In mouse V1 EM, E→I→E wiring was like-to-like with the sign that holographic far suppression needs, but EM orientation specificity was 4–7 times too weak in a truncated linear model

Question: holographic stimulation of co-tuned ensembles in mouse V1 layer 2/3 (Oldenburg et al. 2024) activates nearby cells with the same orientation and suppresses far cells with the same orientation more than orthogonal cells. The authors' model needed orientation-specific E→I and I→E connections. Does the orientation specificity of EM wiring explain these effects? Approach: we re-analysed the released per-cell data (160 ensembles, 13 mice) with ensemble × distance fixed effects, a mouse bootstrap and a no-stimulation placebo. In the MICrONS EM volume (one mouse, v1507), we measured the orientation specificity κ (wiring ∝ 1 + κ cos 2Δθ at fixed lateral distance) of E→E and E→I→E wiring among 7,333 tuned L2/3 pyramidal cells through 7,835 interneurons. We put the EM κ values into the published truncated linear model on synthetic ensembles, with no free orientation parameter. Results: in 25 co-tuned ensembles (8 mice), the orientation slope of ΔF/F was +0.063 (95% CI +0.036 to +0.080) at 15–30 µm and −0.0065 (−0.011 to −0.0017) at 50–150 µm; the placebo slope was +0.005 (−0.003 to +0.016). In EM, E→I→E κ was 0.051 (0.044–0.058) at 50–150 µm for basket and Martinotti cells alike, larger than E→E κ (0.024 at 0–30 µm, 0.028 at 50–150 µm). Orientation shuffles inside 25–200 µm blocks kept 63–80% of the E→I→E κ, so the spatial clustering of orientation carried most of it; with proofread presynaptic axons only, κ was 0.014 (−0.013 to 0.041). With EM κ values the model gave the correct sign of far suppression (untuned inhibition gave the wrong sign) but only 12–44% of its size and at most 2% of the near activation; even upper 95% EM bounds corrected for tuning noise gave at most 39% and 32%. The data needed κ ≈ 0.62 for excitation and 0.35–0.56 for suppression, 7 and 4 times the largest corrected EM values. Conclusion: EM wiring supports the sign of like-to-like suppression when specificity is measured as a perturbation feels it, but synapse counts and sizes are much less orientation-specific than influence. Strength specificity, recurrent amplification or near-target measurement effects must supply the rest. Limitations: EM and function come from different mice; few co-tuned ensembles; digital-twin tuning; mostly unproofread axons; truncated linear model on synthetic geometry.

MICrONS · functional connectomics · holographic optogenetics · mouse visual cortex · orientation tuning · like-to-like connectivity · feature competition · disynaptic inhibition · influence mapping · structure-function · electron microscopy · linear network model