Somatotopic arrangement of thermal sensory regions in the healthy human spinal cord determined by means of spinal cord functional MRI

Authors

  • Patrick W. Stroman,

    Corresponding author
    1. Centre for Neuroscience Studies, Queen's University, Kingston, Ontario, Canada
    2. Department of Diagnostic Radiology, Queen's University, Kingston, Ontario, Canada
    3. Department of Physics, Queen's University, Kingston, Ontario, Canada
    • Centre for Neuroscience Studies, Queen's University, Kingston, Ontario, Canada K7L 3N6
    Search for more papers by this author
  • Rachael L. Bosma,

    1. Centre for Neuroscience Studies, Queen's University, Kingston, Ontario, Canada
    Search for more papers by this author
  • Anastasia Tsyben

    1. Centre for Neuroscience Studies, Queen's University, Kingston, Ontario, Canada
    Search for more papers by this author

Abstract

Previous functional MRI studies of normal sensory function in the human spinal cord, including right-to-left symmetry of activity, have been influenced by order effects between repeated studies. In this study, we apply thermal sensory stimulation to four dermatomes within each functional MRI time-series acquisition. Each of the four dermatomes receives a unique stimulation paradigm, such that the four paradigms form a linearly independent set, enabling detection of each individual stimulus response. Functional MRI data are shown spanning the cervical spinal cord and brainstem in 10 healthy volunteers. Results of general linear model analysis demonstrate consistent patterns of activity within the spinal cord segments corresponding to each dermatome, and a high degree of symmetry between right-side and left-side stimulation. Connectivity analyses also demonstrate consistent areas of activity and connectivity between spinal cord and brainstem regions corresponding to known anatomy. However, right-side and left-side responses are not at precisely the same rostral–caudal positions, but are offset by several millimeters, with left-side responses consistently more caudal than right-side responses. The results confirm that distinct responses to multiple interleaved sensory stimuli can be distinguished, enabling studies of sensory responses within the spinal cord without the confounding effects of comparing sequential studies. Magn Reson Med, 2012. © 2011 Wiley Periodicals, Inc.

INTRODUCTION

An in depth understanding of normal sensory processing in the healthy human spinal cord is essential for providing normative standards for both clinical assessments of patients and for the evaluation of the efficacy of novel treatment strategies. Functional magnetic resonance imaging of the spinal cord (spinal fMRI) has demonstrated potential for mapping this function noninvasively, by means of thermal sensory stimulation of various dermatomes (one dermatome at a time) during the fMRI time-series acquisition (1–3). However, the sensations that are perceived in healthy volunteers are the net result of an integrated response from a network of regions within the central nervous system (CNS) (1, 4), and may not be fully described based on observed responses to stimulation of a single sensory dermatome. Similarly, the effects of trauma or disease can have distributed effects across sensory and motor networks, which may not be revealed by the responses to a single point of thermal sensory stimulation. Key tests of the reliability and sensitivity of fMRI results, based on verifying the correspondence between fMRI results and neuroanatomy, such as the expected right-to-left symmetry (5–7), also require stimulation of multiple sensory dermatomes. However, comparisons of results from repeated fMRI acquisitions within a single imaging session with one participant are influenced by significant order effects due to systematic changes in alertness, anticipation, and attention (1, 4), which can interfere with comparisons. It is therefore necessary to carry out fMRI studies with stimulation of multiple sensory dermatomes within the same fMRI acquisition to map the neural responses at corresponding spinal cord levels to determine the nuances and interactions between stimulation sites and cord responses while avoiding order effects.

In this study, we use a custom-made device to stimulate four dermatomes, each with a distinct thermal sensory stimulation paradigm, within a single fMRI acquisition. With this approach, we present the first detailed study of right–left symmetry of regions of the human spinal cord involved with thermal sensations.

MATERIALS AND METHODS

MRI Acquisition

Ten healthy volunteers with no prior history of neurological injury or disease (age 23 ± 6 years, range 19–37 years, six females and four males) were studied with a 3 T MRI system (Magnetom Trio, Siemens, Erlangen, Germany). Previous studies, and a power calculation, show that this number of subjects should be adequate to reliably determine if activity occurs in the same locations or not, between different groups or stimuli (1, 8). During the initial positioning for each MR imaging session, four thermodes connected to a custom-made thermal stimulation device were positioned symmetrically right and left, on the little finger side of the palm of the hand (the eighth cervical dermatome, C8), and on the upper arm near the shoulder (the fifth cervical dermatome, C5).

Spinal fMRI data were acquired using our established method based on signal enhancement by extravascular water protons contrast (1, 3, 9). For each fMRI time-series, a three-dimensional volume from caudal to the T1 vertebra to superior to the thalamus was imaged 45 times in 6 min 45 s, in nine contiguous sagittal slices, each 2-mm thick, by means of a half-Fourier single shot fast spin-echo imaging sequence. The MR signal was combined from the upper elements of a phased-array spine coil, posterior neck coil, and posterior head coil, and coil elements were included or excluded as needed to optimize the signal-to-noise ratio (SNR) for each participant. Each sagittal slice spanned a 28 × 21 cm2 field of view, with 1.5 × 1.5 × 2 mm3 resolution. The imaging parameters include an echo time of 38 ms, and a repetition time of 9 s (1 s/slice). The image quality was enhanced by means of spatial suppression pulses anterior to the spine, and motion compensating gradients in the head–foot direction. An optical pulse sensor incorporated into the MRI system was placed on the left index finger, and the peripheral pulse was recorded throughout each fMRI acquisition. During the entire imaging session, the volunteer was shown a movie of their choice (selected prior to the imaging session) via a rear-projection screen viewed in a mirror, with audio provided over speakers in the room, in an attempt to attain consistency of the participants' attention focus across studies (1).

Sensory Stimulation

Each thermal stimulation paradigm (Fig. 1) consisted of three periods of warm sensations at 44°C, with durations of 45 s each, interleaved with periods at skin temperature or passive cooling toward skin temperature. The set of four paradigms is linearly independent, enabling the response to each thermode to be detected. The choice of stimulation at 44°C provides a sensation that is just below the pain threshold for most people (measured to be 45°C in a group of healthy volunteers, unpublished data) and can be tolerated for 45 s, and provides activity in ascending and descending sensory and pain pathways that has been documented in previous studies (1, 2, 4).

Figure 1.

Thermal–sensory stimulation paradigms for each of the four sensory dermatomes.

The effects of the sequence in which the stimuli were applied were investigated by acquiring three fMRI datasets for each participant, with the thermode positions switched left–right across the body between successive acquisitions. The side of the body that received the thermal stimulation first, in the first fMRI acquisition, was also alternated between successive participants.

Analysis

Spinal fMRI data were analyzed by means of a general linear model (GLM) implemented in custom-made software written in MATLAB® (The Mathworks Inc., Natick, MA) (10). The basis functions for the GLM consisted of models of signal change corresponding to the heating/cooling pattern for each thermode, convolved with the tissue response function (11), a constant function, a linear ramp, and models of cardiac-related motion of the spinal cord (12). Prior to analysis, reference lines were drawn along the anterior, posterior, right, and left, edges of the spinal cord up to the rostral medulla, and then extended to be tangent to the limits of the midline thalamus, and the position of the pontomedullary junction (PMJ) was marked as a rostral–caudal reference. These four reference lines serve to guide spatial smoothing, coregistration, and spatial normalization steps. The analysis consisted of (1) smoothing the data with a 3-pixel-wide boxcar function in the direction parallel to the long axis of the spinal cord (accounting for the curvature of the cord), (2) coregistration of the data to correct for subtle movement of the body, (3) defining the GLM basis functions for each slice, accounting for slice timing differences, (4) calculating the GLM fit parameters (i.e., β-values) and significance for each voxel, and (5) spatially normalizing the results for display and group analyses.

Coregistration was applied in transverse planes (translations and rotation, three degrees of freedom) based on the MATLAB “cpcorr” function, and control points outside of the spinal canal that were defined automatically based on the four reference lines described above. The subtle shifts that were applied were adjusted to ensure smoothness along the cord in the rostral–caudal direction, and all volumes were aligned to the first volume in the time-series.

Spatial normalization was achieved using a modified method from that described previously (13). The previous method was based on normalizing dimensions between the PMJ and the clearly anatomically identifiable intervertebral disc between the C7 and T1 vertebral bodies. However, based on detailed anatomical studies, the absolute distance along the cord from the PMJ is expected to be a more consistent reference coordinate (14, 15). The coordinate system we have defined is therefore along the axis of the cord, with increasingly negative values moving caudally from the PMJ (zero at the PMJ), with right–left coordinates increasing from right to left (zero at midline), and anterior–posterior coordinates increasing from anterior to posterior (zero at the anterior limit of the cord; Fig. 2). The coordinates for each voxel were determined in the original image space, and then data were mapped to the normalized space with 1-mm resolution in all three dimensions by means of linear interpolation for image data, or by means of nearest-neighbor interpolation for statistical maps and region-of-interest masks. A predefined region-of-interest mask was thus used to identify the estimated spinal cord segmental levels or brainstem regions for each area of activity that was detected. A predefined mask was also applied to the normalized results (except as indicated in the Results section) to exclude any apparent activity within the adjacent cerebrospinal fluid or outside the spinal canal.

Figure 2.

Definition of spinal cord normalized coordinate system, based on absolute dimensions along the cord from the PMJ, from the midline in the right–left direction, and from the anterior limit of the spinal cord in the anterior–posterior direction. Regions of interest are indicated, including the pons and medulla (labeled “Med.”), and locations of each cervical spinal cord segment based on detailed anatomical studies (14). Units are in millimeters. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Connectivity analyses were applied to group average time-series data created from spatially normalized data from all participants. The purpose of the connectivity analysis was to further test whether the expected networks of sensory regions were detected for each of the four stimuli, and whether or not the four responses were distinct. A “seed” point was identified within each spinal cord region of interest by identifying the voxel with the time-series with the highest correlation, R, to the stimulation paradigm. This seed point was grown into a “prime cluster” by identifying all of the contiguous voxels that are correlated with the seed point time-series at R ≥ 0.33. The average time-series of the cluster was then used for the connectivity analysis. Functional connectivity between the prime cluster and every other voxel in the brainstem and spinal cord was determined by means of the voxel-by-voxel correlation, with a threshold of R ≥ 0.5.

RESULTS

Results of the GLM analysis demonstrate areas of activity in the spinal cord in every participant, for each of the four stimuli. A total of 30 sets of time-series data were acquired from the 10 participants, with the right-side stimulation preceding the left-side stimulation in 14 datasets, and the left stimulation preceding the right in the other 16 datasets. Group results (random effects analysis of all data for each sequence of stimulation) are shown for analysis with model paradigms corresponding to hand stimulation in Fig. 3 and shoulder stimulation in Fig. 4, and properties of active clusters in the spinal cord are summarized in Table 1. (The distributions of active regions across the entire cervical spinal cord and brainstem without any masking applied are shown for each stimulus and order of stimulation in Figs. S1–S8 in Supporting Information.) With almost every stimulus (six of the eight), the cluster with the highest significance occurs at 4.0 ± 0.6 mm from midline on the side of the spinal cord corresponding to the stimulus (right or left). These active clusters are consistently in the dorsal region at a distance of 5.5 ± 1.4 mm from the ventral limit of the cord. Stimulation of the right hand produced the most significant active clusters (in grouped GLM results) at −170.3 to −173.3 mm from the PMJ, corresponding to the C8 segment. Stimulation of the left hand, with the left-side stimulus applied first, produced the most significant active cluster at −183.9 mm from the PMJ, corresponding to the rostral edge of the T1 spinal cord segment. When the left hand was stimulated second in the sequence (after the right hand), the most significant cluster of activity was detected 1.2 mm to the right of midline, at −173.6 mm from the PMJ, near areas of activity detected with right-side stimulation.

Figure 3.

Group results (random effects analysis, n = 10) for stimulation of right and left hands. Results are shown for selected contiguous 1-mm thick transverse slices through the C8 spinal cord segment and the rostral medulla. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Figure 4.

Group results (random effects analysis, n = 10) for stimulation of right and left shoulders. Results are shown for selected contiguous 1-mm thick transverse slices through the C5 spinal cord segment and the rostral medulla. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Table 1. Regions Within Spinal Cord Segment of Interest, Detected With Group Analysis of GLM Results
Stimulus location and sequenceNo. of active voxelsT (random effects group analysis)A/P (mm)R/L (mm)S/I (mm)Segment
Right hand right-side first303.1425.0−3.2−170.3C8
22.5486.06.0−170.5C8
Right hand left-side first93.2016.1−4.4−173.3C8
303.1756.62.6−179.4C8
Left hand left-side first153.1442.63.7−183.9T1
73.0796.62.1−181.6C8
Left hand right-side first342.8255.6−1.2−173.6C8
142.7827.2−0.6−167.5C7
Right shoulder right-side first73.2215.4−5.0−127.4C4
132.9063.74.5−133C5
42.7446.8−1.2−133.3C5
Right shoulder left-side first93.4707.6−3.8−128.7C5
603.0875.54.7−129.5C5
223.0772.7−2.8−140.7C5
Left shoulder left-side first32.8625.7−2.0−141.3C5
22.6057.52.0−137C5
Left shoulder right-side first412.8845.64.0−133.5C5
222.8047.0−0.3−135.6C5

A similar pattern was observed for activity detected with stimulation of the shoulders. Right shoulder stimulation produced the most significant areas of activity at −127.4 to −128.7 mm from the PMJ, close to the boundary between the C4 and C5 segments. Left shoulder stimulation produced the most significant activity in left dorsal regions at −141.3 to −133.5 mm from the PMJ, within the C5 segment. Considering only cases in which the stimulus was applied first to the side of interest, left-side activity is observed to be 13.6 mm more caudal for hand stimulation, and 9.6 mm more caudal for shoulder stimulation. However, active regions were spread across a range of rostral–caudal positions as shown in Fig. 3.

Contrasted results between right-hand and left-hand stimulation are shown in Fig. 5. These results confirm that activity within the C8 segment is significantly different between right-side and left-side stimulation. Effects of the sequence in which the stimuli are applied (i.e., right-side first or left-side first) are also apparent in the contrasted results in the rostral medulla. Differences in activity that depend on whether a particular hand is stimulated first, or second, in the sequence of stimulation are apparent in the vicinity of the rostral ventromedial medulla (RVM), inferior olivary nucleus, and medullary reticular formation, with the greatest differences (volume involved) observed for right-hand stimulation. With left-hand stimulation, differences were not observed in the region of RVM or olivary nucleus.

Figure 5.

Selected contrasts between study groups, shown in selected 1-mm thick transverse slices through the anatomical regions of interest. Areas of activity in the rostral medulla are contrasted for right-hand stimulation, in studies with the right-hand stimulation applied first, or second, in the sequence of stimulation (RH first vs. second). Corresponding studies are contrasted for left-hand stimulation as well (LH first vs. second). Contrasted results are also shown for the rostral medulla and C8 spinal cord regions between right-hand and left-hand stimulation, for studies in which the stimulus of interest is applied first in the sequence (RH first vs. LH first). Regions of the C8 segment in which there is no activity detected are not shown, for clarity. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Results of functional connectivity analyses are shown in Figs. 6 and 7, and key regions are summarized in Tables 2 and 3, and demonstrate consistent patterns of connectivity between the spinal cord and the rostral medulla, pons, midbrain, and thalamus, as well as connections within the same cord segment and adjacent segments, in some cases. The “prime” cluster locations indicate that right-side and left-side areas in response to hand stimulation are offset by 7 mm (S/I) and for shoulder stimulation are offset by 2 mm. In both cases, the left-side response is more caudal. Although a number of connected regions in the brainstem are contralateral to the seed cluster in the spinal cord, some areas of activity such as in the pons, in the vicinity of the dorsolateral pontine tegmentum, were observed on the same side of the body, regardless of the side being stimulated.

Figure 6.

Functional connectivity between the most significant cluster in the C8 segment in response to stimulation of the right hand (top) or left hand (bottom) and all other areas of the spinal cord and brainstem. Lines are shown between the centroids of active clusters with time-courses that are correlated at R ≥ 0.5. The lines are color-coded, with red indicating connectivity to regions of the thalamus, orange indicates midbrain, yellow indicates pons, and green indicates connectivity to regions of the medulla, whereas blue indicates connectivity with other areas of the spinal cord. Select transverse slices are also shown through prominent clusters of activity in connected regions. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Figure 7.

Functional connectivity between the most significant clusters in the C5 segment in response to stimulation of the right shoulder (top) or left shoulder (bottom) and all other areas of the spinal cord and brainstem. Again the lines are color-coded to indicate the location of the regions connected to the prime cluster (red: thalamus, orange: midbrain, yellow: pons, green: medulla, and blue: spinal cord), and transverse slices show prominent clusters of active voxels in connected regions. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Table 2. Connectivity to C8 Prime Cluster
Correlation RNo. of active voxelsA/P (mm)R/L (mm)S/I (mm)Region
(a) Right-hand stimulation, with right stimulation applied first
0.67285−4−170C8 dorsal right (prime cluster)
0.571171−30Thalamus left
0.55950−46Midbrain ventral left
0.5611158−50Midbrain dorsal left
0.58485−64Pons dorsal left
0.55610−6−66Medulla dorsal right
0.5645−4−141C5 dorsal right
0.53436−150C6 ventral left
0.51456−171C8 dorsal left
0.5888−1−180C8 dorsal right
(b) Left-hand stimulation, with left stimulation applied first
0.652972−177C8 dorsal left (prime cluster)
0.531616−5−36Thalamus right
0.5687−5−53Midbrain ventral right
0.5740−2−59Pons dorsal right
0.551653−58Pons dorsal left
0.56711−3−66Medulla dorsal right
0.5384−2−74Medulla ventral right
0.54920−128C5 ventral left
0.52522−139C5 ventral left
0.61673−168C7 dorsal left
0.55532−171C8 ventral left
Table 3. Connectivity to C5 Prime Cluster
Correlation RNo. of active voxelsA/P (mm)R/L (mm)S/I (mm)Region
(a) Right shoulder stimulation, with right stimulation applied first
0.64207−3−131C5 dorsal right (prime cluster)
0.55964−34Thalamus left
0.531898−43Midbrain ventral left
0.53412−3−54Midbrain dorsal right
0.53493−58Dorsal pons dorsal left
0.5741−6−58Pons dorsal right
0.52482−70Medulla dorsal medial
0.53521−71Medulla ventral medial
(b) Left shoulder stimulation, with left stimulation applied first
0.653263−133C5 dorsal left (prime cluster)
0.555515−3−41Midbrain dorsal right
0.57132−4−41Midbrain ventral right
0.575125−51Midbrain dorsal left
0.55680−50Midbrain ventral medial
0.531532−75Medulla ventral medial
0.5263−2−80Medulla ventral medial
0.54656−119C4 dorsal left
0.55772−125C4 dorsal left
0.541323−150C6 ventral left
0.5886−5−166C7 dorsal right
0.5577−1−180C8 dorsal right

DISCUSSION

The results of this study demonstrate the differences in active regions in the brainstem and spinal cord between right-side and left-side stimulation, without the confounding effects produced by comparing sequential studies. Group GLM analyses demonstrate responses in specific regions of the spinal cord, for each of the four dermatomes that were stimulated (Figs. 3 and 4, Table 1), as expected based on the known anatomy, and similar to that shown in previous studies (16). The spatial distributions of active regions demonstrate right–left symmetry between stimulation of the right and left sides, except that the right-side and left-side responses are not at the same rostral–caudal positions, although they are within the same spinal cord segment. Stimulation of the right hand produced right dorsal activity near the top of the C8 spinal cord segment, whereas left-hand stimulation resulted in left dorsal activity roughly 13 mm more caudal. Similarly, shoulder stimulation produced ipsilateral dorsal activity within the C5 segment that was ∼7 mm more caudal for left-side stimulation.

Contrasted results (Fig. 5) demonstrate the offset between right-side and left-side activity in the C8 spinal cord segment, and also demonstrate the influence of the sequence of the stimuli, particularly in the rostral medulla. The observed effect of the sequence in which the four dermatomes were stimulated implies that there are interactions between the interleaved stimuli, even though they are applied to different dermatomes. The nature and location of these interactions are further supported and elucidated by the functional connectivity analyses.

Functional connectivity results (Figs. 5 and 6) further illustrate the unique responses detected with stimulation of each of the four dermatomes and symmetry between right-side and left-side stimulation. Prime clusters selected for connectivity analyses demonstrate a similar rostral–caudal offset between right-side and left-side stimulation, as with the group GLM analysis. In the C8 segment, the prime clusters for right- and left-side stimulation are separated by ∼7 mm, whereas in the C5 segment, with stimulation of the right- and left-side shoulder stimulation, are roughly 2 mm apart (S/I). Tables 2 and 3 and Figs. 5 and 6 demonstrate that connected regions in the brainstem are predominantly contralateral and in midline regions. Brainstem regions identified with connectivity analyses include the thalamus, midbrain in the vicinity of the periaqueductal gray matter, pons in the vicinity of the dorsolateral pontine tegmentum, and medulla in the vicinity of the inferior olivary nucleus and reticular formation. Although we can speculate that the rostral–caudal offset of responses to right- and left-side stimulation observed in the spinal cord may relate to the number of nerves entering the cord from the skin, and that they are organized somatotopically as in the brain for efficient connection to interneurons and ascending pathways, our data do not show these details of the neuroanatomy, and further investigation is needed. We must also consider that in this study, given the group size of 10 healthy participants, it is possible that this finding is specific to this group.

The group GLM and connectivity analyses therefore consistently demonstrate specific areas of activity in response to each of the four dermatomes that were stimulated, in good agreement with the known neuroanatomy. Common areas of connectivity between all four stimuli are in the left pons, in the vicinity of the dorsolateral pontine tegmentum, and in the right ventral medulla, near the RVM or the olivary nucleus. Interactions between the four stimuli are therefore believed to involve the descending control pathways through the medulla, possibly with a stronger response (larger signal changes) to right-hand stimulation in the group studied. The predominantly ipsilateral regions in the rostral medulla that responded to right-hand and left-hand stimulation, as demonstrated by the contrasted results, are evidence of descending pathway involvement because the ascending pathways through the brainstem are expected to be mostly contralateral. The primary interactions are expected to occur in regions that provide descending control to both sides of the body such as the periaqueductal gray matter and nucleus raphe magnus.

The group GLM and connectivity results from this study demonstrate a high degree of sensitivity and reliability as demonstrated by the agreement with known neuroanatomy in both the spinal cord and brainstem. The new features of the neuroanatomy that are demonstrated by these results are not inconsistent with previous knowledge, but rather cannot be observed by any other means in humans, except for detailed postmortem anatomical studies. The sensitivity of the results also supports the validity of the spatial normalization method that we have used, which is based on absolute position within and along the cord anatomy from the fixed reference point at the PMJ. The results further demonstrate that the stimulation method used in this study, using multiple thermal sensory stimuli in linearly independent paradigms, is effective for investigating multiple sensory responses within a single fMRI acquisition.

To our knowledge, the results of this study are the first to demonstrate right-to-left asymmetry (due to the slight rostral–caudal offset) in the areas of the spinal cord involved with sensory responses in humans. Only three previous studies appear to have compared right-side and left-side stimulation and these all used axial slices with a thickness of 4.5–7.5 mm, and no comparisons of the rostral–caudal distributions of activity were discussed (17–19). Nonetheless, this information provides a valuable reference for future studies of the effects of trauma or disease. Moreover, the results demonstrate important interactions between stimulation of different dermatomes, particularly in brainstem regions. These interactions are expected to involve the descending control pathways from brainstem regions, and may therefore provide a valuable means of probing changes in ascending and descending pathways in studies of the effects of injury and disease. This study therefore contributes for the further development of spinal cord fMRI as a valuable research and clinical assessment tool.

CONCLUSIONS

Specific areas of activity and patterns of connectivity are demonstrated in response to each of the four dermatomes that were stimulated, confirming that the responses to multiple interleaved stimuli can be distinguished. The results also demonstrate high spatial precision as evidenced by the dependence on the side of the body and dermatome being stimulated, and reveal new details of human spinal cord neuroanatomy. Future applications of this approach for the study of the effects of spinal cord trauma, or diseases such as multiple-sclerosis, may therefore reveal differences in regions relative to the level of injury, and between sides of the body, making it a practical approach for clinical assessments.

Ancillary