Hello from San Diego and welcome to the Stanford Neuroblog! In this post, I'll be covering the first half of SfN Short Course #1: Genes, Photons, and Electrons. This workshop focused on novel techniques for probing and manipulating neural circuits and connecting structure and function. Overall, the talks were excellent and the speakers complemented each other nicely. The course nicely summarized recent progress in experimental technology. The content of the talks also conveyed a mix of excitement at how accessible previously unfathomable circuit level questions have become thanks to novel technologies and bewilderment at how complex the relationship between neural circuit structure, function, and behavior appears to be.
Below are essentially summaries of what was said and presented rather than my own commentary. I have tried to minimize factual errors or misinterpretation of the speaker's remarks, but these summaries are reconstructed from my notes during the talks and are surely imperfect. Please kindly notify me of any errors or oversights in the comments section. Enjoy and stay tuned for Part II!
We can roughly divide the history of scientific investigation into neural circuits into three phases. First came the romantic age, led by Ramon y Cajal, driven primarily by the Golgi stain and pure anatomical observation. Next came the classical age. Hodgkin and Huxley, Katz, Fatt, Eccles, Llinas, Rall, Bliss and others pioneered the study of neural function at the cellular and synaptic level, while Hubel and Wiesel, O'Keefe, Georgeopolous, Newsome, Shadlen opened the doors to modern day systems neuroscience. Despite great advances, what's largely missing from this "Classical Age"? Genetic identity of cell types, recording of activation patterns of all neurons and synapse relevant to behavior, complete descriptions of underlying connectivity patterns, and the ability to clearly demonstrate causal relationships.
Now, according to Hausser, we enter the "age of enlightenment," hinting at a pun on optogenetics with an image of a mouse sporting a head mounted fiber optic glowing blue. In this age, we will enjoy a new array of technologies that he's divided into three overarching categories, following the title of the short course.
Genes
- sequencing of entire genomes
- genetic model systems
- transgenic animals and viral approaches
- genetically encoded labels / probes
Photons
- 2 photon
- super-resolution
- optogenetics
Electrons
- Dendritic patch clamp recording
- in vivo patch clamp recording
- High density arrays and optrodes
So equipped with these dream tools, what would be the dream experiment? Hausser lists a few suggestions: reconstruct the connectivity of entire circuit, measure the activity in all neurons during behavior, sway decisions and recall memories by manipulating neurons, etc.
Semi-Automated Reconstruction of Neural Processes from Large Numbers of Fluorescence Images
Jeff Lichtman, Harvard
Jeff opens with a definition of the word connectomics from the OED circa 2015:
connectomics: noun plural but singular in construction
A branch of biotech concerned with applying techniques of computer-assisted image acquisition and analysis to structural mapping of sets of neural circuits or to the complete nervous system of selected organisms using high-speed methods with organizing the results in databases, and with applications of the data (as in neurology or fundamental neuroscience) - compare proteomics or genomics,
see also connectome
He then describes a related array of efforts and research directions that are commonly associated with connectomics:
- Human connectome, led by Olaf Sporns at Indiana University, to map axon projection pathways with DTI. Perhaps this should be referred to as a projectome?
- Testing Peters rule: expected number of connections proportional to product of their dendritic and axonal tree densities
- Investigation of neurogeometry and potential synaptic connectivity [Stepanyants and Chklovskii, TINS 2005]
- Blue brain project, led by Henry Markram, EPFL
- "Cajal 2.0", first pass connectome: including efforts by the Allen Brain institute, Partha Mitra Brain architecture project, fly optical project at JFRC
- Sparse labeling / reconstruction: micro-optical sectioning tomography [Li, Science 2010]. Reconstructing subsets of cells using automated/semi-automated analysis of fluorescence images. Brainbow.
- Dense reconstruction: Denk, Heidelberg, Seung. This involves dense reconstruction of neuropil structures from stacks of EM images. A great deal of automation and quality control required from tissue handling, image acquisition, image alignment, segmentation, reconstruction, verification, annotation, etc.
Lichtman's lab is primarily involved in these last two efforts: sparse labeling and dense reconstruction. One tool useful for sparse labeling is the array of Thy1-XFP mouse lines which provide expression of a particular flurophore (CFP, YFP, GFP, etc.) in an apparently random susbset of neurons in the brain. This "random" expression results presumably due to random insertion of the transgene into genome.
He demonstrates the power of this technique by showing a series of confocal images of the neuromuscular junction of a Thy1-YFP mouse. The NMJ is innervated by YFP and AChR expressing motoneurons, and it is clear from the images that each NMJ is innervated by only one axon, thought the nerve bundle possess many axon fibers.
In the spirit of the short course having an educational component, Jeff pauses to offer tips on taking a proper confocal image. When performed correctly, confocal offers enhanced contrast, optical sectioning, and a resolution improvement by sqrt(2) over the widefield diffraction limit. However, it is easy to saturate the fluorophores by turning the laser power too high, resulting in disproportionate out of focus signal reaching the detector. Additionally, he asserts it is important to image using the full dynamic range of the sensor (meaning few pixels lying at either end of the histogram range. This allows for lossless imaging and better reconstruction, and deliberate saturation can almost amount to scientific fraud by "throwing away" outlying pixels.
He then turns to the task of segmenting and tracing fluorescence labeled axons in confocal image stacks. The details of the algorithm are mentioned in Lu et al. 2009. Quickly, he demonstrates structural polymorphism present in left vs. right versions of same nerve bundle projecting to muscle. He also points out that individual axon paths demonstrate numerous suboptimalities, exhibiting wasted loops and back-tracking, as well as clear violations of Peter's rule.
Next up, Brainbow! Brainbow is a technique for achieving unique labeling of individual cells by combining random amounts of three fluorophores (mCherry, eYFP, Cerulean) in each cell, achieving the same effect as a TV screen combining RGB intensities to create a particular hue. The construct, which has the form thy1-lox-lox-mCherry-lox-eYFP-lox-Cerulean, uses the stochasticity of Cre splicing to achieve this random expression. Because each neuron has a unique and consistent color (defined by relative levels of red, green, and blue expression), this eliminates the need to trace axons/dendrites since there is a 1:1 correlation of color intensities at both ends of the neuron. However, if the labeling becomes too dense, the fibers in neuropil can become too thin or too weakly expressing for reconstruction.
Another side issue is how to visually display a connectome once you have obtained it. He presents a number of display formats, motivated by graph theory. The point is that there a number of possible choices, but it's clear that there are non-random features evident in the connectivity matrices observed even for small numbers of neurons reconstructed.
Lastly, he discusses the technology behind ATLUM (Automated Tape-Collecting Lathe Ultramicrotome) which automates the slicing and handling of thin brain slices embedded in plastic resin. He shows a video by Daniel Berger in Sebastian Seung's lab which opens from a photo of a silicon plate held by a lab member on which tissue has been mounted for EM. We then gradually zoom in to the point where we see individual vesicles in a presynaptic bouton.
Imaging Neural Activity in Worms, Flies, and Mice with Improved GCaMP Calcium Indicators
Loren Looger, HHMI Janelia Farm
Loren opens with the point that neural circuits underlie behavior, where a circuit is defined as a collection of neurons, their chemical identity in terms of neurotransmitters released on postysynaptic targets, their connectivity graph, the sign of their connectivity, as well as changes over time of these properties as a function of development and experience. He asserts that these properties of neural circuits are essential to making in progress in understanding circuit function and the structure/function relationship, noting that very little insight has been extracted from the complete connectome for C. elegans completed some decades back. He notes that 2 maybe 3 key points of understanding may have been derived, but mostly in the last few years.
He then borrows analogous terminology from forward and reverse genetics to describe the types of optogenetic research that are now possible. Forward optogenetics is observing neural activity optically during behavior, e.g. calcium imaging in head fixed mice on a floating spherical treadmill a la David Tank's lab. Reverse optogenetics is perturbing neural activity in order to determine causal influences of circuit elements on behavior, e.g. a mouse running in circles subsequent to ChR2 activation via head mounted fiber optic a la Karl Deisseroth's lab.
Circuits are ultimately the minimal level at which to study certain interesting behaviors, but molecules compose a circuit. Specifically, a researcher can utilize molecular probes in order to observe and quantify neural function. Loren's lab had previously solved the crystal structure of GCaMP2, a genetically encoded calcium sensor created by placing calmodulin, a calcium binding protein, inside GFP. His lab then gradually engineered an enhanced GCaMP3 by screening point mutations in GCaMP2. He briefly compares GCaMP to FRET based sensors, noting that GCaMP avoids photostability problems common with FRET because the fluorophore is not exposed and therefore not bleachable in the off-state. He briefly mentions the ongoing development of GCaMP5, which touts better SNR and faster off kinetics for enhancing the ability to distinguish single action potentials.
He then mentions an array of calcium indicators spliced to other flurophores, opening the door to spectrally separation in activity reporters, e.g. RCaMP via mRuby with a bimodal 2 photon excitation spectrum (750 and 1125 nm), CyCaMP, BCaMP, etc.
Another direction for enhancing these reporters is in achieving subcellular targeting specificity. For example, by restricting the calcium indicator within neuron nuclei, the overall signal has half the intensity and half the speed of regular GCaMP, but with the advantage of much lower background fluorescence from processes. This clean separation of glowing nuclei greatly facilitates segmentation of individual cellular signals, a direction his lab is pursuing in collaboration with Daniel Dombeck and David Tank at Princeton.
Bioluminescence as a Tool to Monitor Neural Activity in Freely Behaving Animals
Florian Engert, Harvard
Florian opens with an introduction of his favorite model organism, the larval zebrafish. Specifically he employs the nacre mutant, which is perfectly translucent except for eye pigment. This translucency facilitates for a range of imaging and potentially optogenetic techniques. He employs transgenic fish which express GCaMP3 in every neuron, allowing imaging of calcium transients of the entire system. He asks the crowd for suggestions on what to call this kind of dataset: an activitome?
He states that our ultimate goal is to characterize how the brain produces behavior. The experimental desire for behaving animals clashes with the stability demands of high resolution optical imaging. The trend towards head-fixed imaging with an animal performing a task in virtual reality, as David Tank has demonstrated in mice and others previously in flies, is probably the best solution, if you can reproduce the behavior sufficiently well while tethered.
Another approach is to utilize bioluminescence. Equipping bioluminescent apoaequorin protein with GFP across a calcium binding linker creates a FRET pairing interaction, effectively creating an illumination-free calcium reporter. The approach requires the cofactor CLNZ, which is itself fluorsecent, facilitating quantification of loading efficiency. This sacrifices all spatial resolution of the imaging, as photons are usually detected and counted by a wide angle of entry photomultiplier tube. The technique's advantage lies in restoring the fish's full motility. Resolution has to be reintroduced using the specificity of the expression of the apoaequorin, presumably genetically.
He demonstrates one example of this technique by targeting the apoaequorin to the fish hypocretin system, a network consisting of 8 neurons on each side. Two types of neuroluminescence events are clearly distinguished, though both are correlated with bursts of locomotion. "Large" luminescence transients are related to long latency, short travel swims, whereas "small" events are related to shorter latency, longer travel swims.
Typically bioluminscence imaging requires complete darkness, which conflicts with the need for visible wavelength visual stimulation to elicit and modulate fish swimming behavior. Spectral separation wasn't sufficient to recover good signal, so the lab developed an imaging approach relying on temporal separation of the visual stimulus and bioluminescent photon detection. The essence is to strobe visual stimuli at a frequency well beyond flicker fusion (e.g. 1 kHz), then gate the PMT closed during times when visual light is on.
