The medial temporal lobe in nociception: a meta-analytic and functional connectivity study

INTRODUCTION

The medial temporal lobe (MTL) is a region canonically responsible for memory [116], but several pain neuroimaging studies have reported activation within this region. An intriguing intersection of pain and memory comes from the case of patient H.M. who had a high tolerance for heat pain, in addition to global amnesia, following bilateral MTL resection for epilepsy [51]. The MTL region consists of the hippocampus (HC) and subiculum, parahippocampal gyrus (PHG), entorhinal and perirhinal cortices, and amygdala [130]. Activation of various MTL regions is observed in response to nociceptive stimuli of various modalities [66,81,85,113,124,146], and in different tissue types—cutaneous, muscular and visceral [23,61,137]. In particular, several studies have associated MTL activation during experimental pain to emotionally-driven modulatory processes [11,114,119,121]. However, there is little specificity or congruence across studies regarding which MTL subregions are involved in pain, and no specific role has been ascribed to MTL regions in nociceptive processing. Of importance, there has yet to be a formal investigation of the spatial consistency of MTL activations in healthy participants during experimental pain.

In addition to studies in healthy participants, extant evidence indicates aberrant MTL activity in various chronic pain conditions [56,104,108]. Recent evidence in patients with back pain suggests that the transition from subacute to chronic pain may be mediated by MTL structures [134]. Specifically, reduced MTL resting state functional connectivity (rsFC) to other cortical regions [107] and smaller MTL volume [135] predict this transition. However, as with experimental pain, it is not clear which regions within the MTL are implicated in chronic pain and what their role may be.

The overall aim of this study was to determine regions of the MTL consistently involved in nociceptive processing and pain modulation in health and disease. First, we sought to determine which regions of the MTL show consistent spatial activation in response to (1) experimental pain in healthy participants compared to baseline control conditions, and (2) chronic pain patients compared to healthy participants. To that end, we performed two coordinate-based meta-analyses of functional MRI (fMRI) studies of pain that indicated MTL engagement. We expected our first meta-analysis to show consistent MTL activation in the HC and PHG across studies reporting MTL activity. We expected our second meta-analysis to show consistent MTL activation in the HC across chronic pain neuroimaging studies. These analyses are the first formalized investigation of MTL functioning in nociceptive processing.

To generate hypotheses about the mechanistic role of the HC in nociceptive processing from existing data, it is important to explore its network interactions. Thus, we aimed to determine the connectivity of those regions elucidated by our meta-analyses using data from four distinct cohorts of chronic low back pain patients (CBP). Given previous observations of abnormal MTL engagement in CBP, we predicted that CBP would show abnormal rsFC to regions involved in processing the affective dimension of pain, i.e., the anterior insula, midcingulate cortex, amygdala and medial prefrontal cortex [25,31,33,111,134].

METHODS

Our ALE meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2015 guidelines and checklist [103].

Article selection criteria

Our article selection followed the PRISMA flow diagrams [102] as shown in Fig.1. Studies were excluded based on any of the following nine criteria: (1) animal studies; (2) did not use standardized stereotactic (Montreal Neurological Institute (MNI) or Talairach) brain coordinates; (3) case reports; (4) diagnostic or surgical MRI; (5) structural imaging; (6) studies of acute medical pain conditions; (7) functional magnetic resonance imaging (fMRI) studies without a baseline control (for experimental pain studies) or control group (for chronic pain studies); (8) studies not written in the English language; and (9) studies that are not peer-reviewed. As such, we selected human neuroimaging studies with whole-brain and region-of-interest analyses, reporting stereotactic brain coordinates. Three investigators (LA, MG, MM) independently performed the database searches and followed PRISMA guidelines by screening and determining study eligibility. In addition, quality assessment for each article selected was conducted by using a modified version of Downs and Black’s quality assessment score [18,32]. We accounted for statistical significance of the results by qualitatively assigning a maximum score of 2 for articles, which applied multiple comparison corrections to analyses (the Quality Assessment Scoresheet is available in Supplementary).

Fig.1.

PRISMA flow diagram representing the process for article selection.

Database searches

Traditional databases

We conducted a systematic search through the Pubmed, Web of Science, Embase and Medline databases on November 9, 2017. A keyword search of the following terms was conducted on all databases (medial temporal lobe OR hippocamp* OR parahippocamp*) AND pain AND ((FMRI OR functional magnetic resonance imaging OR functional MRI) OR (BOLD OR blood oxygen level dependent) OR (PET OR positron emission tomography) OR (ASL OR arterial spin labelling)).

Neurosynth

We also performed a search of the Neurosynth database (http://www.neurosynth.org) [144] on November 9, 2017 as this provides the ability to cross-search keywords (‘pain’, ‘painful’ and ‘chronic pain’) with brain coordinates. This increased the sensitivity of the meta-analysis as Neurosynth extracts coordinates from the tables in a catalogued study; therefore, even if the HC was not highlighted as a finding in the title, abstract or keywords of a study, we would still be able to identify the study. The MNI coordinates used for the HC search were derived from our previous study of hippocampal parcellation [3], as follows: the right anterior HC (antHC) is (24, 13, −21); the left antHC is (−22, −12, −20); the right posterior HC is (29, −26, −9); and left posterior HC is (−27, 25, −12). We used a liberal search sphere (radius= 15 mm) to maximize the number of studies identified, and to include adjacent structures, such as the PHG and amygdala (see Supplementary Fig.1). The same article selection criteria set out for the traditional databases were applied, with the added constraint that the HC was listed as a significant finding in at least one contrast.

Coordinate-based meta-analysis

We performed a coordinate-based meta-analysis using the activation likelihood estimation (ALE) algorithm to identify consistent regions of brain activity within the MTL in healthy participants and chronic pain patients. Brain coordinates (foci) were computed using GingerALE (v3.0) to generate probabilistic maps of activation [73] (http://brainmap.org/ale).

ALE mask

We restricted the permutation space for the null-distribution of our ALE meta-analysis to the MTL region. To do so, we constructed an MTL mask on the standard MNI152 brain in FSLeyes v.0.22.6. The mask comprised of regions selected from the Jülich Histological Atlas, including the centromedial, laterobasal and superficial amygdalae, hippocampus cornu ammonis, entorhinal cortex, dentate gyrus, hippocampal-amygdaloid transition and subiculum. The final MTL mask includes the amygdala, hippocampus and parahippocampal gyrus (see Supplementary Fig.2).

Data organization

We conducted four separate meta-analyses as part of two investigations: (1) experimental pain in healthy participants and (2) chronic pain versus healthy participants. First, foci were manually extracted from selected articles, and subsequently categorized into four separate datasheets according to the following contrasts: 1) Greater MTL activation in healthy participants during experimental pain than in control conditions, such as warm, innocuous touch or baseline conditions (experimental pain > control conditions); 2) Greater MTL activation during control conditions than painful conditions in healthy participants (experimental pain < control conditions); 3) Greater MTL activation in chronic pain patients than in healthy participants (chronic pain patients > healthy participants); and 4) Greater MTL activation in healthy participants than in chronic pain patients (chronic pain patients < healthy participants). Foci derived from significant and non-significant statistical thresholds of contrasts reporting MTL activity, were extracted from all studies. Three investigators (LA, MM, MG) verified manual extraction of the foci, before running each meta-analysis.

ALE Meta-Analysis

In preparation for the ALE meta-analysis, all foci were standardized into MNI space. Foci reported using Talairach coordinates were converted with “Talairach to MNI (FSL)” [17]. For each set of foci per experiment, the number of participants were added for weighting purposes. Once finalized, each datasheet was entered and computed separately into the software using a single dataset analysis [37,38]. We used the Turkeltaub algorithm to estimate the probability of activation at each voxel in the brain by reducing both within-experiment (inter-participants) and within-group (inter-laboratory) variables [133]. The algorithm attributes an ALE score, a voxel-based value for each focus, to create a Modeled Activation (MA) map [38], calculated using the maximum value with the “random effects” selection [37]. Furthermore, each focus is blurred with a Gaussian distribution, weighted by sample size, to minimize spatial uncertainty [74]. Lastly, the cluster-level inference algorithm described by Eickhoff and colleagues [36] was used to compute the final thresholded ALE P value map with a permutation-based cluster correction p< 0.05 (cluster-forming height threshold of p< 0.005 with 1000 permutations), which uses a Monte-Carlo simulation approach. The resulting maps were visualized and labelled in MNI space using FSLeyes v.0.22.6. Slice images were visualized and labelled in MNI space using “ch2bet” on MRIcron v2016.

Functional connectivity

Given that we found consistent pain related activity in our meta-analyses in the right antHC region, and that rsFC of the antHC region has been previously shown to be a predictor for the transition from subacute to chronic back pain [107], we conducted a whole brain seed-to-voxel rsFC study of the right antHC region, to investigate whether there were connectional differences between healthy participants and CBP. Furthermore, since we have shown that the right antHC region is activated in response to experimental pain in healthy participants and has abnormal activity in chronic pain, we sought to determine whether abnormal right antHC connectivity in CBP was related to pain characteristics, including pain duration and pain intensity.

Participants

Four resting state fMRI datasets comprising a total of 156 participants (CBP (n=77) and age- matched healthy participants (n=79)) were included in the analysis. A summary of each dataset characteristics is provided in Table 1.

Table 1.

Dataset characteristics

Dataset	N	Age (mean± SD years)	Sex (W/M)	Healthy participants			CBP					Scanner	TR (s)	Head Coil channels
				N	Sex (W/M)	Age (mean± SD years)		N	Sex (W/M)	Age (mean± SD years)	Pain duration (n=69; mean ± SD years)	Pain intensity (n=59; mean ± SD VAS (0-10))
1	57	49.6 ± 8.4	26/31	31	13/18	49.9 ± 7.8	26	13/13	49.3 ± 9.1	15.0 ± 12.1	6.5 ± 1.7	3T Siemens Tim Trio	2.5	8
2	27	44.2 ± 11.4	14/13	12	6/6	42.2 ± 11.4	15	8/7	45.9 ± 11.6	5.2 ± 5.1	n/a	3T Siemens Tim Trio	2.26	8
3	48	42.9 ± 11.8	23/25	27	11/16	41.2 ± 11.9	21	12/9	45.0 ± 0.5	11.6 ± 9.5	2.6 ± 2.5	3T Siemens Tim Trio	2.5	12
4	24	44.7 ± 11.5	16/8	9	5/4	46.7 ± 0.5	15	11/4	43.5 ± 11.2	11.7 ± 7.3	4.7 ± 3.0	3T Siemens Tim Trio	2	12
Total	156	45.9 ± 10.8	79/77	79	35/44	45.4 ± 11	77	44/33	46.2 ± 10.7	11.5 ± 10.0	4.8 ± 2.8

A summary of the four datasets included in the seed-to-voxel functional connectivity.

Abbreviations: BDI Beck Depression Inventory, CBP chronic low back pain patients, M men, W women, VAS visual analogue scale.

Dataset 1

The first dataset initially consisted of 68 participants, 34 CBP and 34 healthy participants from the study by Mansour and colleagues [88] and was obtained from the “cbp_resting” provided by the OpenPain Project (OPP) database (OPP; Principal investigator: A. Vania Apkarian; http://www.openpain.org). We consented to and followed OpenPain Data Use Agreement and procedures were approved by the University of Toronto’s Human Research Ethics Board prior to data analysis. As described in the original manuscript, all participants were provided with written consent and all experimental protocols were approved and conducted according to the Northwestern University’s Institutional Review Board committee. Patients’ clinical assessment included the Short-Form of the McGill Pain Questionnaire (SF-MPQ) [98], where the visual analog scale (VAS) (0= no pain, 10= worst pain imaginable) was used to evaluate pain intensity. Participants were given the questionnaire one hour before scanning.

Dataset 2

The second dataset initially consistent of 36 participants, 20 CBP and 16 healthy participants and was previously published in a study investigating rsFC in CBP and acquired from Stone, Seminowicz and colleagues [19]. The study was approved by McGill University Faculty of Medicine Institutional Review Board, the Montreal Neurological Institute (MNI) and Hospital Research Ethics Board, and the McGill University Health Centre Research Ethics Office. The participants gave written consent before starting the study. All participants completed the SF-MPQ for pain intensity prior to scanning, however VAS data were not available.

Dataset 3

The third dataset initially included a total of 63 participants from Osaka, Japan (24 CBP, 39 healthy participants). This dataset was obtained from “BrainNetworkChange_Mano” on OPP database. We consented to and followed OpenPain Data Use Agreement. The original study was approved by the Ethics Committee for Human and Animal Research of the National Institute of Information and Communications Technology, Japan (reference 20140611) [87]. All participants gave written consent before participating in the study. Pain intensity was measured on a VAS scale as part of the SF-MPQ on the day of scanning.

Dataset 4

The fourth dataset initially included 34 participants, 17 CBP and 17 healthy participants from Cambridge, UK in “BrainNetworkChange_Mano” on OPP database. We consented to and followed OpenPain Data Use Agreement. The original study was approved by the East of England NRES Committee, Norfolk, UK (reference 13/EE/0098) [87]. All participants gave written consent before participating in the study. The VAS scale as part of the SF-MPQ was to determine pain intensity on scan day.

Resting state fMRI data acquisition parameters

Dataset 1

Functional T2*-weighting brain images were acquired using a 3T Siemens Trio whole-body scanner, with an 8-channel head coil, during rest, as follows: TR= 2.5s; TE= 30 ms; flip angle= 90°; in-plane matrix resolution= 64 × 64; number of slices= 40; slice thickness= 3mm; field of view= 256 × 256 mm; and number of volumes= 244, 300, or 305 [88]. Additionally, for realignment purposes, T1-weighted brain images for each participant were acquired using the same scanner with the following parameters: isotropic resolution 1mm; TR= 2.5s; TE= 3.36 ms; flip angle= 9°; in-plane matrix resolution= 256 × 256; number of slices= 160; and field of view= 256 × 256 mm [88].

Dataset 2

Scans were acquired using a 3T Siemens Tim Trio scanner with an 8-channel head coil. Participants were instructed to “relax, keep your eyes open, and don’t think about any one thing in particular” [19]. The functional T2*-weighted imaging scans where acquired using: echo planar imaging (EPI), TR= 2.26 s; TE= 30 ms; flip angle= 90°; in-plane matrix= 64×64; number of slices= 38; slice thickness= 4 mm; field of view= 256 × 256 mm; and number of volumes= 133 [19]. The anatomical T1 scans were acquired using the following parameters: isotropic resolution 1mm; TR= 2.3 s; TE= 2.98 ms; flip angle= 9°, in-plane matrix resolution= 256 × 256; number of slices= 176; and field of view= 256 × 256 mm [19].

Dataset 3

The MRI scans were performed using a 3T Siemens MRI Scanner Tim Trio scanner with a 12-channel head coil at CiNet (Osaka, Japan). During resting state scanning, participants were given the following instructions: “please relax during the scan; do not sleep and keep looking at the fixation point (a tiny cross-hair) presented at the center of the display; do not think of anything in particular.” Functional T2* images were acquired with the following parameters: TR= 2.5s; TE= 30 ms; flip angle= 80°; in-plane matrix resolution= 64 × 64; number of slices= 41; field of view= 212 × 212 mm; and number of volumes= 234 [87]. The anatomical T1 scans were acquired using the following sequence: isotropic resolution 1mm; TR= 2.25s; TE= 3.06 ms; time of inversion= 900 ms; flip angle= 9°; in-plane matrix resolution= 256 × 256; number of slices= 208; and field of view= 256 × 256 mm [87].

Dataset 4

The MRI scans were performed with a 3T Siemens MRI Scanner Tim Trio scanner with a 12-channel head coil at Addenbrooke’s Hospital in Cambridge, UK. During resting state scanning, participants were instructed to: “please relax during the scan; do not sleep and keep looking at the fixation point (a tiny cross-hair) presented at the center of the display; do not think of anything in particular.” Functional T2* images were acquired with the following parameters: TR= 2s; TE= 30 ms; flip angle= 80°; in-plane matrix resolution= 64 × 64; number of slices= 32; field of view= 212 × 212 mm; and number of volumes= 300 [87]. The anatomical T1 scans were collected using the following sequence: isotropic resolution 1mm, TR= 2.3s; TE= 2.98ms; time of inversion= 900ms; flip angle= 9°; in-plane matrix resolution= 256 × 256; number of slices= 176; and field of view= 256 × 256 mm [87].

fMRI data preprocessing

FMRI data were preprocessed for whole brain rsFC using CONN toolbox v17f (http://www.conn-toolbox.org) implemented in Statistical Parametric Mapping software package (SPM v.12; http://www.fil.ion.ucl.ac.uk/spm/software/spm12)) and ran on MATLAB (R2016b v.9.1; Mathworks, Nantick, MA) [141]. Each participant’s anatomical T1-weighted and functional T2*-weighted scans were imported into CONN. Data was preprocessed as follows. Briefly, functional T2* scans underwent co-registration to the participant’s structural T1 scans. Functional images were spatially realigned and unwarped. Realignment is defined by default by six dimensions, three translations and three rotations along the x, y and z axes. Furthermore, each anatomical image was segmented for grey and white matter, and cerebrospinal fluid (CSF) and normalized to the T1-weighted MNI152 template. FMRI data were also aligned to the MNI template. Data was resliced after normalization using the default Tissue Probability Maps (structural target resolution= 2 mm, functional target resolution= 2 mm) and subsequently smoothed with a Gaussian kernel of 8-mm at full-width half-maximum (FWHM). Once preprocessed, we checked the histogram plots of rsFC values (voxel-to-voxel correlation coefficient between BOLD times series and 512 subset of voxels) for alignment centered at zero for quality assurance. If the data are completely without noise, then the correlation of a random subset of connections should center around zero; i.e. no correlations. Skewness in the plots indicate noise related to physiological processes or subject motion; this skewness may artificially inflate connectivity strengths. Several steps were taken to mitigate these confounds. First, physiological noise was corrected by using aCompCor, an algorithm which performs a principal components analysis on fMRI signal from the white matter and cerebrospinal fluid (CSF), areas of non-neuronal origin [10]. The white matter and CSF masks were generated by tissue segmentation performed using SPM, and then eroded to minimize partial volume effects for each subject. These masks were visually inspected. In the denoising step, we regressed the following confounds: realignment of six dimensions with their first temporal derivatives (12 components), white matter (5 components) and CSF (5 components). A band-pass filter of 0.008–0.09 Hz was additionally applied to the data. To control for excessive motion, we used a criterion of framewise displacement (FD) greater than the absolute value of 0.5mm threshold [29,87]. The mean FD (mFD) was computed using the weighted sum across six dimensions of the mean absolute scan to scan differences [117]. Participants were excluded if they met any of the following motion criteria: mFD > ∣0.5mm∣ and at least 20 % of the suprathreshold FD value > ∣0.5mm∣ [87,110]. Both mFD and maximum FD (maxFD) were also computed for comparative purposes between healthy participants and CBP for each site (see Supplementary Table 1). The maxFD was calculated using the weighted sum across six dimensions of the maximum absolute scan to scan differences [117]. The Mann-Whitney U test compared FD values between healthy participants and CBP for each site at p< 0.05. In addition, we excluded participants based on quality assurance, if they still presented a heavily skewed histogram plot of correlation coefficient following denoising. The residual fMRI signal of the included participants was used for whole brain rsFC analysis.

Participants were removed from each dataset, as follows. Further, a summary is provided in Supplementary Table 2.

Dataset 1

We excluded 11 participants (three healthy participants and eight CBP) from the initial dataset after matching for age with the other three datasets (n=1 control), after preprocessing (n=3; 1 control and 2 CBP, scans could not be aligned), after denoising (n=1 CBP, to meet quality assurance standards), and after motion correction (n=6; 1 control, 3 CBP had a mFD value > ∣0.5 mm∣ and at least 20% of the suprathreshold FD value > ∣0.5 mm∣) and 2 CBP had at least 20% of the suprathreshold FD value > ∣0.5 mm∣. We included 57 participants in our analysis, which consisted of 26 CBP and 31 age-matched healthy participants.

Dataset 2

We excluded nine participants (four healthy participants and five CBP): two healthy participants from the initial cohort to age match with the other three datasets and two that had at least 20% of the suprathreshold FD value > ∣0.5 mm∣ . Five CBP were excluded after motion correction (n=5; one participant had a mFD value > ∣0.5 mm∣ and at least 20% of the suprathreshold FD value > ∣0.5 mm∣) and 4 CBP had at least 20% of the suprathreshold FD value > ∣0.5 mm∣ . In sum, we included 27 participants in our analysis: 15 CBP and 12 age-matched healthy participants.

Dataset 3

We excluded 15 participants (12 healthy participants and three CBP) to age-match with the other datasets (n=7 healthy participants), to exclude healthy participants with pain (n=4), to exclude one CBP after preprocessing since the skull was not properly removed (n=1), to exclude after motion correction (n=3; one healthy participant had at least 20% of the suprathreshold FD value > ∣0.5 mm∣, one CBP had a mFD value > ∣0.5 mm∣ and one CBP had a mFD value > ∣0.5 mm∣ and at least 20% of the suprathreshold FD value > ∣0.5 mm∣ . We included 48 participants in our analysis, 21 CBP and 27 age-matched healthy participants.

Dataset 4

We excluded ten participants (eight healthy participants and two CBP) to age-match with the other datasets (n=1 healthy participant), to exclude healthy participants with pain (n=6), and to exclude after motion correction (n=3; one healthy participant with at least 20% of the suprathreshold FD value > ∣0.5 mm∣, one CBP with a mFD > ∣0.5 mm∣ and one CBP with a mFD > ∣0.5mm∣ and at least 20% of the suprathreshold FD value > ∣0.5 mm∣. Finally, we included 24 participants in our analysis, 15 CBP and 9 age-matched healthy participants.

Whole brain functional connectivity

We performed seed-to-voxel rsFC of the right antHC to the rest of the brain using the CONN toolbox. The antHC mask was derived from our previous parcellation study, which delineated the region of the right antHC using a hippocampal mask based on the Harvard-Oxford subcortical atlas on FSL v4.0 [3]. This region overlaps with the result from our chronic pain meta-analysis. The center-of-gravity of the mask was (MNI=24, 13, −21) [3]. We conducted a first-level, fixed-effects analysis, which was a seed-to-voxel based correlation between the right antHC ROI timeseries and every other voxel in the brain. We then performed a second-level random effects analysis to evaluate differences between healthy participants and CBP while regressing each dataset (site) and sex as covariates of no-interest. Site was included as a covariate of no-interest to account for scanner/site specific noise. Sex was also included as a covariate as there are sex differences in resting state networks [22,45,54,60,123,140], and our samples were not sex-matched. Voxelwise correlation coefficients were z-scored using the Fisher r-to-z transformation. To test whether the functional connectivity data were normally distributed, we performed the Shapiro-Wilk’s test, with significance set at p< 0.05, in SPSS (v25, IBM corp, Armonk, NY). Group differences between CBP and healthy participants were assessed with parametric cluster-based statistics using family-wise error (FWE), and maps were thresholded using cluster-size at P_FWE< 0.05 (cluster-forming height threshold of p< 0.001).

Post-hoc correlations

We performed post-hoc correlations to determine whether the aberrant antHC functional connectivity (CBP vs. healthy participants) was associated with pain characteristics (i.e., pain intensity and pain duration). Pain intensity was collected with the VAS scale on SF-MPQ on scan day in three datasets (1,3,4). Dataset 2 also collected part of the SF-MPQ, but VAS scores were not collected, as such this dataset was excluded for this analysis. In all datasets, pain duration was evaluated in number of years. We extracted the connection strength between the antHC seed and the resultant cluster (connection strength here is defined as the fisher-transformed correlation coefficient between BOLD activity in the antHC seed region and the resulting cluster) for each CBP subject, to assess whether they were correlated with clinical characteristics, including disease duration (n=69 CBP) and pain intensity (n=59 CBP) using Spearman’s rank correlation. Significance was set at p< 0.05.

RESULTS

Article selection

Our database and reference search identified a total of 49 articles that met our criteria of selection, as shown in Fig.1. Table 2 provides a summary of the 21 studies included in the first meta-analysis, which reported MTL activity in response to experimental pain, compared to a control condition. Table 3 provides a summary of the 28 studies of chronic pain patients reporting abnormal MTL activity compared to healthy participants that were included in our second meta-analysis.

Table 2.

Summary of experimental pain studies in healthy participants included in the meta-analysis

MTL Finding		N	Sex (W/M)	Age (mean ±SD/range/ S.E.M. in years)	Foci	Stimuli			Ref.	QS (/20)
Pain > Control	Pain < Control					Contrast	Modality	Body Part
(1) L HC	(2) L PHG	12	6/6	28.8 (21-41)	(1) 17 (2) 9	(1) heat pain > warm (2) warm > heat pain	thermal laser	R dorsum of the hand	[28]a	17
B HC		8	0/8	22-40	9	heat pain > baseline	thermal	L dorsum of the hand	[114]	17
B HC		14	1/13	25.8 (21-41)	12	thermal laser > baseline	thermal laser	B dorsum of the hand	[13]	17
B HC		12	6/6	25.42 ± 5.43 (20-35)	16	heat pain > baseline	thermal	L dorsum of the hand	[39]	19
(1) B HC (2) L HC		22	(1) 9/10 (2) 9/8	(1) 26 ± 1/34 ± 4 (2) 26 ± 1/31± 4	(1) 8 (2) 6	(1) muscle pain (saline injections) > baseline (2) cutaneous pain (saline injections) > baseline	chemical	(1) R muscle lower leg (2) R cutaneous lower leg	[52]	17
B HC		11	0/11	28 ± 4	24	pressure pain after saline injections > baseline	mechanical and chemical	R lower leg	[81]	18
L HC, B PHG		9	5/5*	29.7 ± 8.7	8	pricking pain > burning pain	thermal laser	L dorsum of the foot	[137]	17
	B PHG /HC	61	33/ 28	26.6 ± 4.7	18	low thermal pain < baseline	thermal	R volar forearm	[70]	17
R HC		7	3/4	18-36	9	heat pain > warm	thermal	L dorsum of the hand	[91]	17
B PHG		11	11/3**	26.9 ± 4.7	5	transcutaneous stimulation > baseline	electrical	R ankle	[113]	17
B HC		12	7/5	20–31	10	heat pain expectancy > baseline	thermal	L or R wrist	[146]	17
R PHG		31	16/ 15	30 (22–38)	25	painful esophageal stimulation > baseline	mechanical	esophagus	[23]	17
R PHG		34	18/ 16	23.4 ± 2.5	36	heat pain > warmth baseline	thermal	L lower leg	[71]	19
R PHG		14	8/6	26.1 (21-37)	13	heat pain > baseline	thermal	L volar forearm	[64]	19
(1) L PHG	(1) L PHG (2) B PHG	32	(1)16 (2)16	(1) 30.9 ± 7.8 (2) 27.8 ± 7.1	(1)57 (2)70	painful esophageal stimulation > baseline	mechanical	esophagus	[66]	18
R HC		12	6/6	29.9 ± 2.5	22	heat pain > baseline cerebral blood flow	thermal	L dorsum of the hand	[85]b	18
B HC		5	5/0	35 (26-56)	38	painful electrical shock > baseline	electrical	R volar forearm	[132]	17
B PHG		15	7/8	25.7 (22-30)	12	pain > baseline	chemical	R lower back muscle (4th lumbar vertebra)	[145]	17
(1) L HC (2) R HC		24	24/0	31.3 (24-45)	44	(1) laser heat > baseline (2) noxious cold > baseline	thermal	L dorsum of hand R foot	[15]	18
R HC		24	10/ 14	25 ± 5	13	heat pain nocebo***< high heat pain	thermal	L volar forearm	[61]	17
B HC/ PHG		23	12/ 11	27.8 ± 4.2	15	electrical pain > baseline	electrical	L dorsum of the hand / forehead	[124]	18

A detailed description of all 21 experimental pain fMRI studies (with the exception of one ASL and the other PET) of healthy participants included in the ALE meta-analysis.

Abbreviations: ALE activation likelihood estimation, ASL arterial spin-labeled magnetic resonance imaging, B bilateral, HC hippocampus, L left, M men, PET positron emission tomography, PHG parahipppocampal gyrus, QS quality score, R right, W women.

^*One participant (gender unspecified) was excluded from the study.

^**Three participants (gender unspecified) were excluded from the study.

^***Low heat pain nocebo was control to high heat pain

^aPET study

^bASL study

Table 3.

Summary of chronic pain studies included in the meta-analysis

Pain condition	Sex (W/M)		Age (mean ± SD/range/ S.E.M. in years)		Foci	Stimuli			MTL Finding		Ref.	QS (/20)
	P	C	P	C		Contrast	Modality	Body part	P > C	P < C
CD	9/16	6/19	31.72 ± 8.05	29.24 ± 6.85	2	resting state conditions			L HC/ PHG		[8]	19
FD	31/18	25/14	22.55 ± 1.78	22.18 ± 0.85	4	resting state conditions				L PHG	[80]	19
	20/11	19/11	22.50 ± 1.46	22.23 ± 0.94	13	resting state conditions			L HC/ PHG		[79]	18
FM	6/0	8/0	53.0 ± 4.8	51.4 ± 3.9	14	FDOPA uptake < control	chemical			B HC	[143]	18
	29/0	31/0	49.8 (25-64)	46.3 (20-63)	7	incongruent > congruent stimuli	stroop color word test			L HC	[90]	18
	37/6	10/5	46.3 ± 11.4	44.1 ± 14.8	34	(1) cuff stimulation > baseline (2) stimulus offset (still in pain)	mechanical	L lower leg	(1) R HC/PHG (2)B HC/R PHG		[126]	17
	15/0	15/0	52.07 ± 7.14 (42-64)	52.67 ± 6.23 (46-69)	1	controllability during expectation of pain > control condition	controllability in reaction time task	R HC		[48]	19
IBS	5/3	5/3	41.3 (27-64)	39.4 (24-54)	3	rectal distention > baseline	mechanical/auditory	rectum	R HC		[5]	18
	15/15	15/15	21.7 ± 3.0	21.4 ± 1.5	4	cognitive task < baseline	stroop task		R HC	[4]	17
	33/0	18/0	32.5 (20-60)/40.3 (21-60)	32.5 (21-54)	3	expectation of high distention > control	mechanical	rectum	R HC		[75]	18
	11/6	11/6	36.0 ± 10.8	37.4 ± 10.2	13	analgesia > control	suggestion -/conditioning enhanced placebo	R PHG		[76]	19
	76/42	31/29	29.39 ± 9.93/35.95 ±12.97	30.65 ± 10.71/37.28 ± 10.75	3	resting state conditions	R HC		[55]	17
	15/2	10/11	47.96 ± 3.7	34.76 ± 2.8	2	visual cue > control cue	visual task with rectal distention	rectum	B HC		[58]	19
	7/14	10/11	41.82 ± 11.92	35.91 ± 14.76	8	resting state conditions		R HC	[83]	18
ITN	10/7	10/9	63.41 ± 7.25	62.53 ± 7.41	3	resting state conditions		R PHG	[139]	19
LBP	5/6	6/5	20.4	21.5	27	task picture > no task picture	visual task picture		B PHG		[128]	19
	16/15	18/13	51.8 ± 9.9	49.3 ± 8.2	2	task picture > no task picture	visual task picture			R HC	[9]	17
Migraine	9/1	9/1	37.9 ± 4.7	37.8 ± 4.8	2	pain-related words > baseline (negative words)	verbal stimuli		R HC/PHG		[35]	19
	55/0	44/0	34.5 ± 11.0	34.3 ± 14.3	15	resting state conditions		B HC	[46]	19
	8/3	8/3	42.5 ± 11.9	42.3 ± 11.9	2	heat pain > baseline	thermal	painful facial side (patients)/matched (healthy participants)	R HC		[106]	18
	19/5	22/5	36.2 ± 11.3	33.7 ± 12.5	9	heat pain > baseline	thermal	L volar forearm	L HC		[127]	18
MOH	30/7	20/12	41.27 ± 9.27	41.34 ± 10.89	1	resting state conditions			R PHG	[20]	18
MPS	10/6	8/6	32.6 ± 7.2	30.6 ± 9.0	15	electrical pain > baseline	electrical	L shoulder		B HC, L PHG	[108]	16
OA	16/0	0/17	60.87	64.17	51	resting state conditions	L HC/PHG		[57]*	17
PD	34/0	34/0	21.5 ± 1.2	22.2 ± 1.7	5	resting state conditions	L HC		[63]	19
SPD	5/7	4/6	45.83 ± 14.95	35.2 ± 10.78	10	electrical pain > baseline	electrical	**lower leg		R PHG	[82]	17
	11/6	11/6	44.7 ± 9.1	45.4 ± 9.2	22	pricking > baseline	mechanical	L lower leg	B HC, L PHG		[131]	15
WAD	17/4	9/9	37.0 ± 11.0	35.0 ± 9.0	15	resting state conditions			B PHG		[78]	17

A detailed description of 28 chronic pain studies selected for the ALE meta-analysis.

Abbreviations: ALE activation likelihood estimation, ASL arterial spin-labeled magnetic resonance imaging, B bilateral, C healthy controls, CD Crohn’s disease, FD functional dyspepsia, FDOPA 6-[¹⁸F][121] fluoro-L-dopa, FM fibromyalgia, fMRI functional magnetic resonance imaging, HC hippocampus, IBS irritable bowel syndrome, ITN idiopathic trigeminal neuralgia, L left, LBP lower back pain, M men, MOH medication overuse headache, MPS myofascial pain syndrome, OA osteoarthritis, P patients, PD primary dysmenorrhea, PHG parahippocampal gyrus, SPD somatoform pain disorder, R right, QS quality score, W women, WAD whiplash associated disorder.

^*ASL study, the rest of the studies used fMRI

^**laterality not specified

Article quality assessment

Quality scores for each article included in each of the two meta-analyses is provided in Tables 2 and 3. Average scores for all articles included in the first meta-analysis were 17.52 ± 0.75 (mean ± SD score out of 20) and 17.93 ± 1.05 for the articles in the second meta-analysis. The most notable limitation of these articles is that most did not report whether sample size calculations were conducted a priori. In addition, some articles did not account for variables of potential interest, such as sex/gender.

Contrasts results

Our experimental pain and chronic pain meta-analyses resulted in the following number of experiments (contrasts within studies were accounted), foci, and participants as follows:

From the 21 studies investigating experimental pain, and reporting HC or PHG activation:

1. Healthy participants during experimental pain > control conditions (baseline): 22 experiments, 398 foci, 385 participants;

2. Healthy participants during experimental pain < control conditions (baseline): 3 experiments, 100 foci, 105 participants.

   From the 28 studies investigating chronic pain, and reporting HC or PHG activation:

3. Chronic pain patients > healthy participants: 17 experiments, 173 foci, 629 participants;

4. Chronic pain patients < healthy participants: 11 experiments, 81 foci, 583 participants.

ALE Meta-Analyses

Experimental pain studies

Our ALE meta-analysis of experimental pain studies in healthy participants is reported in Table 4, Fig.2 and Supplementary Fig.3. The contrast experimental pain > control conditions identified consistently greater activations in the right antHC, amygdala and parahippocampal gyrus at a cluster-corrected p< 0.05 at cluster-forming height threshold of p< 0.005 with 1000 permutations. The experimental pain < control conditions contrast did not have a sufficient number of experiments to perform the analysis.

Table 4.

ALE results of experimental pain and chronic pain studies

Brain region	MNI			ALE value	Cluster size (mm3)
	X	Y	Z
Healthy participants: Experimental pain > control conditions
R antHC/PHG/amygdala	30	−10	−24	0.025	808
Chronic pain patients < healthy participants
R antHC	26	−16	−16	0.021	840

ALE results of all significant contrasts. One significant activation cluster resulted from the ALE meta-analysis of healthy participants during experimental pain compared to control conditions (n=21, cluster-corrected p< 0.05, threshold p< 0.005 with 1000 permutations). One significant activation cluster in chronic pain patients compared to healthy participants resulted from the ALE meta-analysis of chronic pain studies (n=28, cluster-corrected p< 0.05, threshold p< 0.005 with 1000 permutations).

Abbreviations: ALE activation likelihood estimation, antHC anterior hippocampus, BA Brodmann area, MNI Montreal Neurological Institute, PHG parahippocampal gyrus, R right.

Fig.2. Peak activation in the anterior hippocampus in healthy participants during experimental pain conditions, compared to control conditions.

The ALE meta-analysis of noxious experimental conditions and baseline conditions in healthy participants with MTL activation (n=22 experiments). (A) Close-up of foci activation in the MTL, color-labelled according to stimuli. (B) The map shows one significant cluster (red) in the right anterior hippocampus (antHC), parahippocampal gyrus and amygdala (cluster-corrected p< 0.05, cluster-forming height threshold of p< 0.005 with 1000 permutations). (C) The map shows the delineation of the significant cluster in red and the different regions of the MTL: parahippocampal gyrus (yellow), amygdala (blue) and hippocampus (green). The black line represents MNI Y=−21, which represents the border between the anterior and posterior hippocampus [116].

Chronic pain studies

Our ALE meta-analysis of chronic pain studies versus healthy participants is shown in Table 4, Fig.3 and Supplementary Fig.4. The contrast chronic pain patients < healthy participants found consistently less activation in the right antHC at a cluster-corrected p< 0.05 at cluster-forming height threshold of p< 0.005 with 1000 permutations. There were no significant clusters for the chronic pain patients > healthy participants contrast.

Fig.3. The anterior hippocampus shows significantly less activity in chronic pain, compared to healthy participants.

The ALE meta-analysis compared chronic pain studies versus healthy participants with MTL activation (n=11 experiments). (A) Close-up of foci activation in the MTL, color-labelled according to stimuli. (B) The map represents one significant cluster in the right anterior hippocampus (antHC). Patients with chronic pain had consistently lower activity in this region compared to healthy participants (cluster-corrected p< 0.05 at cluster-forming height threshold of p< 0.005 with 1000 permutations). (C) The map shows the cluster in relation to other MTL structures: the hippocampus (green), the amygdala (blue) and the parahippocampal (yellow). The black line represents MNI Y=−21, which represents the border between the anterior and posterior hippocampus [116].

Group differences in antHC functional connectivity

Our whole brain seed-to-voxel functional connectivity analysis of the right antHC seed yielded reduced connectivity in CBP compared to healthy participants in the medial prefrontal cortex (mPFC; peak MNI coordinates: −10, 56, −02; cluster size= 328 mm³, cluster-corrected P_FWE< 0.05) including bilateral pregenual anterior cingulate cortex (pACC) and the left medial frontal pole (Fig.4). We found that the functional connectivity result was normally distributed (p>0.05). Data is shown in Supplementary Fig.5. Post-hoc analyses were performed to determine whether the aberrant connectivity was related to pain characteristics (pain intensity and pain duration). Reduced antHC-mPFC connectivity values in CBP were not significantly correlated with pain intensity (n=59 CBP, r= −0.129, p= 0.329) nor pain duration n=69 CBP, r= −0.088, p= 0.471). Post hoc analyses investigating sex-differences are reported in the Supplementary Materials.

Fig.4. CBP have weaker functional connectivity to the medial prefrontal cortex.

Whole brain seed-to-voxel resting state functional connectivity in four datasets (n=156) of the right anterior hippocampus (antHC) seed in CBP patients compared to healthy participants. Patients had weaker functional connectivity between the right antHC the medial prefrontal cortex, including the bilateral pregenual anterior cingulate cortex and the left medial frontal pole, compared to healthy participants (cluster-size P_FWE < 0.05, with a cluster-forming height threshold of p< 0.001). Functional connectivity strength (mean ± SD) is represented for both patients with chronic low back pain and healthy participants. Significant clusters are shown on a semi-inflated MNI brain in CONN. Abbreviations: antHC anterior hippocampus, CBP chronic low back pain patients, mFP medial frontal pole, pACC pregenual anterior cingulate cortex.

DISCUSSION

Neuroimaging evidence suggest that MTL structures are involved in acute pain and exhibit abnormal activity in chronic pain. To determine which regions of the MTL are involved, we performed two meta-analyses. The first meta-analysis investigated MTL activation in experimental pain studies of healthy participants and found consistent activation in the right antHC. The second meta-analysis investigated MTL activation in chronic pain patients compared to controls and found that patients have consistently less activation in the right antHC. Since functional connectivity of the antHC region has been previously shown to be a predictor for the transition from subacute to chronic back pain [107], we further conducted a seed-to-voxel rsFC analysis of the right antHC region in a large sample of CBP patients pooled from four different cohorts. Our analysis showed reduced antHC-mPFC functional connectivity compared to controls. These data suggest the right antHC is involved in healthy nociception but is dysfunctional in chronic pain.

MTL activity during experimental pain

The first key ALE finding is that experimental pain leads to right antHC activation. The HC, PHG and amygdala are key structures of the Papèz circuit, which is involved in memory and emotional processing. The HC/PHG is canonically responsible for learning and retention; the HC being a key structure for consolidation of contextual and spatial memory [116]. Notably, these MTL structures show pain-related activity in animal studies [27,34,68,84,97]. In contrast, several human studies have linked HC/PHG activity to a negative affect modulation of pain [11,119,121]. For example, one study found greater activity in the entorhinal cortex (the major interface between the HC and neocortex) during painful heat stimuli under elevated anxiety, compared to a low anxiety condition [114]. Activity in entorhinal cortex was correlated with the mid-insula in anxiety-induced hyperalgesia [114]. This finding is in line with the Gray-McNaughton theory of anxiety: the hippocampal formation amplifies anxiety-related signals in threatening situations [49].

Furthermore, the amygdala plays an integral role in the affective component of pain and is an opiate rich brain region [13,41,83,86,99,119,129]. The amygdala engages autonomic and emotional responses such as unpleasantness and fear [109], suggesting an intrinsic role in pain-related negative affect processing. In the context of stress or fear, the amygdala may induce hypoalgesia to minimize pain sensation during noxious stimuli [136]. There are well-known antHC-amygdala interactions in both encoding and retrieval of affective information [31,111] which can certainly extend to situations involving experimental pain.

To our knowledge, we present the first meta-analysis of experimental pain studies with MTL activity and report that MTL structures are consistently co-activated. Thus, our findings lend some new insights into potential memory and affective processing in this context.

MTL activity in chronic pain

Our second aim was to identify which MTL region show consistent activation in chronic pain since recent studies have highlighted the MTL as a potential site for understanding the onset and development of chronic pain [6,134]. Our main finding yielded consistently less activation in the right antHC in chronic pain patients, compared to controls. This antHC region overlaps with the area activated in experimental pain in healthy participants.

The HC is longitudinally divided into the antHC and posterior HC with distinct functional properties [3,116]. The antHC is implicated in mood related functions [69,115], and stress modulation [40], likely through interactions with the amygdala. Thus, abnormal antHC activity in chronic pain may be related to dysregulation in stress modulation. Chronic pain can be considered as a stressor, eliciting a prolonged stress response [26] – i.e., chronic pain poses an allostatic load on the brain [16,96,135]. The HC is particularly sensitive to the neurotoxic effects of prolonged exposure to stress hormones [24,92], thus affecting its structure [96] and function [53]. While working memory deficits have been reported in chronic pain [12,93], there has not been a systematic investigation of types of memory processes most sensitive to hippocampal dysfunction such as associative recollection. A few neuroimaging studies addressed the interruptive function of pain in relation to memory function [14,44], where reduced right antHC activity and poor visual encoding were observed in response to pain stimuli [44]. Interestingly, HC activity is related to psychologically induced analgesia [50]. As such, our finding could reflect HC dysregulation related to prolonged stress exposure, with chronic pain as the stressor. The HC subregions could certainly be further investigated in the context of memory and pain in future mechanistic studies.

antHC functional connectivity in CBP

Our third aim was to identify whether abnormal right antHC activity was accompanied by abnormal rsFC in CBP. We found reduced right antHC-mPFC connectivity in CBP compared to controls, in line with a previous study that showed disrupted antHC-mPFC connectivity predicts the transition from subacute to chronic back pain [107]. These results suggest this aberrant connectivity is also involved in sustaining maladaptive pain, but how it is implicated in chronic pain is not well understood.

The literature indicates antHC-mPFC interaction in memory encoding and retrieval, future decision-making and autobiographical memory [95,105]. Specifically, the mPFC integrates spatial and contextual information [30,62,142], attributing behavioral, cognitive and emotional relevance to a particular stimulus [142]. As such, the antHC-mPFC connectivity facilitates the encoding and retrieval of global schemas, contextual event information and emotional cues and enables future memory-guided behaviours [105,120]. Reduced connectivity between these regions is seen in conjunction with poor autobiographical memory in patients with MTL damage [94] and is associated with deficits in emotional decision-making as observed previously in CBP [7]. Of interest, the right antHC is consistently activated to a greater extent when simulating future events compared to recalling past memories [2,122]. The antHC-mPFC functional connectivity also mediates extinction learning [65,100]. Extinction learning is the process of forming new memories that decouple conditioned responses (e.g., fear) to a stimulus (e.g., tone). This results when the stimulus is repeatedly presented without the unconditioned stimulus (e.g., electric shock) which caused the conditioned response [112,118]. Chronic stressors, including chronic pain, impair extinction learning [1,42,59,101]. In particular, chronic stress blocks long-term potentiation between the HC and mPFC [47,89] and contributes to dendritic regression in mPFC and HC [77]. In CBP, there is impaired extinction learning and slower extinction to pain and verbal responses [43], and muscular reactivity [43,125] compared to controls. Perhaps reduced antHC-mPFC connectivity in CBP reflects a memory network deficit which fails to inhibit the memory of pain or to enable non-pain memory schemas that would allow retrieval of alternate memories.

Study limitations

Chronic pain conditions are heterogeneous and, accordingly, may have different patterns of brain activity. This is compounded by individual variability in chronic pain. Hence, the antHC network found in CBP is only relevant for the specific cohorts and not representative of chronic pain as a whole. For example, in burning mouth syndrome, patients had stronger antHC-mPFC connectivity than controls, but only in the presence of spontaneous pain [67]. The finding that successful recovery in patients with subacute back pain is accompanied by increased antHC-mPFC connectivity [107] suggests it is at least robust in back pain.

Furthermore, advances in MRI technology have led to variability in data quality and coordinates collected between laboratories through years of publications. To minimize this variability, we used the Turkeltaub algorithm, which accounts for such heterogeneity. In addition, foci derived from heterogeneous statistical thresholds were included in the meta-analyses—not accounting for these studies could lead to false negative findings, and thus bias the outcome of the result. Importantly, given our article selection and use of an MTL mask, our results do not speak to the consistency of MTL activation in pain in general, nor to co-activated regions outside the MTL.

In order to achieve sufficient statistical power, it has been reported that an ALE meta-analysis requires 8–15 experiments per contrast [129,138]. Based on this heuristic, our contrast of healthy participants during experimental pain < control conditions was underpowered, and this negative result should be interpreted with caution.

We hope the current findings stimulate future studies that could provide a mechanistic account of antHC in pain. The seeming contradiction of greater antHC activation in acute pain and reduced activity associated with chronic pain portrays a complex role and raises questions as to whether activation magnitude is a cause or consequence of pain experience. Experimental fMRI studies of pain and analgesia would be useful here. Furthermore, investigating functional connectivity during acute pain would be helpful to clarify how correlated activity in the target regions vary with the stimulus and task. Finally, investigating the relationship of these pain-related effects to other known functions of the antHC such as autobiographical memory retrieval or context modulation of anxiety are clearly warranted. Such studies are necessary to provide ‘meat on the bones’ for the relationship of antHC and pain that we have identified.

CONCLUSION

In conclusion, we performed the first meta-analyses of the MTL in pain and determined that in studies reporting MTL abnormalities in chronic pain, the most common subregion was the antHC. The antHC has abnormal rsFC to the mPFC in CBP, reflecting cognitive and affective abnormalities. These data shed novel and important mechanistic information on the role of the HC in chronic pain.