iiinova — where your digital self keeps evolving

12/ We propose that sweep coding in layer 5a may be related to texture decoding. The longer integration time makes it possible to combine current sensory inputs with modulatory signals — possibly motor-related — from higher order POm.

100

11/ This reveals a new dimension in thalamocortical computation: 🔹 Fine, fast features like sticks are inherited from thalamus 🔸 Broader, global features like sweeps are computed in cortex via temporal integration

100

10/ So where do sweeps come from? In layer 5a, we found that sweep-tuned neurons integrate stick inputs from VPM and POm over longer timescales.

100

9/ Recordings in VPM and POm showed that both thalamic nuclei primarily encode sticks. POm adds some diversity, but sweep tuning is not clearly present.

100

8/ Sticks dominated in layer 4 and 3. Sweeps were found in layers 5a and 5b. But can these features be inherited from the thalamus?

100

7/ We identified two distinct types of responses in the cortex: 🔴 Sticks — brief, fast, single-whisker deflections ⚫ Sweeps — broad, multi-whisker movements with large angular changes These were tuned to perpendicular axes in the feature space.

100

6/ We confirmed this with an independent sparse noise stimulus — random single-whisker deflections — and separated the two functional populations.

100

5/ But cells were not uniformly selective across this space. They tended to cluster around two specific feature angles — suggesting a subspace tuning.

100

4/ We found that the whisker movements that elicited the strongest responses belonged to a low dimensional feature space. We could project each cell’s preferred stimulus into this space: the closer to the edge, the more selective.

100

3/ Velocity came out on top, as seen in rats (Harrell et al. 2020), and contrary to stick-slip models where velocity and acceleration are encoded equally.

2/ We designed Gaussian white noise stimuli — optimized to test position, velocity, acceleration — to find which parameter was best encoded by neurons.

1/ We used a unique setup: 24 whiskers deflected with micrometer precision and millisecond timing. This allowed us to deliver naturalistic, reproducible input across the full whisker pad, while recording neurons multiple in the barrel cortex.

100

🧵New preprint from @tom-quetu.bsky.social and me, done in Dan Shulz’s lab @touchmovelab.bsky.social at @neuropsi.bsky.social : We uncover how a tactile code emerges in cortical layer 5a from temporal integration of thalamic input. www.biorxiv.org/content/10.1... Let’s break it down 👇

biorxiv.org

154

100

100

10/ So where do sweeps come from? In layer 5a, we found that sweep-tuned neurons integrate stick inputs from VPM and POm over longer timescales — and are probably modulated by POm.

100

9/ Recordings in VPM and POm showed that both thalamic nuclei primarily encode sticks. POm adds some diversity, but sweep tuning is not clearly present.

100

8/ Sticks dominated in layer 4 and 3. Sweeps were found in layers 5a and 5b. But can these features be inherited from the thalamus?

100

7/ We identified two distinct types of responses in cortex: 🔴 Sticks — brief, fast, single-whisker deflections ⚫ Sweeps — broad, multi-whisker movements with large angular changes These were tuned to perpendicular axes in the feature space.

100

6/ We confirmed this with an independent sparse noise stimulus — random single-whisker deflections — and separated the two functional populations.

100

5/ But cells were not uniformly selective across this space. They tended to cluster around two specific feature angles — suggesting subspace angle tuning.

100

100

3/ Velocity came out on top, as seen in rats (Harrell et al. 2020), and contrary to stick-slip models where velocity and acceleration are encoded equally.

2/ We designed and tested optimized kinematic stimulus spaces — position, velocity, acceleration — to find which parameter was best encoded by neurons.

Olivier Marre @oliviermarre.bsky.social · 03/06/2025

1/ We used a unique setup: 24 whiskers deflected with micrometer precision and millisecond timing. This allowed us to deliver naturalistic, reproducible input across the full whisker pad.

100

Reposted by Matías Goldin

In case this is helpful for anyone, here is a post on how to get a job in french academia : trialsanderrors.substack.com/p/getting-a-... Aimed at non-french scientists who are curious, but french ones may find it useful too.

01712

13/ Big thanks to all co-authors for making this possible! 🎉 Remi Baroux & Ulisse Ferrari And to our funding sources: @agencerecherche.bsky.social ANR, IHU FOReSIGHT, @c-brains.bsky.social DIM-C Brains, @sorbonne-universite.fr SCAI, Institut de la Vision, @cnrs.fr CNRS, @inserm.fr INSERM

000

12/ ⚠️ Without proper color balance, we risk misinterpreting color interactions in retinal processing! This has important implications for future studies on natural color illumination in vision research.

100

11/ 💡 Conclusion: Achieving a precise color balance is crucial when studying retinal color processing at small spatial & fast temporal scales. This ensures both M- and S-opsin signals arrive simultaneously & reliably to ganglion cells, matching natural daylight conditions.

100

10/ We also tested whether our experimentally found balanced illumination aligns with natural photopic levels. It turns out that our empirically determined ratio is much closer to natural daylight conditions than to an M*/S* = 1 scenario!

100

9/ We analyzed spectral data from real-world environments to determine the M*/S* ratio mice experience throughout the day. Surprisingly, it remains remarkably constant at ~9.8.

100

8/ Why does UV require less activation to elicit comparable responses? This suggests that UV pathways undergo different amplification mechanisms compared to green pathways. But what about natural light conditions?

100

7/ We repeated this with full-field illumination chirps and found the same M*/S* ratio requirement for balanced responses.

100

6/ We also tested whether these effects were reversible. After presenting a high S* unbalanced checkerboard, we switched to a green-only checkerboard for 45 min. Green RFs recovered within minutes, though some adaptation effects remained.

100

5/ At low S*, 78% of UV Receptive Fields (STA) were lost and chirp responses disappeared. Only at M*/S* ≈ 20 (S*=5x10^2 isomerisations/second.cone) we recover reliable responses to both colors. At this ratio, both firing rates & detected receptive fields were also balanced for UV and green.

100

4/ This suggests that simply equalizing opsin activation isn’t enough for accurate color processing. So what balance allows reliable responses to both colors? To find the M*/S* ratio that balances reliability, we systematically increased S* level while keeping M* level constant.

100

3/ We used two different stimuli to probe ganglion cell responses: interleaved UV-green color checkerboards and full-field color chirps. First, we equalized S- & M-opsin activation based on isomerization rates. But surprise! Responses to green were abolished.

100

2/ Color vision helps animals find food & evade predators. It starts in the retina, where visual input is processed by neural circuits But how do retinal ganglion cells respond to multiple color sources? We investigated this recording retinal responses to UV & green light using multielectrode array

100