Cingulo-Opercular Control Network Supports Disused Motor Circuits in Standby Mode

Two weeks of upper extremity casting induced plasticity beyond somatomotor regions. Whole-brain resting-state functional MRI (rs-fMRI) revealed that disused motor regions became more strongly connected to the cingulo-opercular network (CON), an executive control network that includes regions of the dorsal anterior cingulate cortex (dACC) and insula. Disuse-driven increases in functional connectivity (FC) were specific to the CON and somatomotor networks and did not involve any other networks, such as the salience, frontoparietal or default mode networks. FC increases during casting were mediated by large, spontaneous activity pulses that appeared in disused motor regions and network-adjacent CON control regions. During limb constraint, disused motor circuits appear to enter a standby mode characterized by spontaneous activity pulses and strengthened connectivity to CON executive control regions.

Three participants (P1, P2 and P3) wore casts covering the entire dominant upper extremity for two weeks. b, Participants were scanned every day for 42-64 consecutive days before, during and after casting. All scans included 30 minutes of resting-state functional MRI (rs-fMRI). c, During the cast period, disused somatomotor circuits exhibited spontaneous activity pulses 30 .

91
The brain's functional network organization can be visualized using spring-embedded graphs, 92 which treat functional connections as spring forces, positioning more strongly connected nodes 93 closer to one another (Fig. 3a, Supplementary Fig. 2). While L-SM1ue FC increases were hemispheres, nearly all of these changes mapped onto a single cluster in network space, the quantified by examining the distribution of the greatest FC increases (top 5%) across the 98 canonical functional networks (Fig. 3b). In all participants, the CON showed more FC 99 increases than expected by chance (P < 0.001 for all participants). Decreases in L-SM1ue FC 100 were also localized in network space (Fig. 3c, Supplementary Fig. 2). Most of the regions   Supplementary Fig. 3). Disuse-driven FC changes involved L-SM1ue more than expected by 116 chance (P1: 9/50 connections, P = 0.018; P2: 37/50, P < 0.001; P3: 39/50, P < 0.001). Figure  4b shows the total magnitude of whole-brain FC change for each 119 vertex/voxel. The total magnitude 120 of whole-brain FC change was 121 significantly greater in L-SM1ue 122 than in the remainder of the brain 123 (Fig. 4B, Supplementary Fig. 3 Supplementary Fig. 3). b, For each vertex/voxel, the magnitude of whole-brain FC change was computed as the sum of squared FC changes between that vertex and every other gray-matter vertex/voxel. Shown for an example participant (P2; all participants shown in Supplementary Fig. 4).   To test if spontaneous activity pulses could explain observed increases in FC, we implemented 172 a pulse censoring strategy, in which frames surrounding each detected pulse (13.2 s before - 173 17.6 s after each pulse peak) were excluded from FC calculations (Fig. 7a). This approach is  pulse censoring on the actual data (Fig. 7b). Adding simulated pulses to all brain regions, 225 using mean pulse time series specific to each region, recreated the anatomical distribution of 226 FC changes observed during casting (Fig. 7d, Supplementary Fig. 6; P1: r = 0.44, P < 0.001; 227 P2: r = 0.95, P < 0.001; P3: r = 0.64, P < 0.001). Thus, pulse with a full distribution of 228 magnitudes could account for observed FC changes during casting and the partial reversal of 229 FC changes after pulse censoring.

231
We also utilized the pulse censoring and pulse addition analyses to test if pulses could explain 232 observed decreases in FC within the somatomotor system. Pulse censoring did not 233 significantly reverse the decrease in FC between L-SM1ue and R-SM1ue observed during the 234 cast period (Supplementary Fig. 7). Adding simulated pulses to baseline time series decreased 235 FC between L-SM1ue and R-SM1ue, but this effect was much smaller than that observed during 236 casting ( Supplementary Fig. 7). Daily 30-minute scans of rs-fMRI before, during, and after casting revealed that disuse not only 241 causes plasticity within the primary motor cortex 3,30 , but also increased functional connectivity 242 between disused circuits and executive control regions in the CON. These increases in FC 243 would have been difficult to observe using the focal recording techniques traditionally 244 employed to study plasticity. Thus, understanding the total impact of disuse on brain function 245 will require consideration of plasticity at multiple spatial scales. Future studies using positron 246 emission tomography (PET) to monitor local metabolic changes will further strengthen the 247 bridge between whole-brain plasticity and molecular mechanisms 32 .
The high degree of network specificity seen in disuse-driven plasticity is a powerful 250 demonstration of the validity of individual-specific rs-fMRI network parcellations. It could have 251 been the case that casting produced a complex pattern of FC changes involving many brain 252 systems. Instead, we found that virtually all of the regions showing increased connectivity with 253 the disused motor cortex belonged to a single network, the CON. Large-scale functional 254 networks seem to be a fundamental unit of brain organization, important not only for 255 understanding patterns of activation during cognitive processes, but also for understanding 256 whole-brain patterns of plasticity.

259
The network-specific increases in FC between the disused L-SM1ue and the CON likely reflect 260 the emergence of synchronized spontaneous activity pulses in these regions during casting.

261
FC is a measurement of the temporal correlation of spontaneous activity between brain   The importance of control systems in the human brain has been recognized for several control settings in relation to task objectives 17,18 . The increased FC between disused L-SM1ue 292 and the CON during casting suggests either 1) circuits within the CON were also disused 293 during casting, 2) the CON helps to maintain disused motor circuitry, or 3) spontaneous activity 294 pulses can spread along functional connections to neighboring brain regions in network space.

296
Spontaneous activity pulses may have emerged in the CON because its use was also reduced 297 during casting. We previously suggested that spontaneous activity pulses found in motor 298 regions may have been caused by disuse-driven changes in local inhibitory physiology 30,41 .
The CON is generally thought to represent task sets, abstract parameters and motor programs 300 governing goal-driven behaviors 17,18 . Task sets can be applied to multiple motor effectors (e.g., 301 right hand, left index finger, etc.), so it would be reasonable to think that casting one extremity 302 would not cause disuse of the CON. However, at least two sets of circuits within the CON may 303 have been disused during casting: circuits that represent bimanual behaviors and circuits that 304 convey task sets to effector-specific downstream regions (e.g., L-SM1ue). During casting, many 305 of the task sets previously used to control the dominant upper extremity (e.g., inserting a key 306 into a lock) could have been used to control the non-dominant extremity. However, task sets 307 requiring bimanual coordination (e.g., fastening a belt) could not be applied to an alternative 308 set of motor effectors and may have become disused. Another set of CON circuits that may 309 have been disused are the circuits that convey task set information to effector-specific brain 310 regions. The CON contains both somatotopically organized regions (e.g., SMA, SII) and non-311 somatotopic regions (e.g., anterior insula). Recent work shows that the SMA and SII, as well 312 as regions of the basal ganglia, may act as hubs that connect the rest of the CON to the 313 somatomotor network 26,42 . Such circuits could potentially undergo disuse during casting of one 314 extremity.

316
An alternative possibility is that inputs from the CON to L-SM1ue served a homeostatic function 317 during disuse, helping to maintain motor circuits that are typically maintained through active 318 use. We previously suggested that spontaneous activity pulses may help to maintain the 319 organization of disused brain circuits 30 . Perhaps pulses are triggered by the CON, which is the 320 system typically responsible for initiating activation of the somatomotor network 17 .

322
Spontaneous activity pulses may also have originated in the disused somatomotor circuits and 323 spread along functional connections to brain regions that were not affected by casting. The CON is immediately adjacent to the somatomotor network in network space (Fig. 3a). Previous  Extensive whole-brain imaging before, during and after casting revealed that disuse-driven 334 spontaneous activity pulses occur not only in primary motor and somatosensory areas, but 335 also in higher-order brain regions responsible for executive control over behavior (i.e., the 336 CON). The emergence of spontaneous activity pulses during casting produced increases in FC 337 that were highly focal in network space, specifically occurring between disused motor regions 338 and the CON. Decreases in connectivity between disused motor circuits and the remainder of 339 the somatomotor system, however, were not explained by spontaneous activity pulses. Thus, 340 disuse may drive network plasticity through two complementary mechanisms. Decreased 341 coactivation of brain regions during disuse might drive Hebbian-like disconnection between 342 disused and still-active somatomotor circuits. Spontaneous activity pulses, which contributed to 343 increased connectivity with the CON, may help preserve disused circuits for future 344 reintegration and use. Together, these two network plasticity mechanisms-Hebbian-like and 345 pulse-mediated-may represent a network-focal "stand-by mode" that allows the brain to  Vertex-wise seed maps were averaged across sessions before, during and after casting (Pre, Cast, 401 Post; Fig. 2, Supplementary Fig. 1).

402
FC between all pairs of cortical, subcortical and cerebellar parcels was measured as pairwise casting and recovery ( Fig. 2-3, Supplementary Fig. 1-2) were computed by averaging together columns 409 of the difference matrices corresponding to the L-SM1ue parcels (see ROI selection, above).

411
Cortical parcels were displayed in network space using force-directed ("spring-embedded") graphs 44 , 412 generated with Gephi (https://gephi.org/). Graph weights were taken from parcel-wise correlation 413 matrices averaged across all sessions prior to casting (Pre). Graphs were thresholded to include only 414 the top 0.2% of pairwise functional connections. We initially examined graphs using an edge threshold 415 of 0.1%, but several parcels with large FC changes were disconnected from the main graph. Thus, we 416 selected the 0.2% connection threshold to display all key findings (Fig. 3, Supplementary Fig. 2).

418
To test which functional networks showed large changes in FC with L-SM1ue during casting, parcel-wise 419 L-SM1ue seed maps were compared to individual-specific network maps. Parcel-wise L-SM1ue 420 difference seed maps (Cast -Pre) were thresholded to contain the top 5% of parcels showing FC increases/decreases. FC increases (Cast > Pre) and decreases (Cast < Pre) were examined 422 separately. The number of supra-threshold parcels in each cortical network is shown in Figure 3.

423
Alternative thresholds (top 1%, 10% and 20%) all yielded similar results. Networks containing a greater 424 number of FC increases/decreases than expected by chance were identified by comparing the number 425 of supra-threshold parcels found using each participant's true network map to the number of parcels 426 found using spatially permuted network maps (see Statistical analyses, below).

428
The spatial specificity of whole-brain plasticity was examined by comparing changes in FC between all 429 cortical, subcortical and cerebellar parcels to individual-specific functional network maps. We examined

435
To examine whole-brain FC changes at a finer spatial resolution, we examined differences in FC 436 between all cortical, subcortical and cerebellar vertices/voxels. We computed the magnitude of whole-437 brain FC change between each vertex/voxel as the sum of squared FC changes between that 438 vertex/voxel and every other gray-matter vertex/voxel (Fig. 4, Supplementary Fig. 3). We tested if 439 whole-brain FC change was higher in L-SM1ue than in the rest of the brain by comparing the mean 440 magnitude of whole-brain FC change inside of each participant's true L-SM1ue ROI to the mean 441 magnitude of whole-brain FC change inside of spatially rotated ROIs (see Statistical analyses, below).

443
To generate a parcel-wise map of spontaneous activity pulses, we extracted rs-fMRI time series from 444 every cortical, subcortical and cerebellar parcel surrounding each detected pulse (13.2 seconds before 445 to 17.6 seconds after each pulse peak). Pulse peaks were detected previously 30 . We then performed an compared to the spatial distribution of L-SM1ue FC changes using a Pearson correlation across parcels. 448 We also compared the number of pulses detected during each rs-fMRI scan 30 to the FC measured 449 between L-SM1ue and L-dACC (Fig. 5c), using a Pearson correlation.  Magnitudes were grouped into bins with a width of 0.5%-signal-change. We then fit a log-normal 466 distribution to the observed pulse magnitudes using a least-squares approach (Fig. 8a). This provided 467 an estimate of the full pulse magnitude distribution, including pulses that were too small to detect. We  Fig. 8c). Finally, we applied the same 473 pulse censoring strategy described above to the simulated time series and computed FC after 474 censoring (Cens; Fig. 8c). Because pulse magnitudes were drawn from a full log-normal distribution, 475 only a portion of the added pulses were detected and censored. We compared Sim vs. Pre FC 476 measurements and Cens vs. Sim FC measurements using t-tests (see Statistical analyses, below). The 477 simulation and censoring procedures were repeated using triangular and exponential magnitude 478 distributions ( Supplementary Fig. 5). We also generated difference seed maps showing FC changes 479 between L-SM1ue and all cortical parcels due to pulse simulation (Sim -Pre; Fig. 8d

512
When parametric statistical tests were not appropriate to test a specific hypothesis, we tested results 513 against a null distribution generated via permutation resampling. In each case, our null hypothesis was 514 that observed effects had no spatial relationship to ROIs/functional networks and any overlap occurred 515 by chance. For vertex-wise ROIs, we modeled the null hypothesis by rotating the ROI around the 516 cortical surface 1,000 times 45,46 . For functional networks, we modeled the null hypothesis by permuting 517 the network assignments of parcels 1,000 times. Each permuted ROI/network map was used exactly as 518 the actual map in order to compute a null distribution for the value of interest. The P-value reported for 519 each test represents the two-sided probability that a value in the null distribution has a greater 520 magnitude than the observed value. Permutation resampling was used to generate null distributions for 521 the following values: • Overlap of each functional network with increases in L-SM1ue FC during casting (Cast > Pre; 523 number of parcels in top 5%; Fig. 3b) 524 • Overlap of each functional network with decreases in L-SM1ue FC during casting (Cast > Pre; 525 number of parcels in top 5%; Fig. 3c) 526 • Overlap of L-SM1ue parcels with the 50 largest changes in FC between all pairs of cortical, 527 subcortical and cerebellar parcels (Fig. 4a, Supplementary Fig. 3)

529
Since we tested the overlap of L-SM1ue FC increases/decreases with 17 different functional networks, a 530 Benjamini-Hochberg procedure was applied to correct for multiple comparisons, maintaining false 531 discovery rates < 0.05. Each of the three participants constituted a separate replication of the 532 experiment, rather than multiple comparisons, so no correction was necessary for tests repeated in   This study used our previously published dataset 30 , available on the OpenNeuro database 546 (www.openneuro.org/datasets/ds002766).