Doron Lancet's talk on June 9th:


Imaging glomerular responses, PNAS 1982
Signaling in sensory neurons, Nature 1985
All senses, Nature 1991
Combinatorial receptor repertoire, PNAS 1993
Human olfactory genome, Gen Research 2001
Receptor repertoire test, J Theor Biol, 2002
Different noses, Nat Genetics 2003
Personal repertoire, BMC Genomics 2012

Aravi Samuel's talk on June 10th:


A preprint on panneuronal imaging in C. elegans is available here:



Understanding thermotaxis in Drosophila larvae required finding the relevant thermosensory neurons. Interestingly these, are embedded in the olfactory system.

Klein et al

Optogenetics has proven useful for defining the sensorimotor transformation during both chemotaxis and thermotaxis (see above) Klein papers. Linear-nonlinear analysis allows defining filters with a remarkable degree of predictive power over larval locomotion. These results were published recently by Marc Gershow's lab and our own.

Marc's paper

Our paper

Given the power of white noise analysis and LN modeling, our inspiration is to find the computational basis of PN activity patterns, going from ORNs to LNs to PNs. A useful handhold in this problem is defining a set of private odors for the ORN periphery in the larva.

Mathew et al

Much of the theoretical framework for what needs to be done in the next few years is well developed and can be found in this inspiring paper by Markus Meister, Gilles Laurent, and colleagues:

Geffen et al.


Matthieu Louis' talk on June 10th:


My presentation intended to summarize our understanding of the link between the computation achieved by single olfactory sensory neurons (OSNs0 and larval chemotaxis. In view of the title of the workshop, I will refer to this process as a functional deconstruction. During my presentation, I covered 4 topics. For each topic, I outlined the main conclusions and references I touched on during my talk (apologies for any omissions!).

Below are two excellent reviews about larval olfaction:



1) Deconstruction of the orientation mechanism underlying larval chemotaxis:
1.1) Using an infrared spectroscopy method, we are able to reconstruct the odorant landscape in which larval chemotaxis can be studied. This method allowed us to establish a correlation between the stimulus input and the locomotor output during orientation in an odor gradient.
Reference:
(work done in the Vosshall lab)

1.2) Like for bacterial chemotaxis, the trajectory of a larva can be approximated as a sequence of runs and directed turns.

1.3) A turn-triggered average analysis of the sensory experience shows that positive gradients tend to suppress turning whereas negative gradients tend to facilitate turning.
References:
(work done by the Samuel lab)
(work done in my lab)

1.4) Turn are directed through lateral head sweeps or casts. This active-sampling behavior seems to be analogous to the trail tracking behavior observed in rodents.
References:
(review)
(work done by the Bhalla lab)
Fun paper on odor-search behavior in moles:

In addition to the run-cast-turn orientation mechanism, larvae appear to be able to curl their runs toward the direction of the odor gradient - a process called weathervaning. Weathervaning was initially discovered in C. elegans.
References:
(work done in my lab)
(work done by Yoshida lab)

2) Deconstruction of the peripheral representation of olfactory stimuli:
2.1) The larval olfactory system comprises 21 olfactory sensory neurons (OSN). Each OSN expresses a different type of odorant receptor (OR). As a rule, one OSN expresses one type of OR in addition to the co-receptor Orco. The repertoire of odorant receptor found at the larval stage was characterized in the following references:
(work done in the Vosshall lab)
(work done by the Carlson lab)

2.2) We developed a method to restrict the function of the larval olfactory system to a single pair of OSNs - one on the left and one on the right side of the head. We discovered that the information mediated by a single OSN is sufficient to direct larval chemotaxis. This suggests that changes in the odor concentration can be encoded by a single type of OSNs.
References:
Fishilevich et al, 2005 (see above, work done in the Vosshall lab)
Louis et al, 2008 (see above, work done in the Vosshall lab)

2.3) Using a stochastic strategy to restore the function of a single OSN restricted to one side of the head. We found that unilateral function does not impair chemotaxis, even though it reduces orientation performances. This result indicates that larval chemotaxis is not purely based on left-right comparisons (or tropotaxis). As a consequence, change in stimulus intensity must be represented in the firing pattern of a single OSN.
Reference:
Louis et al, 2008 (see above, work done in the Vosshall lab)
For similar studies on stereo-olfaction in adult flies, see:
(pioneering work by Axel Borst, technical tour de force)
(work done by the Wilson lab, use of optogenetics)


3) Deconstruction of the sensory encoding of naturalistic stimuli:
3.1) We developed a method to record in vivo from single functional OSN in the larva. We also used optogenetics to perform spike-sorting.
Reference:
Schulze et al, 2015 (work from my lab), see the following figure.

3.2) Upon expression of channelrhodopsin in the Or42a-expression OSN, we showed that the overall response dynamics of the OSN is similar when it is stimulated by light and by odor.
Reference:
Schulze et al, 2015 (work from my lab), see the following figure.

3.3) Using microfluidics, we stimulated the Or42a OSN with a replay of the concentration time course experienced by a larva during runs in a real odor gradient. The response dynamics is largely nonlinear. By testing stereotyped ramps, we were able to conclude that:
- the OSN is sensitive to the first derivative of the stimulus during positive gradients
- the firing pattern of the OSN is mostly driven the stimulus intensity during negative gradient
- abrupt decreases in odor concentration leads to offset inhibition of the OSN as was described in adult flies
Reference:
Schulze et al, 2015 (work from my lab), see the following figure.
For similar work in adult flies, see:
(work from the Lazar group)
(work from the Wilson lab)

For a link with the peripheral representation of gradient in the context of larval thermotaxis, see:
(work from the Samuel lab)

3.4) Due to the nonlinear response characteristics of the Or42a OSN, we used an ODE formalism to describe the input-output transfer function of the neuron. The OSN response dynamics can be accurately modeled by a combination of feedforward inhibition and integral feedback. These two regulatory motifs were chosen as (i) a feedforward regulatory motif appears to control olfactory transduction in C. elegans; (ii) integral feedback form a building block of chemoreception in bacteria, insects and vertebrates. Through parameter optimization, we found that the feedforward motif is sufficient to account for the OSN response dynamics elicited by optogenetic stimulations. In contrast, a combination of the feedforward inhibition and integral feedback is necessary to recapitulate the OSN response dynamics elicited by odor stimulations.
Reference for insects:
(a mine of information coming from the moth)
(work from Bargmann lab)
Nagel and Wilson, 2011 (see above, work from the Wilson lab)
Schulze et al, 2015 (work from my lab), see the following figure

4) Deconstruction of the sensory control of action selection:
4.1) Taking advantage of optogenetics, we substitute the odor for light to elicit olfactory behavior through controlled light stimulations. The same approach was used to study larval chemotaxis by other labs.
References:
(initial work from Stortkuhl lab)
(work from the Gershow lab)
(work from the Samuel lab)
(work from my lab)

4.2) We build a closed-loop tracker to study the behavior of larvae in virtual olfactory realities. This tracker allows us to monitor and classify the behavior of a single larva at a rate of 30Hz. Light stimulation pattern are evoked by predefined rules in open- and closed-loop conditions. This work complements similar efforts to tease apart the role of interneurons in the olfactory circuit of C. elegans.
Reference:
(work by the Ramanathan lab)
Schulze et al, 2015 (work from my lab), see the following figure

4.3) In open-loop stimulation experiments, the probability of implementing a transition from a run to a turn can be predicted from a generalized linear model (GLM). The GLM describes quantitatively how high firing rates of the Or42a OSN suppresses turning while low firing rates (and offset inhibition) facilitate turning. This control rule is consistent with findings made by the Gershow and the Samuel labs based on white noise analysis. Our GLM highlights the contribution of nonlinearities in the OSN response to behavioral control.
Reference:
Gepner et al, 2015 (see above, work from the Gershow lab)
Hernandez-Nunez et al, 2015 (see above, work from the Samuel lab)
Schulze et al, 2015 (work from my lab), see these figures: GLM and relevance of nonlinearities of the OSN response dynamics

Link with similar work in C. elegans:
Kato et al., 2014 (see above, work from the Bargmann lab)
(review)

4.4) In closed-loop stimulation experiments, we demonstrate the relevance of the GLM to predict the behavior of larvae in real odor gradients. Due to the probabilistic nature of the behavioral control, predictions can be made on population of run, which requires to perform a likelihood analysis.
Reference:
Schulze et al, 2015 (work from my lab), see the following figure

[Brian Smith] Reference from discussion 10 June mPN vs uPN:
Sinakevitch et al


Jing Wang's talk on June 11th:


[Text...]


Glenn Turner's talk on June 11th:

Characterization of half of the total population of ORNs in adult Drosophila:
The molecular basis of odor coding in the Drosophila antenna. Hallem EA, Ho MG, Carlson JR. Cell. 2004 Jun 25;117(7):965-79.

Population activity patterns in Mushroom Body with calcium imaging
http://www.jneurosci.org/content/31/33/11772.long
- Patterns remain sparse with natural odors.
- No strong clustering of highly responsive cells or cells with similar odor tuning properties

Connecting physiological and behavioral measures of odor specificity of learning
http://www.jneurosci.org/content/33/25/10568.long
- Morphing blends of odors used to assess behavioral discrimination limits of flies
- Overlap between different odor response patterns predicts (qualitatively) the generalization of Pavlovian associations across different smells
- Sparse format of representations means that a binary rule for synaptic plasticity is still effective (i.e. change synaptic weights if the neuron fires - don’t need to tie strength of plasticity to strength of spiking response).

Theory study in the zebrafish birdsong showing how sparseness can enable a system to learn faster:
http://jn.physiology.org/content/92/4/2274.long
But with more training dense format supports just as high performance

Mushroom Body neurons integrate inputs from different glomeruli
http://www.nature.com/neuro/journal/v16/n12/full/nn.3547.html
- Dendritic imaging shows individual MB neurons collect different types of inputs on their dendritic trees (cells only have 6 different claws on average - there are roughly 53 different types of Projection Neurons from the glomerular layer of the circuit)
- Optogenetic stimulation shows that multiple claws have to be activated for the MB neuron to spike (steep dependence with >4 claws being especially effective. However some cells can be driven with only one claw (with v. high input spike rates)).
- Another paper also used dendritic imaging to come to similar conclusions, I forgot to mention it in the talk: http://www.pnas.org/content/110/29/12084.long

Connectivity to MB neurons is probabilistic
http://www.nature.com/nature/journal/v497/n7447/full/nature12063.html
- Using anatomical tracing, Caron does not find any statistical structure of the inputs to individual MB neurons.
- Gruntman found some evidence for bias in connectivity.
Connectivity clearly probabilistic, with large element of stochasticity. But there may be some bias in connectivity.


Some relevant numbers for the modelers (Ramon, Maxim!)
# KC Claws: mean = 6
http://www.nature.com/nature/journal/v497/n7447/full/nature12063.html
http://www.nature.com/neuro/journal/v16/n12/full/nn.3547.html
http://onlinelibrary.wiley.com/doi/10.1002/cne.22184/abstract;jsessionid=DD55B00DDBCEE44249F2675B3A7C492F.f04t02

# KC neurons = 2000
http://informahealthcare.com/doi/abs/10.1080/01677060802471718

# Different PN types (i.e. # glomeruli) = 54
http://www.nature.com/neuro/journal/v13/n4/full/nn.2489.html
I think there may be better references for this number.

# PN synaptic terminals = approx. 1115
http://jn.physiology.org/content/99/2/734.long

# Microglomeruli = approx. 780 to 1600
http://onlinelibrary.wiley.com/doi/10.1002/cne.22184/abstract;jsessionid=DD55B00DDBCEE44249F2675B3A7C492F.f04t02
A microglomerulus is a PN bouton plus surrounding KC claws. So this number should be the same as # PN synaptic terminals, except that the counts were done based on very different anatomical methods.

#PN active zones per bouton = 22-28
http://onlinelibrary.wiley.com/doi/10.1002/cne.22184/abstract;jsessionid=DD55B00DDBCEE44249F2675B3A7C492F.f04t02
This paper has a lot of very useful numbers in it for those interested in connectivity at this layer.

Summary:
Each KC receives input from 6 PNs on average:
~10% of the 54 different inputs
Each PN sends output to approx 80 KCs:
6 claws * 2000 KCs = 12,000 total inputs to KCs
approx 1500 total PN boutons so each PN bouton contacts about 12000/1500 = 8 claws.
There are 150 PNs to MB, so average 10 boutons per PN
So 10 boutons per PN * 8 claws contacted per bouton = 80 KCs contacted by each PN (4% of the 2000 total)

Another way to look at it is that since each PN bouton has max 30 active zones, the maximum divergence would be:
30 active zones per bouton * 10 boutons per PN -> one PN contacts max 300 KCs

Assumptions:
- KC claws always contact PN boutons and not something else

KC = Kenyon cell. The intrinsic neurons of the Mushroom Body



Brian Smith's talk on June 15th:


[.....]


John McGann's talk on June 16th:


[.....]


David Kleinfeld's talk on June 16th:



[.....]


Ben Strowbridge's talk on June 18th:



[.....]

Lucia Jacobs' talk on June 19th:


[.....]

Ramon Huerta's talk on June 19th:


I attach the presentation up to the point I manage to complete it.
Please follow the paper with the machine learning method comparison here
Method comparison



Paper with the simplified model. It also contains connectivities.

Ramon