CSE511 Brain and Memory Modeling Lecture 1 Notes Mon 8/30/10 Displayed syllabus from http://www.cs.sunysb.edu/~cse511; discussed course texts, exam, sample projects, and goals (learn brain terminology and discuss critical mechanisms for distributed memories), and asked everyone to send email to LW@IC.SUNYSB.EDU with Subject: line containing "cse511". Showed Paul Adams website for Synaptic Darwinism (http://syndar.org/padams/intro%20to%20lecture%20notes.htm) with notes for his Biology 338 course: From Synapse to Circuit: Self-Organization in the Brain. Discussed neuron outputs via axonal firing spikes (self-propagating ionic torrents through temporarily open gaps in the normally insulating cell membrane) that reach synaptic tips ( boutons ), open membrane pores to admit Ca 2+ ion that burst tiny vesicles of neurotransmitter molecules. The molecules diffuse across the narrow 20-50 nm (Bear, p. 105) synaptic gap and open receptor pores in the dendritic post-synaptic neuron. If enough (usually hundreds or even thousands of) axons contribute ion currents to one dendritic neuron, its axon soon generates a new output spike from the axon hillock, where its axon starts at the base of its bulbous central body, or soma. Discussed overlapping distributed firing patterns of millions of brain neurons as probable representations of memories, thoughts, and perceptions. Explained Hebbian cell assembly theory from 1947, Donald O. Hebb. If neuron A synapses on neuron B and A fires then followed soon afterwards (within a few milliseconds) by B firing, the synaptic strength from A to B grows stronger, so A needs fewer helper neurons to fire with it in the future to trigger B to fire. Modern neo-hebbian theory says synapse strengths are correlative : if A fires just before B fires, the synapse from A to B grows stronger; if A fires but B does not soon fire, the synapse from A to B grows slightly weaker. In magnitude, a single weakening is about 1/100 as much as a single strengthening. Gave excerpts from Sir Fredric C. Bartlett, Remembering (1932), on the dynamic nature of memories ( conventionalization ) and their interaction to overwhelm fleeting perception: the story of Amerindians in a canoe that became Irishmen in a rowboat when remembered by British villagers, and an experiment with 100 people, where each was repeatedly shown tachistoscopic flashes of one image for 1 millisecond at a time until the viewer knew exactly all details of the image. One person recognized that he was seeing a photograph of sailboats in the harbor (which he personally had created and could not be the image) instead of correctly perceiving the repeatedly flashed image. The rest reported the details of radically different images that their minds created from their own memories without consciously realizing that what they were seeing was not what was being displayed in the repeated flashes. (See http://www.bookrags.com/research/bartlett-frederic-1886-1969-lmem-01/ ) Showed images from two of my 1978 Simulation papers (1) Large-scale simulation of brain cortices -- Wittie, SIMULATION, 31 (3)_ p73-78 Sept, 1978 and (2) Large network models using the brain organization simulation system (BOSS) -- Wittie, SIMULATION, 31 (3)_ p117-122 Sept, 1978). 1
Figure 1.1 from page 74 of paper (1) Large-scale simulation of brain cortices is reasonably accurate, except for the density of the thicket of axonal boutons shown at the end of the central output axon and the smooth surfaces of the input dendrites, shown without the characteristic spines where axons abut the outer membranes of the dendrites. For more detailed images, see figures 2.14 and 2.15 in the Bear text. 2
Figure 1.2 from page 75 of paper (1) Large-scale simulation of brain cortices shows increased onset delay and spike attenuation for axonal input pulses as they are received at thinner and thinner dendritic branches more distal from the primary axon spike initiation region the axon hillock at the base of the soma and start of the output axon. Parameters in the BOSS initialization system for statistically defining representation brain tissue models can specify attenuation and delay rates versus distance for each distinct type of neuron in the tissue. For more detailed images, see figure 4.27 on page 7 of Paul Adams Central synaptic transmission 1 http://syndar.org/padams/central%20synaptic%20transmission%201.pdf 3
Figure 1.3 from page 76 of paper (1) Large-scale simulation of brain cortices shows a major simplification of the shape of neurons in a BOSS brain tissue model that reduces computation times during model initialization. Each vaguely spherical neuron with its dendritic input branches and its axonal output(s) is approximated by a point center, the location of the neuron s soma, with rectilinear regions for its input (dendritic) and output (axonal) branch projections. The major computational steps of the BOSS model initialization procedure are 1) For each type of neuron in the tissue, use parameters giving the X,Y,Z lattice spacing of the neuron centers plus one giving the probability of having a neuron at each lattice point to locate all neuron instances in the model; 2) For each neuron center in the model, use the parameters for that neuron type giving the number of input (dendritic) and output (axonal) fields, and for each field, the X,Y,Z offsets of the field center from the center for the neuron (soma) and its X,Y,Z widths. 3) (This is the computationally limiting step that will require hierarchical definition of tissue substructure modules for models with millions of neurons) For each dendritic input field of each neuron instance in the tissue model, determine whether it physically overlaps the volume of any axonal output field of any neuron. If there is an overlap volume such as the small axonal output field for neuron B with the huge dendritic input field for neuron A in figure 3 and if there is a non-zero parameter for the density of axo-dendritic connections ( synapses ) from neurons of the type of B to neurons of the type of A, calculate the expected number of synapses in this overlap volume. Uniformly randomly within the X,Y,Z overlap volume locate the calculated number of synapses; any fractional part of the number generates between 0 and 1 additional synapse to give the expected number of synapses on average. 4
4) (After steps 1-3, all neurons and all synapses between them have been precisely located within the space of the tissue model) Calculate the distances for all axonal neuron soma to synapse to dendritic neuron soma, apply the parameters for the propagation rates and attention factors for each neuron type to determine the (millisecond) time delays and initial synapse strengths along each axo-dendritic path whenever the axonal neuron fires an ionic output spike. 5) The resulting neuron instance types, locations, synapse strengths, and signal path delays and attenuations can be used to specify the precise details of a discrete-event simulation system such as used for BOSS cerebellar cortex models in 1969-1978 or membrane differential equation solving simulation systems, such as the NEURON system. Figure 1.4 Figure 3 from page 120 in Large network models using the brain organization simulation system (BOSS) three of the major neuron types Purkinje, Golgi, and tiny ultra-densely packed granule cells in the cerebellar ( little brain ) cortex, which is used to shape rapid finely controlled body movements, such as needed for bicycle riding and piano playing. For more details, see Figures 25.13 25.15 in the Bear text. 5
Figure 1.5 Lobes of the human cerebral cortex from Neuroscience - 4ed by Purves, et al. The frontal lobe was thought for many years to be an unneeded part of the human brain, after the railroad dynamite tamping accident of Phineas Gage in 1848 and was cavalierly ablated in lobotomies that attempted to cure severe mental and social disorders in the 1950s. Lobotomies severely altered the personalities of many recipients, too often resulting in walking human vegetables. The frontal lobes, of the right and left hemispheres of the cerebrum, are now know to handle higher level planning tasks, including resolution of conflicting responses to stimuli. 6
Figure 1.6 Generalized neuron - like the pyramidal cells of the cerebral cortex - with multiple axonal output collaterals, showing nearly orthogonal connections to generate multiple firing spikes from one spike initiated at the axon hillock. Image adapted from a detail of Figure 1.2 of Neuroscience - 4ed by Purves, et al. 7
Figure 1.7A White and gray matter in a coronal slice through a human cerebral cortex and underlying basal ganglia. The gray matter consists of neuron somas and dendrites. The white matter is filled with the fatty insulating myelin sheaths that speed conduction of ionic firing spikes over long-distance axon collaterals connecting neurons in different parts of the cortex and underlying evolutionarily older brain regions, the putamen of the basal ganglia, caudate nucleus, and basal forebrain nuclei. Figure 1.7B Anterior-posterior location of coronal section shown in 17A. This image is a transparent view of a human left cerebral hemisphere showing the approximate location of the section relative to the interior deep gray matter. The deep older brain regions include the large basal ganglion, long arching caudate nucleus, and small almond tip (the amygdala) shown in purple and the more deeply underlying (but too far posterior to appear in section A) thalamus shown in green. The images for 1.7A and 1.7B are adapted from details of Appendix Figure A14 of Neuroscience - 4ed by Purves, et al. 8
Figure 1.8 The major components of the soma of a typical neuron plus its proximal axon and dendrites amid a background of dendrites, somas, and axons of other neurons. The regions labeled B to G are each shown in electron-micrographic detail in the text. B shows an (output) axon; C, terminal boutons at the ends of two axons that form side-by-side synapses onto one dendrite; D, cross sections through myelinated axon(s), showing the spiral of oligodendrocytic glial membranes insulating each axon; E, several branching segments of (input) dendrites for a nearby neuron; F, the soma (cell body) of another neuron with its large central nucleus full of DNA; and G, a myelinated axon showing a gap (node of Ranvier) between two insulating glial membranes where active exchanges of extracellular Na + and intracellular K + ions amplify firing spikes as they rapidly propagate along lengthy axons. Image adapted from a detail of Figure 1.3 of Neuroscience - 4ed by Purves, et al. 9
Figure 1.9 A typical chemical synapse connecting an axonal presynaptic neuron to a dendritic postsynaptic neuron. The synaptic cleft is 20 to 50 nanometers wide, from the axon terminal bouton lined internally with bubbles (vesicles) of neurotransmitter molecules to the dendritic postsynaptic membrane packed with transmitter receptors. When the flood of internal Na + ions creating the positive cross-membrane voltage at the leading edge of a axonal firing spike enters the bouton, the positive voltage opens channels in the cell membrane that admit normally rare Ca 2+ ions into the bouton ending. The Ca 2+ ions cause the vesicles full of transmitter molecules to merge with the presynaptic bouton membrane and release transmitter molecules into the synaptic cleft. When a receptor in the postsynaptic membrane binds to the right transmitter molecule, the receptor opens and admits charged atoms, usually Na + ions, into the dendritic fluid behind the synapse. Ions from hundreds or thousands of active dendritic synapses must reach the voltage-sensitive axon hillock at the base of the dendritic soma to cause it to initiate a firing spike (an activation potential) on its own output axon. Dendrites often have their postsynaptic membranes on raised spines that extend into the dense tangle of somas plus their dendritic processes and local axon collaterals within the grey matter of brain tissues. Image adapted from Figure 5.3 of Neuroscience - 4ed by Purves, et al. End of Lecture 1 Notes. 10