A comprehensive knowledge base of synaptic electrophysiology in the rodent hippocampal formation

Abstract

The cellular and synaptic architecture of the rodent hippocampus has been described in thousands of peer-reviewed publications. However, no human- or machine-readable public catalog of synaptic electrophysiology data exists for this or any other neural system. Harnessing state-of-the-art information technology, we have developed a cloud-based toolset for identifying empirical evidence from the scientific literature pertaining to synaptic electrophysiology, for extracting the experimental data of interest, and for linking each entry to relevant text or figure excerpts. Mining more than 1,200 published journal articles, we have identified eight different signal modalities quantified by 90 different methods to measure synaptic amplitude, kinetics, and plasticity in hippocampal neurons. We have designed a data structure that both reflects the differences and maintains the existing relations among experimental modalities. Moreover, we mapped every annotated experiment to identified potential connections, that is, specific pairs of presynaptic and postsynaptic neuron types. To this aim, we leveraged Hippocampome.org, an open-access knowledge base of morphologically, electrophysiologically, and molecularly characterized neuron types in the rodent hippocampal formation. Specifically, we have implemented a computational pipeline to systematically translate neuron type properties into formal queries in order to find all compatible potential connections. With this system, we have collected nearly 40,000 synaptic data entities covering 88% of the 3,120 potential connections in Hippocampome.org. Correcting membrane potentials with respect to liquid junction potentials significantly reduced the difference between theoretical and experimental reversal potentials, thereby enabling the accurate conversion of all synaptic amplitudes to conductance. This data set allows for large-scale hypothesis testing of the general rules governing synaptic signals. To illustrate these applications, we confirmed several expected correlations between synaptic measurements and their covariates while suggesting previously unreported ones. We release all data open-source at Hippocampome.org in order to further research across disciplines.

Keywords: circuit biophysics; computational biology; information storage and retrieval; knowledge bases; models; neuron types; synapses/physiology.

Figure 1.. From neuron types to synapse types.

(A) Red, the axons of the dentate gyrus (DG) basket neuron type innervate SG (stratum granulare), and those of the HICAP type invade SMi (stratum moleculare - inner). Blue and green, the dendrites of both neuron types span all four DG layers. The local axons of granule cells innervate the hilus (H) while its dendrites span SMi and SMo (stratum moleculare - outer). HICAP axons and granule dendrites are co-located, as are the basket axons and the granule perisomatic region; therefore, these neurons give rise to two distinct inhibitory synapse types. Morphologies rendered with neuTube (Feng, Zhao, & Kim, 2015) using data from the Bausch and Lien archives (Bausch, He, Petrova, Wang, & McNamara, 2006; Liu, Cheng, & Lien, 2014) of NeuroMorpho.Org (Ascoli, Donohue, & Halavi, 2007). (B) The presynaptic spikes (upper traces) generate postsynaptic signals (lower traces) digitized and plotted from pair recording (Liu et al., 2014), from which we identified “Fast- Spiking” as DG basket and “Non-Fast-Spiking” as DG HICAP (IN: interneuron). (C) The 122 known neuron types in the rodent hippocampal formation (presynaptic: rows; postsynaptic: columns) form 3,120 synapse types. The heat map (SUB: subiculum; EC: entorhinal cortex) represents the number of distinct layers in which excitatory and inhibitory axons co-localize with relevant postsynaptic elements. For instance, the inhibitory synapse types in (A) have only one co-location each (in SMi and SG, respectively), corresponding to a −1 value in the directional connectivity matrix.

Figure 2.. Literature mining and knowledge extraction.

For every experiment, we provide (i) a summary; (ii) connectivity ratios, cell-types counts, and any other relevant notes; (iii) bath and pipette solutions; (iv) recorded modalities and pertinent data such as postsynaptic potential (V_m), liquid junction correction (V_j), and measured or calculated reversal potentials (E_rev), each tagged with a reference ID; (v) needed assumptions for neuron identification; (vi) a machine-readable query; and (vii) mapped synapse types and related confidences (blue border: high confidence; others: low confidence), along with identifiers for the publication (PMID), experiment (eID), and extracted data IDs (dIDs). The data in this example is from (Liu et al., 2014).

Figure 3.. Mapping summary.

(A) Integrated knowledge mapping (clockwise from top): “proper” (blue), “high-confidence fuzzy” (green), and “low-confidence fuzzy” (purple). Grid patterns indicate validated (as opposed to potential) connections. (B) An individual synapse type may be linked to multiple experiments with variable mapping confidence. (C) Amplitude and kinetics are the most prevalently reported synaptic electrophysiology properties. (D) Mapping degeneracy by stimulation method: unitary signals (mostly paired recording), evoked (extracellular) and spontaneous. Filled circles represent all methods together.

Figure 4.. Data modalities.

(A) Synaptic signals can be generated in eight different modalities depending on stimulation methods (e, u, s, and m) and response type (C or P). (B) The most prevalent modality among all extracted data (upper chart) is eC and the most prevalent combination of multiple modalities in the same experiment (right chart) is between eC and eP.

Figure 5.. Measurement methods diversity.

Synaptic conductance is the most prevalent measure for amplitude, single-exponential decay time constant (τ_Decay) for kinetics, paired-pulse ratio of 2^nd synaptic signal to the 1^st (2/1 Amplitude Ratio) for plasticity, and failure rate of the 1^st signal for other features.

Figure 6.. Data access.

The described knowledge base of synaptic electrophysiology is freely available online. (A) Synapse types are searchable by the properties of the presynaptic and postsynaptic neuron types. (B) They are linked to experiment IDs categorized by mapping confidence. (C) The details and summaries of any experiment (for example, experiment with eID 331) can be reviewed while checking excerpts as evidence and (D) the extracted data is directly accessible. This example is from (Struber, Jonas, & Bartos, 2015).

Figure 7.. Correcting the liquid junction potential reduces the measured synaptic reversal potential error.

After correcting the experimentally measured synaptic reversal potential (E) for liquid junction potential (V_j), the difference between E_theoretical and E_experimental becomes close to zero on average. All data needed to calculate a pair (each grey line) come from one experiment with different solutions and temperature, which lead to different V_j and E.

Figure 8.. Faster GABAergic synapses are stronger.

(A) GABAergic unitary synaptic potency significantly correlates with decay, both at temperatures ≥31°C or <31°C. (B) The conductance of slower synapses (decay time constant above median) is significantly smaller than that of faster ones. Each data point is the average or single result of a separate experiment.