Neuroimaging of the human visceral pain system–A methodological review

1 Introduction

During the last decades there has been a tremendous development of non-invasive methods for assessment of brain activity. The use of these techniques has brought new insight into the complex brain processes underlying pain perception and paved new ways in our understanding of acute and chronic pain. The techniques most extensively used are functional magnetic resonance imaging (fMRI), positron emission tomography (PET), electroencephalography (EEG), evoked brain potentials (EPs), magnetoencephalography (MEG), single photon emission computed tomography (SPECT), and the multimodal combinations of these techniques.

Painful sensations from the human gastrointestinal (GI) tract are common symptoms in the clinic but the underlying aetiology is often difficult to identify [1]. In clinical practice, it is most likely that this is due to heterogeneity of the patient groups and the many confounding factors such as sedation, nausea and general malaise that all can affect the pain experience [2]. Also the relationship between disease severity and symptoms are ambiguous [3]. Hence, objective methods for pain assessment are needed in order to improve our understanding of GI pain and to facilitate the development of new therapeutic targets. The information on visceral sensory processing has traditionally been based on animal studies, but during the last decades our understanding has been facilitated by the advances in non-invasive technologies. These techniques have vastly increased our understanding of central processing of GI sensation and pain in both healthy volunteers and patients. The methods have demonstrated the complexity involved in pain processing and revealed a number of subcortical and cortical regions involved in this process. As pain is a conscious feeling, the ultimate goal in pain-imaging is to follow the pain stimulus throughout the neuraxis. The pathways are dynamic with modulation of the response to internal or external stressors with a high degree of plasticity. The ultimate outcome of pain perception is brought about by a delicate balance between facilitatory and inhibitory pain mechanisms, which is also influenced by sensory-discriminative, affective-motivational and cognitive-evaluative components. For amore detailed summary of the normal GI sensation see Section 2 below and recent reviews [4,5,6,7,8].

The purpose of this review is to give a state-of-the-art overview of the methods used for assessment of visceral pain in the neuraxis with focus on EEG and especially the advantages and limitations in GI pain research (see Table 1), including examples from relevant studies.

2 The visceral pain system

The GI tract has a complex dual innervation with both intrinsic and extrinsic sensory neurones (the latter is referred to as visceral afferents). Fig. 1 illustrates the principal visceral projections including the subcortical and cortical structures involved in the response to visceral pain. Intrinsic afferents mainly project locally in the wall of the gut and are involved in regulation of the visceral functions, such as secretion, motility, mucosal transport and bloodflow[9,10]. Extrinsic afferents project to the central nervous system (CNS) via the vagal nerve to the brainstem or through splanchnic nerves to the spinal cord [11].

Fig. 1

Illustration of the principal visceral projections including the subcortical and cortical structures involved in the response to visceral pain. The thalamus (Th) functions as a relay station, which conveys the information to various parts of the brain. Third order neurons ascend to the secondary somatosensory cortex (SII), the limbic system including anterior cingulate cortex (AC), insula, and the prefrontal cortex (PFC). The dotted lines illustrate connections to other cortical centres simultaneously activated, see text for more information. Amygdala (A), hypothalamus (H), periaquaductal gray (PAG), reticular formation (R).

Most afferent vagal fibres are unmyelinated C-fibres that project viscerotopically to the medial division of the nucleus of the solitary tract. Second-order neurones project to sites in the brainstem, hypothalamus and amygdala including the vagal motor nuclei, the rostral areas of the ventrolateral medulla and the parabrachial nuclei [12]. Cortical projections from these brainstem areas include the orbitofrontal, infralimbic anterior cingulate and insula cortex, the latter having reciprocal connections with the secondary somatosensory cortex (SII). The vagal afferents are believed to mediate non-noxious physiological sensations such as satiety and nausea (low response thresholds) [13,14,15]. Also, vagal afferents may be involved in the central inhibitory modulation of pain [16]. On the contrary, visceral pain is mainly mediated through the spinal afferents in the splanchnic nerves which pass the dorsal root ganglion and project to distinct laminae of the spinal cord dorsal horn (mainly laminae I and V, and occasionally to contralateral laminae V and X). These afferent projections are organized in a segmental manner, but distributed over several spinal segments in both rostral and caudal directions [17]. This diffuse spinal termination pattern may explain the poor localization of visceral sensations often seen in clinical practice, and convergence of visceral and spinal afferents in the spinal dorsal horn may explain the phenomenon of viscerosomatic convergence, whereby visceral pain is often referred to nearby somatic structures (referred pain) [11,18]. From the spinal cord, pain transmits to the brain through diverse pathways. Most afferents travel in the spinothalamic tract to the thalamus. From the thalamus projections to the insula, hypothalamus, amygdala as well as to higher cortical levels such as cingulate and prefrontal cortices have been described. Insula has an important function for integrating the visceral sensory and motor activity together with limbic integration and is particularly important in pain perception from the gut [19]. The anterior cingulate cortices (ACC) and prefrontal cortices are a part of the medial pain system, which mediates the affective, emotional and cognitive components of pain experience [20,21]. In addition to the spinothalamic tract, some afferents ascend in the spinoreticular tract mediating arousal and autonomic responses through interaction with the reticular formation [22]. Finally, a population of afferents ascends in the spinomesencephalic tract, which relates to a complex neuronal network including the periaquaductal gray, rostroventral medulla (RVM) and dorsolateral pontine tegmentum. This network comprises the structural basis of descending pain control and possess a modulatory effect on the spinal pain processing through so called on- and off-cells in the rostroventral medulla, which are pro-nociceptive or anti-nociceptive, respectively [21,23].

3 Magnetic resonance imaging

MRI allows imaging of both brain structure and activity. Brain activity measured by fMRI has most commonly been acquired by the blood oxygenation level dependent (BOLD) technique, which is based on different paramagnetic properties of oxy- and deoxyhemoglobin in the blood. The BOLD signal reflects simultaneously changes in local blood flow, volume and deoxyhemoglobin content, which derive from changes in neuronal activity [26]. Regions of activation are identified by subtracting regional BOLD signal during a control/resting condition from the signal during a stimulus condition. There is a rapid development in both hardware and software. Recently, other techniques such as arterial spin labeling (ASL) and signal enhancement by extravascular water protons (SEEP) have been applied. ASL allows the measurement of whole brain cerebral blood flow (CBF) in absolute units through the use of magnetically labeled endogenous water in blood acting as a diffusible tracer [27,28] allowing assessment of the temporal dynamics of the neural activation induced by pain. This has been used to detect changes in regional CBF associated with a standard cutaneous heat pain model [29] and infusion of hypertonic saline [30]. ASL is particularly suited to studies of prolonged pain since it becomes increasingly more sensitive than BOLD to changes in neural activation as the stimulus duration exceeds 1min [31]. The SEEP technique has been used in fMRI of the spinal cord, which is essential in the complete mapping of the pain system, and spinal cord and brain stem sensory-related neural activity has been consistently observed in a number of studies [32,33,34,35,36,37,38,39,40,41,42,43,44]. Recently, structural information obtain by other MRI techniques, such as diffusion tensor imaging (DTI) including tractography, volumetry of brain structures and measurement of gray matter density, has been superimposed to gain more structural information on the neural structures and connections between the centres involved in pain processing [45,46,47]. Most neurophysiological research is performed at 3T field strength, but 7T scanners are now emerging with new possibilities for enhanced resolution.

Results obtained with fMRI have been found to be strongly correlated with those from PET-CBF in identical paradigms [48,49,50]. So far, the ‘pain responses’ obtained using PET and fMRI methods have been very similar, but more comparisons of results using both techniques in the same population of subjects are needed.

Advantages: fMRI has an excellent spatial resolution (2-5 mm), especially in the more superficial layers, but limitations are seen in the deeper structures, such as the brainstem and thalamus, due to pulsation artefacts. fMRI has the possibility to take into account anatomy and other individual characteristics, which is a clear advantage over PET. The temporal resolution (ranges from a 300ms theoretical value to a more realistic 1-3 s) in event-related fMRI studies with echo-planar systems is another advantage compared to PET. fMRI appears as an intermediate solution between the temporal resolution of PET (tens of seconds) and electrophysiology (tens of milliseconds). A clear advantage of fMRI is that it operates in a non-invasive and non-radioactive environment, allowing subjects to be studied repetitively. Singlesubject analysis of fMRI will also be an important gain in pain studies, since pain is notoriously dependent upon individual factors.

Limitations: The temporal resolution of fMRI is clearly inferior to EEG/MEG meaning that fMRI is not a specific tool for investigating the primary neuronal activity directly related to the painful stimuli (first few hundred ms post-stimulus). Furthermore, the fMRI activity to gastrointestinal stimulation is not stimulus-specific, anticipation of stimuli can trigger similar activity and repeated activation can result in habituation [51,52]. fMRI remains currently mainly limited to “activation” studies, but recently also resting state fMRI has been applied in pain research including connectivity analysis between multiple brain network [53]. In contrast to PET studies a limitation in fMRI studies is the lack of information regarding neurotransmitters or involved receptors. Furthermore, MR scanners are expensive, and the time allowed for research studies in clinical departments is often limited.

4 Single photon emission computed tomography and positron emission tomography

SPECT and PET are nuclear imaging techniques that can trace radiolabeled molecules injected into the blood stream, whereby the distribution, density and activity of receptors in the brain can be visualized. This provides an insight into the organization of functional networks in the brain, which cannot be achieved by morphologic investigations or imaging of blood flow and metabolism [67]. Preclinical and clinical trials may therefore benefit greatly from such molecular imaging techniques [68]. In fact pharmaceutical compounds used as radiolabeled tracers can combined with kinetic models allow quantification of receptor sites and enzyme function in human subjects using SPECT or PET. The majority of studies have investigated opioids and their receptors, but recently other neurotransmitter systems, e.g. dopamine, have also been examined [69].

Advantages: An advantage with SPECT is that isotopes with a long half life are used, making it possible to widen the observational time window. This allows biomedical scientists to observe biological processes in vivo several hours or days after administration of the labeled compound [70]. PET has excellent spatial resolution (2-5 mm), which allows tagging important biological molecules that bind to targeted receptor groups or glucose metabolism in active neuronal tissue. PET is superior in imaging radiopharmaceuticals and/or other ligands as it offers the ability to study receptor distribution and explore the site of action. This tends to be the future main application of PET.

Limitations: SPECT exhibit lower sensitivity, i.e. ability to detect and record a higher percentage of the emitted events, compared to PET. However, use of specialized collimators is viewed as a technique to improve the sensitivity of SPECT without degrading image resolution [71]. The temporal resolution of SPECT and PET is poor (minutes) compared to especially EEG/MEG, and group analyses are needed for meaningful results. However, PET has some intrinsic advantage over SPECT for dynamic studies [71]. Another major disadvantage is that the subject receives a considerable dose of radiation and the expense of a PET scanner.

5 Electroencephalography

EEG is a method to assess electrical activity in the brain generated by firing between neurons. The activity of the neural networks is usually randomly active in time, but can be synchronized and enhanced in response to an external event. Thus, the activity can be recorded in either the resting state (resting EEG) or as evoked potentials (EPs) following an external stimulus. Both types of recordings may be used to study normal visceral pain processing and to identify alterations of pain processing in different patient groups or due to pharmacological intervention. The resting EEG is typically used to study the pathophysiology of pain in chronic pain patients, while the EPs are used to study the nociceptive pain response including the sequential brain activation following GI stimulations [79].

The resting EEG is often assessed by time-frequency analysis (see section below), while several analysis methods such as sophisticated time-frequency analysis and visual inspection exist for evaluation of EPs. The EPs can be analyzed either as single-sweeps or after an average process. The single-sweep analysis enables an evaluation of both synchronized and unsynchronized brain activity relative to the stimulus, which may provide additional information on abnormal pain processing in some patients [80,81]. The average process is performed over multiple stimuli which improve the signal-to-noise ratio and hereby reveals less prominent peaks. Averaging a sufficient number of sweeps enables visual inspection of the EPs to analyze the peaks in terms of amplitude and latency as illustrated in Fig. 2.

Fig. 2

A typical evoked potential (vertex-electrode) recorded after stimulation of the rectosigmoid junction in a patient with chronic pancreatitis. This is an average of the 50 electrical stimuli. Note the different peaks denoted N1, P1, N2 and P2, each defined by latency (ms) and amplitude (μV).

Advantages: EEG is generally highly available, relatively easy to use and low-cost. The main advantage is the high temporal resolution which allows assessment of the primary pain processing, including sequential activation and analysis of coherence and cross talk between brain centres.

Limitations: The main limitation of EEG is the relative poor spatial resolution. However, various inverse modelling algorithms and signal decomposition procedures (see below) have overcome this limitation to some extent and ongoing research in this field holds promise for further improvements of the methods. While multichannel EEG, cerebral EPs and inverse modelling offers a noninvasive approach to study brain activity with time resolution on millisecond scale, it must not be overlooked that the position of the calculated dipolar sources does not represent the accurate position but rather the “centre of gravity” of brain activity.

5.2 Time-frequency analysis in visceral pain studies

Continuously recorded signals for both resting EEG and EPs have a complex nature, which prevents immediate manual assessment of traces. In spite of the chaotic nature of the signals, it has been shown that the EEG oscillates in certain frequency bands associated with specific brain functions such as pain perception and attention to pain. The EEG frequency bands are typically defined as: delta (up to 4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta (12-30 Hz), and gamma (above 30 Hz) [85]. A number of algorithms to calculate frequency distributions exist, most common ones being the Fourier transformation, wavelets, and matching pursuit as illustrated in Fig. 3 [86]. Frequency analysis can be applied in order to study how EEG characteristics are altered in different patient groups suffering from acute and chronic visceral pain. Rickenbacher et al. found that bladder dysfunction was associated with long-term changes in brain activation characterized by decreased delta oscillations and prominent theta oscillations [87]. Tayama et al. compared 10 healthy controls and 10 IBS patients, and found a significant lower alpha power percentage and a higher percentage of beta power in patients compared to controls [88]. Subsequently, Drewes et al. used topography matching pursuit on EPs in a study of painful chronic pancreatitis in order to assess whether the pain could be of neuropathic origin [89]. They found persistent increased theta activity in the patients indicating pain in chronic pancreatitis is likely of neuropathic origin.

Fig. 3

Time-frequency analysis in visceral pain studies. Illustration of the different algorithms to calculate frequency distributions, including the Fourier transformation, wavelets, and matching pursuit.

5.4 Inverse modelling

The location of brain centres underlying EPs can be determined by multi-channel EEG recordings combined with inverse modelling. This method possesses the opportunity to study pain specific cortical activation dynamically, as it reflects the sequential activation of neuronal pain networks underlying the EPs. The signal recorded on the scalp is distorted by conduction through the scalp and is furthermore a mixture of multiple electrical sources inside the brain. This poor spatial resolution is, in contrast to the high temporal resolution, the main disadvantage of the EEG. There are a number of commercial software and freeware available for inverse modelling, most commonly used are EEGLAB (Matlab, The MathWorks Inc., Natick, MA, USA), BESA (MEGIS Software GmbH, Gräfelfing, Germany), and CURRY (Compumedics, El Paso, TX, USA).

In a study of healthy volunteers based on the BESA algorithm, EPs to electrical stimulation of the upper gut and sigmoid colon were explained by bilateral brain sources in the SII, insular cortices and a single dipole in the ACC [92]. Interestingly, a viscerotopic organization of the different gut segments was seen, thus revealing a “visceral homunculus” mimicking that seen for the somatic sensory system. In IBS patients the ACC dipole (BESA algorithm) following painful electrical stimulation of the sigmoid colon showed a more posterior position in patients than in control subjects [93]. This finding suggests that the cortical representations of painful stimuli are altered in IBS patients possible due to reorganization of the cortical areas involved in visceral pain processing. After experimental acid sensitization of the oesophagus in healthy volunteers, a posterior shift and latency reduction of the ACC dipole was seen [94]. This finding may translate to the clinic where patients with non-erosive gastro-oesophageal reflux disease (NERD) often have unpleasant chest sensations and pain despite the fact that a normal oesophagus is evident at endoscopy [94]. Hence, the symptoms in these patients may be partly explained by alterations in the central processing of visceral pain. Along this line, a modification of bilateral insular dipoles localizations was seen in chronic pancreatitis patients following upper gut stimulations [83]. As the upper gut and pancreas share afferent neuronal pathways these findings likely represent a reorganization of the visceral cortical projections induced by recurrent pain attacks due to pancreatitis. This parallels findings from somatic pain studies where reorganization of the cortical pain matrix has been commonly reported [95].

In a study of diabetes patients with autonomic neuropathy and GI symptoms, the evoked brain potentials to painful oesophageal stimulation were analyzed using the fixed Multiple Signal Classification (MUSIC) algorithm in CURRY [96]. The patients had a posterior shift of the ACC dipole and additional sources close to the posterior insula and medial frontal gyrus, indicating that central neuronal reorganization may contribute to our understanding of the GI symptoms in these patients.

One of the major limitations with inverse modelling has been the instability of algorithms tomodel several simultaneously active sources and sources in deep brain structures such as thalamus or brainstem. To bypass these problems, signal decomposition methods have been developed in order to separate the signal into a sum of waveforms, whereby signals corresponding to specific evoked brain activity can be separated from artefacts and noise. Once the signals are decomposed, inverse modelling can be applied on each of the retrieved waveforms. Some of the most common approaches for signal decomposition are blind source separation algorithms such as independent component analysis (ICA) and second order blind identification (SOBI) [97]. Drewes et al. have successfully used ICA to study sequential brain activation and cross talk between brain centres following electrical stimulation of the oesophagus in healthy controls [97]. Recently, multichannel matching pursuit (MMP) was introduced, which decomposes the EP data into a sum of waveforms (usually termed atoms), each of them being defined in time, frequency and space, see Fig. 4. We showed that inverse modelling on MMP atoms is superior to inverse modelling on instantaneous EP data, ICA and SOBI components with high accuracy to localize superficial, deep, and simultaneously active sources [97,98]. Recently, weused inverse modelling withMMPto study the effects of morphine on oesophageal EPs in healthy subjects, where the brain source shifted frontally in atoms in delta band due to morphine, whereas the source was stable in the placebo condition (unpublished data).

Fig. 4

Source localization of evoked potentials in combination with multichannel matching pursuit to estimate brain activation. Recording from multiple electrodes on the scalp gives continuous EEG signals as shown to the left of the figure on top. When the subject is stimulated a number of times, these signals can be averaged and evoked potentials can be identified as peaks, as seen on the bottom of the figure to the left. The averaged signal is then decomposed into a sum of atoms using multichannel matching pursuit as seen in the centre of the figure. Each of these atoms is well defined in time and frequency and when they are summed up, they represent the signal shown in the bottom left of the figure. Lastly, inverse modelling is done on each of the atoms in order to give a result similar to what is shown to the right of the figure. Now, not only are the active brain source locations known, but also their time and frequency distributions.

6 Magnetoencephalography

MEG is a non-invasive technique for mapping brain activity by recording magnetic fields produced by electrical currents in the brain. The electrical currents (as measured by the EEG) accompanying brain activity are extremely weak, and resulting magnetic fields are 1/8 of the magnetic field of the earth [99]. Nevertheless, these tiny fields can be measured in magnetic shielded rooms by superconducting detectors (supraconducting quantum interference devices), for technical details see [100]. Modern MEG uses a helmet-shaped dewar that typically contain arrays of detectors covering most of the head.

MEG has been extensively used for investigation of the SI and SII areas in health and disease [99]. These areas are rather superficial and their neuronal columns result in equivalent current dipoles with a marked tangential direction, the magnetic fields of which are optimal for MEGmeasurement. In particular, MEG has added to characterization of the somatotopic organization of SI and SII areas as well as the association between pain and cortical reorganization in somatic and neuropathic pain studies [95,99,101]. However, the use of MEG in the study of visceral pain has until today been very limited. For example in a study by Hobson et al. MEG was used to determine the sequential brain activation following oesophageal electrical stimulation in healthy volunteers [102]. They found the earliest cortical activities in the SI and SII cortices and posterior insula, followed by later activities in the anterior part of insula and cingulate cortex.

Advantages: The main advantages follow from the fact that MEG signals are not highly distorted by conduction through the skull between sensors and scalp. Consequently, in contrast to EEG, MEG posses a high spatial resolution comparable to fMRI and PET. In addition, it has unique capabilities for real-time imaging of neuronal activity with millisecond response (comparable to EEG) [103].

Limitations: MEG is a very technically demanding technique and is only available in few specialist centres. Furthermore, it is limited by its incapability to resolve radial currents generated by deep brain sources, e.g. in the cingulate cortex. Radial dipoles generate magnetic fields which do not penetrate the head, since volume currents shield the magnetic fields. In order to investigate pain generators in the cingulate cortex and other deep brain structures, alternative methods such as multichannel EEG therefore have to be used [99].

7 Summary

This review gives an overview of the methods used in the assessment of neurophysiologic mechanisms behind visceral pain with focus on the advantages and limitations of the single techniques. Even though there has been a tremendous development in the methods for imaging the processing of pain, it should be kept in mind that all of these methods are not direct measures of brain activity and will never be able to assess the true neuronal activity and pathophysiological mechanisms. It should be realised that no single method has the potential to give the ultimate answer. However, as outlined in the review there seems to be some hope ahead since the development in hardware, computer power and post-processing is still ongoing allowing high quality data and new ways to interpret the data. Furthermore, the multimodal combinations of methods utilising the advantages of each single technique (see Table 1) will likely in the future bring us closer to true events in the brain, e.g. the fusion of hemodynamic and electrophysiological methods using for instance classification methods. The future aims are to identify still better and more specific biomarkers for abnormal processing of sensory input underlying the pain and suffering in patients groups, eventually to identify different sub-groups with different aetiology, and give directions for the development and testing of new treatment options with benefit for individuals and society.