YNIMG-12323; No. of pages: 6; 4C: 3, 4, 5 NeuroImage xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

NeuroImage journal homepage: www.elsevier.com/locate/ynimg

3Q2

Carsten M. Klingner a,b,⁎, Stefan Brodoehl a, Ralf Huonker b, Theresa Götz b, Lydia Baumann b, Otto W. Witte a,b

4 5

a

6

a r t i c l e

7 8 9 10

Article history: Received 26 January 2015 Accepted 9 June 2015 Available online xxxx

11 12 13 14 15 16

Keywords: MEG DCM Somatosensory cortex Effective connectivity Perception

i n f o

R O

Hans Berger Department of Neurology, University Hospital, Jena, Germany Biomagnetic Center, University Hospital, Jena, Germany

a b s t r a c t

E

D

P

The advent of methods to investigate network dynamics has led to discussion of whether somatosensory inputs are processed in serial or in parallel. Both hypotheses are supported by DCM analyses of fMRI studies. In the present study, we revisited this controversy using DCM on magnetoencephalographic (MEG) data during somatosensory stimulation. Bayesian model comparison was used to allow for direct inference on the processing stream. Additionally we varied the duration of the time-window of analyzed data after the somatosensory stimulus. This approach allowed us to explore time dependent changes in the processing stream of somatosensory information and to evaluate the consistency of results. We found that models favoring a parallel processing route best describe neural activities elicited by somatosensory stimuli. This result was consistent for different time-windows. Although it is assumed that the majority of somatosensory information is delivered to the SI, the current results indicate that at least a small part of somatosensory information is delivered in parallel to the SII. These findings emphasize the importance of data analysis with high temporal resolution. © 2015 Published by Elsevier Inc.

T

b

O

F

2

Parallel processing of somatosensory information: Evidence from dynamic causal modeling of MEG data

1Q1

28

C

32 30 29

E

31

Introduction

34 35

A major question in neuroscience is how inputs are processed in the brain and alter the actual brain state. An important piece of information needed for modeling and understanding these processes is where inputs enter the cortical brain matrix and whether data are processed in parallel or serial. Although this is a seemingly simple question, it is not entirely clear in the case of somatosensory information processing. There are two major conflicting theories regarding the somatosensory network: a serial and a parallel pathway theory (Rowe et al., 1996). The serial pathway theory proposes that somatosensory inputs project from the thalamus to the primary somatosensory cortex (SI) before being relayed to the secondary somatosensory cortex (SII). The parallel pathway theory proposes that somatosensory inputs project from the thalamus directly to both the SI and SII. Both theories are supported by anatomical studies demonstrating that SI is connected to SII via reciprocal cortico-cortical connections (Jones and Powell, 1969) and also that different thalamic nuclei project in parallel to SI and SII (Almeida et al., 2004). Within the thalamus, the ventroposterolateral nucleus (VPL) and the ventroposteromedial nucleus (VPM) are recognized as the main somatosensory relays. Both nuclei are involved in transmitting

42 43 44 45 46 47 48 49 50 51 52

Q3

R

N C O

40 41

U

38 39

R

33

36 37

17 18 19 20 21 22 23 24 25 26 27

⁎ Corresponding author at: Hans Berger Department of Neurology, University Hospital, Jena Friedrich Schiller University, Erlanger Allee 101, D 07747 Jena, Germany. Fax: +49 3641 9323402. E-mail address: [email protected] (C.M. Klingner).

innocuous and nocuous stimulus information to the cortex and are tightly connected to the SI and weakly connected to the SII (Burton and Jones, 1976; Friedman and Murray, 1986; Friedman et al., 1986; Jones et al., 1979). Therefore, pain and tactile information can be conveyed to SII directly from the thalamus but also by an indirect serial pathway via SI. Dynamic causal modeling (DCM) is a method that allows estimating and making inferences about the network dynamic and the coupling among small numbers of brain areas in a Bayesian framework (Friston et al., 2003). Studies using DCM on fMRI data have reached different conclusions regarding the two processing theories. Two studies have found evidence for the serial processing theory (Kalberlah et al., 2013; Khoshnejad et al., 2014), and another two studies found evidence for the parallel processing theory (Chung et al., 2014; Liang et al., 2011). In the present study, we revisited the controversy of serial or parallel relaying of somatosensory information in humans using DCM on magnetoencephalographic data. We tested whether the input to SII is relayed by SI (serial model), or whether a parallel input to SII is indeed necessary to explain the cortical responses (parallel model). The primary advantage of MEG over fMRI is its higher temporal resolution. This increased resolution enables us to not only compare a serial and a parallel model but also allow for analysis dependent on the elapsed time from the somatosensory stimulus. Early cortical responses to median nerve stimuli peak at approximately 20 ms in the SI (Hari and Forss, 1999). Given a certain inter-individual variance and additional time for somatosensory information to be transmitted to the SII, we started our

http://dx.doi.org/10.1016/j.neuroimage.2015.06.028 1053-8119/© 2015 Published by Elsevier Inc.

Please cite this article as: Klingner, C.M., et al., Parallel processing of somatosensory information: Evidence from dynamic causal modeling of MEG data, NeuroImage (2015), http://dx.doi.org/10.1016/j.neuroimage.2015.06.028

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78

84

89 90

Sixteen healthy volunteers without any history of neurological or psychiatric diseases participated in this study (mean age 24.3 ± 2.4 years, 9 female, 7 male). All of the subjects were right-handed according to the Edinburgh Handedness Inventory (Oldfield, 1971). This study was approved by the local ethics committee, and all subjects gave their written, informed consent in accordance with the Human Subjects Guidelines of the Declaration of Helsinki.

91

MEG recordings

92

97

Magnetic fields (MEG) were recorded with a 306-channel helmetshaped neuromagnetometer (Vectorview, Elekta Neuromag Oy, Helsinki, Finland). MEG data were sampled at 2 kHz, following a low-pass filter at 1660 Hz and a high-pass filter at 0.1 Hz. A 3D Digitizer (3SPACE FASTRAK, Polhemus Inc., Colchester, VT, USA) was used to locate anatomical locations (nasion and preauricular points).

98

Experimental protocol

99

108

An electrical median nerve stimulus consisting of a monophasic square wave pulse (200 μs in duration) generated by a clinical neurostimulator (Digitimer Constant Current Stimulator model DS7A) was unilaterally applied at the right wrist. The interstimulus interval was randomized between 700 and 1200 ms. The current amplitude was adjusted individually according to the recommendations of the International Federation for Clinical Neurophysiology IFCN at a motor plus sensory threshold (Mauguiere, 1999). Initially, 100 stimuli were applied and further used as a spatial localizer. These data were not used for the DCM analysis. A total of 561 additional stimuli were delivered.

109

MEG data analysis

110 111

119 120

Raw MEG data were filtered with Maxfilter Version 2.0.21 (Elekta Neuromag Oy. Finland) using the Signal Space Separation (SSS) method (Taulu and Simola, 2006). The MEG data analysis was performed on a workstation using MATLAB (Mathworks, Natick, MA, USA) and SPM12 software (Wellcome Department of Cognitive Neurology, London, UK, http://www.fil.ion.ucl.ac.uk/spm). The data from all sensors were first high-pass filtered at 0.5 Hz and then low-pass filtered at 40 Hz. Eyemovement artifacts were removed after visual inspection. The data were then downsampled to 200 Hz and epoched from a 50 ms prestimulus onset to a 500 ms poststimulus onset. The prestimulus time window (−50 to 0 ms) was used for baseline correction.

121

Localizer preprocessing and dipole fitting

122

The localizer data (100 events) were epoched into segments ranging between − 50 and 110 ms relative to stimulus onset and baseline corrected to the prestimulus period. Data were averaged across trials using robust averaging. Source reconstruction was performed by using variational Bayes equivalent current dipoles (VB-ECD) as implemented in SPM 12 and described by Kiebel et al. (2008). While distributed reconstruction methods consider all possible source locations simultaneously, allowing for large and widely spread clusters of activity, VBECD in contrast relies on the hypothesis that only a few sources are active simultaneously and that these sources are focal. The number of ECDs considered in the model must be defined a priori. The VB-ECD method was chosen because these requirements fit well with our

112 113 114 115 116 117 118

123 124 125 126 127 128 129 130 131 132 133

C

106 107

E

104 105

R

102 103

R

100 101

O

95 96

C

93 94

N

87 88

U

85 86

Model specification

150

F

Subjects

O

83

R O

Materials and methods

134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149

With this DCM analysis, we aimed to clarify whether somatosensory stimulus information may directly enter SII (Chung et al., 2014; Liang et al., 2011) or is transmitted serially from SI to SII (Kalberlah et al., 2013; Khoshnejad et al., 2014). We used DCM for evoked responses (David et al., 2006) and sensory evoked potentials (SEP) as a neuronal model. The SEP model uses a variant of the standard ERP model to capture faster dynamics of neuronal populations (Marreiros et al., 2008). Previous studies have employed models of different complexities (SIc–SIIc; SIc–SIIc–SIIi). Here, we primarily analyzed a model with two nodes (SIc–SIIc) because it is not entirely clear whether responses in SIIi occur in the time window of 100 ms (Hagiwara et al., 2010) and whether such responses could project backward within 100 ms to influence the processing of information in SIIc. However, we nevertheless used an additional model including an SIIi node (3 node model) for two reasons. If the SIIi node does not affect the processing between SIc and SIIc, it should not affect the results of whether inputs are processed in series or parallel. Moreover, the use of two different models (2 nodes and 3 nodes) ensures the reproducibility of results. In the three-node model, the SIIi is connected to the SIIc by bilateral connections. In accordance with evidence from a number of previous studies, bidirectional connections were defined between all nodes as well as selfconnections for all nodes. Models that differed on whether stimulus information directly enters the SII-node or not were compared. In model S, only the SI node received stimulus input (serial processing). In model P, both the SI node and the SII node receive stimulus information (parallel processing). Because the data have only one experimental condition (“stimulus”), the A matrix and the B matrix will have the same interpretation, i.e. the strength of effective connectivity given stimulation. Therefore, we analyzed the A matrix only (not the B matrix) and did not vary the modulation of the connections across models. The models were specified and estimated using the DCM toolbox for SPM (SPM 12 release 6225). The anatomical source locations of the contralateral SI and bilateral SII were selected on the basis of the individual dipole locations of the localizer data. Fig. 1 shows the locations of these sources in Montreal Neurological Institute (MNI) coordinates.

151 152

Bayesian model selection and comparison

186

For each model, DCM computed statistical evidence using the implemented Bayesian scheme. Bayesian model selection (BMS) in DCM was achieved by a free energy approximation to the log evidence of each model in terms of model fit and complexity (Friston et al., 2003). Based on the estimated model evidence, Bayesian model selection (BMS) calculates the probability of each model being more likely than any other tested model (Penny et al., 2010; Stephan et al., 2010). BMS was performed for 4 data segments separately, all starting at 1 ms and with lengths ranging from 40 ms to 100 ms by steps of

187 188

P

82

study. The dipoles were fitted iteratively 10 times at each step, selecting the dipole model with maximum negative free energy (Kiebel et al., 2008). The following procedure was run per participant: (1) a single dipole with uninformative prior was fitted to data at the individual N20 latency (18–22 ms after stimulus on average) (SIc); (2) a second dipole was added to the optimized SI dipole and fitted to data within the 60–80 ms time window within the left hemisphere (SIIc); and (3) a third dipole was added to the optimized SI + SII dipole and fitted to data within the 80–100 ms time window within the right hemisphere (Hagiwara et al., 2010) (for similar approaches see also Helbling et al., 2015; Woodhead et al., 2014). Representative dipoles were identified for contralateral SI and bilateral SII on the basis of these three time windows within the anatomically correct locations of a given region (determined using cytoarchitectonic probabilistic maps from SPM Anatomy Toolbox) (Eickhoff et al., 2005). The source localization was performed by using the template T1 image delivered by SPM12 software.

T

81

DCM modeling with a time window of 40 ms after somatosensory stimulation and incrementally increased this time window up to 100 ms (Auksztulewicz et al., 2012).

D

79 80

C.M. Klingner et al. / NeuroImage xxx (2015) xxx–xxx

E

2

Please cite this article as: Klingner, C.M., et al., Parallel processing of somatosensory information: Evidence from dynamic causal modeling of MEG data, NeuroImage (2015), http://dx.doi.org/10.1016/j.neuroimage.2015.06.028

153 154 155 Q4 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185

189 190 191 192 193 194 195

3

O

F

C.M. Klingner et al. / NeuroImage xxx (2015) xxx–xxx

218

Event-related potentials—localizer

219

231

The MEG response from the localizer (n = 16) showed expected somatosensory evoked potential components, including a clearly pronounced centroparietal N20 (mean amplitude relative to baseline, −65 fT; p b 0.001), followed by P30, P40 and P60 components in each subject (Fig. 1). Individual N20 component latencies were used to fit a single dipole in cSI [average MNI location (− 47, − 20, 51), SD (5, 9, 9)]. Adding to this dipole, a dipole in cSII [average MNI location (−51, −20, 19), SD (9, 8, 9)] was fitted to the 60–80 ms time window. A dipole in iSII [average MNI location (48, −23, 22), SD (11, 9, 9)] was then added. These three sources corresponded to the hand area of the primary somatosensory cortex located in the postcentral gyrus and the bilateral secondary somatosensory area within the parietal operculum (Eickhoff et al., 2005).

232

DCM model comparison and connectivity parameters

233 234

Random-effect Bayesian model comparison (n = 16) was applied to the two models for different time intervals. For the model with two nodes (SIc and SIIc), model P (characterized by input to the SI and SII

212 213 214 215

220 221 222 223 224 225 226 227 228 229 230

235

C

E

210 211

R

208 209

R

206 207

N C O

204 205

U

202 203

P

Results

200 201

D

217

198 199

nodes) clearly obtained the highest EP value for all time intervals (Fig. 2). The same analysis of our three-nodes model (SIc, SIIc and SIIi) revealed similar results with stronger EP values for model P (parallel input) in all time intervals (Fig. 3). Comparing the strength of the inputs, we found a significant difference only for the time interval of 1–100 ms with a stronger input in SI compared to SII (Fig. 4). By testing for

T

216

20 ms (Auksztulewicz et al., 2012; Murata et al., 2005). We did not use a sliding window technique (e.g., 40–80 ms) to ensure that the somatosensory input is always included in the analysis and to exclude the possibility that we analyzed only SI–SII feedback loops. However, the drawback of this approach was that subsequent time windows included data points from previous time windows. Changes in the processing of information would be detected only if their amount became relevant compared to the information processing of the whole time window. This approach limited our ability to detect processing changes in later stages of the corresponding time window. We reported the exceedance probability (EP) for each tested model. The EP yields a measure for the likelihood of one model being more prevalent than the others (Penny et al., 2010). BMS using a random effect (RFX) was performed to determine the best model for each data segment. To test whether there was a difference in the strength between inputs (SI/SII), we performed Bayesian model averaging (BMA) within the winning model and entered the subject-specific BMA parameter estimates of the inputs (C matrix) into a paired t-test. We further tested for significant effects of the effective connectivity (A matrix) within the winning models by performing a one-sample t-test on the connectivity estimates.

E

196 197

R O

Fig. 1. Localizer ERP and resulting dipoles. Individual N20 component latencies were used to fit a single dipole in cSI [average MNI location (−47, −20, 51), SD (5, 9, 9)]. Adding to this dipole, a dipole in cSII [average MNI location (−51, −20, 19), SD (9, 8, 9)] was fitted to the 60–80 ms time window. Another dipole in iSII cSII [average MNI location (48, −23, 22), SD (11, 9, 9)] was then added. Resulting average dipoles are plotted on a single-subject template. In the lower middle part of the image, the sensor outline is shown. The signal time-course of a sensor located in the proximity of the cSI (channel 45, Neuromag) is shown in the upper middle part of the image (unfiltered, epoched and averaged data of a single subject).

Fig. 2. The results of the Bayesian model selection. The upper left part of the image shows the model structures of model S (left—serial processing) and model P (parallel processing). The lower left part of the image shows the exceedance probabilities of both models for each of the four time windows. For each of these time windows, model P showed a higher exceedance probability than model S.

Please cite this article as: Klingner, C.M., et al., Parallel processing of somatosensory information: Evidence from dynamic causal modeling of MEG data, NeuroImage (2015), http://dx.doi.org/10.1016/j.neuroimage.2015.06.028

236 237 238 239 240 241

C.M. Klingner et al. / NeuroImage xxx (2015) xxx–xxx

P

R O

O

F

4

242

E

C

T

Fig. 3. The results of the Bayesian model selection for the model with three nodes (SIc, SIIc and SIIi). The upper left part of the image shows the model structure of model S (left—serial processing) and model P (parallel processing). The lower left part of the image shows the exceedance probabilities of both models for each of the four time windows. For each of these time windows, model P showed a higher exceedance probability than model S.

E

D

Fig. 4. The results of the posterior estimates for the winning model (parallel input) at the level of input strength. The input strengths in SI and SII are shown for the four different time windows. The x-axis represents the ratio between the strength of SI and SII input. Positive values represent a stronger input in SI compared to SII. The error bars represent the standard deviation. Significant differences between SI and SII input are marked by an ‘*’ (p b 0.05).

248

Discussion

249 250

In the present study we performed DCM to investigate where input signals enter the cerebral cortex and particularly whether the information is relayed directly to SII or is exclusively relayed serially from SI to SII. Our results are strongly suggestive of a parallel processing pathway with somatosensory information entering directly into SII. It is generally assumed that the main flow of somatosensory information is serially transduced from the thalamus to SI and from SI to SII; however, this does not exclude the coexistence of direct projections from the thalamus to SII. SI is presumed to process and encode the type, location, duration and intensity of somatosensory inputs from the contralateral side of the body (Schnitzler and Ploner, 2000; Zhang et al., 2007) and respond also to the context of touch (Gazzola et al., 2012). The preprocessed information is then transmitted to the SII as the next higher cortical area. However, our results suggest additional direct input to SII. This raises the question of why it would be beneficial to have a direct SII input. This question will be discussed in terms of the theory of cortical responses based on predictive coding proposed by Karl Friston (Friston, 2005). The main point of this theory is that evoked cortical responses can be understood as transient expressions

251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267

R

O

C

N

244 245

U

243

R

246 247

significant connection (A matrix), we found several significant connections (Fig. 5). We performed an additional comparison of models using a time window of 500 ms. This time window favored the serial processing of information (model S 78%, model P 22%). Comparing all models with two nodes against all models with three nodes showed that adding a third region does not improves the model fits (Fig. 6).

of prediction errors and that the process of minimizing this prediction error at all levels of the cortical hierarchy corresponds to the recognition of a stimulus (Friston, 2005). In this context, a direct input to SII allows for a more precise estimation of the prediction error at an earlier time point. This prediction error influences the information processing in the whole hierarchical network mediated by forward, lateral and backward connections and ultimately resulting in a faster recognition of the cause of a stimulus. Therefore, a direct input to SII is meaningful only if time matters. Such a time dependence of somatosensory information might be necessary for the quick identification of a stimulus (e.g., the dangerousness of that stimulus). Whether a stimulus is, for example, dangerous or not is not only a matter of intensity but is frequently strongly context-dependent and therefore requires higher-order cognitive resources that cannot be performed by the SI alone but rather within the context of a hierarchical organized network. It is important to note that both models used in the current study are greatly simplified. The hierarchical network involved in inferring the causes of somatosensory input encompasses multiple other brain areas, particularly the insula and the amygdala, which are involved in the multisensory integration and affective evaluation of somatosensory stimuli (Bauer et al., 2014; Tamietto et al., 2015). However, our analyses demonstrated a decreased model fit for models with three nodes compared with two node models in the analyzed time windows. This result is in line with a study investigating the response time of the SIIi, demonstrating that responses in SIIi only begin to occur around 100 ms (Hagiwara et al., 2010). This result further suggests that stimulus induced SIIi responses do not project backward within 100 ms to influence the processing of information in SIIc. Our data suggest that early direct thalamus-SII input might be beneficial. Such an early response in SII after somatosensory stimulation at 20–30 ms (i.e., not later than the earliest peak in SI) has been reported (Karhu and Tesche, 1999) but the majority of studies investigating the response time of SI and SII neurons after somatosensory stimuli have

Please cite this article as: Klingner, C.M., et al., Parallel processing of somatosensory information: Evidence from dynamic causal modeling of MEG data, NeuroImage (2015), http://dx.doi.org/10.1016/j.neuroimage.2015.06.028

268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300

C.M. Klingner et al. / NeuroImage xxx (2015) xxx–xxx

5

316 317 318 319 320

Fig. 6. The results of the Bayesian model selection at the level of model families. We compared the models with two-nodes against the models with three nodes. The image shows the exceedance probabilities of these models for each of the four time windows. For each of these time windows, the two-node models showed a higher exceedance probability than the three-node models.

D

P

R O

also receive and process information from the SI. Given that most information is delivered by the SI to the SII, this means that the parallel processing mode is mainly active in the beginning (~ 100 ms) of a new stimulus and that the same neurons in SII will process a sustaining stimulus mainly by relying on preprocessed information from SI. However, it might also be possible that the relay of information changes within the thalamus. It is well known that thalamic information processing and relaying are constantly adjusted by the sensory experience (Herrero et al., 2002; Temereanca and Simons, 2004; Zhang et al., 2007). This adjustment is primarily mediated by corticothalamic axons from all subdivisions of the SI, which outnumber thalamocortical axons by 10-fold (Guillery, 1967; Liu et al., 1995; Temereanca and Simons, 2004). Therefore, the cortex is able to regulate its own input from the thalamus. Such a time-dependent change in the processing stream would be compatible with our results and particularly supported by the identification of a significantly stronger SI input at the 100 ms time window. The observed transversion of the connections over time also suggests that there is a dynamic change over time. The translation of the modulatory effects of connections shows the pattern of a feedback loop (1–60 ms: SI- N SII; 1–80 ms: SI- N SII; 1–100 ms: SII- N SI). The finding of a significant change in the strength of the connection from SI to SII in the time epoch of 1–60 ms and 1–80 ms fits well to the theory of an increased information flow from SI to SII. The observed changes of the strength of the effective connectivity between SI- N SII (1–60 ms and 1–80 ms) combined with the increased SI input in the 1–100 ms window might suggest that the input from SI into SII increasingly affects (and possibly decreases) the importance of the direct SII input. The finding of a serial processing in the 1–500 ms window seems to support such a transversion in the processing from a parallel to serial mode. However, while in the first 100 ms it can be assumed that SI and SII are mainly processing the applied stimulus it is not clear whether this result characterizes indeed a changed processing of the input stimulus or is rather due to feedback loops or a steady state somatosensory input into SI. Moreover, it remains unclear whether a steady state input is processed in the same way as a new or changing input. However, the question of a possible transversion of the processing mode can be directly investigated only by examining the effects of a sustaining stimulus. Relating this explanation to the available fMRI-DCM analyses would suggest that parallel processing of information would be found only by acquiring data shortly after the stimulus onset. In respect to the limited temporal resolution of fMRI investigations, even minor time differences of the acquisition of the corresponding slices might explain the differences in the recent fMRI-DCM studies and particularly the finding of a serial processing of information. For example, as Khoshnejaed and colleagues administered stimuli during a 500 ms inter-volume delay (Khoshnejad et al., 2014). Their finding of serial information processing might well be explained by our hypothesis of a predominantly parallel input into SII for only ~100 ms after stimulus onset. The second fMRI-

E

314 315

T

312 313

C

310 311

E

308 309

R

306 307

R

304 305

found an earlier response of neurons in SI than those in SII (Gobbele et al., 2003; Inui et al., 2004; Jung et al., 2009; Schnitzler et al., 1999). However, these observations do not necessarily imply a sole serial pathway as an incomplete sampling of the neural activity of SII neurons cannot be ruled out (Mauguiere et al., 1997). In cats and rabbits, SII responsiveness is never abolished and infrequently affected by SI inactivation (Murray et al., 1992; Turman et al., 1992, 1995). Further, tactile inputs to SII traverse a direct path from the thalamus, organized in parallel with SI inputs. The presence of these pathways in monkeys and humans has been questioned (Inui et al., 2004; Khoshnejad et al., 2014). It was demonstrated that tactile responsiveness within the hand area of SII is reduced by surgical ablation of the hand area of the SI area of the cortex in the macaque and marmoset monkeys which was interpreted in favor of a serial processing scheme (Pons et al., 1992). However, in the marmoset, SII responsiveness is preserved but reduced after inactivation of SI (Rowe et al., 1996). Particularly, SII responsiveness was unaffected in 25% of neurons, reduced in 65 % of neurons and abolished in ~10% of SII neurons (Rowe et al., 1996). These results suggest that information that is directly transferred from the thalamus to the SII is mainly processed by neurons that

N C O

302 303

U

301

O

F

Fig. 5. The results of the posterior estimates for the winning model (parallel input) at the levels of A-matrix connections. Arrows represent the connection within the model. Connections marked by a green dot indicate a significant (p b 0.05, uncorrected) effective connectivity (A-matrix) induced by the somatosensory input. All significant connections were found to be positive, which is marked by a plus sign within the green dot.

Please cite this article as: Klingner, C.M., et al., Parallel processing of somatosensory information: Evidence from dynamic causal modeling of MEG data, NeuroImage (2015), http://dx.doi.org/10.1016/j.neuroimage.2015.06.028

321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369

386

Acknowledgment

387 388

The authors wish to thank the reviewers for their helpful comments and insights.

389

References

390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434

Almeida, T.F., Roizenblatt, S., Tufik, S., 2004. Afferent pain pathways: a neuroanatomical review. Brain Res. 1000, 40–56. Auksztulewicz, R., Spitzer, B., Blankenburg, F., 2012. Recurrent neural processing and somatosensory awareness. J. Neurosci. 32, 799–805. Bauer, C.C., Barrios, F.A., Diaz, J.L., 2014. Subjective somatosensory experiences disclosed by focused attention: cortical–hippocampal–insular and amygdala contributions. PLoS ONE 9, e104721. Burton, H., Jones, E.G., 1976. The posterior thalamic region and its cortical projection in New World and Old World monkeys. J. Comp. Neurol. 168, 249–301. Chung, Y.G., Han, S.W., Kim, H.S., Chung, S.C., Park, J.Y., Wallraven, C., Kim, S.P., 2014. Intraand inter-hemispheric effective connectivity in the human somatosensory cortex during pressure stimulation. BMC Neurosci. 15, 43. Eickhoff, S.B., Stephan, K.E., Mohlberg, H., Grefkes, C., Fink, G.R., Amunts, K., Zilles, K., 2005. A new SPM toolbox for combining probabilistic cytoarchitectonic maps and functional imaging data. NeuroImage 25, 1325–1335. Friedman, D.P., Murray, E.A., 1986. Thalamic connectivity of the second somatosensory area and neighboring somatosensory fields of the lateral sulcus of the macaque. J. Comp. Neurol. 252, 348–373. Friedman, D.P., Murray, E.A., O'Neill, J.B., Mishkin, M., 1986. Cortical connections of the somatosensory fields of the lateral sulcus of macaques: evidence for a corticolimbic pathway for touch. J. Comp. Neurol. 252, 323–347. Friston, K., 2005. A theory of cortical responses. Philos. Trans. R. Soc. Lond. B Biol. Sci. 360, 815–836. Friston, K.J., Harrison, L., Penny, W., 2003. Dynamic causal modelling. NeuroImage 19, 1273–1302. Gazzola, V., Spezio, M.L., Etzel, J.A., Castelli, F., Adolphs, R., Keysers, C., 2012. Primary somatosensory cortex discriminates affective significance in social touch. Proc. Natl. Acad. Sci. U. S. A. 109, E1657–E1666. Gobbele, R., Schurmann, M., Forss, N., Juottonen, K., Buchner, H., Hari, R., 2003. Activation of the human posterior parietal and temporoparietal cortices during audiotactile interaction. NeuroImage 20, 503–511. Guillery, R.W., 1967. Patterns of fiber degeneration in the dorsal lateral geniculate nucleus of the cat following lesions in the visual cortex. J. Comp. Neurol. 130, 197–221. Hagiwara, K., Okamoto, T., Shigeto, H., Ogata, K., Somehara, Y., Matsushita, T., Kira, J., Tobimatsu, S., 2010. Oscillatory gamma synchronization binds the primary and secondary somatosensory areas in humans. NeuroImage 51, 412–420. Hari, R., Forss, N., 1999. Magnetoencephalography in the study of human somatosensory cortical processing. Philos. Trans. R. Soc. Lond. B Biol. Sci. 354, 1145–1154. Helbling, S., Teki, S., Callaghan, M.F., Sedley, W., Mohammadi, S., Griffiths, T.D., Weiskopf, N., Barnes, G.R., 2015. Structure predicts function: combining non-invasive electrophysiology with in-vivo histology. NeuroImage 108, 377–385. Herrero, M.T., Barcia, C., Navarro, J.M., 2002. Functional anatomy of thalamus and basal ganglia. Childs Nerv. Syst. 18, 386–404. Inui, K., Wang, X., Tamura, Y., Kaneoke, Y., Kakigi, R., 2004. Serial processing in the human somatosensory system. Cereb. Cortex 14, 851–857.

C

E

R

R

O

C

N

383 384

U

381 382

F

385

In the present study, we investigated the processing of somatosensory stimuli in the somatosensory cortex. Our results strongly favor a parallel rather than a serial processing route for somatosensory stimulus information along the somatosensory pathway. We hypothesize that the parallel pathway exists only at the beginning of a stimulus of functional importance and that preprocessed information from the SI later dominates information processing in SII.

O

379 380

376

R O

Conclusion

374 375

P

378

372 373

Jones, E.G., Powell, T.P., 1969. Connexions of the somatic sensory cortex of the rhesus monkey. II. Contralateral cortical connexions. Brain 92, 717–730. Jones, E.G., Wise, S.P., Coulter, J.D., 1979. Differential thalamic relationships of sensorymotor and parietal cortical fields in monkeys. J. Comp. Neurol. 183, 833–881. Jung, P., Baumgartner, U., Stoeter, P., Treede, R.D., 2009. Structural and functional asymmetry in the human parietal opercular cortex. J. Neurophysiol. 101, 3246–3257. Kalberlah, C., Villringer, A., Pleger, B., 2013. Dynamic causal modeling suggests serial processing of tactile vibratory stimuli in the human somatosensory cortex—an fMRI study. NeuroImage 74, 164–171. Karhu, J., Tesche, C.D., 1999. Simultaneous early processing of sensory input in human primary (SI) and secondary (SII) somatosensory cortices. J. Neurophysiol. 81, 2017–2025. Khoshnejad, M., Piche, M., Saleh, S., Duncan, G., Rainville, P., 2014. Serial processing in primary and secondary somatosensory cortex: a DCM analysis of human fMRI data in response to innocuous and noxious electrical stimulation. Neurosci. Lett. 577, 83–88. Kiebel, S.J., Daunizeau, J., Phillips, C., Friston, K.J., 2008. Variational Bayesian inversion of the equivalent current dipole model in EEG/MEG. NeuroImage 39, 728–741. Liang, M., Mouraux, A., Iannetti, G.D., 2011. Parallel processing of nociceptive and nonnociceptive somatosensory information in the human primary and secondary somatosensory cortices: evidence from dynamic causal modeling of functional magnetic resonance imaging data. J. Neurosci. 31, 8976–8985. Liu, X.B., Honda, C.N., Jones, E.G., 1995. Distribution of four types of synapse on physiologically identified relay neurons in the ventral posterior thalamic nucleus of the cat. J. Comp. Neurol. 352, 69–91. Marreiros, A.C., Daunizeau, J., Kiebel, S.J., Friston, K.J., 2008. Population dynamics: variance and the sigmoid activation function. NeuroImage 42, 147–157. Mauguiere, F., 1999. Clinical neurophysiology and functional neuroimaging at the Societe Francaise de Neurologie (1948–1998). Rev. Neurol. (Paris) 155, 857–868. Mauguiere, F., Merlet, I., Forss, N., Vanni, S., Jousmaki, V., Adeleine, P., Hari, R., 1997. Activation of a distributed somatosensory cortical network in the human brain: a dipole modelling study of magnetic fields evoked by median nerve stimulation. Part II: Effects of stimulus rate, attention and stimulus detection. Electroencephalogr. Clin. Neurophysiol. 104, 290–295. Murata, H., Babasaki, K., Yamada, A., 2005. Highly polymorphic DNA markers to specify strains of the ectomycorrhizal basidiomycete Tricholoma matsutake based on sigmamarY1, the long terminal repeat of gypsy-type retroelement marY1. Mycorrhiza 15, 179–186. Murray, G.M., Zhang, H.Q., Kaye, A.N., Sinnadurai, T., Campbell, D.H., Rowe, M.J., 1992. Parallel processing in rabbit first (SI) and second (SII) somatosensory cortical areas: effects of reversible inactivation by cooling of SI on responses in SII. J. Neurophysiol. 68, 703–710. Oldfield, R.C., 1971. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 9, 97–113. Penny, W.D., Stephan, K.E., Daunizeau, J., Rosa, M.J., Friston, K.J., Schofield, T.M., Leff, A.P., 2010. Comparing families of dynamic causal models. PLoS Comput. Biol. 6, e1000709. Pons, T.P., Garraghty, P.E., Mishkin, M., 1992. Serial and parallel processing of tactual information in somatosensory cortex of rhesus monkeys. J. Neurophysiol. 68, 518–527. Rowe, M.J., Turman, A.B., Murray, G.M., Zhang, H.Q., 1996. Parallel organization of somatosensory cortical areas I and II for tactile processing. Clin. Exp. Pharmacol. Physiol. 23, 931–938. Schnitzler, A., Ploner, M., 2000. Neurophysiology and functional neuroanatomy of pain perception. J. Clin. Neurophysiol. 17, 592–603. Schnitzler, A., Volkmann, J., Enck, P., Frieling, T., Witte, O.W., Freund, H.J., 1999. Different cortical organization of visceral and somatic sensation in humans. Eur. J. Neurosci. 11, 305–315. Stephan, K.E., Penny, W.D., Moran, R.J., den Ouden, H.E., Daunizeau, J., Friston, K.J., 2010. Ten simple rules for dynamic causal modeling. NeuroImage 49, 3099–3109. Tamietto, M., Cauda, F., Celeghin, A., Diano, M., Costa, T., Cossa, F.M., Sacco, K., Duca, S., Geminiani, G.C., de Gelder, B., 2015. Once you feel it, you see it: insula and sensorymotor contribution to visual awareness for fearful bodies in parietal neglect. Cortex 62, 56–72. Taulu, S., Simola, J., 2006. Spatiotemporal signal space separation method for rejecting nearby interference in MEG measurements. Phys. Med. Biol. 51, 1759–1768. Temereanca, S., Simons, D.J., 2004. Functional topography of corticothalamic feedback enhances thalamic spatial response tuning in the somatosensory whisker/barrel system. Neuron 41, 639–651. Turman, A.B., Ferrington, D.G., Ghosh, S., Morley, J.W., Rowe, M.J., 1992. Parallel processing of tactile information in the cerebral cortex of the cat: effect of reversible inactivation of SI on responsiveness of SII neurons. J. Neurophysiol. 67, 411–429. Turman, A.B., Morley, J.W., Zhang, H.Q., Rowe, M.J., 1995. Parallel processing of tactile information in cat cerebral cortex: effect of reversible inactivation of SII on SI responses. J. Neurophysiol. 73, 1063–1075. Woodhead, Z.V., Barnes, G.R., Penny, W., Moran, R., Teki, S., Price, C.J., Leff, A.P., 2014. Reading front to back: MEG evidence for early feedback effects during word recognition. Cereb. Cortex 24, 817–825. Zhang, N., Gore, J.C., Chen, L.M., Avison, M.J., 2007. Dependence of BOLD signal change on tactile stimulus intensity in SI of primates. Magn. Reson. Imaging 25, 784–794.

T

377

DCM study favoring the serial processing theory has administered their stimulus for 1 s with a TR of 2 s. According to our proposed hypothesis, a sustaining stimulus for 1000 ms should therefore result in an out balance of the serial processing scheme. Our results, in conjunction with the available fMRI-DCM studies, strongly suggest at least an initial parallel processing of somatosensory information. However, the hypothesis of a switch between a parallel and a serial processing mode requires further investigation using methods with high temporal resolutions.

D

370 371

C.M. Klingner et al. / NeuroImage xxx (2015) xxx–xxx

E

6

511

Please cite this article as: Klingner, C.M., et al., Parallel processing of somatosensory information: Evidence from dynamic causal modeling of MEG data, NeuroImage (2015), http://dx.doi.org/10.1016/j.neuroimage.2015.06.028

435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510

Parallel processing of somatosensory information: Evidence from dynamic causal modeling of MEG data.

The advent of methods to investigate network dynamics has led to discussion of whether somatosensory inputs are processed in serial or in parallel. Bo...
1MB Sizes 2 Downloads 9 Views