 Thank you very much for the invitation and your kind of introduction. So today I'm going to be focusing on some of the principles that we've kind of gathered from our work over the last several years, dealing with atypical and atypical development of brain networks and how they're formed. I just want to acknowledge my group, considering of many different levels of expertise in different domains. These individuals in particular for the work that I'm presenting here today. So we want to understand how the brain functions. It's important to kind of understand how it comes to be through development and the learning process. How the initial structure, the core structure of the brain is then used to build and scaffold networks for learning and formation of complex cognitive functions. And the general idea is that there are a lot of things in terms of both learning as well as things that go awry in various psychopathologies including autism. They are now considered to be aberrations in the formation of particular circuits that form the core functions that go awry in many of these disorders. So we've known for a number of years now this kind of pattern of gray matter maturation with increasing levels of cortical pruning leading to thinning which is kind of assessed in this gray matter thickness measures. And the idea is that the nodes of these systems are continuously undergoing change and that should have some implication for how the brain systems are functionally organized. And so the general ideas on which some of this work is based is actually quite simple. That is that if you look at the changing pattern of contributions of a given cortical or sub-cortical node to function there are two general underlying principles that govern how they come to represent information and how they come to process information the way they do. And this has to do with two essential notions of the structural integrity and intrinsic connections of a node and a fingerprint of connections of these areas. And one could argue that most of the burden really is on the connectional fingerprint because there is a general canonical cortical representation of these nodes in terms of excitation and inhibition. So what really should drive a lot of the changes in development are really changes in connections. And now of course one can study this in the general context of connectomics and graph theoretic metrics but they leave unclear the links with learning and disorders and cognitive function. So the challenge is to relate changes in the connectome with changes in cognitive function. And now this of course has import for the study of psychopathology, a number of developmental psychopathology including autism where there's evidence that there are cell migration deficits or imbalance in excitation inhibition, particular nodes and they happen to be in core hubs of the brain like the posterior cingulate cortex. They can drive a lot of cognitive and affective dysfunction across a wide range of target connected regions. So some of the things we've learned from kind of doing individual node based analysis as well as whole brain analysis of these circuits that have evolved are something I've actually summarized here in this TICS article. And so in general the structural backbone of this system is actually quite well developed by age two and the functional connectivity patterns of some of these core systems like the default mode network or the fronto parietal system they exist by age two. What changes of course is the interconnections between different systems and this kind of functional circuit maturation goes on well into adulthood and one of the key features that characterizes the development of the system is the formation of segregated and specialized functional units as well as their dedicated circuits. And I'll show you some examples of these and this is led by fine tuning of local circuits because as you might surmise from the changes in grey matter maturation and pruning some of these circuits get structurally fine tuned and underlying that of course is changes in excitation and emission balance which gives rise to more fine green structural nodes which then has implications for how proximal areas stuck to the rest of the brain. I'll show you an example of that from our work on the proximal amygdala nuclei BMA and LMA. So then the other kind of notion that's evolved from this work is there's a systems level pruning of connections brain-wide and one of the dominant things that drives this is differences between subcortical and subcortical connections with an overgrowth of subcortical to cortical connections in children which then gets pruned over development to cortical-cortical connections. And then kind of more recent work showing that functional networks which are intrinsically dynamic they become more dynamic and flexible with development. And I'll show you an example of that. So this is just kind of very briefly taking it through some of the results which have led to initiating these general principles. So this work from Gau et al showing that by the age of two the global efficiency starts to look like that of adults and modularity and a few other network measures. If you look at global small world kind of architecture by the age of seven to nine that's really indistinguishable from adults. But there are features and of course a lot of things are changing despite these gross brain-wide metric similarity across between children and adults and some of the measures of hierarchy of networks as well as changes in local connectivity versus long-range connectivity when they go change with the short-range connections being stronger and children and the longer-range connections being stronger in adults. So there's a complex profile of changes that occur that build on a core backbone that's already well formed by the age of two. So this is kind of showing you an example of the differentiation and functional circuits that occurs. It's just an example to show you here with the BLA and the CMA nuclear of the Migdala. In early childhood there's a strong cross-talk between these nodes which gives rise to overlapping connections as you look brain-wide through multiple functional systems and it gets increasingly segregated. These are the patterns associated with the two nuclei and the connections strengthen overall and they get more differentiated and segregated. And this is an example showing the complex pattern that works and we've replicated this in two samples. And what really drives that is this general notion I mentioned of stronger local connections. So the BMA and the CMA are tightly linked in early childhood which gives rise to undifferentiated circuits and that's the pattern that I just showed you. With development what happens is that these nodes get locally uncoupled which then shows up as very distinct connectivity patterns and we think that's driven by changes in GABA-ergic signaling which gives rise to more fine-tuned nodes which has a major influence on brain-wide connectivity. So this is just an illustration of some of those principles and I think they can be easily put into models and tested out in a very computationally rigorous manner, some of which we've started to do. So this is kind of illustrating the point that I mentioned about the changing landscape of sub-cortical connections. There are over-represented in children. The degree is higher, the path length is shorter in children compared to adults and this is really very different from a lot of other systems. If you look at the mean connectivity differences, you see this kind of dissociation with the sub-cortical links being stronger in children and the cortical-cortical links being stronger in adults. And you can use these patterns actually to differentiate between children and adults and the subsystems that show up the biggest difference are the ones that link sub-cortical to cortical structures which has a lot of implications for how these areas process information related to habit formation and reward processing and so on. Some of whose effects we see much later on during adolescence in terms of proclivity for reward processing versus relative weakness in development of cognitive control systems. So these are factors, developmental factors, in terms of whole-brain organization that we think contributes to that. So the other kind of final point that I wanted to make was this kind of the other approach to looking at whole-brain organization in the context of isolating individual components and we know that there can be, you know, the brain-wide, for example, task, arresting state activity can be segregated into these functional systems and there's a lot of model building and theory building that can go around it and one of the systems that we've really... the three systems that we've really focused on is a particular process that we think involves the core components of the salient's network, the anterior insular in particular, which serves as a gating mechanism for salient sensory and limbic stimuli which then down-regulate the default mode network and up-regulate the frontal parietal working memory system in order to engage in appropriate context-dependent information processing and then the default mode network can put a memory or timestamp on it as it absorbs that information and link it to the hippocampal memory system. So this is a general formulation of an intentional system that engages and provides access to resources for salient events in one's environment and so there's a whole model and theory we've built around this and these systems pop up over and over again and virtually every psychiatric and many neurological disorders, particularly the frontal temporal dementias. So we've used this as a model for trying to understand how these cross-network interactions develop in children and we focused on these three core networks that I mentioned to you earlier with the visual system serving as a control region and here's an assessment of developmental differences both kind of within network nodes like this one here and cross-network nodes and we see that both within network and cross-network interactions are strengthened in adulthood so these networks are relatively more segregated in children and there's a more dynamic crosstalk as the brain matures and this is the other kind of principle that I wanted to illustrate to you and underlying that is changes in white matter pathways that link these areas and we can tie the degree of change and the structural links to the functional so there's a structural underpinning to this and this has import for some of the things I showed you are with so far of being trying to understand the principles based on organizational of the intrinsic connectivity patterns of course this has import in terms of information processing as well so in this study we had adults and kids do arithmetic problem solving tasks and you see that the causal interactions this is assessed using two different approaches multi-variate dynamical systems approach as well as a multivariate range of causal approach you can see that the cross-network links in terms of relating and providing access to the parietal working memory system are actually strengthened in adults compared to children so some of the features we see intrinsically always have to be thought about in terms of how they facilitate aspects of information processing and so I think back and forth between trying to understand the intrinsic architecture and how it contributes to more efficient condition and this is again showing this is related to the model of the front of insulin cortex as a node that picks up salient stimulus in the environment and engages other systems as just showing that the same pattern is observed in both children and adults the strength of that link actually is predictive of performance with a number of links contributing somewhat more weekly in children to accuracy and reaction time than much more limited set of links and stronger predictions on the adult side so the way these networks get configured and the strength of their causal influences during task processing comparing on how proficiency in task performance increases and so the idea then is that these networks provide constraints on how we can model and think about information processing now the other thing that we've kind of done more recently is to use we know that even these so-called statically identified networks they have constant interaction as I showed you with the changing patterns of intrinsic connectivity with age as well as the causal influences during information processing and we know that these systems do interact and we've used in Markov models now to try and understand the nature of these states and to model them to identify dynamic functional circuits and how sticky and how flexible they are over development and this slide just briefly summarizes the difference between adults and children you see very, so these are individuals over time you can see the states in different colors they change much more rapidly and children you see these sticky states that are relatively inflexible that last for much longer times than what you can see here in adults and you can actually quantify this with mean lifetime of states and you see that once a child is in a particular state it tends to stick in that state and so the mean lifetimes are much larger and the transitions between states is actually weaker in children and so we want to apply this to our cognitive tasks as well but the general idea it illustrates is that children tend not to switch between states and this may lead to behavioral inflexibility that we've now tried to understand particularly in the context of autism research so this is just basically a summary of the findings related to the principles I've tried to illustrate so just very briefly before I go into the next part of the talk which is more focused on the atypical development focusing on children with autism is that the functional backbone is established by age two so some of these core systems are in place strongly driven by the early developmental processes but on top of that you get a lot of complex features of pruning and segregation and formation of individuated circuits which have a strong bearing on how cognitive and affective information is processed because now you have segregated channels which can interact only at a few places as opposed to more globally because the nodes that process different types of information are segregated and then you have this notion that the cross network links also become much more dynamic and strengthen with development and this kind of finding related to our model of the salience network is something that switches across states that is also borne out in terms of its being weak and therefore it has consequences in terms of how different systems are engaged during active information processing and so that's kind of the general set of ideas related to the principles that I told you about earlier so now I want to kind of move to tell you a little bit about the work on autism that we've been conducting and we've been primarily focused on trying to understand not just the global architecture but how do we go back and link these notions that we are both in terms of principles of adult brain organization but the developmental trajectory as well to understand the core triad of deficits and the actual language communication restricted repetitive behaviors and now it's increasingly clear through a large body of work that autism is a neurodevelopmental disorder of brain circuits the question is when in time it occurs and we have access to information from seven-year-olds and I think the challenge is going to be to go younger to figure out how they came to be and so the general questions here are child with autism differ? How do we relate to clinical symptoms? How does this change with age? Can we identify biomarkers? Can we model the brain? Is it some of these principles that I initiated earlier to see how they are or not are not born out in terms of actual human neuroimaging data? I'll just illustrate that the notion of autism and brain connectivity has now is so tightly linked that it's hard to conceive of models that don't involve connectivity in some sort or the other and that makes sense because you have miswiring of local circuits resulting from your imbalance which as I showed you an example of even in normal development you can get cross-nodal connections and you can as well imagine with dysfunctional nodes and the kind of the general idea there's a high level of comorbidity in terms of epilepsy and autism all of that is kind of given rise to the thought that there are EI imbalances that drive dysfunctional connectivity and so the question is what are the characteristics of that and how do we think about the relationship to clinical symptoms in some principle functionally meaningful manner? So one notion is that you have core nodes or even rich clubs such as those that link these various modules and if you have dysfunction in one of these you could have wide-ranging set of cognitive and affective deficits as opposed to a node in some peripheral a node that might just influence or affect only a module of constrain set of nodes and so the kind of just going back to this general idea of the cortical fingerprint if you have a dysfunctional node perhaps one that's sitting in a core hub of the brain it's going to have a major influence on cognitive, social and affective function whereas a node that's sitting in the periphery may not have such a large impact So I think you're familiar with some of these systems particularly the default mode network which is anchored in the posterior medial cortex which is going to summarize this slide actually summarizes a wide body of our work that we've been conducting and publishing relating these systems so we've related the social abilities deficits, disabilities to the default mode network language communication deficits to the extended voice selective system anchored in the posterior temporal sulcus and the attention and restricted interest clinical feature to the salience network anchored in the insular So just kind of to take you so the next set of slides basically takes you through our publications that show you how those were bone out providing some evidence but that essentially was the gist of our findings and so if you do a standard analysis of gray matter volume in autism there's all kinds of confounds of age as factors and the results over the years has been highly variable but one of the things we then did was to say well maybe it's not a volumetric change at least in a wide age range rather it's the structural organization of gray and the adjoining white matter and what we find is a strong evidence for posterior cingulate cortex dysfunction and there are actually histological studies showing cell migration deficits in the posterior cingulate cortex rather than even the fusiform garrison area that's implicated in face processing that people initially thought was dysfunctional in autism and so the question then is there is a node going back to this node and it's fingerprint there is a dysfunctional node here can we tie that into deficits in autism and there is a large body of work which I don't have the time to get into which involves a range of social cognition tasks initially performed all in adults with theory of mind self in the context of the other and a whole range of them pointing to involvement of the default mode network in particular when you're making judgments about yourself in the posterior cingulate cortex as well as the ventromedial prefrontal cortex the functional roles are somewhat dissociated but this is the node that has been identified in structural studies now and in particular the strongest evidence that I have seen for histological cell migration type of deficits is really the posterior cingulate cortex so we kind of then the other piece of the evidence of course here is that you have the work from Harman and many others now with the structural connectome showing that the posterior cingulate cortex and the posterior medial cortex in general are like a structural core in terms of they have the shortest path length to the rest of the brain so it's a good candidate node to look at in terms of all of these pieces of evidence that have accumulated and so the other thing to kind of keep in mind is that when we talk about regions like the posterior medial cortex and this is an issue with the ongoing connectomics work is the lack of focus on the anatomy and this just result just shows you that the cortical connection fingerprints are really very different in these adjoining areas and that's kind of mapped out here with the posterior cingulate being very strongly linked to the ventromedial prefrontal cortex and precuneus to kind of the motor system and the parietal SPL visual motor regions and so you really and then the retrosperinal cortex to the medial temporal lobe and this finding in children actually roughly replicates what the dissociations that have been seen in adults so even by the age of 8 to 10 these kinds of segregations already occurred although we haven't really directly compared the children and adults here in this study but these associations have to be paid attention to because the functions of these regions are very very different and so I'm arguing against a pure putting all these nodes into a connectome saying you need to focus on the functional anatomy of these regions to understand this both function and this function and here's just the data showing the again another aspect of heterogeneity and dissociation in the sense that the pattern of differences that you see are actually very different not just qualitatively but qualitatively but qualitatively because you see a pattern of hyperconnectivity in children with autism in the in the posterior cingulate cortex and the retrosperinum cortex where it's completely flipped over in the in the precilious so the functional anatomy matters and the functions do matter as we think about this and it turns out that it's the connections of the posterior cingulate cortex that's actually predictive of the social sub-scale measure of social abilities in children with autism and this again it's hints at a out-of-network cross wiring as a metric related to dysfunction so again some set of links have gone awry and these are good candidates at this point and we're trying to understand them with functional imaging tasks what exactly is going on there so the second piece so the second piece is really related to the other core symptom language impairments and they present and they're not in the nature of individual word reading and so on these children tend to be at least at the high functioning and hyperlexic it's really the level of expression and communication using language and and even paying attention to salient sounds in the environment and so while they can hear normally the audition is fine the way they attend to stimuli and this is the canonical example of this where the child with autism does not respond to his mother's voice much to Shagrin and so this has been kind of well kind of documented anecdotally and so the social queue how is it processed and how is it related to either normal or abnormal circuits related to the postures temporal surface which is the voice selective cortex differentially responsive to the human voice which is matched acoustically matched environmental sounds for example and that's been a body of work for the last 10 years and we're still trying to piece the circuit part of this now and trying to develop our analysis framework based on this so what really struck us here in this study was that the posterior the posterior temporal sulcus node shown here actually is much weakly connected to not the language areas but the reward processing system shown here with responses weaker responses in the regions the anterior insular the orbital frontal cortex and and this is just kind of a more extended version of that in fact you can actually trace this down to the ventral tegmental area so we think there is a miswiring of the voice circuit with the reward pathways and this may underline a lack of response to saliency in a stimulus and lack of attention to them because it's not something that the child feels is sufficiently rewarding to orient to compared to something else in the environment like a vacuum cleaner or something like that okay so then we can actually ask is the connectional fingerprint related to social language communication deficits and it turns out it's specifically related to communication scores in these individuals and the pattern you see really are some of these areas involved in reward and affect so it kind of expands the scope of some of the ways we think about language communication not in terms of language, core language circuits but really in terms of whether the child actually attends to the core vocal features that most of us actually find engaging and rewarding to the affective part and then the intrinsically rewarding part is the novel aspect we've shown here and how do we build a model how do we take this set of findings to create a model and then design a proper task to look at these systems and the general model that we've now come up with is you just need to focus not only on the voice selective system but on this extended system which involves reward related processing and so we've designed an experiment where we asked the child to listen to the mother's voice and control voices and environmental sounds and it's just preliminary data that we published earlier this year just looking at controls while our final sample of children with autism is being acquired and we see that the response to the mother's voice is actually exaggerated not just in the voice selective cortex but also in these same target areas that I showed you earlier with all your frontal cortex your insular and your nucleus encumbrance so we've taken this idea of a dysfunctional circuit and tried to map that into a process that we think is dysfunctional in children with autism and this provides a new framework for thinking about deficits and communication cues and this is just showing that even in children with without autism typically developing we can actually take measures from this system the connexional fingerprint you can go back to the same areas that I identified in the intrinsic system and you can actually predict individual differences in social communication abilities and so now we're trying to extend this now with the autism sample to kind of see how well this model plays out in terms of actual processing of these types of stimuli and it provides lead in to some aspects of therapy as well which revolve around pivoting a child towards a engaging stimulus by rewarding that response and so we're kind of back here to this kind of system the network model that I showed you and I want to end with an example that relates to the third core symptom of autism which is the restricted and repetitive behaviors particularly talking about the motoric behaviors as much as the restricted interest the narrow focus on one topic and the idea that that evolves from this work is that again if you look at across all the brain networks the ones which are most dysfunctional it turns out to be the salience network that you can use to classify children with autism from typically developing and it's a hyper connected system and so the general idea is that just showing the classification rates showing the salience network is highest and we can actually take measures from this and show that you can predict depending on these measures of hyper connectivity of the system that it's related to these restricted behaviors and which is core and the general theoretical idea is that this is a system that cannot disengage itself from what it's doing and so there are large classes of stimuli in the environment to which the child does not then orient attention to and this is a model for thinking about restricted interest and attention to it that's very narrowly focused and this just shows that we've now extended this with again cognitive tasks involving a p300 type oddball paradigm to looking at socially relevant faces with scenes and here again if you look at the difference between how much can a child modulate the response between the intrinsic state and the task state that actually predicts restricted behaviors that the inability to modulate intrinsic systems is actually predictive of the restricted interest so again kind of a model here that we've been working with that there's a weak mapping of the external world and saliency that prevents engagement of these other systems and disengagement of the default mode network in particular that allows the child to focus on those narrow interests as opposed to attending to the environment in a dynamic way so just the last slide here I want to conclude with some of these things I've said so the first part dealt with some of the principles that we've gathered based on individual studies of formation of segregated circuits pruning of global connectivity strengthening of cortical-cortical weakening of sub-cortical connections and then tie that to individual tasks and then to individual features of atypical development and these circuit models provide a way to think about these things in a principal way but I think the intrinsic connectome is just the starting point you need to probe it with a proper stimuli to get a sense for how manipulable those circuits are in relation to either hyper or hypo-connectivity that you see in atypical populations such as autism.