Modelling the influence of the hippocampal memory system on the oculomotor system

Visual exploration is related to activity in the hippocampus (HC) and/or extended medial temporal lobe system (MTL), is influenced by stored memories, and is altered in amnesic cases. An extensive set of polysynaptic connections exists both within and between the HC and oculomotor systems such that investigating how HC responses ultimately influence neural activity in the oculomotor system, and the timing by which such neural modulation could occur is not trivial. We leveraged TheVirtualBrain, a software platform for large-scale network simulations, to model the functional dynamics that govern the interactions between the two systems in the macaque cortex. Evoked responses following the stimulation of the MTL and some, but not all, subfields of the HC resulted in observable responses in oculomotor regions, including the frontal eye fields (FEF), within the time of a gaze fixation. Modeled lesions to some MTL regions slowed the dissipation of HC signal to oculomotor regions, whereas HC lesions generally did not affect the rapid MTL activity propagation to oculomotor regions. These findings provide a framework for investigating how information represented by the HC/MTL may influence the oculomotor system during a fixation and predict how HC lesions may affect visual exploration. Author Summary No major account of oculomotor (eye movement) guidance considers the influence of the hippocampus (HC) and broader medial temporal lobe (MTL) system, yet it is clear that information is exchanged between the two systems. Prior experience influences current viewing, and cases of amnesia due to compromised HC/MTL function show specific alterations in viewing behaviour. By modeling large-scale network dynamics, we show that stimulation of subregions of the HC, and of the MTL, rapidly results in observable responses in oculomotor control regions, and that HC/MTL lesions alter signal propagation. These findings suggest that information from memory may readily guide visual exploration, and calls for a reconsideration of the neural circuitry involved in oculomotor guidance.


Abstract 31
Visual exploration is related to activity in the hippocampus (HC) and/or extended medial 32 temporal lobe system (MTL), is influenced by stored memories, and is altered in amnesic 33 cases. An extensive set of polysynaptic connections exists both within and between the 34 HC and oculomotor systems such that investigating how HC responses ultimately 35 influence neural activity in the oculomotor system, and the timing by which such neural 36 modulation could occur is not trivial. We leveraged TheVirtualBrain, a software platform 37 viewed items (Fagan, 1970;Fantz, 1964), and more viewing is directed to areas that have 63 been altered from a prior viewing ( relationship between visual sampling and HC activity is weakened in aging, presumably 78 due to decline in HC structure or function (Liu, Shen, Olsen, & Ryan, 2018 To examine the extent to which HC/MTL activity could influence the oculomotor 112 system, we leveraged a computational modeling and neuroinformatics platform, 113 TheVirtualBrain, and simulated the functional dynamics of a whole-cortex directed 114 macaque network when stimulation is applied to HC and MTL nodes of interest. 115 Critically, we examined whether and when evoked activity culminated in responses in 116 key regions within the oculomotor system. Finally, we observed the extent to which the 117 propagation and timing of such activity was altered following lesions to one or more 118 HC/MTL regions in order to understand the neural dynamics that may underly altered 119 visual exploration in cases of HC/MTL dysfunction, such as in amnesia or aging. 120 121

Results 122
We modelled the influence of HC/MTL activity on the oculomotor system using a 123 connectome-based approach using TheVirtualBrain (see Methods for details). Following 124 Spiegler and colleagues (2016), we assigned a neural mass model to each node and set 125 each to operate near criticality, which is considered to be the point at which information 126  Figure 1 shows an example of activity dissipation following CA1 stimulation. 140 Evoked responses were first detected in other HC subfields and MTL regions but then 141 spread to prefrontal and extrastriate cortices, and later to posterior parietal cortex. The 142 full list of activation times for each of the 77 nodes can be found in Supplementary Table  143 2. However, in all subsequent analyses, we present only the results pertaining to our 144 nodes of interest, identified as those along the shortest paths between HC/MTL and 145 oculomotor regions (see Shen et al., 2016)  PrS stimulations (> 440 ms; Table 1). No evoked response was detected in area LIP 165 following stimulation of S or PaS. Responses were not observed in the majority of the 166 pre-defined cortical hubs following CA3 stimulation, and activity did not culminate in 167 observable responses in the oculomotor areas ( Figure 2B). See Table 1 for activation 168 times for all nodes of interest.   Table 4). 237 A combined lesion to all HC subfields (CA3/CA1/S/PaS/PrS) did not 238 considerably change the pattern of signal propagation from the MTL cortices to 239 oculomotor regions. In cases where speeding/slowing was observed, the timing 240 differences were less than 15 ms, and mostly less than 10ms. Signal from the MTL 241 cortices still culminated within the oculomotor regions well under 50 ms (except for ERC 242 -> FEF at 79 ms, and LIP whose responses remained >140 ms) (Supplementary Table 5).  Table  246 6). TF and/or TH lesions resulted in slowing (10-400ms) of signal following CA1, S, and 247 PaS stimulation to one or more of areas 24, 46 and FEF, and a lack of response in FEF 248 following PaS stimulation (only the combined TF/TH is shown; Supplementary Table 7). 249 Area 35 and/or 36 lesions also resulted in slowing (10-90ms) of signal following CA1, S, 250 and PaS stimulation to one or more of areas 24, 46 and FEF, although not as severe as the 251 slowing observed following TF/TH lesions (only the combined 35/36 lesion is shown; 252 Supplementary Table 8). 253

Other Cortical Lesions 254
In our original stimulations, signals in regions 5, 7a, 23, and V4 were predominantly 255 observed following observable responses in oculomotor areas 24, 46 and FEF, suggesting 256 these cortical areas are receiving feedback signals rather than primarily serving as hubs to 257 transfer signal from the HC/MTL to the oculomotor regions. To explore this in more 258 depth, we simulated a combined lesion of 5/7a/23/V4 and examined signal propagation.  As CoCoMac only provides categorical weights for connections (i.e., weak, 409 moderate or strong), we ran probabilistic tractography on diffusion-weighted MR 410 imaging data from 10 male adult macaque monkeys (9 Macaca mulatta, 1 Macaca 411 fascicularis, age 5.8 ± 1.9 years) using the FV91 parcellation to estimate the fibre tract 412 capacities and tract lengths between regions. Image acquisition, preprocessing and 413 tractography procedures for this particular dataset have been previously described (Shen 414 et al., in press., 2019). Fiber tract capacity estimates (i.e., 'weights') between each ROI 415 pair were computed as the number of streamlines detected between them, normalized by 416 the total number of streamlines that were seeded. Connectivity weight estimates were 417 averaged across animals and applied to the tracer network, keeping only the connections 418 that appear in the tracer network. The resulting structural connectome was therefore 419 directed, as defined by the tracer data, and fully weighted, as estimated from 420 tractography. Tract lengths were also estimated using probabilistic tractography. 421

Node dynamics 422
The dynamics of each node in the macaque network were given by the following generic The local coupling is scaled by g, while the global connectivity scaling factor ξ 429 acts on all incoming connections to each node, which are also weighted individually by 430 the connectivity weights matrix w (as described above). Exogenous stimulation currents 431 of interest in the present study enter the system through the input variable I. Transmission 432 between network nodes was constrained according to the conduction delays matrix Δ = 433 L/v, where L is a matrix of inter-regional tract lengths and v is axonal conduction 434 velocity. As in Spiegler et al. (2016), cubic, quadratic, and linear coefficients for V and W 435 were set such that the dynamics reduce to a classic Fitzhugh-Nagumo system. Additional 436 model parameters are listed in Supplementary Table 1. 437 Brain dynamics operate near criticality (Ghosh et al., 2008). In this state, the 438 nodes will naturally oscillate with constant magnitude. Setting the local parameter g so 439 that the system operates near criticality will allow the node to respond with a strong 440 amplitude, and a longer lasting oscillation. If far from the critical point, the amplitude 441 responses will be weak, slow, and fade quickly, and if spreading within a network, the 442 excitation will decay quickly as it travels. Given our network's structure, re-entry points 443 allow a node to be re-stimulated, making the excitation last longer and travel farther system of delay-differential equations shown above were solved numerically using a 447 Heun Deterministic integration scheme, with step size dt=0.1 ms. 448

Model tuning & stimulation parameters 449
Simulations were run for 7000 ms, with stimulus onset occurring after 5000 ms to allow 450 for settling of the initial transient resulting from randomly specified initial conditions. A 451 single pulsed stimulus was used, with duration of 100 ms. To determine when nodes 452 became active following stimulation, we first computed the envelope of each node's 453 timeseries using a Hilbert transform. Each node's baseline activity was taken as the mean 454 amplitude of the envelope in the 200 ms prior to stimulation. The activation threshold of 455 each node was defined as the baseline activity ± 2 std and activation time of each node 456 was taken as the time its envelope amplitude surpassed the activation threshold. 457 To create a biologically realistic model, we stimulated V1 to find activation times 458 of the following areas: V1, V2, V3, V4, middle temporal and medial superior temporal. sites were repeated on these lesion models. 488

Code availability 489
Simulations were carried out using the command-line version of TheVirtualBrain (TVB) 490 software package in Python, which is available for download at http://thevirtualbrain.org. 491 The customized TVB code for the simulations presented here is available upon request.  Table 1. Activation times (ms) following stimulation of hippocampal subfields and