Simultaneous Brain, Brainstem and Spinal Cord

1) Anaesthesia, Pain & Critical Care Sciences, School of Physiology, Pharmacology & 6 Neuroscience, Biomedical Sciences Building, University of Bristol, Bristol, BS8 1TD, UK. 7 2) Clinical Research and Imaging Centre, School of Psychological Science, University of 8 Bristol, Bristol, BS8 1TU, UK 9 3) Medical Physics, University Hospitals Bristol & Weston NHS Trust, Bristol, BS2 8HW, 10 UK. 11 4) Department of Neuroimaging, Institute of Psychiatry, Psychology & Neuroscience, 12 King's College London, London, SE5 8AF, UK. 13 5) Wellcome Wolfson Brain Imaging Centre, School of Psychology, Lawrence Stenhouse 14 Building, University of East Anglia, Norwich, NR4 7TJ, UK. 15 6) Department of Anesthesiology, University of California San Diego, 9500 Gilman Dr, La 16 Jolla, CA 92093, USA. 17


Summary 21
Pain perception is decreased by shifting attentional focus away from a threatening event. 22 This attentional analgesia engages parallel descending control pathways from anterior 23 cingulate (ACC) to locus coeruleus, and ACC to periaqueductal grey (PAG) -rostral 24 ventromedial medulla (RVM), indicating possible roles for noradrenergic or opioidergic 25 neuromodulators. To determine which pathway modulates nociceptive activity in humans 26 we used simultaneous whole brain-spinal cord pharmacological-fMRI (N=39) across three 27 sessions. Noxious thermal forearm stimulation generated somatotopic-activation of dorsal 28 horn (DH) whose activity correlated with pain report and mirrored attentional pain 29 modulation. Activity in an adjacent cluster reported the interaction between task and 30 noxious stimulus. Effective connectivity analysis revealed that ACC interacts with PAG and 31 RVM to modulate spinal cord activity. Blocking endogenous opioids with Naltrexone impairs 32 Introduction 38 Pain is a fundamental and evolutionarily conserved cognitive construct that is behaviourally 39 prioritised by organisms to protect themselves from harm and facilitate survival. As such 40 pain perception is sensitive to the context within which potential harm occurs. There are 41 well recognised top-down influences on pain that can either suppress (e.g. placebo (Wager 42 and Atlas, 2015) or task engagement (Bussing et al., 2010)) or amplify (e.g. catastrophising 43 (Gracely et al., 2004), hypervigilance (Crombez et al., 2004) or nocebo (Benedetti and 44 Piedimonte, 2019)) its expression. These processes influence both acute and chronic pain 45 and provide a dynamic, moment by moment regulation of pain as an organism moves 46 through their environment. 47 A simple shift in attention away from a noxious stimulus can cause a decrease in pain 48 perception -a phenomenon known as attentional analgesia. This effect can be considered 49 to be a mechanism to enable focus, allowing prioritisation of task performance over pain 50 interruption (Eccleston and Crombez, 1999;Erpelding and Davis, 2013). This phenomenon is 51 reliably demonstrable in a laboratory setting (Miron et al., 1989) and a network of cortical 52 and brainstem structures have been implicated in attentional analgesia (Bantick et  We have shown that two parallel pathways are implicated in driving brainstem activity 56 related to attentional analgesia (Brooks et al., 2017;Oliva et al., 2021b). Projections from 57 rostral anterior cingulate cortex (ACC) were found to drive the periaqueductal grey (PAG) 58 and rostral ventromedial medulla (RVM), which animal studies have shown to work in 59 concert using opioidergic mechanisms to regulate spinal nociception (Fields, 2004;Fields 60 and Basbaum, 1978;Heinricher et al., 1994;Ossipov et al., 2010). Similarly, a bidirectional 61 connection between ACC and locus coeruleus (LC) was also directly involved in attentional 62 analgesia. As the primary source of cortical noradrenaline, the LC is thought to signal 63 salience of incoming sensory information (Aston-Jones and Cohen, 2005; Sara and Bouret, 64 12 nociceptive input was extended through human spinal fMRI by Sprenger and colleagues 305 (2012), who in a second psychophysical experiment with naloxone provided evidence that 306 the modulation of pain percept may involve opioids. We show that naltrexone attenuates 307 spinal responses to attentional analgesia, which underly the behavioural differences 308 between the high|hard and easy|hard conditions. 309 Uniquely, our 2x2 factorial study design enabled the identification of neural activity reading 310 out the interaction between task and temperature which strikingly was only seen at a spinal 311 level in a cluster located deep and medial to the Spinal noci cluster. The activity in this 312 Spinal int cluster was highest in the high|hard condition (ie when the attentional analgesic 313 effect is seen) and this activation was no longer significant in the presence of naltrexone. 314 This may be consistent with the presence of a local interneuron population in the deeper 315 dorsal horn that could influence the onward transmission of nociceptive information 316 (Hughes and Todd, 2020; Koch et al., 2018). Such a circuit organisation is predicted by many 317 animal models of pain regulation with the involvement of inhibitory interneurons that shape 318 the incoming signals from the original gate theory of Melzack and Wall (Melzack and Wall,319 1965) through to descending control (Millan, 2002 A key objective of the study was to determine how the information regarding the 341 attentional task demand could be conveyed to the spinal cord. Analysis of regional BOLD 342 signal showed activity in both the main effect of task and of temperature in all three of the 343 key brainstem sites PAG, RVM and LC with no interaction between task and temperature in 344 the brainstem providing little indication as to which area might be mediating any analgesic 345 effect (in line with previous (Oliva et al., 2021b)). However, an interaction effect was 346 observed on the effective connectivity between RVM and dorsal horn, with coupling highest 347 in the high|hard conditions. The importance of this descending connection to the 348 attentional analgesic effect is emphasised by the effect of naltrexone which blocked both 349 the modulation of RVM-spinal cord connectivity and attentional analgesia (a behavioural 350 finding previously noted by Sprenger et al (2012)). This fits with the classic model of 351 descending pain modulation that has been developed through decades of animal research 352 (Fields, 2004;Ossipov et al., 2010) that is engaged in situations of fight or flight and also 353 during appetitive behaviours like feeding and reproduction. Here we identify that the 354 opioidergic system is also engaged moment by moment, in specific contexts, during a 355 relatively simple cognitive tasks and uncover one of its loci of action in humans. 356 Analysis of effective connectivity also showed evidence for modulation of pathways from 357 ACC to PAG and PAG to RVM by task and the interaction between task and temperature, 358 respectively (in agreement with (Oliva et al., 2021b)). The communication between ACC and 359 PAG was also disrupted by the opioid antagonist naltrexone. This is similar to the previous 360 finding from studies of placebo analgesia where naloxone was shown to disrupt ACC-PAG 361 communication which was also linked to the mediation of its analgesic effects (Eippert et al., nucleus. The noradrenergic manipulation with reboxetine did show a significant analgesic 392 effect which was independent of task difficulty. This indicates that this dose of reboxetine is 393 capable of altering baseline gain in the nociceptive system, but has no selective effect on 394 attentional pain modulation. We performed a post hoc Bayesian paired t-test analysis 395 contrasting reboxetine with placebo which showed moderate level of confidence in this null 396 effect on attentional analgesia (Bayes Factor 6.8). Reboxetine also modulated a task-397 dependent connection between ACC and PAG, though this did not appear to influence task 398 performance and so its behavioural significance is uncertain. In interpreting these findings 399 one potential explanation is that noradrenaline is not involved in attentional analgesia, 400 however it could also be because of a ceiling effect where the reuptake inhibitor cannot 401 increase the noradrenaline level any further during the attentional task. In this sense a 402 noradrenergic antagonist experiment, similar to that used to examine the role of the 403 opioids, would be ideal. However, selective alpha2-antagonists are not used clinically and 404 even experimental agents like Yohimbine have a number of issues that would have 405 confounded this study in that they cause anxiety, excitation and hypertension. Therefore, 406 we conclude that were not able to provide any additional causal evidence to support a role 407 of the LC in attentional analgesia, but this likely reflects a limitation of our approach and 408 lack of good pharmacological tools to resolve the influence of this challenging target. Participants 432 433 Healthy volunteers were recruited through email and poster advertisement in the University 434 of Bristol and were screened via self-report for their eligibility to participate. Exclusion 435 criteria included any psychiatric disorder (including anxiety/depression), diagnosed chronic 436 pain condition (e.g. fibromyalgia), left handedness, recent use of psychoactive compounds 437 (e.g. recreational drugs or antidepressants) and standard MRI-safety exclusion criteria. 438 The study was approved by the University of Bristol Faculty of Science Human Research 439 Ethics Committee (reference 23111759828). An initial power analysis was done to 440 determine the sample size using the fmripower software (Mumford and Nichols, 2008). 441 Using data from our previous study of attentional analgesia ( (Brooks et al., 2017), main 442 effect of task contrast in the periaqueductal grey matter mask) we designed the study to 443 have an 80% power to detect an effect size of 0.425 (one sample t-test) in the PAG with an 444 alpha of 0.05 requiring a cohort of 40 subjects. Of fifty-seven subjects screened, two were 445 excluded for claustrophobia, three were excluded for regular or recent drug use (including 446 recreational), and five were excluded due to intolerance of the thermal stimulus. This was 447 defined as high pain score (≥ 8/10) for a temperature that should be non-nociceptive (<43 448 °C). In addition, six participants withdrew from the study as they were unable to attend for 449 the full three visits. One participant had an adverse reaction (nausea) to a study drug 450 (naltrexone) and dropped out of the study. One subject was excluded for being unable to 451 perform the task correctly.  (Lautenbacher et al., 1995). Participants received a range of 467 temperatures between 36 and 45°C, and were asked to rate the sensation they felt for each 468 stimulus, on a scale from 0 (no pain) to 10 (the worst pain imaginable). The stimulus 469 provoking a pain rating of 6 out of 10 at least 3 times in a row, was used for the "high" 470 temperature stimulation in the experiment. If the participant only gave pain scores lower 471 than 6 to all stimuli, then the maximum programmable plateau temperature of 45°C was 472 used, but with higher temperature spikes of 3, 4 and 5 degrees above, reaching the highest 473 temperature allowed for safety (50°C maximum). 474 The session also included a calibration of the rapid serial visual presentation (RSVP) task 475 (Potter and Levy, 1969), where participants were asked to spot the number 5 among 476 distractor characters. The task was repeated 16 times at different velocities (i.e. different 477 inter-character intervals) in pseudorandom order, ranging from 32 to 256ms. To identify the 478 optimal speed for the hard version of the RSVP task (defined as 70% of each subject's 479 maximum d' score), the d' scores for the different velocities were plotted and the curve fit 480 to a sigmoidal function, using a non-linear least squares fitting routine in Excel (Solver). 481 Once parameterised, the target speed for 70% performance was recorded for subsequent 482 use during the imaging session. 483

Imaging sessions 484
Following the screening/calibration session, participants returned for three imaging 485 sessions, spaced at least a week apart. Participants underwent drug screening 486 (questionnaire) and pregnancy testing. After eating a light snack, they were given either an 487 inert placebo capsule, naltrexone (50mg) or reboxetine (4mg) according to a randomised 488 18 schedule. The dose of the opioid antagonist Naltrexone (50mg) was as per the British 489 National Formulary (BNF) where it is licensed to prevent relapse in opioid or alcohol 490 dependency. Naltrexone is well absorbed with high oral bioavailability and its levels in the 491 serum peak after 1 hour with a half-life of between 8 and 12 hours (Verebey et al., 1976). 492 Reboxetine is used for the treatment of depression, and we used the lowest dose 493 recommended by the BNF (4mg). It has high oral bioavailability (~95%), serum levels peak noradrenergic reuptake inhibitor on emotional processing but no effect on performance of a 503 rapid serial visual presentation task. 504 All tablets were encased in identical gelatine capsules and dispensed in numbered bottles 505 prepared by the hospital pharmacy (Bristol Royal Infirmary, University Hospitals Bristol and 506 Weston NHS Foundation Trust). Neither the participant nor the investigator knew the 507 identity of the drug which was allocated by a computer-generated randomised schedule. No 508 subject reported being aware of whether they had received active drug or placebo (but the 509 effectiveness of masking was not formally assessed post hoc after dosing). 510 One hour after drug dosing, calibration of the RSVP task was repeated (to control for any 511 effect on performance). Before scanning, participants received the high thermal stimulus at 512 their pre-determined temperature, which they rated verbally. If the rating was 6±1, the 513 temperature was kept the same, otherwise it was adjusted accordingly (up or down). 514 Neither reboxetine nor naltrexone caused a significant change in pain perception or task 515 velocity during the calibration, as verified with paired t tests (placebo versus reboxetine and 516 placebo versus naltrexone, see Figure 2 Supplementary Figure 3). On average, the plateau 517 temperature used for high temperature stimuli was 43.8± 1.25°C. The median inter-stimulus 518 interval for the hard RSVP task was 48ms, range . 519 In the MRI scanner, participants performed the RSVP task at either difficulty level (easy or 520 hard) whilst innocuous (low) or noxious (high) thermal stimuli were delivered concurrently 521 to their left forearm. The four experimental conditions (easy|high, hard|high, easy|low, 522 hard|low), were repeated four times each, in a pseudo-random order. The hard version 523 (70% d' performance) of the task and the high (noxious) thermal stimulus were calibrated as 524 described above. In the easy version of the task the inter-character presentation speed was 525 always set at 192ms, except when a participant's hard task velocity of was equal or slower 526 than 96ms, whereby the easy task was set to 256ms. The low (innocuous) thermal stimulus 527 was always set to be a plateau of 36°C with spikes of 2, 3 and 4˚C above this baseline. 528 Participants performed the task (identifying targets) and provided pain ratings 10 seconds 529 after the end of each experimental block on a visual analogue scale (0-100), using a button 530 box (Lumina) held in their dominant (right) hand. 531

Acquisition of functional images 532
Functional images were obtained with a 3T Siemens Skyra MRI scanner, and 64 channel 533 receive-only head and neck coil. After acquisition of localiser images, a sagittal volumetric 534 T1-weighted structural image of brain, brainstem and spinal cord was acquired using the 535 MPRAGE pulse sequence, (TR = 2000ms, TE = 3.72ms, TI = 1000 ms, flip angle 9°, field of 536 view (FoV) 320 mm, GRAPPA acceleration factor = 2) and 1.0mm isotropic resolution. Blood 537 oxygenation level dependent (BOLD) functional data was acquired axially from the top of 538 the brain to the intervertebral disc between the C6 and C7 vertebral bodies, with TR = 539 3000ms, TE = 39ms, GRAPPA acceleration factor = 2, flip angle 90°, FoV 170 mm, phase 540 encoding direction P>>A, matrix size 96 by 96. 541 Slices were positioned perpendicular to the long axis of the cord for the C5-C6 spinal 542 segments, whilst still maintaining whole brain coverage, and had an in-plane resolution of 543 1.77 x 1.77 mm and slice thickness of 4mm and a 40% gap between slices (increased to 45-544 50% in taller participants). To determine the optimal shim offset for each slice, calibration 545 scans were acquired cycling through 15 shim offsets. For the caudal 20 slices covering from 546 20 spinal cord to medulla, manual inspection of images determined the optimal shim offset to 547 be used for each subject (Finsterbusch et al., 2012). The remaining supraspinal slices were 548 acquired with the first and higher order shim offsets determined using the scanner's 549 automated routine. The ability of z-shimmed whole CNS imaging to adequately capture 550 BOLD signal was assessed through pilot data examining the temporal signal to noise ratio 551 (tSNR) across cord and brain, see Figure 1 Supplementary Figure 1. 552 During scanning, cardiac and respiratory processes were recorded using a finger pulse 553 oximeter (Nonin 7500) and pneumatic respiratory bellows (Lafayette), respectively. These 554 physiological signals and scanner triggers were recorded using an MP150 data acquisition 555 unit (BIOPAC, Goleta, CA), and converted to text files for subsequent use during signal 556 modelling. 557

Analysis of pain scores 559
Pain scores recorded during the experiment were investigated collectively for the three 560 visits using a three-way ANOVA in Prism version 8 for Windows (GraphPad Software, La 561 Jolla, California). Any significant interaction was further investigated with two separate 562 three-way ANOVAs (placebo versus naltrexone and placebo versus reboxetine). Finally, each 563 drug condition was analysed individually with three separate two-way ANOVAs. Two-tailed 564 post-hoc tests were used to further investigate any interactions. 565

Pre-processing of functional data and single-subject analysis 566
Functional data were divided into spinal cord and brain/brainstem, by splitting at the top of 567 the odontoid process (dens) of the 2 nd cervical vertebra. The resulting two sets of images 568 underwent separate, optimised, pre-processing pipelines. 569 Spinal cord data was motion corrected with AFNI 2dImReg (Cox, 1996) coverage. Subsequently these regressors (which are 4D images) were split at the level of the 595 odontoid process, to be used separately for brain and spinal cord physiological noise 596 correction. For the brain data the PNM consisted of 32 regressors, with the addition of a CSF 597 regressor for the spinal cord, giving a total of 33 regressors for this region. 598 All functional images were analysed using a general linear model (GLM) in FEAT with high-599 pass temporal filtering (cut-off 90s) and pre-whitening using FILM (Woolrich et al., 2001).

Group analysis 619
We used a conservative approach to investigate differences in CNS activity in main effects 620 and interactions due to administration of reboxetine or naltrexone. All first-level analyses, 621 single group averages and pooled analyses were performed with the experimenter masked 622 to the study visit (i.e. drug session). An initial analysis examined the brain, brainstem, and 623 spinal cord activation in the planned contrasts (main effects of temperature, task, and their 624 interaction) across all visits i.e. a "pooled" analysis. Individual subjects' data were averaged 625 using a within-subject "group" model (treating variance between sessions as a random 626 effect), and resultant outputs averaged (across subjects) using a mixed effects model. This 627 allowed the generation of functional masks, to use for investigation of differences between 628 drug conditions. 629 Generalised psychophysiological interaction (gPPI) analysis (McLaren et al., 2012) was used 630 to assess effective connectivity changes between brain, brainstem, and spinal cord during 631 the attentional analgesia experiment. The list of regions to be investigated were specified a 23 priori on the basis of our previous study (Oliva et al., 2021b), and included the ACC, PAG, LC 633 and RVM -to which was added the left side of the spinal cord at the C5/C6 vertebral level. 634 Following partial unblinding to drug, an initial analysis was performed for the placebo visit. 635 This analysis strategy, which examined connectivity between CNS regions identified in the 636 pooled data and previously (Brooks et al., 2017;Oliva et al., 2021b), was initially limited to 637 examination of the placebo data and largely replicated our earlier findings (Oliva et al., 638 2021b). By identifying those connections that are normally active during attentional 639 analgesia, we could then test whether they are subject to specific neurotransmitter 640 modulation. This involved partial-unmasking to the remaining two conditions (information 641 on the specific drug used was withheld), so paired t-tests could be performed between the 642 connections of interest. Finally, after the analysis was completed the full unmasking was 643 allowed for the purpose of interpretation of paired differences between conditions. 644

Pooled analysis -spinal cord 645
For each subject, parameter maps estimated for each contrast and each visit (i.e. drug 646 session), were registered to the PAM50 template with SCT. Each contrast was then averaged 647 across visits using a within-subject ordinary least squares (OLS) model using FLAME (part of 648 FSL) from command line. The resulting average contrasts (registered to the PAM50 649 template) were each concatenated across subjects (i.e. each contrast had 39 samples). 650 These were then investigated with a one-sample t-test in RANDOMISE, using a left C5-6 651 vertebral mask, based on the probabilistic atlas from the SCT. The choice to use a relatively 652 large vertebral level mask, rather than a more focussed grey matter mask, was based on 653 consideration of (1) the voxel size of our fMRI data compared to the high-resolution data 654 (0.5mm) used to define probabilistic grey matter masks in SCT, and (2) to allow for inter-655 subject differences in segmental representation of the stimulation site on the left forearm. 656 It should be noted that by using larger masks we effectively decreased our sensitivity to 657 detect activation, due to the more punitive multiple comparison correction. Results are 658 reported with threshold free cluster enhancement (TFCE) P < 0.05 corrected for multiple 659 comparisons. Significant regions of activation from this pooled analysis were used to 660 generate masks for subsequent comparison between conditions, using paired t-tests. 661

Pooled analysis -brainstem 662
Similar to the spinal cord, for each subject, parameter maps from the brainstem for each 663 planned contrast and visit were averaged with an OLS model in FEAT software. The resulting 664 average was the input to a between-subjects, mixed effects, one-sample t-test in FEAT. 665 Subsequently, group activations for each of the six contrasts were investigated with 666 permutation testing in RANDOMISE, using a probabilistic mask of the brainstem taken from 667 the Harvard-Oxford subcortical atlas (threshold set to P=0.5). Results are reported with TFCE 668 correction and P < 0.05. Significant regions of activity were binarized and used as a 669 functional mask for the between conditions comparison. 670 Pooled analysis -brain 671 Brain data was averaged and analysed with the same FEAT analyses that were applied in the 672 brainstem. Following within subject averaging, group activity was assessed with a mixed 673 effects two-tailed one sample t-test at the whole-brain level, with results reported for 674 cluster forming threshold of Z > 3.1, and corrected cluster significance of P < 0.05. This 675 produced maps of activity (one per planned contrast) that were then binarized to produce 676 masks that were used in follow up paired t-tests. 677 Within subject comparison -paired tests 678 Paired t tests were performed to resolve potential changes in activity in reboxetine versus 679 placebo and naltrexone versus placebo, separately. Design and contrast files for input in 680 RANDOMISE were built in FEAT. A group file with appropriately defined exchangeability 681 blocks was additionally defined. Permutation testing in RANDOMISE was used to assess 682 group level differences between placebo and the two drugs, separately for brain, brainstem, 683 and spinal cord. The investigation was restricted to the functional masks derived from the 684 main effect analysis for each contrast. 685 Effective connectivity analysis (gPPI) 686 For connectivity analysis, functional data for brain, brainstem and spinal cord were pre-687 processed as previously described (Oliva et al., 2021b). To restrict analysis to connections 688 typically observed during attentional analgesia, we initially estimated the connection 689 pattern for the placebo session, then within this network tested for differences in the other 690 drug conditions. To achieve this goal, placebo data were first analysed for main 691 effects/interaction with the simple (non-gPPI) analysis to define the pattern of BOLD 692 activity. Subsequently, time series extraction was restricted to anatomical regions/contrasts 693 identified previously (Oliva et al., 2021b) to different seed regions, models used for estimating connectivity for brain and spinal cord 712 seeds were otherwise identical. Estimates of effective connectivity for the group were 713 obtained with permutation testing with RANDOMISE, using as targets the same ROI masks 714 used for time-series extraction. For example, a gPPI analysis with an RVM seed timeseries 715 (taken from the region responding during the main effect of temperature), examined 716 connectivity to brain/brainstem and spinal cord with PAG, LC, ACC, and left C5-6 vertebral 717 masks. To test whether drug administration altered connectivity during attentional 718 analgesia, the significant connections detected in the placebo session were examined for 719 differences in the other drug conditions i.e. the same masks were used for time-series 720 extraction for gPPI analysis of the naltrexone/reboxetine conditions. At the group level, two-721 tailed paired t-tests were used to detect differences with RANDOMSISE (TFCE, P<0.05) 722 26 between placebo and naltrexone, and between placebo and reboxetine visits, as described 723