Fig 1.
Basic microcircuit of neocortex and turtle dorsal cortex.
(A) The neocortex consists of 6 layers with multiple types of pyramidal neurons (color) and thalamic inputs (red) terminating in spatially restricted regions. For clarity, interneurons are omitted in this schematic diagram. (B) The turtle dorsal cortex consists of one cellular layer (#2) of densely packed pyramidal neurons (blue), sandwiched between two neuropil layers (#1 and 3) that are densely packed with dendrites and axons, and also contain interneurons (grey). Sensory afferents (red) from the lateral geniculate nucleus (LGN) make en-passant synapses in superficial layer 1 on distal segments of pyramidal neuron dendrites and on superficial inhibitory interneurons.
Fig 2.
Whole-cell recordings from pyramidal neurons in turtle visual cortex.
Schematic diagram of an isolated piece of turtle dorsal cortex (left panel) with the ventricular side up and containing pyramidal neurons (blue) and interneurons (grey). A whole-cell recording of the pyramidal neuron membrane potential in response to current injection (right panel) is obtained with a patch electrode (grey triangle) that is positioned at the pyramidal neuron soma under visual guidance with DIC optics.
Fig 3.
Distribution of electrophysiological properties.
(A) Illustration of how a subset of the parameters were measured from the action potential shape of the first action potential in response to somatic current injection: threshold voltage (Vth), width at threshold (WAP,thresh), width at half-max (WAP), height, maximum fall rate (Min(dV/dt)), and afterhyperpolarization latency (LAHP). (B) Distribution of the measured values for the 14 electrophysiological parameters from membrane potential recordings in response to somatic current injection from 225 pyramidal neurons: rheobase current (Ir), membrane time constant (tm), action potential voltage threshold (Vth), action potential frequency adaptation ratio (AP FAR), action potential amplitude (AP Height), action potential duration at threshold (WAP,thresh), action potential fall rate (Min(dV/dt)), time to peak of the afterhyperpolarization (LAHP), action potential duration halfway between threshold and peak (WAP), action potential duration increase (WAP Inc.), resting membrane potential (Vrest), action potential amplitude decrement (Vpeak Dec.), input resistance (IR), interspike interval variability (ISI Var). The apparent deviations from normal distributions suggest that there are discrete groups of pyramidal neurons within this population.
Fig 4.
Two main types of pyramidal neurons in turtle dorsal cortex.
(TOP) Ward’s unsupervised clustering applied to a sample of 225 pyramidal neurons from turtle dorsal cortex, with each neuron characterized by 14 electrophysiological parameters. The x-axis in each plot represents the individual neurons. The y-axis represents the Euclidean distance between the two merged neurons/clusters in parameter space. Dashed lines in the four identical dendrograms indicate possible threshold choices. The dashed line in the colored dendrogram marks the threshold as suggested by the Thorndike procedure, which indicates two types of pyramidal neurons. (BOTTOM) The number of types increases with decreasing threshold in units of the linkage distance. The most robust choice of the threshold value is suggested by the widest range of threshold values in normalized parameter space for which the number of pyramidal neuron types is constant. This choice also indicates two main types of pyramidal neurons in turtle dorsal cortex.
Fig 5.
Comparison of clustering algorithms.
(A) Silhouette plot of Ward’s clustering. Within each cluster (green/A and magenta/B), cells are ranked (vertical axis) in decreasing order of their silhouette values (horizontal axis). Large positive silhouette values indicate that the data point is close to its cluster’s centroid, whereas negative silhouette values indicate that the data point is closer to the centroid of the other cluster. Right panel: The dendrogram from Ward’s clustering is shown for comparison. (B) Silhouette plot for one rendition of k-means clustering (k = 2). Right panel: The plot of clustered data points (black and gray) within the plane spanned by the input resistance (IR) and the action potential frequency adaptation ratio (AP FAR) illustrates the partial separation of data points from different clusters in this plane alone. (C) Silhouette plot for one rendition of k-means clustering (k = 2) on the scrambled data set. Right panel: The plot of clustered data points (scrambled data set) within the same plane of parameters as in C, reveals the lack of separation caused by scrambling. (D) Comparison between the average silhouette for the Ward’s and k-means (k = 2) clustering of the original dataset and the average silhouette of randomized databases. Scrambling of the data set causes a consistent loss of quality in the clustering. Error bar of the average silhouette for k-means clustering is evaluated by the SD over 1000 renditions of the original data set and by independent randomization for each rendition of the scrambled data sets.
Fig 6.
Physiological differences between the two main types of pyramidal neurons based on Ward’s clustering.
Histograms of the distribution of the 14 electrophysiological properties shown in Fig 3B and corresponding Gaussian fits for the two main types (green/A and magenta/B). Abbreviations as in Fig 3B.
Table 1.
Properties of the two main types of pyramidal neurons in turtle visual cortex.
Mean and standard deviation for all values.
Fig 7.
Three physiological parameters produce good separation of the two main pyramidal neuron subtypes.
(A) Representative membrane potential responses to somatic current injections for three pyramidal neurons from each subtype (green/A and magenta/B). Input currents ranged from 50 to 70 pA. (B) Partial separation between the two main pyramidal neuron subtypes is observed in a plot of clustered data points (green/A and magenta/B) within the space spanned by the resting membrane potential (Vrest), the input resistance (IR), and the action potential frequency adaptation ratio (AP FAR).
Fig 8.
Visual response properties of the two physiologically defined pyramidal neuron types.
(A) Schematic of the turtle ex vivo eye-attached whole-brain preparation. A diffuse flash of light from the LED (red) is projected onto the intact retina within the eye-cup (gray bowl), while the membrane potential from a pyramidal neuron is recorded with a patch electrode (gray triangle) inserted into the unfolded visual cortex. (B) Pyramidal neuron membrane potential responses to brief flashes of light (10 ms, 640 nm; red arrow) persist long beyond the duration of the flash, are variable from trial-to-trial, and display sparse spiking. Trial averages are shown in black. Representative membrane potential visual responses to flashes are shown for three pyramidal neurons from each physiologically defined type (green/A and magenta/B). The responses are fluctuating and similar for both types. (C) The time courses of trial-averaged membrane potential (after spike clipping) of all pyramidal neurons recorded in response to flashes (8 type A (green), 16 type B (magenta)). Averages across pyramidal neuron visual responses of the same physiological type are plotted in bold (green/A and magenta/B).
Table 2.
Parameters used to define the model neurons.
Fig 9.
A model network with pulsed external inputs reproduces the similarity of the responses for two types of excitatory neurons.
(A) Conceptual cartoon illustrating key model features, including inhibitory neurons (blue) and two types of excitatory neurons (green/A and magenta/B), with excitatory external inputs (“LGN”, orange) limited to one excitatory type (green/A). Intracluster and intercluster connection probabilities are parameterized by Pin and Pout, respectively. (B) Simulated membrane potential responses for the two types of excitatory model neurons (green/A and magenta/B) in response to a brief (100 ms, black horizontal bar) increase in the external input (increased rate of the Poisson pulse trains) for multiple combinations of intracluster and intercluster connection probabilities, each ranging between 0.0 and 0.5. For clarity, simulation results for 200 of the 1000 neurons of each excitatory type are plotted. (C) The connection probability combination of Pin = 0.1 and Pout = 0.25 best reproduces the experimentally observed persistent activity and the similarity of the type A (green) and the type B (magenta) response to brief external inputs.