High Frequency Cortical Networks

determining the dynamic changes of the functional network topology during interictal and ictal periods as revealed by measures of connectivity in the ripple and fast-ripple frequency bands using high density cortical recordings in humans will likely provide deep insight into the nature of these topological changes. This problem has been partially investigated using amygdalo-hippocampal depth electrodes for oscillations <50 Hz (Wilke, Worrell et al. 2011, van Diessen, Hanemaaijer et al. 2013), but has never been extensively examined in the HFO band in human epilepsy patients. This gap in knowledge represents an important problem because it limits our understanding of the neural mechanisms underlying epileptogenesis in general and specifically as a tool to precisely define seizure-generating foci to guide surgical resections. 

On the other hand, multivariate effective connectivity measures, such as the directed transfer function[N1]  (kaminski and blinowska 1991, Franaszczuk 1994, Franaszczuk Bergey 1998, wilke 2010 jung 2011, wilke 2011) and partial directed transfer (varotto 2012), has been used to localize seizure onset zones, assuming that iEEG signals remain stationary during ictal periods. However, the grounds for such an assumption are not strong given the natural tendency of ictal iEEG signals to become non-stationary. To overcome this limitation, a time-variant version of the directed transfer function called adaptive directed transfer function has been developed to account for the non-stationarity of iEEG signals in localizing seizure onset zones (Astolfi 2008, Wilke 2008,Wilke 2009. 2011, van mierlo 2011, 2013). However, the frequency band included in all these studies were up to mid gamma range (< 50Hz) and did not include HFOs.

An extensive number of studies have entertained the idea of channel-wise analysis of iEEG using univariate and bivariate methods for seizure predictions (listed in Mormann 2005). These studies have relied on brute force signal processing of iEEG recordings using purely algorithmic approaches in order to construct a measure that would predict the ictal onset and location. The results of these studies provided strong evidence for the existence of a theoretical measure for seizure prediction and showed remarkable accuracy in seizure prediction, in some cases (REF). However, methodologically, these studies relied heavily on the model of focal brain areas initiating a seizure, a concept that has been seriously challenged in recent years (REF?). Moreover, these prediction algorithms required extensive training/tuning of their parameters using within-patient iEEG signals. Despite the initial optimism due to excellent performance of these logarithms on limited set of patients, all such detection algorithms failed when tested using rigorous statistical measures in multicenter studies (Mormann 2005 and 2007).

Another methodology, implementing a graph theoretical analysis using iEEG signals, has been used to determine the temporal evolution of the network architecture in the preictal-ictal-postictal (P-I-P) period. As a measure of seizure network activity, the clustering coefficient C, reflecting network segregation, and shortest path length L, reflecting network integration, have been used to determine the functional architecture. The overall picture points to random (low C and L) → small world (high C and low L) → regular (high C and L) network architecture during the P-I-P period (Ponten 2007, ponten 2009, wu 2006 exp neuro).  These transitions were suggested to be due to monotonically increasing clustering coefficient and shortest path length values from random networks to regular or ordered networks (Ponten 2007, 2009). These studies provided support that interictal and preictal randomly connected networks allow rapid phase transitions and may be more inclined to synchronize and initiate ictal activity (Netoff 2004, Percha 2005, chavez 2006).

Most recently, Mierlo and colleagues used the out-degree reflecting the number of outgoing connections of each iEEG electrode contact to define the SOZ, which was comparable to the SOZ identification by the epileptologist (Mierlo 2013 epilepsia). This study introduced a novel mathematically objective method of localizing the SOZ, which was independent of the traditional methods requiring the expert opinion of an epileptologist (van Mierlo 2011, neuroimage and 2013 epilepsia).

However, while the recent studies showed significant advantage of the use of graph theoretical analysis in localizing seizures, the available literature has major conceptual and technical shortcoming because: (1) There is a disconnect between seizure locus and time prediction parameters and biological mechanisms of epileptogenesis.  This, in our opinion is the major reason for failure of thus far proposed detection algorithms when applied to a new patient population non involved in algorithm training/tuning.  The solution again in our opinion should take into consideration an epileptogenic marker such as HFOs with strong experimental and theoretical evidence and not rely on brute force signal processing. Therefore, we are proposing to determine the dynamic topological changes of brain networks in interictal-ictal-postictal periods using in the HFO band (HFO connectomics of seizures). (2) Only a few studies have determined the topological structure of functional brain networks during seizures using graph theoretical analysis (refs), but none of these studies have been able to generate testable hypothesis regarding mechanisms of seizure generation; (3) There exists a large gap between the functional brain networks and their underlying structural correlates. Several descriptive studies have investigated the morphological alterations of cellular and synaptic components of cortical microcircuit in human epilepsy. However, the direct correlation of such alterations with electrographic signals in human patients is totally missing. Therefore, we are proposing to determine the structural alterations of cortical microcircuits at the cellular and synaptic levels in the SOZ, SSZ and silent areas determined by iEEG recording. This goal will provide first hand data regarding the epileptogenic alterations and will help formulate testable hypotheses directly related to mechanisms of seizure generation; (4) The structural (e.g., cellular and synaptic) underpinnings of abnormal seizure networks in human patients remain unknown, limiting translations of bench findings made in animal models to the human epilepsy.  

In conclusion, the collective evidence reviewed in this section strongly suggests the presence of abnormal seizure networks and identifies the critical gaps in our knowledge regarding the role of pathological HFOs in epileptogenesis as well as the cellular underpinnings of HFOs. Such knowledge is important because it would help monitor the onset of seizure for its prevention, provide accurately targeted surgical resection of epileptogenic core network, and ultimately allow for the development of novel treatment opportunities for patients with intractable epilepsy.