Mapping pain activation and connectivity of the human habenula

Abstract

The habenula, located in the posterior thalamus, is implicated in a wide array of functions. Animal anatomical studies have indicated that the structure receives inputs from a number of brain regions (e.g., frontal areas, hypothalamic, basal ganglia) and sends efferent connections predominantly to the brain stem (e.g., periaqueductal gray, raphe, interpeduncular nucleus). The role of the habenula in pain and its anatomical connectivity are well-documented in animals but not in humans. In this study, for the first time, we show how high-field magnetic resonance imaging can be used to detect habenula activation to noxious heat. Functional maps revealed significant, localized, and bilateral habenula responses. During pain processing, functional connectivity analysis demonstrated significant functional correlations between the habenula and the periaqueductal gray and putamen. Probabilistic tractography was used to assess connectivity of afferent (e.g., putamen) and efferent (e.g., periaqueductal gray) pathways previously reported in animals. We believe that this study is the first report of habenula activation by experimental pain in humans. Since the habenula connects forebrain structures with brain stem structures, we suggest that the findings have important implications for understanding sensory and emotional processing in the brain during both acute and chronic pain.

the habenula is located in the dorsal posterior thalamus (epithalamus) adjacent to the third ventricle. It is involved in a wide variety of behaviors including olfaction, ingestion, mating, endocrine function, aversive motivation, and brain stimulation, as well as pain and analgesia (Sugama et al. 2002; Thornton et al. 1985). Its role in pain and analgesia is supported by several animal studies, including mapping of nociceptive afferents innervating the habenula via the lateral hypothalamus (Craig 2003), stimulation-induced analgesia (Cohen and Melzack 1986; Mahieux and Benabid 1987; Neal et al. 1999), opioid-mediated analgesia (Cohen and Melzack 1985; Martin et al. 1997; Neal et al. 1999), electrophysiological recordings (Gao et al. 1996), anatomical connectivity with modulatory systems (Araki et al. 1988; Ferraro et al. 1996; Herkenham 1979; Kim 2009), markers of activation (cfos gene) with noxious stimuli (Smith et al. 1997; Lehner et al. 2004), and functional imaging in rats (Paulson et al. 2007).

Furthermore, the habenula's relationship to pain processing can be inferred from its relationship to emotion mechanisms. The habenula receives information from forebrain structures and the dorsal diencephalic conduction system, one of two major pathways that interconnect the limbic forebrain and sites in the mid- and hindbrain (Bianco and Wilson 2009) (see Fig. 1). As such, the habenula is ideally situated to relay information related to processing from brain regions involved in emotion to brain stem systems. Examples include habenular alteration of serotonergic and dopaminergic systems, which are responsible for pain modulation [this interaction manifests via loops that include the nucleus accumbens (NAc) and periaqueductal gray (PAG) (Ma et al. 1992; Yu and Han 1990)]; the medial habenula sending projections to the interpeduncular nucleus (IPn), which modulates cells in the raphe magnus that influence ascending nociceptive processing (Hentall and Budhrani 1990); connections to the ventral tegmental area (VTA) and substantia nigra compacta, through which the habenula may have a role in signaling errors of prediction of reward-related signals (Bromberg-Martin and Hikosaka 2011); stress evasion (Hikosaka 2010); and stimulation of the lateral habenula, which decreases reward (Friedman et al. 2011). Taken together, the region is a major point of convergence where external stimuli are received, evaluated, and redirected for motivation of appropriate behavioral response.

Fig. 1.

Diagrammatic connectivity map of the habenula. This figure represents structural connections found in animal studies. Afferent inputs are labeled on left and efferent connections on right. Black dashed rectangle outlines the epithalamus, which contains the lateral and medial habenula as well as the pineal gland. Red lines designate peripheral afferent connections, green lines represent descending and modulatory nociceptive pathways, and blue lines are central inputs to the habenula that may modify pain processing. FrCtx, frontal cortex; NAc, nucleus accumbens; Lateral Hypoth, lateral hypothalamus; EP, entopenduncular nucleus; CPu, caudate/putamen; Hippo, hippocampus; M, medial habenula; L, lateral habenula; P, pineal; IPN, interpeduncular nucleus; PAG, periaqueductal gray; Raphe, raphe nuclei; VTA, ventral tegmental area; SNc, substantia nigra pars compacta.

To the best of our knowledge, no functional magnetic resonance imaging (fMRI) measures of habenula activation with pain or analgesia have been specifically reported in humans to date. However, neuroimaging studies have reported on habenula activation and connectivity as it relates to error and reward prediction (Ide and Li 2011; Li et al. 2008), negative feedback measures (Ullsperger and von Cramon 2003), mood as related to its influence on serotonergic activity (Morris et al. 1999), and deep brain stimulation for depression (Sartorius et al. 2010).

For noninvasive mapping of habenula activation during pain processing in healthy human volunteers, we used 1) blood oxygen level-dependent (BOLD) fMRI mapping to identify activation in response to noxious heat and 2) diffusion tensor imaging (DTI) for anatomical connectivity analysis to verify that the fMRI activation was localized to the habenula. We tested the following hypotheses: 1) distinct activation to pain would be observed in the human habenula; 2) afferent and efferent connections, based on our understanding from the animal literature, would be similar in the human utilizing DTI; and 3) for some structures, functional connectivity would correlate with anatomical connectivity.

METHODS

Subjects

Eleven male subjects (27 ± 7.6 yr) were recruited by advertisement. Experiments were approved by the Institutional Review Board (IRB) and were conducted in accordance with International Association for the Study of Pain (IASP) recommendations for pain studies involving human subjects (Charlton 1995). Prescreening included patient consent forms, EKG, and completion of the Beck Depression Inventory (BDI). Subjects accepted into the study had no history of psychiatric, neurological, or other medical conditions. Exclusion criteria included a BDI score >21, testing positive for alcohol or drug abuse, alcohol screening, and urine screening for barbiturates, benzodiazepines, amphetamines, cocaine, tetrahydrocannabinol, phencyclidine, or opioids. Subjects were told to fast and abstain from drinking 6 h prior to their fMRI session.

Psychophysical Measures and Controls

Quantitative sensory testing (QST) was performed with a Thermal Sensory Analyzer (TSA, Medoc, Haifa, Israel). A thermode located on the left foot started at a baseline temperature of 32°C and underwent a 5-s ramp for a 25-s noxious stimulation at 46°C. Subjects rated pain intensity with an MRI-compatible visual analog scale (VAS), with anchors labeled “no pain” and “maximum pain.” Subject scores were stored numerically from 0 to 10. This stimulus and rating pattern was repeated four times.

Scanning and Stimulus Procedures

Data acquisition.

Subjects were scanned with a 3-T Siemens Trio System using an eight-channel phased-array head coil (Erlangen, Germany). Localization, anatomical scans, and diffusion-weighted imaging (DWI) were acquired during the first 30 min. An initial three-plane localizer was used for placement of experimental slices. A sagittal three-dimensional T1-weighted scan was performed (TR/TE/flip = 2500 ms/30 ms/90°; field of view = 20 cm; slice thickness = contiguous 2.8 mm; in-plane resolution = 1.2 mm; matrix = 256 × 192; and averages = 1).

FUNCTIONAL SCANS (FMRI).

Functional scans were performed with a gradient echo, T2*-weighted sequence (TR/TE/flip = 2.5 s/30 ms/90°; slice thickness = 3 mm; in-plane resolution 3.125 mm × 3.125 mm), performed on a 41-slice prescription. With a block design, subjects were stimulated with 46°C on their left foot. The thermal block design consisted of 25 s of stimulation separated by 35 s of a 35°C baseline temperature with a 5-s ramp between (as shown in Fig. 2). Heat was delivered with a 3 × 3-cm Peltier thermode (TSA, Medoc, Haifa, Israel). Subjects rated their pain during the entire paradigm with the VAS scale used for QST. They could terminate the trial at any time. The rating was not used in analysis but was measured to ensure that stimulation exceeded pain thresholds.

Fig. 2.

Stimulation paradigm. Eleven subjects were thermally stimulated with noxious 46°C heat on their left foot. The block design paradigm consisted of 25 s of stimulation separated by 35 s of a 35°C baseline temperature with a 5-s ramp between delivered with a 3 × 3 cm Peltier thermode [Thermal Sensory Analyzer (TSA); Medoc, Haifa, Israel]. The stimulation block was repeated 4 times. Participants rated their pain on a visual analog scale (VAS; 0–10) during acquisition. The average rating was 7.6 ± 2.0.

DIFFUSION-WEIGHTED IMAGING.

DWI was collected with a single shot-twice reinforced echo planar imaging (EPI) pulse sequence with imaging parameters TR = 7,900 ms, TE = 92 ms, 5/8 partial Fourier, threefold SENSE acceleration, resolution = 1.75 × 1.75 × 2.5 mm³, and 50 axial slices to cover the entire cortex and cerebellum. Eight nondiffusion volume images were collected before 72 distinct diffusion-weighted volumes were collected at b = 1,000 s/mm² with acquisition time ∼10 min.

Data analysis.

FMRI DATA ANALYSIS.

FSL software (FMRIB's Software Library, www.fmrib.ox.ac.uk/fsl) version 4.1.6 was used for analysis of all imaging data. For each subject's functional time series, the first four volumes were discarded for signal calibration. Preprocessing included 1) brain extraction (using BET tool), 2) slice timing correction and motion correction using FMRIB's Linear Image Registration Tool (MCFLIRT), 3) spatial smoothing with a 5-mm full-width half-maximum Gaussian kernel, 4) high-pass filter temporal smoothing (cutoff 75.0 s), and 5) anatomical images used to linearly register motion-corrected EPI activation maps. Baseline drift was accounted for with Gaussian mixture modeling as described below.

Single-subject General Linear Model (GLM) analysis was carried out with FMRI Expert Analysis Tool (FEAT) version 4.1.6 (Woolrich et al. 2001). Temperature profiles that were recorded during the scan were resampled and then scaled from 0 to 1 for use as explanatory variables (EVs) in the GLM analysis.

For noxious heat, an early and late (biphasic) pain model was devised. The BOLD time course response to noxious contact heat stimulation has been shown to contain a delayed response shifted by several time points where the late and early phases present the first and second halves of the stimulus event. This feature has been found throughout subject populations and in multiple brain structures (Becerra et al. 2001; Moulton et al. 2005; Upadhyay et al. 2010). The phases might correspond to different networks of pain processing, such as emotional/salience and sensory circuits.

The early- and late-phase pain model was devised by segmenting the thermal paradigm. Thus this second GLM analysis consisted of two EVs corresponding to early-phase response (EV1) and late-phase response (EV2). Based on previous models (Becerra et al. 2001), they were devised to have equal length response, i.e., response to EV1 = response to EV2.

HABENULA ANATOMICAL LOCALIZATION.

The habenula location was defined on each subject's high-resolution T1-weighted structural image. This process was aided by a human brain atlas (Mai et al. 2008)(see Fig. 3 and Fig. 4A). Because there are no cellular borders around the habenula that are distinguishable with MRI, we identified the habenula on the basis of histological studies that indicate that its 0.2-ml volume is bordered by the third ventricle walls (Shelton et al. 2012). This criterion allows the habenula to be defined anatomically (see Fig. 3).

Fig. 3.

Thalamic activation. Segregated activation of the habenula and neighboring thalamic nuclei. Top: coronal slices include y = −14, −20, and −26 in MNI space showing the various thalamic activation clusters. The habenula appears in top right slice as highlighted with a red circle on atlas images. Noxious thermal stimulation activated the habenula as well as other thalamic structures typically identified in noxious thermal paradigms. Atlas images on bottom right show coronal (y = −28) and axial (z = 2) slices with the habenula marked with a red circle, and the other thalamic structures are marked with blue circles. The corresponding functional image for coronal slice y = −28 has third ventricle cerebrospinal fluid (CSF) voxels removed for the purpose of visualization. The small box on bottom right of this functional image shows the activation map before removal. Note that the activation is located on the opposite side as displayed in this atlas image. Atlas images adapted with permission from Mai et al., Atlas of the Human Brain (3rd ed.). Copyright Elsevier (2008). Pul, pulvinar; APul, anterior pulvinar; Hb, habenula; Th, thalamus; MD, medial dorsal.

Fig. 4.

Functional activation to pain. A: coronal slice of the human brain diagramming the location of the habenula, a Nissl stain slice, and a coronal slice graphic adapted with permission from Mai et al., Atlas of the Human Brain (3rd ed.). Copyright Elsevier (2008). hbc, Habenula commissure; MHb, medial habenular nucleus; LHb, lateral habenula nucleus; MD, medial dorsal thalamic nucleus; Lim, limitans nucleus; PTc, pretectal area; bfx, body of formix; SFPul, superficial pulvinar nucleus; iml, internal medullary lamina of thalamus; Pul, pulvinar; APul, anterior pulvinar nucleus; st, stria terminalis; eml, external medullary lamina of thalamus; VPL, ventral posterior lateral thalamic nucleus. B: group activation maps from noxious thermal stimulation administered in a 4-block design paradigm. Data were registered to the individual subject high-resolution anatomical images. Other regions of the brain were masked out for viewing purposes. The image shows group results for the single pain paradigm. C: BOLD signal change in the habenula. Individual subject functional regions of interest (fROIs) were created for habenula activation to a 4-block noxious thermal stimuli paradigm. The individual time series were normalized and de-meaned before being averaged for a group time series as shown in B. They are graphed against a 2-wave pain paradigm representing early and late pain phases. D: single-trial average showing the activation of the right and left habenula averaging over the 4-block activation paradigm.

HABENULA FUNCTIONAL LOCALIZATION.

fMRI activation in the habenula was identified by using the activation map from the entire stimulation period to determine whether a cluster overlapped the habenula in spatially coregistered anatomical space. Using the human brain atlas (Mai et al. 2008), we also identified other thalamic structures known to activate under noxious thermal conditions, such as medial dorsal thalamic nucleus magnocellular part (MDMC), medial dorsal (MD), ventral posterior thalamic nucleus (VPL), ventral posterior lateral posterior thalamic nucleus (VLPL), anterior pulvinar (APul), and pulvinar (Pul). To determine the habenula activation, first all of the peak activations (hot spots) in that region were determined in each slice and then, using the coordinates obtained from the atlas described above and the spatially coregistered anatomical data, we identified activation peaks corresponding to the habenula versus other thalamic nuclei. This can be seen in activation images of a horizontal and several coronal slices set in comparison to atlas slides in Fig. 3.

For time series extraction, care was taken to only use the clusters localized within the anatomically defined habenula region of interest (ROI) so as not to misattribute activation from other thalamic structures. When clusters appeared to overlap more than a single definable region, thresholding was increased until separate peaks were distinguished.

Functional ROIs (fROIs) were defined for thalamic nuclei showing consistent activation to noxious heat in the same manner as previously described for the habenula. Voxels from each region, including MD and Pul, were tested against the habenula voxels to ensure there were no overlapping definitions, resulting in unique habenula activation and cluster definitions.

Group analysis.

With FMRIB's Linear Image Registration Tool (FLIRT), the individual statistical maps were registered to standard Montreal Neurological Institute (MNI) coordinate space. Precaution was taken to visually inspect all subjects for good registration around the thalamus for best habenula alignment. We paid particular attention to having a good alignment with the ventricle walls because they are easily identified in imaging. We also verified that the whole thalamus was well registered, to ensure that no significant distortion of subthalamic structures took place.

The FEAT analysis tool was used for higher-level single group average using FLAME (FMRIBs Local Analysis of Mixed Effects) with the two-EV model to capture early and late BOLD responses, as described above.

Group activation thresholds were determined with an in-house Gaussian mixture modeling (GMM) approach (Pendse et al. 2009). In GMM, the “activation” and “deactivation” distributions are used for alternative hypothesis testing instead of null hypothesis testing based on a fixed parametric form of the “null” distribution. It is one of several “adaptive techniques” to thresholding statistical maps in contrast to standard approaches, such as Bonferroni corrections or cluster-based ones. GMM does not make any assumption of the nature of the statistical distribution, including normality. It fits the statistical parametric map histogram with Gaussian functions and determines the posterior probability in each voxel for each Gaussian (class) identified. Visual inspection allows us to select which classes represent deactivation, activation, and null hypothesis. The posterior probability of each class is set to a 50% threshold for each voxel. This allows us to determine the most probable classification of a voxel in one of the groups (activation, deactivation, null). This in turn could result in an equivalent statistical threshold that is lower than the standard approach.

Single-trial average.

A single-trial average (STA) was performed from group activation masks in standard space for the left and right habenula. These two masks were transformed and corrected (as discussed for previous fROI masks) to the individual subject spaces, and functional time courses were extracted. BOLD signals were measured during baseline, stimulation, and back to baseline.

Activations were de-meaned for each subject before calculation of a group average for the left and right habenula. These results were plotted against time in order to determine left and right variance.

Functional connectivity.

We performed functional connectivity analysis to determine whether fMRI could visualize an interaction between activity in the habenula and some of the regions associated with pain processing. Partial correlation analysis was used (Smith et al. 2011). First, the habenula seed was located by drawing ROI masks on high-resolution anatomical images, localizing the habenula along the third ventricle in the horizontal plane segmentation showing the pineal gland, and further aligning along the sagittal and coronal planes for verification—similar to the initial step for the fROIs. In the previous fMRI section, confidence in ROI placement was achieved despite spatial resolution limits because overlaying functional maps showed highly localized activation within the designated habenula ROI (as shown in Fig. 3).

Because of the close proximity to cerebrospinal fluid (CSF) and the potential expansion of the habenula mask upon transformation back to functional space, masks of the third ventricle were designed. These binary masks were similarly made in high-resolution anatomical space, transformed to functional space, and thresholded at 50%. Both ROIs were binarized and multiplied in order to locate regions of overlap from transformation. The overlap was subtracted from the original habenula ROI. Because functionally inaccurate ROI designation can be extremely damaging to network estimation, the masks were carefully inspected after spatial transformations, to ensure well-defined overlap and good registration. Next, the group average functional map was positioned on the ROI. A localized activation cluster (seen in Fig. 4B) aligned on the designated habenula seed region. Each individually averaged time course was used to estimate the connections between any given pair of seeds.

Partial correlation was used to perform a normalized correlation between different time series that had been adjusted by regressing out all other time series in the data (Smith et al. 2011). The correlation was determined through the linear relationship between two variables once effects of independent factors had been excluded, revealing correlations explained by the other variables as well as hidden correlations masked by the effects of the other variables.

This method was implemented for some of the major descending and modulatory nociceptive pathways, as diagrammed in Fig. 1. The ROIs were defined in individual subject high-resolution space with a human brain atlas (Mai et al. 2008). Connectivity was assessed for the 1) frontal cortex, 2) NAc, 3) lateral hypothalamus (Hypo), 4) caudate and putamen (CPu), 5) IPn, 6) PAG, 7) raphe nucleus (RN), 8) VTA, and 9) substantia nigra pars compacta (SNc) (ROIs defined below). The afferent inputs are labeled along the left and the efferent connections along the right. The black dashed rectangle in Fig. 1 outlines the epithalamus containing the lateral and medial habenula as well as the pineal gland; the red line designates afferent connections from the peripheral nervous system, green represents descending and modulatory nociceptive pathways, and the blue line represents central inputs to the habenula that may modify pain processing.

DTI connectivity and tractography analysis.

In addition to functional connectivity, we performed a tractography analysis to further investigate structural white matter connections between the habenula and other structures as displayed in Fig. 1. Tractography was performed with DWI data processing tools of FSL. A voxelwise analysis was performed with a Markov chain Monte Carlo sampling method that models the distribution of fiber orientations or principal diffusion direction with the voxel. The probability density functions (PDFs) for the model parameters were then derived from the measured diffusion-weighted signal of each voxel. Each analysis used the default setting specifying a total of 5,000 samples.

Initial processing steps started with a brain extraction in order to separate the brain from skull and nonbrain tissue. For each subject DWI data set, a head motion and eddy current distortion correction was employed with an automated affine registration algorithm from FLIRT. The extracted brain structural images were used as reference volumes. A least-squares fit of the tensor model to the diffusion data was used to calculate a diffusion tensor for each voxel. The eigenvalues of each tensor, representing the three axes of diffusion, and the fractional anisotropy (FA) values were calculated.

Probabilistic tractography was run to tract probable structural connections to the habenula, using FDT, FMRIB's diffusion toolbox from FSL (www.fmrib.ox.ac/fsl/fdt/). This tool allows for estimation of the most probable pathways from a seed mask. Tracking was run for the extent of the subject brain space for an overall connection chart from the habenula seed ROI.

In addition, specific locations were registered for probable path definition from seed masks by creating waypoint and termination masks. These ROIs were defined from connections between the habenula and regions within the nociceptive pathways as described in Functional connectivity. Termination masks included Hypo, caudate, Nac, PAG, putamen, IPn, SNc, and VTA.

To ensure that the systems were separated and not capitalizing on the same pathways, tractography was run a second time for each of these regions, but with the addition of exclusion masks made from all adjacent ROIs being examined in this analysis that utilized similar white matter pathways. For instance, the habenula remained the seed mask and the caudate was again used as a termination mask but with the addition this time of the putamen as an exclusion mask. The putamen was similarly run with the caudate as the exclusion mask.

To investigate the habenula's potential involvement in a PAG, NAc, pain modulation loop, the habenula was again seeded and both PAG and NAc were used as waypoints. Similarly, the analysis was run with the PAG as the seed and NAc as the seed while the other two remained waypoints. This analysis only tracks fibers from the habenula that are inclusive of both targets.

ROI MASK DEFINITION FOR DTI.

ROI masks for structural connectivity targets were defined with the use of volumetric T1-weighted MPRAGE images from each individual subject. Using the specific anatomy of each subject bolstered accuracy. It avoids additional registration errors from imperfect registration when seeds are first defined in standard space before transforming back to each anatomical space. However, in order to perform group analysis, probabilistic tracking had to run in standard space. Structural ROIs underwent structural to standard space transformations to standardize all masks. They were individually examined for good registration and appropriate regional overlap to the MNI atlas before being used in tracking analysis.

GROUP ANATOMICAL CONNECTIVITY ANALYSIS.

Probabilistic tractography results for individual subjects were thresholded equivalently for all subjects, minimizing confounds such as partial volume effects near borders of white and gray matter and CSF. Tracts were normalized with total streamline counts and were further thresholded to include only voxels that received at least 0.01 of the total streamlines extending from the ROI masks used to trace that tract. The individual standard space maps were thresholded, normalized, and summed to produce a group average probability map.

RESULTS

Pain Responses

The average VAS pain rating for the stimuli used was 7.6 ± 2.0 (mean ± SE). No subject terminated the experiment because of excessive pain.

Evoked Response to Noxious Heat

Painful thermal stimulation resulted in activation in several brain regions including those corresponding to the habenula. Figure 4 shows the habenula activation with the rest of the brain masked out. Individual subject maps showed habenula activation in 10 of the 11 subjects. Figure 4B shows group activation in all three planes—coronal, horizontal and sagittal. Note the location of the activations and the corresponding anatomical labels shown in Fig. 4A. To avoid averaging effects, the time series of the group shown in Fig. 4C was defined from individual subject habenula fROI time series. The graph demonstrates activation for both early and late pain waves, EV1 and EV2. When further evaluated in a single trial average for the right and left habenula combined, a two-phase response was revealed (Fig. 4D). It should be noted that there are no significant differences between the two (left and right habenula) responses, although the stimulus was administered to the left foot.

Activations Correlating with Early and Late Phases

Further evaluation of the activation patterns associated with the early (EV1) and late (EV2) phases is provided in Tables 1–6. Clusters are divided into activation and deactivation during early and late pain phases. Overlapping regions can be seen in Tables 5 and 6: Table 5 shows overlapping activation clusters, and Table 6 shows overlapping deactivation clusters. Figure 5 shows representative brain slices for activation and deactivation from EV1 and EV2 where regions showed segregation of responses.

Table 1.

Positive activation in response to early phase

			Coordinates, mm
Brain Region	Lat.	Zstat	x	y	z	Volume, cm3
Frontal
Superior	R	3.40	6	52	36	1.14
Superior	L	2.92	−30	4	36	3.62
Superior orbital	R	2.87	14	24	48	0.68
Inferior triangular	R	2.91	36	30	28	0.47
Inferior triangular	L	3.07	−52	16	2	3.94
Inferior orbital	R	3.15	38	30	−16	2.49
Inferior orbital	L	2.88	−44	40	−4	3.25
Inferior operculum	L	2.99	−42	10	8	1.38
Precentral	R	2.82	38	−8	48	5.90
Precentral	L	2.94	−38	−2	54	2.28
Supp motor area	R	2.85	14	−4	46	6.85
Supp motor area	L	2.90	−26	−14	46	0.30
Parietal
Postcentral	L	3.02	−62	−20	22	0.68
Supramarginal	R	3.49	52	−38	30	2.36
Supramarginal	L	2.84	−54	−34	32	0.94
Occipital
Rolandic operculum	R	3.21	54	4	8	2.15
Rolandic operculum	L	2.94	−50	−6	8	2.20
Inferior	R	2.91	28	−94	−4	3.18
Inferior	L	2.81	−20	−86	−2	2.80
Temporal
Pole superior	L	3.05	−48	18	−12	0.90
Middle	R	3.04	68	−22	−18	2.30
Middle	L	2.91	−44	−26	28	1.05
Superior	R	3.47	52	−28	16	2.99
Superior	L	3.01	−58	−30	22	1.6
Lingual	L	3.12	−8	−96	−18	1.216
Transverse	R	2.92	30	−30	8	0.68
Cingulum
Anterior	L	3.31	0	26	30	4.856
Middle	L	3.28	−4	14	40	6.144
Anterior	R	3.02	2	14	22	6.08
Insula
Insula anterior	R	3.43	34	10	2	6.35
Insula anterior	L	3.58	−38	−2	−8	1.90
Subcortical
Putamen	L	3.10	−28	6	8	4.31
Putamen	R	3.24	26	2	4	2.43
Habenula	R	2.02	6	−26	2
Habenula	L	2.35	−6	−26	2
Thalamus	R	3.29	20	−22	6	4.63
Thalamus	L	3.53	−8	−14	6	5.65
VTA/SNc		3.03	6	−22	−10	3.86
Brain stem/cerebellum
spV	L	2.95	−2	−40	−52	5.49
Vermis 4 5		2.96	0	−54	−24	1.54
Vermis 9		2.95	−2	−56	−34	2.38
Cerebellum	R	3.23	6	−72	−44	3.56
Cerebellum crus 2	L	2.82	−36	−56	−42	1.94
Cerebellum 8	R	2.82	34	−64	−56	1.03
Cerebellum 6	R	2.91	32	−66	−20	3.42
Cerebellum 6	L	3.01	−32	−66	−24	1.78
Cerebellum crus 2	R	3.09	38	−68	−44	1.82
Cerebellum crus 2	L	3.12	−36	−76	−38	2.77
Cerebellum crus 1	R	2.83	26	−80	−34	9.73
Cerebellum crus 1	L	3.36	−26	−88	−22	4.20

Significant cluster activation during the early phase that includes the habenula, cortical frontal regions extending posterior to the supplementary motor cortex, cerebellum, brain stem, and the parietal, occipital, insula, and temporal regions showing bilateral activation is shown. In the subcortical regions, the putamen and thalamus show significant clustering on both left (L) and right (R) sides.

Table 2.

Negative activation in response to early phase

			Coordinates, mm
Brain Region	Lat.	Zstat	x	y	z	Volume, cm3
Frontal
Rectus	R	−3.81	12	26	−22	3.20
Rectus	L	−3.67	−8	20	−20	1.42
Superior orbital	L	−3.31	−18	22	−14	1.15
Inferior orbital	R	−2.80	26	20	−22	0.29
Inferior orbital	L	−3.08	−22	16	−16	0.47
Medial orbital	R	−2.99	12	4	−14	1.21
Inferior		−2.80	−20	34	−18	0.66
Olfactory	R	−3.02	20	12	−20	0.75
Olfactory	L	−2.77	−12	10	−16	0.50
Precentral	L	−2.55	−38	−24	64	0.34
Parietal
Postcentral	R	−3.15	42	−28	62	0.87
Postcentral	L	−2.85	−30	−38	60	0.88
Temporal
Pole middle	R	−2.81	30	12	−42	0.64
Pole middle	L	−2.94	−34	12	−34	3.00
Inferior	R	−2.35	34	6	−42	0.39
Inferior	L	−2.65	−26	2	−28	0.30
Fusiform	L	−2.84	−28	2	−42	0.59
Lingual	R	−2.94	12	−72	−6	1.43
Parahippocampus
Parahippocampal	L	−2.92	−16	−8	−24	0.36
Subcortical
Caudate	L	−3.21	−10	18	−10	0.33
Hypothalamus	R	−2.73	6	2	−8	0.26
Hypothalamus	L	−2.80	−4	−2	−10	0.39
Hippocampus	L	−2.67	−32	−6	−26	0.45

Significant deactivation during the early phase response is shown. Regions include bilateral frontal, parietal, occipital temporal, and brain stem/cerebellum. The parahippocampal shows unilateral left-side deactivation and the subcortical regions with deactivation including the caudate, hypothalamus, and hippocampus.

Table 3.

Positive activation in response to late phase

			Coordinates, mm
Brain Region	Lat.	Zstat	X	y	z	Volume, cm3
Frontal
Middle	L	3.33	−30	42	20	0.91
Superior	L	2.98	−24	20	54	0.50
Inferior operculum	L	3.24	−52	14	22	0.72
Precentral	R	2.70	46	−16	52	0.56
Precentral	L	2.63	−46	2	32	0.25
Paracentral lobule	R	2.92	6	−34	66	0.29
Parietal
Postcentral	R	3.10	58	−22	48	0.38
Inferior	L	2.83	−56	−34	46	0.56
Supramarginal	L	3.41	−58	−38	32	1.86
Superior	L	3.24	−28	−58	50	1.16
Angular	L	2.77	−34	−60	38	0.22
Precuneus	L	3.26	−6	−66	50	1.78
Occipital
Middle	R	2.79	46	−78	6	0.29
Inferior	R	2.82	40	−82	−8	0.33
Temporal
Fusiform	R	2.69	44	−56	−24	0.25
Inferior	R	2.92	44	−68	−6	0.68
Lingual	R	3.59	20	−82	−14	2.28
Cingulum
Middle	L	2.67	0	−42	52	0.27
Subcortical
Habenula	R	2.27	6	−24	2
Habenula	L	2.17	−8	−26	4
Thalamus	R	2.80	8	−18	8	0.38
Thalamus	L	3.36	−8	−20	6	1.33
Brain stem/cerebellum
Cerebellum 8	R	2.62	34	−56	−54	0.51

Positive activation in the late response shows up on the left in the frontal (middle, superior, inferior operculum, and precentral) and parietal (inferior, supramarginal, superior, angular, and precuneus) regions and cingulum. The occipital and temporal regions showed unilateral right-side activation. The subcortical regions with positive activation include bilateral clusters in the habenula and thalamus.

Table 4.

Negative activation in response to late phase

			Coordinates, mm
Brain Region	Lat.	Zstat	x	y	z	Volume, cm3
Frontal
Superior medial	L	−3.46	−12	46	6	0.23
Rectus	R	−3.49	14	32	−20	0.22
Rectus	L	−3.75	−10	34	−16	0.31
Middle orbital	L	−4.10	−6	22	−14	0.43
Inferior orbital	R	−3.92	24	14	−22	1.50
Temporal
Pole middle	L	−3.94	−34	8	−34	1.78
Cingulum
Post	R	−3.57	6	−48	26	0.28
Parahippocampus
Parahippocampal	L	−3.62	−28	−2	−30	0.55
Cortical
Cerebral	L		−30	−28	6	4.01
Frontal
Middle	L		−28	42	24	16.10
Inferior Orbital	L		−48	22	−4	1.56
Middle	L		−42	16	46	1.44
Inferior Operculum	L		−52	14	20	3.81
Precentral	R		40	4	34	18.24
Precentral	L		−48	2	34	3.22
Parietal
Inferior	L		−60	−40	40	11.70
Temporal
Pole superior	L		−50	12	−8	1.95
Superior	L		−62	−38	18	5.24
Lingual	R		14	−86	−14	9.94
Insula
Insula anterior	L		−40	10	0	1.64
Subcortical
Pallidum	L		−18	4	0	4.94
Thalamus	R		12	−8	8	2.46
Thalamus	L		−6	−18	8	20.90
Habenula	R		−4	−24	4
Habenula	L		4	−23	4
Brain stem/cerebellum
Cerebellum 8	R		40	−48	−56	15.48
Cerebellum crus 1	R		28	−64	−34	11.47
Vermis 7			6	−78	−28	7.74
Cerebellum crus 2	R		4	−82	−26	3.38

Deactivation in the late phase in the frontal superior medial, rectus, middle and inferior orbital is shown. Individual clusters are in the temporal (left), cingulum (right), and parahippocampus (left).

Table 5.

EV1 and EV2 overlap: positive activation

		Coordinates, mm
Brain Region	Lat.	x	y	z	Volume, cm3
Cortical
Cerebral	L	−30	−28	6	4.01
Frontal
Middle	L	−28	42	24	16.10
Middle	L	−38	42	26	0.78
Inferior orbital	L	−48	22	−4	1.56
Middle	L	−42	16	46	1.44
Inferior operculum	L	−52	14	20	3.81
Inferior operculum	L	−48	10	22	3.09
Precentral	L	−46	8	40	4.40
Precentral	R	40	4	34	18.24
Precentral	L	−48	2	34	3.22
Parietal
Inferior	L	−60	−40	40	11.70
Inferior	L	−44	−42	56	7.99
Temporal
Pole superior	L	−50	12	−8	1.95
Superior	L	−62	−38	18	5.24
Lingual	R	14	−86	−14	9.94
Lingual	R	20	−90	−12	3.89
Insula
Insula anterior	L	−40	10	0	1.64
Insula posterior	L	−44	−6	−4	0.51
Subcortical
Pallidum	L	−18	4	0	4.94
Thalamus	R	12	−8	8	2.46
Thalamus	R	8	−18	8	1.11
Thalamus	L	−6	−18	8	20.90
Thalamus	R	18	−20	10	7.11
Thalamus	R	12	−22	4	3.77
Habenula	R	4	−23	4
Habenula	R	−4	−24	4
Brain stem/cerebellum
Cerebellum 8	R	40	−48	−56	15.48
Cerebellum crus 1	R	28	−64	−34	11.47
Vermis 7		6	−78	−28	7.74
Cerebellum crus 2	R	4	−82	−26	3.38

This table shows the regions of significant activation overlap between the early and late pain phases (EV1 and EV2). Regions include the middle, inferior orbital and precentral frontal, inferior parietal, cerebellum, pole superior, superior and lingual temporal, insula and subcortical regions; pallidum thalamus and habenula.

Table 6.

EV1 and EV2 overlap: negative activation

		Coordinates, mm
Brain Region	Lat.	x	y	z	Volume, cm3
Frontal
Rectus	R	10	46	−18	1.4
Rectus	L	−8	18	−14	7.50
Inferior orbital	R	26	18	−22	13.31
Inferior orbital	L	−20	8	−20	2.05
Temporal
Pole middle	R	32	12	−36	4.26
Pole Middle	L	−34	12	−34	11.94

This table highlights the overlapping deactivation regions of EV1 and EV2. Regions include the frontal rectus and inferior orbital as well as the pole middle temporal.

Fig. 5.

Deactivation and activation maps. Activation and deactivation during the first and second halves of stimulation [labeled EV1 (left) and EV2 (right)], representing an early and a late phase. Activation is shown in red, and deactivation is marked by blue. Afferent connections are shown in green, the efferent connections are yellow, and the habenula is red. Activation is shown in the early pain phase in caudate, putamen, VTA, Snc, and habenula. Classic pain network activation is also displayed in somatosensory regions. Deactivation in EV1 is detected in frontal cortex and hypothalamus. There was overlap between early and late pain phases in activation of the habenula. Other regions of overlap can be seen in Tables 5 and 6, including activation overlap in the middle, inferior operculum, and precentral frontal regions, some clusters in the temporal and insula, cerebellum, and the pallidum and thalamus in the subcortical regions. Deactivation was only seen in the frontal (rectus and inferior orbital) and pole middle temporal. VTA, ventral tegmental area; SNc, substantia nigra pars compacta.

Early-phase brain activations.

As shown in Table 1, a number of cortical, subcortical, and cerebellar regions showed increased activation. Of the cortical regions, the inferior orbital region, the anterior insula, temporal, and cingulate regions were noteworthy. Significant activation in subcortical regions was observed in the habenula, putamen, and thalamus, as well as the VTA. Clusters were also present in the cerebellum. Decreased activations were also associated with the early phase of the response. These included a number of cortical regions, as well as the parahippocampus, caudate, hypothalamus, and hippocampus (Table 2).

Late-phase brain activations.

The second wave of activation during thermal stimulation showed increased activation in the frontal lobe (middle, superior, inferior operculum, and precentral), parietal regions (inferior, supramarginal, superior, angular, and precuneus), and cingulate cortex (Table 3). The occipital and temporal regions showed unilateral right-side activation. The subcortical regions with positive activation included bilateral clusters in the habenula. Fewer regions showed decreased activation for EV2 (Table 4).

Regions of overlap for EV1 and EV2.

For activation, aside from the habenula itself, we observed overlap mainly in the cerebellum, temporal, and frontal lobes. For decreased activation, regions showing overlap included the inferior orbital gyrus and gyrus rectus in the frontal lobe and the temporal lobe.

Functional Connectivity Results

Partial correlation results show significant functional connectivity between BOLD signals in the PAG and habenula and the putamen and habenula, exposing interrelated activity in pain processing. Binarized results are displayed graphically in Fig. 6B, showing significant correlations in red (q < 0.05; 1-pval images).

Fig. 6.

Functional connectivity. A is an individual subject habenula ROI used as the seed mask for connectivity analysis. B illustrates the statistically significant interactions between ROIs by binarizing the partial correlation group average, displaying false discovery rate (FDR)-corrected (q < 0.05) 1-pval image for visualization. Boxes with a white X represent significant habenula connectivity. 3V CSF, third ventricle CSF; IPn, interpeduncular nucleus; NAc, nucleus accumbens; PAG, periaqueductal gray; Raphe, raphe nuclei; SNc, substantia nigra pars compacta; VTA, ventral tegmental area; FrCtx, frontal cortex.

DTI Connectivity Results

Probabilistic tract connections to the habenula can be seen clearly that correspond to the diagrammatic afferent inputs in Fig. 1: caudate, putamen, hypothalamus, and NAc, as seen from group average maps in Fig. 7A. Tract connections between the habenula and posited efferent connections to other brain structures were found with IPn, PAG, Raphe, VTA, and SNc, shown in Fig. 7B.

Fig. 7.

Diffusion tensor imaging (DTI) connectivity. A: group probabilistic tracking images to posited afferent connections: caudate, putamen, hypothalamus, and nucleus accumbens (NAc). B: group probabilistic tracking from the habenula seed to the posited efferent connections: interpeduncular nucleus (IPn), periaqueductal gray (PAG), raphe nucleus (Raphe), ventral tegmental area (VTA), and substantia nigra pars compacta (SNc).

Individual subject tracts are shown in Figs. 8–10 between the habenula and multiple brain regions, including the caudate, SNc/VTA, NAc, IPn, Hypo, putamen, and RN. The habenula shows connection to the caudate through the anterior thalamic radiation. Coronal slices display this connection in MNI coordinate space from y = 16 to −20. Three horizontally aligned images show tracking from z = 4 to −2. The corticospinal tract is highlighted in the probabilistic tractography image in Fig. 8B, connecting the habenula to the SNc/VTA. Sagittal slices ranging from x = 11 to −6 and horizontal slices ranging from z = −2 to −29 are highlighted to show the probable fiber projections. In Fig. 9 probable fiber tracts from the habenula to NAc are displayed through coronal slices y = −14 to 31, and two slices from the horizontal plane, z = −2 and 1. The tracts to IPn are shown in sagittal slices x = −2 to 0 and slices z = −19 to −2 in the horizontal place. Tracking from the habenula to Hypo, putamen, and RN are shown in all three planes in Fig. 10.

Fig. 8.

Tractography in individual subjects. A: single-subject tracking results from the habenula seed to the caudate. Anterior thalamic radiation connects the regions as shown by the MNI space coronal slices from y = −20 to 16 and the horizontal slices located on right from z = −4 to 2. B: probable tracts from a single subject showing the corticospinal tract connecting the habenula to SNc/VTA. Three sagittal slices from x = 11 to −6 and 6 horizontal slices from z = 2 to −29 are displayed.

Fig. 9.

Tractography in individual subjects. A: tracking images from a habenula seed to the NAc. Probable fiber tracts are displayed through coronal slices y = −14 to 31 and 2 slices from the horizontal plane, z = −2 and 1. B: tractography results for the habenula and interpeduncular nucleus (ipn). Tracts are displayed in the sagittal slices x = −2 to 0 and slices z = −19 to −2 in the horizontal plane.

Fig. 10.

Tracking images from the habenula to the hypothalamus (A), putamen (B), and raphe nuclei (C) are displayed in all 3 planes (coronal, sagittal, and horizontal).

Additionally, these fiber tracts projected through the whole brain in a single ROI connectivity analysis, where the habenula was seeded without any restricting waypoint, termination, or exclusion targets. Most subjects showed some additional probable fiber connections to the frontal cortex. The group average whole brain tractography is shown in the horizontal plane in Fig. 11. Tracking extended into the frontal cortex via corpus callosum (white matter tracts connecting bilateral frontal areas). The corticospinal, anterior thalamic radiation, and optic tracts also showed extensions toward the spinal cord, midbrain, and frontal regions. In addition, some of the individual subjects showed tracts extending toward the cerebellum and spinal cord.

Fig. 11.

Whole brain connectivity for the habenula. The habenula seeded to a whole brain structural connectivity analysis is displayed via the horizontal plane in MNI space from slices z =−17 to 4. The optic, anterior thalamic radiation, corpus callosum, and corticospinal tracts show connections from the habenula to the midbrain, frontal regions, and toward the spinal cord. Graphic on left shows a coronal-plane image of some of the major regions and the connecting tracts. IPn, interpeduncular nucleus; VTA, ventral tegmental area; SNc, substantia nigra pars compacta; PAG, periaqueductal gray.

Because tracking to regions that were relatively close showed similar results, as described in methods, we also ran tractography with additional exclusion masks from adjacent ROIs. The results still showed strong fiber tracking to the habenula, giving credence to the original results that both regions show structural connections. The results were consistent for all exclusion pairs, including the SNc, VTA, putamen, NAc, and caudate. In four of the subjects, fiber tracts were distinguished between the PAG, habenula, and NAc. This was run with the same habenula seed with PAG and NAc mutually inclusive waypoint masks. Multiple waypoint masks discarded tracts that did not pass through all masks.

Finally, the supplementary motor cortex was masked and run as a control region, an area known to have no physical fiber paths to the habenula. This analysis resulted in no connections. The habenula showed no significant probable tracts in any of the 11 subjects, further confirming the specificity of our DTI methodology.

Summary of Results

Through structural connectivity and fROI determination, our results show that the human habenula can be localized in imaging and is incorporated into pain processing systems. Noxious heat initiates bilateral habenula activation in both the early and late phases of stimulation. In addition, the habenula showed interrelated activity with regions that have known afferent (putamen) or efferent (PAG) connections with the habenula.

Study Caveats

It should be noted that this study only included male subjects. This article was devised to show an initial evaluation of brain function. We will pursue sex differences in subsequent studies. Given the increased prevalence of chronic pain in women, we are hopeful that the habenula is a structure that can provide insights into sex differences in pain responses.

In addition, given the habenular response during the early and late phases of the noxious stimulation as well as its previously described structural connections, it would be worth pursuing psycho-physiological interactions (PPI) in a future study. This would help explore the connectivity patterns exposed in the functional connectivity analysis.

Finally, it should be noted that we assume the localization of the habenula is accurate based on all the convergent structural (such as connections mimicked from animal literature) and functional evidence (the thalamic segmentation as demonstrated in Fig. 4) that have been presented, despite limits of spatial resolution.

DISCUSSION

The habenula is a major relay that interconnects the limbic forebrain and the mid- and hindbrain and is situated to convey information related to emotional processing, where it alters the serotonergic and dopaminergic systems responsible for pain modulation. While its structural connectivity and functional interaction to major pain processing components (Bianco and Wilson 2009; Shelton et al. 2012) have been shown in the animal literature and mostly implied in human research, we believe this study is the first to report habenula activation in response to noxious stimulation in humans. We show that BOLD responses in the habenula to painful stimuli can be localized despite limitations of functional resolution, lack of specific landmarks, and spatial registration of functional images with structural images (Borsook et al. 2004; DaSilva et al. 2002). In this study, anatomical, functional, and connectivity approaches showed evidence convergent with what had previously only been reported in animal studies (see Fig. 1 and Bianco and Wilson 2009). Furthermore, a recent review on imaging the habenula noted that prior pain studies show prominent activations in the region, despite not being specifically reported (Shelton et al. 2012). Nonpain human neuroimaging studies have also reported activation within the human habenula (Ide and Li 2011; Li et al. 2008; Morris et al. 1999; Sartorius et al. 2010; Savitz et al. 2011; Ullsperger and von Cramon 2003).

Anatomical Connectivity of Habenula in Humans

The anatomical connectivity results (Figs. 8–11) showed a parallel with the animal literature. The images defined efferent connections (viz., IPn, PAG, RN, VTA, and SNc), as well as afferent connections (caudate, putamen, frontal cortex, Hypo, and NAc) (Bianco and Wilson 2009; Brinschwitz et al. 2010; Craig 2003; Goto et al. 2005; Herkenham and Nauta 1977). In addition to the designated termination masks, the whole brain connectivity diagram (Fig. 11) showed white matter fiber tracts extending rostrally to the frontal cortex and caudally toward the spinal cord. The latter highlights potential nociceptive pathway connectivity with the spinal cord in addition to its efferent projections to multiple brain stem nuclei noted above. The lateral hypothalamus, involved in pain modulation (Dafny et al. 1996), is one of the known afferent inputs projecting to the habenula complex (Herkenham and Nauta 1977; Parent et al. 1981). These results are consistent with noxious stimulus-evoked activity being significantly modulated when the habenula is stimulated. Other direct spinal-habenula connections, where spinal lamina I injections of the anterograde tracer Phaseolus vulgaris leucoagglutinin (PHA-L) in the cat terminate in the lateral habenula, have also been reported (Craig 2003). Thus the data replicate connectivity in humans previously shown in animals, where IPn-habenula (Kuan et al. 2007; Qin and Luo 2009; Rossiter et al. 1977), RN-habenula (Araki et al. 1988), Hypo-habenula (Goto et al. 2005), etc. have been described (Bianco and Wilson 2009).

Functional Activation (fMRI)

Noxious heat produced activation in a number of structures previously reported by us (Becerra et al. 2008; Moulton et al. 2011) and others (Apkarian et al. 2005; Peyron et al. 2000), as well as significant bilateral activation in the habenula. What drives habenula activation by pain is unclear: Is it via direct nociceptive pathways or through secondary systems activated by pain? With respect to the former, nociceptive afferents to the lateral habenula via the lateral hypothalamus (a region involved in afferent nociceptive processing; Strassman et al. 1996) suggest that these inputs may account for the robust activations observed. In addition, afferent spino-hypothalamic pathways are bilateral. This may account for the bilateral activation, although direct interhabenula (commissural) connections are also known to be present (Kim 2009). Our single-trial analysis (STA) showed no temporal differences in activation between the right and left habenula, suggestive of bilateral activation of the region following unilateral painful stimulation. The habenula has medial and lateral divisions, but it is through the lateral habenula that afferent nociceptive pathways may access the structure (Jiyi et al. 1982). Although we cannot define differences between activations of the two divisions, we do see highly segregated activations of the habenula with different brain regions when the activation is temporally separated into its two phases. Clearly visible biphasic responses were observed, which support the early and late phase model approach. Prior studies have reported a biphasic response to noxious heat that may segregate functional processing in the brain (Becerra et al. 2001). Here, temporal segregation of the two peaks (EV1 and EV2) showed major differences in activation patterns with minimal overlap (see Tables 5 and 6). The differences in brain regions that correlated with early or late phases may reflect a number of processes. 1) Brain regions may simply be differentially activated without any relationship to habenula activation. When compared with prior data reported for regions that showed early- versus late-phase BOLD responses, the regions could be segregated into early-phase emotional regions and late-phase predominantly sensory systems (Becerra et al. 2001). 2) Early and late activations could relate to the different habenula circuitry. Based on known anatomical reports (Bianco and Wilson 2009), BOLD activation was observed in three regions that are known to have connections with the habenula: the frontal regions, the caudate/putamen, and the VTA/SNc. With respect to frontal regions, aside from defined connections, the habenula (lateral) exerts a tonic inhibitory effect on dopamine in the medial prefrontal cortex (mPFC) (Lecourtier et al. 2010). In addition to interacting with the mPFC, mPFC and lateral habenula nucleus (LHb) inputs converge onto GABAergic neurons in the dorsal raphe nucleus (DRN), which in turn inhibit the activity of 5-HT neurons (Varga et al. 2003). We observed activation in the mPFC (anterior cingulate) predominantly in EV1. We observed activation in the caudate and putamen, which are both structures known to have connections with the globus pallidus (GP). These activations were more prominent in EV1. The habenula receives inputs from the basal ganglia, a set of structures also known to be involved in pain (Borsook et al. 2010) and reward processing (Hong and Hikosaka 2008). Through these circuits, the reward value of a stimulus may be integrated through the habenula via dopaminergic pathways (Bromberg-Martin et al. 2010). We observed activation in the caudate and putamen, which are known to have connections with the GP, a major output station of the basal ganglia (see Anatomical Connectivity of Habenula in Humans). The VTA/SNc was the only habenula projection region noted in Fig. 1 where activation was observed, again in EV1. An anticorrelation between VTA/SNc and the habenula is consistent with the role the habenula plays in firing dopamine-suppressing neurons to these regions (Matsumoto and Hikosaka 2007).

Functional Connectivity

In addition to reproducing anatomical connections reported in animal data, partial correlation functional connectivity results showed an interrelated activity with the habenula for two regions: habenula-putamen and habenula-PAG. The putamen has a well-established role in motor processes, although recent evidence suggests that it shares a role in sensory components of pain-related processing (Starr et al. 2011). It may impact neural processes influencing and determining behavioral relevance and saliency of nociceptive inputs (Starr et al. 2011). Functional connectivity results with the PAG expound upon pain relationships reported with the habenula, including the habenula pain interaction with the PAG (Yu and Han 1990), and the PAG's role in pain modulation (Behbehani 1995; Ren and Dubner 2002; Tracey et al. 2002). As mentioned above, a unidirectional loop (habenula, PAG, and NAc) has been described that is important in pain modulation (Ma and Han 1991).

A Putative Role of the Habenula in Pain Processing

The literature indicates that the habenula may play a role in multiple and diverse functions that are implicated in pain processing including cognition, addition, aversion, learned helplessness, and reward (Frahm et al. 2011; Friedman et al. 2010; Hikosaka et al. 2008; Li et al. 2011; Matsumoto and Hikosaka 2007; Paul et al. 2011). Pain is a multidimensional experience. The unique location and connectivity of the habenula—receiving inputs from frontal regions and sending output to brain stem regions—clearly make it a “structure of interest” in the pain field. This is because it is well positioned to orchestrate the interactions between reward processing and pain modulation in brain stem regions through the regulation of the transmission of monoamines such as dopamine and serotonin (Lee and Goto 2011). As summarized by Hikosaka: “As a highly conserved structure in the brain, the habenula provides a fundamental mechanism for both survival and decision-making” (Hikosaka 2010). The structure encodes negative reward-related events such as negative reward prediction error signals evaluated in cortical regions (Hong et al. 2011). Outputs to brain stem regions modulate sensory gating (Ellison 1994), protective behaviors (Pobbe and Zangrossi 2010), and endogenous tone in response to pain and may be thus construed as involved in stress evasion (Hikosaka 2010). Its role in chronic pain is yet to be determined. However, given the insights of the structure to conditions that are comorbid with chronic pain, including addiction (Fowler et al. 2011; Varga et al. 2003) and depression (Sartorius and Henn 2007), abnormal functioning in the habenula may have a significant role in the chronic pain state.

Conclusion

Increasingly, studies are pointing toward the habenula's role in behavioral functions and emotional processes. This is the first study reporting the successful use of MRI to measure pain responses in the habenula. Previously reported afferent (e.g., putamen) and efferent (e.g., PAG) pathways in animals were replicated in humans with probabilistic tractography and significant functional correlations between the habenula, and these structures were confirmed. Future studies evaluating the function of the habenula in pathological pain conditions will provide further insights into the salience of its role in pain processing.

GRANTS

This work was supported by National Institute of Neurological Disorders and Stroke Grants 1R01 NS-065051 and K24 NS-64050 and NIDA Grant K01 DA-024289.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: L.S., G.P., N.M., and E.A.M. analyzed data; L.S. drafted manuscript; A.L. edited and revised manuscript; L.B. and D.B. conception and design of research; L.B. and D.B. performed experiments; L.B. and D.B. approved final version of manuscript.