(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



Synaptic and dendritic architecture of different types of hippocampal somatostatin interneurons [1]

['Virág Takács', 'Laboratory Of Cerebral Cortex Research', 'Hun-Ren Institute Of Experimental Medicine', 'Budapest', 'Zsuzsanna Bardóczi', 'Áron Orosz', 'János Szentágothai Doctoral School Of Neurosciences', 'Semmelweis University', 'Abel Major', 'Luca Tar']

Date: 2024-03

GABAergic inhibitory neurons fundamentally shape the activity and plasticity of cortical circuits. A major subset of these neurons contains somatostatin (SOM); these cells play crucial roles in neuroplasticity, learning, and memory in many brain areas including the hippocampus, and are implicated in several neuropsychiatric diseases and neurodegenerative disorders. Two main types of SOM-containing cells in area CA1 of the hippocampus are oriens-lacunosum-moleculare (OLM) cells and hippocampo-septal (HS) cells. These cell types show many similarities in their soma-dendritic architecture, but they have different axonal targets, display different activity patterns in vivo, and are thought to have distinct network functions. However, a complete understanding of the functional roles of these interneurons requires a precise description of their intrinsic computational properties and their synaptic interactions. In the current study we generated, analyzed, and make available several key data sets that enable a quantitative comparison of various anatomical and physiological properties of OLM and HS cells in mouse. The data set includes detailed scanning electron microscopy (SEM)-based 3D reconstructions of OLM and HS cells along with their excitatory and inhibitory synaptic inputs. Combining this core data set with other anatomical data, patch-clamp electrophysiology, and compartmental modeling, we examined the precise morphological structure, inputs, outputs, and basic physiological properties of these cells. Our results highlight key differences between OLM and HS cells, particularly regarding the density and distribution of their synaptic inputs and mitochondria. For example, we estimated that an OLM cell receives about 8,400, whereas an HS cell about 15,600 synaptic inputs, about 16% of which are GABAergic. Our data and models provide insight into the possible basis of the different functionality of OLM and HS cell types and supply essential information for more detailed functional models of these neurons and the hippocampal network.

Funding: This work was supported by the Frontline Research Excellence Program of the Hungarian National Research, Development and Innovation Office (nkfih.gov.hu, NRDI Fund 133837), the Hungarian Brain Research Program NAP3.0 (agykutatas.hu, NAP2022-I-1/2022) and the European Union project RRF-2.3.1-21-2022-00011 within the framework of the Translational Neuroscience National Laboratory to G.N.; the European Union Human Brain Project ( www.humanbrainproject.eu , SGA3, 945539) and the Hungarian Brain Research Program NAP2.0 (agykutatas.hu, 2017-1.2.1-NKP-2017-00002) to T.F.F. and G.N.; the European Union project RRF-2.3.1-21-2022-00004 within the framework of the Artificial Intelligence National Laboratory to S.K. and G.N.; the New National Excellence Program of the Ministry for Innovation and Technology and the Ministry of Human Capacities, (nkfih.gov.hu, ÚNKP-19-2-I-SE-36 and ÚNKP-18-2-I-SE-20 to A.M, ÚNKP-19-3-I-SE-9 to M.M, ÚNKP-23-2-II-DE-146 to H.S. and ÚNKP-23-3-I-SE-48 to Á.O.); and the National Academy of Scientist Education Program of the National Biomedical Foundation under the sponsorship of the Hungarian Ministry of Culture and Innovation to G. N. and H.S ( www.edu-sci.org ). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability: All data are fully available without restriction, all relevant data are included in the manuscript and its supporting information files, whereas the models and the source code for all the software developed for this study are available at https://zenodo.org/records/10580838 . Models are also available at https://modeldb.science/2016137 .

Copyright: © 2024 Takács et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

We cell-type-specifically labeled the 2 populations and reconstructed their whole dendritic trees at the light microscopic level. Then, we created 3D reconstructions of randomly selected dendritic (n = 108) and axonal (n = 24) segments of OLM and HS cells and measured several of their ultrastructural parameters using a scanning electron microscope. Semi-automatic acquisition of serial electron microscopic images enabled us to 3-dimensionally reconstruct thousands of synapses (n = 5,844) and correlate their anatomical parameters (size, linear density, surface density) to their location on the dendritic trees. We measured the electrophysiological properties of OLM and HS cells using in vitro hippocampal slices. To compare the signal propagation properties of OLM and HS cells, multi-compartmental passive and active models of OLM and HS cells were created by combining the data sets of the sampled dendritic segments and somata, the 3D morphology of cells, and the electrophysiological parameters.

Here, we present the most complete reconstruction and analyses of the somato-dendritic and synaptic architecture of 2 classes of interneurons so far. We examined 2 types of SOM interneurons, the OLM and the HS cells because they play important roles in memory processes and the coordination of functionally related brain areas. These 2 cell types are located in the same layer, both express SOM and have similar dendritic morphology at the light microscopic level but they are functionally different. We tested whether their different functions are reflected in differences in their finer morphological details.

To understand the functional role of hippocampal interneurons and to build realistic models of the CA1 network, quantitative data are required about their morphological architecture and their connectivity [ 51 ]. The density and distribution of glutamatergic and GABAergic synaptic inputs, as well as the postsynaptic target selectivity are characteristically different features of different cell types [ 3 , 5 , 44 , 51 – 54 ]. Our knowledge of the anatomical parameters mentioned above is incomplete for many interneurons, especially in mice.

Theta and gamma oscillations as well as sharp-wave-associated ripples accompany different stages of memory processing in the hippocampus [ 46 – 48 ]. OLM and HS cells have characteristically different firing patterns during sharp wave-ripples: OLM cells are silent or decrease their firing [ 49 , 50 ], whereas double projection (HS) cells discharge with high frequency during these events [ 36 ].

Another group of CA1 somatostatin-positive neurons are primarily not local interneurons; instead, they send long-range axonal projections to other brain areas such as CA3, the dentate gyrus, the subiculum, the MS, and the medial entorhinal cortex [ 36 – 41 ]. Their main role may be to temporally coordinate the activity of functionally related brain areas [ 37 , 42 ], which is important during memory processes. A large subgroup of these somatostatin-positive cells are the double-projecting cells which send long-range axons to both the MS and the subiculum and have sparse local axonal branches in CA1 str. oriens and radiatum [ 36 , 43 – 45 ]. Although in rats probably all septally projecting interneurons innervate the subiculum as well [ 36 ], no such data is available in the mouse yet. Here, we investigated these cells using retrograde labeling. Because we labeled them from the MS and could not examine their potential projections to other areas, we will use the classical terminology: we call them hippocampo-septal (HS) neurons in this study. Dendritic trees of HS cells are similar to those of OLM cells: These dendrites are horizontally distributed in str. oriens and are sparsely spiny [ 36 , 44 , 45 ]. In rats, they receive a very large number of glutamatergic inputs and a smaller amount of GABAergic innervation, a substantial proportion of which originates from GABAergic neurons of the MS [ 44 ]. Their local axons form output synapses predominantly with PC dendrites in adult rats [ 36 , 44 ], although HS cells were also found to target interneurons in in vitro slice preparations of juvenile rats [ 45 ].

The most abundant type of locally projecting SOM-positive interneurons in CA1 are the oriens-lacunosum-moleculare (OLM) cells that were discovered by McBain and colleagues [ 27 ]. They have a horizontally oriented dendritic tree in str. oriens and a dense axonal arbor in str. lacunosum-moleculare. OLM cells receive most of their glutamatergic inputs from local CA1 PCs [ 28 ] and target the distal dendrites of the same population of cells [ 29 , 30 ]; therefore, they are ideally suited for feed-back regulation of the CA1 network [ 31 ]. Although in much smaller numbers, they also innervate interneurons, some of which interneurons target the proximal PC dendrites [ 29 , 32 , 33 ]. During contextual memory formation, CA1 PCs receive information about the context (multisensory information) from Schaffer collaterals of CA3 PCs to the proximal dendrites of CA1 PCs, and they integrate it with the unconditioned stimulus (e.g., an aversive stimulus) that is broadcasted by the temporo-ammonic pathway from the entorhinal cortical layer III PCs to the distal dendritic region of the CA1 PCs in str. lacunosum-moleculare. Because OLM cells target PC dendrites in str. lacunosum-moleculare, they can gate the information flow onto CA1 PC by directly excluding entorhinal inputs from a range of PCs. They can also facilitate the Schaffer collateral inputs via the disinhibition of proximal dendrites [ 33 ]. Furthermore, OLM cells can be activated by cholinergic and glutamatergic inputs from the medial septum (MS) [ 21 , 34 ] and inhibited by GABAergic inputs from the brainstem nucleus incertus [ 24 ], the latter of which proved a crucial component of memory formation in the hippocampus. These subcortical inputs can control which OLM cells can be activated during the memory formation and thus can indirectly determine the number of active CA1 PCs participating in the formation of a memory engram. Although the developmental origin of OLM cells was previously unclear, single-cell RNA sequencing data showed a very high homogeneity within the OLM cell population [ 35 ]. However, the expression of certain genes was detected only in subsets of OLM cells, which may suggest dynamically regulated temporal or permanent differences between OLM neurons [ 25 , 35 ].

The dorsal hippocampus is essential for the formation of contextual memory and the hippocampal somatostatin-positive interneurons proved particularly important in the selection of engram cell assemblies responsible for the formation and recall of memories [ 8 , 19 – 24 ]. In the CA1 region of the dorsal hippocampus, somatostatin is expressed in several types of interneurons [ 8 , 25 ]. All of these innervate, the dendrites of pyramidal cells (PCs) locally but their axonal arbors show preferences for different layers and co-align with different glutamatergic input pathways [ 8 , 26 ].

A diverse population of GABAergic interneurons controls the signal integration and action potential firing of principal cells in the cerebral cortex [ 1 – 5 ]. A large group, 12% to 30% (depending on brain area) of these interneurons expresses the neuropeptide somatostatin (SOM) [ 6 , 7 ] and targets predominantly dendrites of local principal cells [ 8 ]. Somatostatin is a small peptide that localizes in dense-core axonal vesicles of SOM-positive neurons and acts as a co-transmitter modulating neuronal activity via both pre- and postsynaptic mechanisms [ 9 ]. Interestingly, SOM-positive interneurons play an indispensable role in memory processes and learning in all brain areas where they occur [ 10 ]. Two features of SOM-positive cells might explain this phenomenon: (i) with the location of their output inhibitory synapses on dendrites they are able to finely modulate the effect of glutamatergic inputs to principal cells [ 11 ]; and (ii) their input synapses show plasticity which can be important in long-lasting changes of the neuronal networks [ 12 – 18 ].

Neurolucida reconstructions of cells from 10–17, 60 μm thick sections. All dendrites were in str. oriens and alveus. OLM cells are shown in different shades of purple. HS cells are shown in shades of green. The dendrites mostly arborized in a plane that was parallel with the border of the str. oriens and alveus. Here, all neurons are illustrated from a view perpendicular to that plane. These full reconstructions were the foundations ( Fig 1G, 1L and 1M ) of the computational models created in the modelling part of this study ( Fig 1N and 1V ). Dendrite segment analysis of the cells can be found in S15 Data . Sholl analysis of the cells are shown in S3 Fig and S20 Data . Three-dimensional Neurolucida reconstructions and dendrograms of the cells are presented in S21 and S22 Data files, respectively. HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare.

Dendritic trees of 4 OLM and 4 HS cells were completely reconstructed from series of light microscopic sections of optimally fixed brains in 3D using Neurolucida ( Fig 6 ). Reconstructions were adjusted for tissue shrinkage/dilatation that might have occurred during tissue processing (see correction factors in Methods ). Both cell types had a large dendritic tree restricted to str. oriens/alveus. Their dendrites extended over 10 to 17, 60 μm thick sections. The total dendritic length, the total number of nodes and endings, and the number of first-order dendrites were not different between the 2 populations ( Table 1 ). Sholl analysis was carried out to test whether there is a difference between OLM and HS cells in the number of dendrites intersecting spheres, the length, surface area, volume and average diameter of dendritic segments, the number of dendritic branching points, and endings in spheric shells ( S4 Fig ). This analysis did not reveal any significant differences, except that HS cells had branches in shells further than 500 μm, while the dendrites of the OLM cells did not reach that far from their somata.

We estimated the proportions of OLM and HS cells within somatostatin-positive interneurons in CA1 str. oriens using immunochemistry. In mice with the most effective tracer-labeling, minimum 49.1%, 42%, and 37.9% of somatostatin-positive cells were OLM cells (out of 1036 SOM-positive cell from 3 mice). Since the expression of the Chrna2 gene is not detectable in every anatomically identified OLM cell [ 62 ], and the AAV virus we used for identification in this study does not necessarily infect all Cre-positive cells at the injection site, this number should strictly be considered as an absolute minimum value. We found that minimum 9.5%, 8.1%, and 7.5% of SOM-positive cells were HS cells (out of 1,276 SOM-positive cells from 3 mice). The latter should also be considered as an absolute minimum proportion, as it is not possible to label all the cells projecting to the MS with retrograde tracers.

We also investigated the morphologies of OLM cell axons to further facilitate their functional modeling ( Fig 5B–5K and S7 Data ). Axon terminals (en passant boutons) of these neurons were barely distinguishable in thickness from the intervaricose axon segments. We found that OLM cell axons establish 4 synapses per 10 μm axons, whereas HS cell axons establish 4.5 synapses per 10 μm axons. The sizes of output synapses of OLM cells were significantly larger (almost 2-fold difference) than those of HS cell axons ( Table 1 ). OLM and HS cells had 2 and 2.7 pieces of mitochondria per 10 μm axon, respectively. The sizes of individual mitochondria in OLM axons were also significantly larger (2-fold difference) than those of HS cells ( Table 1 ).

The 3 reconstructed BDA-labeled HS cells had local axonal branches that arborized mostly in str. oriens and radiatum and did not enter str. lacunosum-moleculare (Figs 5E and S1 ). Out of 95 HS (n = 3 cells from 2 mice) synapses 73% and 23% targeted PC shafts and spines, respectively. These PC spines always received an unlabeled asymmetric input as well, and 2% of the synapses targeted interneuronal dendritic shafts; 2% of the targets could not be classified.

BDA injection into the MS revealed only a few HS cells with extensive labeling of their local axon collaterals in the mouse CA1 area. However, anterograde filling of the hippocampal axon collaterals of septo-hippocampal neurons was also unavoidable. Therefore, to make sure that we investigate the target selectivity of only HS axonal segments in CA1, we partially reconstructed the local axonal arbors of BDA-labeled HS cells using Neurolucida (MicroBrightField). We analyzed these segments using correlative light and electron microscopy (Figs 5E–5K and S1 ).

Postsynaptic targets of axons were followed in the serial images and were identified as described previously [ 61 ]. Briefly, spiny dendritic shafts without type 1 synaptic inputs on the shaft were considered PC dendritic shafts, whereas dendritic shafts that received type I inputs were identified as interneurons. Dendritic spines were identified by their size and specific shape emerging from a dendrite. Out of 136 OLM (from 2 mice) synapses, 78% and 19% targeted PC shafts and spines, respectively. These PC spines always received an unlabeled asymmetric input as well, and 1% of the synapses targeted interneuronal dendritic shafts; 2% of the targets could not be classified.

For the explanation of the workflow, see Fig 1I and 1J . (A) Light microscopic image of the sampled area used for OLM cell axon reconstruction (dashed line) in the str. lacunosum-moleculare of the CA1 area in a Cre-dependent viral tracer-injected Chrna2-Cre mouse. Cell bodies of labeled OLM cells in the str. oriens are indicated by white arrowheads. (B) 3D scanning electron microscopic reconstruction of 2 OLM axonal segments (brown and black) and one of their common target pyramidal dendrites (turquoise). Postsynaptic targets of other synapses of the reconstructed axons are indicated by arrowheads (red: PC dendritic shaft, yellow: spine). (C, D) Electron micrographs of the synapses (black arrowheads) with the reconstructed pyramidal dendrite (Pd, turquoise). S labels spines. Immunogold labeling for gephyrin (black particles) is visible at the postsynaptic sites of the synapses. (E) Neurolucida reconstruction and (F) light microscopic image of the same BDA-labeled HS cell. The complete dendritic tree (green) and partial axonal arbor (black) of this cell were reconstructed. Inset shows the location of the cell in the hippocampus. (G) Light microscopic image of the axonal segment indicated with dashed circle in E. Postsynaptic targets of synapses are indicated on the correlated light microscopic image in G, as above. (H) The same axonal segment (black) was serially sectioned and reconstructed in 3D using an array scanning electron microscope. Two of the postsynaptic PC dendrites (Pd 1, orange and Pd 2, blue) were also reconstructed and shown in H. (I–K) Electron micrographs of synapses (arrowheads) of the labeled HS cell axon with the reconstructed PC dendrites (orange and blue). The “blue” dendrite receives 2 synapses (J and K) from different branches of the reconstructed axon. Sections for HS cell axon reconstruction were not immunostained against gephyrin (see Fig 1 and Methods ). (L) Individual synapses of OLM axons are significantly larger than those of HS cells. (M) Individual mitochondria are significantly larger in OLM cell axons than in HS cell axons. (N) Proportions of different postsynaptic targets in CA1 of OLM and HS cells. Medians, interquartile ranges, and p-values for L and M can be found in Table 1 . Raw data are presented in S7 Data . Scale bar in K applies to all electron microscopic images. HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare; PC, pyramidal cell.

The target-selectivity of OLM and HS cells has been investigated in rats [ 29 , 36 , 44 ]. Here, we investigated it in mice. Due to the high number of strongly labeled OLM cells, a very dense axonal labeling was found in str. lacunosum-moleculare ( Fig 5A ). Using scanning electron microscopic array tomography, serially sectioned, randomly selected, tracer-labeled axonal segments were 3-dimensionally reconstructed ( Fig 5B ).

Glutamatergic synapses of HS cells were significantly larger on proximal dendrites than on somata ( Table 2 ). This difference was much smaller and less significant in the case of OLM cells ( Table 2 ). GABAergic synapses were also larger on the dendrites of HS cells than their somatic synapses ( Table 1 ). We found an opposite difference in OLM cells: GABAergic synapses were larger on the somata than on the dendrites ( Table 2 ).

We then compared somatic and dendritic synapses. Because dendritic inputs of OLM an HS cells correlated with somatic distance, we compared the synapses of only the proximal dendrites (0 to 100 μm) with the somatic ones.

(A) Whole somata of tracer-labeled cells were reconstructed to estimate the total somatic surfaces and volumes of OLM and HS cells (OLM: 01–03, HS: 08–10). Partial segments of somata (OLM: 06–07, HS: 11–13) were sampled to estimate the parameters of p-glutamatergic (red) and p-GABAergic (blue) synaptic inputs. The dendritic membranes (yellow) were made partially transparent to reveal the nuclei (whole somata) and synapses on the other side of the somata. (B–E) Representative sections of the reconstructed somatic membranes of an OLM (B, D) and an HS (C, E) cell. (B, C) Artificially colored scanning electron micrographs of the DAB–labeled cell bodies (yellow) of an OLM (B) and an HS cell (C). Preembedding immunogold labeling of gephyrin (black particles, arrow) was used to distinguish between GABAergic and glutamatergic synapses. Red and blue arrowheads label the edges of p-glutamatergic (red) and p-GABAergic (blue) synapses, respectively. Mitochondria are colored green in the electron micrographs. (D, E) Parts of the 3D reconstructions of the cells and their input synapses. Sections from B and C are indicated by gray planes. Also see Tables 1 and 2 and S6 Data . EM, electron microscopic; HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare.

Somata of OLM and HS cells were reconstructed and analyzed similarly to the dendrites. The total somatic surfaces and volumes of OLM and HS cells were similar (n = 3 OLM and 3 HS cells; Fig 4 and Table 1 and S6 Data ). Somatic inputs were reconstructed from other partially (about 25% to 60%) reconstructed somata (n = 4 OLM and 4 HS cell bodies, Fig 4 ). OLM and HS somata received many p-glutamatergic inputs and about 8% to 39.7% p-GABAergic inputs ( S6 Data ). Using the total somatic surface and the surface density of input synapses, we estimated that an OLM cell soma receives about 173 glutamatergic and 21 GABAergic inputs, whereas an HS cell soma receives 414 glutamatergic and 147 GABAergic inputs.

Although data suggested a similar trend for GABAergic inputs ( Fig 3G ), the same analysis revealed significant correlations neither for the distance, nor for the order of the dendritic segments ( Table 3 ). This may have been due to the smaller numbers of p-GABAergic synapses (284 and 324 inputs on OLM and HS cells, respectively).

Although the size of individual synapses varied to a very large extent, we found a strong correlation between the somatic distances and the sizes of the synapses. The sizes of p-glutamatergic inputs increased on average towards more distal dendrites on OLM cells, whereas synaptic size decreased with increasing distance from the soma on HS cells ( Fig 3F and Table 3 ). Putative glutamatergic inputs of proximal (0 to 100 μm) OLM dendrites were significantly smaller (median: 0.125 versus 0.164 μm 2 ), whereas distal (farther than 250 μm) dendrites were significantly larger (median: 0.153 versus 0.129 μm 2 ) than those on HS cells ( Table 1 ). Similar significant correlations were found between the size of p-glutamatergic synapses and the order of the dendritic segments ( Table 3 ).

We reconstructed 1,765 and 1,874 p-glutamatergic inputs on OLM and HS dendrites, respectively, and analyzed the correlation between their size and their distance from the soma (somatic distance). The somatic distance of all synaptic inputs on the same dendritic segment was considered identical and equal to the somatic distance of the dendritic segment. We used this simplification because the median dendritic segment length was very small (OLM: 15.7, HS: 11.1 μm) compared to the up to 432 μm long dendrites examined. The somatic distance of the dendritic segment was measured along the dendrite from the soma to the starting point of the given dendritic segment.

We found no correlation between the ratio of p-GABAergic inputs (out of the sum of all glutamatergic and GABAergic inputs) and either their distance from the soma or the dendritic orders of their OLM or HS cells ( Table 3 ). The median ratio of p-GABAergic inputs (on dendritic segments with at least 10 synaptic inputs) was not different (14.3% for OLM and 15.4% for HS cells; Fig 3D and Table 1 ). Nevertheless, because HS cells receive denser synaptic inputs they received a significantly larger number of p-GABAergic and p-glutamatergic inputs (see above).

Mitochondria occupied a significantly larger proportion of the cytosol in the branching points compared to the dendritic shafts ( Table 2 ). This is likely due to the fact that branching points are naturally thicker in the dendrites; therefore, they more likely had enough space for mitochondria than thinner dendrites.

Thicker dendrites (with larger median cross-sectional area) were occupied by proportionally larger mitochondria for both OLM and HS cells. The correlation was significant ( Table 3 ). This mitochondrial ratio was similar along the proximal to distal dendrites; therefore, we found no correlation between mitochondrial volume density (%) and the somatic distance or the order of the dendritic segments ( Table 3 ).

OLM and HS cells had sparsely distributed spines on their dendrites. On the proximal dendrites (dendrites closer than 100 μm from the soma), only 6% and 4.5% of the excitatory and 2.7% and 0% of the inhibitory inputs targeted the spines of OLM and HS cells, respectively. More distally (>100 μm), spine synapses became more frequent. On distal dendrites, 23.9% and 12.9% of excitatory synaptic inputs and 16.6% and 7.8% of inhibitory synaptic inputs innervated spines of OLM and HS cells, respectively.

We compared the number of synaptic inputs of OLM and HS cells for unit surface area (surface density of synaptic inputs). We found that the surface density of synaptic inputs on HS dendrites was significantly higher than that for the OLM cells. OLM cell received 0.522 p-glutamatergic and 0.075 p-GABAergic synapses/μm 2 , whereas HS cells were innervated by 0.763 p-glutamatergic and 0.135 p-GABAergic synapses/μm 2 (medians; for interquartile ranges, see Table 1 ). Consequently, synaptic inputs proportionally covered significantly larger areas on HS dendrites ( Fig 3D and Table 1 ). These differences were detected for both p-glutamatergic and p-GABAergic synaptic inputs ( Table 1 ).

Correlations between the linear densities and distance or branching order may partly be explained by the fact that the median thickness of the dendrites decreases away from the soma, towards higher order dendrites (although with very high variability).

The higher the branching order of the OLM dendrites were, the smaller linear density of p-glutamatergic inputs they received ( Table 3 ), whereas the linear density of p-GABAergic inputs of HS cells showed a similar correlation with the branching order ( Table 3 ).

Because dendrites became thinner more distally from the soma, the linear density of p-glutamatergic dendritic synaptic inputs decreased significantly with the distance from the soma ( Fig 3 and Table 3 ). The linear density of p-GABAergic synaptic inputs did not correlate with the distance from the soma in OLM cells but showed a weaker correlation with distance in HS cells ( Fig 3G and Table 3 ).

After analyzing the number of synaptic inputs per dendritic length (linear density of synapses) on 47 OLM and 35 HS cell dendritic segments, we found that thinner dendrites (with smaller cross-sectional area), as expected, received fewer inputs per micrometer dendrite in both cell types. This was significant for both p-glutamatergic and p-GABAergic synaptic inputs of both cell types (Figs 2 and 3E and Tables 2 and 3 , S2 and S3 Data files).

Morphologically detailed models are typically represented as interconnected cylinders or truncated cones, and these segments are characterized by their length and diameter (or the diameters at the 2 ends in the case of truncated cones). To aid the construction of such detailed model neurons, we created equivalent cylinder models for each of our electron microscopic samples of OLM and HS cell dendrites (with the exception of branch points). We calculated the diameter of the cylinder in 3 different ways: each cylinder had the same length as the 3D reconstructed segment, but one version had the same volume as the original segment, a second version had the same external membrane surface, while the third one had the same average perimeter. These methods yielded different, but highly correlated estimates of the equivalent dendritic diameter ( S17 and S18 Data files), and further confirmed our previous conclusions regarding the high variability of dendritic thickness.

The general assumption about dendritic morphology is that the thickness of a dendrite decreases towards distal dendrites and towards higher order dendrites. We confirmed this view in both OLM and HS cells. However, we also found a very high variability of thickness (cross-sectional areas) within the same order of dendrites and at different distances both in our light microscopic (Neurolucida drawings) and electron microscopic (3D reconstructions) samples ( Fig 2 and S2 and S3 Data files).

Here, we defined dendritic orders in the following way: first-order dendrites originated from the somata, whereas the branching order of all offspring dendrites was one more than the order of the dendrite that they originated from (for example, all offspring of a second-order dendrite were third-order dendrites, and so on). However, although this ordering is unbiased, dendrites with the same order could be very different, partly because dendrites with the same order could have very different number of offspring dendrites that they supported. For example, we found that a first-order proximal dendrite could have the same thickness as a distal 10th-order dendrite. These data can be found in Table 1 and S2 and S3 Data files. Branching points (n = 11 OLM and 6 HS) of the dendritic trees were analyzed separately ( Fig 2 and S4 and S5 Data files).

The following parameters were measured for each dendritic segment ( S2 – S4 Data ): distance from the soma, length, dendritic order, volume with and without mitochondria, the volume of individual mitochondria, surface area, cross-sectional areas and perimeters of the dendrite along the dendritic segments, number of p-glutamatergic/p-GABAergic inputs, size of individual p-glutamatergic/p-GABAergic inputs, all surfaces of p-glutamatergic/p-GABAergic inputs including synapses on somata, dendritic shafts and spines. Morphological parameters normalized by surface area and length were also calculated.

(A–C) Comparison of a typical OLM- and a typical HS dendritic segment of similar diameters. (A, B) Artificially colored scanning electron microscopic images of the DAB–labeled dendrites (yellow) of an OLM (A) and an HS cell (B). Preembedding immunogold labeling of gephyrin (black particles, black arrows) was used to distinguish between GABAergic and glutamatergic synapses. Red and blue arrowheads label the edges of gephyrin-negative and gephyrin-positive synapses of p-glutamatergic (red) and p-GABAergic (blue) boutons, respectively. Mitochondria are colored green in the electron micrographs. (C) 3D reconstructions of the dendritic segments shown in A and B. The section plane of A and B are indicated by gray in C. The OLM and HS dendritic segments in C are also shown in Fig 2 (dendrite 12–15 and dendrite 78, respectively). Putative glutamatergic and p-GABAergic synapses are shown in red and blue, respectively. The dendritic membranes (white) were made partially transparent to reveal mitochondria (green) inside the dendrites and synapses on the other side of the dendrites. Both cell types have irregularly shaped dendritic shafts that are sparsely spiny. (D) Ultrastructural parameters of reconstructed OLM and HS dendritic segments show characteristic differences. Medians (square) and interquartile ranges (boxes), minimum and maximum values (whiskers) of measured parameters are shown. Asterisks indicate significant difference, n.d: no significant difference was found. See medians, interquartile ranges, and p-values in Table 1 and raw data in S2 – S5 Data files. The portions of surfaces covered by p-glutamatergic and p-GABAergic synapses are significantly larger on HS cell dendrites than on OLM cell dendrites. There is no difference between the ratios of GABAergic inputs on the dendrites of the 2 populations of interneurons. Mitochondria of HS cell dendrites occupy significantly larger volume inside the dendrites than those of OLM cells. (E) The linear density of p-glutamatergic (filled circles) and p-GABAergic (open triangles) synapses show positive correlation (Spearman rank) with the thickness of the dendrite both in the case of OLM (purple) and HS cells (green). However, the linear densities are markedly different. (F) The sizes of individual p-glutamatergic synapses varied to a large extent, and they showed a significant correlation with the distance from the soma along the dendritic tree in both cell groups. Input synapses of OLM cells became larger, whereas inputs of HS cells became smaller with the distance from the somata. (G) The p-GABAergic inputs showed similar trends without being statistically significant. For E–G, p-values can be found in Table 3 . Raw morphological parameters are presented in S2 – S5 Data files. HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare.

Proximal dendrites of OLM and HS cells were mostly smooth, whereas their distal dendrites carried sparsely distributed spine-like structures. On OLM cells, some of these protrusions were very long. The dendritic shafts received many inputs, most of which formed p-glutamatergic, type I (asymmetric) synapses (Figs 2 and 3 ). They were innervated by fewer p-GABAergic, type II (symmetric) inputs (Figs 2 and 3 ). Spines were different from PC spines as they did not have a typical head and were often innervated by more than one synaptic input. Here, sometimes, we refer to synapses established on OLM and HS cell dendrites as “inputs” to clearly differentiate them from synapses that the axons of these cells establish on other cells.

Because the reconstructions of the total dendritic arbors of all cells examined here would have taken more than a decade, we took representative samples from these dendritic arbors, and measured their properties at different points of the dendritic trees of OLM and HS cells. First, dendritic trees were reconstructed in 3D at the light microscopic level using the Neurolucida software. Then, we selected dendritic segments randomly so that dendritic segments at different distances from the somata are represented relatively evenly from the whole dendritic tree. Dendritic segments from every 100 micrometers from the somata were represented in the samples. We also ensured that every dendritic order was represented by several segments ( Fig 2 ). The order of the selected segments and the distance of their starting point from the somata were determined using the dendrograms created by the Neurolucida Explorer software. Then, these sections were re-embedded and serially sectioned for electron microscopic reconstruction, and 53 OLM cell dendritic segments (median length: 15.7 μm) and 37 HS cell dendritic segments (median length: 11.1 μm) were fully reconstructed. Plasma membranes of the dendrites, outer membranes of mitochondria, and synaptic membranes were reconstructed in 3D and their dimensions were measured.

We collected the following data groups: high-resolution local morphological parameters of dendritic, somatic, and mitochondrial membranes; parameters of dendritic and somatic inhibitory and excitatory input synapses that OLM and HS cells receive; data for axon terminals that OLM and HS cells establish; light microscopic reconstructions of total dendritic arbors of OLM and HS cells; detailed in vitro physiological parameters. We are presenting all raw data and their statistical comparisons in supporting data files ( S2 – S10 Data and S19 – S22 Data ). In addition, we also used these data to build neuronal models of these cells to reveal their basic computational features and the functional consequences of their differences ( Fig 1N, 1U and 1V ).

OLM and HS cells were first reconstructed in light microscope and representative samples were taken from somata and dendritic segments that were further processed for scanning electron microscopy (SEM), after which they were reconstructed from 70 nm thick section series.

To determine the synaptic, dendritic, and somatic parameters of OLM and HS cells that correspond to a live animal, we implemented a workflow that aims to be unbiased, representative and that compensates all distortions of the tissue processing procedures. Tissue shrinkage/dilatation was estimated during the whole process and corrected to collect data that represent in vivo parameters (see Methods ).

3D rendering of all reconstructed OLM dendritic (yellow or gray) and branching point (light blue) segments (n = 53 + 11, respectively), and all HS dendritic (yellow or gray) and branching point (light blue) segments (n = 37 + 7, respectively). Besides the dendrites (yellow) with fully reconstructed synapses (n = 4,536), there were some dendrites (gray) without synaptic reconstructions. Before ultracutting, OLM and HS cells were reconstructed under a light microscope using Neurolucida to determine the dendritic order and the distance of the sampled dendritic segments from the soma along the dendritic tree. The dendritic segments and branching points of OLM and HS cells are illustrated in groups in the upper and lower rows, respectively. Dendritic segments are arranged based on their distance from the soma. The dendritic membranes were made partially transparent to reveal their mitochondria (green) inside the dendrites and the input p-glutamatergic (red) excitatory (n = 3,876) and p-GABAergic (blue) inhibitory (n = 660) synapses on both sides of the dendrites. They were imaged in SEM and reconstructed from serial ultrathin sections immunostained for the GABAergic synapse marker gephyrin. Thickness of the dendrites varies greatly even within a given distance range. The surfaces of the HS cell dendrites are more densely covered by synapses than those of OLM cells. All raw morphological parameters are presented in S1 – S5 Data files. EM, electron microscopic; HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare.

GABAergic synaptic surfaces were identified using the reliable preembedding immunogold labeling against gephyrin [ 57 , 58 ], a major postsynaptic protein in GABAergic synapses ( Fig 2A and 2B ). Based on data in the literature gephyrin labels, only GABAergic synapses in cortical areas. Synapses that were not labeled for gephyrin were considered putative glutamatergic synapses, because (i) the ratio of false negative synapses are considered negligible because of the reliable gephyrin labeling that was tested on several sections of the same synapse; and (ii) although a negligible amount of non-GABAergic synapses may have been established by monoaminergic fibers, most of those fibers do not typically form synapses [ 59 , 60 ] in the CA1 stratum oriens. Therefore, we considered the labeled and unlabeled synapses putative GABAergic (p-GABAergic) and putative glutamatergic (p-glutamatergic) synapses, respectively.

After cell type-specific labeling of OLM and HS cells using different tracing methods (A, B), samples went through different tissue processing protocols optimized for electron microscopic analysis of cells/whole cell reconstruction/colocalization experiments at light microscopic level or in vitro physiological recordings (C, G, O, P, R). Purple and green colors indicate OLM and HS cells, respectively. Black arrows indicate the workflow of sample preparation. Purple and green arrows indicate processes in which only OLM or HS cells were involved. Yellow boxes show the methods, work packages, which led to 3 major data sets in this study: the first data set contained all the representative dendrite type-specific data, including dendritic and somatic morphologies and synapses (F), second data set represented complete dendritic morphologies (M), the third data set represented all physiological properties of OLM and HS cells (U). After the combined analyses of the morphological data (N), it was combined with the physiological data (U) to create complete cellular models (V). For further details see Methods . HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare.

First, we labeled OLM and HS cells specifically. The workflow is illustrated in Fig 1 . It has been repeatedly demonstrated that nicotinic acetylcholine receptor alpha 2 subunit (Chrna2) positive neurons in the dorsal hippocampus CA1 area, in stratum oriens are OLM cells [ 33 , 55 , 56 ]. Here, we labeled OLM cells specifically, using an injection of a Cre-dependent fluorescent reporter adeno-associated virus (AAV) into a Chrna2-Cre mouse. Additionally, because HS cells (by definition) project to the MS, they were labeled retrogradely from the MS. Depending on the type of data we collected, we used different retrograde tracer techniques including the injection of biotinylated dextran amine (BDA) or FluoSpheres into wild-type (WT) mice, or the injection of a Cre-dependent retrograde tracer into SOM-Cre mice ( Fig 1A and 1B ). After tracers labeled the target cell types, we specifically selected the type of tissue processing most suitable for collecting a given type of data set ( Fig 1C, 1G, 1P and 1R ).

Second, we investigated whether the spike output of the model neurons could be modulated by synaptic input whose rate varied in a sinusoidal manner at theta frequency (7 Hz). We stimulated 5% of the excitatory synapses randomly distributed on the whole dendritic arbor, activating synapses independently at a rate which varied between 1 and 21 Hz during the theta cycle. This was meant to represent inputs from local pyramidal neurons during behavioral states characterized by theta oscillations in the population activity. Then, we calculated the spiking rate of our active OLM and HS model neurons at each phase of the sinusoid to evaluate theta modulation ( S11 Fig ). We used Rayleigh’s Test of Uniformity to detect significant phase dependence and measured the mean vector length to determine the depth of modulation. Each of the 4 HS cells showed significant theta modulation (p < 0.001 according to Rayleigh’s test; mean vector lengths: 0.467 to 0.642) and so did 3 of the 4 OLM cells (p < 0.001; mean vector lengths: 0.2 to 0.676). The fourth OLM cell fired only a few spikes at this level of synaptic stimulation; however, when we repeated the simulation with 15% of excitatory synapses activated, this model also showed significant theta modulation (p < 0.001; mean vector length: 0.36). Overall, no clear difference was found in the degree of modulation by theta frequency inputs between the 2 cell types.

First, we determined how the output firing rate of the model neurons depended on the number and location of active synaptic inputs. In the simplest case, we activated a variable number of excitatory synaptic inputs selected randomly according to the predicted distributions of synapses on the dendrites of individual reconstructed neurons. Synaptic inputs were activated independently and randomly at a constant rate of 10 Hz. We found, as expected, that a certain number of excitatory synapses had to be active for the neurons to start generating action potentials, and the firing rate increased monotonically as the number of active inputs was increased further ( S10 Fig ). We also repeated the same simulation experiments targeting only proximal (0 to 100 μm from the soma) or distal (300 to 400 μm from the soma) synapses. The qualitative picture was similar to the baseline case in both of these situations, but we found that a smaller number of proximal synapses was required to drive spiking in most cells compared to distal or widely distributed ones ( S10 Fig ). Finally, we used the estimates of synapse numbers that we had derived by combining the morphologies with the synaptic densities measured by EM to plot the output spike rates as a function of the percentage (rather than the absolute number) of activated synapses in the given cell and distance range ( S10D–S10F Fig ). These percentages are expected to provide a better measure of the level of activity in the relevant presynaptic populations, independent of the morphology of the target neuron. There were substantial differences between the individual models in the threshold percentage (typically less than 5%) and in the steepness of the activation curve, but there was no systematic difference between models of OLM and HS cells in these characteristics.

Our active models of OLM and HS neurons, combined with our data on synaptic distributions, also allow the investigation of spiking responses in the 2 cell types in response to various spatial and temporal patterns of synaptic input. While a complete analysis of all the relevant scenarios is beyond the scope of the current study, and may also require further experimental data and the improvement of the active neuronal models, we demonstrate the feasibility of this approach by showing the results of 2 sets of simulations.

The best-fitting parameters for all models are listed in S16 Data . The resulting models were then validated using the Somatic Features Test of HippoUnit ( S8 and S9 Figs), including features that were not used during parameter optimization. The results of this validation indicated that the active models captured the physiological behavior of OLM cells reasonably well (335 to 361 out of 447, i.e., 75% to 81% of features were within 2 standard deviations of the experimental mean). Considering only the features that describe the intensity and pattern of spiking (these include spike numbers, spike delays, and interspike intervals), the match between the model output and the experimental results was even better (90 to 100 out of 100 features within 2 standard deviations). The responses of HS cell models matched the corresponding experimental data less accurately (237 to 285/505, or 47% to 56% of all features and 62 to 69/106, or 58% to 65% of spike pattern features within 2 standard deviations). These results suggest that the set of biophysical mechanisms that determine the physiological behavior of the 2 cell types may not be identical, and achieving a more accurate description of HS cell physiology will require direct measurements of voltage-gated channel properties and distributions in this cell type.

Nevertheless, our in vitro physiological data from the 2 cell types indicated that they showed relatively similar behavior (at least in terms of the somatic voltage response to somatic step current injections), which encouraged us to build and simulate a set of prototype active models of both cell types based on our data using the same set of biophysical mechanisms. We used the same reconstructed morphologies as above (4 OLM and 4 HS cells), and took the models of ionic conductances and the basic patterns of their spatial distributions from a previously published detailed model of OLM cells [ 68 ]. We tuned the maximal conductance values of the voltage-gated channels and leak channels (14 parameters in total) via Neuroptimus using a feature-based error function as before, but this time including features extracted from responses to both depolarizing and hyperpolarizing current injections recorded in the appropriate cell type.

As presented above, our anatomical and physiological data allowed us to construct morphologically detailed, essentially passive models of OLM and HS cells, and to investigate dendritic signal propagation and synaptic integration in the subthreshold voltage range. A natural next question is how action potentials are generated in these neurons, and how their spike output is influenced by the spatial and temporal pattern of the synaptic inputs that they receive. These questions can be addressed effectively in fully active, morphologically and biophysically detailed neuronal models. The construction of such models is also facilitated by our anatomical and physiological data; however, building such an active model also requires knowledge of the identity, quantity, and spatial distribution of voltage-gated ion channels in the cell membrane. Some (but certainly not all) of the required parameters have been measured in OLM cells [ 66 – 68 ], and this has enabled the construction of detailed multi-compartmental models of OLM cells [ 68 – 72 ], while there is practically no information available about voltage-gated channels in HS neurons.

Finally, we investigated whether taking into account the presence of mitochondria has a significant effect on signal propagation in the models. To do this, we first observe that, from the perspective of electrical signal propagation, the main effect of the presence of mitochondria in dendrites is to decrease the effective cross section that is available for axial current flow. In our models, we can compensate for this effect by increasing the specific axial resistance. More specifically, we multiplied the specific axial resistance by the ratio of the total cross-section of the dendrite and the cross-section excluding mitochondria (proportion of mitochondria in Table 1 and S2 and S3 Data files). Since we did not have data on the exact location of mitochondria in our reconstructed morphologies, we applied the correction based on a modified axial resistance uniformly in the dendrites, using the ratio of the total dendritic volume and the volume without mitochondria in our samples of EM data from the 2 cell types. Using this method, we did not see any significant change in the behavior of our model neurons, suggesting that mitochondria do not have a significant effect on signal propagation in the dendrites of OLM and HS neurons.

Using these calculated effective values of the tonic membrane conductance (including the contributions of synaptic conductances as well as the leak conductance), we repeated our in silico experiments that measured the somatic effects of activating a single (excitatory or inhibitory) synapse at different locations in the dendritic tree ( Fig 9C and 9D ). The results showed that the inclusion of tonic background activity fundamentally changed the effects of single synaptic inputs in both cell types. In the HCS, even proximal synapses evoked at least one order of magnitude smaller PSPs than in the silent state, and the magnitude of the predicted response became almost negligible beyond approximately 200 micrometers from the soma, so that the majority of single synapses evoked very small somatic voltage responses. In addition, because of the larger attenuation of synaptic signals in HS cells due to their higher tonic membrane conductance, single somatic PSPs at a given distance from the soma were smaller in HS cells than in OLM cells in the HCS, in contrast to what we observed in the silent state. On the other hand, the distributions of PSP amplitudes in the 2 cell types became strongly overlapping, indicating that the effects of single synaptic inputs could be quite similar on average, although the strongest synapses were predicted to be more potent in OLM cells in this state.

Another consequence of the different contributions of excitatory and inhibitory synapses to the total membrane conductance in the different cell types is a difference in the local equilibrium potential of the cell membrane. This can be calculated as the mean of the reversal potentials of all ionic currents weighted by their respective conductances. We calculated the local equilibrium potential at every segment of the models and observed a clear difference between the 2 cell types. The proximal dendrites of the OLM cells had more depolarized equilibrium potentials than the proximal dendrites of the HS cells but became gradually more hyperpolarized further away from the soma, while the dendrites of the HS cells had a more uniform distribution of dendritic equilibrium potentials with slightly more depolarized values further away from the soma ( S7 Fig ). These differences affect the steady-state membrane potential distribution within the cells and the driving force associated with excitatory and inhibitory synaptic inputs at different dendritic locations.

Fig 8E shows the contributions of the excitatory synaptic, inhibitory synaptic, and leak conductance to the total conductance density (excluding Ih, which is voltage dependent) in various parts of our model neurons. Under the conditions of the HCS as defined above, excitatory and inhibitory synapses are predicted to make contributions of comparable magnitude to the total membrane conductance, and both of these are several orders of magnitude larger than the contribution of the leak conductance (and Ih). Different models of the same cell type showed very similar conductance distributions. The total conductance was significantly larger in HS cells than in OLM cells, and it was almost constant at all locations in OLM cells, while it decreased substantially with increasing distance from the soma in HS neurons. The difference between the 2 cell types was due to the different dependence of synapse density on cellular parameters (such as distance from the soma).

The density of synapses in any given section of our model neurons can be estimated from our EM data as described above ( Fig 8 ). The peak conductance and the decay time constant (tau) for excitatory and inhibitory synapses were based on published values for OLM cells, and parameters for HS cells were assumed to be the same, except for a 15% larger value of the peak conductance of inhibitory synapses due to their larger size as discussed above. Finally, the firing rates of excitatory and inhibitory presynaptic populations can vary considerably (e.g., in different brain states), so we will use a particular set of plausible values for demonstration purposes (3 Hz for excitatory and 30 Hz for inhibitory afferents).

To model the behavior of neurons in the HCS, one needs to take into account the total membrane conductance, including contributions from synaptic, leak, and voltage-gated conductances. This requires the calculation of the temporal average of the excitatory and inhibitory synaptic conductance per unit area of membrane everywhere in the dendritic tree [ 54 ]. For a specific type of input, this can be computed as the product of 3 quantities: the number of synapses per unit membrane area, the time integral of the conductance change evoked by the activation of a single synapse, and the average firing rate of the presynaptic neuronal population. If the synaptic conductance change is approximated well by a single decaying exponential function, the second term in this product is simply the maximum conductance change multiplied by the decay time constant. As a result, the mean contribution of a set of synaptic inputs to the total membrane conductance can be calculated as follows [ 54 ]:

Neuronal populations in vivo are continuously active in a brain state-dependent manner. As a result, individual neurons in the hippocampal network are constantly bombarded by many excitatory and inhibitory synaptic inputs, which lead to a sustained increase in the membrane conductance (and a possible shift in the equilibrium potential) in all parts of the dendritic tree. These effects can greatly influence the processing of synaptic inputs within the cell, pushing individual neurons and the whole network into what has been referred to as the “high-conductance state” (HCS) [ 54 , 65 ].

(A, B) (top) The peak amplitude of the EPSPs (A) and IPSPs (B) measured at the soma in response to synaptic inputs to segments at various distances from the somata of the 4 OLM and 4 HS cell models, simulated in the silent state. Each color represents 1 cell, and each circle is a different dendritic segment. In the bottom panels, we show the density distributions of the somatic PSP amplitudes. Panels C and D show the results of the same in silico experiment in the high-conductance state. HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare; PSP, postsynaptic potential.

We activated synapses with these properties, which were assumed to be uniform within a given cell type, at different dendritic locations in our model neurons, and measured the amplitude of the somatic postsynaptic potential (PSP) using the Post-synaptic Potential Attenuation Test of HippoUnit. We found that, under such isolated conditions, both excitatory and inhibitory synapses generated larger amplitude somatic PSPs in HS neurons than in OLM cells ( Fig 9A and 9B ). We also note that the amplitude of EPSPs and (especially) IPSPs showed only a moderate dependence on the distance of the synapse from the soma.

To reveal the functional consequences of the morphological differences between the 2 cell types that were indicated by our EM analysis, we examined the responses of our model neurons to synaptic inputs arriving at different locations in the dendritic arbor. We first considered the effects of single synapses activated in isolation; this corresponds to in vitro experimental conditions with a low level of neuronal activity in the hippocampal network, which will be referred to as the “silent state.” In the absence of relevant data regarding the properties of unitary synaptic currents in HS cells, we assumed that most parameters characterizing the primary effects of synaptic input (rise time, decay time, reversal potential) were the same in the 2 cell types ( S14 Data ) and corresponded to values measured in OLM cells [ 13 , 33 , 64 ]. However, as results from other cell types showed that the size of the synapse was strongly correlated with peak synaptic conductance (for a given type of synapse) [ 12 , 13 , 33 , 64 ] we took advantage of our EM data to estimate the relative strength of (excitatory and inhibitory) synapses in the 2 cell types. Since we found no significant difference in the size of excitatory synapses, we used the same maximal conductance value (calculated from measurements in OLM cells; [ 12 ]) in the 2 cell types. On the other hand, our EM data indicated that inhibitory synapses onto HS cell were, on average, 15% larger than those targeting OLM cells, and we therefore used a 15% larger value for the peak synaptic conductance in HS cells ( S14 Data ), which resulted in a 15% larger amplitude of (local) inhibitory synaptic currents compared to the experimentally measured value in OLM cells [ 13 ].

The best-fitting parameters for all models are listed in S13 Data . These covered similar ranges in the 2 cell types, although there was a tendency for higher conductance values in OLM cells compared to HS cells, especially for Ih. The resulting models were then validated using the Somatic Features Test of HippoUnit ( S5 and S6 Figs), including features that were not used during parameter optimization, but still restricted to hyperpolarizing current injections. This validation confirmed that our models provided a good description of the subthreshold behavior of OLM and HS neurons at the level of the somatic voltage response to local current inputs, as the overwhelming majority of the model features were within 2 standard deviations of the corresponding experimental mean value (79/81 in OLM cell models and 72/81 in HS cell models).

Next, we constructed morphologically detailed but biophysically simplified functional models of OLM and HS neurons based on our experimental data. As our aim was to capture the subthreshold behavior of these cells, we included only those mechanisms that substantially influence the electric behavior of neurons in this membrane potential range. Specifically, our models contained, in addition to the membrane capacitance, the axial resistance (whose value was fixed to 100 Ohm cm), and the passive leak conductance, a model of the hyperpolarization-activated current (Ih). The model of Ih was based on the experimental results of [ 63 ] regarding the properties of Ih in OLM cells; we assumed that Ih had similar characteristics in HS cells, with possible variations in the amount of Ih as well as in the relative contribution of fast and slow components of Ih activation dynamics. Overall, for each of our model cells, we considered 5 unknown parameters: the specific membrane capacitance (cm), the conductance (gpas) and reversal potential (epas) of the leak current, the maximal conductance (gmax_H) of Ih, and the relative contribution of the fast component to the dynamics of Ih (fastRatio_H). All of these parameters were assumed to be spatially uniform. These unknown parameters were fitted to match the electrophysiological features extracted from voltage responses to negative current steps in the whole-cell recordings described above.

(A) The total length of the dendrites in the morphological reconstructions. OLM = 5,917, 3,788, 6,816, 7,547 μm. HS = 7,409, 6,654, 8,068, 7,558 μm. (B, C) The total number of excitatory synapses (panel B: OLM = 7,062, 4,776, 7,311, 9,766. HS = 10,620, 10,330, 16,570, 15,243) and inhibitory synapses (panel C: OLM = 1,229, 801, 1,293, 1,913. HS = 2,172, 2,133, 3,375, 3,136) predicted for each model cell. In panels A–C, the circles represent the cells, and each cell is shown in the same color throughout the panels. (D) The number of excitatory and inhibitory synapses in each model as a function of the distance from the soma (bin size = 50 μm). See raw data in S23 Data . (E) The effective conductance densities in each segment of the model neurons in the high-conductance state (see main text). Color red marks the excitatory and blue the inhibitory current, black is the leak conductance and orange is the sum of the 3 (called “all currents” in this figure, although it excludes the voltage-gated current Ih, which is also implemented in the model). Each circle represents a segment of the model at different distances from the soma. HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare.

In addition, a larger proportion of these inputs targeted proximal versus distal parts of the dendritic tree in HS neurons (skewness values were strictly positive for the distributions of both excitatory and inhibitory synapses in HS cells, and they were significantly larger than the skewness values in OLM cells; T test, excitatory synapses: p = 0.0081, inhibitory synapses: p = 0.001) ( Fig 8 ).

The majority of inputs on both cell types arrived onto the dendritic tree, while only 3% and 4% of all excitatory inputs and 3% and 7% of all GABAergic inputs targeted the cell body of OLM and HS cells, respectively. Most dendritic synapses were located on the shaft: 77% of all excitatory and 83% of all inhibitory synapses in OLM cells and 85% of all excitatory and 87% of all inhibitory synapses in HS cells; 20% of all excitatory and 14% of all inhibitory synapses innervated dendritic spines in OLM cells and 11% of all excitatory and 5% of all inhibitory synapses were on dendritic spines in HS cells.

The predictive equations from multiple linear regression allowed us to estimate the density and, consequently, the number of excitatory and inhibitory synapses, in each segment of the fully reconstructed morphologies (4 OLM and 4 HS cells). Analyzing the resulting distributions and total predicted numbers of synapses for the 2 cell types, we estimated that, although the total length of the dendrites was similar ( Table 1 ), HS cells have a significantly larger number of both excitatory and inhibitory synaptic inputs compared to OLM cells. OLM cells receive 7,186 (median) p-glutamatergic synapses (interquartile ranges: 5,919 to 8,538) and 1,261 p-GABAergic synapses (interquartile ranges: 1,015 to 1,603), whereas HS cells are innervated by 12,931 (interquartile ranges: 10,475 to 15,906) p-glutamatergic synapses and 2,654 (interquartile ranges: 2,152 to 3,255) p-GABAergic synapses, and 14.9% and 17% of the synaptic inputs of OLM and HS cells are p-GABAergic, respectively.

As a first step, we computed the number and spatial distribution of excitatory and inhibitory synapses on OLM and HS cells by combining our light microscopy-based reconstructions with our EM data. To this end, we examined the relationship between the density of synapses and several potential predictors such as dendritic diameter, distance from the soma, and dendritic order using multiple linear regression. We found that, overall, the density (number per unit membrane area) of excitatory and inhibitory synapses in both cell types depended only weakly on each of these morphological variables, but the most reliable predictions could be made by considering the diameter of the dendrite and its distance from the soma (see Methods , Eq 1 ). There were also substantial differences in the prediction coefficients based on cell type (OLM versus HS) and synapse type (excitatory versus inhibitory) ( S12 Data ), consistent with the differences in average synapse density and in correlations with morphological features, as we described above ( Table 3 ).

The anatomical and electrophysiological data collected in our study provide essential information for the construction of detailed models of hippocampal neurons and networks (see Discussion ). As an initial demonstration of the utility of these data sets, we built and simulated a set of morphologically detailed models with a limited set of biophysical mechanisms to capture the subthreshold behavior of OLM and HS cells, focusing on the effects of excitatory and inhibitory synaptic input.

For the explanation of the in vitro experimental workflow, see Fig 1O and 1U . (A, D) Typical voltage responses of OLM and HS cells to depolarizing and hyperpolarizing current steps (amplitude: 160 pA). (B, C) Example voltage responses from an active OLM and an active HS cell model, corresponding to the same stimulation protocol used in the experiments. (E, F) Reconstructions of a biocytin labeled OLM (E) and an HS cell (F). Dendrites are shown in purple and green, whereas axon arbors are shown in black. Cells were pre-labeled for in vitro recordings by Cre-dependent tracer virus injected in the CA1 area of Chrna2-Cre mice (OLM) or Fluospheres injected into the medial septum (HS cells). HS, hippocampo-septal; OLM, oriens-lacunosum-moleculare.

We analyzed the membrane potential responses of 8 OLM and 8 HS cells (using a total of 46 measurements) to a series of current step injections during somatic whole-cell recordings in acute hippocampal slices ( Fig 7 , see Methods ). From each recording, we extracted a total of 267 electrophysiological features corresponding to several standardized current amplitudes. We found that several of these features had significantly different values in the 2 cell types (Mann–Whitney test with Bonferroni correction to account for multiple comparisons, p < 0.00018; Table 4 and S11 Data ). HS cells fired more rapidly at the beginning of large positive current injections, while OLM cells fired action potentials of larger amplitudes. However, we also found many features (including input resistance, the size of the hyperpolarizing sag response, the mean firing frequency, and action potential width) that were not significantly different between the 2 cell populations. More specifically, it is interesting to note that, although the spiking responses of the 2 cell types were clearly distinguishable, their subthreshold responses were quite similar.

Finally, the data collected in this study also provides essential constraints for data-driven models at the network level. In particular, our results on the densities and numbers of input and output synapses of OLM and HS cells provide information on the connectivity of these cell populations in the hippocampal network and can be used during the reconstruction and validation of the connectome of area CA1 and its afferent and efferent brain regions [ 77 , 105 , 106 ].

It is important to note that in the case of the active models and their responses to synaptic stimulation, the variance within a cell type was greater than the difference between cell types. By increasing the number of cells in each population along with building HS models that are closer to the experimental observations could emphasize the differences of the 2 interneuron types. For now, we may conclude that since all the cells within 1 class were fitted to the same electrophysiological data using the same conductances, morphological differences provide the most likely explanation of the variability of the cell models within the same cell type.

Including the responses to positive current injections in the physiological target data provides additional constraints for the construction of fully active spiking model neurons. However, the spiking behavior of neurons depends strongly on the amounts and exact biophysical properties of the voltage-gated channels in their cell membranes. While there are some experimental findings available on voltage-gated ionic currents in OLM cells [ 66 – 68 ], there are (to our knowledge) essentially no such data published for HS cells. Therefore, in our models we used the same types of conductances and spatial distributions for HS cells as for the OLM cells. We used the previously published OLM cell model of Lawrence and colleagues [ 68 ] as a starting point and fitted the maximal conductances of its channels separately for each morphology to match the features derived from responses to depolarizing and hyperpolarizing currents in the corresponding cell type. It became quite clear from the parameter fitting and even clearer after the validation process that the models constructed in this way cannot entirely describe the active properties of the HS cells while they provide a good fit to the experimental data from OLM cells. To create better models for HS interneurons, more data would be required about their ion channel composition, the properties and localization of their ionic currents, and more information about their dendritic behavior. We also found that the models of both cell types (but especially OLM cells) were better at describing the spike rates and patterns of real neurons than at capturing finer details of the voltage response (e.g., action potential shape). The good accuracy of the models at predicting spikes allowed us to investigate the effects of synaptic inputs on the spiking output of these cells and suggests that the models may be suitable for inclusion in biophysically detailed network models.

Our morphological reconstructions and electrophysiological recordings enabled us to construct morphologically detailed models of OLM and HS cells at 2 different levels of biophysical realism. One set of models was designed for the investigation of subthreshold phenomena and included only Ih as a voltage-gated ion channel. The other set of models aimed to capture both subthreshold and suprathreshold (spiking) responses and included a wide variety of voltage-dependent biophysical mechanisms. Our primary goal was to demonstrate how the various data sets resulting from our experiments can be combined to enhance the quality of data-driven models of specific cell types. However, even with the simpler passive models and restricting our attention to the subthreshold voltage range, we were able to make some interesting predictions. For example, our simulations of intracellular responses to synaptic inputs predicted that, although synapses in all parts of the dendritic tree had similar effects on the somatic membrane potential in isolated cells (such as those recorded in most in vitro experiments), only the more proximal synapses had a measurable effect on somatic voltage during in vivo-like conditions with a high level of network activity.

On the other hand, a significant role of other mechanisms, such as differences in the short-term plasticity of the inputs of OLM and HS cells, or differences in voltage-gated currents, cannot be ruled out, and would constitute exciting targets for future experimental and modeling research.

Because most excitatory inputs come from local PCs in both cell types, the larger number of excitatory inputs in HS cells may contribute to higher activity during SWRs, when the average activity of local PCs is high. However, the fact that OLM cells are silent during these high-activity periods while HS cells fire at high rates indicates that OLM cells probably receive more active inhibition during SWR activity, which also points to a (partially) different origin of the inhibitory synapses that they receive.

Therefore, the most likely remaining explanation is that the responses of OLM and HS cells are dominated by excitatory and/or inhibitory input from different sources, with very different activity patterns during SWRs. This is interesting in itself since the dendrites of the 2 cell types lie in the same layer of the hippocampus, and thus any difference in the sources of synaptic input would indicate selective innervation of one or both cell types. In fact, as we discussed above, there is some anatomical evidence for such selectivity, especially for inhibitory inputs.

Our results allow us to zoom in on the possible reasons for the distinct firing activity of OLM and HS cells in vivo, and particularly during sharp wave-ripple (SWR) activity. The relatively minor differences in intrinsic physiological features that we found are unlikely to fully account for the in vivo findings. Therefore, the difference in the level of SWR-related activity (high in HS cells and very low in OLM cells) probably comes from differences in the amount of excitatory and/or inhibitory input that they receive. We found that HS cells received a significantly larger number of both excitatory and inhibitory synaptic inputs than OLM cells; however, the ratio of excitatory and inhibitory synapses was similar in the 2 cell types. In addition, our modeling results indicated that the larger number of active synaptic inputs in HS cells would increase the membrane conductance more than in OLM cells, thereby effectively shunting further synaptic inputs and thus reducing the functional effect of the difference in synapse number. The typical sizes of synapses were also similar in the 2 cell populations, the only significant difference being the slightly larger size of inhibitory synapses in HS cells, which, if we take synapse size as a proxy of synaptic strength, would actually predict larger inhibition and thus lower activity in HS cells during SWRs, which is the opposite of the experimental observations. Overall, the observed differences in synapse numbers and sizes are also unlikely to account for the distinct firing activity during SWRs.

Responses of the 2 cell types to sustained somatic current injections were generally similar, with some potentially important differences. Specifically, HS neurons showed a clear propensity to generate high-frequency bursts following the onset of intense stimulation, which may contribute to their high activity during sharp wave-ripples.

In CA1 str. oriens, there are other subsets of somatostatin-positive interneurons in addition to OLM and HS cells (bistratified cells, back-projection neurons, oriens-retrohippocampal projection cells) [ 8 ]. We estimated the density of HS and OLM cells and found that they account for at least 49.1% and 9.5% of CA1 str. oriens somatostatin-positive cells, respectively. Nichol and colleagues found the same proportion for OLM cells (47%) [ 104 ], whereas Jinno and Kosaka found a larger population (22.5%) of HS cells using a different tracer [ 43 ]. It should be noted that these values must be regarded as minimum values since probably not all cells could be labeled using tracing methods.

The sizes of axonal output synapses of OLM axons were significantly larger than those of HS cells. In addition, the volumes of individual mitochondria were also larger in OLM axons. OLM axons innervate more distant dendritic segments of CA1 PCs in str. lacunosum-moleculare than HS cells in str. radiatum and oriens, and they may form larger synapses to compensate for the larger electrotonic distance.

We also found that only 1% of OLM cell synapses and 2% of HS cell synapses targeted GABAergic interneurons. Because the ratio of GABAergic neurons is about 9.9% in the CA1 area of the dorsal hippocampus [ 103 ], this suggests that these cells preferentially innervate PCs.

Although HS cells were found to target interneurons in in vitro slice preparations of young rats [ 45 ], both cells are known to innervate mainly PCs and to a lesser extent (3.3% to 14%) interneurons in adult rats [ 29 , 30 , 37 , 44 ]. We found that OLM and HS synapses primarily targeted PCs (97%, 96%, respectively) and often targeted their spines (19%, 23%, respectively) that is especially efficient to modulate excitatory inputs of the temporo-ammonic pathway (OLM cells) or Schaffer collaterals (HS cells) on the same spines [ 11 ]. However, OLM cells established several orders of magnitudes higher number of synapses locally.

Here, we reconstructed the local axonal tree of BDA-labeled HS cells as well. Because the DAB labeling of local axons faded away from the soma towards the distal branches and we could analyze only a limited number of sections, these are considered only partial reconstructions. The best filled HS axonal arbor was 8,625 μm long locally (only 8 pieces of 60 μm thick sections, i.e., a 480 μm thick block was examined in the rostro-caudal direction). If we multiply this length by the linear density of 0.45 synapses/μm (data from our EM reconstructions), it gives a total of 3,881 local synapses in that 480 μm thick block. However, previously in the rat we found that bouton-bearing collaterals of HS cells spread all over more than half of the entire dorsal CA1 rostro-caudally, and in the entire medio-lateral extent of CA1 [ 44 ], some cells reached even the CA3 area. The dorsal mouse CA1 is about 3.5 mm. Therefore, if these mouse HS cells also arborize in about half (1.75 mm) of the entire dorsal CA1, then HS cells may have as many as about 14,149 synapses locally in the CA1 (and they may have further synapses in CA3 as well). Therefore, HS cells may have just as many local synapses as other classical local interneurons but probably reach much more PCs (with fewer synapses per PC) in the hippocampus, while they target extra-hippocampal areas as well.

OLM cells are local interneurons with dense axonal trees that primarily target the str. lacunosum-moleculare [ 27 , 101 , 102 ]. In contrast, HS cells are long-range projecting cells that send axons to the MS and probably all of them also to the subiculum, but also have sparse local axonal branches in the hippocampus [ 36 , 45 ]. Local axons of mouse HS cells reconstructed in this study arborized in str. oriens and radiatum and never entered str. lacunosum-moleculare. Compared with the dense axonal tree of OLM cells, HS cells formed sparse branches extending far from the soma, similar to rat HS cells [ 36 , 44 ]. The axonal tree of our best filled biocytin-labeled OLM cell was 22,780 μm long in a 300 μm thick slice (Neurolucida reconstruction, Fig 7 ). Multiplying this number by linear density of 0.404 synapses/μm (data from our EM reconstructions) gives 9,203 synapses. The axonal branches of this cell extended to 818 μm in the medio-lateral direction. We can assume that the axon tree also extends at least this far in the rostro-caudal direction. If we assume that we reconstructed only a 300 μm thick slice of the 818 μm diameter disc-like axon arbor field, then we can estimate that this OLM cell established approximately 16,900 local synapses (depending on the position of the slice in the disc-like arbor). Because inter-varicose segments of a mouse OLM axon also form synapses and an axon bouton can sometimes make 2 synapses (see Fig 5B ), it is hard to compare this numbers with earlier investigations where only the total number of boutons (16,874), but not the synapses of an OLM cell was estimated, especially because this number was calculated in rat [ 101 ].

We found that the sizes of p-glutamatergic inputs increased towards the distal dendrite on OLM cells, whereas the sizes of HS synapses showed an opposite effect. Similar trends were found in the case of GABAergic inputs, although they did not show significant correlations probably because of their smaller numbers. Generally, larger synapses contain proportionally more postsynaptic receptors [ 99 ] and elicit larger synaptic events. The increase in synapse size at distal dendrites can make these distal synapses more efficient and can compensate for the electrotonic filtering of inputs to distal dendrites in OLM cells. According to our modeling results, this would make all synapses (of the same type) about equally efficient in terms of the amplitude of the somatic PSP, at least in an isolated neuron. On the other hand, our simulations indicate that such a moderate increase in the synaptic conductance would fail to offset the much greater attenuation of distal inputs in the presence of a large amount of background synaptic activity in active network states. Further research is needed to understand why the opposite strategy is adopted by HS cell dendrites and what the consequences for synaptic integration might be. A detailed modeling approach like the one used here could be particularly useful in this context; however, to investigate and compare different strategies for dendritic integration, the model may need to be extended to incorporate active dendritic processes (which are known to be present, at least in OLM cells [ 66 ]), as demonstrated previously for CA1 pyramidal neurons [ 100 ].

In addition to typical GABAergic cells, MS cholinergic cells are also GABAergic and all of their terminals establish gephyrin-positive p-GABAergic inputs in CA1 [ 98 ]. OLM cells were reported to receive cholinergic inputs which were essential for their activation during memory process [ 21 , 34 ]. Although cholinergic synapses are smaller than the non-cholinergic GABAergic synapses, the GABAergic synapses did not show bimodal size distribution.

The source of the p-GABAergic inputs is probably more diverse and more consequential, because both cell types may receive inputs from a variety of local interneurons and extrinsic GABAergic afferents. Both OLM and HS cells are among the targets of GABAergic medial septal inputs [ 13 , 44 , 91 – 93 ]. Relaxin-containing axons from the nucleus incertus also provide an important external GABAergic input to both cells types [ 24 ]. From the possible local inputs, type III interneuron-selective interneurons were shown to be particularly important for both cell types [ 13 , 94 – 96 ]. GABAergic inputs arriving from different sources can be very different in terms of their synapse size, temporal dynamics, efficiency, etc. [ 13 , 97 ]. Furthermore, they can preferentially innervate certain cell types or distinct membrane domains of these cells (soma, proximal, or distal dendritic segments) [ 5 ]. We hypothesize therefore, that the differences in the sizes of GABAergic synapses on OLM and HS somata and dendrites (and between somata and dendrites within the same cell type) are due to this variability and reflect the fact that the 2 cell types (and their different membrane domains) are probably differently innervated by inputs from different sources. This may also play a role in their distinct behaviors during different network activity patterns [ 36 , 49 ].

The dendritic trees of OLM and HS cells arborize in the termination zone of local axons of CA1 PCs in str. oriens/alveus [ 88 ] and were shown to receive the majority (at least 75% to 83%) of their p-glutamatergic inputs from CA1 PCs [ 28 ]. The str. oriens of CA1 receives a small number of entorhinal axons as well (alvear pathway) [ 89 ], which preferentially innervate GABAergic cells [ 61 ]. Therefore, these entorhinal axons could also be among the p-glutamatergic inputs of OLM and HS cells. Schaffer collaterals of CA3 PCs likely provide input to OLM and HS cells only very rarely, as they innervate PCs to a much greater extent than interneurons [ 61 , 90 ]. Based on the above, the source of glutamatergic innervation of the 2 cell types is probably similar, although HS cells receive substantially more inputs.

Since many anatomical data are only available from rats, we have to rely on these earlier studies when discussing the source of the inputs to OLM and HS cells in the mouse. However, the anatomical organization is very similar in the 2 species [ 85 ].

More synapses were associated with significantly more mitochondria in HS cell dendrites compared to OLM cell dendrites. This finding explains why HS cells express a higher than average level of cytochrome C (a central component of the oxidative phosphorylation chain) compared to other somatostatin neurons in str. oriens [ 87 ]. This may suggest a higher metabolic rate that may be explained by the large number of inputs that they need to integrate, while they must also maintain the large axonal tree far away from their somata. Indeed, during sharp wave ripples double-projection (HS) cells increase their firing rates [ 36 ] demanding more intense ATP production, whereas OLM cells are inhibited during these events [ 49 ].

Comparing the anatomical parameters of reconstructed dendritic segments of OLM and HS cells, we found several significant differences ( Fig 3D and 3E ). The numerical density of both p-glutamatergic and p-GABAergic synaptic inputs (and consequently the synaptic coverage on the plasma membrane) was significantly higher on HS cells. Most of the HS cell dendritic segments were almost completely covered by converging afferent terminals (see Fig 3B and 3C ), whereas larger terminal-free membrane regions were detected on OLM dendrites ( Fig 3A and 3C ). In an earlier study, using serial transmission electron microscopy, we estimated the density of GABA-negative and positive inputs of HS cells in rat [ 44 ]. Similarly to mouse HS cells, they were completely covered by terminals and compared to other examined interneurons in the CA1 area of rat hippocampus they received the most abundant synaptic inputs [ 54 , 85 , 86 ]. Here, the ratio (%) of GABAergic inputs per all inputs was similar in the case of mouse OLM and HS cells (14.3% and 15.4%, respectively, this study), and these were similar in rat HS cells (14%) [ 44 ]. This suggests that more excitatory inputs require proportionately more inhibition, although very different GABAergic percentages have been also detected in other cell types [ 54 ]. The higher number of synaptic inputs likely means that HS cells sample the average activity of large populations of PCs more efficiently than OLM cells. HS cells seem to function like antennae or probes that keep broadcasting the activity state of the hippocampus CA1 to the MS (and probably to the subiculum). Conversely, OLM cells are more selectively involved in local computations, including the selection of engram cells, which may require fewer inputs with a more specific distribution from subcortical afferents and local PCs.

We found that in the case of OLM and HS cells, the thickness of the dendritic segments is highly variable. Since we also found that the thicker dendrites receive proportionally more inputs, the linear density of inputs is also highly variable along the dendritic tree and it could not be precisely predicted without knowing the thickness of the dendritic segment. This highlights the importance of accurately tracking dendritic thickness during whole cell reconstructions, if they are used to estimate the number of synaptic inputs.

OLM and HS cells are hippocampal somatostatin-positive GABAergic neurons in the CA1 area that have similar dendritic morphology (at least at the light microscopic level) but different axonal projections, firing patterns, and functional roles in network activity [ 8 , 26 ]. To characterize the anatomical properties (including inputs and outputs) of the 2 cell types in more detail, we first reconstructed their dendritic trees in 3d at the light microscopic level (n = 4–4 cells) and then we created a database of their 3-dimensionally reconstructed dendritic segments, cell bodies, and axonal segments including their input and output synapses (∑ n = 90 dendritic segments, 18 branching points, 14 somata, 24 axonal segments, 5,844 input, and 227 output synapses). We measured several anatomical parameters that are important in the electrical conductance of cell membranes, e.g., the surface area of dendritic segments, the volume of cytosol and mitochondria, and the density, number and size of excitatory/inhibitory inputs. All data were corrected for volume changes during the sample processing. These parameters were precisely incorporated into the morphologically detailed compartmental model neurons we reconstructed to examine how the input organization influences signal propagation in the 2 cell types.

Understanding how the brain implements various cognitive functions requires detailed knowledge of the properties of the different classes of its constituent neurons and their synaptic interactions [ 73 , 74 ]. The commonly used abstraction of the point-like neuron in computational models completely ignores the powerful computational properties of dendrites that allow the isolation of synaptic inputs that enable multiple parallel operations [ 75 ]. Recently, large-scale, biologically realistic data-driven models, in particular, have emerged as powerful tools in the investigation of brain circuits [ 76 – 84 ], but they explicitly rely on detailed information about neuronal morphologies, electrophysiological characteristics, connectivity, and synaptic properties, and the quality of these data is a crucial determinant of the faithfulness and predictive power of such models.

Methods

Ethics statements All experiments were performed in accordance with the Institutional Ethical Codex, Hungarian Act of Animal Care and Experimentation (1998, XXVIII, section 243/1998) and the European Union guidelines (directive 2010/63/EU), and with the approval of the Institutional Animal Care and Use Committee of the Institute of Experimental Medicine of the Hungarian Academy of Sciences (PE/EA/107-5/2021). All efforts were made to minimize potential pain or suffering and to reduce the number of animals used.

Animals and surgery A total of 13 C57Bl/6NTac mice (young adult, from both sexes), 15 nicotinic acetylcholine receptor alpha 2 subunit (Chrna2)-cre mice (young adult, from both sex, heterozygous/homozygous), and 1 SOM-Cre mouse (young adult female, heterozygous) were used in the present study. Mice were anaesthetized with isoflurane followed by an intraperitoneal injection of an anesthetic mixture (containing 8.3 mg/ml ketamine, 1.7 mg/ml xylazine-hydrochloride in 0.9% saline, 10 ml/kg bodyweight) and then were mounted in a stereotaxic frame. For selective visualization of HS cells, we used 3 different retrograde labeling methods. See Fig 1 for detailed workflow. For reconstruction of dendritic trees and axons of HS cells, we injected 50 or 2 × 50 nl biotinylated dextrane amine (BDA-3000, Invitrogen) into the medial septal area (MS) of C57Bl/6NTac mice (n = 6). For the colocalization experiments (Fig 1), we injected yellow-green FluoSpheres (Thermo Fisher Scientific) into the MS of 2 C57Bl/6NTac mice or AAVrg-syn-FLEX-jGCaMP8m-WPRE (Addgene, 162378) into the MS of a heterozygous SOM-Cre female mouse. For the in vitro experiments, we used FluoSpheres-injected mice (see below). The coordinates for the injections were based on the stereotaxic atlas [107]: 1 mm anterior from the bregma, in the midline, and 4.5 and/or 4.9 mm below the level of the horizontal plane defined by the bregma and the lambda (zero level). For selective labeling of OLM cells that are Chrna2-positive in the str. oriens, we injected 20 to 60 nl rAAV2/5-EF1a-DIO-eYFP tracer virus (UNC Vector Core; 4.4–8.5 × 1,012 colony forming units/ml) into the CA1 area of the hippocampus bilaterally into Chrna2-Cre mice. The coordinates for these injections were: −2.0 posterior from the bregma, +-1.5 or 1.75 laterally, and 1.4 mm below the zero level. For the injections, we used a Nanoject 2010 precision microinjector pump (WPI, Sarasota, FL 34240) and borosilicate micropipettes (Drummond, Broomall, PA) with tips broken to 40 to 50 μm. After the surgeries, the animals received 0.5 to 0.7 ml saline for rehydration and 0.03 to 0.05 mg/kg meloxicam as a nonsteroidal anti-inflammatory drug (Metacam, Boehringer Ingelheim, Germany) intraperitoneally to support recovery, and we placed them into separate cages for 7 to 14 days (BDA and FluoSpheres-injected mice) or 23 to 39 days (virus-injected mice) before perfusions.

Perfusions and sectioning For perfusion, mice were deeply anaesthetized as above. Mice used for electron microscopic analysis of dendrites and somata were perfused transcardially first with 0.9% NaCl in 0.1 M phosphate buffer solution (PBS, pH = 7.4) for 30 s followed by a fixative containing 2% paraformaldehyde (PFA), 1% or 0.5% glutaraldehyde (GA), and 15% (v/v) picric acid in 0.1 M phosphate buffer (pH = 7.4; PB) for 5 min. Then, the brain was removed from the skull, cut sagittally and coronally into 4 pieces, and immersion fixed in a fixative containing 4% PFA and 0.2% to 0.25% GA in PB for 2 to 4 h at room temperature, on a shaker, then the fixative was washed out with PB. Mice used for Neurolucida drawing of whole dendritic trees of OLM and HS cells and electron microscopic analysis of HS cell axons were perfused similarly (Fig 1), except that the fixative used for immersion did not contain glutaraldehyde. Mice used for somatostatin and eYFP/FluoSpheres double labeling were perfused transcardially first with PBS for 30 s followed by a fixative containing 4% PFA for 40 min, followed by PB for 10 min. For the correction of shrinkage/dilatation of the tissue (see below), the volume of brains used for Neurolucida or electron microscopic reconstruction of cells was measured, then forebrain blocks were serially sectioned using a Leica VT1200S vibratome at 60 μm. Before further processing, all sections were mounted on slides in PB and photographed using a Zeiss Axioplan2 microscope. For measurement of shrinkage/dilatation during processing, see methods below.

In vitro slice preparation For acute slice experiments, FluoSpheres (HS cells, n = 4 mice) or rAAV2/5-EF1a-DIO-eYFP tracer virus-injected mice (OLM cells, n = 4 mice) were decapitated under deep isoflurane anesthesia. The brain was removed and placed into an ice-cold cutting solution, which had been bubbled with 95% O2/5% CO2 (carbogen gas) for at least 30 min before use. The cutting solution contained the following (in mM): 205 sucrose, 2.5 KCl, 26 NaHCO3, 0.5 CaCl2, 5 MgCl2, 1.25 NaH2PO4, and 10 glucose. Then, coronal slices of 300 μm thickness were cut using a Vibratome (Leica VT1000S). After acute slice preparation, slices were placed into an interface-type holding chamber for recovery. This chamber contained standard ACSF at 35°C that gradually cooled down to room temperature. The ACSF solution contained the following (in mM): 126 NaCl, 2.5 KCl, 26 NaHCO3, 2 CaCl2, 2 MgCl2, 1.25 NaH2PO4, and 10 glucose saturated with carbogen gas. All salts were obtained from Sigma-Aldrich or Molar Chemicals KFT.

Intracellular recordings After incubation, slices were transferred individually into a submerged-type recording chamber with a superfusion system allowing constantly bubbled (95% O2–5% CO2) ACSF to flow at a rate of 3 to 3.5 ml/min. The ACSF was adjusted to 300 to 305 mOsm and was constantly saturated with 95% O2–5% CO2 during measurements. All measurements were carried out at 33 to 34°C, the temperature of ACSF solution was maintained by a dual-flow heater (Supertech Instruments). The pipette solution contained (in mM): 110 K-gluconate, 4 NaCl, 20 HEPES, 0.1 EGTA, 10 phosphocreatine, 2 ATP, 0.3 GTP, 3 mg/ml biocytin adjusted to pH 7.3 to 7.35 using KOH (285 to 295 mOsm/L). Pipette resistances were 3 to 6 MΩ when filled with pipette solution. Visualization of slices and selection of cells was done under an upright microscope (BX61WI; Olympus, Tokyo, Japan equipped with infrared-differential interference contrast optics and a UV lamp). FluoSpheres (HS cells)- or eYFP (OLM cells)-labeled cells from CA1 str. oriens were selected. Only cells located deeper than approximately 50 μm measured from the slice surface were targeted. Recordings were performed with a Multiclamp 700B amplifier (Molecular Devices). All cells were initially in voltage-clamp mode and held at −65 mV holding potential during the formation of the gigaseal. Series resistance was constantly monitored after the whole-cell configuration was established, and individual recordings taken for analysis showed stability in series resistance between a 20% margin during the whole recording. To characterize electrophysiological properties of the cells, responses to a pulse-train of current steps was recorded in current-clamp mode, while the cells membrane potential was held around −65 mV at baseline (current values of steps in pA: 20, −20, 40, −40, 60, −60, 80, −80, 100, −100, 120, −120, 140, −140, 160, −160, 180, −180, 200, 220, 240, 260, 280 with 1,000 ms in between and 300, 400, 500, and 600 with 5,000 ms in between, each stimulus lasted for 800 ms). The recorded cells were filled with biocytin. After the recording, the slices were fixed in 4% paraformaldehyde in PB for at least 12 h, followed by washout with PB several times.

Digital signal processing, analysis, and statistics for in vitro experiments Data were digitized at 20 kHz with a DAQ board (National Instruments, USB-6353) and recorded with a custom software developed in C#.NET and VB.NET in the laboratory. All data were recorded, processed, and analyzed offline using standard built-in functions of Python (2.7.0.), Matlab, or in custom made software developed in Delphi.

Fluorescent immunohistochemistry and microscopy Hippocampal slices (300 μm thick) and sections (60 μm thick) were washed in 0.1 M PB (pH 7.4) and incubated in 30% sucrose overnight for cryoprotection. Then, they were freeze-thawed over liquid nitrogen 3 times for antigen retrieval. Slices or sections were subsequently washed in PB and Tris-buffered saline (TBS, pH 7.4), blocked in 1% human serum albumin in TBS (HSA; Sigma-Aldrich) containing 0.05% Triton X-100 (Bio-Rad Laboratories) and then incubated in a mixture of primary antibodies and reagents dissolved in TBS containing 0.05% Na-azide (Sigma-Aldrich) for 48 to 72 h (chicken anti-eGFP (1:2,000, Thermo Fisher Scientific, A10262) an

[END]
---
[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3002539

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.