Timing in Biological Systems Lou Scheffer Howard Hughes Medical - - PowerPoint PPT Presentation

Timing in Biological Systems

Lou Scheffer Howard Hughes Medical Institute

http://www.tiempo-secure.com/technology/asynchronous-design-technology/

Nervous systems and electronics

A lot in common

Basic operation is electrical Multiple inputs + nonlinearity

Both have circuits where timing is critical Tools for studying timing

Experimental Theoretical

Biology has cool tricks

Grows (no $6B factory) Resilient Learns

Credits: Lynn Riddiford, Wojciech Maly, Rex Kerr

Even politicians recognize this is a good problem to work on!

Trying to understand the brain

Hot topic – many methods are being used

Structural – reverse engineering Genetic Behavioral Electrophysiology Imaging Lineage Combinations of these techniques

Philosophy, Washinton and Lee

History – Basic Science

Caton saw electrical activity in animal brains, 1875 Berger produced first EEGs of humans in 1924 Hodgkin-Huxley Nobel-winning model in 1953

Established basic model of how a neuron works

From Nobelprize.org

Neuron operation

Internal operation

Chemical Electrical

A voltage controlled current source Timing is critical to the main neuron

• peration – “action

potential” or spike

Action potential result of time constants

• http://hyperphysics.phy-astr.gsu.edu/hbase/biology/actpot.html

Also ‘gap’ junctions

Direct connection between cell interiors

Like a resistor (diode) Same sign Gain < 1 Very fast Not used much in mammalian brains

Used in insect brains Used in the retina and visual system of mammals

Wikipedia Gap cell junction en.svg 1 Nov 2009

Example circuits where timing is critical

Motion detection Sound localization Energy savings & synchronization Learning

http://www.theguardian.com/us-news/2014/oct/29/paraglider-dell-schanze- charged-with-harassing-an-owl-in-utah

t _/ + + t _/ + + Σ +

• +

+

Σ

• L1

Mi1 T4 L1

+ +

Σ

• Mi1

T4 Tm3

Σ

+ -

Motion detection depends on delays

Credit: Dmitri Chklovskii, Janelia, HHMI

Hearing and localization depend on precise delays

Carr, C. E., and M. Konishi. "A circuit for detection of interaural time differences in the brain stem of the barn owl." The Journal of Neuroscience10.10 (1990): 3227-3246.

Nature builds balanced trees

Carr, C. E., and M. Konishi. "A circuit for detection of interaural time differences in the brain stem of the barn owl." The Journal of Neuroscience10.10 (1990): 3227-3246.

• Different neurons respond differently

to the same inputs

Energy savings or synchronization

Data from locust (not in all insects)

Perez-Orive, Javier, et al. "Oscillations and sparsening of odor representations in the mushroom body." Science 297.5580 (2002): 359-365.

! "# $

%&'())&**$+,-,## .-

• Spike Timing Dependent Plasticity

Spike Timing Dependent Plasticity

IBM trying to implement this with memristors

Multi-purpose Neuro-architecture with Memristors Idongesit Ebong, Durgesh Deshpande, Yalcin Yilmaz, and Pinaki Mazumder

Adjust M weights in one cycle

Repeater insertion in biology

We know there must be repeaters

RC of neurons gives 1 ms for a 1 cm length If strictly electrical, goes like L^2

10 sec for a 1 meter nerve

Not observed, so potential is regenerated Propagates like a wavefront (constant speed) A main success of Hodges/Huxley model

As in EE, two regimes

Small insects, neurons are isopotential and propagation delays are << gates Large animals (mammals) transport delays dominate

Long nerves are different

Bigger diameter helps, but only as sqrt(d) Myelinated nerves reduce capacitance

Myelin sheath Schwann cell Node of Ranvier Axon terminal Dendrite Soma Nucleus

Wikipedia: Neuron Hand-tuned.svg 1 Nov 2009

How does this help?

Myelin is a good insulator

Isolates sections of line Less C and less leakage Needs to reach threshold by the next node Gives factor of a few better speed Very similar to repeater insertion in VLSI

Myelin is white, which is why the brain is grey matter (no myelin) interconnected by white matter

Statistical timing in biology

Every step in bio timing is statistical

Small number (usually 1?) of vesicles released during an action potential Each is small (~5000 molecules) so a ~70 molecule std deviation This triggers a few (1-10? Ion channels) That relax statistically…..

Combines effects of statistical timing and noise

Ion channels are fundamentally statistical

Open channel always the same size Controlled by a Markov-like process

THE PHYSICAL BASIS OF ION CHANNEL KINETICS: THE IMPORTANCE OF DYNAMICS, Liebovitch and Krekora http://www.ccs.fau.edu/~liebovitch/dyn.html

Biological systems stable to perturbations in timing

Cold blooded animals work over a range of 10- 40 degrees C.

Reaction rates vary considerably over this range.

But animal behavior largely unchanged Not well understood how this works

Tools for studying timing

In one way, similar to studying timing in ICs

Most methods cannot measure signals without affecting them

One big problem – no test chips Some readout techniques too slow Some have promise of sufficient speed, but not yet Fast-enough methods have other limitations.

Gold standard: single cell recording

Excellent resolution Can see sub-threshold Both input and output Drawbacks

Invasive Hard to connect to desired cell Short life (1/2 hour) Not easy to parallelize Tough in behaving animals

Electrophysiology (continued)

Animals such as mice and rats can wear headgear Now working for flies using ‘virtual reality’

Michael Reiser, janelia, HHMI

Electrical readout by electrode array

Electrodes go near cells, not in them Readout by capacitive coupling Each electrode reads many cells (needs spike sorting) Timing is good, but not cell ID Spikes only (no subthreshold)

BYU

• !
• Brown.edu

Spike sorting

All nearby neurons couple to the same probe – data is ambiguous Spikes of different sizes/speeds probably come from different neurons Uses classification algorithms; called ‘spike sorting’

Can potentially record from hundreds of neurons But we don’t see this many - The ‘dark matter’ problem….

Wikipedia: Spike clusters.png 1 Nov 2009

Extracellular recording (many cells)

IMTEK probes, from CMOS technology

www.pb.izm.fhg.de

500 µm 40 µm

Source: K. Seidl, et al., JMEMS 2011, vol. 20, pp. 1439 -1448

Consortium for active probes

Replace head with LPFs, A/Ds, muxes Readout of hundreds of channels

Imaging

Use genetic techniques to add indicators that glow when cells are active Can look at many cells in parallel

Can be too many cells – need genetic subsets

Disadvantages

Limited to surface of brain, except for a few transparent animals, such as zebrafish larva Voltage resolution is not great Time constants too long

But better indicators are continually developed

Zebrafish brain imaging

Misha Ahrens, Philipp Keller, Janelia, HHMI

Perturbing biological timing

Affect timing of

• perating neurons

Turn specific neurons on and off

Equivalent of stuck-at faults

Current and voltage injection work well, but are difficult and tedious

Channel opens at 26-27C and causes neuron to fire.

• Example method that is too slow

Temporal Dynamics of Neuronal Activation by Channelrhodopsin-2 and TRPA1 Determine Behavioral Output in Drosophila Larvae Stefan R. Pulver, Stanislav L. Pashkovski, Nicholas J. Hornstein, Paul A. Garrity, Leslie C. Griffith Journal of Neurophysiology Published 1 June 2009 Vol. 101 no. 6, 3075- 3088 DOI: 10.1152/jn.00071.2009

Optogenetic better for timing

Add light gated ion channels Drives neurons more naturally Can force nerves on or off Independent parallelism is hard Only good for the outer few layers

Simulation tools (only)

Brette, Romain, et al. "Simulation of networks of spiking neurons: a review of tools and strategies." Journal of computational neuroscience 23.3 (2007): 349-398.

Simulation problem harder than in EE

Connectivity data (from EM, retro-viruses) does not tell channel type Images do not directly tell strength Some connection forms (gap junctions) not visible in traditional EM Hundreds of types of channels

Even sign of response is

• ften unclear

Dynamics of all versions not known

Many fundamental questions about timing remain

Neural coding - for a train of spikes, what is important?

Average rate? First spike time? Inter-spike distance? Correlation? All the above?

841 16 676 169 25 81 9 1 1 1 20 20 20 800 40 40 150 10

R1-6 L1 C2 L5 C3 Lai L3 R7-8 L2 L4 Photo- receptors Medulla Lamina

Biological Neural Nets

Feedback

Lots of local feedback Feedback with timing has no good models Calling all theorists!

Status of timing understanding

Hardwired circuits (insect vision)

Same from animal to animal Making good progress

PLA like systems (insect olfactory)

Every animal different, but in standard ways Harder, but technical advances seem sufficient

Fully programmable and time varying (mammalian cortex)

New ideas are needed

Gcat.davidson.edu

npu.edu

Conclusions

Timing is critical to understanding brain

• peration

No clear separation of concerns, unlike timing in human designed systems. Better theoretical and experimental tools are needed Fascinating and hugely important problem.