Temporal derivative computation in the dorsal raphe network revealed by an experimentally-driven augmented integrate-and-fire modeling framework

Abstract

By means of an expansive innervation, the relatively few phylogenetically-old serotonin (5-HT) neurons of the dorsal raphe nucleus (DRN) are positioned to enact coordinated modulation of circuits distributed across the entire brain in order to adaptively regulate behavior. In turn, the activity of the DRN is driven by a broad set of excitatory inputs, yet the resulting network computations that naturally emerge from the excitability and connectivity features of the various cellular elements of the DRN are still unknown. To gain insight into these computations, we developed a flexible experimental and computational framework based on a combination of automatic characterization and network simulations of augmented generalized integrate-and-fire (aGIF) single-cell models. This approach enabled the examination of causal relationships between specific excitability features and identified population computations. We found that feedforward inhibition of 5-HT neurons by heterogeneous DRN somatostatin (SOM) neurons implemented divisive inhibition, while endocannabinoid-mediated modulation of excitatory drive to the DRN increased the gain of 5-HT output. The most striking computation that arose from this work was the ability of 5-HT output to linearly encode the derivative of the excitatory inputs to the DRN. This network computation primarily emerged from the prominent adaptation mechanisms found in 5-HT neurons, including a previously undescribed dynamic threshold. This novel computation in the DRN provides a potential mechanism underlying some of the functions recently ascribed to 5-HT in the context of reinforcement learning.

Nonlinear computation by a habenula-driven recurrent inhibitory network in the raphe

“…To investigate the computation carried out by DRN circuits under habenular drive (e.g., threat information), we leveraged our mathematical description of the short-term plasticity of 5-HT synapses in a network model (see Methods). Our model consisted of 5000 conductance-based leaky-integrate-and-fire 5-HT neurons, building on recent simplified single-cell modeling of 5-HT neurons (Harkin et al, 2021). 25% of 5-HT neurons received excitatory input from LHb ( Σ 7.5 nS ), while all 5-HT neurons received background Poisson excitation as well as 5-HT 1A R-mediated inhibition from other 5-HT neurons ( Σ 2.5 nS ) (Fig.…”

“…Briefly, identified serotonin neurons from SERT-CRE-tdTomato mice were recorded in whole-cell current-clamp configuration, and a noisy input stimulus was presented through multiple test and training repetitions. These repetitions were used to train a generalized integrate-and-fire model to each individual serotonin neuron by maximizing the likelihood of the model (for full details, see Harkin et al, 2021).…”

“…Note that these reflect unitary conductance values, and neurons receive multiple LHb ( g LHb ) and recurrent GIRK ( g GIRK ) inputs, as well as a constant background rate of Poisson inputs ( g pois ). For the AHP dynamics, the following values were extracted from fits on identified 5-HT neurons using a previously described protocol (see Pozzorini et al, 2015; Harkin et al 2021): Note that since the generalized integrate-and-fire model in Pozzorini et al (2015) extracts a spike-triggered current for the AHP, we convert this to a conductance by dividing by the driving force at spike threshold.…”

Dentate gyrus mossy cells exhibit sparse coding via adaptive spike threshold dynamics

“…2Bii), as well as the underestimation of the first spike latency, suggest that there is an underlying mechanism delaying the first spike. To account for this fact, we tried implementing a simplified fast inactivating K + -current, also known as an IA current, as proposed for serotonergic neurons in the raphe nucleus (Harkin et al, 2021) since these currents are also known to be present in mossy fiber boutons (Geiger and Jonas, 2000). However, the IA currents caused the rheobase currents in our model to rise significantly, such that we were able to initiate spiking only for currents greater than 150 pA, contrary to what we observed in hMCs (some cells could spiked with current injections as low as 20 pA, data not shown).…”

Untitled

“…In [62], an extensive modeling and experimental study of DRN neurons and networks containing them were carried out. The DRN neurons were described by generalized integrate and fire models, which required only a few parameters that could be estimated accurately.…”