Rapid learning of neural circuitry from holographic ensemble stimulation enabled by model-based compressed sensing

Abstract

Discovering how neural computations are implemented in the cortex at the level of monosynaptic connectivity requires probing for the existence of synapses from possibly thousands of presynaptic candidate neurons. Two-photon optogenetics has been shown to be a promising technology for mapping such monosynaptic connections via serial stimulation of neurons with single-cell resolution. However, this approach is limited in its ability to uncover connectivity at large scales because stimulating neurons one-by-one requires prohibitively long experiments. Here we developed novel computational tools that, when combined, enable learning of monosynaptic connectivity from high-speed holographic neural ensemble stimulation. First, we developed a model-based compressed sensing algorithm that identifies connections from postsynaptic responses evoked by stimulation of many neurons at once, considerably increasing the rate at which the existence and strength of synapses are screened. Second, we developed a deep learning method that isolates the postsynaptic response evoked by each stimulus, allowing stimulation to rapidly switch between ensembles without waiting for the postsynaptic response to return to baseline. Together, our system increases the throughput of monosynaptic connectivity mapping by an order of magnitude over existing approaches, enabling the acquisition of connectivity maps at speeds needed to discover the synaptic circuitry implementing neural computations.

Compressive sensing of neuronal connectivity maps from subsampled, cell-targeted optogenetic stimulation of a network model

“…Prior work has applied compressive sensing to the synaptic connectivity mapping problem, albeit in highly simplified simulation settings or from postmortem anatomical data 36,55,56 . More recent work has applied this approach combined with parallel stimulation of visually identified cells with measured single cell PSCs in vitro 38 . Here, we pushed the limits of the simulation environment by incorporating much larger network models with recurrent connectivity and varying many critical parameters to closely resemble biological networks.…”

“…In contrast to the sequential approach, this parallel stimulation approach leverages inferred cell-type information, network sparsity and the theory of compressive sensing (CS) 34,35 to discover synaptic connectivity with far fewer measurements. Compared to other related work 36–38 , our approach is particularly novel in recapitulating much more biologically plausible network topologies and dynamics such as recurrent connectivity, varying levels of background noise and memory retention of past inputs. We demonstrate that at 10% sparsity typically found in cortical networks 27–32 , CoCoMap achieved >90% performance with only half the number of measurements that would be needed using the sequential approach.…”

High-throughputin vivosynaptic connectivity mapping of neuronal micro-circuits using two-photon holographic optogenetics and compressive sensing

“…In our experimental conditions we used Δt ≈100 ms (with t SLM ≈40 ms). The use of faster SLM (refresh rate 300Hz) 42 , that will lower t SLM down to ≈ 2-3ms, or the use of approaches for fast (50-90 μs) scanning through multiple holograms 101 , also combined with current deconvolution algorithms 75 , could reduce the sequential time interval to Δt ≈ t ill or Δt ≈ t rec if t rec > t ill .…”

Bayesian target optimisation for high-precision holographic optogenetics

“…[Ahmadian et al, 2011] considers optimising spike-timing using two-photon stimulation (in addition to electrical stimulation), but focused on single neurons and did not use GP estimation methods. Finally, a number of papers have developed statistical models of optogenetic data to infer functional or synaptic connectivity [Hu and Chklovskii, 2009, Shababo et al, 2013, Aitchison et al, 2017, Draelos and Pearson, 2020, Triplett et al, 2022, Printz et al, 2023, but do not consider identifying the exact stimulation parameters needed to evoke specific activity patterns. Here, we unify these three approaches to develop a novel computational framework for optimal holographic stimulation of neural ensembles.…”

“…Therefore, an ORF model should leverage prior knowledge of how neurons respond to stimulation. Importantly, spike probability should increase monotonically with laser power until saturation [Triplett et al, 2022], and the ORF should approximately match the shape of the somatic membrane (though enlarged to account for the ∼10 µm diameter of the holographic disk [Bounds et al, 2021]). However, any residual expression of opsin molecules in the proximal dendrites of a neuron creates unique differences in its ORF shape compared to other neurons.…”