Weak pre-before-post pair biases sort synaptic weights in a conductance-based model of spontaneous network bursting

A conductance-based model of a bursting neuronal culture reveals a curious bias in synaptic weight sorting during spontaneous population bursts. The study focused on a model with 48 excitatory and 12 inhibitory stomatogastric-ganglion (STG) model neurons, featuring sparse recurrent connectivity, short-term depression, and pair-based spike-timing-dependent plasticity (STDP). Two cohorts of 10 connectivity wirings were analyzed, with the model's predictions and potential findings recorded beforehand.

Across the cohorts, stronger synapses consistently accumulated a greater net potentiation than weaker ones. However, the pre-registered magnitude criterion for this sorting was not achieved. The spike statistic most strongly linked to this weight sorting was not the first-spike latency, as initially anticipated. Instead, stronger synapses exhibited a remarkably small but consistently high excess of pre-before-post spike pairs.

This excess ranged from 0.2-0.35 percentage points, accumulating over approximately 10^4 pairs per synapse within 60 seconds (median 8,960 in the out-of-sample cohort, around 3 x 10^6 per wiring).

A preregistered intervention aimed at shortening bursts by 41% at a matched burst rate significantly amplified the sorting measure across all five wirings. Another distinct perturbation, which directly targeted burst duration, corroborated the relationship between burst duration and the sorting phenomenon. However, separating the effect of burst duration from the per-burst spike count proved challenging.

All interventions were computationally modeled, with no observational testing in 16 archived recordings of developing cortical cultures. The predicted strength-ordered pair excess was absent, and the predicted negative relationship with burst duration was also absent. The former represents a bounded null - the recordings lacked effects above +0.32 percentage points but could not discern the model's approximate 0.2-point magnitude. The latter was adequately powered under the preregistered criterion.

Interestingly, a consistent positive relationship between spike latency and synaptic strength was observed, a finding not predicted by the model. Despite establishing a reproducible sorting phenomenon within the specific computational model and proposing a burst-shortening prediction, the study does not demonstrate that this mechanism operates in living cortical networks. The decisive test involves interventions on cultured networks employing an independently validated measure of synaptic strength.