Abstract
Article Figures and data Abstract eLife digest Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract Generating a comprehensive description of cortical networks requires a large-scale, systematic approach. To that end, we have begun a pipeline project using multipatch electrophysiology, supplemented with two-photon optogenetics, to characterize connectivity and synaptic signaling between classes of neurons in adult mouse primary visual cortex (V1) and human cortex. We focus on producing results detailed enough for the generation of computational models and enabling comparison with future studies. Here, we report our examination of intralaminar connectivity within each of several classes of excitatory neurons. We find that connections are sparse but present among all excitatory cell classes and layers we sampled, and that most mouse synapses exhibited short-term depression with similar dynamics. Synaptic signaling between a subset of layer 2/3 neurons, however, exhibited facilitation. These results contribute to a body of evidence describing recurrent excitatory connectivity as a conserved feature of cortical microcircuits. https://doi.org/10.7554/eLife.37349.001 eLife digest The outer sheet of brain tissue, the neocortex, is composed of circuits formed from trillions of connections among billions of neurons, of which there are about one hundred different neuron types. The scale and complexity of cortical circuitry pose experimental challenges, leading to an incomplete understanding of how cortical cell types are connected and the computations that take place at the connections. About half of the cell types in the brain are excitatory, which means they can activate other cells. The cortex consists of several distinct layers of cells, within which excitatory cells cooperate to process the signals they receive from other cortical layers and brain areas. Using recordings of electrical activity arising from the connections between pairs of excitatory neurons, Seeman, Campagnola et al. measured the likelihood and strength of connectivity among related groups of excitatory cell types in slices of cortex taken from human and mouse brains. The initial results confirm previous findings that individual layers of human cortex can have more and stronger excitatory connections than the same layers of mouse cortex. In most layers of mouse cortex, repeatedly activating the excitatory cells leads to progressively weaker responses. However, in the upper layers of mouse cortex, the opposite effect is sometimes seen: more excitatory activity causes the connections to generate stronger responses. By feeding these data into a computer model, Seeman, Campagnola et al. described and compared the activity of the groups of related excitatory cell types. These results are the first of a new, large-scale project where findings can be integrated across experiments to gain a more detailed picture of cortical circuitry and computation. Neuroscientists will be able to use the results to build advanced computer models of cortical circuits. Such models will, for example, generate predictions for how the attributes of excitatory connectivity revealed by Seeman, Campagnola et al. influence how information is processed in the cortex. In so doing, the models will add to our understanding of how the human brain works both in health and in disease. https://doi.org/10.7554/eLife.37349.002 Introduction Generating well-informed, testable hypotheses about how the cortex represents and processes information requires experimental efforts to characterize the connectivity and dynamics of cortical circuit elements as well as efforts to build models that integrate results across studies (Sejnowski et al., 1988). Estimates of connectivity and synaptic properties vary widely between experiments due to differences in model organisms, experimental parameters, and analytic methods. This variability limits our ability to generate accurate, integrative computational models. Addressing this problem requires standardized experimental methods and large-scale data collection in order to characterize synaptic connections between the large number of potential cell types (Tasic et al., 2016). Although it may be possible to infer part of these results based solely on anatomical constraints (Markram et al., 2015), evidence has shown that the rate of connectivity and properties of synaptic signals can depend on the identity of the pre- and postsynaptic neuron (Reyes et al., 1998; Galarreta and Hestrin, 1998; Larsen and Sjöström, 2015). To collect standardized data at scale, we have established a pipeline to characterize local, functional connectivity in the adult mouse and human cortex. Initially, we seek to characterize connectivity among cell classes, that is, groups of related cell types (Tasic et al., 2016). Here, we report on the characteristics of local excitatory inputs among pyramidal neurons from within the same layer (recurrent connections) obtained during the pipeline’s system integration test—an end-to-end test of the pipeline’s hardware, software, and workflow carried out prior to initializing the pipeline. Recurrent excitatory connectivity is thought to be important in behavior (Evans et al., 2018) and disease (Jin et al., 2006). It is a common feature in computational models of cortical working memory, receptive field shaping, attractor dynamics, and sequence storage (Camperi and Wang, 1998; Olshausen and Field, 1996; Mongillo et al., 2008; Brunel, 2016; Pernice et al., 2018). Empirical measurements of recurrent connectivity and synaptic properties are needed in order to constrain and validate these models. However, characterizing recurrent connectivity in a standardized, high-throughput manner is challenging because the synaptic connections can be sparse and weak (Braitenberg and Schüz, 1998; Song et al., 2005; Lefort et al., 2009). Furthermore, most measurements of recurrent connectivity have been performed in juvenile rodents, leading to a recent debate over the rate of connectivity in the adult cortex (Biane et al., 2015; Barth et al., 2016; Jiang et al., 2016). The data reported here demonstrate that sparse recurrent connectivity is present among excitatory neurons in all layers of adult mouse and human cortex. Using a novel automated method for systematically estimating connectivity across experiments, we further demonstrate that different populations of adult mouse pyramidal neurons exhibit characteristic distance-dependent connectivity profiles and short-term dynamics. Finally, we quantify and compare differences in short-term dynamics with a mechanistic computational model. Results We performed in vitro whole-cell recordings from up to eight excitatory neurons simultaneously. We probed 2836 putative connections in mouse V1 from excitatory cell classes defined by transgenic labeling, morphology, and cortical layer (Tasic et al., 2016). We further probed 616 putative connections in human frontal and temporal cortex from excitatory cell classes defined by morphology and cortical layer (Table 1). Recurrent connectivity was tested and observed in layer 2/3 through layer 6 of mouse primary visual cortex and layer 2 through layer 6 of the human cortex. To assess connectivity, trains of action potentials were evoked in each cell, one at a time, while recording synaptic responses in all other cells. Connections were identified by the presence of excitatory postsynaptic potentials (EPSPs) evoked with a short latency and low jitter following the presynaptic spike, consistent with monosynaptic connections (Figure 1—figure supplement 1). We encountered no examples of EPSPs eliciting spikes in any recorded pyramidal cells, further indicating that evoked polysynaptic activity should be rare in these experiments. Table 1 The number of connections probed and the number of connections used in subsequent analyses per the analysis flow diagram in Figure 1—figure supplement 1C–G. For each column, the Figure 1—figure supplement 1 letter indicates the end level in the analysis flow diagram while the main figure reference indicates n connections included in that figure. For example, the ‘Strength’ column indicates the number of connections for each type used to measure the strength (or amplitude) of the connection as shown in Figure 1F. The inclusion criteria for these connections can be followed in the diagram in Figure 1—figure supplement 1E. Similarly, these data are provided for kinetics (rise time and latency) and short-term plasticity (STP). https://doi.org/10.7554/eLife.37349.003 Layer/Cell TypeTotal probed (Figure 1—figure supplement 1C)Total connected (Figure 1—figure supplement 1C)Total connection probability (%)Strength (Figure 1—figure supplement 1E, Figure 1F)Kinetics (Figure 1—figure supplement 1F, Figure 1F)Connection probability (%) w/in 100 µm (Connected/probed, Figure 1—figure supplement 1D, Figure 4A,C)STP (Figure 1—figure supplement 1G, Figures 5 and 6)Mouse L2/3180158.312913/130 (10.0)9Rorb315206.3131318/247 (7.3)9Tlx31108393.5171436/746 (4.8)5Sim1783557.0181841/527 (7.8)7Ntsr145020.4220/313 (0.0)N/AHuman L21322216.7181813/69 (18.8)N/AHuman L32493714.9332920/106 (18.9)N/AHuman L412343.3221/51 (2.0)N/AHuman L51121311.6666/49 (12.2)N/A Properties of intralaminar excitatory synaptic signaling in mouse cortex Layer and projection-specific classes of excitatory neuron populations were identified either by post-hoc morphologic evaluation in layer 2/3 (animals n = 11) or transgenic labelling to target layers 4–6 (layer 4: Rorb (n = 28), layer 5: Tlx3 (n = 57), Sim1 (n = 20), layer 6: Ntsr1 (n = 13); Figure 1A). Layer 5 recordings were subdivided into subcortical projecting cells (Sim1; http://connectivity.brain-map.org) or corticocortical projecting cells (Tlx3; Kim et al., 2015). In layer 6, only the subcortically projecting cells were targeted (Ntsr1; Vélez-Fort et al., 2014). We probed 2836 potential connections (layer 2/3: 180, Rorb: 315, Tlx3: 1108, Sim1: 783, Ntsr1: 450) across these excitatory populations in mouse cortex (Table 1). Connections were detected between 131 putative pre- and post-synaptic partners (layer 2/3: 15, Rorb: 20, Tlx3: 39, Sim1: 55, Ntsr1: 2; Table 1). For >75% of the recorded cells, we recovered a biocytin fill (Figure 1A) and for all cells we obtained an epifluorescent image stack (Figure 1B). Figure 1 with 1 supplement see all Download asset Open asset Electrophysiological recordings of evoked excitatory synaptic responses between individual cortical pyramidal neurons in mouse primary visual cortex. (A) Cartoon illustrating color, Cre-line, and cortical layer mapping in slice recording region (V1). Example maximum intensity projection images of biocytin-filled pyramidal neurons for L2/3 and each Cre line. (B) Example epifluorescent images of neurons showing Cre-dependent reporter expression and/or dye-filled recording pipettes. Connection map is overlaid on the epifluorescent image (colored: example connection shown in C). (C) Spike time aligned EPSPs induced by the first AP of all ≤ 50 Hz stimulus trains for a single example connection (individual pulse-response trials: grey; average: colored). (D) First pulse average, like in C., for all connections within the synaptic type; grey: individual connections; thin-colored: connection highlighted in C; thick-colored: grand average of all connections. (E) Overlay of grand average for each connection type. (F) EPSP amplitude (in log units), CV of amplitude, latency, and rise time of first-pulse responses for each Cre-type (small circles) with the grand median (large). See Figure 1—figure supplement 1 for data processing and analysis diagrams. https://doi.org/10.7554/eLife.37349.004 Figure 1—source data 1 Electrophysiological recordings of evoked excitatory synaptic responses between individual cortical pyramidal neurons in mouse primary visual cortex. https://doi.org/10.7554/eLife.37349.006 Download elife-37349-fig1-data1-v1.csv Table 2 Properties of mouse EPSPs. Median, mean, and standard deviation of EPSP properties plotted in Figure 1F for each layer and Cre-type. Number of connections used in the amplitude and CV analysis are found in Table 1 ‘Strength’, or for latency and rise time in Table 1 ‘Kinetics’. https://doi.org/10.7554/eLife.37349.007 Amp median (mV)Amp mean (mV)Amp SD (mV)Latency median (ms)Latency mean (ms)Latency SD (ms)Rise Time median (ms)Rise Time mean (ms)Rise Time SD (ms)CV medianCV meanCV SDL2/30.260.34±0.321.481.87±1.01.241.45±0.570.550.56±0.15Rorb0.310.54±0.491.311.50±0.61.321.63±0.940.590.55±0.24Sim10.330.52±0.511.862.05±0.821.911.86±1.10.360.43±0.2Tlx30.140.24±0.241.812.07±0.741.441.35±1.10.510.51±0.18 We first characterized the strength and kinetics in recurrent connections of each Cre-type and layer (Figure 1). To measure these features with minimal influence of STP, only the first response on each sweep (inter-trial interval (ITI) = 15 s) was included for this analysis. For each connection, individual sweeps were included based on a number of criteria, namely a maximum autobias current to reach a holding potential of −70 ± 5 mV, a stable baseline, and absence of spontaneous spiking (see Materials and methods; Figure 1—figure supplement 1E,F). A minimum of 5 QC-passed sweeps were required for each connection to be included. Figure 1C shows EPSPs recorded from one example connection found in each of the chosen excitatory cell groups. For the large majority of connections, it was not possible to unequivocally distinguish synaptic failures from detection failures, thus we used the mean response from all sweeps (Figure 1C) to evaluate the EPSP features. Consistent with previous reports that recurrent connectivity is weak (Song et al., 2005; Lefort et al., 2009), we found that a majority of the connections had amplitudes less than 0.5 mV. In this small sample, we did not observe statistical difference in the EPSP amplitudes (Figure 1E,F) between groups (KW p=0.07), although there was a trend toward overall smaller Tlx3 EPSP amplitudes (median ± SD 0.14 ± 0.24 mV). The range of amplitudes for layer 2/3 (0.032–0.902 mV), Rorb (0.105–1.626 mV), Sim1 (0.068–1.254 mV), and Tlx3 (0.02–0.833 mV) spanned an order of magnitude. We could not assess the range of recurrent Ntsr1 connections due to the low number of connections measured; however, the amplitude and relatively long latency (Figure 1F) are consistent with connections between corticothalamic (CT) layer six neurons in the rat cortex (West et al., 2006; Table 1). The mean EPSP amplitude was consistently larger than the median (Table 2) due to a skewed (long-tailed) distribution of response amplitudes. Similar observations in the rat visual cortex, and mouse somatosensory cortex, has led to the suggestion that rare, large-amplitude connections are important for reliable information processing (Song et al., 2005; Lefort et al., 2009; Cossell et al., 2015). The majority of EPSP latencies were less than 2.5 ms (Table 2), and similar across populations (KW p=0.17), consistent with a direct, monosynaptic connection between recorded neurons. We could not directly quantify synaptic failures and thus calculated the coefficient of variation of synaptic amplitudes (CV; Figure 1F) to assess release probability. The CV of each connection describes the variability in a particular response in relation to the mean (ratio of standard deviation to mean) and is negatively correlated with release probability (Markram, 1997). The range of coefficient of variation in our data suggests differences in release probability between cell classes and is consistent with STP modeling results (see Figure 6). Properties of intralaminar excitatory synaptic signaling in human cortex To what extent is recurrent connectivity in mouse V1 representative of connectivity in other regions and species? To make this comparison, we performed multipatch recordings from human frontal and temporal cortex. Specimens were collected during surgical resection of epileptic or tumorous tissue, but were distal to the site of pathology. We sampled recurrent intralayer connectivity in all layers containing pyramidal cells. Pyramidal cells were identified by their morphology visualized via biocytin (Figure 2A) or fluorescent dye (Figure 2B). We found 22 connections between layer 2 pyramidal cells (132 probed), 37 connections between layer 3 pyramidal cells (249 probed), four connections between layer 4 pyramidal cells (123 probed), and 13 connections between layer 5 pyramidal cells (112 probed). We found 1 connection in layer 6 (16 probed connections) but have not yet probed this layer sufficiently to make confident measurements of connection probability or synaptic properties. We selected 1.3 mM [Ca++]e for our human experiments because of reports that synaptic strength is higher than in mouse and to minimize the complex events that can be initiated by individual spikes in human tissue (Molnár et al., 2008) that make identifying monosynaptic connectivity challenging. Indeed we found that human cortex had a higher connectivity rate and mean amplitude (Figure 2C,D) compared to mouse cortex (despite a higher [Ca++]e in mouse), consistent with previous reports (Molnár et al., 2008). Layers 2, 3, and 5 had a sufficient number of connections to characterize strength and kinetics. However, we found no differences in response properties among these three layers (amplitude p=0.22, latency p=0.51, rise time p=0.22, Table 3). We did observe differences in CV between layers 2, 3, and 5 (p=0.0004, Table 3) suggesting layer-specific differences in release probability of recurrent connections, similar to findings in mouse V1. Figure 2 Download asset Open asset Electrophysiological recordings of evoked excitatory synaptic responses between individual human cortical pyramidal neurons. (A) Cartoon illustrating color and cortical layer mapping in slice recording region (temporal or frontal cortex). Example maximum intensity projection images of biocytin-filled pyramidal neurons for layers 2–5. (B) Example epifluorescent images of neurons showing dye-filled neurons and recording pipettes. Connection map is overlaid on the epifluorescent image (colored: example connection shown in C). (C) Spike time aligned EPSPs induced by the first AP of all ≤ 50 Hz stimulus trains for a single example connection (individual pulse-response trials: grey; average: colored). (D) First pulse average, like in C., for all connections within the synaptic type; grey: individual connections; thin-colored: connection highlighted in C; thick-colored: grand average of all connections. (E) Overlay of grand average for each connection type. (F) EPSP amplitude, CV of amplitude, latency, and rise time of first-pulse responses for each layer (small circles) with the grand mean (large circles). See Figure 1—figure supplement 1 for data processing and analysis diagrams. https://doi.org/10.7554/eLife.37349.008 Figure 2—source data 1 Electrophysiological recordings of evoked excitatory synaptic responses between individual human cortical pyramidal neurons. https://doi.org/10.7554/eLife.37349.009 Download elife-37349-fig2-data1-v1.csv Table 3 Properties of human EPSPs. Median, mean, and standard deviation of EPSP properties plotted in Figure 1F for each layer and Cre-line. Number of connections used in the amplitude and CV analysis are found in Table 1 ‘Strength’, for latency and rise time in Table 1 ‘Kinetics’. https://doi.org/10.7554/eLife.37349.010 Amp median (mV)Amp mean (mV)Amp SD (mV)Latency median (ms)Latency mean (ms)Latency SD (ms)Rise time median (ms)Rise time mean (ms)Rise time SD (ms)CV medianCV meanCV sdL20.220.30±0.221.841.79±0.781.471.53±0.590.800.64±0.29L30.340.54±0.681.571.58±0.971.602.07±1.360.390.44±0.23L40.970.97±1.052.702.70±1.802.022.02±0.470.370.37±0.20L50.620.80±0.691.641.75±0.591.161.23±0.460.290.34±0.10 It is reasonable to question if the recurrent connectivity we see in tissue from epilepsy and tumor patients differs from that of healthy individuals. Although we cannot rule this out, we saw no significant differences in overall connectivity between tumor and epilepsy-derived specimens (p=0.833, Fisher’s Exact Test). We also found recurrent connections within multiple cortical regions and disease states in the human. Taken together, this may indicate that our results capture a common architecture of the mouse and human microcircuit. Detection limit of synaptic responses When using whole cell recordings to characterize synaptic connectivity, a major limitation is that some EPSPs may be too weak to be detected at the postsynaptic soma. Detection limits are influenced by several factors including the kinetics of EPSPs, the amplitude and kinetics of background noise, the frequency and properties of spontaneous EPSPs, and the number of evoked presynaptic spikes. One consequence is that we expect to generally underestimate connectivity, and in some cases, cell class differences in synaptic strength can be misinterpreted as differences in connectivity. Another consequence is that it may not be possible to obtain an accurate measurement of the distribution of synaptic weights, since the weakest synapses are undetectable. To address these issues, we characterized the sensitivity of our experiments by testing whether a machine classifier could detect simulated EPSPs of varying, known strength (see Materials and Methods). The classifier was trained to detect connections based on features extracted from the averaged response to evoked spikes (Figure 3B; features listed in Supplementary file 1) and from the distributions of features measured on individual responses (Figure 3C). For each putative connection probed, we collected recordings of background activity when no cells were being stimulated and superimposed EPSP-like deflections. These recordings were then processed for features (Figure 3 – table supplement 1) which were fed to the classifier to generate connectivity predictions. By testing several sets of artificial EPSPs in which we systematically varied the average amplitude, we were able to measure the relationship between EPSP strength and the probability that a connection could escape detection (Figure 3D). Figure 3 Download asset Open asset Characterization of synapse detection limits. (A) Scatter plot showing measured EPSP amplitude versus minimum detectable amplitude for each tested pair. Detected synapses (manually are shown as pairs with no detected EPSPs are The region the the region in which synaptic connections are to be as example putative connections are highlighted in A and described further in One connection was selected for large amplitude and low background Another connection is to detect by due to low amplitude and background The shows a recorded that was as (B) A of postsynaptic current recordings in response to presynaptic spikes. recordings from a single tested pair. The indicates the time of presynaptic measured as the of maximum in the presynaptic indicate the of the (C) showing distributions of response measured from (see Materials and indicates measurements on background indicates measurements following a presynaptic (D) Characterization of detection limits for each the probability that simulated EPSPs be detected by a as a of the rise time and mean amplitude of the EPSPs. example has a different characteristic detection limit that on the recording background and the of the (E) of the number of across the The measured distribution of EPSP amplitudes is shown in with a with = 1 The in is by the measured distribution by the overall probability of a synapse at each See Supplementary file 1 for features included in Figure data 1 Characterization of synapse detection limits. Download This analysis for connection that we probed, an of the minimum detectable EPSP amplitude (Figure with low background and will generally the detection of small EPSPs (Figure has a detection limit of recordings will have higher detection and will report connectivity (Figure has a detection limit 100 EPSPs with rise time (or other properties that distinguish the EPSP from are more to be detected (Figure 3D). These results confirm that the differences in experimental between studies example, the number of presynaptic spikes evoked for each can have a on the connectivity but also suggests that future studies could these differences by characterizing their detection limits. The results of this analysis also a means of estimating the of the distribution of synapse at the low end, where synapses more to Figure shows the distribution of EPSP amplitudes across all detected synapses as well as the the probability that synapses be detected as a of EPSP amplitude the measured distribution by the probability of detection a distribution with an overall in connectivity. Although this as the detection probability an is that the of the distribution in the region where detection probability is suggesting that the sensitivity of our experiments is to capture the majority of synapses and that we have the distribution of synaptic We are however, in our of this analysis on several about the behavior of the classifier and the of the simulated EPSPs. the be a larger Connection probability of excitatory synapses Estimates of connectivity vary widely across in part due to In to the of detection sensitivity described connection probability is by the over which connections are This distribution of connections may also into the of functional microcircuits. In connectivity in layer and Sim1 neurons within 100 µm was similar Figure Rorb: Sim1: However, within this Tlx3 connectivity was Tlx3: Consistent with previous experiments in rat neurons (West et al., Ntsr1 connectivity was sparse as only connections were detected of and were relatively of and to all other connectivity versus profiles (Figure a in the connection probability with We did not out an analysis of connectivity because we the statistical to detect differences between our Furthermore, connectivity at the cell class level can results and Figure 4 with 3 see all Download asset Open asset connectivity profiles of mouse and human connections. (A) Recurrent connection probability and distribution of connections for mouse layer connection probability circles) and for connections probed within 100 µm (n connections in Table 1). (B) Connection probability over for mouse and layer of putative connections connection probability with in µm (C) connection probability and distribution of connections between human pyramidal neurons. connection probability circles) and for connections probed within 100 (D) Connection probability over for human pyramidal neurons,