Connected Brains: news

The Web Connected Brains

NetSci 2017 - For everyone who loves graphs

Date: 1-7-2017

Contributed by: Willem de Haan

This year’s edition of one of the most exciting yearly network science conferences, “NetSci", was held this June (19-23) in Indianapolis, USA. This annual international meeting aims to bring together people working on network analysis from every field imaginable, from those trying to analyse behavioural patterns in worms to those interpreting behavioural patterns from presidential tweets. In theory they have much in common, but can they inspire each other in practice? Yes, they can. Here, I would like to give an overall impression, which is inevitably incomplete, subjective and focused on the brain (surprise). For a full program and other details, please visit the NetSci 2017 website.

To start with a few general observations: the brain was actually well represented at the conference, judging from the number of plenary lectures, sessions and poster presentations on the topic. As expected, these were mainly provided by physicists, mathematicians, neuroscientists and, clearly a minority, medical doctors. Most presenters seemed to have kept the multidisciplinary nature in mind: while content could get fairly technical (especially for the mathematically less gifted), the quality of most talks ensured good accessibility for everyone. The conference planning also facilitated cross-disciplinary communication, for example with “lightning talks”, 5 minute-presentations that enabled to quickly get an idea across in the plenary hall right after keynote lectures, thereby enabling people to open their minds to new developments in unfamiliar fields. Another positive aspect (with our own research group in mind) was the substantial contribution of neurophysiology (EEG, MEG, eCog, etc. ) in a (still) fMRI-dominated world.

During the two-day satellite symposia, sessions focused around a central theme, such as ‘machine learning and network analysis’, ‘statistical interference for network models’ or ‘network controllability’, and were attended and/or chaired by prominent network researchers like Barabási, Boccaletti, Vespignani, Fornito, and many others.

For me personally, the Network Neuroscience satellite, organised by Olaf Sporns and others, was the most relevant day. Several recurring themes were noticeable throughout the conference. First of all, many presentations dealt with the core challenge of relating network structure to dynamics.

In a very entertaining talk, John Beggs discussed high resolution neuronal activity measurements in slices of cortical tissue, which, using a combination of transfer entropy and partial information decomposition, enabled an investigation of the extent to which a neuron modifies incoming information depending on its topological location. Clear hub and rich-club structure was found, and most computation seemed to be happening in neurons that received input from high out-degree neurons, while in-degree did not seem to play a big role.

Michael Schaub talked about ‘Slow-switching assemblies’, i.e., sustained epochs of increased firing within assemblies of neurons which transition slowly between different assemblies throughout the network. He showed that the emergence of these is closely linked to spectral properties of the network (especially the spectral gap), that they happen on multiple timescales, and that small-world topology facilitates them.

As if the nematode C. Elegans was not cool enough already, Eduardo Izquierdo had a spectacular demonstration of their projects trying to use bottom-up, evolutionary approaches to understand the (motor) behaviour based on the fully known connectome. This is not limited to the (non-spiking!) neurons however, but also including the entirely known neuromuscular junctions, gap junctions, muscles etc. The big discussion is whether movement is generated by central generators or proprioceptive feedback, and this question is not yet resolved, although the latter seems more plausible.

Stephen Larson showcased the ‘OpenWorm’ project, including beautiful high resolution images that indicate the multiple axonal sites which the synapses in C.elegans are targeting, giving rise to different influences. Furthermore, optical imaging is now enabling a realtime view of the full connectivity in behaving worms!

Speaking of impressive online connectome databases and software, Nicholas Cain of the Allen institute demonstrated the great graphical user interface of the full mouse connectivity atlas, available at connectivity.brain-map.org (software goo.gl/D41Myu), Alex Leow introduced ‘NeuroCave’, a VR-compatibel connectome visualization (creativecodinglab.github.io/NeuroCave).

Second, there was a strong interest for community structures within networks. Aaron Clauset discussed his recent Nature Communications paper (co-author Mark Newman) in which ‘metadata’ (additional information about network nodes) can be used to more accurately detect community structure within networks.

Claus Hilgetag was stranded on a New York airport due to bad weather predictions, but delivered a great live online tele-presentation, in which he reviewed main brain network elements, pointed out the many different definitions for network communities, and reviewed his ‘SER model’, which suggests that functional connectivity is substantially influenced by the structural modularity of a network. Longer stable functional states seem to be supported by a hierarchical modular structure.

Flaviano Morone discussed methods to find the set of ‘essential’ nodes in a network (‘influencers’) using optimal percolation theory. This set appears to be smaller than expected, and often includes nodes with weak links that can easily be overlooked otherwise. See the Nature paper and PNAS paper . This theory can have consequences for interventions trying to control brain network activity.

Daniele Marinazzo talked about visibility graph techniques to map multivariate time series to a multilayer network (Lacasa et al. Scientific Reports 2015). He also argued that the horizontal visibility graph measure has benefits over the traditional method (). He showed applications of these techniques to describe network features like modular structure in fMRI data. He also mentioned neurovault.org as an online archive to share network science data.

In the session on function and dysfunction in the human brain, I was happy to share some results from our recent project which deals with modeling network interventions in a computational model of Alzheimer’s disease. The model, which involves neural masses coupled according to human topology, generates neurophysiological data that can be analysed in the same way as patient data, and I demonstrated a comparison of different intervention strategies that alter neuronal excitability in order to maintain functional network integrity.

Danielle Bassett discussed several recent advances in network science useful for understanding human learning, such as description of adaptive neuronal processes, dynamics during neurodevelopment, and susceptibility to external modulation. She emphasised the mesoscale and the existence of (multilayer) modules in particular as the most relevant scale to find explanations for these phenomena. One example was a method to describe the flexibility of network nodes in terms of belonging to different functional communities at different times. She demonstrated that this is a beneficial feature, although in people with schizophrenia flexibility appears to outweigh other important topological features. Other findings were a predictive value of resting-state connectivity between calcarine and frontal areas for motor learning (e.g. after stroke), and the factors that seem to determine who is a good (motor) learner: release of frontal top-down control, structural network integrity, and functional autonomy. Also, Bassett mentioned the importance of the ‘controllability’ field within network neuroscience with regard to understanding human learning. For community detection, she favored stochastic block modeling and non-negative matrix factorization over the traditional graph theoretical modularity measures.

In a session on networks and genetics, Jonas Richiardi told about new ways to study the genetic influence on brain networks, showed that resting-state fMRI networks can be predicted by using measures of gene expression, and that many of the genes identified in this case are related to voltage-gated ion channels and synaptic function. Neda Jahanshad demonstrated results of GWAS studies from the ENIGMA consortium.

Olaf Sporns then shortly discussed the recently launched Network Neuroscience journal, and invited the participants of the satellite to contribute for a special edition scheduled to appear in the beginning of 2018.

Closing the neuroscience network satellite was a panel discussion with all speakers and questions from the audience. Most of the questions referred to current dilemma’s and expectations or recommendations for future studies, and very different answers were produced by the group. After this, everyone was thirsty and drinks were being served, but not without an equally refreshing poster session featuring more brain network studies.

On the first day of the official NetSci conference, morning and afternoon sessions focused again on brain networks. There were many highlights; the following list is just a brief summary. Marcus Kaiser demonstrated that new high-resolution connectome data from the Human Connectome Project enables unprecedented levels of detail: reproducible networks with 50,000 nodes are now available. This enables the analysis of networks within brain regions, and it appears that many regions exhibit a clear, spatially localised modular structure. As one would expect, long-range connections bridge different modules, and short-range association fiber connections are within modules. See also and also. Brendan Chambers discussed the results of mapping and comparison of propagating activity in large neuronal ensembles from mouse neocortex and also in a naturalistic network model. He showed that functional connectivity patterns can be quite different than the underlying structural topology, and that certain motifs (“fan-in clustering”) of synaptic connectivity coordinate post-synaptic activity to a larger extent than others, with a prominent role for weak connections. When synaptic weights are made artificially stronger, so that cooperative input is less crucial, dynamics become less dominated by fan-in triangles, and reflect the underlying pairwise structure more faithfully. This shapes the routing of activity in the cortex.

Nathanial Rodriguez talked about optimal modularity, which is the network state in which modularity is just strong enough to allow global cascades. Spiking neuronal networks of rodents organise themselves according to this optimal level of modularity. There is a relation between optimal modularity and graph spectral properties (spectral gap). Enrico Amico’s presentation was about Connectivity Independent Component Analysis (ConnICA), a new method which is able to merge structural and functional connectome data in a ‘hybrid’ matrix, and extract a kind of fingerprint of the human brain. This technique can help to find main traits in different, combined datasets. For an example of this technique on Alzheimer patients see here.

Fabrizio de Vico Fallani introduced a criterion (ECO) to filter connectivity based on the optimization of fundamental properties of complex systems, ie efficiency and economy. This can help to address the question how to objectively fix a connectivity threshold. This “optimal” density gives sparse networks (average node degree k~3) yet emphasizing the intrinsic structural properties, which can help to compare biological networks. Recent PLoS Computational Biology paper. How about whole-head electrocorticography? Well, not exactly, but Richard Betzel integrated data from intracranial EEG recordings of 88 patients into one whole-brain functional network, and showed a correspondence with structural networks derived fro diffusion imaging methods. The relationship was dependent on frequency band, and longer functional connections were tied to relatively lower frequencies. Of course, the data was derived from epilepsy patients, placing doubts on the generalizability.

The final talk of the afternoon was provided by Sarah Morgan, and consisted of an interesting use of network motifs, on which a principal component analysis (PCA) was performed to show that two PCs explain a remarkable 97% of motif variability. With these, a low-dimensional motif ‘morphospace’ was built. The first PC correlates with small-world topology, efficiency and the Euclidean distance of the longest edges. An economical clustering model from the same group (Vertes, 2012) can largely reproduce the motif profiles and morphospace, unlike other models.

All in all, NetSci 2017 was a very inspiring meeting, with lots of creative and well-presented ideas. The enthusiasm about the role that network analysis plays, and will play in the future in brain research was obvious and infectious. Next year’s edition will be in Paris, which will make it an offer hard to refuse for European network aficionados. Save the date!

Funding for the development of a computer model of the brain to improve epilepsy surgery outcome

Date: 11-3-2017

Contributed by: Arjan Hillebrand

A collaboration between the Departments of Clinical Neurophysiology, Neurology, and Neurosurgery of the VU University Medical Center and the Network Architectures and Services group of Delft University of technology has received a grant from ZonMw and the Nationaal Epilepsie Fonds (NEF). The project is part of the Programme Translational Research, and aims to develop computer models of brain networks that can be used to improve the outcome of epilepsy surgery.

Neurosurgery is still the most often used treatment for patients with epilepsy that does not respond to medication. When it is likely that the seizures start at a given location in the brain then removal of that area can stop the seizures. However, epilepsy surgery is not always successful, since it is difficult to determine which area needs to be resected. A computer model of the patient’s brain can help. The aim of the project is to increase the outcome of epilepsy surgery so that more patients will be seizure free after the operation and will therefore have a better quality of life.

Nowadays, patients are examined using a variety of different neurophysiological and imaging techniques during the presurgical work-up. Results from these techniques are used to determine the location of the area where seizures start. These techniques can also be used to develop an individualised computer model of the patient’s epileptogenic brain network, using information from the brain’s anatomy (measured using DTI) and activity (measured using MEG). The spreading of seizure activity can be simulated in this computer model, and we can subsequently examine the effect of different resection strategies (virtual resections) on the seizures. A major advantage of this model is that it allows us to carry out a large number of such virtual surgeries, and thereby to predict and optimize seizure freedom in a safe and cost-effective manner.

Highlights from “Update on the Human Brain Connectome: From physiology to diseases” - Rome, November 2016

Date: 10-12-2016

Contributed by: Willem de Haan

On 28 and 29 November, the medical faculty of the ‘Universita Cattolica del Sacro Cuore’ in Rome was the stage of a focused conference on human brain connectivity research. Although the scale was modest, with a single plenary session room, the speaker lineup consisted of various renowned experts. A more elaborate report of the many fascinating talks will appear in the upcoming year, so this short note will just mention a few of my personal highlights.

The conference was opened by its organizers Paolo Rossini and Mark Hallett, and started with a very nicely illustrated historical overview of connectivity as concept to explain brain function by prof. Marina Bentivoglio. She also pointed out the underestimated role of astrocytes in synaptical function, and argued that microconnectivity patterns seem too complex to describe with the node-and-edge method of graph theory. Bharat Biswal was unfortunately unable to attend the conference, but had recorded a multimedia presentation in advance that focused on fMRI connectivity research so far, and discussed new answers to presumed methodological limitations of the field. Franco Pestilli discussed network analysis methods, including DTI maps Pestilli ea Nature methods 2015). He discussed difference between deterministic and probabilistic maps, and demonstrated LiFE (Linear Fascicle Evaluation), freely available software that can analyse the quality of connectome data. He also argued that, contrary to often-heard statements, the imaging research field is quite critical of itself.

Christian Gerloff talked about connectome analysis in stroke, using various methods such as graph theory, dynamic causal modeling and game theoretical approaches. One of the interesting parts was the finding that individual connectome data can enhance clinical predictions based on EEG. He also pointed out that people who recover from stroke often regain normal connectivity levels, which is particularly the case for functional connectivity. Importantly, he argued strongly in favor of the use of modeling and multimodal approaches to understand the connectome. Ulf Ziemann discussed a very interesting setup using pharmaco-TMS-EEG in order to study relations between neurotransmitter/receptor traffic and connectivity. Gereon Fink also focused on stroke an connectivity, including TMS and tDCS studies. He showed that these techniques have the potential to improve motor function, praxis and even neglect (after a single session!). Risto Ilmoniemi suggested to use TMS-EEG to develop a map of connectome function. Various techniques were discussed: single pulse, repeated TMS, multiple coil techniques and closed-loop TMS.

Fabrizio Vecchio gave a nice overview of EEG and MEG connectivity analysis, and was so kind to quote several papers from our group as examples. Following this talk, there was some discussion about the potential of MEG to measure deep sources of neuronal activity in a reliable way. Clearly there were believers ad non-believers! Michael Nitsche discussed many tES studies, including those by Polania ea in the previous years. One of them is transient alternating current stimulation (tACS), which modulates existing neuronal activity, rather then inducing spiking behavior or shutting neuronal circuits down.

Overall, an exciting first day, which strangely enough for an Italian event did not include coffee breaks in the morning, but did include a great villa dinner for all.

The next day, two very inspiring talks were delivered by Wolf Singer and Marcello Massimini, who talked about the neural correlates of qualitative and quantitative aspects of consciousness, respectively. Singer started by quoting the visionary Sherrington, who claimed that consciousness is about brain dynamics. He then explained the dynamic binding hypothesis and the mechanism of flexible binding based on phase adjustment. In his opinion, the transient and precise sync (in the gamma range) of widely distributed neuronal assemblies is the best neural correlate of consciousness at the moment. He also nicely pointed out that the ‘outside world’, in particular the statistical correlations of aspects in the world around us, is represented in the brain by the weight distribution between the connections, which may explain why certain binding paths are dominant and others are not. Connections are everything, and, quoting Tononi, everything that impairs long distance coordination, impairs consciousness. Massimini talked about many TMS-EEG experiments, including sleep studies (‘since REM-sleep can be reported afterwards, we are conscious during this phase’). He pointed out that synchrony is not a good marker of the level of arousal (as measured for example by the bispectral index during surgery). Conditions like alpha coma and status epilepticus in absence epilepsy reflect conditions where synchrony can be high, but consciousness is definitely disturbed. He also argued in favor of pertubational approaches to study the connectome. He compared TMS to a loud ‘knock’ on the brain, in order to study its ‘echo’, which makes it unsensitive to qualitative consciousness research, but helpful for quantitative aspects. He also mentioned the special effect of ketamine: people can be fully anesthetized, are able to undergo surgery, but are able to report about it afterwards. He further discussed the use of ‘compressability’ measures as a measure of complexity.

In the session on neurodegenerative disease Paolo Maria Rossini discussed many findings in Alzheimer EEG/MEG research, including TMS-EEG techniques that demonstrate for example increased neuronal excitability in AD. Reinhard Dengler discussed the growing interest in network activity in motor neuron disease (eg ALS), which has a very fascinating link with frontotemporal dementia (FTD). He demonstrated that functional connectivity changes occur before structural changes. For the closing talk of the session I was happy to show the results of our recent computational modeling work, which involves a new approach to test etiological hypotheses and interventions in a dynamic neurophysiological model of the human connectome (de Haan ea, PLoS Comput Biol 2012).

Yong He’s talk was a very comprehensive summary of many mainly fMRI-based results, including several works by Martijn van den Heuvel. In his presentation about epilepsy and connectivity research, Ivan Rektor gave a clear overview of recent studies, including some by our collaborator Eric van Diessen. Mark Hallett discussed findings in movement disorder connectivity research, such as tremor and dystonia. Finally, Riccardo di Iorio presented promising work on phantom limb syndrome, which showed that the use of a robotic hand seems to improve pain sensations and motor performance.

Overall, it was very good to see the growing interest in connectivity and network analyses (including graph theory), and the added value of brain stimulation techniques such as TMS/tDCS combined with EEG/MEG. Although the term ‘network’ doesn’t mean the same to everyone (ranging from microscopical anatomical circuits of several neurons to ICA-based regions of interest to graph theory-derived macroscopic topology descriptions), the well-balanced program and lively discussion certainly facilitated cross-disciplinary thinking, which is exactly what is needed to bring our overlapping worlds together.

Neuroscientific crowdfunding at CROWDF€ST

Date: 3-10-2015

Contributed by: Linda Douw

Last weekend, the first live crowdfunding festival CROWDF€ST took place in Amsterdam. Besides the largest 3D scanner of Europe, various food trucks and wonderful music performances, visitors could also enjoy live anatomy lessons on stage and ask neuro nerds from the VUmc department of Anatomy and Neurosciences their burning questions about the brain. They were there to generate attention for their crowdfunding campaign "The brain network from brain cell to scan". For more information go to: Anatomy-neurosciences

Their aim is to improve treatment of brain diseases by first gaining a better understanding of the healthy brain. The brain is in fact similar to social networks like Facebook, both in terms of individual brain cells under the microscope, as well as when mapping interactions between larger brain regions with for instance MEG or MRI. To understand all those different levels of brain networks in people with brain diseases, we need a comprehensive analysis of the healthy brain network: what is different from this healthy situation in brain disease? What goes wrong in Alzheimer's or Parkinson's disease? The investigators have therefore started a database, in which healthy brains from people who have not died from brain diseases are collected. With the proceeds of this campaign, three healthy brains and brain networks at all levels will be mapped with the latest technologies, such as advanced microscopy and MRI scans.

During CROWDF€ST, almost a third of the amount needed for this campaign was already raised! Still want to participate? Go to crowdfunding .

Research grant to improve cognitive symptoms with tDCS in Parkinson's patients

Date: 1-11-2014

Contributed by: Linda Douw

Many people who suffer from Parkinson's disease suffer from cognitive symptoms, especially problems with the so-called executive functions. These cognitive functions include planning of daily activities, maintaining overview in a variety of situations and the flexibility to switch between different tasks. One way to address these cognitive problems is ‘transcranial direct current stimulation’ (tDCS), during which very small electrical currents are applied to the outside of the skull. This method seems to have a positive effect on cognitive problems in general and is well tolerated by patients. However, it is not yet clear what the best type of stimulation is, whether the technique mainly improves executive functions, and how functional communication within the brain alters because of this intervention.

Neuropsychologists Martin Klein (Department of Medical Psychology) and Linda Douw (Department of Anatomy & Neurosciences) received a grant from the Parkinson Association to investigate tDCS for cognitive problems in Parkinson’s disease, in cooperation with the departments of Rehabilitation Medicine, Clinical Neurophysiology and Neurology. They will treat a group of Parkinson patients for a week with placebo stimulation, or one of two types of tDCS. Executive functioning is measured before and after the treatment. Furthermore, magnetoencephalography (MEG) will be performed on a number of occasions, in order to see whether certain functional connections change after stimulation. The researchers hope be more equipped to help Parkinson patients with cognitive problems after completion of this study, by increasing knowledge about the effects of tDCS and gaining experience with this treatment modality.

Young investigator award for Edwin van Dellen at Biomag 2014

Date: 4-10-2014

Contributed by: Edwin van Dellen

Edwin van Dellen received the Young Investigator Award at the 19th International Conference on Biomagnetism in Halifax, Canada. He received the prize in the category ‘Applications’ for his work on MEG network analysis in lesional epilepsy. Three Young Investigator Awards (500CAD each) were awarded to post-doctoral fellows, within five years of obtaining a PhD. The award was given for work presented at the conference and curriculum vitae, and was based on the originality of contribution, presentation skills, and significance and anticipated impact of findings.

The presented work was based on a longitudinal MEG study in lesional epilepsy patients, a collaborative project of the departments of clinical neurophysiology, neuro-oncology, and neurosurgery. Several functional network parameterswere shown to correlate with epilepsy burden, and these parameters normalizedonly in patients who became seizure-free after surgery. These results increase insight in functional network changes insuccessful epilepsy surgery, and may eventually be utilized for optimization of neurosurgical approaches.

BioMag 2014

Date: 4-10-2014

Contributed by: Bob van Dijk, Arjan Hillebrand, Prejaas Tewarie and Edwin van Dellen

The 19th International conference on Biomagnetism (Biomag), which is a biennial conference mainly focused on MEG methodology and applications, took place this year in Halifax, Canada. This city is surrounded by water, woods and then some more woods, which made it the perfect setting for our talks on trees in the brain. It is interesting to see progress in the field compared to the conference of 2012 (see also the report on Biomag 2012 on this website), for example in dynamic network analysis. Adam Baker presented interesting work from the Oxford group, using a novel approach that identifies the points in time at which unique patterns of activity recur. They found transient (100-200 ms) brain states with spatial topographies similar to those of well-known resting state networks. Furthermore, they found a dynamic anti-correlation in activation between the default mode network and parietal regions of the dorsal attention network, whichwas consistent with an inability of the system to switch directly between two transient brain states. One of the hot topics of Biomag 2012 was the relationship between GABA expression and gamma band activity. A previous PNAS paper from the group of Krish Singh suggested that GABAergic signaling underlies gamma oscillations, but this could not be replicated in a larger sample by another group, putting this previous finding in perspective.

Surjo Soekadar, as well as several other groups, presented a method to perform transcranial direct current stimulation during MEG recordings. This method may have important applications, since it allows for direct analysis of stimulus-related alterations of neural activity in source space.

Another hot topic was cross-frequency coupling. There were a number of papers describing novel ways to detect coupling and directionality of coupling between oscillators at different frequencies. Most groups working in this field use the Brainstorm toolbox developed by the group of Sylvain Baillet, where others use Fieldtrip. Sylvain Baillet showed that cross frequency coupling is changing between active and resting states; a number of posters and papers described theta phase and gamma power coupling during memory tasks, and there were also a few papers describing cross-frequency coupling in (rhythmic) motor tasks, or disturbed cross-frequency coupling in patients with motor problems (in stroke patients by Nina Forss or in Parkinson patients by Vladimir Litvak). A very important contribution was by Michael Wibral, who cautioned about the interpretation of cross-frequency coupling and stressed the importance of correctly applying the tools from signal analysis. He showed that cross-frequency coupling can easily be found in the absence of underlying coupling. One very obvious confounder is non-stationarity. If people want to study cross-frequency coupling they are advised to read this paper, and follow the guidelines described there; unfortunately the cross-frequency analysis toolboxes do not restrict users to the proper parameter choices (as became obvious in the Brainstorm satellite workshop).

Another interesting topic during the conference was networks in epilepsy. Sam Doesburg presented work published in Brain (2014) were they investigated the effect of interictal epileptiform discharges on resting state networks and cognitive performance in children with epilepsy. To this end, they performed a multi-modal fMRI/MEG analysis, where they used the spatial information of fMRI for the analysis of resting state network dynamics of MEG. In addition, they also quantified large-scale network topology using graph measures. Their main findings were that large-scale network topology was shifted towards more regular networks both before and after interictal epileptiform discharges, which was interpreted as causal role for interictal epileptiform discharges. Secondly, networks of children with strong resilience to interictal epileptiform discharges were characterized with high activity of the posterior cingulated and precuneus. Lastly, the presence of interictal epileptiform discharges was strongly related to cognitive performance in these children.

Yet another hot-topic was decoding algorithms. Decoding here means predicting a stimulus or a state provided features from MEG data. One inspiring symposium was devoted to this subject with a nice introductory lecture by Alexandre Gramfort. In this symposium Stefan Haufe showed how the weight vectors from a trained decoder can be used to interpret in a completely data driven manner the decision or categorizing processes. And both LauriParkonnen and Alona Fisher described time-resolved decoders that can accurately distinguish word categories or pictures. In fact one of the data challenges of BIOMAG2014 organised by Emanuele Olivetti challenged to decode pictures from faces vs scrambled pictures from faces using MEG-data as input. The winner Alexandre Barachant (from France, but not from the MEG field) scored 78% correct.

In terms of hardware, there were three noteworthy developments. The first developments are the attempts to replace the squid-based sensors using novel technologies, which is of particular importance for the future of the field given the finite supply of helium. Atomic magnetometers seem to have, potentially, the right combination of high sensitivity, reasonable noise levels, and ease of manufacturing, for use in MEG systems (see e.g. here). In his key-note lecture, Gareth Barnes provided an upbeat view about the future of MEG. He argued that MEG is, and should present itself, as a mature neuroimaging technique with not only high temporal resolution, but also high spatial resolution. He gave the example that the use of a printed 3D head-cast enables repeated recording on the same subject, as typically done in invasive animal studies, thereby significantly increasing the signal-to-noise ratio. He showed that the spatial resolution that can be achieved in this way even enables the distinction between different cortical laminae.

A third technical development was the use of Kinect to improve 3-D digitization of the head. Kinect is a video based tool widely used for gaming (Xbox), which uses a RGB video camera and a depth sensor to capture infrared light. Accurate 3-D digitization of the head is required for optimal MEG-MRI co-registration. During the conference, the developers claimed that the Kinect improves 3-D head digitization by 137% and 50% over the classical Polhemus system and NextEngine laser, respectively. The explanation for this increased accuracy was related to capturing 2000 times more surface points in one third of the time compared to the Polhemus system. Therefore, the developers encouraged all MEG users to use Kinect based head digitization over older methods.

Symposium "BrainTrees 2014"

Date: 23-6-2014

Contributed by: Kees Stam

Research in complex brain networks is flourishing, and many new papers describing graph theoretical studies of complex brain networks, both structural and functional, are published every week. However, since graph theoretical analysis of brain networks is still quite new, no dedicated conferences or symposia on this topic exist. Usually, such studies are presented in the context of conferences on related topics, which does not alway create the environment for a critical and detailed discussion of network issues. The symposium "BrainTrees 2014" is an attempt to have a one day meeting, with invited talks by a large number of experience speakers, all of whom have - sometime extensive - experience with brain network analysis. While the symposium will deal with more general aspects, the main theme is the question how network comparison should be done, and in particular whether use of so called "minimum spanning trees" is of any help in this respect. This explains the title of the symposium. However, many more topics will be covered, ranging from structural to functional networks, and healthy brains to different psychiatric and neurological disorders. Participation is free, but only a few places are left. Contact details can be found in the program. Abstracts of all presentations are available.

PhD position for project "The molecules of behaviour"

Date: 9-2-2014

Contributed by: Linda Douw

Want to do investigate brain networks? There is a job opening for a junior scientist to work on a project entitled "The molecules of behavior". As you may have read on Connected Brains, connectivity and network analysis in neuroscience have been applied on multiple spatial scales, but the interrelations between these scales have rarely been investigated. Determining the molecular basis of anatomical and functional brain networks will help us understand the brain in health and disease. Gliomas, primary brain tumors originating from the supporting glial tissue, are characterized by changes in connectivity and networks. The tumor not only interferes with the area surrounding it, but changes the entire functional brain network. Furthermore, frequently occurring symptoms, such as global cognitive deficits and epileptic seizures, are related to these brain-wide alterations. However, both from a general neuroscientific persepective and in the field of neuro-oncology, a number of questions remain. Glioma patients are an interesting population to study in this setting, not only because gliomas have clear network effects, but also because combined analysis of tissue and neuroimaging is feasible. This population therefore provides a unique window of opportunity to investigate how growing lesions affect molecular processes, microscopic network properties, large-scale networks based on neuro-imaging, and cognition at the same time. In this project, the brain network will be investigated on multiple spatial scales in glioma patients, as well as molecular processes in and around the tumor, which will for the first time elucidate (1) how anatomical connectivity/networks are affected by glioma, and how the association between anatomy and function is altered, (2) how microscopic network patterns relate to macroscopic connectivity patterns in glioma patients, (3) how molecular tissue properties relate to local and global network topology in these patients, and (4) which factors are most predictive for patients’ cognitive and neurological status. Interested? See here

The Networked Brain. Cell Symposium, San Diego, 8-9 November 2013

Date: 18-11-2013

Contributed by: Ilse van Straaten

On November 8 and 9 the Cell symposium ‘the networked brain’ was held in San Diego in honor of the 25th anniversary of the journal Cell. The aim was to integrate network studies at different scales, from structure to function, from applied to computational. The organizers had put together an interesting program with some recognized leaders in the different neuroscience network fields. Although the quality of the presentations was high, the presented work was using network and graph theory in the methodology rather than interpreting the results within this framework. (Also, it appeared fashionable to present photographs of the researchers that were cited.)

Olaf Sporns kicked off and explained the basic terminology and goals of network analysis. He stressed the beauty of this field: the applicability across scales and modalities to integrate results. The concept of the connectome was addressed: based on Hagmann’s parcellation of brain regions and fiber tracking methods, the structural matrix shows behavior similar to many other complex networks including a skewed degree distribution, modules, and a structural core. Next, computational modeling was introduced as a method to test hypotheses. Sporns gave an example that a connectome-based model of functional connectivity could predict fMRI, without going into detail on the possible mechanisms. He also referred to the work of Martijn van den Heuvel on the relationship between structural and functional connectivity and the overlap between DTI and fMRI networks. He concluded by saying that independent component analysis may be gone in five years, but network theory based brain analysis has just begun.

Steven Petersen explained his work on the search for network vulnerable sites, since damage to central nodes has large effects on the network structure. His group used an fMRI-based atlas of communities (components that have similar BLOD signals) and from that, computed the participation coefficient to explain the cognitive effects of small focal lesions distributed over the cortex. In a study with 10-20 subjects with lesions in different locations in the brain, they found that the lesions in regions with a high PC had a larger effect on language and other cognitive performance. The interesting part was the counterintuitive results of the PC: defined by high PC, the hubs locations were not the ‘classical’ hub sites, such as the posterior cingulate and hippocampus, but at a fronto-lateral site and angular gyrus. On the other hand, these regions are well known for their effect on language and cognition, and it remains unknown whether PC is causal in this respect.

The talk of Daniel Geschwind showed how networks can be generated from gene transcription: transcriptions that are involved in coding the same protein are said to be connected. This way, a complex network arises (the transcriptome) and groups of genes, and the relationships between these groups rather than individual genes, can be investigated for involvement in health and disease. For example, gene transcription of certain gene modules is different in autism from controls very early in gestation, resulting in a change in transcription of another gene module, involved in inflammation. This may affect functional connectivity and ultimately behavior.

Michael Greenberg initially feared that his contribution was mainly ‘for comic relief’, since instead of networks, he has studied one gene at a time for the last 35 years. Although humorous at times, the main message was very serious and interesting. He reported the effect of activity-dependent gene transcription on changes in neurotransmitters and thereby changes in synaptic plasticity and neuronal circuitry. Gene transcription is one step in the production of synaptic receptors, and this production involves a complex cascade. Apparently, this transcription is influenced by activity of the cell. As an example, he showed that in Rett syndrome, the transcription of a gene coding for one protein reacted differently to activity than controls. What his information may implicate for macroscopic networkers is that this can be seen as a mechanism that can account for differences in gene expression in different brain regions because of differences in cellular activity. This might link somehow the findings of de Haan et al (PLoS Comp Biol 2013) that found differences in loss of connectivity dependent of activity. This might also link to the fact that the apoE4 gene has a higher risk for amyloid deposition in Alzheimer’s disease in the intercellular space, and in cerebral amyloid angiopathy in the small-vessel walls: the genetic composition is similar, but the effect of activity on gene transcription may be different for the two conditions.

Verginia Lee showed work on two hypotheses on how degeneration-associated products (such as tau, amyloid beta, and alpha-synuclein) spread throughout the brain. Protein aggregation and spreading is common for all neurodegenerative diseases, and the spread has a distinct pattern, as shown for example by Braak and Braak for amyloid. One hypothesis was that this spread was due to spread of misfolded proteins. She showed results of experiments with Lewy bodies, in which seeds (pieces) of misfolded Lewy bodies were fed to cells, incorporated in the cells’ axon and transported to the rest of the cells, and even to other cells, partly by trans-synaptic transportation, following the white matter tracts. The cells get sick and die also according to this pattern. Although it did not work so well for tau, and amyloid has not been tested yet, she pointed out that in mice in only 3-6 months a spread throughout the entire cortex could be seen, mimicking the disease spread in humans. This work might suggest that we can see degenerative diseases to be ‘intra-individual infectious’, just as prion-related diseases. The second hypothesis was called ‘the strain hypothesis’. She wanted to have an explanation for the fact that in AD, DLB, and Parkinson’s disease there is an overlap since in all these diseases, tau aggregates and Lewy bodies can be found to some extent. She showed that alpha-synuclein seeds can under certain conditions give rise to Lewy body aggregates and in other conditions to tau aggregates, in genetically prone mice.

Michael Greicius reported his work on fMRI networks as tools to show differences between groups. He showed a lot of ICA-based data (thereby ignoring the warning of Olav Sporns). He showed the DMN to be involved in AD, and the salience network to be involved in FTLD. Randy Buckner had a very speculative talk, due to the fact that he was the last speaker of the day, and most of the background and methods that he uses in his work was already addressed. He elegantly took ‘advantage of this disadvantage’ to show his opinions on why the human brain is different from other animals. Our brains are bigger, but why are bigger brains special? First, they have more neurons, even more than brains of whales and elephants, if you were to count them. But second, this exponential increase in neurons during the last couple of millions years is not random. The primary cortices are roughly unchanged in relative size, but especially the association cortices have grown bigger. This results in a different flow of information, since it involves interconnectivity of distributed areas, rather than serial flows of activity which is more characteristic for signal processing in the primary cortices. How can this expansion of association cortex have evolved? Fetal growth factors, which migrate to various cortical areas early in gestation, enhance cortical expansion. Somehow, these signaling growth molecules end up more in the association cortices in humans. After this first stimulus, growth is mainly fueled by activity. Since the association cortices are heavily interconnected, they can be expected to be electrically more active, and therefore grow bigger.

Wolf Singer discussed his view that the brain is organized temporally as it is organized spatially. The density of brain networks is high, and therefore communication on this network requires dynamic coordination. Oscillations are a key part of this coordination. The synchrony between distributed oscillations is dynamically changed according to context, such as arousal or expectancy of a trigger. It does not change the firing rate, but heavily changes oscillations, mainly in the gamma band. Communication only occurs when signals are received in the appropriate phase, and mainly with a zero-lag phase difference. Adjusting oscillation frequency and phase modifies coupling between neurons. In addition, the concept of ‘nested relations’ can account for the finding that slow frequency oscillations gate high frequency oscillations. A discussion after the talk addressed the interesting finding of the growing brain that needs to adjust conduction velocity in order to maintain optimal phase coupling between brain areas: studies suggest that the speed of myelination is adjusted by activity.

György Buzsaki defended the statement that ‘the brain is the best prediction-device that we have’. It is constantly fueled by and returns feedback to its environment. In professor Buzsaki’s view, cognition arises when brain activity loops do not longer involve the environment but remain inside the brain. Memory, as an example for cognition, can be seen as navigation in past time, similar to navigation in space. The sequential activation of cell assemblies of memory, or other cognitive function, appears from internal loops by microsecond plasticity.

Next year, a Cell symposium will cover the topic of ‘translational neuroscience’, so should you consider your work relevant in this field, I can recommend the trip to Los Angeles.

New scholarship for brain network project investigating the relationship between molecules and behavior

Date: 13-7-2013

Contributed by: Linda Douw

Linda Douw has received a prestigious Branco Weiss Fellowship of the Society in Science. She will be investigating the assocations between molecular markers and brain networks in brain tumor patients for the next five years, both at the VU University Medical Center as well as at Harvard Medical School and Massachusetts General Hospital (Boston, USA).

In collaboration with the departments of Anatomy and Neuroscience, Neurology, Clinical Neurophysiology, Neurosurgery, Pathology and Radiology of the VU University Medical Center, she will collect both neuroimaging data and tumor tissue of glioma patients. In the tumor tissue, a number of relevant molecular markers that determine the type of tumor, patients’ survival, tumor progression, and response to medication (chemotherapy, anti-epileptic drugs) will be collected. At this time, these properties can be determined only with the aid of brain surgery, and the removal of brain tissue. However, using network theory on the brain may lead to the use of non-invasive measurements to obtain the same information.

The brain can be regarded as an extremely complex network, which is in many ways similar to other networks, such as the railway network of any country and Facebook. By using knowledge we already have on these simpler networks, we have learned a lot about how our brains work, or conversely do not work optimally in neurological disease. At this time, it is unknown how cellular processes lead to the neural network topologies we measure with for instance MRI and magnetoencephalography. Uniting the previously separate fields of microscopy and neuro-imaging on a single continuum, this project will elucidate the molecular basis of anatomical and functional brain connectivity and networks. Furthermore, the project aims to tailor diagnosis and treatment options for brain tumor patients.

For more information on this prestigious grant, see website

National Epilepsy Fund supports network study that aims to aid patients with epilepsy

Date: 24-6-2013

Contributed by: Arjan Hillebrand

A team of researchers and clinicians from the departments of Clinical Neurophysiology/MEG Center, Neurology and Neurosurgery has secured funding from the National Epilepsy Fund (NEF). The 4-year project aims to develop a new approach, based on network theory, to delineate the epileptogenic zone in patients with epilepsy who would otherwise not be able to undergo neurosurgery.

Surgical removal of the epileptogenic zone (the area that causes seizures) in pharmacoresistant patients may lead to seizure freedom. Unfortunately, at least 25% of these patients cannot undergo epilepsy surgery because localisation of the epileptogenic zone is impossible, often due to the lack of interictal epileptiform activity. Thus, more patients would benefit from neurosurgery if the epileptogenic zone could be identified reliably, even when interictal discharges are absent. In this project, we will use information about the brain network as a whole in order to localise the epileptogenic area, thereby avoiding the need to detect epileptiform discharges. This approach is based on recent work from our own group, and others, which has shown that abnormal interictal functional connectivity and brain network characteristics are both related to epilepsy. For example, we have recently shown that only when the regions that had a pathological “hub-status” had been removed during surgery the patients became seizure free after surgery.

In this project we will use a recently developped atlas-based beamforming approach that enables us to better estimate interictal network characteristics. It also allows us to determine more accurately which characteristics have most predictive power regarding the location of the epileptogenic zone as well as regarding postoperative seizure reduction. We will validate our approach against the current clinical gold-standard, which is a combination of EEG, interictal MEG and depth electrode recordings. We envisage that this approach will result in an increase in the effective use of EEG/MEG in the facilitation of recective surgery. Students with a techical or medical background who are interested in pursuing a PhD can contact principle investigator Dr. Arjan Hillebrand.

Update on resting-state connectivity analysis in glioma patients

Date: 20-11-2012

Contributed by: Edwin van Dellen

In a collaboration of the departments of neuro-oncology, clinical neurophysiology, neurosurgery, and neuropsychology, we have recently published two MEG papers on functional connectivity and graph analysis in brain tumor patients. In the first study, we constructed functional networks in sensor space from pre-surgical resting-state magnetoencephalography (MEG) recordings of 13 low-grade glioma (LGG) patients, 12 high-grade glioma (HGG) patients, 10 epilepsy patients with non-glial lesions (NGL), and 36 healthy controls. LGG patients showed decreased network synchronizability and decreased global integration compared to healthy controls in the theta frequency range (4-8Hz), similar to NGL patients. HGG patients’ networks did not significantly differ from those in controls. Network characteristics correlated with clinical presentation regarding seizure frequency in LGG patients, and with poorer cognitive performance in both LGG and HGG glioma patients. We think that differences between LGG and HGG patients’ networks are explained by differences in plasticity, guided by the particular lesional growth pattern. Interestingly, decreased synchronizability and decreased global integration in the theta band seem to make LGG and NGL patients more prone to the occurrence of seizures and cognitive decline.

In the second study, we explored the relation between alterations in connectivity of MEG resting-state networks and changes in cognitive processing after resective surgery in ten low grade glioma patients. Lower alpha band functional connectivity in the default mode network (DMN) was increased after surgery, and this increase was related to improved verbal memory functioning. Right frontoparietal network (FPN) connectivity was increased after resection in the upper alpha band, which correlated with improved attention, working memory and executive functioning. These findings indicate that connectivity of MEG RSNs may be used for presurgical prediction of cognitive outcome in future studies.

A recent fMRI study showed similar results of alterations in the DMN in glioma patients, which were more outspoken in LGG patients than in HGG patients. Esposito and others studied the DMN in twenty-four patients with a left-hemisphere cerebral tumor (14 grade II and 10 grade IV gliomas) and 14 healthy age-matched right-hand volunteers during performance of language tasks for presurgical mapping. Reduced DMN connectivity was detected in tumor patients with respect to controls. A significantly increased integration of the hippocampal areas with the DMN was found, while a reduced integration with the DMN was found for prefrontal regions. Modifications were closely related to tumor grading similar to our findings: the DMN lateralized to the hemisphere contralateral to tumor in the low-grade, but not in the high-grade tumor patients.

In conclusion, these studies all suggest that analysis of resting-state functional connectivity in MEG-recordings may become of clinical value in brain tumor patients. We will have to elucidate which factors determine how resting-state connectivity patterns are altered, as this knowledge may be used for the prediction of surgical outcome regarding epilepsy and cognitive performance. That this is goal is within reach was recently shown by Tarapore and others, who demonstrated the predictive value of MEG connectivity analysis for postsurgical motor- and language performance.

Complex systems and networks symposium 2012

Date: 3-10-2012

Contributed by: Betty Tijms

The Complex Systems and Brain Networks Symposium was a three day conference about complex systems that took place in Delmenhorst, Germany. This symposium was organised as a present for Carsten Giessing, who received a fellowship of the Hanse-Wissenschaftskolleg Institute for Advanced Study.

During these three days, 21 speakers presented the latest state of the art research in the field of complex systems and brain networks, ranging from single neuronal models (i.a., Stefan Bornhold, Christian Finke) to cultured cell models (Dietmar Plenz), large scale brain network analysis (i.a., Gustavo Deco, Petra Ritter) and clinical applications of brain network investigations (i.a., Ed Bullmore, Kees Stam).

The first day focussed at the neuronal level of complex brain networks and started with a talk from Dietmar Plenz. He proposed that neuronal avalanches are an indication that functional brain dynamics is near criticality, which is a state where the brain can optimally respond to sensory inputs and also maximises its information capacity (Pajevic, S., & Plenz, D. 2012. doi:10.1038/nphys2257). He demonstrated that weak links between neuronal populations are important to balance propagation of such avalanches over long-distances while avoiding explosive growth of activity.

The second day focussed on new approaches for brain network analysis, where the common theme was the integration of different brain modelling and network analysis methods. Both Petra Ritter and Peter Robinson gave an excellent general overview highlighting this need for integrated approaches. Gustavo Deco further demonstrated how a static anatomically defined (DTI) network of the human brain provides the basis for critical functional dynamics. He demonstrated how simulated brain function moves away from criticality in sleep, and that during awaking function reorganises according to the underlying anatomical structure. Petra Vertes presented a generative model for brain functional connectivity, that was able to explain differences in resting state networks between people with childhood onset schizophrenia and healthy subjects (Vértes, P. E., et al., 2012. doi:10.1073/pnas.1111738109). Both these presentations demonstrated how computational modelling can be used to generate testable predictions about brain graphs. At the neuronal level, Stefan Rotter showed in an simulated neuronal networks how higher order moments in functional data reflect structural connectivity, which provides support for an important assumption commonly used in brain imaging that describing functional networks can inform about the underlying anatomical network (Pernice, V. et al., 2011. doi:10.1371/journal.pcbi.1002059.t001).

The talks on the third day demonstrated the clinical relevance of brain networks. Interestingly, a common theme in two general talks of the applications of graph theory in a clinical setting is the vulnerability of hubs. Hubs are bottlenecks in complex networks and therefore the weak points. The importance of hubs was further nicely illustrated by the work of David Sharp who presented work on brain dysfunction after traumatic brain injury. Mika Rubinov argued for the importance of analysing weighted networks, and presented a new method to accurately define modularity in weighted brain networks. Sophie Achard presented how brain networks can be used in clinical practice to identify cortical areas in single subjects that are specific for their epilepsy. This method considerably improved the accuracy of surgery in epileptic patients, minimising damage and maximising positive outcome.

Biomag 2012

Date: 20-9-2012

Contributed by: Edwin van Dellen

The 18th International conference on Biomagnetism (Biomag), which is a biennial conference mainly focussed on MEG methodology and applications, took place this year in Paris. The MEG community was apparently eager to meet up, as multiple satellite meetings were planned already two days prior to the conference opening.

One of these satellite meetings was organized by Srikantan Nagarajan from UCSF, and focussed on source space functional analysis of MEG/EEG data. Steven Stufflebeam from the Martinos Center in Boston presented his view on MEG research for the upcoming years, and stated that multimodal imaging is the way to go. He suggested that especially the combination of MEG data with DTI maps, which they now measure on a high resolution Connectom scanner, should be used to study the relationship between structural and functional brain networks (see http://www.humanconnectomeproject.org/). Other talks were more focussed on specific methodological issues for the beamformer approach to localise the neuronal sources underlying the measured MEG data, such as the choice of an appropriate number of ROIs and functional connectivity measures. Jan-Matthijs Schoffelen from the Donders Center in Nijmegen presented a method to ‘blob’ connections in order to define ROIs based on voxels with similar activity. Other speakers suggested to use the imaginary coherence in source space because normal coherence measures would still (partly) represent spurious interactions between source space regions, as was also shown by Arjan Hillebrand (NeuroImage, 2012).

A frequently posed statement at the conference was that the MEG community should now use the advantages of the high temporal resolution of MEG, and no longer should try to replicate fMRI results. Ironically, this statement was most often made by speakers that presented work that did exactly that, namely replicating fMRI findings. Matthew Brookes showed his recently published work of Independent Component Analysis of Hilbert transformed power maps, which show a lot of similarities to known resting-state networks described in fMRI studies (Brookes et al., NeuroImage 2011, PNAS 2012). He and his colleague Mark Woolrich from Oxford suggested that the activity within these resting-state networks should be seen as a transient phenomenon, and that the temporal scale at which the activity is switched on and of is crucial for studies on information processing. Francesco de Pasquale shares this view, and presented a study on fluctuations of non-stationary interactions within the default mode network, and between the default mode network and other resting-state networks (De Pasquale, Neuron 2012).

Another interesting field of studies combines local tissue characteristics, for example local neurotransmitter concentrations or protein expression, with MEG data. Krish Singh was one of the key note speakers who applied these techniques to elucidate the relation between gamma band activity and GABAergic neurotransmission in the visual system. Linda Douw showed in a combined MEG-neuropathology study that protein expression is related to network parameters of the tumor region in glioma patients with epilepsy. Combined MEG and MR Spectroscopy may provide an interesting non-invasive method to study this type of correlations also in brain pathologies. Viktor Jirsa presented a project on behalf of a consortium working on computational modelling studies of brain activity. In the upcoming months, a platform called the virtual brain will be launched that allows for the integration of multiple computational models with available DTI maps. A long-term goal of this project is to allow for modelling of brain activity for individual subjects. In summary, it was a highly inspiring conference and it would be a good reason to visit Halifax, Canada in 2014.

Clinical MEG in patients wearing metal braces

Date: 18-9-2012

Contributed by: Arjan Hillebrand

Patrik Fazio stayed in our department for several months in 2010. His work on the feasibility of clinical MEG in patients wearing dental braces was recently published in Clinical Neurophysiology.

Artefacts caused by the presence of metallic orthodontic material, such as braces, can be of such magnitude that it becomes impossible to record an interpretable MEG. This has been a problem for some patients, usually young people, who relatively often wear dental braces nowadays. Fortunately, recently developed signal processing techniques, namely signal space separation (tSSS) and beamforming, can remove even these large artefacts, as was shown in this study. This means that, although it is still preferable to remove artefact-inducing material whenever possible, a clinical MEG scan is feasible for a much larger clinical population than previously thought possible.

Two other recent papers reached similar conclusions with regards to the clinical applicability of tSSS for movement compensation and removal of artefacts from vagal nerve stimulators

Brain networks involved in auditory hallucinations

Date: 3-7-2012

Contributed by: Remko van Lutterveld

A new MEG research paper has appeared from the collaborator network. The study, in PLoS ONE, describes brain activity related to the presence of auditory verbal hallucinations in schizophrenia. Auditory verbal hallucinations (AVH) are a common symptom of this illness, occurring in approximately 70% of patients. These hallucinations can be highly distressing and often lead to a disrupted social life. It was found that the presence of AVH correlated with a decrease in beta-band power in the left temporal cortex as well as with a decrease in alpha-band power in the right inferior frontal gyrus. Moreover, the onset of hallucinations was related to a decrease in theta-band power in the right hippocampus. These results suggest that auditory verbal hallucinations are triggered by a short aberration in the theta band in a memory-related structure, followed by activity in language areas accompanying the experience of AVH itself. These results provide information about the frequency bands and brain structures related to the experience of hallucinations, and may help improve our models of the pathophysiology of AVH.

Amsterdam Brain Connectivity Conference

Date: 1-5-2012

Contributed by: Arjan Hillebrand

The two-day ABC Conference, held at the UVA, focussed on different approaches to study functional connectivity using neuroimaging techniques such as functional Magnetic Resonance Imaging (fMRI) and Magnetoencephalography (MEG). Morning sessions were devoted to the introduction of general topics, whereas the 3 afternoon talks were given by proponents of a particular analysis approach. Interestingly, these speakers were all given the same dataset to analyse, with the aim to reveal the strengths and weaknesses of each approach.

Day 1 focussed on connectivity as estimated from fMRI data, where most talks described methods to study causality (i.e. directed connectivity; in contrast to the ‘weaker’ undirected functional connectivity, which merely indicates that a statistical relationship exists between signals). Martijn van den Heuvel discussed how estimates of functional connectivity can be used to construct brain-wide networks, revealing that our brains seem to consist a backbone formed by regions that are part of a ‘rich club’ (well-connected regions what preferentially connect with one-another). Alard Roebroeck highlighted several problems (and provided some solutions) with the estimation of directional connectivity from fMRI data. One issue, in particular when using Dynamic Causal Modeling (DCM), is that of missing sources. Lourens Waldorp provided a solution to this problem, using ancestral graphs to find missing sources (which basically tests each connection to find out if such a connection could also be explained by a missing source/connection). The second issue is that fMRI provides an indirect measure of neuronal activity, namely through the hemodynamic response, which is slow (and slowly sampled) and may vary across brain regions. Anil Seth showed that the variability of the hemodynamic response renders interpretation of (Granger) causality meaningless when there is noise in the data and when the hemodynamic response is undersampled. Hence, unless MRI scanners and scan sequences improve to provide data with better signal-to-noise ratios and smaller TRs (repetition times), the results of fMRI Granger causality analysis should be interpreted with caution. The problem of variable hemodynamic responses could be solved by incorporating this variability in the models used by DCM (an overview of this technique was given by Jean Daunizau), although DCM itself relies on having very strong hypotheses about the data (but see here).

Day 2 started with 2 talks describing the use of functional connectivity analysis in memory research, using intracranial EEG, (Nikolai Axmacher) and in vision research, using MEG (Tobias Donner). Germán Gómez-Herrero subsequently described how Granger causality can be determined between the timeseries of independent components, where these components are extracted through a combination of multivariate autoregressive (MVAR) modelling and Independent Component Analysis (ICA). Although it is not clear whether the assumptions underlying MVAR and ICA modelling are always fulfilled, this approach does not rely on an initial reconstruction of neuronal sources, which is exactly the approach proposed by Matias Palva. He discussed how activity can be reconstructed to the cortical surface using a linear inverse estimator, and after grouping activity into cortical patches, the activation profiles for these patches can be used as input for connectivity analysis. This approach is similar to the approach taken in our group, with the differences being that we use beamforming for source reconstruction, parcellate activity to anatomically defined (atlas-based) regions, and that we do not constrain activity to a (inaccurate) cortical surface. In contrast to the approaches above, Mike Cohen restricted his analysis to the sensor level, where he proposed a novel method, based on an analysis of the local covariance structure, to find connectivity on a local scale.

The main aim of this conference was to contrast different approaches through analysis of a common fMRI or MEG dataset. Although serious attempts were made to extract meaningful connectivity patterns from these data, a direct comparison of results was impossible. Reasons for this included the underestimation of required analysis time (seemingly trivial pre-processing steps take more time than expected), complexity of the task (different speakers focussed on different tasks-conditions) and the lack of a good working hypothesis (the context for the analysis and research questions were not clear to all speakers, hampering their interpretation of the results).

Perhaps a similar approach could be taken in a follow-up conference, but using a less complex dataset to contrast methods with --- how about resting-state data…?

We need your brain!

Date: 31-10-2011

Contributed by: Ilse van Straaten

We are constantly trying to improve the research on the topic of functional brain networks. To get new impulses from the field we would like everybody who is interested to be part of our upcoming brainstorm session.

To optimize input during the session we came up with the following format which enables us to identify possible new research questions beforehand and give everybody the opportunity to prepare ideas.

Step 1: Think about a research question regarding brain function What is the most important question that you want to answer with your research? Where do you want your research to lead to? What are some unanswered questions that need to be addressed? Send your question, including its explanation and any comments to : Ilse van Straaten before November 15th 2011. Ilse will collect all input and will then send all questions (anonymized) to everybody.

Step 2: Think of points of discussion When you have received the research questions of the other participants, could you think of (at least one) point of discussion / additional idea / adjustment. Do you think that the answer(s) to the research question really can contribute to the already available knowledge and will it fill a gap that desperately needs filling? If yes: why, what additional actions can improve the results. And if no: why not and what adjustments are needed and possible to give the question further meaning? Send you comments again to Ilse van Straaten.

Step 3: Brainstorm session On Monday 28th of November from 13.00 – 15.00 a meeting will be held in our conference room (-1 Z 135, department of Clinical Neurophysiology, VU University Medical Center) during which we will discuss the future research questions and their comments and hopefully come to new and brilliant ideas on the future of brain network research! Have a look at the Prezi presentation of this BrainStorm.

A followup meeting has now been planned for Monday April 2nd, 14.00-15.00.

Two new MEG research papers

Date: 10-9-2011

Contributed by: Arjan Hillebrand

Two new MEG research papers have appeared from our collaborator network, both addressing basic sensory processing. The first paper, in Cerebral Cortex, describes a novel binaural pitch that is elicited by phase-modulated noise. A binaural pitch is an auditory percept that emerges from combined inputs to the ears but that cannot be heard if the stimulus is presented to either ear alone. In this manuscript we used MEG and psychophysical measurements to characterize this novel pitch, heard when band-limited noise has a rapidly changing interaural phase difference. The MEG data revealed that the pitch was perceptually lateralized (despite the stimulus being binaurally symmetrical), in agreement with the lateralization of the evoked changes in MEG spectral power, and its salience depended on dichotic binaural presentation. The psychophysics data revealed that the binaural pitch depended on the processing of binaural information in a lower spectral sideband of the phase-modulated noise. These results may help to improve our models of binaural processing, i.e. how the brain integrates information from both ears. Such models are key to our understanding of how the typically developed and impaired auditory system extracts information from the auditory world.

The second paper describes a study on retinotopic mapping of the primary visual cortex using MEG. The mapping from stimulation of a particular area of the visual field to primary visual cortex (V1) is known to a reasonable degree of accuracy, allowing us to compare MEG source reconstructions with the true electrical state of the brain. Here, we localized, using a beamforming method, the induced responses in the visual cortex generated by a high contrast, retinotopically varying stimulus. Although well described in primate studies, it has been an open question whether the induced gamma power in humans due to high contrast gratings derives from V1 rather than the prestriate cortex (V2). We showed that the beamformer source estimate in the gamma and theta bands does vary in a manner consistent with the known retinotopy of V1 (see image). However, these peak locations, although retinotopically organized, did not accurately localize to the cortical surface. Importantly, the use of this well-described physiological phenomenon (retinotopy of gamma oscillations) has given us a metric of the distortion (without which we might have been content with our findings). We would suggest that highly convoluted cortical areas such as V1 pose particular challenges for MEG, and so we consider the ability to construct retinotopic maps a useful but challenging benchmark for testing the accuracy of MEG source localization methods.

Brain Connectivity Workshop 2011 Montreal

Date: 4-7-2011

Contributed by: Prejaas Tewarie

The brain connectivity workshop brings together experimental neuroscience, neuroscience methodology and computational neuroscience for a better understanding of the relationship between anatomical connectivity, brain dynamics and cognitive function. This year’s theme was the developing brain.

The first day of the program was an educational session. The most intriguing session was given by Viktor Jirsa who gave an introduction to dynamical systems. A dynamic system is merely a state or a superposition of states that changes in time. This can be described by ordinary differential equations, discrete maps, integral equations or integro-differential equations. Dynamics occur in almost every branch of nature, also in the brain. Dynamics can be visualized in the form of time series, phase flow and the potential. The phase flow is the trajectory of a system in phase space. If we disturb a dynamic system, in particular the dynamics of the brain, under certain circumstances the perturbation decays in time. This only occurs in non-linear dynamics which is a subject far more complicated than linear systems. The most important contributions from this field are different metrics of synchronization. Unfortunately not much emphasis was laid on different metrics of synchronization which I think was a missed opportunity. On the other hand Viktor Jirsa and Christophe Bernard did present an interesting model of the underlying brain dynamics in seizures in children. In their model the underlying mechanism of seizures was simply just a noise driven emergent property of immature neuronal networks.

The second day Allard Roebroeck and several others gave lectures on structural connectivity and the accompanying difficulties of measuring this. Both probabilistic and deterministic tractography have their own biases. Some algorithms are biased towards short distance connections whereas others are more biased towards longer distance connections. Therefore the choice of your tractography algorithm influences small worldness. A new algorithm was therefore presented: graph tractography. This should not be confused with graph theoretical analysis of brain networks after an adjacency matrix is obtained. The tractography algorithm itself uses graph theory (Iturria-Medina et al.Neuroimage 2008).

The last day several other interesting topics were put forward. Alan Evans for instance suggested that cortical thickness correlations could resemble cortical connections since the corresponding adjacency matrix showed a small world network (Evans et al. Cerebral cortex 2007). Kaiser raised some fundamental questions about brain networks and criticized graph theoretical studies on the brain. To start with the criticism he argued that too less links are made between the abstract findings of graph theoretical measures and the underlying pathology. He admitted that this is not a trivial task. However this should be engaged by linking processes on micro-, meso- and macro level to get a better comprehensive understanding of brain function/structure and brain pathology. The fundamental and interesting questions raised by him were: How does one node know that another node is highly connected? Where do hubs come from? Where do modules come from? What are the evolutionary constraints of brain organization? How and why are hierarchical networks formed?

Other topics in the workshop were optogenetic fMRI, Genetic influences in maturation of human functional networks, near infrared spectroscopy imaging of connectivity in children, mean field models and dynamic causal modeling. The session was closed by Michael Breakspear who gave a lecture about phase transitions in neonate cortical networks during recovery from hypoxia. He didn’t present a biophysical model, but a phenomenological model borrowed from physics.

Functional connectivity as outcome measure in a medical trial: Souvenir II

Date: 30-5-2011

Contributed by: Hanneke de Waal

Many clinical trials on Alzheimer’s disease have neuropsychological outcome measures. Of course, for the patient it is most important how his or her functioning in daily life is affected by an intervention. But from a disease perspective it is very interesting to know if the intervention has a direct influence on the brain. In animal studies it is possible to measure for example synaptic density directly, but in humans this is not possible. However, we can measure the electrical activity generated by neuronal communication at the synapse in humans indirectly with functional connectivity studies using EEG and MEG.

The Souvenir II trial is a randomised controlled trial designed to assess the efficacy of a multinutrient (Souvenaid) in patients with mild Alzheimer’s disease. Souvenaid is a multinutrient drink which contains specific nutrients, like omega-3 fatty acids, choline and certain vitamins, that act as precursors in the synthesis pathway of phosphatides, the building blocks of synaptic membranes. In animal studies these multinutrients have shown to increase neuritic outgrowth and dendritic spine density. In the Souvenir II study, functional connectivity, measured by EEG and MEG, is used as secondary outcome measure. In all patients of participating centres across Europe EEG is measured at baseline, half way and at the end of the 24-week intervention. All EEG’s are collected and analysed at the VUMC. In a small substudy in 40 patients MEG is measured as well. Outcome measures will be, besides spectral analysis, graph theoretical network analysis. The results of functional connectivity analysis are expected in 2012-2013.

What lies beyond the epileptic EEG?

Date: 27-5-2011

Contributed by: Cornelis Jan Stam

On Thursday 20 May 2011 Maeike Zijlmans of the department of neurology of the Utrecht University Medical Center, successfully defended her PhD thesis "New Presurgical techniques to characterize the focus of epilepsy". The thesis and defense were given a well-deserved "cum laude" judicium. On the occasion of this PhD thesis a full day symposium entitled "What lies beyond epileptic EEG?" was organized in Utrecht on 25 May 2011. During this well-attended symposium with a mixture of speakers from Canada, France and The Netherlands many new and exciting topics in epilepsy research were addressed. In addition to presentations that challenged the conventional ideas about interictal spikes and source localization, many contributions dealt with EEG / fMRI as well as high frequency oscillations. It was clear that interictal and even ictal EEG / fMRI recordings are increasingly becoming important tools for the evaluation of epilepsy, in particular in relation to the work-up for epilepsy surgery. With the advent of 7 Tesla and even 11 Tesla scanners this approach may become even more important. High frequency oscillations (HFOs) were discussed in several talks. These high frequency / low amplitude phenomena, that can be categorized as "ripples" (80-250 Hz) and "fast ripples" (250-500 Hz) are very likely related to epileptogenic brains areas. Their relation to epileptic spikes is not yet clear however. Surprisingly, it was argued that such HFOs may even be visible in scalp EEG during seizures. This would require a much higher sample frequency and completely different filter settings. These novel concepts are very challenging, and may require an adjustment of EEG practice in relation to epilepsy. The meeting was closed by Jean Gotman from Montreal who presented his view on the future of the field. He showed that, with properly modeled hemodynamic response functions, fMRI analysis of epileptic activity may actually be possible without recording the EEG. In his final slide Gotman predicted three facts about the future: (i) it will be exciting; (ii) it will belong to women; (iii) it will belong to the Montreal / Utrecht cooperation. Concerning the first two points one couldn't agree more. Keep in mind however that it is difficult to make predictions, especially of the future.

"Stochastic activity month" (SAM) at Eurandom

Date: 18-5-2011

Contributed by: Ilse van Straaten

During April and the beginning of May 2011 it was ‘stochastic activity month’ (SAM) at Eurandom, the workshop centre based at the Stochastics Section of the Department of Mathematics and Computer Science of the Technical University Eindhoven. The goal of SAM is to bring together a sizeable group of national and international researchers on the topic and this month’s theme was ‘Stochastic Modelling and Analysis of Networks’. Within this framework, the workshop ‘Random Graphs and the Brain’ was held on May 11 and 12 2011. The specific aims of the meeting was to combine researchers from random graphs and complex networks, and from neuroscience to obtain a better understanding of how random graphs, or probability theory in general, can play a role in neuroscience. On behalf of prof Stam and the other members of the research group of the department of clinical neurophysiology at VU University Medical Center interested in clinical application of graph theory, Ilse van Straaten was attending part of this workshop.

Some presentations focused on the explanation why neurons behave the way they do with respect to their firing activity and connection with their neighbouring neurons. Others focused on calculations on models from statistical mechanics. And there was one talk with an overview of top-down research on MEG/EEG/fMRI in healthy subjects and subjects with neurological disease using modern network theory by the only clinician in the room.

The group of participants was pleasantly small, the atmosphere was inspiring and there was a great deal of enthousiasm, allowing for much interaction and discussion. Some of the discussion was on the growing and developing network and how to model the complex brain network. Some parallels were found with other developing networks, such as the internet, in which the hubs turn out to be not the most important nodes to begin with, but the oldest. For the last issue, scale-free percolation might be a promising next step, adding to the models that have been used so far.

Wiring the brain, making connections

Date: 21-4-2011

Contributed by: Maria Boersma

Last week, the international conference titled "Wiring The Brain: Making Connections" took place in Powerscourt, County Wicklow, Ireland. A few young investigators of the VUmc (W. de Haan, M. Schoonheim and M.Boersma) interested in brain networks from a graph theoretical perspective, attended this conference.

This biannual conference aims to bring together scientists from separate disciplines, including developmental neurobiology, psychiatric and neurological genetics, molecular, cellular and systems neuroscience and cognitive science. A series of articles about wiring the brain is being published by BioMed Central. The main theme of the current meeting was to explore how brain connectivity is established, what happens to circuit and network function when the underlying processes go wrong and how this can lead to psychiatric and neurological disease.

This conference gave a nice update of the latest results in a broader scope of neuroscience. The lectures showed topics from very diverse specializations of neuroscience. Strikingly, only a few of the speakers took time to properly introduce their field of science and give an overview of recent studies. The integration of information from several disciplines during this conference was mostly achieved during the interactive poster sessions.

From a graph theoretical network perspective, topics became more interesting towards the end of the conference. During the first three 'bottom-up sessions -“making connections”, “circuit dynamics”, “from genotype to phenotype”- the lectures were mostly aimed on geneticists and cell-biologists in the audience. The speakers focused on how gene products regulate (or dysregulate) intra- and interneuron-signaling, neuron migration, axon guidance and synapse formation in animal models. György Buzsáki shortly introduced graph theory and small-world organization in large assemblies of interacting neurons. The last three top-down sessions included neuroimaging studies describing brain connectivity in several psychiatric disorders and development both at whole brain level and at neuronal level. In these sessions only two lectures touched upon graph theoretical applications. Bradley Schlaggar summarized findings from resting-state fMRI studies and network analyses on the developing brain. Rosa Cossart showed, using several cell imaging techniques, that early born interneurons are most likely to become the hubs in mature networks. Since graph theoretical analysis is applicable at every scale in the brain and is relatively easy to understand, it can/might/should become the designated tool to link together findings from different disciplines.

Storm van Leeuwen Magnus award of the Dutch society for Clinical Neurophysiology for research on complex brain networks in Alzheimer's disease

Date:29-3-2011

The Dutch society for Clinical Neurophysiology awarded Kees Stam, professor of Clinical Neurophysiology, the 2011 Storm van Leeuwen Magnus prize on 29 March during its annual meeting. He obtained this award for his paper "Graph theoretical analysis of magneto-encephalographic functional connectivity in Alzheimer's disease" that was published in the journal Brain in 2009. This research was done in collaboration with, among others, Willem de Haan and Philip Scheltens of the Alzheimer Center of the VU University Medical Center. Criteria for the award are that the paper should have been published in a peer-reviewed scientific journal with a high ranking and that the topic is innovative and clinically relevant. For the 2011 award there were 11 submissions, including four papers published in Brain. The prize is awarded once every two years and consists of an object of art and cheque. [adapted from: Tracer 21 april 2011, nummer 8, page 3]

Linda Douw gets NWO Rubicon grant for working on complex brain networks in Boston

Date:28-3-2011

Dr. Linda Douw of the VU University Medical Center in Amsterdam (Netherlands) has procured a Rubicon-grant from the Netherlands Organisation for Scientific Research (NWO), to spend one year in Boston (USA), in the lab of Dr. Steven Stufflebeam (Harvard University, Massachusetts General Hospital). During this year, she will investigate the relation between cognition and localized network topology, in both healthy participants and patients with lesional epilepsy.

Epilepsy is the most frequently occurring neurological disorder, affecting 1% of the world population. Epilepsy is most often caused by a lesion in the brain. Only approximately 70% of patients are seizure-free using anti-epileptic drugs (AEDs). In the remaining patients, surgical removal of the (area surrounding the) lesion may be effective. The link between epileptic seizures and whole-brain network topology is well-studied in silico, in vitro, and in humans. However, patients’ problems are not exclusively limited to the seizures. Lesional epilepsy patients often experience cognitive deficits, which they themselves rank as more burdensome than the actual seizures on their list of complaints. These cognitive deficits are still relatively poorly understood, but the lesion and AED use may contribute to the problem. Most importantly, the neural correlates of cognitive deficits remain to be elucidated. Patients with epilepsy due to a lesion (‘lesional epilepsy’) have cognitive deficits, which are related to global brain networks. Interestingly, zooming in on the roles certain areas have in the network has never been combined with cognitive investigations. Brain hubs (important areas influencing the entire (sub-)network, see figure) may play a key role in the cognitive brain network.

The association between brain hubs and cognition will be investigated in lesional epilepsy patients. On a neuroscientific level, possible associations between brain hubs and cognition in both healthy participants and lesional epilepsy patients will increase our basic understanding of human cognition. Ultimately, this research also has a clinical goal, since 25% of lesional epilepsy patients undergoing neurosurgery to halt seizures develop postoperative cognitive problems, despite presurgical (deterministic) mapping of functions. Brain hub analysis will be used as a new paradigm for presurgical mapping in lesional epilepsy patients to ameliorate postoperative cognitive outcome.

For comments and suggestions: send an e-mail

copyright C.J.Stam
Contact information: Department of Clinical Neurophysiology VU University Medical Center
Postal address: De Boelelaan 1118 Postal code: 1081 HV Amsterdam The Netherlands
P.O Box: 7057 Postal code: 1007 MB Amsterdam The Netherlands
Phone: 020 4440727 Fax: 020 4444816