Yesterday’s post discussing the recent paper “The Role of Hub Neurons in Modulating Cortical Dynamics“ (2021) ended with a number of questions. In particular, questions about the cortical microcircuit’s activity states. Spontaneously synchronous bursts were observed when the extracellular calcium concentration was set to 1.4 mM. Spontaneous asynchronous activity was observed when the calcium concentration was 1.25 mM (this state wasn’t mentioned in yesterday’s post). A major question we were left with was, what are the equivalent behavioral states associated with these brain states?
This question was addressed in detail by the same research team in their 2015 paper “Reconstruction and Simulation of Neocortical Microcircuitry“. The researchers incrementally varied extracellular calcium concentration in their simulated cortical microcircuit within a range based on data from intact organisms. At the lower concentration of 1.25 mM they saw spontaneous asynchronous activity, which is typically associated with wakefulness (Figure 1B). As they increased the extracellular calcium concentration, neuronal activity became more synchronous until they saw spontaneous synchronized bursts about once every second, which is typically associated with deep sleep (Figure 1A). Notice the steep transition between states in Figure 1C (spontaneous synchronized bursts has the higher cross-correlation value).
In conclusion, there are solid data to associate the two cortical circuit states spontaneous synchronized bursts and spontaneous asynchronous activity to deep sleep and wakefulness respectively. The question of the existence of spectral analyses on these data remains. Please tweet me if you find any!
Brain cells with more than the average number of connections with other neurons are known as hub neurons. Hub neurons have been shown to significantly decrease the average path length of communication from any one neuron to another, which results in what is called a small world network. Recent research investigated hub neurons and their contributions to simulated cortical circuit activity (“The Role of Hub Neurons in Modulating Cortical Dynamics“, 2021).
Synchronous bursts of activity at about 1 Hz (Figure 1A above) was spontaneously generated by a data-driven simulation of a 0.3 cubic millimeters piece of cortex. The cortical microcircuit contained 31,000 neurons, about 37 million synapses, and 55 morphological cell types. Interestingly, 1 Hz is the delta oscillation frequency, which is most prominent in cortex during deep sleep. Unfortunately, spectral analysis was not addressed in this paper.
The number of network bursts was precipitously reduced and periodicity was far less sharply defined when the investigators turned off 2,977 hub neurons selected randomly among the hub neuron sub-population (Figure 1B above). In contrast, network bursting remained unaltered when they turned off 2,977 randomly selected neurons (Figure 1D above). Hub neurons clearly contributed to the 1 Hz spontaneously synchronous bursts of activity.
Next the research team turned off hub neurons one cortical layer at a time (except for layer 1) and then observed network activity. Turning off random hub neurons across all layers had the most robust affect on all network activity measures: reduced bursts, reduced firing rates, reduced coefficient of variation, and decreased correlation. Removing layer 5 hub neurons had the largest effect of the layer specific manipulations. Interestingly, removing a relatively small number of layer 4 hub neurons reduced spontaneously synchronized burst activity well before effects were seen from hub neuron removal from any other layer or from across all layers.
In summary, hub neurons in general and layer 5 hub neurons in particular contributed to the spontaneously synchronized bursts of activity in simulated cortical microcircuit experiments (“The Role of Hub Neurons in Modulating Cortical Dynamics“, 2021). These intriguing results suggest taking closer looks at the effects of specific layer 5 neuron types that have hub connectivity. It would also be informative to see a more detailed look at the cortical circuit’s neural dynamics. In particular, is the pre-manipulated simulated cortical circuit activity equivalent to awake activity in real brains? Deep sleep activity in real brains? Cortical slice activity? And what do the spectral analyses show?
It’s been three years since I last wrote to you here. I’ve been busy in theoretical and computational neuroscience after several years focused on the semantic web and artificial intelligence technologies company I founded in 2001. My big surprise is that neuroscientists continue to painstakingly search for and extract data from research papers by hand!
Of course there will always be a need to read and pull data and ideas straight from research papers but I thought that by now, 20 years later, easy-to-use web pages would provide globally aggregated brain research data. Not yet. There is exciting linked-data based work going on and a wide range of useful ontologies used by brain researchers with sophisticated working understandings of semantic web technologies but there appears to be a large gap in tools for those not steeped in this knowledge. Neuroscientists don’t have easy rapid access to data answering, for example, the question, “what is the average spike rate of pyramidal neurons in layer 5B primary motor cortex?”
The infrastructure that enables expert searches through piles of data is everywhere. Google, Apple, and the Allen Institute for Brain Science are just a few of the institutions that use semantic web technologies (ontologies, reasoners, linked-data) but these technologies remain mostly hidden and unrecognized by the public. Indeed, most programmers are at best only vaguely aware of them.
Speed Date with Semantic Web Technologies
Let’s take a look at semantic web technologies in the brain sciences today by diving into the important new Neuron Phenotype Ontology. The Neuron Phenotype Ontology was created to help find and categorize neurons with particular sets of traits (phenotypes).
1. To download the ELK Reasoner, select Check for plugins … from the Protégé File menu and the Automatic Update window will pop up.
2. Check the Install box next to ELK Reasoner and click on the window’s Install button.
3. Select the ELK Reasoner by opening the Protégé Reasoner menu and selecting ELK Reasoner.
4. Run the reasoner by opening the Protégé Reasoner menu and selecting Start Reasoner.
Ideally, an otology has access to a world of linked-data through the Internet. The Neuron Phenotype Ontology is in its infancy and so it’s available as an evaluation version with a relatively small set of data from three sources (Henry Markram’s Blue Brain Project, the Allen Institute for Brain Science, and Josh Huang’s laboratory). With limited links and data, just any search won’t do. For example, most of the linked-data are from visual and somatosensory cortices, there are a small number of neurons from auditory cortex, but there are no neurons from motor cortex. With this in mind, let’s take a look.
Neuron Phenotype Ontology OWL File
1. Look for the Submissions section on the Neuron Phenotype Ontology page in the BioPortal, and find the latest version (top) and under Downloads click on the OWL link. The npo.ttl file will download to your computer.
2. In Protégé, open the File menu and select Open. Find and load the npo.ttl file.
A Brief Tour of Neuron Phenotype Ontology
We can do a simple query but first a very brief tour of our ontology so that you may start building your own queries. Select the Entities tab at top left and you see several tabs below. Open the Classes sub-tab. These are the things, in OWL known as classes, that you may work with in the Neuron Phenotype Ontology. You may only see the top thing owl:Thing. Select the > to the left of owl:Thing and a whole sub-tree of classes should appear. Labels for many of these may not make sense to the human reader but are important classes for machine reasoning.
Let’s find an important human readable class in this ontology using Protégé search. Click on the Search… button in the far upper right. The Search window pops up (see Figure 3 above). Type in morphological phenotype. At the top of the resulting list should be the class, represented by an orange circle at left, named ‘Morphological phenotype’. The blue rectangles represent properties. More on these in a minute. Double click on the ‘Morphological phenotype’ class. Move or close your Search window and you’ll see that a sub-tree opened up in your Classes sub-tab list and the class Morphological phenotype is highlighted. Searching is the best way to get around in an ontology.
Look at the list of sub-classes under the Morphological phenotype class (select the > at left to see them) . These are the pre-defined cell morphology phenotypes currently in the Neuron Phenotype Ontology. Also notice that there are a whole lot of other classes under the high-level Phenotype class that break out neuron phenotypes in different ways including axon, cell, dendrite, and electrophysiological. You can use these classes to find the data you want.
To carry out a query you’ll need properties in addition to classes. You can think of classes as things in the world. Physical or mental things. Properties are, well, properties of those things. A bicycle is a thing with a color property. A blue bicycle. Blue is a thing that is a property of the bicycle. Likewise, a morphological phenotype is a thing but it is also a property of cells, including neurons. Notice that listed under the ‘Morphological phenotype’ class (yellow circle) in Figure 3 above is the hasMorphologicalPhenotype property (blue rectangle). Now you’re ready to query the data for a particular neuron phenotype.
Query for Neurons with Dendritic Spines
Many of the most complex neurons in cortex have dendrites with a lot of spines on them like in the image of two human cortical neurons at top in Figure 1. Each spine has a synapse (yellow dot) with an incoming axon from another neuron. We will query for data about neurons with spiny dendrites. Select the DL Query tab in Protégé and type Neuron and hasMorphologicalPhenotype some ‘Spiny phenotype’ into the Query (class expression) text box.
Neuron is the class of cells you’re looking for with spiny morphology. That is, you’re looking for every neuron that has the morphological phenotype of dendrites with spines (‘Spiny phenotype’ is a sub-class of ‘Dendrite phenotype’).
You should get around 454 results. Most and perhaps all data for this query was collected at the Allen Institute. Normally what you want a query to return is linked-data from around the world. In this case we are using a proof-of-concept version of the ontology with a small local dataset. Also, the classes you see in your results represent the data. Typically end results provide links (URLs) to the actual data.
Why do neuroscientists continue to painstakingly search for and extract data from research papers by hand when technology is available to automate much of the process? The number one reason is most likely the complexity of neuroscience data. A lot of work went into creating the Neuron Phenotype Ontology … and there is much still to do. Another probable reason is the work and expense that goes into building and maintaining the necessary web-based user friendly tools. Finally, there is the question of how much linked-data is openly available to the neuroscience community. I believe that there is a large and growing collection of neuroscience linked-data now openly available. Check back soon and I’ll share with you what I find.