Functional reorganisation and recovery following cortical lesions: A study in macaque monkeys

Damage following traumatic brain injury or stroke can often extend beyond the boundaries of the initial insult and can lead to maladaptive cortical reorganisation. On the other hand, beneficial cortical reorganisation leading to recovery of function can also occur. Here, we used resting state FMRI (rsFMRI) to examine how functional connectivity in the macaque brain changed across time in response to lesions to the prefrontal cortex, and how this reorganisation correlated with changes in behaviour. Two monkeys were trained to perform location-based and object-based delayed match-to-sample tasks. We also collected rsFMRI data under general anaesthesia at two pre-lesion time-points, separated by 3-4 weeks. After two cycles of testing and scanning, the animals received a principal sulcus lesion followed by an additional 4 cycles of testing and scanning. Later, the same animals received a second lesion to the opposite hemisphere and additional cycles of testing and scanning. Both animals showed a marked behavioural impairment following the first lesion, which was associated with a decrease in functional connectivity, predominantly within frontal-frontal networks in both hemispheres. Approximately 8 weeks following the lesion, performance improved, as did functional connectivity within these networks. Following the second lesion, functional connectivity again decreased and this was associated with a marginal behavioural deficit that did not recover. Our data show that behavioural impairments reflect not just the removal of the lesioned area, but also disturbance to an extensive cortical network. This network can recover by restoring and/or strengthening pre-existing connections, leading to improvement in behaviour.


INTRODUCTION
Cortical damage that accompanies traumatic brain injury or stroke often extends beyond the 45 boundaries of the initial injury. This can lead to maladaptive cortical reorganisation and cognitive impairment 46 (Grefkes & Fink, 2014). On the other hand, beneficial cortical reorganisation following injury can also occur 47 and this can lead to recovery of function. Understanding the nature of cortical reorganisation after injury and 48 how this might be promoted is a challenge for research on developing treatments for patients suffering from 49 brain injury. 50 Resting state functional connectivity (rsFMRI) provides an indirect method of measuring cortical 51 organisation across the whole brain by correlating BOLD activation patterns between pairs of brain areas. 52 Strong correlation implies, at minimum, a "functional" connection, and often an anatomical connection 53 (Deco, Jirsa, & McIntosh, 2011). Over the past decade, rsFMRI has been used to examine changes in network 54 organisation in healthy individuals as well as patients who have suffered lesions or who have a variety of 55 disorders such as schizophrenia, Alzheimer's and Parkinson's disease (Fornito, Zalesky, & Breakspear, 2015; 56 He et al., 2007;Siegel et al., 2016). However, to fully understand the consequences of cortical reorganisation 57 following damage, it is necessary to measure correlations within cortical networks both pre-and post-injury. 58 This is rarely possible with human patients, and so we must rely on animal models, where we can collect data 59 both before and after a lesion. 60 Recent studies have looked at functional connectivity following lesions in non-human primates.  These studies are important in demonstrating the utility of studying functional connectivity following 70 lesions. However, to understand how behavioural recovery occurs after brain injury in patients, we need to 71 correlate changes in functional connectivity with behavioural measures as recovery occurs. In the present 72 study, we therefore used rsFMRI to study how cortico-cortical connectivity in the macaque monkey brain 73 changed in response to discrete lesions to regions near the principal sulcus (specifically, areas 46 and 9/46) 74 of the prefrontal cortex over an extended period of time; and how these changes related to behaviour. 75 We chose to study lesions to regions near the principal sulcus because it is well known that lesions 76 there reliably abolish the ability of monkeys to perform delayed response and delayed alternation tasks (E. and object-based delayed match-to-sample task. We collected rsFMRI data at periodic intervals during the 81 pre-lesion period to coincide with behavioural testing sessions. The animals then first received a lesion to 82 both banks of the principal sulcus, including areas 46 and 9/46. Following a post-operative recovery period, 83 we resumed periodic testing and scanning sessions. Later, they received a second lesion to the same region 84 in the opposite hemisphere and they were once again tested and scanned at regular intervals. Adding the 85 second lesion allowed us to assess the contribution to recovery of the homotopic region in the undamaged 86 hemisphere.

89
We performed a longitudinal assessment of the effect of lesions to the principal sulcus (areas 46 and 90 9/46) on behavioural performance on two cognitive tasks and related it to changes in functional connectivity 91 (Figure 1). Once the animals had reached a predefined level of performance on the behavioural tasks (>70%), 92 we collected functional neuroimaging (rsFMRI) data under general anaesthesia at two intervals prior to the 93 first lesion, separated by 3-4 weeks (Data from two additional scans, earlier in the animals' training, are not 94 included in the present report). Several days prior to each scanning session, the animals were tested on both 95 the location-and object-based delayed match-to-sample (DMS) tasks. Following these two cycles of 96 behavioural testing and scanning, each animal received a lesion to both the dorsal and ventral banks of the 97 4 left principal sulcus (PS), targeting areas 9/46, 46d, and 46v (Figure 2). Following a post-operative recovery 98 period (approximately 4 weeks), we resumed cycles of behavioural testing and scanning (4 cycles 99 approximately once/3-4 weeks). For the next several months, similar cycles of behavioural testing (but 100 without scanning) continued. After 7 months following the first lesion, the animals received a second lesion 101 to both banks of the right PS (Figure 2). Following a post-operative recovery period (approximately 4 weeks), 102 the animals were once again tested and scanned (4 cycles, approximately once every 3-4 weeks).

104
Behavioural deficits following unilateral lesion associated with decreased functional connectivity within 105 frontal cortex 106 The ability of both monkeys to perform the two DMS tasks was significantly impaired following a 107 unilateral lesion of the left PS regions ( Figure 3A). We compared behavioural performance (measured as % 108 correct) in the two sessions prior to the lesion (pre-lesion-1) with the first two sessions following the lesion 109 (early post-lesion-1) and the two after 8 weeks (late post-lesion-1) using a mixed-model ANOVA ( Figure 3A; 110 see MATERIALS AND METHODS). We observed a significant main effect of monkey (F (1,13) =23.20, p=3.37x10 -111 4 ), along with a significant interaction between monkey and experimental stage (F (2,13) =5.74, p=0.0164). We 112 also observed a significant main effect of task (location vs. object, F (1,13) =87.34, p=3.92x10 -7 ). Critically, a main 113 effect of experimental stage (F (1,13) =17.90, p=1.84x10 -4 ) was observed, including a notable decrease in 114 performance shortly following the lesion, followed by substantial recovery. All remaining main effects and 115 interactions failed to achieve statistical significance (p's>0.05).

122
This decrease in behavioural performance during the first 8 weeks following the lesion (early post-123 lesion-1) was associated with significant changes in functional connectivity within our network of interest. In   Behavioural recovery associated with restored frontal connectivity 137 In later behavioural sessions conducted 8-12 weeks after the lesion (late postlesion-1), both monkeys 138 showed signs of recovery on the location task, and in the case of monkey 1 on the object task ( Figure 3A). On 139 the location task, the performance of monkey 1 increased from 49±6% to 65±5% (p=0.069), and the 140 performance of monkey 2 increased from 65±5% to 75±5% (p=0.28). For both monkeys, performance 141 improved to the point where it was no longer statistically different from their pre-lesion performance (pre-142 lesion-1 vs. late postlesion-1, p=0.43 and p=0.095 for monkeys 1 and 2, respectively). On the object task, the 143 performance of monkey 1 increased from 73±5% to 95±2% (p=0.006). It too was restored to pre-lesion levels 144 (p=0.38). The performance of monkey 2 showed a marginal decrease from 93±3% to 90±3% (p=0.78).  Disruption to behavioural performance following subsequent lesion to right principal sulcus 151 Approximately seven months following the lesion to the left PS regions, both monkeys received a 152 second lesion to the right PS regions (Figure 2). We used a second mixed-model ANOVA to evaluate the 153 behavioural impact of this procedure across the three stages of behavioural testing: the 8-week period 154 immediately before the second lesion (pre-lesion), the first 8 weeks following the second lesion (early post-155 lesion-2), and the period 8-12 weeks following the second lesion (late post-lesion-2).

191
Compared to the first lesion, the most striking feature of network reorganisation following the  On the one hand, the network effects of a focal frontal lesion were surprisingly widespread. Following 210 a unilateral lesion, disruption extended not just to connections between undamaged frontal regions within 211 the lesioned hemisphere, but to connections of these to the opposite frontal lobe, and even to connections 212 exclusively within the undamaged hemisphere. On the other hand, disrupted connectivity was far from 213 universal. In both hemispheres, connections of frontal to parietal and frontal to temporal cortex were largely 214 unchanged, following either unilateral or bilateral frontal lesion. These results suggest that, following focal 215 frontal lobe lesions, behavioural impairments and recovery could be specifically related to a widespread 216 disruption of connectivity; restricted to the frontal lobes but widespread within those lobes. Similarly, 217 recovery of behaviour following a unilateral lesion was associated with bilateral recovery of frontal 218 connectivity.

219
In the following sections, we first acknowledge potential limitations in our approach. We then discuss 220 why a unilateral lesion to PS regions might lead to disruptions in behaviour. Next, we consider how recovery 221 of connections within and between the frontal lobes could help compensate for the effect of the lesion and 222 thus lead to an improvement in behavioural performance. Finally, we discuss the marginal behavioural 223 impairment and lack of network recovery following a second lesion to the opposite hemisphere.

225
Limitations and issues of interpretation 226 We acknowledge several potential limitations to this study. First, we only tested two animals and no 227 control animals were included. When designing studies in non-human primates, one must balance the need 228 for adequate sample size with the ethical, logistical, and financial costs associated with the work. We did not 229 anticipate a need for control animals because both animals were trained to the same criterion prior to the 230 first lesion. In the absence of any lesion, there was no reason to suspect that the animals would exhibit 231 significant changes in behaviour.

232
Second, we cannot be sure that underlying white matter was not affected during the surgical 233 procedures even though an operating microscope was used. And indeed, there is evidence from a few 234 sections that the lesion may have extended beyond the fundus of the sulcus (Fig. 2). However, the purpose 235 of our study was not to determine the function of the cortical tissue near the principal sulcus, but rather to 236 study functional recovery; and in stroke patients, white matter is always affected. Regardless of the extent 237 to which underlying white matter was affected, we were nonetheless able to observe both behavioural 238 impairment and recovery that correlated with changes in functional connectivity.

239
Third, after the second lesion there were changes in functional connectivity; yet there was only a 240 marginal change in behaviour. One possible confound is a practice effect, in that as the experiment 241 progressed the monkeys received more and more practice on the two tasks. However, it is important to note 242 that both monkeys were trained to criterion prior to the first lesion and they were tested relatively 243 infrequently following the lesion. It is therefore unlikely that practice alone can account for the behavioural 244 recovery following the first lesion and the marginal effect on behaviour of the second lesion. We instead 245 argue that it is more likely the monkeys adopted a different strategy for completing the task that allowed 246 them to bypass the effects of the lesions (see below). Consistent with (but not definitively in support of) this 247 suggestion is the observed increase in connectivity between the parietal and prefrontal cortex in the 248 hemisphere (left) in which we placed the first lesion (Fig. 5). 249 Fourth, we collapsed the behavioural data across testing cycles and difficulty (i.e., delays); which 250 raises some issue with interpretation. We chose to collapse these data in order to increase statistical power 251 and to better align the behavioural data with the imaging data. It is possible that doing so obscured a  Finally, we acknowledge a potential issue of voxelsize. The problem is that neighbouring voxels may 257 be supplied by the same vessels. This means that there may be an artefactual correlation between adjacent 258 7 regions. To best avoid this problem, we chose as seed areas regions that were less likely to be supplied in this 259 way (Fig. 1).  Both monkeys eventually showed recovery of function after the first lesion, and this brings us to the 289 fundamental issue, which is how does this recovery occur. First, we must consider the possibility that the 290 lesions were not complete and it was residual tissue that was mediating behavioural recovery. However, as 291 can be seen from Figure 2, there was very little, if any, tissue left in the principal sulcus. It is one advantage 292 of working with animals that it is possible to ensure, as here, that the lesions are indeed complete.

293
A second possibility is that after the first unilateral lesion, the homotopic region in the other 294 hemisphere could compensate. But if area 46 in the right hemisphere had indeed taken over, we might have 295 expected to see a severe impairment when it was then itself lesioned, which was not the case ( Figure 3C); so 296 at best this may only be part of the answer. A third possibility, and one that is most strongly supported by 297 the data, is that widespread cortical recovery and reorganisation within the frontal lobes contributed to the 298 recovery of function following the first lesion. Given the complexity of the cortical network and the degree 299 of interconnectivity, particularly within and between the frontal lobes, there are always potential alternative 300 routes of information transmission so long as the lesion is not extensive. If there are multiple potential 301 strategies to complete the task, for example, it is conceivable that alternative pathways that bypass area 46 302 might be recruited, or might recover, as frontal connectivity is restored.

303
Results following the second lesion were somewhat mixed. The behavioural impairment was modest 304 in comparison to the large impairment that followed the first, unilateral lesion and yet, there was little 305 evidence of subsequent recovery. The disruption of frontal lobe connectivity was again substantial and this 306 time long-lasting or permanent. Potentially, behaviour is only mildly affected by a second lesion to the right 307 area 46 because by this point, the animals have either adopted a strategy that is less dependent on area 46 308 and/or alternative neural pathways that bypass area 46 have been recruited. Moreover, assuming this is true, 309 there would be little drive for cortical reorganisation in this case, which would account for why little change 310 in network connectivity was observed even at the late post-lesion-2 period. A second possibility is that, for 311 frontal networks to recover, area 46 is necessary in at least one hemisphere, relating to the common clinical 312 8 observation that especially severe cognitive deficits can follow bilateral frontal lesions. In this case, modest 313 behavioural impairment following the second lesion might be ascribed to some additional factor, such as the 314 additional training received between the two lesions. In line with many previous suggestions, our data confirm that the effects of a focal frontal lobe lesion 318 cannot be understood simply as loss of function in the specific area removed. Instead, there is a widespread, 319 though also specific, disruption of connectivity between many regions within and between the two frontal 320 lobes, potentially bringing a widespread impairment in their function. At least following unilateral lesion, this 321 disruption recovers over time, with associated recovery in behaviour. We suggest that it is only possible to 322 understand how a brain lesion affects behaviour by studying the whole network and its interconnections. A 323 lesion, however discrete, disturbs the network as a whole and this effect can last for weeks. However, 324 recovery can occur as intact parts of the network regain their normal function, providing alternative ways in 325 which the system can perform the task. Given the complexity of the network, there are multiple ways in 326 which information in one area can be transmitted to inform another.

327
With this study, we have provided a framework for investigating recovery after lesions that relies on faced a touchscreen to which the monkey had access. In the 'match-to-location' task ( Figure 1A, left), the 353 monkey was required to touch a red cross that appeared in a random location on the touchscreen. The cross 354 then disappeared and a distractor (blue square) appeared in the centre of the screen and the monkey was 355 required to touch this. After a variable delay (2, 4, 8, or 16 s), three stimuli identical to the sample appeared 356 in three different locations. The three locations included the sample location from the current trial, the 357 sample location from a previous trial, and a third random location. The monkey was required to touch the 358 cued location on the current trial to receive a food pellet reward.

359
In the 'match-to-object' task ( Figure 1A, right), the monkey was required to touch a cue that 360 appeared in the centre of the touchscreen. There was then a variable delay (3, 5, 9, or 17 s); the extra 1 361 second was added to approximately match the distractor plus delay durations in the match-to-location task.

362
Two stimuli then appeared on the touchscreen on either side of midline (along the horizontal meridian, 363 equidistant from centre). These included the sample stimulus and a distracter stimulus (randomly allocated 364 to either left or right of midline). The monkey was required to touch the stimulus that had been cued to 365 receive a food pellet reward.

366
The two tasks were not matched for overall difficulty: based on performance data, the location task 367 was more difficult than the object task (Figure 3). For each testing cycle, there were 1 or 2 test sessions per 368 task (100-120 trials per session), on different days. The second test session was added from the second 369 postlesion cycle onwards. For cycles with 2 test sessions per task, data from the two were combined as there 370 was no significant difference in performance between the sessions when summed across all tasks, monkeys, 371 and stages of testing (testing session 1 vs. testing session 2: 82±2% vs. 83±2, p=0.11, paired t-test).

372
Because of the relatively long delays between testing sessions (several weeks), we started each cycle 373 with shorter 'warm-up' sessions (40-100 trials). These were held over two days prior to actual testing sessions 374 in order to re-introduce the animals to the process of testing. Data obtained during warm-up days were 375 excluded from all analyses.

376
All performance data (percent correct) were arcsine transformed before being analysed using two 377 separate 3-way mixed-model ANOVAs (one per lesion) with each testing cycle corresponding to a unit of 378 replication. Each ANOVA included three fixed-effects: task (match-to-location, match-to-object), subject 379 (monkey 1, monkey 2), and experimental stage (pre-lesion-1/2, early postlesion-1/2, late postlesion-1/2); 380 and one random-effect corresponding to testing cycle. Post-hoc t-tests were carried out on ANOVA-derived 381 estimated marginal means to identify specific changes in performance associated with subject, task and 382 experimental stage. All p-values were adjusted for multiple comparisons using the Holm-Bonferroni method 383 to control for family-wise error (Holm, 1979).    Under deep anaesthesia, the head was placed in a head holder and the skull exposed by opening the 417 scalp and galea in layers. The temporal muscles were retracted and a bone flap was removed. The dura was 418 cut to expose the cortical surface. Both banks of the principal sulcus were removed with aspiration ( Figure   419 2). The dura was then sewn and the bone flap replaced.   The first temporal derivatives of these time-courses were also included. A motion confound covariate was 448 defined as the time-course of average displacement over the expected brain volume (approximated as a 449 sphere of radius 40 mm). The first temporal derivative of this vector was also included, as were the element- Defining the network of interest 458 To evaluate changes in connectivity following the lesions, we compared pre-and post-lesion 459 connectivity in a predefined "network of interest" (shown in Figure 1B   touchscreen. In the 'match-to-location' task (left), the monkey was required to touch a cue that appeared in 597 a random location on the touchscreen. The cue then disappeared and after a variable delay, three stimuli 598 identical to the cue appeared in three different locations. The three locations included the sample location 599 from the current trial, the cued location from a previous trial, and a third random location. The monkey was 600 required to touch the location of the cue on the current trial to receive a food pellet reward. In the 'match-601 to-object' task (right), the monkey was once again required to touch a cue that appeared in a random location 602 on the touchscreen. After a variable delay, two different stimuli appeared in random locations: the sample 603 stimulus and a distracter stimulus. The monkey was required to touch the sample stimulus to receive a food