Using a Semi-Automated Strategy to Develop Multi-Compartment Models That Predict Biophysical Properties of Interneuron-Specific 3 (IS3) Cells in Hippocampus

Semi-automated strategy. Flowchart outlining the steps in the approach that we used to develop the IS3 cell model databases. Note that gray boxes indicate steps that were performed manually through hand tuning, and green boxes indicate steps that were automated within NEURON and MATLAB.

Experimental measurement histograms. A–R, The histograms are generated from data where the criterion is that they exhibit hyperpolarization (A–E; CIPs include −100 and −90 pA for the “lack of sag feature”), passive depolarization (F; CIPs include 10 and 20 pA for the “no spiking feature”), depolarization with spiking (G–P; CIPs range from 10 to 140 pA with 10 pA intervals for the “normal spiking feature”), or depolarization block (Q, R; CIPs range from 450 to 700 pA with 50 pA intervals for the “depolarization block feature”). Note that the dashed lines indicate the measurements obtained from the selected IS3 cell experimental trace used to compute the distance metric for each model. The signature features, and characteristic measurements and their values are given in Table 2 (second, third, and fourth columns).

Semi-automated strategy and model databases. A1–A3, Top models in S.1, S.2, and SD. The current injection protocol is −100, +20, +50, and +500 pA. B1–B3, Parameter spaces for S.1, S.2, and SD (from left to right), as visualized using CBDR. Each pixel represents a single model and the distance metric of that model. Conductance axes are organized such that overall low-conductance models are in the bottom left quadrants and overall high-conductance models are in the top right quadrants. Black pixels represent models that are eliminated in step 2 and are assigned a distance value of 100 as a result. For S.1, S.2, and SD, respectively, 182 of 432, 59 of 81, and 42 of 81 models did not get rejected (i.e., the remaining colored pixels). C, Example experimental and model database (i.e., for S.1, S.2, and SD) voltage trace measurement histograms for spike half-width and first spike time during 50 pA stimulation. The additional flowchart above the figures shows example rationales for the changes that were made among the S.1, S.2, and SD distributions, including reasoning for why the S.2 models best captured IS3 experimental features.

Example adjustments made in between parameter refinement cycles for SDprox.1 and SDprox.2 model databases. A, For SDprox.1, five cycles of steps 1 to 6 were required (Cycle 1 = 47 remaining of 243 models; Cycle 2 = 64 remaining of 243 models; Cycle 3 = 22 remaining of 243 models; Cycle 4 = 52 remaining of 243 models; Cycle 5 = 70 remaining of 243 models). B, for SDprox.2, three cycles of steps 1 to 6 were required (Cycle 1 = 24 remaining of 243 models; Cycle 2 = 107 remaining of 243 models; Cycle 3 = 146 remaining of 243 models). The CBDR plots show the quality of the parameter space of the model database during each cycle. Note that blue-boxed areas indicate parameter spaces of interest that were focused on. Red-boxed areas indicate parameter spaces that were purposely avoided.

Kv3.1 and Kv2.1 expression in putative IS3 cells. A, B, Immunohistochemistry data showing GFP (green; left) and Kv3.1 expression (red; middle) in the stratum radiatum (A) and pyramidale (B) of a VIP-GFP mouse. Note the presence of Kv3.1 membrane labeling in the soma and proximal dendrites of VIP- expressing cells. C, D, Immunohistochemistry data showing GFP (green; left) and Kv2.1 expression (red; middle) in the stratum pyramidale (C) and radiatum (D) of a VIP-GFP mouse. Scale bar, 10 μm.

VGC distributions in proximal dendrites. A, Action potential amplitude deterioration along a dendrite section (tree 1) in the S.2 top model following a somatic 800 pA current injection for 2 ms (G_Na,t soma = 0.25 S/cm²; G_Na,p soma = 0.0001 S/cm²; G_Ka soma = 0.15 S/cm²; G_Kdrf soma = 1 S/cm²). B, Action potential amplitude deterioration along a dendrite section (tree 1) in the SD top model following a somatic 800 pA current injection for 2 ms (G_Na,t soma/dendrites = 0.06 S/cm²; G_Na,p soma = 0.0002 S/cm²; G_Ka soma/dendrites = 0.1 S/cm²; G_Kdrf soma/dendrites = 0.1 S/cm²). C1, SDprox.1 top model with channels that are uniform from the soma until the first 70 μm of the dendrites using the Boltzmann function. Conductance values are as follows: G_Na,t soma/dendrites = 0.07 S/cm²; G_Na,p soma = 0.000075 S/cm²; G_Kdrf soma/dendrites = 0.25 S/cm²; G_Ka soma/dendrites = 0.07 S/cm². Note that the current injection protocol is −100, +20, + 50, and +500 pA. C2, Parameter spaces for SDprox.1, as visualized using CBDR. Note that 70 of 243 models did not get rejected. C3, Action potential amplitude deterioration along a dendrite section (tree 1) in the above SDprox.1 top model following a somatic 800 pA current injection for 2 ms. D1, SDprox.2 top model with channels that are uniform from the soma until the first 70 μm of the dendrites using the Boltzmann function. Conductance values are as follows: G_Na,t soma/dendrites = 0.055 S/cm²; G_Na,p soma = 0.00015 S/cm²; G_Kdrf soma/dendrites = 0.295 S/cm²; G_Ka soma = 0.07 S/cm². Note that the current injection protocol is −100, +20, + 50, and +500 pA. D2, Parameter spaces for SDprox.2, as visualized using CBDR. Note that 146 of 243 models did not get rejected. D3, Action potential amplitude deterioration along a dendrite section in the above SDprox.2 top model following a somatic 800 pA current injection for 2 ms.

Electrotonic analysis of the M2 morphology. A, A guide of the M2 morphology designating all of the morphological subsections, as well as showing both the spatial scale and diameter. B, Electrotonic distance along the dendritic arbor for voltage flowing into the soma. C, Electrotonic distance along the dendritic arbor for voltage flowing away from the soma. Note that the electrotonic distance is equal to the log value of attenuation, where attenuation is measured as voltage upstream/voltage downstream. More specifically, voltage upstream is an applied 1 mV signal, and voltage downstream is the downstream response to the 1 mV signal. In this sense, electrotonic distances >1 would imply a 10-fold attenuation in the signal.