Area-Specific Features of Pyramidal Neurons—a Comparative Study in Mouse and Rhesus Monkey

Abstract

A principal challenge of systems neuroscience is to understand the unique characteristics of cortical neurons and circuits that enable area- and species-specific sensory encoding, motor function, cognition, and behavior. To address this issue, we compared properties of layer 3 pyramidal neurons in 2 cortical areas that span a broad range of cortical function—primary sensory (V1), to cognitive (frontal)—in the mouse and the rhesus monkey. Hierarchical clustering and discriminant analyses of 15 physiological and 25 morphological variables revealed 2 fundamental principles. First, V1 and frontal neurons are remarkably similar with regard to nearly every property in the mouse, while the opposite is true in the monkey, with V1 and frontal neurons exhibiting significant differences in nearly every property assessed. Second, neurons within visual and frontal areas differ significantly between the mouse and the monkey. Neurons in mouse and monkey V1 are the same size, but differ in nearly every other way; mouse frontal cortical neurons are smaller than those in the monkey and also differ substantially with regard to most other properties. These findings have broad implications for understanding the differential contributions of heterogeneous neuronal types in construction of cortical microcircuitry in diverse brain areas and species.

Introduction

An influential idea in neuroscience has been that prototypical pyramidal neurons form the basic building blocks of canonical cortical circuits or microcolumns which, once fully understood, can be extrapolated from one brain area or even species to another (e.g., Douglas and Martin 2004). The canonical cortical column serves as a key theoretical underpinning of a variety of brain mapping and modeling initiatives, although this concept remains a matter of intense investigation and debate (reviews: Rockland 2010; Molnar 2013). Much of what is known about cortical pyramidal neuron function and structure has been derived from rodents, and the extent to which these data can be extrapolated to the primate remains an open question. Compared with rodents, cortical areal diversity is highly expanded in primates, as demonstrated by differences in the number, density, lamination, and connectivity patterns of neurons in functionally distinct areas (reviews: DeFelipe et al. 2002; DeFelipe 2011; Barbas 2015). It is likely that individual neuron types differ markedly as well, and it is imperative to gain clarity on these differences in diverse cortical areas both within- and between-species.

Early studies established that pyramidal neurons differ with regard to dendritic arbor size across cortical areas in both the rhesus monkey and the human brain (Cajal 1894, 1995; Conel 1941, 1967). More recent studies have confirmed and extended these findings by demonstrating significant heterogeneity of pyramidal neurons, cortical circuitry and function in diverse areas and species (Preuss 2001; Jacobs and Scheibel 2002; Elston 2003; Elston and Fujita 2014). In the rhesus monkey, there is an increase in the size and complexity of layer 3 (L3) pyramidal neurons from primary visual cortex (V1) to higher order lateral prefrontal cortex (LPFC; Elston 2000, 2002; Elston et al. 2001; Amatrudo et al. 2012). Further, the size of dendritic spines and of excitatory synapses is significantly larger in the LPFC than V1 of this species (Medalla and Luebke 2015). These unique structural features impact the physiology of pyramidal neurons in these areas—smaller and less complex V1 pyramidal neurons have a significantly higher input resistance, depolarized resting membrane potential and higher action potential (AP) firing rates than do LPFC neurons. Moreover, spontaneous excitatory postsynaptic currents (sEPSCs) are lower in amplitude and frequency and have faster kinetics in V1 than in LPFC pyramidal neurons (Amatrudo et al. 2012; Medalla and Luebke 2015). These distinctive properties of rhesus monkey V1 and LPFC neurons are likely fundamental determinants of area-specific neuronal and network behavior. Whether such marked differences between V1 and frontal neurons exist in the rodent is not known.

Comparative studies have established that pyramidal neurons in a variety of brain areas increase in size across phylogeny (reviews: (Wittenberg and Wang 2008; DeFelipe 2011). Thus, pyramidal neurons are thought to “scale” from a small brain, such as that of a mouse, to a large brain, such as that of a rhesus monkey (reviews: Wittenberg and Wang 2008; DeFelipe 2011). Scaling occurs for L3 pyramidal neurons in a variety of cortical areas across species (Elston and DeFelipe 2002; Elston and Zeitsch 2005; Ballesteros-Yanez et al. 2006; Elston 2007; Wen et al. 2009; Elston and Manger 2014). Whether patterns of basal and apical dendritic architecture differ or are conserved is less understood. Compartmentalization of distinct dendritic domains is key to the computational and integrative capacities of pyramidal neurons. Elaboration of apical and basal arbors enables integration of synaptic input signals within different cortical layers (review: Spruston 2008). These signals are heavily filtered as they traverse the dendritic arbors toward the soma and axon initial segment (Rall 1962, 1964; Migliore et al. 1995; Mainen and Sejnowski 1996; Vetter et al. 2001; Krichmar et al. 2002; Tsay and Yuste 2002; Acker and White 2007; Psarrou et al. 2014). Thus, a detailed understanding of areal- and species-specific differences in dendritic topology is a prerequisite for building and analyzing realistic models of neurons and circuits in different cortical areas. While these structural properties have important implications for the passive behavior of neurons, it is also critical that detailed data on active physiological parameters (which likely differ in different dendritic domains) are used to constrain realistic single-neuron and circuit models. In comparison to structural properties, less is known about the electrophysiological properties of pyramidal neurons in the primate (e.g., Murayama et al. 1997; Shinomoto et al. 2005; Amatrudo et al. 2012; Zaitsev et al. 2012; Medalla and Luebke 2015) and these have rarely been directly compared with those in the rodent (but see Verhoog et al. 2013; Testa-Silva et al. 2014). Importantly, in vitro slice recordings of human pyramidal neurons have shown that synaptic plasticity rules are reversed from those in rodent (Verhoog et al. 2013) and that the temporal resolution of synaptic information exchange in human cortex far surpasses that of mice (Testa-Silva et al. 2014). On the other hand, it has been reported that there is no significant difference in the excitatory synaptic properties of layer 4 neurons in rat versus cat neocortex (Bannister and Thomson 2007). These studies, demonstrating both differences and similarities, highlight the need for empirical investigation of neuronal properties across species.

The present report directly compares high-resolution structural and electrophysiological data from L3 pyramidal neurons in slices prepared from mouse versus rhesus monkey across 2 functionally distinct cortices—the primary sensory visual and the higher order frontal cortices. We address several outstanding questions on areal- and species-specific cortical neuronal properties: What are the detailed topological features of dendritic arbors and spine distribution patterns of neurons in the 2 brain areas and species? Does dendritic scaling occur in L3 pyramidal neurons from mouse to rhesus monkey, and if so what is its nature? How do the physiological properties of neurons compare and contrast in the 2 areas and species? Understanding the diversity of individual pyramidal neuron properties across cortical areas and species is essential to understand how functionally specialized cortical networks encode information for complex behavior.

Materials and Methods

Experimental Subjects

Rhesus Monkey: A total of 12 adult (5–13 years of age) rhesus monkeys (Macaca mulatta) were used in this study. Rhesus monkeys were sacrificed as part of a larger ongoing study of cognition and normal aging in nonhuman primates. The subjects were obtained from the Yerkes National Primate Research Center at Emory University (Atlanta, GA, USA) and housed at Boston University School of Medicine in the Laboratory Animal Science Center (LASC).

Mouse: A total of 11 adult (1–3 months of age) CD1 mice obtained from Charles River Laboratories were used in this experiment. At 5–13 years of age (monkey) and 1–3 months of age (mouse), both species are considered “young adult.” While 1-month-old mice differ from 3-month-old mice in that they are not yet sexually mature, statistical analyses revealed no difference in any structural or physiological property of neurons from these age groups (Crimins et al. 2012) and, thus, data were pooled for analyses. Some of the subjects for this study were analyzed in our previously published datasets from rhesus monkeys and mice (Amatrudo et al. 2012; Crimins et al. 2012; Medalla and Luebke 2015). All research was conducted in strict adherence to animal care guidelines from the NIH Guide for the Care and Use of Laboratory Animals and the U.S. Public Health Service Policy on Humane Care and Use of Laboratory Animals. The Boston University School of Medicine LASC and the Emory University Yerkes National Primate Research Center are both accredited by the Association for Assessment and Accreditation of Laboratory Animal Care.

Cytoarchitecture of Brain Areas

For qualitative comparisons of cytoarchitecture, images from the Allen Brain Atlas (http://www.brain-map.org/) and from archives of rhesus monkey sections were examined. Images of Nissl-stained coronal sections through mouse frontal cortex (FC) and V1 were obtained from the Allen Brain Atlas. Coronal Nissl-stained whole-brain sections through rhesus monkey V1 (Brodmann's Area 17) and caudal LPFC (Brodmann's Area 46) were imaged at ×4 using a CCD camera-computer system (Surveyor Software, Objective Imaging) mounted on a Nikon E600 microscope. To measure the mean size of somata in layer 3 of the 4 brain areas, archived and newly processed araldite resin-embedded tissue blocks were examined. Rhesus monkey tissue blocks from V1 and LPFC (n = 3 cases, 5–6 years old) were obtained from the archives of Dr Alan Peters, which were processed for electron microscopy and embedded in araldite resin as described (Peters et al. 2008; Luebke et al. 2015). Mouse tissue blocks from FC and V1 (n = 3 cases, 2–3 months old) were processed similarly (Ludvigson et al. 2011). Semithin sections (1 µm thick) were cut from each resin block using an ultramicrotome (Leica) and a diamond histo knife (Diatome), and stained with Toluidine blue. Toluidine blue sections (2–3 sections per case) were imaged at ×40 and montaged (Surveyor Software). Layer 3 was delineated as: 250–500 µm deep from the pia for monkey V1, 300–700 µm deep for monkey LPFC, and 200–400 µm deep in the mouse V1 and FC, as measured in this study and based on previous studies (O'Kusky and Colonnier 1982; Barbas and Pandya 1989; Medalla and Barbas 2010; Van De Werd et al. 2010). Somata in layer 3 with an evident apical trunk and nucleus were selected to be measured for maximum diameters using Reconstruct (Fiala 2005).

Preparations of Brain Slices

Rhesus monkey: Subjects were tranquilized with ketamine (10 mg/mL), and then anesthetized with sodium pentobarbital (15 mg/kg, i.v. to effect). Subsequently, the head was secured with a head holder and a craniotomy was performed to expose the dura. The subjects were then sacrificed by exsanguination through infusion of the ascending aorta with ice-cold oxygenated Krebs-Henseleit buffer (concentrations, in mM: 6.4 Na₂HPO₄, 1.4 Na₂PO₄, 137 NaCl, 2.7 KCl, 5 glucose, 0.3 CaCl₂, 1 MgCl₂; pH = 7.4, chemicals from Sigma, St Louis, MO, USA). During perfusion, the dura was quickly cut to expose the brain, and multiple blockings cuts were made to remove 10 mm³ blocks of live tissue from the LPFC (caudal Brodmann's Area 46) and V1 (Brodmann's Area 17). The tissue blocks were then cut into 300-μm-thick slices with a vibrating microtome while in ice-cold oxygenated Ringer's solution (concentrations, in mM: 26 NaHCO₃, 124 NaCl, 2 KCl, 3 KH₂PO₄, 10 glucose, 1.3 MgCl₂; pH = 7.4, chemicals from Sigma). Slices were then placed in room temperature, oxygenated Ringer's solution to allow equilibration for 1 h. Following equilibration, each slice was placed in a submersion recording chamber (Harvard Apparatus, Holliston, MA, USA) on the stage of a Nikon E600 infrared-differential interference contract (IR-DIC) microscope (Micro Video Instruments, Avon, MA, USA), and secured by a nylon mesh while being superfused with oxygenated Ringer's solution (2–2.5 mL/min).

Mouse: Mice were deeply anesthetized with isoflurane and decapitated. Brains were rapidly removed and either the FC or occipital cortex (V1) was blocked from the brain. Blocks were then mounted against an agar slab for cutting of 300-μm-thick slices with a vibrating microtome in ice-cold oxygenated Ringer's solution. Slices from mice were subsequently handled in a manner identical to that used for rhesus monkey slices (above). In mouse experiments, pyramidal neuron recordings were obtained from layer 3 in the dorsal premotor area (area F2 in Paxinos and Franklin 2012) of frontal slices and from layer 3 of the primary visual cortex (Fig. 1). While rhesus monkey and mouse brain tissues were harvested differently preslicing (monkeys were perfused with Krebs solution and blocks removed, while mice were decapitated without exsanguination), we do not believe that this difference affected the outcome of our experiments. In both the mouse and the monkey, the time from sacrifice to obtaining viable slices was ∼5 min, neurons exhibited physiological properties that were well within the norm for healthy neurons and there was no evidence for morphological dystrophy (swelling or shrinkage which would be evident in the high-resolution confocal scans) in slices from either species.

Whole-Cell Patch-Clamp Recordings and Cell Filling

Whole-cell patch-clamp recordings were performed on L3 pyramidal neurons following identification under IR-DIC optics as described previously (Amatrudo et al. 2012; Crimins et al. 2012). Individual L3 pyramidal neurons were filled with 1% N-biotinyl-l-lysine (Biocytin, Sigma) during recordings. Pipettes were manufactured on a Flaming and Brown micropipette puller (Model P-87, Sutter Instruments, Novato, CA, USA). Each pipette was filled with potassium methanesulfonate-based internal solution (concentrations [in mM]: 122 KCH₃SO₃, 2 MgCl₂, 5 ethylene glycol tetraacetic acid, 10 Na4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, 1% biocytin, pH7.4 [chemicals from Sigma Aldrich]). Electrophysiology data were acquired using PatchMaster software on an EPC-9 or EPC-10 path-clamp amplifier (HEKA Elektronik, Lambrecht, Germany) and were imported into FitMaster software (HEKA Elektronik Lambrecht, Germany) for further analysis.

Assessment of Electrophysiological Properties

A series of 200-ms depolarizing and hyperpolarizing current steps was used to determine passive membrane properties (resting membrane potential, input resistance, and time constant), and single AP properties (threshold, amplitude, rise time, fall time, and duration at half maximal amplitude). Resting membrane potential (Vr) was determined as the voltage of the cell with no injected current. The input resistance (Rn) was calculated as the slope of the best-fit line of the voltage–current linear relationship. The membrane time constant (tau) was assessed by fitting a single exponential function to the membrane potential response to the −10 pA hyperpolarizing current step. Single AP properties were measured from the first AP elicited in the current-clamp series. Threshold was measured as the sharp positive voltage deflection. Amplitude was measured as the difference between the voltage at the peak of the AP and the threshold. Duration at half maximal amplitude was measured as the time between the upward and downward deflections at half the amplitude. Rise time and fall times were measured as the time from threshold to the peak and time from the peak to the threshold in the downward deflection, respectively. Rheobase was determined as the minimum current required to evoke a single AP during a 10-s depolarizing current ramp (0–200 pA). Sag potential was measured as the difference between the initial hyperpolarizing voltage and the steady-state voltage response to a −170-pA current step. Repetitive AP firing was assessed using a series of 2 s hyperpolarizing and depolarizing current steps (−170 to + 380 pA, using either 20 or 50 pA increments). sEPSCs were recorded by holding cells at −80 mV for 2 min and were exported from FitMaster and analyzed with the MiniAnalysis software (Synaptosoft). Synaptic event detection was established as the maximum root mean squared noise level (5 pA). Frequency, amplitude, area, rise time, decay time, and time at half width of sEPSCs were measured for each neuron.

Slice Processing and Alexa-Streptavidin Labeling of Biocytin Filled Neurons

Neurons filled with biocytin during electrophysiological recordings were labeled using streptavidin fluorescent conjugates for visualization. Slices were fixed in 4% paraformaldehyde in 0.1 M phosphate buffer (PB) at pH 7.4 and kept at 4°C for 2 days. Slices were then permeabilized with 1% Triton X-100 in phosphate-buffered saline (PBS, 0.1 M PB with 0.9% NaCl, pH 7.4) at room temperature for 2 h and were subsequently incubated with streptavidin-Alexa Fluor 488 (1:500, Life Technologies) at 4°C for 2 days. Slices were then rinsed thoroughly in PBS and wet mounted on glass slides using Prolong Gold anti-fade mounting medium (Life Technologies). The mounting medium was allowed to cure at room temperature for at least 1 week before imaging.

Confocal Imaging and Preprocessing of Image Stacks

Filled pyramidal cells labeled with Alexa 488 were imaged using a Zeiss 510 or Zeiss 710 confocal laser-scanning microscope with a ×40 1.3NA oil immersion objective lens and an argon laser (488 nm excitation). Filled neurons were scanned at a resolution of 0.1 × 0.1 × 0.2 μm. Raw image stacks were deconvolved using the AutoQuant X3.0.3 Software (Media Cybernetics, Bethesda, MD, USA), to reduce signal halo inherent in the z-plane during confocal imaging then converted from 16 to 8 bit. Confocal Z-stack tiles were aligned and merged using Volume Integration and Alignment System (VIAS) software (Rodriguez et al. 2003) to create a single integrated image stack for each cell.

Dendritic Reconstruction and Analysis of Spine Number and Density

Integrated image stacks of whole cells were loaded into NeuronStudio 64-bit (Rodriguez et al. 2003, 2006, 2008), to reconstruct total, apical, and basal dendritic arbors. Semiautomated dendritic reconstructions were performed using the Rayburst algorithm method in NeuronStudio (Rodriguez et al. 2006, 2008), and each dendritic reconstruction was then manually checked and edited for errors. Typically, the semiautomated reconstruction algorithm correctly identified ≥90% of filled dendrites, missing only very thin distal processes or processes with suboptimal signal-to-noise; these processes were easily identified and corrected manually. Total, apical and basal dendritic lengths, diameters, and branch points were obtained from each .swc file generated by the reconstruction. Following dendritic reconstruction, spines across entire arbors were manually marked using NeuronStudio spine counting tool, and spine numbers and densities (spines per micron of dendrite) were calculated. Sholl analyses of apical and basal arbors were employed using concentric spheres placed at 10-μm increments from the soma (Sholl 1953) to determine the dendritic length, volume, diameter, branch points, and spine count as a function of distance from the soma.

Neuronal Inclusion Criteria

For inclusion in electrophysiological analyses, neurons were required to have a resting membrane potential of less than or equal to −55 mV, a stable access resistance, an AP overshoot (>0 mV), and the ability to fire repetitive APs in response to prolonged depolarizing current steps. For morphometric inclusion, neurons were required to have complete somata and dendritic fills with no cut branches in the proximal third of the apical dendritic tree. Given that neurons were filled in slices rather than in vivo, it was unavoidable that some distal dendritic branches were cut in the z-plane. However, to be included in the study neurons were required to be positioned near the middle of the slice in the z plane, thus limiting the number of cut dendritic branches.

Statistical Analysis

All data were compiled in Microsoft Excel spreadsheets to calculate the mean, standard deviation, and standard errors. For each variable, differences across the 4 groups were assessed using analysis of variance (ANOVA) with Tukey's or Fisher's least significant difference post hoc test for multiple comparisons in SPSS (v 19, IBM Company). Relationships between variables were determined using a Pearson product–moment correlation. Validity for using these linear regression analyses was tested by calculating P values and plotting residuals for each pair of variables. Significance was set at α = 0.05 for all statistical tests. To assess (dis)similarities across the 4 groups of neurons based on combined multivariate effects, we employed multidimensional hierarchical clustering and linear discriminant analyses using SPSS. Hierarchical cluster analysis (HCA) was employed to determine global similarities and differences among 4 groups of neurons—mouse V1, mouse FC, rhesus monkey V1, and rhesus monkey LPFC—based on 3 sets of multiple electrophysiological and morphological variables (a total of 73 parameters measured) combined: 1) intrinsic properties (passive and excitability properties) together with dendritic morphological properties, 2) synaptic physiological properties with spine morphological properties, and 3) all electrophysiological and morphological properties combined. The HCA employed squared data (dis)similarity matrices derived from pairwise comparisons of mean electrophysiological and morphological parameters across 4 groups by Pearson's correlation, as described previously (Dombrowski et al. 2001). HCA hierarchically groups areas based on (dis)similarities in their parameter profiles, which are interpreted as spatial distances and plotted as a cluster tree diagram. The distance between 2 branching points in a cluster tree diagram represents the relative similarity of the groups; the longer the interbranch distance, the more dissimilar are the subgroups. Discriminant analysis (DA) was employed to assess how well each measured morphological and physiological variable segregated the 4 groups of neurons and predicted group membership of each neuron. The stepwise DA was run to identify which experimental variables produced significant discriminant functions (group separators) and showed the clearest separation of individual data points belonging to different groups (mouse V1, mouse FC, rhesus monkey V1, rhesus monkey LPFC). A canonical correlation coefficient, r, and Wilks' λ were used respectively as measures of goodness of fit and significance of the discriminant functions for group separations based on each variable set. A discriminant score for each variable based on each discriminant function was calculated, which determines the variable's ability to predict group membership. The highest discriminant score for each function represents the best predictor of group membership.

Results

Experimental Subjects and Cytoarchitecture of Brain Areas Assessed

Whole-cell patch-clamp recordings with simultaneous biocytin filling were obtained from L3 pyramidal neurons in primary visual and frontal association cortices of adult mice and rhesus monkeys. Figure 1 shows the brain areas from which blocks of tissue were obtained and slices prepared, with Nissl-stained sections demonstrating distinctive thickness of the gray matter and laminar arrangements between areas and species (Fig. 1A,B). Semithin (1 µm) sections from each brain area were stained with toluidine blue to reveal neuronal somata and the surrounding neuropil in layer 3 of the 4 areas (Fig. 1C). The maximum diameters of somata measured from these sections exhibited significant differences (P < 0.01 for all comparisons; ANOVA with Tukey's Post hoc test). Monkey LPFC L3 neurons had the largest somata (21.36 ± 0.18 µm, n = 298 neurons; 2 cases) followed by mouse FC neurons (16.98 ± 0.23 µm, n = 97 neurons; 2 cases), mouse V1 neurons (15.60 ± 0.27 µm, n = 52 neurons; 2 cases), and, finally, monkey V1 neurons (14.11 ± 0.11 µm, n = 260 neurons; 2 cases). These data are consistent with measurements obtained from biocytin-filled neurons from which recordings were obtained, although with this lower n statistical significance was achieved for some but not all comparisons (Table 1).

Areal Differences in Morphological Properties of L3 Pyramidal Neurons in Rhesus Monkey but not in Mouse

Dendritic Topology

Figure 2 shows representative reconstructions of L3 pyramidal neurons from the mouse and monkey visual and frontal cortices. All neurons fell within the boundaries of layer 3 for each cortical area and species (see Materials and Methods and Table 1). It is immediately evident that neurons were markedly different in structure between V1 and LPFC in the monkey, but not between V1 and FC in the mouse. Quantitative dendritic data are presented in Figure 3 and in Table 1. Mouse V1 and FC neurons and monkey V1 neurons did not differ from each other in terms of mean total, mean apical, and mean basal dendritic lengths but they were each significantly smaller than monkey LPFC neurons (Fig. 3B, top; P < 0.01). Branch point analyses revealed that mouse V1 and FC neurons did not differ significantly from each other or from monkey LPFC neurons in terms of their dendritic complexity. Monkey V1 neurons had fewer mean total branch points and thus were less complex than mouse V1 (P = 0.095), mouse FC (P < 0.012) and monkey LPFC (P < 0.002) neurons (Fig. 3B, bottom).

Sholl analyses of basal dendritic lengths showed a consistent pattern of dendritic lengths across the arbors of each of the four groups, peaking in the proximal half and then decrementing (Fig. 3C, bottom). Similar analyses of the apical arbor revealed an unexpected and important difference between mouse and monkey neurons (Fig. 3C,D, top; Fig. 4). In all of the mouse neurons (both V1 and FC), the apical dendritic arbor exhibited 2 clear apical compartments, consisting of an area of dense proximal oblique branches and a distal apical tuft area separated by a portion of the main apical trunk (ca. one-third) essentially devoid of dendritic branches (Fig. 4). In contrast, in monkey neurons (both V1 and LPFC) oblique branches were relatively uniformly distributed across the apical trunk. This finding has significant implications for differences in the compartmental signal integration by pyramidal neurons in the 2 species.

Finally, dendritic diameters of the apical trunk, and single apical oblique, apical tuft, and basilar dendritic branches were measured for each neuron and compared between the 4 groups (Table 1). There was no significant difference in mean diameter of any of the dendritic compartments in V1 neurons between species. There was also no difference in the mean diameters of basal, apical trunk, or oblique branches for the mouse versus monkey frontal cortical neurons; however, in the apical tuft region, dendritic diameter was significantly lower in monkey than in mouse neurons (P < 0.01). This species difference was accounted for by the fact that approximately half of the monkey neurons had very thin apical tuft branches, while the other half had branches with diameters approximating those of mouse frontal cortical neurons.

Spine Distribution

As with dendritic length and complexity, there was no difference in the mean spine number or density across either the total, apical, or basal arbors in mouse neurons from the 2 brain areas (Fig. 5A). In contrast, the mean number and density of total, apical, and basal spines was far greater on monkey LPFC than on V1 neurons, as we and others have previously reported (Elston and Rosa 1997; Elston 2003; Amatrudo et al. 2012; Medalla and Luebke 2015). Indeed, compared with all other groups, monkey V1 neurons possessed the lowest number and density of spines in both the apical and basal arbors (Fig. 5A; Table 1; P < 0.05). Interestingly, spine densities did not significantly differ between monkey LPFC and mouse FC and mouse V1 neurons (Fig. 5A). Sholl analyses of spine number mirrored the pattern seen for dendrites themselves (Fig. 5B,C). In the apical arbors of mouse neurons, there were 2 compartments (proximal oblique and apical tuft) in which spine number was high, separated by an area of few/no oblique branches and commensurately low spine number. In monkey neurons, the number of spines was highest proximally but reached a steady asymptote beginning in the middle third of the apical arbor. In contrast, spine density was uniform across both basal and apical arbors of all neurons, with a spine free ∼30-µm proximal area followed by consistent densities across the rest of the arbors.

Areal Differences in Electrophysiological Properties of L3 Pyramidal Neurons in Rhesus Monkey but not in Mouse

Intrinsic Membrane Properties

Mouse V1 and FC neurons were virtually identical with regard to an array of intrinsic membrane properties, while monkey V1 and LPFC neurons were highly distinctive (Fig. 6; Table 2). Input resistance (Rn) did not differ between mouse V1 and FC neurons, but differed markedly between monkey V1 and LPFC (P < 0.001). The mean Rn of monkey V1 neurons was similar to that of mouse V1 and FC neurons, while that of monkey LPFC neurons was approximately half (Fig. 6A,C). Consistent with this finding, the rheobase for monkey LPFC neurons was much higher than that for mouse V1, mouse FC and monkey V1 neurons, which did not differ from each other (Fig. 6C). The resting membrane potential was similar in the 4 groups of neurons (Table 2).

AP firing rates evoked by a series of 2-s depolarizing current steps did not differ between mouse V1, mouse FC and monkey LPFC but were nearly twice as high in monkey V1 neurons (Fig. 6B,D, left). In neurons from each species, there was a significant linear correlation between Rn and AP firing rate (Fig. 6D, right; r² = 0.29 for both correlations, P < 0.01).

Spontaneous Excitatory Postsynaptic Current (EPSC) Properties

Mouse FC and V1 neurons did not significantly differ from each other with regard to any spontaneous EPSC property (Fig. 7; Table 2). Monkey LPFC neurons resembled both V1 and FC mouse neurons with regard to amplitude and kinetics but exhibited a significantly lower frequency of events compared with mouse FC (but not V1) neurons. Monkey V1 neurons exhibited a significantly lower frequency and amplitude and shorter decay time of events than any other group (P < 0.05). The mean sEPSC rise time in monkey V1 neurons was also significantly shorter than monkey LPFC neurons but did not differ from mouse neurons.

Relationships Between Structural and Electrophysiological Features of L3 Pyramidal Neurons

Linear correlation analyses were employed to assess relationships between morphological and electrophysiological variables. First, we determined whether any of the dendritic morphological variables were correlated with the size (major diameter) of the soma. Soma diameter positively correlated with total (r² = 0.56; Fig. 8A), apical (r² = 0.51) and basal (r² = 0.49) dendritic length (P < 0.01 for all significant correlations). Soma diameter was also positively correlated with the diameter of the apical trunk (r² = 0.34, P < 0.01) but not with the diameter of apical oblique, apical tuft or basal dendritic branches. Two sets of significant correlations (P < 0.01) were observed between morphological and electrophysiological variables. Negative correlations between both soma diameter and apical trunk diameter with Rn were observed (r² = 0.22 and r² = 0.28, respectively; Fig. 8B). Similar negative relationships were seen between Rn and dendritic length (total r² = 0.43; apical r² = 0.41; basal r² = 0.35) and dendritic branch number (total r² = 0.42; apical r² = 0.35; basal r² = 0.32). Thus, the larger and the more complex the neuron, the lower the Rn. Spine density was significantly positively correlated with both the frequency (r² = 0.25) and the amplitude (r² = 0.41) of spontaneous EPSCs.

Discriminant and Hierarchical Cluster Analyses

DA was employed to assess how well each set of morphological and electrophysiological variables (73 parameters in total, which represented subcategories of the 40 major variables assessed) predicted group membership and segregated the 4 groups of neurons. DA was employed for each set of variables—intrinsic membrane properties, synaptic current properties, dendritic topology, and spine distribution properties, using stepwise correlative methods (to determine which variables are most correlated with discriminant scores, Fig. 9A). Among each group of variables, stepwise DA revealed that the following variables were the most significantly correlated and the most reliable discriminators of group membership: 1) for intrinsic membrane and AP firing properties: Tau, Rn, AP rise time, depolarizing sag amplitude and repetitive AP firing rates at 80 pA; 2) for synaptic current properties: EPSC frequency, amplitude, rise time and area; 3) for dendritic topology: total dendritic length, apical dendritic length and branch points, length of the mid-apical segment, and basal dendrite vertical extent; and 4) for spine properties: mid-apical spine number, and basal distal spine density. Importantly, a scatter plot of group membership based on discriminant functions for each set of variables showed more segregation of data points between monkey V1 versus LPFC than between mouse V1 versus FC (Fig. 9A), as expected.

To assess the (dis)similarities across the 4 groups of neurons, when all 73 electrophysiological and morphological variables were considered in concert in a multidimensional scale, HCA was performed (Fig. 9B). Hierarchical clustering based on all mean morphological and electrophysiological properties supported results from multifactorial ANOVA comparison and revealed that, indeed, monkey V1 is the most different group (Fig. 9B, left). Dendrogram plots based on squared Euclidean distances showed that mouse V1 and FC were the most similar to each other, having an interbranch distance smaller (d² = 36.36) than between than monkey LPFC and V1 (d² = 185.1). Moreover, monkey V1 branches off earliest in the dendrogram and has the largest branching distance, and is therefore the most dissimilar across the 4 groups. These results were verified in a DA that was run for a subset of cells wherein a complete set of morphological and electrophysiological variables were measured (n = 24 neurons total; Fig. 8B, right). Taking all morphological and electrophysiological variables together, the DA significantly discriminated the neurons into 3 groups, with apical oblique dendritic length, apical number of branch points, and mid-apical spine density being the best discriminators. Consistent with the HCA, the DA showed that monkey V1 versus LPFC neurons segregated from each other to a greater degree than mouse V1 and FC neurons. Further, the monkey V1 neurons formed the most segregated group.

Discussion

Pyramidal neurons constitute a majority of the fundamental building blocks of cortical microcircuits, which in turn endow different cortical areas with the capacity to perform a broad range of highly distinctive functions. While it has been suggested that cortical circuits possess emergent properties that are largely independent of individual neuron properties (Alivisatos et al. 2012), little direct empirical evidence supports this view. Indeed, an important challenge in the field is to gain a broader and more detailed understanding of how pyramidal neuron heterogeneity across different areas and species impacts individual neuron, microcircuit, and ultimately areal functional specialization (Jacobs and Scheibel 2002; Elston 2003; Spruston 2008; Elston and Fujita 2014 for reviews). In the present study, comprehensive characterization of L3 pyramidal neurons in primary visual and frontal cortical areas in mouse and rhesus monkey underscores 2 fundamental principles of cortical organization. First, V1 and FC neurons are remarkably similar with regard to nearly every property in the mouse, while the opposite is true in the monkey, with V1 and LPFC neurons exhibiting significant differences in nearly every property assessed. Second, neurons within visual and frontal areas differ significantly between the mouse and the monkey. Thus, while neurons in the mouse and monkey area V1 are the same size, they differ in nearly every other way, and frontal cortical neurons in the mouse are much smaller than those in the monkey and also differ substantially with regard to wide range of structural and functional properties.

Within-species Comparison of Functionally Distinct Cortical Areas

Our findings revealed that the structural and functional properties of L3 pyramidal neurons in V1 cortex versus FC are markedly heterogeneous in the monkey but homogeneous in the mouse. In a series of previous studies, we and others have established that L3 pyramidal neurons in V1 and LPFC of the monkey are highly distinctive across a broad spectrum of structural and functional properties (Elston 2000, 2002; Elston et al. 2001; Amatrudo et al. 2012; Medalla and Luebke 2015). The dendritic arbors of LPFC neurons are ∼3.4× larger and possess a 2.3× higher density of spines than V1 neurons. Further, there is a significantly higher proportion of large dendritic spines and large excitatory synapses on LPFC neurons compared with V1 neurons and a significantly higher number of perforated synapses in LPFC neuropil (Medalla and Luebke 2015). Functionally, LPFC neurons in the rhesus monkey are far less excitable, in terms of evoked AP firing rates, than are V1 neurons, likely due in large part to their significantly lower input resistance. Excitatory synaptic currents also differ markedly, with spontaneous and miniature AMPAR-mediated synaptic currents occurring at a significantly higher frequency and amplitude in LPFC than in V1 neurons (Amatrudo et al. 2012; Medalla and Luebke 2015). Neuroanatomical tracing studies show that many more areas provide convergent input to LPFC than to V1 in the monkey (Perkel et al. 1986; for review: Felleman and Van Essen 1991; Luebke et al. 2010; and see also the macaque macrconnectivity database: http://cocomac.g-node.org/main/index.php). Thus, electrically compact and highly excitable V1 neurons receive fewer, smaller amplitude inputs while the much larger and less excitable LPFC neurons receive more numerous inputs with broader range of synaptic strength and temporal scale. The large difference in the biophysical synaptic and excitable properties of V1 versus LPFC neurons in monkeys suggest very distinct intrinsic temporal dynamics of signal processing by these 2 cortices. Indeed, physiological studies in awake behaving rhesus monkeys show that the relative timing of signal processing within V1 is much faster than that of LPFC (Murray et al. 2014; Chaudhuri et al. 2015). These marked differences in the fundamental units of computation—the pyramidal neurons—in these brain areas likely contribute to fulfilling the different requirements for signal processing by these areas.

Here, we sought to determine whether L3 pyramidal neurons in the visual cortex versus FC in the mouse also exhibit distinctive properties. In stark contrast with the differences observed between these 2 cortical areas in the rhesus monkey, L3 pyramidal neurons in the mouse V1 and FC exhibited only modest differences in structural properties—V1 neurons being slightly smaller than FC—and were nearly identical with regard to physiological features. In contrast to the rhesus monkey, the similarity in the structural and biophysical properties of V1 and FC neurons in the mouse suggests that relatively similar temporal dynamics may exist within these 2 cortical areas in the mouse. These findings are consistent with the idea that in the mouse, where L3 pyramidal neurons do not differ in these 2 functionally distinct brain areas, a “prototypical” pyramidal neuron may form a key constituent of the neuronal circuitry. On the other hand, in the rhesus monkey this is clearly not the case; cortical areas differ markedly from each other at the individual pyramidal cell and hence likely at the circuit level. That said, others have demonstrated differences in the basilar arbors of pyramidal neurons in different mouse cortical areas (Ballesteros-Yanez et al. 2006; Benavides-Piccione et al. 2006). However, these differences are less pronounced than the differences in basilar arbors of neurons seen across areas in the rhesus monkey as shown by the same group (review: Elston 2003) and as shown in our data here. This raises the important question of the functional relevance of such distinctive specialization of neurons between V1 and LPFC in the primate but not between V1 and FC in the mouse.

Higher order association frontal cortices are much more developed in primates than in mice, enabling primate specific cognitive and behavioral capabilities not present in mice (reviews: Fuster 2001; DeFelipe 2011; Keeler and Robbins 2011). The granular (well-defined layer 4) primate LPFC in particular does not have a homolog in the rodent brain, since the rodent FC is dysgranular/agranular (poorly defined or absent layer 4) (review: Preuss 1995; Uylings et al. 2003; Elston and Garey 2009, 2013; Barbas 2015). A widely held assumption is that the “prefrontal” cortex in the mouse is composed of the medial infralimbic and prelimbic cortices, which correspond to primate dysgranular/agranular medial and orbital prefrontal areas (review: Uylings et al. 2003; Seamans et al. 2008; Barbas 2015). How pyramidal neurons within these medial and orbital frontal areas compare across species should be addressed in future studies. As with the FC, the mouse primary visual cortex is dysgranular, while that of the monkey is highly granular. In fact, this difference in granularity between species is even more dramatic in V1, since this area possesses a prominent and highly differentiated layer 4 in primates, including the rhesus monkey (O'Kusky and Colonnier 1982; Peters et al. 1994). The primary visual cortex indeed differs functionally in rodents and primates, particularly in terms of greater spatial resolution of visual responses and in the proportions of neurons responsive to different visual stimuli in rhesus monkeys compared with mice (Wurtz and Mohler 1976; Niell and Stryker 2008, 2010; Garrett et al. 2014; Juavinett and Callaway 2015; review: Rosa and Tweedale 2005; Van Hooser 2007; Laramee and Boire 2014; Solomon and Rosa 2014). The relative cellular uniformity observed between mouse FC and V1 supports the idea of a greater degree of cytoarchitectural and functional similarity across mouse cortical areas, compared with the highly specialized cortical areas of the primate brain. Mouse V1 has numerous reciprocal connections with nonvisual cortical areas such as primary somatosensory, rhinal, cingulate, and retrosplenial cortices (e.g., Wang et al. 2012). In contrast, in the rhesus monkey nonvisual connections with V1 are rare. Thus, while in the nonhuman primate, V1 has been demonstrated to have connections with area prostriata (marmoset monkeys; Yu et al. 2012), and with auditory areas (cynomolgus monkeys; Falchier et al. 2002), the great majority of connections are restricted to unimodal extrastriate visual areas (reviewed in: Felleman and Van Essen 1991; http://cocomac.g-node.org/main/index.php). It is also important to bear in mind that it is possible that higher segregation of cortical areas is a requirement of brain expansion (Ringo 1991). Indeed, small primates such as the marmoset are likely to have less differentiated cortical areas than rhesus monkeys (Palmer and Rosa 2006), while human cortices are likely more differentiated than those of rhesus monkeys. Consideration of interareal and interspecies areal differences such as these, and specialization of individual neuron properties will enable the construction of realistic models of the neural circuits required for species-specific cognitive and visual function.

Between-Species Comparison of L3 Pyramidal Neurons in Visual and Frontal Cortices

Dendritic Scaling

Since the beginning of the modern era of neuroscience, investigators have sought to understand the relationship between brain size and brain capacity, as well as how increased brain size comes about during evolution. It has been widely believed that, in addition to possessing greater numbers of neurons, larger brains possess neurons with larger dendritic trees; in other words, that neurons “scale up” in size (reviews: Harrison et al. 2002; Wittenberg and Wang 2008; DeFelipe 2011). Indeed, this seems to be the case for many, but not all, neuron types. For example, while pyramidal neurons in the hippocampus (Bekkers and Stevens 1990; Buckmaster and Amaral 2001) and entorhinal cortex (Buckmaster et al. 2004) and some regions of the neocortex (Cajal 1894; Barasa 1960; Mohan et al. 2015) do scale up from rodent to primate, cerebellar granule cells (Bekkers and Stevens 1990) and dentate granule cells (St John et al. 1997) do not. Interestingly, Mohan et al. (2015) recently reported that L3 pyramidal neurons in the temporal lobe scale up from the mouse to the human (but not to the rhesus monkey) brain. Mouse and human neurons in this study were completely filled during electrophysiological recordings by these investigators, enabling direct comparisons. In contrast, rhesus monkey neurons, obtained from an online archive (NeuroMorpho.org), were filled in lightly fixed tissue slices, which likely led to underestimates of their size. Further, in the rhesus monkey, pyramidal cells in layer 3 of inferior temporal cortex (area TE) are much larger than those in V1 (Elston and Rosa 1997); this together with the present findings make it unlikely that these neurons are the same size in mouse and rhesus monkey temporal cortex. In the present study, we demonstrated that L3 pyramidal neurons do not scale in size from the mouse to the monkey V1 but that frontal pyramidal neurons do. Physiological properties were highly distinct in mouse compared with monkey V1 neurons, but largely conserved in mouse compared with monkey frontal neurons, with few differences seen in intrinsic membrane or synaptic response properties. Computational modeling and further empirical experiments are needed to fully understand the relationship between these morphological and electrophysiological properties in individual neurons, and how distinctive properties of these neurons impact signaling by the cortical networks to which they contribute (Ascoli 2003).

The relative differences in scaling and electrophysiology of frontal versus visual L3 pyramidal neurons provide insight into potentially differential capabilities of these neurons in the 2 species. As frontal cortical pyramidal neurons increase in size across phylogeny, the opportunity for convergence of diverse inputs is increased, as is their integrative and computational dynamic range. Synaptic integration and filtering occur as a function of number of dendritic branch points and both diameter and length of dendritic segments across the arbor (Rall 1962, 1964). Importantly, while the dendritic arbors of LPFC neurons were much larger than those of FC, the diameters of individual dendritic branches were not greater. Indeed, monkey LPFC neurons possessed, on average, significantly smaller diameter dendrites, due to the presence of a number of neurons with especially thin dendrites. To preserve electrotonic properties in scaling (termed conservative scaling), a doubling of the length of a dendrite requires a 4-fold increase in diameter (Bekkers and Stevens 1990); since this does not occur the electrotonic properties of LPFC and FC neurons must differ substantially. Scaling and cable theory predict that monkey LPFC neurons filter input signals to a greater extent than mouse FC neurons due to their greater dendritic length and equivalent or reduced dendritic diameters. Consistent with this idea, we observed a higher frequency of spontaneous EPSCs in mouse FC versus monkey LPFC. Moreover, input resistance, which is related to dendritic length, proximal apical trunk diameter, and soma size, is much lower in monkey LPFC neurons compared with mouse FC. However, overall, we observed that the AP firing and other intrinsic properties are largely preserved between mouse and monkey frontal neurons, suggesting significant roles for nonpassive properties (e.g., active conductance and intact synaptic inputs) that should be examined in future studies (Nusser 2012).

In contrast to frontal neurons, V1 neurons did not exhibit dendritic scaling, but differed substantially with regard to a broad array of physiological and spine morphometric properties in the 2 species. Interestingly, discriminant and hierarchical cluster analyses based on multidimensional electrophysiological and morphological variables showed that monkey V1 neurons are by far the most distinctive of the 4 groups of neurons examined. For instance, while AP firing rates are similar in mouse V1 and FC and monkey LPFC they are almost twice as high in monkey V1. Spontaneous EPSCs are the lowest in terms of frequency and amplitude in monkey V1, which is consistent with the finding that these neurons had the lowest spine density. Thus, monkey V1 neurons are highly excitable yet have sparse synaptic activity compared with monkey LPFC and mouse neurons.

Dendritic Compartmentalization

In the present study, we discovered a fundamental difference in the topology of the apical dendritic arbors of L3 neurons in the mouse versus monkey cortex. In mouse pyramidal neurons, few if any oblique branches emanate from the mid- to distal apical trunk, whereas monkey neurons typically possess relatively uniformly distributed oblique branches originating across the entirety of the apical trunk. This important difference in dendritic topology and compartmentalization has significant implications for integration of synaptic inputs by mouse compared with monkey pyramidal neurons. Structurally distinct dendritic domains confer a broad range of functional compartmentalization of distinct afferent inputs. Apical trunk, apical oblique, apical tuft, and basilar dendrites each possess their own distinct passive and active signal filtering and boosting capacities (Larkum et al. 2001, 2009; London and Hausser 2005; Losonczy and Magee 2006; Losonczy et al. 2008; review: Spruston 2008). Insight into the importance of dendritic compartmentalization is provided by Schaefer et al. (2003) in which simulations predict that the ratio of proximal over distal oblique branches strongly influences the coupling between somatic and dendritic spikes in pyramidal neurons (Schaefer et al. 2003). In this simulation, insertion of oblique branches of the main apical dendrite close to the soma (<140 µm) increased coupling between apical dendrite and AP initiation zones, whereas insertion of oblique branches distally decreased this coupling. Thus by virtue of their distinct somatodendritic compartments, pyramidal neurons can act as coincidence detectors capable of a great range of integration of temporally and spatially unique synaptic signals from diverse inputs. Even minor differences in branching characteristics can exert a major influence on signal processing—for example, differences in the geometrical features of branch points (at which impedance mismatch is great), which are known to vary widely in apical arbors, likely play an important role in tuning neuronal integration of input signals and excitability in CA1 pyramidal neurons (Ferrante et al. 2013). Indeed, modest variations in branch point morphology are capable of transforming the electrical relationship between oblique dendrites and the main apical shaft dendrite from fully coupled to fully compartmentalized (Ferrante et al. 2013). Thus, the structural differences between mouse and monkey pyramidal neurons reported here have significant implications for signal processing at both the individual neuron and network level.

Concluding Comments

With the advent of large-scale brain mapping initiatives such as the Human Brain Project and the BRAIN Initiative, which have the ultimate aim of understanding the human brain, it is increasingly important to ascertain which, if any, data on the fundamental features of rodent neurons can be extrapolated to the primate and ultimately to the human brain. The vast majority of what we know about the structure and function of pyramidal neurons has been obtained from the juvenile rodent somatosensory (barrel) cortex, and these data constrain at least one important large-scale model of cortical microcircuitry (Markram et al. 2015). Data from the present study support the growing view that, because constituent neurons can vary widely from area to area and across phylogeny, a single prototypical cortical column cannot be constructed and merely multiplied by the thousands to create a brain capable of vision, motor function and the extraordinarily complex cognitive capabilities evinced by the nonhuman primate and humans. Instead, it is vital that the cortical microcircuits and columns being constructed and modeled be constrained by realistic pyramidal neuron building blocks.

Funding

This work was supported by National Institutes of Health: the National Institute on Aging (NIA) (grants P01-AG00001, R01 AG025062, R01 AG035071) and the National Institute of Mental Health (NIMH) (grant K99 MH101234).

Notes

We thank Dr Kathleen Rockland and Teresa Guillamon-Vivancos for helpful comments on the manuscript. Conflict of Interest: None declared.

References

Author notes

Joshua P. Gilman and Maria Medalla are co-first authors.