In 47 IBL mice, tracer-weighted cortical timescales predicted a small, specific part of the mouse striatal timescale map (ρ = 0.24), but striatal position predicted five times more variance

Question: do mouse striatal neurons get their intrinsic timescale from the cortical areas that send input to them? Approach: we used spontaneous-activity timescales (iSTTC fits, Gülenç et al. 2026) of 6,660 striatal units from 82 International Brain Laboratory probes in 47 mice and 12 labs, and of 16,705 cortical units. For each 100 µm striatal voxel, we calculated the input share of 42 cortical areas from 494 Allen tracer injections. The predicted timescale is the input-weighted mean of cortical area timescales, with the test mouse removed from the cortical map. We compared the real map with 2,000 surrogate cortical maps that keep its spatial autocorrelation. We repeated the test with our own fits of 1,952 good units, and we used task signals of 3,263 good units as a positive control. Results: predicted and measured striatal timescales correlated (Spearman ρ = 0.24, mouse-bootstrap 95% interval 0.09 to 0.35, surrogate p = 0.0015; within labs 0.18, p = 0.013). The mixed-model slope was 0.98 ± 0.09. The result held with disjoint halves of mice (mean ρ = 0.22, p = 0.006) and with cortical units only from probes that did not enter the striatum (ρ = 0.20; within sessions 0.18, p = 0.008). Of eight cortical maps, the timescale map gave the highest positive correlation; the thalamic input map gave ρ = −0.15. Wide-waveform units gave ρ = 0.30. But the prediction explained only 1.9% of unit variance in held-out mice, while a cubic function of position explained 10.5%, also in held-out labs. Measured timescales became shorter away from the midline two times faster than predicted, and the fast-spiking unit fraction did not explain this. Our own fits gave the same sign (ρ = 0.15 to 0.22) but did not pass the surrogate test. In the positive control, reward-versus-error selectivity followed the cortical feedback-decoding map (ρ = 0.18, p = 0.003; within sessions 0.07, p = 0.013); movement and choice signals followed their maps weakly. Limitations: cortical area timescales had low split-half reliability between mice (ρ = 0.36); input-weighted maps vary mainly along one medial-lateral axis, so the test has little power inside one probe; tracer data and recordings come from different mice.

striatum · intrinsic timescales · corticostriatal projections · International Brain Laboratory · Allen Mouse Brain Connectivity Atlas · Neuropixels · structure-function · spatial null models · Moran spectral randomization · mouse