This is real team science: 5 labs expert in recordings from mouse and monkey cerebellum joined forces to make the project happen. Thanks to a great team, led by @1DanaCohen1, @Steve_Lisberger, @dkneurolab , @theHullLab, @Neuralcomplab, and Javier Medina.
1
5
967
Second, the recruitment of this non-coding pool by optogenetic stimulation can enhance perception. This means that this latent pool of neurons (the “dark matter”) can actually be useful for sensory processing!
1
4
609
We therefore provide two key new insights into the mechanism and significance of sparse coding. First, the sparse code is enforced by selective targeting of powerful inhibition, which predominantly suppresses non-coding neurons.
1
1
10
854
The all-optical experiments also revealed that these non-coding (i.e. normally silent) neurons are selectively targeted by inhibition recruited during sensory stimulation – in other words, they are preferentially suppressed by sensory stimulation.
1
6
714
All-optical interrogation revealed that we can produce a behavioral bias which depends on the number of neurons activated. Moreover, what is fascinating is that the behavioral effect is dominated by activation of non-coding neurons.
1
1
4
702
Next, we switched to all-optical interrogation to be able to activate small ensembles of identified neurons using two-photon optogenetics, and measure the impact both on local network activity and on behavior.
1
2
758
We then performed two-photon imaging during the task and confirmed that coding of whisker input is sparse – in other words, most neurons are silent! This sparse code aligns with work from many groups (e.g. Olshausen & Field, Brecht, Helmchen, Svoboda, Barth & Poulet).
1
3
833
We used one-photon optogenetics to show that barrel cortex is required for this task and that the processing window is very brief – less than 100 ms!
1
2
921
We developed a challenging new behavioral task for mice which requires that they integrate bilateral whisker information. This produces sigmoidal psychometric curves, similar to the classical curves in visual psychophysics experiments (Shadlen, Newsome, Movshon, et al.).
1
1
1,206
Most neurons in the cortex remain silent, even during sensory stimulation and behavior. What are these silent neurons good for? This is known as the “dark matter problem” of the brain. We address this problem in our @NeuroCellPress study led by @omgauld
doi.org/10.1016/j.neuron.202…
3
58
187
25,994
These results should be of interest to anyone interested in linking cortical activity to perception – and particularly for those designing BMIs, which will be most effective when stimulating the right neurons, at the right time, in the right task conditions.
6
1,473
The effect of photostimulation during the engaged state depended on the sensory conditions of the task: at low contrast (i.e. a difficult stimulus to detect), photostimulation enhanced detection. At high contrast (i.e. an easy stimulus) photostimulation impaired performance.
1
1
326
The impact of optogenetic activation of these neurons on behavior depended on two factors. First, the mouse had to be engaged in the task for photostimulation to be effective.
1
1
378
We trained mice in a visual detection task and then selectively targeted the most stimulus-responsive neurons in visual cortex (i.e. the neurons most likely to be processing the stimulus) for two-photon photostimulation.
1
1
7
1,089
When does the cortex matter for perception? Our new study addresses this question using all-optical interrogation in visual cortex.
doi.org/10.1038/s41467-024-4…
1
30
77
9,780
We harnessed the latest developments in high-density silicon probe recordings, optogenetic tagging, and machine learning, using the cerebellum as a testbed to reliably identify cell types from their electrophysiological signatures.
1
21
1,890
Have you ever looked at traces from an extracellular recording and wondered what kind of neurons you are recording from? Five labs across three continents working with two species joined forces to solve this decades-old problem in systems neuroscience. 🧵
2
75
212
31,134
We developed a new texture discrimination task and combined two-photon imaging and two-photon targeted optogenetics (“all-optical interrogation”) to show that neurons encoding decision signals can be detected in S1 – and that their activity can drive behavior.
1
21
Where do decisions get made? Decisions about sensory information are thought to arise high in the cortical processing hierarchy. Our @NatureNeuro study challenges this, showing that decision signals consequential for behaviour are found at the earliest cortical processing stages.
1
75
300
Out today – all-optical interrogation lore from ~10 years of work (and thousands of postdoc and student hours) compressed into a single @NatureProtocols article!
rdcu.be/cMeO2
5
64
244
In novices, the BG may help the Cb construct models for rewarded actions, predicting successes and failures. In experts, the Cb then sends confident action predictions to the BG, aiding in reinforcement. If task conditions change, the BG may again teach the Cb to update its model
1
4
11
How are cerebellar reward signals related to those in the basal ganglia? The dense reciprocal connections between these structures have been known about for decades, and new studies keep finding connections between them that suggest they mutually instruct each other.
1
8
Indeed, pretty much anywhere you look in cerebellar circuits, you can find signals related to reward. Moreover, these signals often obey similar rules to reward signals mediated by the classical reward system - the dopamine neurons in the VTA and substantia nigra.
1
6
5 years ago, we were imaging climbing fiber-driven signals in Purkinje cells in our goal-directed behavioral task and noticed something very peculiar: every time our mouse got a reward for correctly executed trials, many Purkinje cells lit up (and were organized into microzones)
1
8
We are excited to share our Perspective in @NeuroCellPress discussing why cerebellar reward signals may be important for learning. Here’s a personal history explaining what led to this Perspective, and why we think these signals are so interesting.
doi.org/10.1016/j.neuron.202…
2
82
338
One week until the deadline for the @UCL @WellcomeTrust PhD Program in @OpticalBiology. We're looking for enthusiastic and talented students interested in developing and applying new optical tools to big questions in neuroscience and cell biology. Details: opticalbiology.org
4
12
When we embed temporal signals in the input patterns, the model learns to capitalize on those too. In fact, we find that rate and temporal processing strategies can even be synergistically combined.
10/12
1
13
Surprisingly, whereas in basal dendrites the model learns to exploit NMDA-dependent supralinear integration, in apical dendrites, the model learns to exploit the sublinear integration that arises from local voltage saturation of active events.
9/12
1
16
With randomly distributed, rate-coded input, the weights evolve to engage dendritic nonlinearities by strengthening nearby coactive synapses. With a synaptic learning rule that is sensitive to dendritic properties, structured connectivity is not necessary to solve the task.
8/12
1
14
We show that the model can solve a canonical nonlinear feature-binding problem (see also bit.ly/314HWhj & bit.ly/3CqXSZ1), which requires the neuron to spike in response to specific conjunctions of input stimuli.
7/12
1
15
Working from a foundation of cable theory, we developed a learning rule (inspired by go.nature.com/3EnmHFZ) that tunes synaptic weights to control somatic spiking.
5/12
1
1
12
We first built a biophysical model of a layer 2/3 pyramidal neuron that embodies both the computational potential and physical constraints imposed by dendrites (thanks to @AllenInstitute for the beautiful reconstructions!).
4/12
1
1
18
Does the brain use a more flexible alternative? Taking up the single-neuron-as-a-neural-network analogy (bit.ly/3jL7d6P), we wanted to know whether dendritic computations could instead be learned.
3/12
1
14
Our paper on learning to compute with dendrites is now out in @NeuroCellPress. We show that the power of dendritic processing can be efficiently harnessed through synaptic plasticity, allowing single neurons to perform sophisticated computations.
1/12
(bit.ly/3mnTIeY)
7
125
421
This is not just a correlation: blocking Na⁺ or NMDA receptor channels in individual dendritic branches shows that local dendritic Na⁺ spikes and especially NMDA spikes causally contribute to the generation of tuned action potential output in vivo.
9/12
1
7
AP-triggered spatiotemporal averages show that somatic APs tend to follow a surge in dendritic spikes, involving on average ~10 dendritic Na⁺ spikes and ~5 NMDA spikes.
8/12
1
5
So how many dendritic spikes in turn are required to trigger somatic output APs? The spatiotemporal plot shows that somatic APs are typically preceded by many local dendritic Na⁺ and NMDA spikes in different dendritic branches.
7/12
1
8
We also developed a new kind of plot – a spatiotemporal synopsis of membrane potential across all dendrites – to show how removing just a single active synapse (asterisk at 27.6 ms) prevents a dendritic Na⁺ spike from occurring.
6/12
1
1
7
Surprisingly, in some cases only a single local synapse can be enough to trigger a dendritic Na⁺ spike or an NMDA spike (acting on background depolarization created by active synapses elsewhere in the neuron).
5/12
1
7
We developed detectors for dendritic Na⁺ and NMDA spikes, and used spike-triggered averaging to reveal that remarkably few synaptic inputs are required to trigger these spikes.
4/12
1
7
Spot quiz: can you tell which traces are from the model, and which traces are from the experiment? (Hint: see Fig. 1C) Amazing in vivo dendritic patch-clamp recordings from @Labrigger, see nature.com/articles/nature12…
3/12
2
6
We simulated synaptic integration in vivo using a new active model of a neocortical layer 2/3 pyramidal neuron, combined with a model of the synaptic input it receives during sensory stimulation, both constrained by the latest experimental data.
2/12
1
1
7
Dendrites are “active” – they generate dendritic spikes, which are used for computations that influence behavior. How are dendritic spikes triggered by synaptic input patterns in vivo, and how do they influence neuronal output? Check out our new paper pnas.org/content/118/30/e201…
1/12
3
68
274
Thanks also to our partner @zeiss_micro and to the funders for their generous and patient support: @wellcometrust Gatsby Charitable Foundation @ERC_Research @BBSRC
8/8
9
The work was a fantastic team effort from @NeuralCompLab @kstaras @janfunkey and Claudia Racca with the heavy-lifting experiments and analysis carried out by @Sim0nAnna, @ArndRoth, @SheridanArlo and Mehmet Fisek with help from @vincenzomarra
7/8
1
11
This approach allows us to gain new insights into relationships between ultrastructure and function of synapses during behaviour. We also show how this method can be used to estimate synaptic weights of individual synapses within a wiring diagram.
6/8
1
10
This provides a new strategy for identifying active synapses in vivo at ultrastructural resolution. Our method complements existing connectomics approaches which can provide the wiring diagram of a circuit, but not information about which synapses are active.
5/8
1
8
We used convolutional neural networks to automatically segment neurons, active zones, mitochondria and vesicles. We could also automatically classify vesicles as photoconverted (active) or non-photoconverted (resting), and thus quantify presynaptic activity at each synapse.
4/8
1
6
The high resolution of FIBSEM means that each ~50 nm vesicle can be visualized in multiple consecutive sections, allowing us to map non-photoconverted and photoconverted vesicles with high precision throughout individual synaptic boutons in a large volume of neuropil.
3/8
1
9
We labelled activated synapses in vivo using a special fluorescent dye which is incorporated into recycled vesicles recruited by sensory stimulation. Photoconversion of this dye then generates an electron-dense product which can be visualized using electron microscopy.
2/8
2
7
Want to identify active synapses or estimate synaptic weights in your high-resolution connectome? Our latest preprint combines 2-photon targeted functional labelling in vivo with machine learning readout of focused ion beam scanning EM (FIBSEM) data. biorxiv.org/content/10.1101/…
1/8
5
54
170
This work was a great team effort spearheaded by Shinichiro Tsutsumi together with @OscarChadney, @TinLong_Yiu, @BaumlerNeuro, Lavinia Faraggiana and @maxime_beau with support from the @wellcometrust, @ERC_Research and a steady supply of pastries from Miel Bakery. 15/15
9
Finally, given that motor responses became increasingly accurate with learning, we asked whether complex spike synchrony developed in parallel with animal performance. Animals did, in fact, increase their level of complex spike synchrony across learning. 12/15
2
6
Next, we wondered whether there is a relationship between complex spike synchrony and sensory-driven motor action. Indeed, we observed that higher levels of CS synchrony were correlated with more accurate movements and increased probability of motor initiation. 11/15
1
3
We found that complex spike synchrony was topographically organized into parasagittal zones, the structure of which was highly conserved across animals: forming a consistent “complex spike map”. 10/15
1
1
7
Given the well-established role of complex spike synchrony in motor timing, we next used two-photon calcium imaging of Purkinje cell dendrites to measure complex spike signals during our task. 9/15
1
4
