Functional Connectivity Of Brain Network Constituted With Principal Components Of Action Potentials During Working Memory Task

Objective:Working memory (WM) is critically important in cognitive tasks. The functional connectivity has been a powerful tool for understanding the mechanism underlying the information processing during WM tasks. The aim of this study is to investigate how to effectively characterize the dynamic variations of the functional connectivity in low dimensional space among the principal components (PC) which were extracted from the instantaneous firing rate series. Spikes were obtained from medial prefrontal cortex (mPFC) of rats with implanted microelectrode array and then transformed into continuous series via instantaneous firing rate method. Granger causality method is proposed to study the functional connectivity. Then two scalar metrics were applied to identify the changes of the reduced dimensionality functional network during working memory tasks:functional connectivity and efficiency.Methods:1. Experimental data:In the process of adult SD (Sprague-Dawley) rats working memory in Y-maze, multi-channel implantable electrodes array were applied to record multi-channel neural electrical activities in the medial prefrontal cortex, through high-pass filter, peak potential detection and classification processing (spike-sorting), to obtain the neurons’action potential. The data come from the WM of6rats,80trails.2. Transformed into continuous series:Considering that both rate and temporal information might be important for assessing the interaction among neurons, we here used instantaneous firing rate method which was applicable for dealing with short, sparse spikes trains to generate a continuous time series more suitable for MVAR modeling based granger causal.3. Selected PC from continuous series:selected n new variables (PC) accounted for over the most energy of the original continuous series. 4. Causal connectivity among PC based on Granger causality:Using the method of granger causality, the causal connectivity matrix and the average of causal matrix (GCpc) were calculated in the low-dimensional PC space, in order to quantitatively describe the functional connectivity.5. The efficiency of information transmission of PC causal network during WM: Definite a causal network based on the matrix of causal connectivity. The efficiency of network was calculated to quantitatively describe the efficiency of information transmission in causal network in the low-dimensional PC space during WM.Results:1. Conversion into continuous series. One WM task of rat1selected as example to show multi-channel actionpotential, neuron’s action potential and its continuous series.2. Selected PC from continuous series:Due to the different number of recorded neurons from6rats, the principal components shared of energy are also different. In this paper, we select first seven principal components that take the most energy of continuous sequence of neurons at working memory states. The first seven principal components accounted of all the energy of a continuous sequence83.0%of Rat1,82.1%of Rat2,91.4%of Rat3,81.4%of Rat4,82.6%of Rat5and96.3%of Rat6.3. Dynamic variations of PC functional connectivity during the WM tasks3.1Dynamic variations of GCpc during WM tasksThe changing tendencies of the functional connectivity were analyzed during the correctly performed WM tasks (4s pre and2s post the tripping point). Accordingly, the period of the WM task was divided into six1s length bins which were defined as Ao, A1, A2, A3. A4and A5from the beginning to the end. GCpc of6rats were dynamic varied. The maximum were0.031±0.005、0.041±0.007、0.033±0.006、0.031±0.005、 0.027±0.003、0.051±0.016, occurred at Is pre the tripping point (Rat1, Rat2and Rat3) or2s pre the tripping point (Rat4, Rat5and Rat6). GCpc at A0were0.023±0.006、0.011±0.005、0.020±0.005、0.011±0.005、0.020±0.005、0.013±0.004. The maximum of GCpc were significant lager than GCpc at A0(paired sample t-test, P＜0.05).3.2Dynamic variations of Epc during WM tasksThe path lengths were determined by taking the inverse of the strength of the granger causality connection. Then, graph theoretical measures were applied to estimate the connectivity as it provided an effective and informative way to explore the network properties. Epc of6rats were dynamic varied. The maximum were:0.63±0.04、0.55±0.04、0.54±0.07、0.50±0.05、0.51±0.04、0.66±0.09, occurred at2s pre the tripping point (Rat1, Rat2and Rat3) or1s pre the tripping point (Rat4, Rat5and Rat6). EPC at the A0of WM were0.36±0.07、0.26±0.06、24±0.07、0.22±0.06、0.25±0.04、0.35±0.08. The maximum Epc were significant lager than Epc at A0(paired sample t-test, P＜0.05).4. Dynamic variations of functional connectivity during the WM tasksIn order to compare to functional connectivity in low dimensional space, we courted GC and E each second from A0to A5during WM tasks. The maximum of GC, E, GCpc and Epc occurred at same time. The maximum of GC were significant lager than GC at A0of Rat1, Rat2, Rat3and Rat4(paired sample t-test, P<0.05). he maximum of E were significant lager than E at Ao of Rat2, Rat3, Rat4, Rat5and Rat6(paired sample t-test, P＜0.05). The maximum of GCpc and Epc were significant lager than the maximum of GC and E (paired sample t-test, P＜0.05).Conclusion:1. The changing trends of GCpc and Epc were the same as GC and CD. The maximum of GC, E, GCpc and Epc occurred at same time. These significant results, taking together, suggested that combining the application of the PCA and the GCCA might provide an effective way to investigate the functional connectivity mechanism during the WM tasks in low dimensional space.2. The maximum values of the functional connectivity and the effect of the parallel information transfer appeared at A2(Rat1, Rat2and Rat3) and A3(Rat4, Rat5and Rat6), which corresponds to2s or1s before the tripping time. Because the tripping time of the infrared sensor marked the’choice run’ behavioral events during the WM tasks, the results suggested the strongest connectivity happened2s or1s before the’choice’ behavioral events.