One Brain, Two Computers
The Rationale for Developing a Microglial Modulator : Part 2
Discovering the Active Role Microglia Play in the Healthy Brain
Copyright 2017 David Mayfield and Blue Bridge Life Science
Microglia are the brain’s principal immuno-competent cells making up roughly 10-15 percent of the CNS cell population. Prior to 2005, they were thought to play a largely quiescent, passive role under physiological conditions. As the brain's resident phagocytic immune cells, they could certainly be called into action -- it was thought -- but only in response to an immune challenge to the brain caused by infection, injury, or established disease.
In 2005, however, this dogma was challenged. The publication of groundbreaking, time-lapse photography based on two photon microscopy of the living rodent brain, showed that microglia are far from quiescent in the healthy CNS (Nimmerjahn, et al, 2005, Davalos et al, 2005). They are, in fact, frenetically motile unlike any other brain cell type.
In the healthy adult brain, microglia have small cell bodies, and numerous long, thin, arm-like processes which they rapidly extend and retract throughout their immediate cellular environs, collectively exploring the entire volume of the brain parenchyma. Their explorations appear to be particularly focused on the synaptic connections between neurons. Under healthy conditions, microglia make transient physical contact with all 100 trillion synapses of the mature neuronal network every 30-60 minutes. And, interestingly, the frequency and duration of microglial contact with a particular synapse has been shown to be a function of the level of neuronal activity in the circuitry of which it is a part (Wake et al, 2009, Panatier & Robitaille, 2012). These unprecedented and totally unexpected observations led researchers to ask two new questions:
|
2 Photon Microscopy-Time Lapse
of Individual Microglia Cell Motility (Nimmerjahn, 2005) Microglia in the Living Brain:
Thin-skulled 2 Photon Microscopy (Parkhurst, 2009) |
Harnessing Microglia's Immunological Machinery to Non-Immunological Functions within the Brain
The body of research pursued in response to these questions is already large, but the short answer to them is as follows: evolution has elegantly harnessed microglia's innate immunological capabilities to serve an entirely non-immunological function within the central nervous system. Microglia, we now understand, help to manage the plasticity (especially longer term, structural plasticity) of the 100 trillion synapses wiring together the adult human brain's 100 billion neurons. How they came, through evolution, to be yoked to this and other important non-immunological tasks relates, paradoxically, to their developmentally immunological origins.
Unlike other brain cells (including other glia), microglia originate embryologically as primitive macrophages from the extra- embryonic mesoderm, rather than the neural ectoderm (Ginhoux et al, 2010, Prinz, Erny & Hagemeyer, 2017). They colonize the brain early in development prior to the closure of the blood brain barrier. And they bring with them the full immunological machinery of their macrophage cousins residing outside the CNS -- machinery which includes, in addition to phagocytic capacity, an extensive array of communication/signalling molecules and their receptors which enable all immuno-competent cells to;
In the brain, however, it is healthy neurons which are the principal interlocutors of microglia -- not diseased or injured cells -- even though the language spoken includes the same immunological vocabulary of cytokines, chemokines, trophic and toxic factors and their receptors (Garay and McAllister, 2010). And the dialogue between microglia and neurons is incessant. The subject of their conversation turns out frequently to be about whether the strength of signal received by a post-synaptic neuron from one or more of its pre-synaptic neuronal partners should be dialed up, or dialed down (in an analog fashion), given what microglia sense to be significant alterations in the active firing patterns between these partnered neurons caused by novel experience.
Unlike other brain cells (including other glia), microglia originate embryologically as primitive macrophages from the extra- embryonic mesoderm, rather than the neural ectoderm (Ginhoux et al, 2010, Prinz, Erny & Hagemeyer, 2017). They colonize the brain early in development prior to the closure of the blood brain barrier. And they bring with them the full immunological machinery of their macrophage cousins residing outside the CNS -- machinery which includes, in addition to phagocytic capacity, an extensive array of communication/signalling molecules and their receptors which enable all immuno-competent cells to;
- recognize signal about the disease state of a neighboring cell - either in the form of chemical attractants released into the extra-cellular space (such as, for example, fractalkine/CX3CL1, or ATP), or in the form of a tag presented on, or bound to, the cell's surface via, for example, major histocompatibility complex 1 or 2 (MHC1 or MHC 11), or a complement antigen (C1q/C3),
- convey a response back to those neighboring cells via the release of pro or anti-inflammatory cytokines and trophic or toxic factors (for example, IL-1, IL-6, TNF-alpha, Il-10, IL-4, BDNF, NO, O2, to name only the most commonly studied),
- include in that response "instructions" which alter the behavior and fate of those cells by, for example, triggering second messenger system cascades within a diseased cell which, in turn, may trigger 1) local phosphorylation or dephosphorylation of extant proteins changng their function, 2) activation or deactivation of transcription factors impacting the expression levels of specific genes in RNA transcripts and thereby altering the ratios of various gene products and their function, and 3) epigenetic methylation of genes, which may shut down particular genes altogether. Activation of these pathways may ultimately lead to programmed cell death and phagocytosis in the case of irreparably damaged cells, or they may invoke pro-survival pathways.
In the brain, however, it is healthy neurons which are the principal interlocutors of microglia -- not diseased or injured cells -- even though the language spoken includes the same immunological vocabulary of cytokines, chemokines, trophic and toxic factors and their receptors (Garay and McAllister, 2010). And the dialogue between microglia and neurons is incessant. The subject of their conversation turns out frequently to be about whether the strength of signal received by a post-synaptic neuron from one or more of its pre-synaptic neuronal partners should be dialed up, or dialed down (in an analog fashion), given what microglia sense to be significant alterations in the active firing patterns between these partnered neurons caused by novel experience.
One Brain, Two Computers: A Digital Computer Built to Compute, An Analog Computer Built to Teach
Neuroscientists since Donald Hebb have understood that the brain’s capacity to change the strength of its own synaptic connections -- as it detects in altered neuronal firing patterns previously unrecognized correlations between events -- is the biological substrate of an intelligence which not only computes (i.e. thinks), but also learns to compute differently as novel experience challenges what it already knows. And neuroscientists have long appreciated that this same activity and experience dependent plasticity is fundamental to the immature brain’s extraordinary capacity to self-organize into a mature computing machine well tuned not to random noise reverberating within the inexperienced brain, but to relevant signal sensed from both the body and the external environment of the organism.
But prior to the 2005 publications and the microglia-focused research which has followed, neuroscientists did not realize that neurons and the circuits composed of them couldn’t manage this activity and experience dependent plasticity entirely on their own. In our brains at least, for experience to durably modify the function, structure and organization of neuronal circuitry, a non-neuronal interlocutor, microglia, turns out to be required.
It is as if our brains are comprised of two computers rather than one:
What the digital machine knows is rather durably stored in the relative strengths of the 100 trillion synapses through which pre-synaptic neurons send signals to a shared post-synaptic partner. The relative weights of the synaptic connections determine which pre-synaptic neurons more or less reliably contribute information leading to the post-synaptic neuron firing an action potential, the brain's digital yes/no computation of biological relevance.
What the digital brain learns is incorporated biologically as the microglial analog machine durably alters the weight of synaptic inputs from a particular pre-synaptic partner relative to the strength of inputs from other pre-synaptic partners connected to the same post-synaptic neuron. Microglia make these alterations (as we shall see) after detecting in the context of novel experience that a relatively unweighted connection is proving, more reliably than the brain "expects" -- given the relative differences between synaptic weights previously built into the network -- to lead to the post-synaptic neuron firing.
But prior to the 2005 publications and the microglia-focused research which has followed, neuroscientists did not realize that neurons and the circuits composed of them couldn’t manage this activity and experience dependent plasticity entirely on their own. In our brains at least, for experience to durably modify the function, structure and organization of neuronal circuitry, a non-neuronal interlocutor, microglia, turns out to be required.
It is as if our brains are comprised of two computers rather than one:
- a neuron-based digital machine built to compute, in binary fashion, the relevance/no relevance of particular experience strictly in terms of what it already knows, and
- a microglia-based analog machine built to teach the digital computer how to compute.differently given novel experience which it can detect but not yet understand.
What the digital machine knows is rather durably stored in the relative strengths of the 100 trillion synapses through which pre-synaptic neurons send signals to a shared post-synaptic partner. The relative weights of the synaptic connections determine which pre-synaptic neurons more or less reliably contribute information leading to the post-synaptic neuron firing an action potential, the brain's digital yes/no computation of biological relevance.
What the digital brain learns is incorporated biologically as the microglial analog machine durably alters the weight of synaptic inputs from a particular pre-synaptic partner relative to the strength of inputs from other pre-synaptic partners connected to the same post-synaptic neuron. Microglia make these alterations (as we shall see) after detecting in the context of novel experience that a relatively unweighted connection is proving, more reliably than the brain "expects" -- given the relative differences between synaptic weights previously built into the network -- to lead to the post-synaptic neuron firing.
Learning in the Mature Brain
Of course the adult brain already knows almost everything it needs to know in order to compute the relevance of experience to the well being of the animal. So, it doesn't experience very much which runs "counter to the expectations" durably built into the weights of synapses through long and recurring experience. of a lifetime. In this context, the digital computer dominates. But the analog computer still plays a critical educational role throughout life, mediating all long-term change in neuronal circuitry predicated on learning, such as new memory formation, new motor or cognitive skill acquisition, and adaptation of the brain to reward, to pain, and to injury. For the mature, experienced animal, novel experience will challenge relatively minor elements of what its digital computer already knows and understands. In order to detect and respond to such a learning challenge, the microglia of the analog computer must operate at the very fine scale of the individual branches of the axon terminal of the individual neuron which is responsive to the novel experience in question - "listening" to whether the action potential of this neuron unexpectedly, and more reliably than before, leads to the firing of an action potential in its downstream post-synaptic partner. If it does, microglia will participate in strengthening their particular connection as a function of this increased reliability while leaving most of neuronal network unaltered.
And the mature morphology of microglial cells allows for this fine-grained level of dexterity. These are the microglia depicted in the seminal 2005 publications. With small cell bodies and long, very thin, and exceptionally motile arm-like processes, they are equipped to explore the finely articulated dendritic arbor of the individual neuron in order to detect newly coordinated, or newly uncoordinated, firing patterns between it and its (potentially) thousands of pre-synaptic neuronal partners. And, having listened to alterations in firing patterns, they deliver instructions in response -- deploying tiny amounts of pro and anti-inflammatory cytokines, as well as glia-derived trophic and toxic factors at the scale of the individual synapse. Using Hebbian algorithms, microglia direct the incorporation of new knowledge by dialing the strength of individual synaptic connections up or down, effectively changing the relative weights of the pre-synaptic inputs to a shared post-synaptic partner and therefore changing the brain's "expectation" regarding which pre-synaptic inputs will in the future more or less reliably lead to the firing of the action potential in their post synaptic partner.
AND, at the same time, microglia make compensatory "homeostatic" adjustments to "scale" the overall balance of excitatory and inhibitory inputs to the shared post-synaptic neuron in order to protect the neuronal network from epileptic instability as it undergoes Hebbian change (Stellwagon and Malenka, 2006, Turigiano, 2006, Ferrini and De Koninck, 2013).
Hebbian plasticity is at the root of the brain's life-long machine-learning capacities. What is it? Briefly, where novel experience causes the action potential of a particular pre-synaptic neuron more reliably to lead to the firing of an action potential in its post-synaptic partner (pre before post), the functional strength of their particular synaptic connection will be enhanced (long term potentiation) by the brain's analog machinery. The initial changes are very rapid and easily reversed, and caused by fast, calcium-dependent, NMDA receptor mediated signals localized to the post-synaptic dendritic spine which will induce the insertion of more excitatory AMPA receptors in the surface of the spine. More durable changes occur, however, as the microglial elements of the analog computer sense that the disturbance caused by a newly coordinated firing pattern is not random, but in fact represents a significant, previously unrecognized correlation between events which is likely to recur. Microglia then deploy signals such as the anti-inflammatory cytokine, IL-10, and the glia-derived trophic factor, BDNF (Parkhurst et al, 2013, Lim et al, 2013) which provoke the new construction (synaptogenesis) of additional paired post-synaptic dendritic spines and pre-synaptic terminals so as to reinforce the long term potentiation of the connection between an increasingly coordinated pair of neuronal partners, (Parkhurst et al, 2013, Lim et al, 2013, Miyamoto et al, 2016).
Likewise, where novel experience results in a new incoordination of pre and post-synaptic firing patterns, the brain's analog machine will weaken the connection (long term depression). Here incoordination means the post-synaptic neuron fires with unanticipated regularity before a specific pre-synaptic neuronal partner fires (post before pre), suggesting that other pre-synaptic inputs are carrying the relevant signal to the post-synaptic cell. Again, the initial weakening of the connection is fast, and easily and quickly reversed, and a function of rapid, calcium dependent NMDA receptor mechanisms determining AMPA receptor population local to the dendritic spine. But as microglia sense that the firing of the pre-synaptic input is in fact less likely now to carry signal relevant to the firing of the post synaptic cell, it will deploy from its repertoire of cytokines, chemokines, trophic and toxic factors, chemical cues to encourage the permanent weakening or even the phagocytic elimination of their synaptic connection (synaptic pruning) (Miyamoto et al, 2016),
And the mature morphology of microglial cells allows for this fine-grained level of dexterity. These are the microglia depicted in the seminal 2005 publications. With small cell bodies and long, very thin, and exceptionally motile arm-like processes, they are equipped to explore the finely articulated dendritic arbor of the individual neuron in order to detect newly coordinated, or newly uncoordinated, firing patterns between it and its (potentially) thousands of pre-synaptic neuronal partners. And, having listened to alterations in firing patterns, they deliver instructions in response -- deploying tiny amounts of pro and anti-inflammatory cytokines, as well as glia-derived trophic and toxic factors at the scale of the individual synapse. Using Hebbian algorithms, microglia direct the incorporation of new knowledge by dialing the strength of individual synaptic connections up or down, effectively changing the relative weights of the pre-synaptic inputs to a shared post-synaptic partner and therefore changing the brain's "expectation" regarding which pre-synaptic inputs will in the future more or less reliably lead to the firing of the action potential in their post synaptic partner.
AND, at the same time, microglia make compensatory "homeostatic" adjustments to "scale" the overall balance of excitatory and inhibitory inputs to the shared post-synaptic neuron in order to protect the neuronal network from epileptic instability as it undergoes Hebbian change (Stellwagon and Malenka, 2006, Turigiano, 2006, Ferrini and De Koninck, 2013).
Hebbian plasticity is at the root of the brain's life-long machine-learning capacities. What is it? Briefly, where novel experience causes the action potential of a particular pre-synaptic neuron more reliably to lead to the firing of an action potential in its post-synaptic partner (pre before post), the functional strength of their particular synaptic connection will be enhanced (long term potentiation) by the brain's analog machinery. The initial changes are very rapid and easily reversed, and caused by fast, calcium-dependent, NMDA receptor mediated signals localized to the post-synaptic dendritic spine which will induce the insertion of more excitatory AMPA receptors in the surface of the spine. More durable changes occur, however, as the microglial elements of the analog computer sense that the disturbance caused by a newly coordinated firing pattern is not random, but in fact represents a significant, previously unrecognized correlation between events which is likely to recur. Microglia then deploy signals such as the anti-inflammatory cytokine, IL-10, and the glia-derived trophic factor, BDNF (Parkhurst et al, 2013, Lim et al, 2013) which provoke the new construction (synaptogenesis) of additional paired post-synaptic dendritic spines and pre-synaptic terminals so as to reinforce the long term potentiation of the connection between an increasingly coordinated pair of neuronal partners, (Parkhurst et al, 2013, Lim et al, 2013, Miyamoto et al, 2016).
Likewise, where novel experience results in a new incoordination of pre and post-synaptic firing patterns, the brain's analog machine will weaken the connection (long term depression). Here incoordination means the post-synaptic neuron fires with unanticipated regularity before a specific pre-synaptic neuronal partner fires (post before pre), suggesting that other pre-synaptic inputs are carrying the relevant signal to the post-synaptic cell. Again, the initial weakening of the connection is fast, and easily and quickly reversed, and a function of rapid, calcium dependent NMDA receptor mechanisms determining AMPA receptor population local to the dendritic spine. But as microglia sense that the firing of the pre-synaptic input is in fact less likely now to carry signal relevant to the firing of the post synaptic cell, it will deploy from its repertoire of cytokines, chemokines, trophic and toxic factors, chemical cues to encourage the permanent weakening or even the phagocytic elimination of their synaptic connection (synaptic pruning) (Miyamoto et al, 2016),
Learning in the Developing Brain: Pre-programming Steps
In the prenatal and postnatal brain of the developing human, it is the analog computer which dominates, first as it mediates the wiring of the neuronal circuitry, and then as it teaches a functional but naive digital computer how to compute in the first place at a time when it knows little or nothing at all.
During this developmental period, microglia do their work sequentially, somewhat in contrast to the adult brain. The first steps in the sequence are preparatory and can be thought of as wiring the hardware of the digital computer: Using a tightly regulated mix of pro and anti-inflammatory cytokines, as well as cytotrophic and cytotoxic factors produced in response to cues expressed by a developing superabundance of 200 billion immature neurons, microglia "instruct' them about when and where to extend their axons toward potentially viable synaptic partners (Pont-Lezica et al, 2014-- gestational week 12 to infancy in humans),
After the axonal pathways of the neuronal population have been layed down (the time course of which varies significantly across brain region and functional circuitry), microglia together with astrocytes then mediate massive regional waves of synaptogenesis -- gestational week 18/19 through mid childhood (Miyamoto et al, 2016, Bilmoria and Stevens, 2015, Deverman and Patterson, 2009, Boulanger et al, 2009) wiring the elemental microprocessors of the brain (neurons) together into a functional, fully interconnected, but nonetheless unintelligent computational network, It can transduce electrical, chemical and, in specialized neurons, mechanical inputs one into the other. It can fire action potentials in response to pre-synaptic inputs which sum to meet the membrane potential for relevance (-55mV), But this electrical activity, to a first approximation, initially represents undifferentiated noise. The firing of any given neuron within the circuitry of the inexperienced brain isn't patterned or coordinated activity, but occurs in response to the random firing of action potentials by its respective pre-synaptic partners which themselves fire in response to the random activity of their own upstream pre-synaptic partners.
At this point, the digital brain faces the fundamental question of machine learning: if, as a first principle, it can think (compute) only in terms of what it already knows, how does the naive digital machine, which knows nothing at all, first acquire the terms in which to think? Or, to pose the question biologically, if what the digital neuronal machine knows is stored in the unequal weights of synapses connecting each post synaptic neuron to its many (potentially thousands) pre-synaptic partners, from whence the original unequal weighting? We've known since the Nobel prize winning research of Hubel and Wiesel (Hubel and Wiesel, 1963a, Hubel and Wiesel 1963b) that the acquisition of the primordial terms of thought -- such as the visual properties of form, position, orientation, and direction of movement necessary for visual computation -- requires the naive digital brain actively to experience foundational spatial and temporal cues from the environment (and body) of the animal. But their work posed rather than answered the basic question: how does the digital neuronal machine not only detect these foundational spatial and temporal cues, but incorporate them into understanding by durably changing the shape and function of neuronal circuitry?
The answer is: it doesn't. The neuron-based digital machine, as we have learned in the aftermath of 2005, cannot pull itself up by its own bootstraps. The digital machine requires a third party interlocutor, microglia, to build intelligence into an initially unintelligent computing machine.
During this developmental period, microglia do their work sequentially, somewhat in contrast to the adult brain. The first steps in the sequence are preparatory and can be thought of as wiring the hardware of the digital computer: Using a tightly regulated mix of pro and anti-inflammatory cytokines, as well as cytotrophic and cytotoxic factors produced in response to cues expressed by a developing superabundance of 200 billion immature neurons, microglia "instruct' them about when and where to extend their axons toward potentially viable synaptic partners (Pont-Lezica et al, 2014-- gestational week 12 to infancy in humans),
After the axonal pathways of the neuronal population have been layed down (the time course of which varies significantly across brain region and functional circuitry), microglia together with astrocytes then mediate massive regional waves of synaptogenesis -- gestational week 18/19 through mid childhood (Miyamoto et al, 2016, Bilmoria and Stevens, 2015, Deverman and Patterson, 2009, Boulanger et al, 2009) wiring the elemental microprocessors of the brain (neurons) together into a functional, fully interconnected, but nonetheless unintelligent computational network, It can transduce electrical, chemical and, in specialized neurons, mechanical inputs one into the other. It can fire action potentials in response to pre-synaptic inputs which sum to meet the membrane potential for relevance (-55mV), But this electrical activity, to a first approximation, initially represents undifferentiated noise. The firing of any given neuron within the circuitry of the inexperienced brain isn't patterned or coordinated activity, but occurs in response to the random firing of action potentials by its respective pre-synaptic partners which themselves fire in response to the random activity of their own upstream pre-synaptic partners.
At this point, the digital brain faces the fundamental question of machine learning: if, as a first principle, it can think (compute) only in terms of what it already knows, how does the naive digital machine, which knows nothing at all, first acquire the terms in which to think? Or, to pose the question biologically, if what the digital neuronal machine knows is stored in the unequal weights of synapses connecting each post synaptic neuron to its many (potentially thousands) pre-synaptic partners, from whence the original unequal weighting? We've known since the Nobel prize winning research of Hubel and Wiesel (Hubel and Wiesel, 1963a, Hubel and Wiesel 1963b) that the acquisition of the primordial terms of thought -- such as the visual properties of form, position, orientation, and direction of movement necessary for visual computation -- requires the naive digital brain actively to experience foundational spatial and temporal cues from the environment (and body) of the animal. But their work posed rather than answered the basic question: how does the digital neuronal machine not only detect these foundational spatial and temporal cues, but incorporate them into understanding by durably changing the shape and function of neuronal circuitry?
The answer is: it doesn't. The neuron-based digital machine, as we have learned in the aftermath of 2005, cannot pull itself up by its own bootstraps. The digital machine requires a third party interlocutor, microglia, to build intelligence into an initially unintelligent computing machine.
Learning in the Developing Brain: Tuning a Naive Digital Machine to Environmental Signal
And this is where the recent discoveries regarding microglia get really interesting. Microglia accomplish their developmental programming task not by first connecting one microprocessor to another and then constructing one semi-intelligent microcircuit at a time in order to build up a complex fully intelligent machine from small parts and simple elements. Rather, microglia begin their machine programming task with a fully functional, fully interconnected digital machine in place. It is composed of a more than sufficient number of neurons, and a more than sufficient number of synapses which connect those neurons into a network of more than sufficient logical complexity to fully represent/compute all those properties of the world which are relevant to the behavior and well being of the animal. And then -- through an iterative, trial and error process which begins as soon as the digital brain first detects and actively responds to spatial and temporal signals from the environment (and the body) -- microglia tune the initially noisy naive brain by "revealing" signal-responsive circuits which are, quite literally, already there and waiting to be found; using the same Hebbian rules described above for the adult brain, and the same repertoire of immunological tools (cyotokines, chemokines, trophic and toxic factors) to communicate with and provoke behavior in neurons, microglia listen to the patterns of electrical activity wiithin the digital network, and then begin to weight those synapses which connect neurons whose firing patterns prove through many trials frequently to be coordinated pre-synaptically with upstream neurons which are themselves responsive to these primordial spatial and temporal cues provided by original experience. And they will un-weight those which are somewhat less responsive. And, crucially, they will fully un-weight and then prune those which, after repeated trials of experience, prove to be only responsive to, or productive of, noise. If a pre-synaptic neuron which had originally been connected to a post synaptic partner is orphaned as a result of pruning their connection (not connected as an input device to any other downstream neuronal partner or partners), it will also be phagocytosed by microglia.
In order to build intelligence into the brain this way -- refining what are essentially "pre-established", signal-responsive circuits by sculpting away those elements within circuits which activity and experience prove are only responsive to, or productive of, noise -- nature must initially supply the animal with a large superabundance of both neurons and synapses in order to be sure that, after the process of circuit refinement is complete, the brain retains sufficient neurons, synapses and logical complexity of connections to represent all that must be represented in the adult brain for the animal to produce behavior required of survival. This is exactly what nature does provide; the naive but functional digital machine is initially composed of 200 billion neurons, 100 billion of which will need to be winnowed by microglia (phagocytosed) during development, and 200 trillion synapses 100 trillion of which must be pruned by microglia, as the analog machine pulls an initially weak environmentally derived signal out from the predominantly noisy electrical activity of the inexperienced neuron-based digital machine. (Schaffer et al, 2012, Trembley et al, 2010, Trembley et al, 2011, Paolicelli et al. 2011)
The temporal window for this "critical" period of activity and experience dependent circuit refinement varies widely by brain region and sub-region, with some circuits refined prior to birth, sensory and motor cortex circuitry fully refined within the first two to three years of life, and other circuitry (for example, circuitry associated with higher cognitive, emotional and executive function) remaining quite plastic through childhood, adolescence and even early adulthood. But regardless of region and precise timing, such large scale, and in many regions robustly fast-paced modifications to neuronal circuitry during development require microglia to manifest a less delicate, more macrophage-like form and function than the microglia of the adult brain which dexterously modify one synapse at a time leaving most synaptic connections of the mature brain unchanged. And, indeed, this is the case, Microglia during development resemble much more their macrophage cousins on the non-parenchymal margins of the CNS and in the periphery. They have larger cell bodies and much shorter and thicker arm-like processes. Furthermore, developmental microglia deploy larger gross quantities of immune molecules, and express their receptors more abundantly (such as CD11b) than do adult microglia, And in the aggregate, the mix of cytokines, chemokines, trophic and toxic factors produced by microglia at this stage favors a pro rather than anti-inflammatory ratio in contrast to adult microglia under healthy conditions (see Bilbo lectures). This stands to reason given the general mode of circuit refinement in development which proceeds largely through a massive process of subtraction, whereby billions of superfluous neurons, and trillions of synapses express antigen presenting molecules such as MHC1 and MHC11 and complement molecules localized to their surface as tags marking them for phagocytosis by microglia.
But as circuitry is refined, and the signal to noise ratio represented by electrical activity within circuitry increasingly favors signal over noise, then microglia must begin to adopt the more ramified form and function which typifies their mature identity. For their role is no longer to use Hebbian rules coarsely to distinguish those synapses carrying signal from those carrying noise and to physically root out those elements of circuitry which are only responsive to, or productive of, noise. Rather, their mature task will increasingly be to use Hebbian rules to distinguish between synapses all of which carry signal, albeit signal graded by degrees of salience, and then to modulate their relative weights in light of experience -- by adding or removing dendrictic spines and terminal boutons between particular neuronal partners, but never (or very rarely) eliminating their connection altogether.
Indeed, the end of the so-called "critical" developmental period of activity and experience dependent plasticity within particular circuitry and brain regions (initially described by Hubel and Wiesel) is triggered once microglia sense that all synapses connecting neuronal partners within relevant circuitry carry signal rather than noise. At this point, the digital machine switches from one which must lose synapses and neurons in order to properly mature, to one which must preserve net synapse density, and its non-proliferative population of neurons at all cost in order to remain healthy. This fundamental shift in the modus operandi of the digital computer requires an accompanying permanent change in microglia gene transcription profile and phenotype enabling a shift away from macrophage-like form and function typical of the developing brain to the ramified, delicate, highly motile phenotype revealed in the path breaking 2005 2-photon microscopy of the healthy adult brain.
In order to build intelligence into the brain this way -- refining what are essentially "pre-established", signal-responsive circuits by sculpting away those elements within circuits which activity and experience prove are only responsive to, or productive of, noise -- nature must initially supply the animal with a large superabundance of both neurons and synapses in order to be sure that, after the process of circuit refinement is complete, the brain retains sufficient neurons, synapses and logical complexity of connections to represent all that must be represented in the adult brain for the animal to produce behavior required of survival. This is exactly what nature does provide; the naive but functional digital machine is initially composed of 200 billion neurons, 100 billion of which will need to be winnowed by microglia (phagocytosed) during development, and 200 trillion synapses 100 trillion of which must be pruned by microglia, as the analog machine pulls an initially weak environmentally derived signal out from the predominantly noisy electrical activity of the inexperienced neuron-based digital machine. (Schaffer et al, 2012, Trembley et al, 2010, Trembley et al, 2011, Paolicelli et al. 2011)
The temporal window for this "critical" period of activity and experience dependent circuit refinement varies widely by brain region and sub-region, with some circuits refined prior to birth, sensory and motor cortex circuitry fully refined within the first two to three years of life, and other circuitry (for example, circuitry associated with higher cognitive, emotional and executive function) remaining quite plastic through childhood, adolescence and even early adulthood. But regardless of region and precise timing, such large scale, and in many regions robustly fast-paced modifications to neuronal circuitry during development require microglia to manifest a less delicate, more macrophage-like form and function than the microglia of the adult brain which dexterously modify one synapse at a time leaving most synaptic connections of the mature brain unchanged. And, indeed, this is the case, Microglia during development resemble much more their macrophage cousins on the non-parenchymal margins of the CNS and in the periphery. They have larger cell bodies and much shorter and thicker arm-like processes. Furthermore, developmental microglia deploy larger gross quantities of immune molecules, and express their receptors more abundantly (such as CD11b) than do adult microglia, And in the aggregate, the mix of cytokines, chemokines, trophic and toxic factors produced by microglia at this stage favors a pro rather than anti-inflammatory ratio in contrast to adult microglia under healthy conditions (see Bilbo lectures). This stands to reason given the general mode of circuit refinement in development which proceeds largely through a massive process of subtraction, whereby billions of superfluous neurons, and trillions of synapses express antigen presenting molecules such as MHC1 and MHC11 and complement molecules localized to their surface as tags marking them for phagocytosis by microglia.
But as circuitry is refined, and the signal to noise ratio represented by electrical activity within circuitry increasingly favors signal over noise, then microglia must begin to adopt the more ramified form and function which typifies their mature identity. For their role is no longer to use Hebbian rules coarsely to distinguish those synapses carrying signal from those carrying noise and to physically root out those elements of circuitry which are only responsive to, or productive of, noise. Rather, their mature task will increasingly be to use Hebbian rules to distinguish between synapses all of which carry signal, albeit signal graded by degrees of salience, and then to modulate their relative weights in light of experience -- by adding or removing dendrictic spines and terminal boutons between particular neuronal partners, but never (or very rarely) eliminating their connection altogether.
Indeed, the end of the so-called "critical" developmental period of activity and experience dependent plasticity within particular circuitry and brain regions (initially described by Hubel and Wiesel) is triggered once microglia sense that all synapses connecting neuronal partners within relevant circuitry carry signal rather than noise. At this point, the digital machine switches from one which must lose synapses and neurons in order to properly mature, to one which must preserve net synapse density, and its non-proliferative population of neurons at all cost in order to remain healthy. This fundamental shift in the modus operandi of the digital computer requires an accompanying permanent change in microglia gene transcription profile and phenotype enabling a shift away from macrophage-like form and function typical of the developing brain to the ramified, delicate, highly motile phenotype revealed in the path breaking 2005 2-photon microscopy of the healthy adult brain.
Implications for Medical Neuroscience: The Microglia-driven Analog Machine Can Break
These differences between the function and morphology of microglia in the developing vs mature brain -- and the success with which microglia do or do not make the transition from a more macrophage-like juvenile form and funciton to a more microglia like mature form and function -- have profound consequences for medical neuroscience as we shall see in Pt. 3. Nevertheless, whether the brain has reached maturity or is still developing, the analog computer teaches, and the digital computer learns, as microglia durably shape and alter the strength of synaptic connections between neuronal partners in response to activity and experience dependent changes in neuronal firing patterns. And the most intriguing aspect of microglial activity in this realm, is that they do their work by having re-purposed their inherited repertoire of immunological tools to accomplish this fundamentally non-immunological task.
Two paradigm-shifting questions quickly follow from the basic discovery that microglia play such an critical machine learning role in the physiological brain:
We now know that the analog computer can break and does so with widely varying pathological consequences depending on the time of life and brain region when and where it breaks. Further, we know that the reason it breaks, quite ironically, is related to the very reason why microglia are also so well equipped as the interlocutors of neurons and managers of synaptic plasticity; namely, they share ontogenetic origins with peripheral tissue macrophages which, like microglia, are programmed as immuno-competent cells in the fetal yolk sac prior to migration to their specific developed tissue of residence. Microglia are therefore armed with all of the immunological tools of a phagocyte (in particular, the language of cytokines, chemokines, trophic and toxic factors and their receptors) which are needed by immune cells outside of the brain to respond effectively to infection by pathogens. Under certain inauspicious circumstances (and given the right mix of environmental risk factors, and, perhaps, genetic vulnerabilities), microglia may, as a consequence of this inheritance, react in error immunologically to a systemic immune challenge as if they were tissue macrophages residing in the periphery, rather than brain cells residing within the parenchyma. They are particularly vulnerable to this error during development, when they provisionally retain their macrophage-like reactive nature needed to sculpt brain circuits.
They may "forget", in other words, that their evolved function within the brain is computer programming under physiological conditions, rather than macrophage-like host defense under pathological ones. If they do react immunologically, they can't help but abandon their non-immunological role as the interlocutor of neurons, essential to the brain's machine-learning capacities - a result which, in and of itself, will impact cognition and behavior, including most importantly the brain's capacity to consolidate new learning. Under the worst conditions, they may even mistakenly deploy their inherited immunological machinery to target healthy synaptic connections, not for modulation in the context of a learning challenge, but for destruction in the context of a misperceived immunological threat.
Two paradigm-shifting questions quickly follow from the basic discovery that microglia play such an critical machine learning role in the physiological brain:
- can the microglia-regulated analog computer break?
- and what happens to brain health and neuronal function if it does break -- in utero, in infancy, in childhood, adolescence, adulthood or old age?
We now know that the analog computer can break and does so with widely varying pathological consequences depending on the time of life and brain region when and where it breaks. Further, we know that the reason it breaks, quite ironically, is related to the very reason why microglia are also so well equipped as the interlocutors of neurons and managers of synaptic plasticity; namely, they share ontogenetic origins with peripheral tissue macrophages which, like microglia, are programmed as immuno-competent cells in the fetal yolk sac prior to migration to their specific developed tissue of residence. Microglia are therefore armed with all of the immunological tools of a phagocyte (in particular, the language of cytokines, chemokines, trophic and toxic factors and their receptors) which are needed by immune cells outside of the brain to respond effectively to infection by pathogens. Under certain inauspicious circumstances (and given the right mix of environmental risk factors, and, perhaps, genetic vulnerabilities), microglia may, as a consequence of this inheritance, react in error immunologically to a systemic immune challenge as if they were tissue macrophages residing in the periphery, rather than brain cells residing within the parenchyma. They are particularly vulnerable to this error during development, when they provisionally retain their macrophage-like reactive nature needed to sculpt brain circuits.
They may "forget", in other words, that their evolved function within the brain is computer programming under physiological conditions, rather than macrophage-like host defense under pathological ones. If they do react immunologically, they can't help but abandon their non-immunological role as the interlocutor of neurons, essential to the brain's machine-learning capacities - a result which, in and of itself, will impact cognition and behavior, including most importantly the brain's capacity to consolidate new learning. Under the worst conditions, they may even mistakenly deploy their inherited immunological machinery to target healthy synaptic connections, not for modulation in the context of a learning challenge, but for destruction in the context of a misperceived immunological threat.