Churchland lab

Last week on Twitter, I confessed my ultimate scientific dream, which is to have lab meeting with my team members, talk about papers, and eat sandwiches. I mean, what could be better than discussing science with folks who have a shared interest and like digging deep into papers? And to have a sandwich at the same time? Its the best! We did a literature blitz this week: each of us presented one figure from a recent paper (many from the biorxiv). Our summaries are below.

Bilal Haider: Spatial attention enhances network, cellular and subthreshold responses in mouse visual cortex
By Anderson Speed, Joseph Del Rosario, Navid Mikail & Bilal Haider Approach: The authors recorded neural responses from the monocular zone in mouse V1. In some trials the mouse’s attention was directed there (due to stimuli repeatedly being shown in the monocular zone) and in others it was directed elsewhere. The early LFP response in layer 4 showed a spatial “attentional enhancement” to task-relevant stimuli, as well as irrelevant probe stimuli. The biggest enhancement over time was in fast-spiking neurons. In noise correlations, the biggest effect of attention was between FS/RS. Finally, they measured the membrane potential of 5 neurons and found that attention depolarizes those neurons, bringing them closer to spiking threshold. Take home message: Spatial attention in mice increases behavioral sensitivity and decreases reaction time. The mechanism proposed here is that there is more spiking and decreased correlations in the receptive field that the animal attends. Skeptics’ corner: How should we think about the behavioral manipulation (presenting a stimulus solely in 1 spatial location for 25 trials, then switching location)? Is it equivalent to single-trial cueing of attention as is usually done in humans or non-human primates (see e.g. https://www.nature.com/articles/nrn3838#Sec6 for a discussion on dissociating expectation, attention and stimulus repetition). It’s also surprising that mice take several trials to adjust their behavior towards the new location, as the blocked spatial context is fully deterministic. See also: This interesting paper from the Glickfeld lab about mouse attention in V1. 

Cortical pattern generation during dexterous movement is input-driven Britton Sauerbrei, Jian-Zhong Guo, Jeremy D. Cohen, Matteo Mischiati, Wendy Guo, Mayank Kabra, Nakul Verma, Brett Mensh, Kristin Branson & Adam W. Hantman Take home message: Inputs from other regions are necessary for keeping the local dynamics in motor cortex (Figure). This paper is a nice example of an experiment designed to understand how multiple areas in the brain interact during behavior.  Approach: the group tracked mouse reaching movements for food. They then perturbed motor cortex, either by perturbing inhibitory neurons, intratelencephalic neurons (Tlx3-cre), or pyramidal tract neurons (Sim1-cre). Following release of the perturbations, for VGAT and Tlx3 perturbations, the movements occurred with a shorter reaction time.  This contradicts a model in which motor cortex acts autonomously. The authors next inactivated motor cortex, and then released it from inactivation, and then in some trials, they also inactivated the input from the motor thalamus. This blocked reaching and the neural trajectory changed considerably. The authors also found stimulating thalamocortical terminals at high frequencies can disrupt behavior and change the neural trajectory. See also: This paper from both Hantman & Dudman about how we should think about pyramidal cell classes in motor cortex. 

Separable  codes for read-out  of mouse primary visual  cortex across attentional Ashley M. Wilson, Jeffrey M. Beck, and Lindsey L. Glickfeld  Take home: Attention drives a change in the magnitude of V1 responses and also how choice is represented, thus impacting the downstream readout. Approach: The authors developed a task in which they could probe the impact of attention by comparing V1 activity responding to identical sensory stimuli on trials that differed only in their attentional state. They achieved this by training mice to do either an auditory or visual detection task, and by including a small number of “invalidly cued” trials in which the non-cued modality changed. It is a really nice task! They also measured orientation tuning during passive viewing. This was critical for interpreting the results of neural activity measured during behavior, because the effect of attention was not uniform across cells: those with orientation preferences that matched that used in the task showed the strongest modulation. Further, the subsequent decoding analyses that the authors performed were really interesting: they found that they were able to decode the animal’s choice from single trial activity during both auditory and visual stimuli. However, the weights used for the two sensory modalities were uncorrelated. This suggests that a different readout was used in each case, a surprising finding. This decoding analysis was among a number of interesting ones; the ability to carry out these analyses in a meaningful way was bolstered by the careful stimulus design and data collection. Skeptics corner: A slightly confusing feature of this task is that on catch trials, the mouse is rewarded for attending to the wrong modality. I guess the fact that such trials are rare makes this okay, and it is clear that they had to design the task this way to match rewards across conditions.  

Sensory and Behavioral Components of Neocortical Signal Flow in Discrimination Tasks with Short-term Memory  Yasir Gallero-Salas, Balazs Laurenczy, Fabian F. Voigt, Ariel Gilad, Fritjof Helmchen.  Take home message: In both a whisking and an auditory task, neural activity during a delay period depends strongly on the animal’s behavioral strategy: specifically, whether the animal is active vs. passive. Approach: The authors measured cortex-wide activity using widefield imaging. They found that during a sensory presentation period, the two modalities activated fairly non-overlapping regions. These included area RL,  for somatosensory stimuli (as previously been shown), and also the mysterious “Area A” for auditory activity. During the delay period, the overall activity pattern was fairly independent of the animal’s strategy and depended most strongly on whether animal’s had been active or passive during the stimulus period. Skeptic’s corner: I remain perplexed as to why there is activity in posterior visual areas during the delay. Notably, this appears to be present in auditory animals (a new and important addition in this paper compared to the previous paper). This is reassuring because the animal’s can’t use vision to solve an auditory task. See also: A previous paper from this group on a similar topic.

Interrogating theoretical models of neural computation with deep inference Sean R. Bittner, Agostina Palmigiano, Alex T. Piet, Chunyu A. Duan, Carlos D. Brody, Kenneth D. Miller, John P. Cunningham Approach: The authors highlight a major problem: to find the distribution of parameters that generate your computation of interest. One way is to do a big search. Here, they instead use a machine learning technique and train a deep network to learn the distribution. The first examples are oscillations in the stomatogastric ganglion; the second example is a model of task switching in the superior colliculus. Take home: This is a new way to generate hypotheses from neural circuit data by looking at parameters that support computation Skeptics corner: There are ways of estimating these in a likelihood free way (like approximate bayesian); also we weren’t totally sure how much data you need to generate the predictions.   

Low dimensional spatio temporal dynamics underlie cortex-wide neural activity Camden J. MacDowell,  Timothy J. Buschman Approach: The authors performed widefield imaging w/ GCaMP6f in passive animals. They observed a lot of spontaneous activity and they partition the 12-minute session into 6 epochs: discovery vs. withheld. They made unsupervised discovery (convNMF) of 16 motifs, but you mostly don’t need them to explain the variance.  19 motifs can explain the vast majority of the variance. Take home: neural activity is “conserved.” Skeptics’ corner: We weren’t sure how this relates to activity during a task, and also the relationship to the resting state network.