted with no further changes over a one-year period (Mansour et al., 2013). The latter study concluded that specific white matter deviations from the norm are most likely preexisting risk factors that put subjects at risk to developing chronic back pain.

Perception related brain activity in chronic pain

The Cartesian expectation for brain activity in chronic pain would be a state of continued, or enhanced, activity in brain regions identified for acute pain (all or a subset of the regions seen in figure 3a). However, when brain activity is determined for subjective report of spontaneous fluctuations in the magnitude of perceived pain (Baliki et al., 2006), one observes specific brain activity for chronic pain distinct from that for acute pain, where chronic pain conditions activate more limbic and emotional brain regions, and that different chronic pain conditions engage specific patterns (Apkarian et al., 2011). Results from our longitudinal study, in fact, demonstrates how brain activity in relation to subjective pain shifts dynamically away from sensory brain regions to emotional/limbic regions (Hashmi et al., 2013). In early SBP (10–15 weeks after start of back pain), brain activity for back pain closely corresponds the activity for acute pain. However, as subjects transition to either SBPp or SBPr (one year later), brain activity diverges between the groups, with SBPr showing minimal brain activity (below detection threshold) while SBPp shows decreased activity in sensory regions and increased activity in the medial prefrontal cortex and amygdala. The latter spatial shift seems to occur even though these subjects judge their back pain as essentially unchanged over the one-year monitoring period. Thus, it seems that chronification of pain, accompanied by gray matter and functional connectivity reorganization, and also accompanied with reduced capacity to activate central opioid neurotransmission (Martikainen et al., 2013), renders the pain to become more subjective/intrapersonal and more emotional.

Chronic pain and resting state brain activity

If chronic pain underlies neocortical anatomical reorganization and functional connectivity changes, then it should be reflected in the properties of resting state activity. Indeed, there is growing reproducible evidence that resting state activity in many chronic pain conditions show a variety of abnormalities (Baliki et al., 2008; Baliki et al., 2014b; Bolwerk et al., 2013; Cauda et al., 2010; Cauda et al., 2009; Gupta et al., 2015; Loggia et al., 2013; Malinen et al., 2010). The most consistent result is a disruption of the default mode network, specifically a dissociation of its prefrontal component in many different types of chronic pain. There is also consistent evidence of increased high frequency oscillations, mainly in the prefrontal cortex, as well as disruption of insular cortex functional connectivity. How similar are these changes across chronic pain types remains unknown, and how these changes are related to the myriad other changes observed in the neocortex remains to be uncovered.

Resting state activity technology promises to become a dominant modality with which brain information exchange properties can be probed for chronic pain as it allows to probe the properties of both local and global functional properties in the naturalistic setting of unperturbed state of chronic pain. It should be emphasized that only by studying large groups of chronic pain patients we will be able to gain more comprehensive insight into underlying mechanisms. Such endeavors are complicated and hard to accomplish within single research labs, and requires collaborative effort across centers. The first such consortium has been ongoing for the study of pelvic pain and initial multi-center results are just being published (Farmer et al., 2015; Kairys et al., 2015; Kilpatrick et al., 2014).

Predicting transition from acute to chronic pain

The extent to which transition to chronic pain can be predicted from brain properties is a very important issue as it paves the way for personalized evidence-based medicine (Denk et al., 2014). Brain white matter properties seem one such predictor (Mansour et al., 2013), which when measured within weeks after the inciting event, predict SBPp and SBPr one year later at 80–100% accuracy. Another predictor identified from the same longitudinal study (Baliki et al., 2012) is the corticostriatal functional connectivity. The latter is constant and stronger in SBPp than SBPr over a one-year period, and at time of entry into the study could predict chronification of pain at about 80% accuracy. These results indicate that limbic brain properties and its responses to the injury is the primary determinant (they explain almost all of the variance of the outcome parameter) for transition to chronic pain, at least for back pain. It should be noted, however, this study remains one of a kind and thus awaits replication in other chronic pain conditions.

Animal studies regarding chronic pain

Animal studies have failed to show critical controllers regarding transition to persistent pain-like behavior. There may be multiple technical reasons for this failure. However, from the viewpoint of the current perspective, we can assert that these studies generally have not differentiated between nociception and pain, and also ignored much of the rest of the brain, especially the limbic brain. More recent literature is now filling these gaps (figure 4:1–9). There is now good evidence of the critical role of the amygdala in multiple animal models of pain, where its properties seem to modulate even spinal cord central sensitization processes (Li and Neugebauer, 2004), and influences prefrontal activity (Ji and Neugebauer, 2011). Additionally, the dendritic size and spine density of pyramidal neurons in the medial prefrontal cortex change within days after a peripheral nerve injury (Metz et al., 2009), and long-term neuropathic pain decreases prefrontal cortical grey matter density in the rodent (Seminowicz et al., 2009), just as in humans. Hippocampus volume in humans with chronic pain is smaller (Mutso et al., 2012), and with transition to chronic pain shows changes in information exchange within the hippocampus as well as between the hippocampus and the neocortex (Mutso et al., 2013). Consistently, following a peripheral nerve injury rodents show deficits on hippocampal-dependent memory extinction tasks and exhibit abnormalities in information processing at the level of single neuron electrophysiology (Mutso et al., 2012; Ren et al., 2011). Very recent studies also show that peripheral nerve injury modulates functional connectivity of NAc and decreases expression of dopaminergic receptors in the ventral striatum (Pei-Ching Chang, 2014), and disturbs glutamatergic information processing and long term depression of ventral striatal neurons, resulting in decreased motivated behavior (Schwartz et al., 2014).

These recent animal are providing evidence that parallel the human brain imaging studies. Additionally, the rodent studies begin they are unraveling the detailed brain circuit properties associated with transition to chronic pain, pointing to potential novel therapies (Centeno et al., 2009; Millecamps et al., 2006; Schwartz et al., 2014). From the viewpoint of the model for transition to chronic pain the rodent results emphasize more the reorganization of the limbic brain with transition to chronic pain-like behavior. Yet, these are relatively early studies and much remains to be done in the field.

Mechanistic parallels between stress, anxiety, depression and chronic pain

Ample evidence supports the notion that the most prevalent clinical manifestations of negative emotion, anxiety and depression, reflect a common spectrum of symptoms with overlapping mechanisms (Watson, 2005). Chronic stress, in particular, has emerged as a dominant and common underlying factor. Here we propose that our framework regarding nociception and pain relative to behavior selection (figure 5a,b) can be extended to incorporate negative moods (see (Coenen et al., 2011) for a somewhat different formulation for interpreting depression as a type of pain).

Figure 5.

Nociception, pain, and negative moods constitute a continuum imparting inhibition of behavior through negative affect, based on expected or apparent inputs across varying spatial and temporal dimensions. The four landscapes (a–d) illustrate negative emotional value assignment relative to the individual (the contemplative Cartesian self). Zero on this space-time plane represents either the body in relation to sensory inputs, or equivalently the self within the arena or the global neural workspace of consciousness, where accumulated or experienced aversiveness is assigned for varying space-time relationships that dictate behavioral selection. Hot colored valleys represent negative affective states or valuations, blue-white undulations signify emotionally more neutral states. a. In the absence of an experienced or expected threat (e.g. while kneeling to smell the roses) nociception in the absence of negative affect subconsciously protects the organism from injury by constraining behavioral repertoires (delimiting bodily positions or postures). b. Failure of nociception results in conscious pain (burning the skin of the Cartesian self by the fire), associated with a rapid withdrawal from the environment. Thus, pain evokes conscious negative affect and behavioral modification at the scale of the immediate body vicinity (aversion at zero space-time). c. When the threat is a learnt association, and is expected to be encountered at a distance or time removed from the body, then the subject experiences anxiety or stress. d. If instead the threat is experienced as, or expected to be, pervasive the associated negative mood is more abstract, described as depression, and the behavioral inhibition is generalized across scales of time and space. As pain is a primary reinforcer, its presence or persistence can rapidly become associated with expanded aversive landscapes, incorporating various combinations of the landscapes b–d, which is complimentary to the imprecision model recently proposed for chronic pain (Moseley and Vlaeyen, 2015). In this framework, we posit that the four phases of transition to chronic pain (illustrated in figure 3) also apply to chronification of negative moods. Both specific chronic pain conditions and the variety of types of chronic negative moods are expected to have unique limbic predisposition signatures and long-term brain adaptations. Computations needed for constructing these cognitive aversiveness maps are variants of Sutton and Barto’s (Sutton and Barto, 1981) temporal difference algorithm, applied to dopaminergic activity for assimilating reward prediction error to induce approach behavior (Schultz et al., 1997), which can also be recast as a Bayesian inference that optimizes energy based on model evidence (Friston et al., 2014).

Just as pain motivates the avoidance of further bodily injury and promotes behaviors that enhance healing (figure 5b), anxiety can be recast as an emotional state, sustained by sympathetic arousal that promotes behaviors that diminish anticipated danger within one’s immediate physical space and at relatively short future time scales (figure 5c). Moreover, depression can be conceptualized as a more global generalization of perceived aversiveness to one’s environment. In such a case, perceived or anticipated danger reflects a more abstract level of cognition that results in constraining personal space (through social isolation, reduced physical activity, and diminished motivated behavior) (figure 5d). Thus, just as nociception and pain protect against bodily injury by limiting behavior, negative moods minimize exposure to danger and promote survival by inhibiting behavior as well. Moreover, in similarity to chronic pain, persistence of negative moods becomes a maladaptive process, at least partially maintained by neuropathological mechanisms.

Within this framework, brain mechanisms underlying the transition from acute to more persistent negative mood states should parallel those we describe for chronification of pain. In fact both animal model studies and human brain imaging studies show strong similarities between mood disorders and chronic pain, and both conditions critically involve limbic brain circuits. Most importantly, the structural and functional alterations in the ventral tegmental-ventral striatal circuitry associated with anhedonia (Russo and Nestler, 2013) is consistent with the threshold phenomenon we have discussed for pain. Just as chronic pain conditions are associated with decreased hippocampal volume (Khan et al., 2014; Mutso et al., 2012), a rich parallel literature indicates that depression is associated with hippocampal volume decrease and decreased synaptic and glial density, based on brain imaging and post-mortem evidence (Brown et al., 2014; Campbell et al., 2004; Czeh and Lucassen, 2007). More equivocal evidence shows decreased volume of the amygdala (Hickie et al., 2007; Whittle et al., 2014) and medial prefrontal cortex (Caetano et al., 2006; Drevets et al., 1997; Rajkowska, 2000) in humans with depression. Major depression in the adolescent is now tightly related with decreased information sharing between the amygdala and the hippocampus (Cullen et al., 2014), whereas decreased information sharing is seen between the hippocampus and the neocortex with chronic pain (Mutso et al., 2013). Moreover, medial prefrontal cortex connectivity to the nucleus accumbens has become a primary neurosurgical stimulation target for treating intractable depression (Mayberg et al., 2005; Ressler and Mayberg, 2007) with the intention of modulating properties of the corticostriatal circuit. Decreases in hippocampus and amygdala volumes have also been described in post-traumatic stress (PTSD) (Chao et al., 2013; Chao et al., 2014; Gilbertson et al., 2002; Starcevic et al., 2014), and white matter microstructural predispositions in PTSD indicate that such structural differences reflect long-term vulnerability (Sekiguchi et al., 2014), as also observed for chronic back pain (Mansour et al., 2013). In humans, amygdala response properties seem to indicate risk for developing PTSD (McLaughlin et al., 2014), and in rodents, susceptibility to stress response is dependent on hippocampal volume and functional processing (Nalloor et al., 2014; Tse et al., 2014). Chronic tinnitus, a persistent unpleasant sensation of ringing or buzzing in the ear, is also now characterized as a dysregulation of the limbic network, mainly due to hyperactivity of the ventral striatum coupled with decreased grey matter in the medial prefrontal cortex (Leaver et al., 2011). The latter result is highly consistent with our notion that would explain tinnitus too as a shift in mesolimbic threshold for conscious perception of painful/unpleasant sounds mediated through either a peripheral injury (rock concert or rave) or central events (stress), and coupled with limbic predispositions. The configuration of these predisposing factors, the inciting event(s) (i.e., bodily injury, traumatic experiences), arousal-enhanced aversive learning, and long-term maintenance of these maladaptive limbic memory traces likely contribute to the broad range of phenotypic expressions that are used to differentiate clinical diagnoses.

Overall, there seems to be a remarkable overlap between the brain structures that either impart vulnerability or are affected by pain chronification and pathological negative moods. It is therefore not surprising that these conditions are often comorbid, and indeed, there is now a small but emerging literature regarding the interaction between negative moods and acute and chronic pain (Jensen et al., 2012; Lopez-Sola et al., 2010; Mutschler et al., 2012; Rodriguez-Raecke et al., 2014; Schweinhardt et al., 2008; Strigo et al., 2013). So far, the most compelling evidence is the observation that, in pessimistic subjects, ventral striatal activity for anticipation of pain relief (Leknes et al., 2011) corresponds with the abnormal phasic activity observed for expected pain relief in the same brain region in chronic back pain patients (Baliki et al., 2010). Thus, the conceptual contiguity between pain and negative moods illustrated in figure 5 may also be extended to their corresponding persistent pathological states.

Despite their broad overlap, there is danger in oversimplifying the mechanistic parallels between pain and negative moods; we suspect that limbic brain properties will be differentially configured with chronic pain conditions in contrast to different types of persistent negative moods. For example, the deep brain stimulation site used for treating depression is more orbitally located than the medial prefrontal cortex that correlates with subjective fluctuations in chronic back pain and that also predicts pain chronification. There is also good evidence that negative moods and chronic pain can co-exist, without interacting with one another (Jensen et al., 2010).

Conclusions

The pain research community has made extensive efforts to establish the presence and identify properties of the nociceptive system. This effort has been quite successful, yet in the process, the contribution of the emotional brain on pain perception has received little serious attention. The general notion that pain can be understood in the context of behavior selection and its underlying mechanisms are grounded in limbic brain properties is not a new idea. In fact, the initial formulation of the concept of the limbic brain by MacLean (MacLean, 1955), which first characterized the main components of the brain that generate emotional responses to the environment, stated that pain expression was dependent on this circuitry. Subsequently Melzack and Casey (Melzack and Casey, 1968) expanding on the gate control theory put forward the “motivational and central control” model for pain, and stated “that pain is comprised of both sensory and affective dimensions was clear to Sherrington,” and quote Sherrington’s assertion that “… affective tone is an attribute of all sensation, and among the attribute tones of skin sensation is skin pain” (Sherrington, 1900, p. 974). Melzack and Casey concluded that pain must engage limbic brain, and specifically the hippocampus and amygdala. Even though this conceptually seminal paper has been cited more than 1300 times in the literature, its basic concepts regarding the role of limbic circuitry in pain has failed to advance until the advent of studies discussed here. For example, in a model of pain proposed about 20 years later, the author simply puts a question mark next to the contribution of the limbic brain to the affective and motivational aspects of pain (Price, 1992).

We have provided a broad range of evidence that nociception, pain and negative mood states can be viewed as a single continuum of aversion, within the framework of behavior selection. Further we suggest that the chronification of such states can be conceptualized as being composed of four distinct phases that are contingent on limbic predispositions; the precise mechanistic details, especially regarding pain, remain to be unraveled. Our synthesis of this rapidly accumulating evidence, both in human brain imaging studies and in rodent models, provides compelling evidence that pain perception, as distinguished from nociception, is part of a continuum of aversive behavioral learning that manifests as pain, anxiety, or depression over time, based on preexisting vulnerabilities dictated by emotional learning and the physical proximity of the perceived source of danger.