Endogenous opioids are necessary for the expression of pain relief (17) and pain-induced aversion (18, 19). Blocking opioidergic transmission reduces dopamine release in the nucleus accumbens that accompanies pain relief (20). The endogenous opioid system consists of four opioid peptide families: β-endorphin, enkephalins, dynorphins, and nociceptin/orphanin and 4 families of receptors: mu, delta, kappa, and nociceptin (21, 22).

Opioid receptors are expressed by central and peripheral neurons, by neuroendocrine (pituitary, adrenals), immune, and ectodermal cells (23–25). All opioid receptor types mediate analgesia but have differing side effects, mostly due to their variable regional expression and functional activity in different parts of central and peripheral organ systems. Endogenous opioids are particularly concentrated in circuits involved in pain modulation (26).

Beta-endorphin levels in the CSF, blood and saliva have been investigated as possible pain biomarkers. Plasma Beta-endorphin has been used to investigate age responses to experimental pain (27). Patients with chronic neuropathic pain due to trauma or surgery have been shown to have lower levels of Beta-endorphin in the CSF (28). Plasma and CSF Beta-endorphin have been investigated in patients with trigeminal neuralgia (29). Interestingly, Beta-endorphin in peripheral blood was related to levels in CSF; furthermore, the levels of Beta-endorphin were inversely correlated with the severity of pain symptoms (29). While chronic low back pain typically involves non-neuropathic pain mechanisms, it is interesting that plasma Beta-endorphin levels have been shown to be a promising biomarker for chronic back pain (30). In other non-neuropathic pain conditions, mu opioid receptors expressed on immune B cells was found to be a biomarker for chronic pain in fibromyalgia and osteoarthritis. In this study, the percentage of mu opioid receptors positive B cells was statistically lower in patients with moderate to severe pain than in pain-free subjects or mild pain subjects (31). In a heterogenous group of patients with pain, a composite biomarker was identified that uses emergent properties in genetics to separate patients with pain requiring extremely high opioid doses from controls (32). Negative studies for opiate mechanism pain biomarkers have shown that salivary Beta-endorphin is not a biomarker for neuropathic chronic pain propensity (33).

Functional brain imaging performed on patients with non-neuropathic primary dysmenorrhea with mu-opioid receptor A118G polymorphism has been used to investigate pain sensitivity and opioid-analgesic treatment related to function in the descending pain modulatory system. Specifically, the functional connectivity of the descending pain modulatory system dependence upon mu-opioid receptor A118G polymorphisms was investigated. This study found that patient groups with different alleles for the A118G polymorphisms exhibited varying functional connectivity between the anterior cingulate cortex and periaqueductal gray (34).

Although magnetic resonance imaging provides information regarding structural and metabolic changes that provide insight into pain perception of the CNS, magnetic resonance imaging cannot image opioid function in cells in vivo at the molecular level. Such important opioid function information can be obtained through positron emission tomography and can be used to investigate pain opioid mechanisms (35).

While the development of neuropathic pain has long been ascribed to the known contributors of central sensitization (action potential kinetics, phenotypic transformation, receptor density zone reorganization and long-term potentiation), the role of neuroinflammation regarding the initiation and maintenance of neuropathic pain has evolved tremendously over the last decade. Pro-inflammatory cytokines have been implicated in the generation of neuropathic pain states at both peripheral and central nervous system sites (36, 37). Neuroinflammation of the peripheral nervous system is triggered by inciting damage to the peripheral nerves, either by trauma, metabolic disturbances (i.e., diabetes), viral infection (i.e., HIV) or surgical lesions leading to sprouting of new pain-sensitive fibers (such as A-β fibers), excessive neuronal firing, and hypersensitization of primary afferent peripheral neurons. During a peripheral nerve injury, local cytokines recruits macrophages which secrete components of the complement cascade, coagulation factors, proteases, hydrolases, interferons, and other cytokines that ultimately facilitate degradation and phagocytosis of the pathogen and injured tissue. Peripheral neuroinflammatory mechanisms affect the damaged neuron and neighboring afferent neurons sharing the same innervation territory (38–45).

Peripheral nerve injury causes neuroinflammation in the spinal cord (41, 46–48). The neuroinflammation is triggered by hyperactivity of the injured primary afferent peripheral sensory neuron which increases neurotransmitters and neuromodulators, causing hyperactivity of postsynaptic nociceptive neuronal hyperactivity as well as the release of several inflammatory activators. A result of this lumbar spinal inflammation process is disruption of the blood-spinal cord barrier leading to increased permeability, which then leads to infiltration of immune cells such as T lymphocytes, macrophages, mast cells, and neutrophils from the periphery into the spinal cord and dorsal root ganglion (46). These mechanisms contribute to further release of inflammatory mediators which contribute to alterations in post-synaptic receptors. This neurotransmitter increase leads to hyperactivity of post-synaptic nociceptive neurons in the spinal cord and altered signaling up to the thalamus and cortex that may contribute to central sensitization and pain hypersensitivity (41, 45, 49). Nerve injury typically involves neuro-immune interaction involving glia (50–52). Glia are known to provide functional microenvironment modulating neuronal signal transduction, synaptic pruning, and neuroplasticity that contributes to central sensitization.

There are three classifications of cannabinoids: phytocannabinoids (plant-derived), endocannabinoids (in human or animal tissues), and synthetic cannabinoids. Similar to the opioid system, versions of the endocannabinoid (ECB) system have been found in the vast majority of species with a nervous system (65). In particular, the ECB ligands 2-AG and AEA have been found throughout the animal kingdom (66). The ECB system regulates physiology across most organ systems and operates independently and interacts with the inflammatory system, the opiate system, the Vaniloid (TRP) system, and with nuclear transcription factors (67–72). The ECB system works as a part of a negative feedback loop that regulates neurotransmitter and neuropeptide release in the nervous system. Endocannabinoid ligands are generated on-demand in response to high levels of activity and produce short-term inhibitory effects via their actions as retrograde transmitters at presynaptic inhibitory G protein-coupled receptors (2).

The two most prevalent endocannabinoid ligands that bind endocannabinoid receptors are anandamide (AEA) and 2-arachidonoylglycerol (2-AG) (67). The 2-AG basal level is ~1,000 times greater than AEA in the brain (73). The enzyme acylphosphatidylethanolamine-phospholipase D is involved in the formation of AEA, and the enzyme diacylglycerol lipase is involved in 2-AG formation (74). Once synthesized and released, endocannabinoids are removed from the extracellular space through an endocannabinoid membrane transporter, subsequently AEA is hydrolyzed by the enzyme fatty acid amide hydrolase (FAAH), and 2-AG is degraded by cytosolic monoacylglycerol lipase (MAGL).

There are two G protein-coupled ECB receptors (CBR1 and CBR2). CBR1 receptors are highly expressed on presynaptic neurons in the brain, spinal cord, and dorsal root ganglion. CBR2 receptors are primarily expressed in immune cells (including myeloid, macrophage, lymphoid, and mast cells) (75). AEA maintains basal endocannabinoid tone and has a high selectivity for the CBR1 receptor over the peripheral CBR2 receptor.

The CBR1 receptor is the most abundant G protein-coupled receptor in the brain and one of the most abundant in both the peripheral and central nervous system. CBR1 are expressed primarily on presynaptic peripheral and central nerve terminals. CBR1 is the central receptor responsible for the behavioral and psychotropic effects of the “high” caused by THC (65, 72). The CBR1 receptor is also present in multiple immune cells, making it important when considering anti-inflammatory properties of endocannabinoids (68, 75).

The CBR2 receptor is largely present on peripheral immune cells and participates in regulation of the immune system. The principal endogenous ligand for the CBR2 receptor is 2-AG. In the brain, stimulation of CBR2 receptors does not produce cannabis-like effects (72, 74). The anti-inflammatory effects of exogenous cannabinoids are mediated by the endocannabinoid system, likely through CB2Rs in the periphery that have immunomodulatory functions (76).

In neural and non-neural systems, in response to tissue injury or excessive nociception, the ECB system generally suppresses inflammation, suppresses sensitization, and suppresses pain (68, 71, 72, 77); however, ECB activity on the nociception system can be complicated, with ECB antinociceptive or pronociceptive depending on the site of expression and the underlying physiological brain state (2).

ECB biomarkers have been used to monitor neuropathic pain. Increases in circulating AEA concentrations occur in patients with neuropathic pain complex regional pain syndrome (CRPS) when compared to controls without pain (78). Increases in circulating 2-AG concentrations have also been reported in neuromyelitis optica (79). Circulating concentrations of AEA and 2-AG have been shown to correlate with the numbers of daily severe headaches (80). ECB biomarkers also are present for non-neuropathic pain conditions: bladder pain (81), fibromyalgia (78, 82), cold pain sensitivity (83), osteoarthritis (84), knee pain (85), and back pain (86). Clinical Cannabinoid Deficiency Syndrome has been linked to migraines, neuromuscular pain, and gastrointestinal disorders (87).

When considering ECB pain biomarkers, it is important to consider that the ECB system interacts significantly with inflammation mechanisms as well as opioid mechanisms. The ECB system regulates inflammation at multiple levels and generally inhibits inflammation. In preclinical and laboratory investigation, it has been shown that the ECB system inhibits pro- inflammatory cytokines while increasing production of anti-inflammatory cytokines. The ECB system also inhibits immune cell activation, immune cell proliferation and migration, and can increase immune cell apoptosis via multiple mechanisms (23, 67, 88–93).

CBR1 receptors are 10 times more concentrated than mu-opioid receptors in the brain, and cannabinoid receptors co-localize with opioid receptors in many regions involved in pain circuitry including the dorsal horn of the spinal cord and in the supra-spinal periaqueductal gray and rostro-ventral medulla (67, 72, 94). ECB and opioid pain biomarkers potentially can be identified via endogenous opioid function which can be assumed when comparing pain sensitivity in the presence of opioid blocking treatment, such as naloxone vs. placebo. Higher endogenous opioid function is associated with decreased benefit from opioid treatments such as morphine (95). Exercise induced analgesia involves both opioid and endocannabinoid mechanisms (96). Exercise induced increases of endocannabinoid ligands such as AEA are blocked by naltrexone. This indicates that opioids are involved in the increase of endocannabinoid ligands following exercise (97). Intrathecal morphine decreases circulating levels of endocannabinoids demonstrating how the opioid and endocannabinoid systems are linked (98). Lower endogenous opioid function is associated with greater analgesia from pain treatment with opiates. One study demonstrated that low endocannabinoid activity is also associated with greater analgesia from pain treatment with opiates (99).

Pathogenic alterations in the distribution of microbial species within the gut (gut dysbiosis) is associated with neuropathic pain in a variety of clinical conditions. One study found that reductions in the diversity and increases in the ratios of two microbial species (ratios of Blautia and Clostridium to Lachnospira) may contribute to HIV-associated neuropathic pain (100). This may be particularly relevant in the context of the endocannabinoid system, as the endocannabinoid system regulates homeostasis of multiple organ systems, including the gut. Because dysregulation of the gut-brain axis can result in chronic inflammation and neuroinflammation, endocannabinoids have anti-oxidant, and anti-inflammatory properties relevant to modulation of inflammation that occurs along the gut-brain axis (101, 102).

In a systematic review and meta-analysis of 29 studies on potential genetic variants associated with neuropathic pain identified (103), 28 genes were significantly associated with neuropathic pain, many involved in neurotransmission, immune response, and metabolism. Genetic variants in HLA genes, COMT, OPRM1, TNFA, IL6, and GCH1, were found to have an association with neuropathic pain in more than one study. In the meta-analysis, polymorphisms in HLA-DRB1^*13, HLA-DRB1^*04, HLA-DQB1^*03, HLA-A^*33, and HLA-B^*44 were associated with significantly increased risk of developing neuropathic pain, whereas HLA-A^*02 reduced risk of neuropathic pain.

To detect genetic associations, particularly those of small effect size, a study must be sufficiently statistically powered to detect those differences. Most genetic studies of neuropathic pain have typically analyzed cohorts with <1,000 cases, which has resulted in only suggestive associations (103, 104). One reason that genetic studies in neuropathic pain lack sufficient sample sizes is the costs associated with studying these large cohorts (105).

Biomarkers in saliva may be useful as they are easily measurable without requiring a needlestick or invasive methods (135). Salivary biomarkers, such as salivary cortisol, salivary α-amylase, secretory IgA (sIgA), testosterone, glutamate, or tumor necrosis factor receptor type II (TNF-RII) has been proposed as possible pain biomarkers (136–138). In particular, sIgA and TNF-RII as useful salivary markers of pain given their high intra-individual reproducibility (139, 140).

In the past decade, the focus of brain imaging investigation of pain mechanisms has shifted from investigating individual regions of the brain to investigating brain circuits, see Figure 1. One of the most important brain circuits is the default mode network which is associated with daydreaming (154, 155). The executive network is the brain circuit used when not daydreaming but instead attending to the outside world (156). The ascending pain network includes the anatomical pathway that conveys the nociceptive input from the peripheral nervous system to the spinal cord and the brain (157). The descending modulation network involves brain regions that connect to the brainstem and then down to the dorsal horn and increase or decrease ascending pain signals depending on the behavioral state of the individual (158). The salience network is a large-scale brain network of the human brain that is primarily composed of the anterior insula and dorsal anterior cingulate cortex. It is involved in detecting and filtering salient stimuli, as well as in recruiting relevant functional networks (157, 159).

Individuals have a set point around which different biological attributes can fluctuate transiently into different states. However, if one remains in a different state other than their set point for a considerable period (e.g., induced by a disease), this different state is considered a new set point. In pain research it is important to consider trait and state pains to gain an understanding of not only an individual's current pain state but also more broadly to their trait pain, which may be more reflective of their general condition (181). Resilience is a trait that is highly associated with chronic pain–related health outcomes. The neural correlates of both pain and trait resilience are critical to understand the brain–behavior relationship in chronic pain; yet, neural correlates of resilience in chronic pain states are unknown (182). Therefore, regional BOLD (Blood Oxygen Level Dependent) variability and circuit connectivity have potential to provide predictive power for pain resilience or vulnerability to chronic pain and treatment efficacy (182, 183).

Two reviews on mechanisms and imaging biomarkers for diabetic neuropathic pain review that diabetic peripheral neuropathy and associated pain have structural and functional central nervous system changes in the spinal cord, subcortex, and cortex (184, 185). Diabetic peripheral neuropathy has been associated with changes in the thalamus. A decreased thalamic NAA/creatinine ratio is suggestive of thalamic neuronal dysfunction (186, 187), and thalamic microvascular perfusion changes have also been observed (188). Smaller spinal cord cross-sectional area has been observed in those with diabetic peripheral neuropathy (189, 190). In fact, in diabetic peripheral neuropathy, diffusion tensor imaging techniques found posterior column damage in the cervical spinal cord (191). Diabetic neuropathic pain is related to decreased NAA in the thalamus (187), increased thalamic vascularity (188), and spinal cord posterior column damage (191).

Diabetic neuropathic pain is associated with increased regional brain gray matter volume loss localized to brain regions involved in somatosensory perception (192); furthermore, in diabetic neuropathic pain, increased total gray matter atrophy is associated with impaired ability to walk (193). Diabetic neuropathic pain has been shown to be related to aberrant default mode functional connectivity (171), decreased functional connectivity between the thalamus and cortex (194), and decreased functional connectivity in attention networks (195). Altered fMRI activation responses to experimental heat pain in limbic and striatal brain circuits are related to the duration of diabetic neuropathic pain (196). Diabetic neuropathic pain is related to a double dissociation such that neuropathic pain intensity is more associated with thalamus-insular cortex functional connectivity and nerve deficits are more related to thalamus-somatosensory cortex functional connectivity (197). Diabetic neuropathic pain is also associated with decreased functional connectivity between the thalamus and amygdala (198), decreased gray matter volumes and decreased white matter connectivity in pain processing and pain modulation brain regions (199), decreased somatosensory cortical thickness related to cortical function dysfunction (200), increased activity in the anterior cingulate cortex (201), as well as ventrolateral periaqueductal gray functional connectivity is altered and correlates with magnitude of spontaneous pain and allodynic pain (202).

Structural brain imaging has revealed changes to the brain associated with HIV peripheral neuropathy. Total cortical volume is smaller with HIV distal neuropathic pain (203). In fact, in HIV distal neuropathic pain the posterior cingulate cortex is the cortical region that was found to be smaller (204). In another sample of people living with HIV, subjective symptoms of HIV peripheral neuropathy were associated with smaller precuneus volumes which overlap with the posterior cingulate cortex (205). Smaller brain volumes for HIV distal neuropathic pain are consistent with a general pattern that brain volumes are reduced for a variety of chronic pain conditions (161, 162). Interestingly, the midbrain, thalamus and posterior cingulate cortex volumes are all reduced in HIV distal neuropathic paresthesia (206). It has been suggested that brain atrophy associated with HIV distal neuropathic paresthesia may precede brain atrophy associated with HIV distal neuropathic pain (206).

More recent multi-modal brain imaging research has revealed structural brain changes associated with HIV peripheral neuropathy (207). HIV peripheral neuropathy is correlated with decreased white matter integrity running from the midbrain to the somatosensory cortex. HIV peripheral neuropathy severity is also associated with decreased generalized fractional anisotropy along the tracts of the external capsule in both hemispheres, appearing to lead along the lateral thalamus to sensorimotor cortex. A similar correlation is found in the superior bilateral cingulum. These results indicate ascending deafferentation in HIV peripheral neuropathy extends further downstream from damaged peripheral nerves than seen previously, into the cortex.

HIV-associated distal neuropathic pain is associated with decreased fMRI resting state functional connectivity in the default mode network and increased functional connectivity in the salience network (208). Decreased connectivity between the medial prefrontal cortex and posterior cingulate cortex and stronger connectivity between the ACC and thalamus is associated with HIV distal neuropathic pain.

In the setting of experimental heat pain, significant interaction has been found within the right anterior insula during expectation of experimental pain offset in that a group with HIV distal neuropathic pain compared group without HIV distal neuropathic pain exhibited increased insula activation in the feet (with painful neuropathic pain) compared to the hand (without neuropathic pain) (209). These findings are consistent with abnormal processing of expectation of experimental pain offset or abnormal pain relief mechanisms potentially due to increased negative expectation regarding the experience of chronic endogenous neuropathic pain.

Anterior cingulate cerebral perfusion and gray matter density correlate with chemotherapy-induced peripheral neuropathy symptoms including pain (210). Patients with chemotherapy-induced peripheral neuropathy symptoms (including pain) demonstrated greater activation during painful stimulation in the precuneus compared to healthy controls and exhibited hypo-activation of the right superior frontal gyrus compared to healthy controls. Painful stimuli delivered chemotherapy-induced peripheral neuropathy symptoms patients evoke differential activation of distinct cortical regions, reflecting a unique pattern of central pain processing compared with healthy controls providing a tool for monitoring cerebral changes during anti-cancer and analgesic treatment (211).

A population of mixed small-fiber peripheral neuropathy (metabolic, inflammatory, chemo, idiopathic) was used to investigate how dysfunction of skin nerves led to abnormal recruitment of pain-related brain regions, suggesting that the brain may be affected in SFN. Greater volume reduction in pain-processing regions, particularly the bilateral anterior cingulate cortices was associated with greater depletion of intraepidermal nerve fibers. There was significant reduction in functional connectivity from the anterior cingulate cortex to the insula pain-processing cortex that is linearly correlated with the severity of intraepidermal nerve fiber depletion (212). Similarly, another population of mixed small-fiber peripheral neuropathy (metabolic, inflammatory, chemo, idiopathic) the degree of skin nerve degeneration was associated with the reduction of connectivity between the thalamus and pain-related areas. Despite altered white matter connectivity, there was no change in white matter integrity assessed with fractional anisotropy. These findings indicate that alterations in structural connectivity may serve as a biomarker of maladaptive brain plasticity that contributes to neuropathic pain after peripheral nerve degeneration (213).

A population of Charcot-Marie-Tooth patients had abnormal diffusion tensor imaging findings indicative of significant cerebral white matter abnormalities. Diffusion tensor imaging abnormalities were correlated with clinical disability, suggesting that there is comorbidity of central nervous system damage with peripheral neuropathy in Charcot-Marie-Tooth patients (214). A population of patients with hereditary neuropathy with liability to pressure palsies were compared to a population of normal controls and the fractional anisotropy values of the patients were significantly lower in bilateral frontal, orbitofrontal, and temporal areas of white matter (215). Patient populations of paresthesia-dominant and pain-dominant patient groups were compared and contralesional cortical thickness were correlated with pain severity (216). Acquired and hereditary peripheral neuropathies are associated with increased functional connectivity of the left precuneus/posterior cingulate cortex in the default mode network. This increased connectivity in the default mode network is correlated with duration of peripheral neuropathy and severity of clinical total neuropathy score (217).

As detailed above, chronic pain and neuropathic pain impact multiple organ systems. Advancing the value of pain biomarkers depends on (1) selection of measurements and metrics that are the most mechanistically valid and informative, and (2) combining the selected measurements such that they mechanistically and statistically maximize accurate classification. Advancement of measurement accuracy is vital and the subsequent steps of the approach are entirely contingent upon the success of this step. This literature for the domains discussed in this manuscript is too voluminous for a single review. In the above reviewed literature, we attempted principally to focus on which biological systems and which biomarkers should be the focus of measurement. For effective application of measurements of these domains it is important to discuss approaches for measurement selection.

In Figure 2, we provide a significantly abbreviated schematic of key available statistical approaches to handling multimodal datasets in building composite biomarkers. We have highlighted four general areas of statistics/machine learning: (1) feature reduction (225), (2) classification (226, 227), (3) regression (228), and (4) clustering (229). Feature reduction can occur during or prior to classification, regression, or clustering. Feature reduction primarily focuses on two primary approaches: (1) integration of measurements toward creation of a composite variable to simplify and enhance model performance, and (2) effective feature reduction through variable selection to use optimal variables. Thus, feature reduction can represent the effective combining of strong measurements to a meaningful and robust latent variable or elimination of unnecessary, or statistically weak, measurements. Some methods, such as random forest, has built in feature reduction (230). Classification methods are often utilized to build toward categorical variables, however methods like neural networks are also designed for predicting continuous variables (231). Regression models are often used for the prediction of continuous measures or in the case of canonical approaches this can be with multiple dependent variables predicted simultaneously (232). Finally, in the case where there is no existent or optimal category or variable that the biomarkers seek to predict unsupervised approaches can be useful. With all these approaches variables can either be approached as linear or non-linear, although transformations and feature reduction approaches can mitigate these differences. It is important, regardless of approach, to understand the biological mechanisms being modeled by defining a model that best reflects the underlying systems to optimize prediction.

Two key methods for statistical reduction of variables are (1) selecting top ranking variables and (2) creation of composite variables by factor or component-based analysis. Random Forest, as depicted in Figure 2, can be utilized to determine importance scores by evaluating the hierarchical functionality of a given variable as a bifurcator for optimizing classification (226). Random Forest is not alone in its utility to provide variable importance ranking but provides a nice mechanism for this analysis. The statistical creation of composite variables can be done through principal component analysis (or independent component analysis among other methods) such that novel values are calculated for a set of variables that account for large swaths of variance with a single value vector (233, 234). This can substantially increase the efficiency of a model and serve to highlight a robust latent feature.