Why Different Neuron Parts Learn Differently?

Synaptic plasticity isn’t governed by a single “one-size-fits-all” learning rule. In the motor cortex of learning mice, the same pyramidal neuron uses different long-term potentiation (LTP) mechanisms depending on where the synapse sits on its dendritic tree—apical versus basal. That compartment-specific wiring matters because it suggests the brain can assign distinct learning computations to different input streams within a single cell, rather than treating every synapse as operating under identical rules.

The core molecular logic starts with NMDA receptors acting as coincidence detectors. Glutamate released by a presynaptic spike primes NMDA receptors, but a magnesium ion blocks calcium entry at rest. Only when the postsynaptic membrane becomes sufficiently depolarized—removing the magnesium block—does calcium flood in, triggering cascades that strengthen synapses by increasing receptor content and even enlarging dendritic spines. This calcium-dependent switch underlies long-term changes lasting hours to days, in contrast to short-term potentiation that fades without lasting structural remodeling.

Two routes can generate the depolarization needed for NMDA receptor activation. In classical Hebbian plasticity, the postsynaptic neuron fires an action potential, and a back-propagating action potential travels from the soma into the dendrites. If glutamate is present during this back-propagation, NMDA receptors open and calcium-dependent strengthening follows—capturing the “fire together wire together” principle. A second route, non-Hebbian plasticity, relies on local coactivity: clustered synapses on a small dendritic segment can sum their individual depolarizations enough to relieve the magnesium block locally, even if the soma never spikes. The result is compartmentalized learning, where nearby synapses can cooperate without the neuron’s global output being the deciding factor.

To test whether these rules actually segregate within living animals, researchers trained mice on a lever-press motor task cued by an auditory signal, with sugar reward. As performance improved over days, the team monitored synaptic activity and neuron firing in real time using glutamate-sensitive fluorescent reporters (green flashes at active synapses) alongside a red calcium indicator (RCAMP) to infer when the neuron fired. They then imaged the same dendritic spines across days to quantify structural growth, using spine enlargement as a readout for LTP.

The findings matched the compartment hypothesis. On apical dendrites, spine growth correlated most strongly with local synaptic coactivity—frequent coactivation among neighboring spines on the same apical branch—while whether the neuron fired an action potential was not the primary predictor. A genetic block preventing action potentials left apical LTP largely intact, reinforcing that apical plasticity can proceed via local, spike-independent mechanisms. Basal dendrites told a different story: spine strengthening tracked the classical Hebbian coincidence between synaptic input and the neuron’s own action potentials. Blocking action potentials substantially reduced basal LTP.

Why split learning rules this way remains unresolved, but the transcript points to functional anatomy. Apical dendrites receive feedback connections carrying contextual information, while basal dendrites more often receive feedforward inputs and are positioned to encode prediction errors in predictive-coding frameworks. The emerging picture is that dendritic compartments may implement different computations—binding contextual patterns in apical branches and reinforcing pathways that drive reliable firing in basal branches—offering a more nuanced blueprint for how single neurons perform efficient, multi-purpose learning.