Microcircuit Recording and Imaging Ed Boyden Synthetic Neurobiology Laboratory Massachusetts Institute of Technology Invited lecture at Stanford in CS379C on April 10, 2013
Today’s outline PRINCIPLES, USES, UNKNOWNS Electrodes Electrode arrays Dyes Indicators Microscopes Endoscopes
Electrodes • A wire can do the trick • Precise impedance, material, tip angle, shank – Want an impedance of 100-5000 kiloohms length, taper, etc. can all • Polyimide, teflon, or impact your recording parylene coated – Ad hoc adjustment of these tungsten, steel, or parameters platinum-iridium
Tetrodes
Motorized microdrives • An individual motor on each microdrive (Yamamoto and Wilson, 2008)
Early electrode array: theme: simple, reliable engineering tricks (Campbell et al., 1991) • Thermomigration of Al • Take a dicing saw and cut through n-type silicon in between each of these p+ trails – Takes a few minutes; heat opens up migration slots – Diamond, 250-micron-wide blade, 30krpm – PNP junctions isolate probes from each other – Cut one way, then the other – Can use wire EDM to do this (Ian Hunter’s lab at MIT) – Modern probes are physically separated by glass (silicon dioxide) layers
Making the tips • Vigorously stirred 3- – Followed by a shallow 3- minute etch (simply, do not minute etch with 5% HF stir) and 95% HNO 3
Coating the tips • These two steps remove 75% of the volume of each needle, thus displacing <2% of nerve tissue upon insertion • Deposit gold on the tips – Stick array through thin foil, then sputter gold on the exposed surface – Platinum also used
Inserting it into the brain • Pneumatic inserter (Rousche and Normann, 1992)
• Scale bar = 1 mm
Modern such array • Silicon oxide layers for even better insulation • Next steps: integrating electronics in • Lesson: neurons are pretty macroscopic as far as MEMS goes
That hasn’t stopped some people from going further • Create silicon nanowires, then align them using microfluidics methods (Patolsky et al., 2006) – By creating p- and n-doped versions, directly make FET amplifiers on the neuron
How do you get nanowires to line up on pads? Do it the other way around • Langmuir-Blodgett • Then etch away silicon aligning: compress a wires outside each domain monolayer of nanowires of interest floating on a trough • Since silicon, can then add other devices
Yield and drift • Hard arrays - most studies indicate that the neurons recorded drift – Yield: 25-60% of electrodes on a given implantation will have useful recordings • Often get multiple neurons per trode, but must separate in software – No way to move these electrodes dynamically yet – Drift over several weeks • Wire arrays – “a few cells” could be recorded over 3 years in a monkey (sporadic result; not replicated) • Insert stiff electrodes that relax (exciting area!!) • Tetrodes – can drift somewhat over many weeks – Can be high yield – many neurons per tetrode can be separated thanks to combinations; can adjust depth of each to optimize positions – Flexible – move somewhat with brain?
Modern ideas • Make the electrode flexible, to match the impedance of tissue – Carbon fiber, polyimide • Make the electrode small, comparable to the size of neurons – “Microthread” • Coat the electrode with anti-biofouling compounds – PEGMA
Spike sorting • Many electrodes will get more than one neuron – Can only analyze spikes separated in time • A few attempts to overcome this problem, but none universally accepted – Clustering in spike-property space allows you to separate the spikes • Tetrodes give you many spike properties, and measured from multiple positions – Of course, spikes that always occur close in time will be ignored/ deleted since they will never be separable, so synchrony or other properties will be eliminated by clustering analysis – Tons of heuristics, free software, MATLAB scripts, etc.
Make the electrode fit the purpose • Example: lots of interest in recording large sets of neurons in a peripheral nerve • Cuff electrodes – get a few (3-10) recording sites – Crude for recording; crude for stimulation too – Most nerves are mixed – both motor and sensory • Sieve electrodes – Cut nerve, let it regenerate – Lots of problems
One partial solution: change the neural code • Reroute arm nerves to muscles in chest to broadcast coherent signals
Patch clamping • Glass electrode, place near cell, suck membrane close, and then ‘break in’ using suction – Get access to the inside – Negative • Wash out cell’s contents • Short recording times (minutes) à cell death • Low yield • Irreversible (can’t tune back and forth) • Works best for surfaces structures – Positives • Great isolation • Fine resolution à subthreshold activity • Voltage control • Dialyze in filling dyes, indicators, blocking drugs, or other chemicals • Suck out mRNA/other intracellular information
Patch clamping in freely moving rats • Must anchor after achieving whole cell recording (Lee et al., 2006)
A robot that can automatically patch clamp neurons in living brain Kodandaramaiah et al. (2012) Nature Methods 9:585–587. Commercialized by Neuromatic Devices, Inc. (ESB has no financial affiliation)
In vivo robotics: converting an art form to software Kodandaramaiah et al. (2012) Nature Methods 9:585–587.
Derived an algorithm: high-performance recording, with high yields Kodandaramaiah et al. (2012) Nature Methods 9:585–587.
Derived for the cortex, the algorithm works in the hippocampus as well Kodandaramaiah et al. (2012) Nature Methods 9:585–587.
Integrative analysis of cell types of the brain: molecule to morphology to physiology + Ragan et al., Nature Methods 2012 + gene expression Suhasa Kodandaramaiah, Ian Wickersham, Craig Forest, Hongkui Zeng and Allen Institute for Brain Science
Nanowires inside cells? (Duan et al., 2012)
Optical principles • Already covered absorption – ‘Intrinsic imaging’ – chromophores like blood that change absorption functionally • Reflection – Indirect measure of • Lots of others: coherent absorption – good for in vivo anti-stoke raman • Fluorescence scattering, second – Increase the parameters for harmonic generation, more flexibility – excitation fluorescence lifetime vs emission imaging, optical coherence tomography, …
Dyes: Calcium A = Ca+2 saturated, B = Ca+2 • Calcium is 50 nM inside a free neuron, and 2 mM outside • Big changes! a neuron • Thus, a natural tag for neural activity • Fura-2 (1985) – Four carboxylate groups – a motif that binds Ca+2 • Kd ~ 236 nM – Excitation fluorescence shifts when Ca+2 binds
Calcium dye usage in vivo • Pressure-eject Oregon Green BAPTA-1 (Stosiek et al., 2003)
Problems that must already be all too obvious • Big signals • Simulate spike ßà Ca +2 in order to invert better – 20-50% changes in fluorescence – can see by (Vogelstein et al., 2008) eye • Calcium is slow – Slow to enter, slow to leave – Saturates dyes rapidly, because of vast dynamic range of calcium combined with slow kinetics – Even with linear filters, signal processing, or deconvolution, can’t resolve individual spikes at rates much greater than 1 Hz
Voltage dyes – Fluorescence: Di-4- • Translocate across the ANEPPS membrane, or sit in membrane and change conformation when voltage changes – Fluorescence: RH1691 • Smaller changes than Ca +2 – < 1% – Absorption: RH155
Two-photon voltage imaging? • Hard to resolve cells – • Need a fast scope (Vucinic need a good dye (Kuhn et and Sejnowski, 2007) al, 2008)
Genetically-encoded sensors • Find a sensor of x, and then attach it to a fluorophore in some way • FRET methods – Donor and acceptor molecules must be in close proximity (typically 10-100 Å, which is 1-10 nm. For comparison the diameter of a DNA double helix is 2.3 nm, an F-actin filament ~6 nm, an intermediate filament ~10 nm, and a microtubule 25 nm – Absorption spectrum of the acceptor must overlap fluorescence emission spectrum of the donor. – Donor and acceptor transition dipole orientations must be approximately parallel (for optimal energy transfer). – Can be slow • Protein-unfolding/perturbing methods – Attach GFP (or other molecule) to a sensor – Circularly permute GFP and attach a sensor inside GFP – Can be slow • Few of these have been crafted from scratch – E.g., a stark shift genetic sensor - never been done yet
Genetically-encoded sensors • Cameleon – one of the earliest proposed sensors (Miyawaki et al., 1997) – blue or cyan variant of GFP (FRET donor), calmodulin (CaM), a glycylglycine linker, the CaM-binding domain of myosin light chain kinase (M13), and a green or yellow version of GFP (FRET acceptor) – Used to be lots of background binding (neurons express a lot of stuff), slow, small changes; gotten better
Genetically-encoded sensors • Circularly permuted GFP sensors (Nagai et al., 2001) – GFP is a barrel – can start and stop the strands at any point
Recommend
More recommend