Abstract
We test the feasibility of an optically pumped magnetometer (OPM)-MEG system for the measurement of human cerebellar activity. We show that the OPM system allows for excellent coverage of this structure by decreasing the average sensor-to-cerebellum distance by around 33% (16mm), compared to a standard MEG helmet. This closer proximity to the cerebellum approximately doubles the signal-to-noise ratio (SNR). As a proof of principle, we used an air-puff stimulus to the eyeball in order to elicit cerebellar evoked and induced responses that are well characterized in non-human models. In three subjects, we observed an evoked component at 50ms post stimulus, which originates in the cerebellum (predominantly ipsilateral). This response was followed by a second component at 100ms post stimulus (predominantly contra-lateral). Sensory stimulation also elicited an event-related broadband spectral power change in the ipsilateral cerebellum at ~100ms in all subjects. We conclude that the OPM-MEG technology offers a promising way to advance the understanding of the information processing mechanisms in the human cerebellum.
Introduction
Our understanding of cerebellar function has undergone a paradigm-shift in recent decades due to the studies of neuroanatomy (Glickstein, Sultan, & Voogd, 2011), neuropsychology (Schmahmann & Sherman, 1998), and functional magnetic resonance imaging (Buckner, 2013; Stoodley & Schmahmann, 2009). Until recently thought of as part of the motor system, accumulating evidence indicates that the cerebellum is essential for a variety of cognitive and social functions (Sokolov, Miall, & Ivry, 2017). Despite this expanding repertoire of cerebellar functions, there is a marked absence of human electrophysiological studies in this area.
In the domain of magnetoencephalography (MEG), less than 1% of publications have reported cerebellar activity (Dalal, Osipova, Bertrand, & Jerbi, 2013). A significant proportion of pioneer work described the cerebellum as one part of physiological (Gross et al., 2002; Jerbi et al., 2007; Muthukumaraswamy et al., 2006; Pollok et al., 2004, 2005; Tass et al., 2003) or pathological (Schnitzler et al., 2009; Timmermann et al., 2003) oscillatory networks, but there have been few subsequent detailed investigations in other domains. This is due to several factors. First, compared to the cerebral cortex, the cerebellar cortex is less favourable to the generation of a measurable MEG signal. This is because its densely folded anatomy causes a high degree of field attenuation due to locally opposing current sources (Dalal et al., 2013; Hashimoto, Kimura, Tanosaki, Iguchi, & Sekihara, 2003; Tesche & Karhu, 1997). Second, cerebellar neurons are thought to have low firing synchrony based on the small amplitudes observed in local field potential studies (Gerloff, Altenmüller, & Dichgans, 1996). Third, the majority of the cerebellum is situated deep in the human cranium. The relatively large distance from the MEG sensors therefore makes it likely that cerebellar-generated MEG signals will be smaller than those measured due to current flow in the neocortex. Finally, yet importantly, current whole-head MEG systems were designed to target the cerebrum and provide poor coverage of the cerebellum (Ioannides 2005 and Figure 2 A). The issue is not exclusive to MEG – to date, few studies with scalp cerebellar EEG have been reported (Lascano et al., 2013; Muthukumaraswamy, Johnson, & Hamm, 2003; Todd, Govender, & Colebatch, 2018). It is widely assumed that scalp EEG recordings for the cerebellum suffer from muscle artefacts (Muthukumaraswamy, 2013). Intracranial recordings in the cerebellum are also scarce because of their limited clinical indications (Dalal et al., 2013; Niedermeyer, 2004). These issues combined mean the electrophysiology of the human cerebellum is almost undocumented.
The development of optically pumped magnetometers (OPMs) provides a new opportunity to investigate cerebellar electrophysiology. OPMs are high sensitivity magnetic field sensors that can be flexibly placed on the scalp and do not need cryogenic cooling so that the sensor-to-brain distance can be significantly reduced. We have recently built such a wearable OPMMEG system, placing the sensors close to the scalp by mounting them in a 3D printed scanner-cast (Boto et al., 2017, 2018). Currently, these sensors can be positioned in a dense array over a specific brain region of interest (Tierney et al 2018). This system is less susceptible to muscle artefacts compared to EEG (Boto et al., 2018). Moreover, the system is accommodated in a field-nulling apparatus (Holmes et al., 2018) which minimises field variation at the sensors due to head movement in the ambient field. These characteristics make OPM-MEG an ideal candidate for the study of cerebellar electrophysiology that is responsible for both motor and non-motor tasks.
As a proof-of-principle study, we recorded MEG data when non-noxious air-puffs to the eye trigger blinks. Air-puffs are the unconditioned stimuli (US) in a well-established cerebellar learning paradigm: eye-blink conditioning. The stimuli are known to elicit activity in the principal cells of the cerebellum, the Purkinje cells. These cells show “simple spike” responses driven by input from the brainstem pontine nuclei as well as “complex spike” responses to climbing fibres projecting from the inferior olive to bilateral, predominantly ipsilateral, cerebellar cortex, in untrained animals (Ohmae & Medina, 2015). Animals and humans present comparable behaviour, and apparently share similar circuitry and responses to physiological and pathological (lesion) interventions (see Freeman & Steinmetz, 2011 for a review of the system). Functional MRI studies in humans also accord well with corresponding electrophysiological studies in animals, with a prominent BOLD response in ipsilateral cerebellar cortex (Cheng, Disterhoft, Power, Ellis, & Desmond, 2008; Dimitrova et al., 2002; Thurling et al., 2015).
We aim to demonstrate that air-puff driven neural signals in the cerebellum, which have been observed in both the invasive animal and non-invasive human literature, can be measured with OPM-MEG. We first show the scalp mounted OPM system reduces the sensor-to-cerebellum distance by one-third, which converts into an approximately two-fold increase of signals. We then present both evoked and induced responses measured with OPM-MEG in the cerebellum. Taken together, we show wearable OPM-MEG of the cerebellum provides a promising future to examine the cerebellum during human cognition and action, and pathological conditions linked to cerebellar dysfunction.
Materials and methods
This section is divided into three parts. First, we describe the OPM-MEG system. Second, we summarise the experimental procedures for cerebellar activity measurement. Finally, we introduce the inversion scheme used to localise the source activity.
Participants
Three healthy subjects (1 female, 2 male) aged 27-50, with no history of psychiatric or neurological diseases, participated in the study. All subjects were naïve to the eye-blink conditioning. The research protocol was approved by the University of Nottingham Medical School Research Ethics Committee and the University of Birmingham Research Ethics Committee. Written informed consent was obtained from all participants. The experiments took place at the University of Nottingham.
OPM-MEG System
The OPM-MEG system has been previously described in detail (Boto et al., 2017, 2018; Holmes et al., 2018; Tierney et al., 2018). Briefly, the system consists of an OPM sensor array within a customised scanner-cast, and a field-nulling apparatus comprising four reference OPM sensors and field-nulling coils (Figure 1).
Optically-pumped magnetometer (OPM)
The OPM sensor used here (QuSpin Inc., Louisville, CO, USA) is an integrated unit with a noise level of ~15 fT/√Hz in the 1-100 Hz band, and a dynamic range of ±1.5 nT (for details see (V. K. Shah & Wakai, 2013; V. Shah, Osborne, Orton, & Alem, 2018)). Each OPM sensor head contains a 3 x 3 x 3 mm3 cell containing a 87Rb vapour, with a 795 nm semiconductor laser beam passing through the cell to pump the atoms to a magnetically sensitive state, and a silicon photodiode. A zero-field resonance is then used to detect the weak biomagnetic fields (Dupont-Roc, Haroche, & Cohen-Tannoudji, 1969). In brief, at zero field the gas is transparent to the laser light and there is maximal signal at the photodiode. Perturbations from zero field cause laser light to be absorbed by the gas (as the atoms have changed energy state) this changes the transparency of the cell to the laser light, resulting in less signal at the photodiode.
Field nulling apparatus
The Earth’s residual static field in the Nottingham magnetically shielded room is ~25nT with a maximal gradient of approximately 10 nT/m. This means that even minimal head movements (e.g. a 4 degree rotation) can easily generate signals which exceed the OPM’s dynamic range. In order to mitigate these effects we used a set of bi-planar coils to generate magnetic fields which counteract the remaining static field as measured by four reference sensors close to the head. By applying this method, the dominant component of the static field and field gradient can be diminished by factors of 46 and 13 respectively, in a volume of 40 × 40 × 40 cm3 encapsulating the head (Boto et al., 2018; Holmes et al., 2018).
Scanner-cast design
Data from an anatomical T1-weighted MRI scan was used to generate a 3D mesh representing the outer surface of each participant’s scalp. This 3D mesh was used in 3D printing to shape the inner surface of a nylon head-cast, as described in (Boto et al., 2017; Meyer et al., 2017; Tierney et al., 2018), with sockets around the outer surface to hold the OPM sensors. Importantly, the mesh and subsequently produced scanner-cast provide accurate socket and thus sensor positions and orientations, which are in the same coordinate space of as the MRI image. Therefore, no coregistration is needed and the sensor positions are defined as the centre of the socket bases with a 6.5 mm offset, which takes into account that the sensitive portion of the OPM not being precisely at the base of the sensor. Here, with limited sensors available, we placed the sensors proximal to the cerebellar and somatosensory cortices. Figure 1 shows the sensor configuration in a typical subject. Across the 3 participants thirteen to nineteen posterior sensors and four to six somatosensory sensors were used. Additionally, two sensors were placed in bilateral infra-orbital slots for eyeblink detection (Figure 1.A).
Experiment: Eyeblink paradigm
Eyeblinks were elicited by a 32 ms air-puff delivered through a nozzle mounted on the scanner-cast (Figure 1.A); essentially a pressurised air cylinder (1 Bar) fed into a 10 m semi-rigid plastic tube (2mm internal diameter), under the control of a bespoke pneumatic valve controller. Some details of the pneumatic delivery system are given in Leonardelli et al., (2015). The nozzle directed the air-puff to the outer canthus of the left eye from a distance of approximately 2-4 cm, individually adjusted to evoke a visible blink after each delivery, but without discomfort. The arrival time of the air-puff was calibrated off-line using a microphone and was relatively insensitive to distance of the nozzle over a limited range (~1 ms/cm). Subjects received four contiguous 12-minute blocks of stimulation. To equate the task to the baseline phase of a previously validated eyeblink conditioning paradigm (Cheng et al., 2008), each block constituted 200 trials: 140 trials of air-puffs, 50 trials of a 550-ms binaural tone (2800 Hz), and 10 paired trials with the tone co-terminated with air-puff delivery; trial order was randomized in sets of 20. A total of 600 air-puff trials were recorded per subject. Every trial began with a random wait of 1-2.5 seconds to avoid habituation to puffs; inter-stimulus intervals averaged to 3.6 s.
MEG data collection
The OPM data were digitized at a sampling rate of 1200 Hz, using a 16-bit digital acquisition (DAQ) system (National Instruments, Austin, TX) controlled by custom-written software in LabVIEW.
MEG data analysis
Pre-processing
Data were filtered between 5 and 80 Hz (or 0 and 200 Hz for eye-blink analysis; see Supplementary Material) and each trial epoched between -200 and +500 ms relative to air-puff onset. Because OPMs are configured as magnetometers (as distinct from gradiometers which are used in many cryogenic MEG systems) they are susceptible to increased environmental interference. We mitigated interference by constructing virtual gradiometers, which linearly regress the signal recorded by the reference array from the signal recorded at the scalp array (Boto et al., 2017). Thereafter, data were concatenated across 4 blocks and all 600 trials of data were ranked according to signal variance. Trials with variances higher than [median + 3x median absolute deviation] were rejected (Leys, Ley, Klein, Bernard, & Licata, 2013). For the evoked response analysis, the remaining trials (548, 588 and 585 trials with air-puff stimulation, for subject 1, 2, and 3 respectively) were baseline corrected to the mean of the window 50 ms prior to stimulus onset and then averaged.
For spectral power changes, single trial time-frequency (TF) decompositions in sensor space were calculated for each subject using a Morlet wavelet transform (Tallon-Baudry et al 1998) and then averaged across trials. The wavelet transform was calculated for each time-point between -150 and 450 ms, with 76 scale bins corresponding to frequencies between 5 and 80 Hz. For each trial and frequency, the mean power of the interval from 50 ms before stimulus onset until stimulus presentation was considered as a baseline level. The power change in each frequency band post-stimulus was expressed as the relative percentage change from the pre-stimulus baseline. We additionally calculated the correlations between the latencies of induced power changes and blinks across trials using Pearson’s r-values (Yuval-Greenberg at al. 2008): for all trials, the latency of the maximal peak of 5-80 Hz TF data at the time window of 0-350 ms was computed and correlated with the latency of the blink peak (as estimated by the time of the peak amplitude for each trial, examples can be seen in Supplementary Figure 2.B.) at the same time window. All of the data analysis was performed using SPM12 within the MATLAB environment (Release 2014a, Mathworks Inc., Natick, MA).
MEG source localisation
We evaluated both the average evoked response using a dipole fit analysis and the average spectral power change using a beamformer. In both cases the volume conductor model was the Nolte single shell model (Nolte, 2003), implemented in SPM12, using the inner-skull boundary from the individual T1-weighted MRI.
Dipole fitting
We reconstructed sources of the evoked field data for each subject using the SPM implementation of variational Bayesian for equivalent current dipole fitting (Kiebel, Daunizeau, Phillips, & Friston, 2008). In brief, the Bayesian inversion scheme assigned prior means and variances of dipole positions and moments. The posterior dipole locations and moments were estimated by maximising the model evidence (as approximated by negative Free energy (Friston, Mattout, Trujillo-Barreto, Ashburner, & Penny, 2007)) for the data, given the model parameters. The advantage of using the Bayesian formalism is that it allows us to test a straightforward set of hypotheses by comparing models in which the sources have different prior locations. We specified the prior mean locations of 4 single dipole models based on the literature (Cheng et al., 2008; Nevalainen, Ramstad, Isotalo, Haapanen, & Lauronen, 2006): (1) right somatosensory cortex (S1, face area), (2) left S1 (face), (3) right cerebellum (in lobule VI) and (4) left cerebellum (lobule VI). Two additional priors in the right and left primary visual cortex (proximal to the cerebellum) were also tested (models 5 and 6). The standard deviation of each prior dipole location was set to 10 mm. The mean and standard deviation of each prior moment were assigned as 0 and 10 nA·m. To avoid local maxima, 50 iterations, with starting locations and orientations randomly sampled from prior distributions (i.e. the means and standard deviations of prior locations and moments), were carried out for each model in each subject (50 x 6 = 300 iterations in each subject). For each subject we used the model parameters corresponding to the highest free energy value across all iterations. The maximal free energy values across subjects were then averaged.
Beamforming
We used the scalar version of a linear constrained minimum variance beamformer algorithm implemented in the DAISS toolbox for SPM (https://github.com/spm/DAiSS) to localise the source of maximal spectral power change over the brain volume. We used a covariance window of 5 to 80 Hz and -50 to +125 ms and contrasted the 75 to 125 ms post-stimulus period with a pre-stimulus period between -50 and 0 ms. The regularization rate λ was set to be 0 (i.e. unregularized). The source orientation was set in the direction of maximal power. The reconstruction grid spacing was 2 mm. We bootstrapped the analysis in order to find a confidence volume on the global peak estimate by randomly resampling trials with replacement and producing 50 peak estimates per subject.
Results
We first compared the spatial distribution of the OPM sensors around the cerebellum to a standard cryogenic (CTF) MEG system (Figure 2) for a typical subject (Subject 2). The cryogenic MEG sensor positions were obtained from another study during which subject 2 underwent conventional MEG scanning (Meyer et al., 2017). It is clear that the coverage of the cryogenic MEG system is sub-optimal for cerebellum (Figure 2.A). To provide a fair comparison of the 18 OPM cerebellar sensors, only the nearest 18 SQUID sensors were taken into account. The average sensor-to-cerebellum distance decreased from 49 mm in the SQUID MEG to 33 mm in the OPM-MEG (Figure 2.B). This change in distance (factor 1.48) translates into an approximate doubling of the signal magnitude measured by the OPM array, when compared to SQUID sensors. Bespoke scanner-casts could further optimise the placing of sensors for cerebellar recording, by adding another row of sockets below the lowest currently available.
We next looked at the average evoked response to air-puff stimulation on the cerebellar OPM sensors and the sensors positioned over the contra-lateral somatosensory cortex. Figure 3.A shows the average evoked response in posterior sensors (left sensors: blue dotted, right sensors: orange solid) for the three subjects. Two main peaks were observed across subjects at around 50-60 ms and 85-115 ms. The field patterns observed at these peaks were qualitatively dipolar (Figure 3.A lower panel). There was a slight latency difference (approx. 2 ms on average) between positive and negative going extrema (e.g. Subject 2, Figure 3.A upper panel), indicating a more complex source distribution. We also observed two response peaks from sensors proximal to the primary somatosensory cortex (anterior sensors: green solid, posterior: blue dotted, Figure 3.B). For each subject, the earliest distinct somatosensory evoked response peaked at 40-50 ms post-stimulus, compatible with the p45m response to facial tactile stimuli (Nevalainen et al., 2006).
We also looked for event related spectral changes in the cerebellar sensors. Figure 4 shows the average time-frequency spectrograms (in percentage change of power) for the sensor with the largest power change in each subject. Increased broadband activity, peaking at ~100 ms post puff contact can be seen in all three subjects. The peak correlation between the latencies of the maximal power changes and blinks, across trials, was significant (p<0.001 in all cases) although weak (R Square = 0.02, 0.02, 0.03 for the three subjects, also see Supplementary Figure 2).
Source reconstruction
Figure 5 shows the source reconstruction results of early and late evoked responses. Figure 5.A compares the free energy values for the early response (50-60ms; top panel) and late responses (85-115ms, bottom panel) for six models with different source location priors, averaged across subjects. Both cerebellar priors are significantly better than the other models (ΔF > 3, as F approximates the log likelihood, a positive difference of 3 means the preferable model is about twenty times more likely). The left (ipsilateral) cerebellum lobule VI had the highest model evidence for both peaks; although this model was not significantly better than the model of a source in the right cerebellum (ΔF = 0.8 for early and 2.2 for late peaks). Figure 5.B shows the fitted source locations for subject-specific, early and late time-windows. The source localisations were within the cerebellum or the brainstem/cerebellar peduncle adjacent to the cerebellum. Table 1 shows the MNI coordinates and anatomical labels of each fit.
Dipole locations converged into the cerebellum even when using priors in the occipital lobe (model 5 & 6, see Figure 5.C.). We also undertook an additional Bayesian single dipole fit using an uninformative prior (initial location randomly sampled from a distribution centring at the middle point of the cerebellum with a 50-mm standard deviation; Figure 5.D.) and found similar results.
In order to locate the event-related power changes we used a beamformer with a covariance window covering a time range of -50 to +125 ms relative to air-puff onset, and a 5-80 Hz frequency band. We contrasted the power within this band between time windows of +75 to +125 ms and -50 to 0 ms relative to baseline (based on the observed sensor level power changes, Figure 4). We plotted each subject’s global maximum location and the bootstrapped confidence volume in the MNI space in Figure 6. The locations were in the left (ipsilateral) cerebellum in all 3 subjects. The 95% confidence volumes based on 50 bootstraps were lower than the grid spacing (2 mm) for subject 1 and 2 and was 12.0 mm3 for subject 3.
Discussion
We have demonstrated that a small (less than 20 sensor) OPM array can detect cerebellar activity during unconditioned eyeblinks elicited by brief air puffs. The evoked responses had both early (~50 ms) and late components (~90-110 ms); spectrally we found a broad-band power change at ~100 ms. The source localisation for both evoked and event-related spectral change was generally consistent with previous neural recordings in animals and fMRI in humans (i.e. ipsilateral).
Evoked fields from the cerebellum used to be considered unattainable at the scalp level because low-amplitude cerebellar activity would be concealed by stronger adjacent neocortical activity. In the small number of published MEG studies, cerebellar activation to somatosensory stimuli has been identified by either using a priori assumptions of an equivalent current dipole (ECD) in the cerebellum (Tesche & Karhu, 1997; Tesche & Karhu, 2000) or by using a beamforming source reconstruction (Hashimoto et al., 2003). The Hashimoto study (Hashimoto et al., 2003) showed a four-component response in the cerebellum following median nerve stimulation, and the authors made putative assignment of these peaks to different cerebellar inputs. The Hashimoto showed robust source level images (again lateralized stimuli predominantly localized to medial cerebellum) but based on 10,000 trials. By contrast, our results are apparent at sensor level and are based on 550-580 trials per participant. We anticipate that a further reduction in required trials could be achieved by bespoke design of a cerebellar scanner-cast, to optimise sensor locations and further increase SNR.
We were concerned about the contribution of artefacts from non-neural sources, especially the neck muscles and eye movements. The corneo-retinal dipole or extra-orbital muscle is not likely to be an issue, as this type of artefact has been shown to be highly focal and limited to fronto-central sensors (Carl, Açik, König, Engel, & Hipp, 2012; Muthukumaraswamy, 2013). Likewise, the air-puff induced eye-blink occurs relatively late (~100 ms) after stimulus onset (Supplementary Figure 1) making it an unlikely source of the early transients. For the same reason, we think it unlikely that the observed changes could be due to some reflexive head/neck muscle activity, as we would expect this to occur post-eyeblink (Valls-Solé et al. 1999). For induced response, the significant, but only weak temporal correlation between the latencies of spectral power changes and blinks (explaining less than 4% of the variance in latency) suggested they were not blink-related artefacts either (Supplementary Figure 2). Further, we observed the same spectral profile (Supplementary Figure 3) at the source level (using an unregularized beamformer) which we would not have expected if the response were due to muscle artefact (Hipp & Siegel, 2013). However, it is important to note that while blink-related non-neural sources can generally be excluded, the cerebellum is known to be a key part of the blink-control circuit (Hesslow, 1994).
To our knowledge, this is the first MEG study to examine the neural response due to the unconditioned stimulus (US, i.e. air-puffs) of eyeblink conditioning. Being a model system for cerebellar learning, the neural circuits of both unconditioned and conditioned blink responses have been extensively investigated in animals. In untrained rodents, single-unit recording from Purkinje cells in ipsilateral cerebellar cortical lobule VI show that air-puffs elicit climbing fibre inputs, with high probability, peaking around 15-50 ms post-stimulus (Mostofi, Holtzman, Grout, Yeo, & Edgley, 2010; Ohmae & Medina, 2015). Importantly, excitation of the very powerful climbing fibre-Purkinje cell synapse, simultaneously activating the dendrites of a spatially aligned set of Purkinje cells has been considered to be the most probable source for a strong open field detected by MEG recordings (Tesche & Karhu, 1997). According to animal local field potential recording, the US reaches the cerebellum first via an early mossy fibre response and then the aforementioned climbing fibre response (Hesslow, 1994; Mostofi et al., 2010). Our MEG evoked responses showed two peaks, the first tightly clustered around 50-60 ms while the second occurred 80-110 ms post-puff with significant inter-individual differences in latency. These first human data are thus broadly consistent with the animal literature. Previously, similar qualitative features including multiple components (Hashimoto et al., 2003; Tesche & Karhu, 2000) and inter-individual latency variability (Tesche & Karhu, 1997) responding to simple somatosensory input have also been found in human MEG at source levels.
We also report a broadband power increase ranging ~100 ms following air-puff stimulation in all subjects, originating from the ipsilateral cerebellum. In the sparse number of MEG studies reporting cerebellar oscillatory changes, Tesche and colleagues discovered increases in the alpha and gamma range power in response to somatosensory stimulation administration and omission (Tesche & Karhu, 2000). For action execution and observation, decreased power in alpha and beta range was observed (Kennedy, Singh, & Muthukumaraswamy, 2011; Muthukumaraswamy et al., 2006).
The Bayesian inversion scheme found the left (ipsilateral) lobule VI was the best fitting single-dipole model of these evoked responses. For the induced response, the left cerebellum was also identified as the neuronal source by beamforming. The source locations are in agreement with previous findings (Cheng et al., 2008). However, it should be noted that in all three subjects, the best evoked-response fits located close to the midline, more medial than lobule VI. Besides, for both early and late peaks of the evoked field, the differences of Free energy values between the right and left cerebellum lobule VI were small (ΔF < 3, Figure 5.A). One potential cause of this finding is that although ipsilateral lobule VI is the dominant neuronal source of the US signal, concurrent (but weaker) neuronal activation has been found in the contralateral lobule VI in both human fMRI (Dimitrova et al., 2002; Thurling et al., 2015) and in animal studies (Mostofi et al., 2010). When fitting evoked fields with two-dipole priors, the majority of best fits were indeed paired bilateral cerebellar sources (see Supplementary Figure 4). These results support the hypothesis of bilateral activation in the cerebellum by unilateral air-puff stimulation. This said, we admit the current results cannot conclusively locate the sources at a resolution of cerebellar lobules (Schmahmann et al., 1999). To give confidence in the identities and localisations of our findings, we intend to extend the studies to include classical conditioning and extinction of the eye-blink, and to track the predicted changes in the responses to the unconditioned and conditioned stimuli.
The current sensor layout is limited by the number of OPM sensors available. Given the relatively greater depth of the cerebellum, its complex architecture, and the diffuse spatial signature seen in the evoked responses (see the fieldmaps in the Figure 3.A., lower panel), the source localisation would likely benefit from a denser and spatially extended sensor coverage. Additionally, it would be useful to design scanner-casts allowing sensor placed over the upper neck to provide a necessary scope on the inferior cerebellum, which was less sufficiently covered by the current set-up (see Figure 2.A.).
Conclusion
We have demonstrated an OPM-MEG system that can be used to study the electrophysiology of the human cerebellum. The similarities between human MEG and animal field potential data in this proof-of-principle task offers promise for future studies to advance our understanding of cerebellar function through non-invasive electrophysiology in humans. Possessing sufficient signal detectability and being wearable and potentially moveable, we expect OPM-MEG systems to be a powerful tool in the investigation of cerebellar functions in both motor and cognitive tasks (Boto et al., 2018; Sokolov et al., 2017; Tierney et al., 2018). This OPM-MEG system also has the capacity to fill the “white regions” (Niedermeyer, 2004; Schomer & Lopes da Silva, 2010) of the map of clinical electrophysiology in the cerebellum, with impacts on diseases including movement disorders (Bostan & Strick, 2018), mental disorders (Romer et al., 2017), dementia (Fyfe, 2016), to name a few.
Acknowledgements
CL is funded by a BBSRC research grant (BB/M009645/1). This work was supported by a Wellcome collaborative award to GRB, MJB and RB (203257/Z/16/Z, 203257/B/16/Z). The WCHN is supported by a strategic award from Wellcome (091593/Z/10/Z). RCM is funded by a Royal Society Leverhulme Senior Fellowship and by Wellcome grant WT212422. We thank Uta Noppeney for use of the pneumatic controller to deliver air-puffs and Jose David Lopez for helpful discussion.