Investigations of human cortical processing of gentle touch
A study with time-resolved electro-magnetic signal analysis
Elin Eriksson Hagberg
Department of Physiology
Institute of Neuroscience and Physiology Sahlgrenska Academy, University of Gothenburg
Gothenburg, Sweden, 2019
time-resolved electro-magnetic signal analysis.
© Elin Eriksson Hagberg 2019 elin.eriksson-hagberg@neuro.gu.se ISBN 978-91-7833-374-5 (PRINT) ISBN 978-91-7833-375-2 (PDF) http://hdl.handle.net/2077/59060 Printed in Gothenburg, Sweden 2019 Printed by BrandFactory
processing of gentle touch
A study with time-resolved electro-magnetic signal analysis
Elin Eriksson Hagberg
Department of Physiology, Institute of Neuroscience and Physiology Sahlgrenska Academy, University of Gothenburg
Gothenburg, Sweden
ABSTRACT
The present work summarizes investigations of the temporal correlates of brain activity elicited by gentle, moving touch on the hairy skin in healthy participants and in epilepsy patients. Light touch to the hairy skin activates two distinct afferent classes: fast conducting, Aβ afferents and slowly conducting C-tactile (CT) afferents.
Aβ afferents signal discriminative aspects of touch, whereas CT afferents are proposed to play a role in affective touch. Using complementary neuroimaging methods with high temporal resolution, we aimed to distinguish between brain responses evoked by Aβ and CT afferents. In Papers I and IV, electroencephalography (EEG) showed that brush stroking on the hairy skin evoked an ultra-late potential, presumably driven by CT afference to the brain. Source localization indicated the cingulate cortex, and the precuneus as the underlying sources of this ultra-late potential. In Paper II, using magnetoencephalography (MEG) to spatially track brain activations in response to brush stroking over time, we showed that Aβ afference rapidly activates a well-defined network, including operculo-insular and cingulate regions. In Paper III, time-frequency analyses of MEG recordings from healthy participants were complemented with analyses of stereotactic EEG (SEEG) recordings from epilepsy patients. Here, we showed that naturalistic stroking touch induced spectral changes in alpha, beta, and gamma frequencies in sensorimotor regions and the posterior insula, similar to what has been described previously in studies using less naturalistic stimuli such as electrical median nerve stimulation. The present work contributes new information about the spatiotemporal evolution of the brain’s responses to caress-like touch and highlights the importance of considering both Aβ and CT afferents in gentle touch processing.
Keywords: Aβ afferent, C-tactile, touch, EEG, MEG, SEEG
En lätt beröring av huden aktiverar ett flertal olika nervtrådar som var och en förmedlar olika aspekter av beröringens kvalitet. Den mest undersökta typen av dessa nervtrådar kallas för ’Aβ-afferenter’, vilka gör att vi med hög träffsäkerhet kan använda känselsinnet för att skilja på olika material och former. Nervsignalerna från Aβ-afferenterna fortleds från huden till hjärnan med hög hastighet, ca 60 m/s. Utöver detta snabba system, finns det ett långsamt känselsystem av nervtrådar som kallas för C-taktila (CT) afferenter.
Dessa nervtrådar leder sinnesintrycken med låg hastighet, ca 1 m/s och de liknar på så sätt mer hudens temperatur- och smärtsystem än Aβ-afferenterna.
CT-afferenter aktiveras optimalt när ett mjukt föremål långsamt stryker över huden. Studier har visat att aktivitet i CT-afferenter är sammankopplat med subjektivt upplevd behaglighet av en beröring och forskare tror därför att CT- systemet utgör en kanal för bearbetning av beröringar som bär en emotionell komponent. En emotionell hudberöring kan t.ex. vara en smekning i syfte att trösta, lugna eller visa tillgivenhet till en nära anhörig. Upplevelsen av behagliga hudberöringar är en grundläggande komponent i sociala interaktioner och tros ha betydelse för hur människor knyter an till varandra.
Således är flera kliniska frågeställningar kopplade till förståelsen av hur hjärnan bearbetar lätta hudberöringar. Det har t.ex. påvisats att personer med autism bearbetar lätta hudberöringar annorlunda jämfört med ”neurotypiska”
personer. Vidare finns det studier som indikerar att patienter med anorexia nervosa har en störd kroppsuppfattning, och att detta även reflekteras i hur dessa personer bearbetar lätta hudberöringar.
CT-afferenternas exakta funktion är dock långt ifrån fastställd. Forskare är t.ex. inte säkra på vilka områden i hjärnan som aktiveras mer av CT respektive Aβ-afferenter i relation till lätta och behagliga hudberöringar.
Detta forskningsprojekt avser därför att studera hjärnans aktivitet till följd av lätta, behagliga hudberöringar för att belysa frågeställningen om hur CT- afferenternas information bearbetas, och hur Aβ och CT-afferenternas signaler integreras i hjärnan för att en känsla av beröring skall uppstå. För att undersöka detta använde vi tre komplementära hjärnavbildningsmetoder:
elektroencefalografi (EEG), magnetencefalografi (MEG), och stereotaktiskt EEG (SEEG). Dessa metoder har mycket hög tidsupplösning, vilket innebär att man kan mäta hur hjärnsignalerna varierar över tid i relation till en yttre stimulering. EEG och MEG mäter hjärnans elektromagnetiska aktivitet, utanför skallen. SEEG är en metod som används i epilepsikirurgiska utredningar, då elektroder opereras in i hjärnan för att mäta elektrisk aktivitet direkt från hjärnvävnaden. EEG och MEG studierna utfördes på friska
epilepsikirurgisk utredning med SEEG.
Resultaten visade att Aβ afferenterna snabbt aktiverar områden i hjärnan som har att göra med den direkta bearbetningen av hudberöringarna som sådana, dvs, dess kvalitativa egenskaper. Aβ afferenterna aktiverade även områden i hjärnan som är involverade i att reglera kroppens fysiologiska balans och välbefinnande, och områden kopplade till emotioner. Resultaten antydde vidare att CT-afferenternas signaler anländer till hjärnan med en fördröjning på ca en halv sekund efter Aβ-afferenterna, vilket var förväntat baserat CT- afferenteras långsamma nervledningshastighet, och förstärker aktiviteten i de hjärnområden som reglerar välbefinnande och emotioner.
Sammanfattningsvis bidrar studierna i denna avhandling med ny information angående tidsförloppen i hjärnans aktivitet till följd av lätta behagliga hudberöringar. Dessa fynd har relevans för den fortsatta forskningen om känselsinnets roll i sociala interaktioner.
This thesis is based on the following papers, referred to in the text by their Roman numerals.
I. Ackerley, R., Eriksson, E., & Wessberg, J.
Ultra-late EEG potential evoked by preferential activation of unmyelinated tactile afferents in human hairy skin.
Neuroscience Letters 2013: 535, 62-66.
II. Eriksson Hagberg, E., Ackerley, R., Lundqvist, D., Schneiderman, J., Jousmäki, V., & Wessberg, J.
Spatio-temporal profile of brain activity during gentle touch investigated with magnetoencephalography.
Manuscript, submitted.
III. Eriksson Hagberg, E., Krýsl, D., Ackerley, R., Nilsson, J., Schneiderman, J. Lundqvist, D., Jousmäki, V., Malmgren, K., Rydenhag, B., & Wessberg, J.
Induced brain responses to natural touch recorded with intracranial stereo-EEG and MEG.
Manuscript.
IV. Eriksson Hagberg, E., Wramner, M., Nyberg, A., Blümel, S., Ackerley, R., Schneiderman, J., & Wessberg, J.
Cortical potentials elicited by gentle touch to the hairy skin: A high- resolution EEG study.
Manuscript.
ABBREVIATIONS ... IV
1 INTRODUCTION ... 1
1.1 Tactile innervation of the human hairy skin ... 1
1.1.1 Aβ afferents ... 2
1.1.2 CT afferents ... 3
1.2 Central processing of gentle touch ... 6
1.2.1 Spinal pathways ... 6
1.2.2 Cortical processing of gentle moving touch ... 7
1.3 Time-resolved neuroimaging ... 9
1.3.1 Signal source ... 10
1.3.2 Electroencephalography ... 12
1.3.3 Magnetoencephalography ... 13
1.3.4 Stereotactic EEG ... 14
1.3.5 Studying the brain in time ... 16
1.3.6 Modulation of brain rhythms by sensory stimulation ... 17
1.3.7 EEG studies of ‘pleasant’ tactile stimuli ... 18
1.4 Thesis motivation ... 19
2 AIM ... 21
3 METHODS ... 23
3.1 Ethics and participants ... 23
3.2 Tactile stimuli ... 24
3.3 Neuroimaging ... 25
3.3.1 EEG recordings ... 25
3.3.2 MEG recordings ... 25
3.3.3 SEEG recordings ... 26
3.3.4 Co-registration ... 27
3.4 Data processing ... 27
3.4.1 Preprocessing ... 28
3.4.3 Time-frequency analysis ... 28
3.4.4 Source reconstructions ... 29
3.4.5 Statistical analysis ... 30
4 RESULTS ... 33
4.1 Paper I ... 33
4.2 Paper II ... 33
4.3 Paper III ... 35
4.4 Paper IV ... 36
5 DISCUSSION ... 39
5.1 Evoked responses ... 40
5.1.1 EEG ... 40
5.1.2 MEG ... 42
5.2 Time-frequency responses ... 43
5.3 Time course of activity in the posterior insula to gentle touch ... 44
5.4 Limitations ... 45
5.5 An updated framework ... 45
6 CONCLUSION ... 47
7 FUTURE PERSPECTIVES ... 48
ACKNOWLEDGEMENTS ... 49
REFERENCES ... 51
ACC anterior cingulate cortex aMCC anterior midcingulate cortex
BA Brodmann area
BEM boundary element model BOLD blood-oxygen-level-dependent CLTM C low-threshold mechanoreceptor
CT C-tactile
dSPM dynamic statistical parametric mapping EEG electroencephalography
EMG electromyography
ERD event-related desynchronization ERF event-related field
ERP event-related potential ERS event-related synchronization
fMRI functional magnetic resonance imaging GFP global field power
HPI head-position indicator
LCMV linearly constrained minimum variance LTM low-threshold mechanoreceptor M1 primary motor cortex
MEG magnetoencephalography MNE minimum norm estimate MRI magnetic resonance imaging OFC orbitofrontal cortex
PET positron emission tomography pSTS posterior superior temporal sulcus RAII rapidly adapting type II
ROI region of interest
S1 primary somatosensory cortex S2 secondary somatosensory cortex SAI slowly adapting type I
SAII slowly adapting type II
SEEG stereotactic electroencephalography TFR time-frequency response
VPL ventral posterior lateral VMpo ventro-medial posterior
1 Introduction
This thesis is focused on the brain processing of naturalistic, moving tactile stimuli, with emphasis on the temporal evolution of activity in somatosensory brain regions. Moving tactile stimuli (e.g., stroking touch) on the hairy skin are specifically implicated in positive affective touch (McGlone et al., 2014).
Although it may intuitively seem like a simple stimulus, a gentle caress on the arm gives rise to a complex pattern of activity in peripheral sensory neurons. Concomitant firing in multiple populations of somatosensory afferents with different mechanoreceptive properties produces a population- based neural code that provides the brain with a full representation of the tactile event.
The first and second parts of this chapter introduce peripheral and central gentle touch processing, respectively. The third part of the Introduction provides a background to the neuroimaging methods that were used for this thesis, followed by a justification of the studies performed.
1.1 Tactile innervation of the human hairy skin
Light stroking touch to the human hairy skin readily activates low-threshold mechanoreceptors (LTMs), sending impulses along their axons towards the central nervous system. These mechanoreceptive afferents are composed of two main groups: myelinated, fast conducting, Aβ afferents and unmyelinated, slowly conducting, C-tactile (CT) afferents. In animal studies, the CT afferents are denoted C low-threshold mechanoreceptive afferents (CLTMs). Although both Aβ and CT afferents are activated by light moving tactile stimuli, their response profile differs (Figure 1). Aβ afferents show a linear or exponential increase in firing rate as stroking velocity increases, whereas CT afferents respond preferentially to stroking velocities within a certain range that are considered pleasant (Löken et al., 2009; Ackerley et al., 2014b). Indeed, the proposed hypothesis regarding the functional role of CT afferents is that they are important for mediating positive affective (pleasant) aspects of touch (McGlone et al., 2014). However, much less is known about the CT system compared to the extensively studied Aβ system. Hence the following text provides a brief overview of the Aβ afferents innervating the human hairy skin and a more thorough introduction of the less understood CT system.
1.1.1 Aβ afferents
Aβ mechanoreceptive afferents are classified physiologically according to their adaptation profile and size of the receptive field (Vallbo et al., 1995).
Human hairy skin contains rapidly adapting (RA) hair and field units that are responsive to hair follicle deflection and vibration, respectively, rapidly adapting type II units (RAII; putatively connected to Pacinian corpuscles) that are responsive to vibration, slowly adapting (SA) type I units (SAI;
putatively connected to Merkel cells) that are responsive to skin indentation, and slowly adapting type II units (SAII; putatively connected to Ruffini endings), responsive to skin stretch (Vallbo et al., 1995; Abraira and Ginty, 2013). In a study of the forearm skin, the median mechanical thresholds were 0.45 mN for SAI units, 1.3 mN for SAII units, and 0.1 mN for field units (Vallbo et al., 1995).
RA and SA units are distinguished by their response profile to sustained skin indentation. RA units fire vigorously to the instant changes in skin deformation, i.e. when a probe touches down on the skin and when it lets go, whereas they do not fire to sustained pressure with a constant force. SA units on the other hand fire continuously during sustained pressure on the skin (Vallbo and Johansson, 1984). The receptive field of hair and field units is
~80 – 110 mm2 and contains multiple high-sensitivity spots, where hair units fire briskly to the displacement of individual hairs, whereas field units are not particularly sensitive to hair deflections (Vallbo et al., 1995). The receptive fields of RAII units in the hairy skin are larger and they respond strongly to skin vibration, including taps that are located remotely away from the receptive field. However, they contain one highly sensitive spot where the firing frequency is maximal (Vallbo et al., 1995). SAI units have a receptive field size around 10 mm2 that contains a few high-sensitivity spots, separated by silent regions. SAII units often fire spontaneously and their receptive fields are ~1.4 mm2, defined as the area where the firing is three times higher than the spontaneous firing rate (Vallbo et al., 1995).
The conduction velocity for Aβ afferents is in the range of 37 – 73 m/s (Kakuda, 1992). Based on the response properties of the Aβ LTMs to tactile stimuli it is well-recognized that Aβ afferents signal discriminative qualities of touch, e.g., the location of a stimulus on the body, the shape and texture of objects (Phillips et al., 1992; Saal et al., 2009; Pruszynski and Johansson, 2014).
1.1.2 CT afferents
The existence of animal CLTMs was first reported by Yngve Zotterman (1939). When recording nerve activity from a thin branch of the cat’s saphenous nerve, in an ex vivo skin-nerve preparation, Zotterman found that three types of spikes could be detected when gently stroking the skin (cf. the Erlanger and Gasser scheme of the compound action potential; Erlanger and Gasser, 1924; 1930). The first spike type had short latency and large spike amplitude, corresponding to that of Aβ afferents. The second spike type, coming from Aδ afferents due to hair follicle deflection, had longer latency and a smaller spike height. The third spike type was derived from C fiber afferents and appeared as single spikes with small amplitudes or as summated potentials at a longer latency than the Aβ and Aδ spikes.
In humans, single unit recordings, using the microneurography technique (Vallbo and Hagbarth, 1968), have shown the existence of CT afferents in the infra- and supraorbital nerves of the face (Johansson et al., 1988; Nordin, 1990), the lateral and dorsal antebrachial cutaneous nerves of the forearm (Vallbo et al., 1993; Vallbo et al., 1999; Löken et al., 2009), the lateral cutaneous femoral nerve of the thigh (Edin, 2001), and in the radial nerve of the dorsum of the hand and the peroneal nerve of the lower leg (Löken et al., 2007). In microneurography recordings from the antebrachial nerve, which
! ! ! ! ! !
"
#
$ %$
Figure 1. Neural firing rate to gentle brush stroking. Dots show mean firing rate to the different velocities (indicated by the x-axes). Brush stimuli were performed at two different calibrated normal forces, indicated by open and filled dots. Left panel: CT afferents, (n = 16). Right panel: hair-unit afferents (n = 4). Adapted from Löken et al. (2009).
innervates the forearm skin, CT units are encountered approximately as frequently as Aβ fibers (Vallbo et al., 1993; Vallbo et al., 1999; Löken et al., 2009). CTs have never been encountered when recording from afferents innervating the glabrous skin in humans; hence the current belief is that they lack altogether in human non-hairy skin, which has implications for their role in touch processing. However, some studies have reported CLTMs in rodent glabrous skin (Cain et al., 2001; Djouhri, 2016).
CT afferents respond readily to very light indentation forces of 2.5 mN or less. The mean conduction velocity of CTs ranges from 0.6 – 1.2 m/s (Vallbo et al., 1999; Watkins et al., 2017). The receptive fields of CTs are small, ranging from 1 – 35 mm2, and contain one to a few high-sensitivity spots where the firing frequency is maximal (Wessberg et al., 2003). CTs fire most vigorously when a moving stimulus traverses the receptive field, e.g., when a soft brush strokes the skin (Vallbo et al., 1999). More specifically, the mean firing frequency of CTs is highest when a soft object strokes the skin at velocities ranging from 1 – 10 cm/s (Löken et al., 2009; Ackerley et al., 2014b). Stimulus velocities below 1 cm/s or exceeding 10 cm/s decrease CT firing frequency and the relationship between CT afferent firing and stroking velocity fits a negative quadratic regression model (inverted-U shape, see Figure 1). Furthermore, they modulate their firing pattern to the temperature of the stroking probe, exhibiting the highest mean firing frequency to neutral (skin like) temperature of 32°C in contrast to cool (18°C) and warm (42°C) temperatures (Ackerley et al., 2014b).
Repeated suprathreshold stimulation induces fatigue in CT afferent firing.
When the inter-stimulus interval is 3 – 5 s, the most prominent decrease in impulse rate is seen between the first and second stimuli, whereas the decrease between the second and subsequent stimuli is less marked. The fatigue is less prominent with longer inter-stimulus intervals, i.e. 10 s (Vallbo et al., 1999; Wiklund Fernström, 2004). Furthermore, CTs have an intermediate adaptation rate to sustained indentation, where the impulse rate declines to zero or near zero within 4 s of indentation. However, if tactile pressure is sustained for 10 – 30 s, the activity resumes and in some cases build up to considerable rates, called delayed acceleration (Vallbo et al., 1999).
Functional role of CT afferents 1.1.2.1
Due to the slow conduction velocity of CTs, their role in coding rapidly changing tactile events is probably of very low significance. Furthermore, unlike the sensation of a ‘second pain’, which is due to slowly conducting C- nociceptive afferents (Craig, 2002), there is no equivalent conscious
perception of a ‘second touch’ that would be due to long latency CT-related signaling in the brain. In fact, on direct interrogation, the delayed acceleration seen in CTs (described previously herein) does not cause any discernible sensation of any changes to the touch when asking subjects during experiments (Vallbo et al., 1999). The observation that subjects do not seem to be able to distinguish any changes in touch quality, during the delayed acceleration, suggests that CT signaling is processed on a subconscious level.
Also, it is generally believed that patients suffering from complete large fiber neuronopathy altogether lack the ability to detect innocuous touch.
Nevertheless, two investigated neuronopathy patients (patient initials: G.L.
and I.W.) lacking myelinated large diameter Aβ fibers, but with a presumably intact C fiber system, could detect when being touched on the hairy skin, however with poor localization accuracy, but not on the palm, in a two- alternative forced choice task (Olausson et al., 2002; Olausson et al., 2008a).
Moreover, on direct interrogation, and without knowing what type of tactile stimulus (a soft brush stroke) that had been applied, G.L. reported that the touch perception, although faint and diffuse, felt like ‘a pressure’ and that the sensation was pleasant, without any perception of pain, temperature change, or itch (Olausson et al., 2002). Further testing of G.L. and I.W. were subsequently performed to explore the qualitative features of presumed selective CT stimulation by gentle brush stroking on the hairy skin. Both G.L. and I.W. found it difficult to consciously apprehend the touch sensations, and that they were difficult to describe. The common characteristics were however that the sensations were not associated with pain, tickle or itch, but rather slightly or moderately pleasant (Olausson et al., 2008a; McGlone et al., 2014). It was also found that gentle brush stroking in these patients, and in healthy participants, elicited sympathetic skin responses, suggesting that CT afferent activity may influence autonomic regulation (Olausson et al., 2008a).
Experiments performed in healthy participants combining psychophysics and microneurography have shown that there is a correlation between CT afferent mean firing frequency and the perceived pleasantness of brush stroke stimuli (Löken et al., 2009; Ackerley et al., 2014b). When asked to rate brush strokes of different velocities (0.1, 0.3, 1, 3, 10, and 30 cm/s) on a visual analog scale (VAS) from unpleasant to pleasant, subjects rate the 1 – 10 cm/s strokes as more pleasant than slower and faster velocities (Löken et al., 2009; Ackerley et al., 2014b). Löken et al. (2009) found that the variance of the mean impulse rate of the CTs could explain as much as 70 % of the variance in the pleasantness ratings. Importantly, they found no correlation between pleasantness ratings and the mean firing frequency in Aβ fibers in the hairy skin. On the contrary, Aβ fibers have a linear response pattern to increased
velocities, with higher mean firing frequencies as velocity increase (Löken et al., 2009; Ackerley et al., 2014b).
Löken et al. (2009) also investigated the pleasantness ratings when brushing on the glabrous skin of the palm in subjects. Their results showed that gentle stroking on the glabrous skin of the palm is also considered pleasant, but less compared to when the stimulus is delivered to the hairy skin of the arm.
Moreover, the VAS results from the palm condition did not show a statistically significant correlation between pleasantness ratings and stimulus velocity. However, later studies have reported that gentle brushing on the palm of the hand also gives rise to the highest ratings of pleasantness around stroking at 3 cm/s (Löken et al., 2011; Ackerley et al., 2014a). Although, it seems that the pleasantness sensation is not the same as on hairy skin, as Löken et al. (2011) found order effects between brushing on glabrous and hairy skin sites, and both Ackerley et al. (2014a) and Walker et al. (2017) found differences in the shape of the pleasantness curve over different stroking velocities (i.e. it was not the typical inverted-U shape).
1.2 Central processing of gentle touch
1.2.1 Spinal pathways
Sensory signals from the skin ascend in two main pathways: the dorsal column and the spinothalamic tract. Primary Aβ sensory fibers innervating the trunk and limbs terminate in intermediate or deep layers (~III – V) of the dorsal horn of the spinal cord, where they synapse with second order neurons, which project their fibers proximally in the dorsal column medial lemniscal pathway (Abraira and Ginty, 2013). Specifically, mechanoreceptive impulses from axons innervating the upper limbs are conveyed ipsilaterally in the cuneate fascicle of the dorsal column pathway. From the cuneate nuclei located in the medulla, where processing also takes place (Jörntell et al., 2014), the signals decussate to the contralateral side and ascend in the medial lemniscal tract to the ventral posterior lateral (VPL) nucleus of the thalamus.
VPL neurons that transmit cutaneous information subsequently project mainly to the primary somatosensory cortex (S1) Brodmann area (BA) 3b, in the postcentral gyrus (Krubitzer and Kaas, 1992).
The outermost layer, lamina I, of the dorsal horn receive input from primary small diameter (Aδ and C) afferents and relay these signals via the spinothalamic tract. In addition to conveying information about temperature and nociception, the lamina I neurons signal other kinds of physiological changes (e.g., local tissue metabolism and inflammatory responses) that
relate to the control of homeostasis (Craig, 2002). CLTMs terminate in lamina II in the dorsal horn in rodents (Sugiura et al., 1986; Li et al., 2011;
Abraira et al., 2017) and relay signals to lamina I projection neurons via interneurons (Andrew, 2010). However, the spinal projection for CT afferents in humans is not known, but based on animal work and clinical observations (Foerster et al., 1932; Lahuerta et al., 1994), it has been suggested that CTs ascend in the spinothalamic tract together with other thin fiber input (McGlone et al., 2014).
Lamina I spinothalamic neurons decussate around one or two spinal levels above the input level, and ascend on the contralateral side to the posterior part of the ventromedial nucleus (VMpo) of the thalamus. VMpo neurons then project to the insula, and BA 3a within S1 (Craig, 2002).
1.2.2 Cortical processing of gentle moving touch
Most of the studies regarding the cortical processing of gentle moving touch have been performed with functional magnetic resonance imaging (fMRI), which relies on the blood-oxygen-level-dependent (BOLD) contrast.
Although fMRI has good spatial resolution, the temporal resolution is low (around a second) due to reasons discussed in Section 1.3 of this thesis.
In healthy subjects, several studies using fMRI, have reported that gentle moving touch on the hairy skin activates somatosensory regions, i.e. S1 predominantly contralateral to the stimulation site, bilateral secondary somatosensory (S2) cortices, and insular areas (e.g., Olausson et al., 2002;
Bjornsdotter et al., 2009; Kress et al., 2011; Ackerley et al., 2012; Gordon et al., 2013; Sailer et al., 2016). These findings are in agreement with the brain activity elicited by other types of tactile stimuli, mainly targeted towards Aβ afferent stimulation, e.g., single-unit intraneural stimulation (Sanchez Panchuelo et al., 2016), transcutaneous electrical nerve stimulation (Forss et al., 1996; Ferretti et al., 2007; Avanzini et al., 2018), vibration (McGlone et al., 2002), air puffs (Nakamura et al., 1998; Huang and Sereno, 2007; Oh et al., 2017), and mechanical tapping (Hinkley et al., 2007; Hayamizu et al., 2016).
Cortical projections of CT afferents 1.2.2.1
fMRI studies have shown that presumed selective activation of CT afferents by soft brush stroking on the forearm of the patients G.L. and I.W. activates the contralateral posterior insular cortex but not S1 or S2 (Olausson et al., 2002; Olausson et al., 2008b). It should be pointed out that the activation of the contralateral posterior insula in these studies was found when doing a
directed search in that region, and not on a whole brain search level, indicating that the activation was weak. Since there is no existing method to exclusively activate CTs without concomitantly activating Aβ afferents, in healthy subjects, additional direct evidence that CTs project to the posterior insula is limited. However, there are a few studies that support this theory.
One important finding is that the BOLD response to gentle touch stimulation of the thigh and the arm is somatotopically organized in the contralateral posterior insula in the patient G.L. and healthy subjects (Bjornsdotter et al., 2009). Another study has shown that the posterior insular response to CT- optimal touch is absent in a neuronopathy patient group (n=5) with a rare mutation causing reduced C fiber innervation (Morrison et al., 2011a). A third essential finding is that the BOLD response in the posterior insula is tuned to pleasant caressing speeds (Morrison et al., 2011b).
The role of the insula in sensory processing 1.2.2.2
The human insula, or the ‘island of Reil’, first described by J.C. Reil in 1796, is a complex structure with connections to somatosensory and motor cortices, basal ganglia, amygdala, other limbic areas, and the thalamus (Augustine, 1996; Craig, 2011). It is roughly divided by a central sulcus into the anterior insular cortex, containing three short gyri, and the posterior insular cortex, containing the two posterior long gyri (Afif et al., 2010). Histologically the anterior insular cortex is an agranular region that exhibits relatively little differentiation between layer II and layer III, and lacks layer IV. The posterior insular cortex is called the granular region of the insula because it constitutes true isocortex with a distinguished layer IV that receives thalamocortical input (Namkung et al., 2017).
The insula is believed to be a primary area for processing interoceptive sensations, where the posterior insula is the initial receiver of signals from so-called homeostatic small diameter afferents (Aδ and C) signaling nociception, temperature, and sensations from inner organs (Craig, 2002). It is further suggested that these ‘objective’ signals are relayed to the anterior insular cortex, where they are integrated with emotional, cognitive, and motivational signals from other cortical and subcortical regions, to generate subjective feeling states and ultimately self-awareness (Craig, 2009).
A wide variety of multisensory stimuli and cognitive tasks have been found to engage the insula (Craig, 2009). This multimodal behavior deems the insula particularly challenging to probe in relation to specific stimuli. Several studies have reported that the insula is activated by innocuous touch delivered to both glabrous and hairy skin (see example references in Section 1.2.2 of this Chapter). However, whether these insular activations are driven by serial
input from other somatosensory regions, by direct input from e.g., CT afferents, or by processes related to e.g., attention and/or salience is not clear.
A reverse approach to unraveling the functions of the insula is to map the sensory manifestations that are induced when electrically stimulating this region, which has been performed during intracranial depth electrode investigations of patients who are candidates for epilepsy surgery. Non- painful paresthesiae (often described as ‘tingling’ or ‘electric sensations’), painful paresthesiae (often described as ‘pins and needles’), painful electric sensations (described as ‘electric discharge’), diffuse visceral sensations, warmth sensations, auditory responses, and gustatory sensations are among the reported responses to electrical stimulation of the insula (Ostrowsky et al., 2000; Isnard et al., 2004; Afif et al., 2010; Pugnaghi et al., 2011). These studies have not established a unanimous somatotopical or functional organization within the insula (cf. Penfield and Rasmussen, 1950).
Furthermore, when stimulating the electrodes in the insula in the aforementioned studies, not all locations generated a perceptual response in the patients.
Activity across the brain in response to skin stroking 1.2.2.3
In addition to S1, S2, and posterior insula, a number of brain regions have been implicated in the processing of gentle moving touch on the hairy skin, such as the orbitofrontal cortex (McCabe et al., 2008; McGlone et al., 2012;
Voos et al., 2013), cingulate regions (Gordon et al., 2013; Case et al., 2016), and the superior temporal sulcus (Gordon et al., 2013; Voos et al., 2013;
Davidovic et al., 2016), however, activations of these regions is not seen consistently (cf. Morrison, 2016).
1.3 Time-resolved neuroimaging
Electroencephalography (EEG), magnetoencephalography (MEG), and stereotactic EEG (stereo-EEG or SEEG), are neurophysiological methods that are able to track neural signaling on a millisecond time-scale. For the purpose of neuroscience research, sampling rates of 0.5 – 2 kHz are usually employed to provide sufficient resolution in time and frequency, although higher sampling frequencies can of course be used if needed. As an alternative to EEG/MEG, the most commonly used imaging method in human neuroscience is fMRI. Although fMRI has good spatial resolution (~3 mm at 3T, closer to 1 mm at 7T), the temporal resolution is low (~1 s). The low temporal resolution is due to the slow evolution of the BOLD response, which is delayed by several seconds in relation to the neuronal activity that
give rise to the hemodynamic change (Logothetis et al., 2001). As such, fMRI is not ideal to temporally distinguish between brain responses that are instantly evoked by peripheral afference and that which comes later.
Although the fMRI method is constantly evolving, and new techniques are promising to achieve higher sampling rates of the fMRI volumes (e.g., Lin et al., 2010; Lin et al., 2014), the slowness of the BOLD response is inherent and cannot be influenced by method development.
EEG, MEG, and SEEG share a primary origin of the measured signals, which is the summated electrical postsynaptic activity generated by dendrites of pyramidal neurons in the cortex (Buzsaki et al., 2012). However, the three methods have different properties when it comes to e.g., spatial sensitivity and coverage, bandwidth, noise sources, and general accessibility (Nunez and Srinivasan, 2006; Hansen et al., 2010; Buzsaki et al., 2012). Thus, these methods are not redundant. Although they are commonly used alone, more information can be gained by combining the methods, exploiting the advantages of one method to overcome the limitations of the other.
1.3.1 Signal source
Pyramidal cells in the cerebral cortex are densely packed with their apical dendrites aligned parallel to one another and perpendicular to the cortical surface. For an excitatory postsynaptic potential, a depolarization at the distal apical dendrite results in an inward flow of positive ions into the cell, which causes an active sink (negative charge outside of the cell) at the level of the synapse and a passive source (positive charge outside of the cell) at the level of the soma. Conversely, for an inhibitory postsynaptic potential, a hyperpolarization close to the soma causes an active source at the level of the synapse and a passive sink at the distal apical dendrite. The sink-source combination can be modeled as a dipole (Figure 2a), where MEG is primarily sensitive to the magnetic field generated by intracellular primary currents (Baillet, 2017). EEG, on the other hand, detects the currents that are volume conducted in the extracellular media (Nunez and Srinivasan, 2006). The magnitude of the dipole from one single neuron is too small to be detected by means of EEG and/or MEG and it is estimated that around 50,000 neighboring cells, corresponding to a cortical of patch of ~0.6 mm2, need to be synchronously active in order to produce a net dipole (with an associated magnetic field) strong enough to be detected by MEG (Murakami and Okada, 2006; Lopes da Silva, 2010). For EEG, estimating the size of the patch of cortex that need to be synchronously active to produce a detectable signal at the scalp level is not straightforward (Burle et al., 2015). However, to provide an indication, it is approximated that at least 6 cm2 of cortex need to be active
in synchrony to produce a signal strong enough to detect at the scalp with EEG (Nunez and Srinivasan, 2006).
Figure 2. . Signal source and spatial sensitivity. Panel a shows a summed dipole current (orange arrow) generated by postsynaptic potentials in the dendrites of neighboring pyramidal cells. Panel c illustrates the magnetic field (B) that circulates perpendicular to the dipole current (I). Panel b shows the orientation of radial and tangential dipole sources. EEG is more sensitive to signals coming from cortical gyri, i.e. radial or orthogonal sources, whereas MEG is more sensitive to tangential sources. Panel d shows an SEEG electrode with six recording contacts (black dots) that are located within the brain tissue. SEEG electrodes register local field potentials relative to a reference electrode (usually located in the white matter).
1.3.2 Electroencephalography
EEG is the measurement of voltage differences between a recording electrode on the scalp and a reference. The EEG signals are very weak, in the range of a few to a hundred microvolts (µV). In order to record these weak signals, strong amplification is needed, which in turn makes EEG even more sensitive to environmental and physiological noise. Environmental noise comes typically from power lines, which results in a spectral peak at 50 Hz (60 Hz in, e.g., the Americas) and harmonics in the recording. Physiological noise is all electrophysiological activity that is not derived from the brain, e.g., electromyogram (EMG), electrooculogram (EOG), and electrocardiogram (ECG). To reduce the physiological noise, when possible, subjects are usually asked to be as still and relaxed as possible, to avoid excessive EMG activity. Finally, EEG signals are normally processed offline to further reduce noise, e.g., by filtering.
A simplified description of the spatial sensitivity of EEG is that it is ‘sees’
activity from the superficial parts of the brain (for a comprehensive background about the biophysics of EEG see Nunez and Srinivasan, 2006).
Dipole sources that have a radial orientation to the skull are more optimal for EEG to pick up (Figure 2b), but EEG does detect signals from sources oriented tangential to the skull if they are superficial. The signals spread like ripples away from the signal source. Thus, the signals that are detected by the on-scalp electrodes are a mix of multiple signals from close and distant sources. Moreover, the signals have to propagate through different tissues (meninges, cerebrospinal fluid, skull, and scalp) in order to be detected at the scalp level. The different tissues have different physical characteristics (namely, dielectric properties) that distort the signal. The combination of superposition of multiple signals, signal distortion by volume conduction, and noise, make the accurate reconstruction of signal sources a so-called ill- posed problem with an infinite number of solutions. This ill-posed problem is called the inverse problem, which is the challenge to trace back the source(s) that fit the measured signal. However, by making certain assumptions (e.g., about the underlying sources), thus restricting the number of solutions, there are ways to obtain acceptable source estimates of neural activity in EEG recordings (Pascual-Marqui, 1999; Grech et al., 2008).
A significant advantage with EEG is that it is relatively cheap (purchase price for a high-density EEG system is ~25000 €) and accessible.
1.3.3 Magnetoencephalography
The MEG method, invented by physicist David Cohen (1972), relies on the fact that all electrical currents generate a magnetic field, B, which is measured in teslas (for a recent review of the method, see Baillet (2017), and for a comprehensive introduction, Hansen et al. (2010)). The magnitude of the magnetic fields due to brain activity is measured on the scale of femtoteslas (fT, 10-15 T), which is about 10 to 100 million times smaller than the Earth’s magnetic field. The magnetic field of a current dipole circulates around, and is perpendicular to, the axis of the direction of the current flow (right hand rule), which explains some of the differences in spatial sensitivity between EEG and MEG (Figure 2c). A current dipole that is oriented orthogonally to the skull (i.e., a radial source) produces an extremely weak magnetic field outside the head. However, a dipole that has a tangential orientation to the skull produces relatively large radial magnetic fields (i.e., pointing into/out of the scalp), which are ideal for MEG sensors to pick up (Figure 2b). Magnetic fields travel through the air, which means that MEG signals can be picked up at some distance away from the head. It is however desirable to keep the distance between the MEG sensors and the brain as short as possible, since the magnetic field decays drastically as a function of the distance to the signal source. An advantage of MEG over EEG is that the magnetic field does not get distorted due to the different dielectric properties of the tissues in the head (the magnetic susceptibility of biological tissues is roughly the same as that of air/vacuum). Although this advantage of MEG over EEG does not eliminate the inverse problem, it means that fewer assumptions need to be accounted for when doing source reconstruction, resulting in more accurate estimates of the true sources of the signals (e.g., Hämäläinen and Ilmoniemi, 1994). MEG also exhibits better signal-to-noise- ratio (SNR) than EEG in the higher frequencies, due to that the magnetic field is less distorted when it passes through different tissues (Cohen, 2014).
Furthermore, compared to EEG, ocular and muscular artifacts are more readily distinguished in the sensor topography in MEG, which makes it easier to separate high-frequency brain components from physiological noise (Baillet, 2017).
Contrary to EEG, MEG is a reference free recording. As such, interpretation of magnetic field topography is more straightforward compared to EEG electric fields (Nunez and Srinivasan, 2006).
Modern whole-head MEG systems use superconducting quantum interference device (SQUID) technology for magnetic sensing. SQUIDs become superconducting when cooled below a critical temperature (Tc), which
depends on the material, typically below 20 Kelvin (-253°C). A common cooling agent is liquid helium (boiling point 4.2 K or -269°C). The cooling of the SQUIDs requires that the MEG-sensors be contained within an insulted helium tank.
There are three common types of MEG sensors in conventional MEG systems: magnetometers, and axial and planar gradiometers. The simplest sensor type is the magnetometer, which consists of a pickup coil, or flux transformer, that is coupled to a SQUID. The pickup coil is a loop that integrates the magnetic field from the brain and transfers it as magnetic flux to the SQUID. A gradiometer consists of two pickup coils that measure the difference in magnetic flux between two positions, which is a way of cancelling out distant signals (i.e., noise sources). An axial gradiometer measures the difference in magnetic flux in the axial direction, whereas a planar gradiometer measures the difference in magnetic flux in two positions on the same plane.
Since the magnetic field from the brain is very weak, and because of the extreme sensitivity of the MEG sensors, MEG recordings need to be carried out in magnetically shielded rooms. Furthermore, all accessory equipment, e.g., stimulation apparatus, wiring, etc, has to be non-magnetic, since any moving metal close to the MEG sensors causes significant interference to the recording. People with metal implants in their body are thus not ideal for participating in MEG studies. However, some signal artifacts due to noise caused by implants can be removed in the offline processing.
Installing a MEG system, together with a magnetically shielded room is very expensive. Furthermore, the running costs, and liquid helium refills are not insignificant. As of now, there is only one conventional low-Tc MEG machine in Sweden, installed at the Karolinska Institute, Solna. However, new MEG technology such as optically pumped magnetometers (Boto et al., 2017), and high-Tc SQUIDS (Öisjöen et al., 2012) are promising alternatives to conventional low-Tc sensors in making MEG more available. Except increased availability and lower cost, a main justification for developing MEG sensors that operate at higher temperatures is that it enables the sensors to come closer to the head surface, which increases SNR and spatial resolution (Riaz et al., 2017; Xie et al., 2017).
1.3.4 Stereotactic EEG
The SEEG method was developed in the 1960s by the French team of Jean Talairach and Jean Bancaud (Talairach et al., 1962). SEEG has been used for
preoperative epilepsy evaluations for decades, mainly in France and Italy (for reviews and methodological developement see; Kahane et al., 2003; Cossu et al., 2005; Cardinale et al., 2013).
Human SEEG recordings are carried out strictly in patients that suffer from drug-resistant focal epilepsy, who are surgical candidates, when non-invasive investigations fail to identify the epileptogenic zone. Based on rigorous information about the patient’s clinical symptoms, seizure semiology, and functional and anatomical imaging, intracerebral depth electrodes can be placed in deep brain structures with high precision, and used to record seizure onset and spread. SEEG is advantageous for examining deep brain sources (Enatsu et al., 2015), e.g., the insula, operculum, and medial structures. Each depth electrode has several recording points (‘contacts’) along its axis (Figure 2d). A neurosurgeon uses 3D navigation equipment for electrode placement.
For the implantation, magnetic resonance imaging (MRI) volumes and computed tomography angiograms are used. After electrode placement, another computed tomography scan of the brain is carried out for the sake of reconstructing the exact location of each recording contact.
Like the regular on-scalp EEG, SEEG signals are sampled relative to a reference. The reference electrode used during data acquisition is often located in the white matter. Subsequently, offline processing of the SEEG signals enables re-referencing to various montage schemes.
Since SEEG electrodes are placed within the brain tissue, the spatial precision is essentially as high as the accuracy at which the electrodes can be placed (typically ~1 mm). In addition to summated postsynaptic activity, neuronal spike activity may also contribute to the SEEG signal (Ray and Maunsell, 2011). Intracranial recordings also exhibit higher signal amplitudes, and better resolution in higher frequencies, compared to extracranial recordings (Buzsaki et al., 2012). Furthermore, the spectral bandwidth of muscle activity overlaps high-frequency neural activity (~20 – 300 Hz), and it may be difficult to distinguish brain activity from EMG noise with extracranial methods (Muthukumaraswamy, 2013). Thus, high- frequency oscillations can be recorded more reliably with SEEG, compared to EEG/MEG.
SEEG has some obvious limitations, namely that it requires brain surgery and that it is only performed in patients with severe epilepsy where the cortical networks, outside the epileptic network, may also be disturbed. Another drawback with the SEEG method is that it suffers from sparse spatial sampling since not all portions of the brain can be investigated in a single
patient. A way of reducing this shortcoming in experimental neuroscience is to retrieve recordings from a large number of patients, thus achieving a larger cortical coverage (see, e.g., Avanzini et al., 2016). However, this approach is probably only feasible in studies when the experimental manipulation is very simple to perform, since the numbers of patients that undergo SEEG are few.
Thus data collection is deemed to take several years, requiring long-term planning and commitment from both clinicians and researchers.
Despite the clear limitations with SEEG, i.e. invasiveness, that it is only performed in patients with severe epilepsy, and sparse spatial sampling, it is a valuable method to retrieve human in vivo recordings of local field potentials.
As such, it provides unique information about the neurophysiological sources on which almost all other neuroimaging modalities rely (i.e., EEG, MEG, and fMRI (via neurovascular coupling)) (Logothetis et al., 2001; Buzsaki et al., 2012; Baillet, 2017; Cohen, 2017).
1.3.5 Studying the brain in time
Recordings of spontaneous EEG/MEG during rest is commonly used for clinical purposes to investigate, e.g., epilepsy or sleep disorders (Rowan and Tolunsky, 2003). In experimental neuroscience it is common to record brain activity in participants while they are performing a task or during some sensory stimulation. As mentioned previously, the EEG signal is weak, therefore a task or stimulation is often repeated many times, to be able to average the brain response over multiple stimulation trials, in order to increase the SNR. The method of such simple signal averaging is called event-related potentials (ERPs) or evoked potentials (Luck and Kappenman, 2011). In MEG, the corresponding approach is called event-related fields (ERFs). Only the signals that are both time- and phase-locked to the stimulus onset survive averaging, all others (i.e., ‘noise’) are reduced. This means that brain signaling that is evoked by the stimulation, but not repeatedly and perfectly aligned in time to the onset of the stimuli is cancelled out in the event-related average.
Brain activity that is time-locked, but not necessarily phase-locked to the stimulus is called induced activity. The induced activity can be studied by spectral analyses, e.g., Fourier transforms, and are also called event-related spectral perturbations (Pfurtscheller and Lopes Da Silva, 1999). Analysis of the oscillatory responses to sensory stimuli gives information of both where in the brain there are changes in activity and what these changes are. The next section of this chapter contains a brief introduction to how sensory stimuli modulate the rhythms of the brain.
1.3.6 Modulation of brain rhythms by sensory stimulation
Hans Berger (1929) made the first description of the brain’s oscillatory changes to sensory input, when he reported about the modulations in occipital alpha and beta frequencies during periods of open and closed eyes. The brain rhythms have since then been, somewhat arbitrarily, partitioned into the frequency ranges: delta, 0.5 – 4 Hz; theta, 4 – 8 Hz; alpha 8 – 12 Hz; beta 12 – 30 Hz; gamma, >30 Hz (Buzsaki, 2006).
In addition to occipital alpha during awake states with eyes closed, high amplitude oscillations between 8 – 12 Hz have been described over central regions (mu) and temporal (tau) brain regions. Movement and tactile stimulation attenuates central (also called ‘Rolandic’ or ‘sensorimotor’) alpha/mu (Pfurtscheller and Lopes Da Silva, 1999), and auditory stimulation similarly attenuates alpha/tau in the temporal lobe (Lehtelä et al., 1997).
Decrease of spectral power, in response to an event like those described above, is called event-related desynchronization (ERD), whereas an increase of power is called event-related synchronization (ERS) (Pfurtscheller and Lopes Da Silva, 1999). Alpha ERS in primary sensory cortices are believed to reflect a deactivated or ‘actively inhibited’ state within that region, where bottom up information from the thalamus to the primary sensory cortex is absent. Alpha ERD, on the other hand, is suggested to be a correlate of increased excitability within the relevant sensory region (Neuper and Pfurtscheller, 2001).
Central/sensorimotor beta oscillations are slightly different compared to central mu. Beta ERD is seen during tactile stimulation or movement, but when the tactile stimulus or movement is abolished, there is often a transient overshoot in beta power compared to baseline. This phenomenon is called beta rebound and is believed to play a role in cortical inhibition (Cheyne, 2013).
Tactile mechanical stimulation (Cheyne et al., 2003), median nerve stimulation (Salenius et al., 1997), tactile attention (van Ede et al., 2010), active (Neuper and Pfurtscheller, 2001) and passive movement (Parkkonen et al., 2015), and motor imagery (Neuper and Pfurtscheller, 2001) are known to modulate sensorimotor mu and beta oscillations. Perturbations of mu and beta rhythms are observed in both primary motor (M1) and S1 cortices in the aforementioned experimental conditions, thus it makes sense to group these phenomena and call them ‘sensorimotor’. There are, however, indications that mu and beta oscillations in M1 and S1 are functionally distinct. For
