The Molecular Information Boundary of Neuronal Electrical Waveforms

Neuronal electrophysiology and gene expression are coupled, but it remains unclear whether an electrical waveform contains a broadly decodable representation of the transcriptome or only selected molecular information. We analyzed paired public Patch-seq current-clamp recordings and transcriptomes using a deliberately conservative pilot framework. Nine waveform-derived features were extracted from raw NWB recordings. Unsupervised transcriptomic principal components were defined without waveform labels and tested by donor-held-out ridge prediction. In the 30-cell module analysis, the best waveform-readable transcriptomic axis was RNA-PC3 (donor-held-out Q²=0.436; observed-predicted Spearman rho=0.641; family-wise max-statistic permutation P=0.008 across 10 modules). However, when broad neuronal-class effects were removed gene-by-gene before latent-module construction, the best residual module in 31 unique cells showed only Q²=0.161 and failed family-wise significance (P=0.335). Global transcriptome geometry was also not recoverable from waveform geometry (rho approximately 0, P approximately 0.96 in the 30-cell analysis). These results support a provisional molecular information boundary: neuronal electrical phenotypes carry selectively decodable transcriptomic identity, but do not behave as compressed reconstructions of the whole transcriptome. Because the cohort is small and drawn from a single public resource, the proposed boundary is a testable hypothesis requiring independent large-cohort replication. Keywords: Patch-seq; electrophysiology; transcriptomics; neuronal identity; latent state; information boundary