|Home | About | Journals | Submit | Contact Us | Français|
Local hyperconnectivity in the neocortex is a hypothesized pathophysiological state in autism spectrum disorder (ASD). MET, a receptor tyrosine kinase that regulates dendrite and spine morphogenesis, has been established as a risk gene for ASD. Here, we analyzed the synaptic circuit organization of identified pyramidal neurons in the anterior frontal cortex of mice with a dorsal pallium derived, conditional knockout (cKO) of Met. Synaptic mapping by glutamate uncaging identified layer 2/3 as the main source of local excitatory input to layer 5 projection neurons in controls. In both cKO and heterozygotes this pathway was stronger by a factor of ~2. This increase was both sub-layer and projection-class specific, restricted to corticostriatal neurons in upper layer 5B, and not neighboring corticopontine neurons. Paired recordings in cKO slices demonstrated increased unitary connectivity. We propose that excitatory hyperconnectivity in specific neocortical microcircuits constitutes a physiological basis for Met-mediated ASD risk.
Genetic factors play an important role in the etiology of autism spectrum disorder (ASD). Accumulating evidence supports an association of disrupted MET receptor tyrosine kinase-mediated signaling and ASD risk. MET is a pleiotropic factor in the ontogenesis of multiple organs and a key signaling component in a number of neurodevelopmental events (Bronner-Fraser, 1995; Maina et al., 1998; Gutierrez et al., 2004; Judson et al., 2010). We first described the relevance of human MET gene (MIM 164860) to ASD by reporting that a functional single nucleotide polymorphism (SNP) in the MET promoter region confers genetic risk for ASD (Campbell et al., 2006). This and other ASD-associated promoter variants and copy number variants (Campbell et al., 2009; Jackson et al., 2009; Sousa et al., 2009; Thanseem et al., 2010) and the observed reduction in MET protein in postmortem neocortex from ASD subjects (Campbell et al., 2007) have since been substantiated in multiple additional cohort studies.
We recently investigated neurodevelopmental features of Met function in the context of forebrain circuitry related to ASD etiology. Met is expressed transiently in the postnatal mouse forebrain, particularly cortical, amygdala and hippocampal projection neurons, with a peak during the second postnatal week (Judson et al., 2009), when rates of dendritogenesis and synaptogenesis are maximal (Blue and Parnavelas, 1983; Micheva and Beaulieu, 1996). Additionally, in a conditional mutant (Met-cKO) of the mouse dorsal pallium, generated by crossing “floxed” Met line with an Emx1cre driver line (Gorski et al., 2002; Judson et al., 2009; Judson et al., 2010), Experimental Procedures) neocortical neuron architecture is altered. Specifically, the proximal dendritic arbors are increased and distal arbors are reduced (Judson et al., 2010). In addition, spine volume is increased by 20%. These morphological changes suggest a potential substrate for cortical circuit dysfunction in the Met-cKO mice.
It has been hypothesized, based on functional imaging studies in ASD, that cortical dysfunction arises both from long-range ‘under-connectivity’ between cortical areas and from short-range ‘over-connectivity’ within areas, especially in frontal and temporal cortex (Frith, 2004; Just et al., 2004; Courchesne and Pierce, 2005; Geschwind and Levitt, 2007; Just et al., 2007; Scott-Van Zeeland et al., 2010). However, a neural correlate for this altered connectivity, such as synaptic ‘hyperconnectivity’ or other functional changes in local circuits, has not been identified in human cortex. Given the role of MET in the genetic architecture of ASD risk and in regulating synapse-related neuronal structures, we investigated the functional implications of disrupted Met signaling on cortical circuit organization using Met-cKO mice. We focused on characterizing excitatory synaptic inputs to major types (corticostriatal and corticopontine) of cortical output neurons in the anterior frontal cortex (AFC), and surveyed the local circuits with synaptic mapping methods. We found that disrupted Met signaling that resulted in structural changes also generated robust pathophysiological correlates that may be relevant for understanding the etiology of ASD.
Mice were bred in-house by mating one parent homozygous for a Met allele in which exon 16 is flanked by loxP (floxed) sites (Metfx/fx, provided by S. Thorgeirsson; (Judson et al., 2009)) with another parent carrying an Emx1-cre transgene (provided by K. Jones; (Gorski et al., 2002)) and a single floxed Met allele (Metfx/+/Emx1cre). The Metfx/fx and Metfx/+/Emx1cre breeding lines were both fully back-crossed onto the C57BL/6J background. The genotypes of their offspring (i.e., Metfx/fx/Emx1cre, Metfx/+/Emx1cre, Metfx/fx/WT; Metfx/+/WT) were determined by PCR as described previously (Judson et al., 2009) or with real-time PCR by Transnetyx (Cordova, TN). Electrophysiological data collected at different institutions were consistent with each other. Data from male and female mice were pooled, as initial analysis of data from individual animals did not reveal any sex differences. All procedures using mice were approved by the Institutional Animal Care and Use Committee of University of Southern California, Vanderbilt University, or Northwestern University and conformed to NIH guidelines.
Retrograde labeling was performed in anesthetized mice as described (Anderson et al., 2010). Contralaterally projecting corticostriatal neurons in AFC were selectively labeled by stereotaxically injecting ~50 nL of red or green fluorescent microspheres (Lumafluor) into the (left) dorsolateral striatum (coordinates, relative to bregma: 0.0 mm posterior, 2.0 mm lateral, 2.5 mm ventral). Corticopontine neurons were labeled by injecting the (right) ventral pons in the region of the pontine nuclei, approached at a 50° angle off the vertical axis through a craniotomy 3.2 mm posterior to lambda, 0.5 mm lateral, and 4.9 mm ventral from the surface of the brain. In all cases, animals were injected ≥16 hr prior to brain slice preparation.
Brain slices were prepared as described (Anderson et al., 2010). Mice (male or female) were sacrificed at P21–32 by decapitation under deep isoflurane anesthesia. Brains were dissected and sectioned in ice-cold, carbogenated choline solution (in mM: 110 choline chloride, 25 NaHCO3, 2.5 KCl, 1.25 NaH2PO4, and 0.5 CaCl2, 7 MgSO4, 25 D-glucose, 11.6 sodium ascorbate, 3.1 sodium pyruvate). Parasagittal slices (300 µm thick) of the right cerebral hemisphere containing the AFC were cut, approximately 1–2 mm lateral to the midline, using a slice angle that was rolled rightward ~15° to align the slice plane optimally with pyramidal neuron apical dendrites and descending axons. This angle yielded 2–3 contiguous slices suitable for recording. We routinely verified that the recorded neurons had intact apical dendrites, based on epifluorescence visualization of the dye-filled neuron after recording. Slices were transferred to and incubated in carbogenated ACSF (in mM: 126 NaCl, 2.5 KCl, 26 NaHCO3, 2 CaCl2, 2 MgCl2, 1.25 NaH2PO4, and 10 D-glucose) for 30 min at 35 °C, and then maintained at 22 °C until recording.
Electrophysiological recordings and LSPS mapping were performed as described (Anderson et al., 2010; Shepherd, 2011), with the experimenter blind to genotype. Slices were transferred to the recording chamber of a custom-built LSPS microscope, and fluorescently labeled neurons were patched with internal solution (in mm: 126 K-gluconate, 4 KCl, 10 HEPES, 4 ATP, 0.3 GTP, 10 mM phosphocreatine, and either 0 or 0.05 Alexa 594, 488, or 350). LSPS mapping was performed at 22 °C in ACSF with 0.2 mM MNI-caged glutamate (Tocris), 4 mM CaCl2, 4 mM MgCl2, and 5 µM CPP (Tocris). Intrinsic properties were measured in current clamp mode immediately before mapping, by presenting families of current steps (−100 to +500 pA in 50 pA increments, duration 0.5 sec). Traces were analyzed offline to identify action potentials (APs). Frequency-current relationships were calculated based on the numbers of APs per current step, and the frequency-current slopes were calculated by linear regression. Spike frequency adaptation was calculated by averaging the ratio of the third and fifth inter-spike interval for all responses with >5 APs. Signals were filtered at 2–4 KHz and sampled at 10 KHz. For LSPS we used Ephus software to control data acquisition (www.ephus.org; (Suter et al., 2010)) and custom routines for offline analysis. In some cases, neurons were filled with biocytin during recording, and subsequently stained with streptavidin-conjugated fluorophores and imaged on a two-photon laser scanning microscope system. Paired recordings were made in plain ACSF at 32 °C.
Morphometric measurements were made from slice images captured during experiments. Cortical thickness (dcortex) was measured as the distance from the cortical surface (pia) to the border of layer 6 and white matter (WM). The depth of the soma in the cortex was measured as the distance from the pia to the soma (dsoma), and the normalized soma position was calculated as dsoma / dcortex. Similarly, we measured the distance from the pia to the layer 5A/B border (d5A/B) and calculated the normalized distance as dL5A/B / dcortex. Identification of this border was guided by a previous quantification of the laminar profile of optical density in video micrographs of agranular cortex (Weiler et al., 2008). Neuronal density was measured in Nissl-stained slices of AFC, by counting the number of neurons in a defined region of interest in layer 2/3 (a 0.1 by 0.2 mm box).
Synaptic input maps were analyzed using custom routines and published approaches (Anderson et al., 2010). These are described further in the Results. We note here that black pixels in input maps, representing dendritic stimulation, are ignored in calculating averages (i.e., averages across map planes, for calculating ‘front views’; or averages across map rows, for calculating ‘side views’). There are fewer black pixels in front view average maps than in individual maps, in which black pixels occur at various locations. On the one hand this is a positive effect: more details (pixels) emerge as more neurons are pooled for averaging. On the other hand, it introduces an under-sampling effect, because for these emergent pixels the number of neurons contributing to their average pixel will be less than the total number of neurons in the group. This effect is strongest near the soma, where one or more pixels usually remain black even after averaging, but also affects pixels around the apical dendrite. The latter is potentially problematic because this is also a region of synaptic input from layer 2/3. However, because black pixels tended to be sparse and variable in location, this effect was generally minor. For layer 2/3 pixels, in the most severe case, the fraction of neurons contributing to the average, relative to total neurons, was 0.3, but on average this fraction was ~0.9. We emphasize that average maps are used as a graphical tool to aid in the visualization of these complex three dimensional data sets.
Unless otherwise indicated, group data including bars in graphs represent the mean ± s.e.m. Differences in the means of two or more groups were tested by Student’s t-test (for normally distributed data) or Wilcoxon rank-sum test (for non-normally distributed data), as indicated. For all comparisons, significance was defined as p < 0.05.
We focused on the AFC for this study, because it is an area of high Met expression in the mouse (Judson et al., 2009), it is the thickest region of the mouse neocortex, and it is implicated in higher order cognitive and motor functions including aspects of motivated behavior. The AFC (Figure 1) is located at the rostral pole of the neocortex just anterior to the rostral forelimb area of motor cortex (Tennant et al., 2010). It is distinct from rodent “prefrontal cortex” (PFC), which is situated medially and is much thinner and cytoarchitectonically simpler (Van De Werd et al., 2010) compared to the lateral agranular neocortex containing the AFC and somatic motor cortex located on the dorsolateral aspect of the frontal lobe (Caviness, 1975; Shepherd, 2009). In choosing a cortical cell type to focus on, we reasoned that layer 5 pyramidal neurons projecting to the contralateral striatum (intratelencephalic-type, “crossed” corticostriatal neurons) were particularly likely to be affected by deletion of Met. These neurons are among the subtypes of cortical neurons that express Met during the peak period of synapse formation (Judson et al., 2009), exhibit modest dendritic and spine changes, and are implicated in the cell-nonautonomous structural changes of striatal medium spiny neurons in Met-cKO (Judson et al, 2010). However, in contrast to relatively well-studied cortical areas such as motor and sensory cortex, there have been no prior investigations of local circuit organization in mouse AFC; similarly, although electrophysiological properties and local circuits of corticostriatal neurons have been described in other cortical areas (Morishima and Kawaguchi, 2006; Hattox and Nelson, 2007; Brown and Hestrin, 2009; Anderson et al., 2010), they have not been characterized in the AFC. Therefore we developed an experimental approach for in vitro analysis of the circuits of corticostriatal neurons in the AFC.
Stereotaxic injection of fluorescent microspheres in the left striatum (Figure 1A and 1B) resulted in retrograde fluorescent labeling of contralaterally projecting (‘crossed’) corticostriatal neurons in the right AFC (Figure 1C). Consistent with normal gray and white matter cytoarchitecture (Judson et al., 2009), cortical thickness at the anterior pole (measured in bright-field images of living slices) was not significantly different between WT (Metfx/*/WT) and cKO (Metfx/fx/Emx1cre) mice (WT: 1.67 ± 0.02 mm, mean ± s.e.m., n = 12 mice; cKO: 1.67 ± 0.02 mm, n = 7 mice; p > 0.05, t-test). Analysis of the laminar distribution of labeled corticostriatal neurons also showed no difference between cKO and WT (Figure 1D and 1E). The density of labeled neurons reached a maximum in upper layer 5B.
Labeled corticostriatal neurons in these parasagittal AFC brain slices were targeted for whole cell patch recordings (Figure 1F). Neurons from WT (n = 32) and cKO (n = 35) mice did not differ significantly in basic electrophysiological parameters, including resting potential (WT: −70.2 ± 0.9 mV; cKO: −71.3 ± 0.9 mV; mean ± s.e.m, t-test, p > 0.05) and input resistance (WT: 201.5 ± 8.9 MΩ; cKO: 192.5 ± 6.8 MΩ, p > 0.05, t-test). Responses to steps of hyper- or depolarizing current (Figure 1G) showed no differences between WT and cKO neurons in frequency-current slopes (WT: 0.055 ± 0.002 Hz/pA; cKO: 0.056 ± 0.002 Hz/pA; p > 0.05, t-test) or spike-frequency adaption ratios (WT: 0.85 ± 0.02; cKO: 0.83 ± 0.01; p > 0.05, t-test) (Figure 1H). These properties were measured just prior to LSPS mapping, thus under conditions optimized for mapping rather than for intrinsic property analysis.
These characterizations validate these methods as an experimental paradigm for studying crossed corticostriatal neurons, a major class of excitatory projection neurons, in the AFC of mice, and demonstrate that the laminar distributions and basic electrophysiological properties of these neurons were similar in cKO and WT animals.
To examine cortical circuit organization in cKO animals, we used glutamate uncaging and LSPS (Callaway and Katz, 1993; Katz and Dalva, 1994; Shepherd et al., 2003; Shepherd, 2011) as a quantitative tool for mapping local sources of excitatory input to individual postsynaptic neurons. Brain slices containing AFC harvested from P21–32 were placed in the recording chamber and bathed with ACSF containing caged glutamate. Fluorescently labeled corticostriatal neurons were targeted for patch recording. Brief (1 msec) flashes of focused ultraviolet laser light caused photo-release of glutamate at discrete locations in the slice, in turn causing a small population of neurons to depolarize sufficiently to fire APs. Excitatory connections from the stimulated presynaptic neurons onto the postsynaptic neuron were recorded as excitatory postsynaptic currents (EPSCs) in voltage clamp mode (Figure 2A and 2B). By stimulating sequentially over an array of hundreds of locations, a spatial array of traces was collected (Figure 2C). These data were converted to synaptic input maps (Figure 2D) by calculating the mean response amplitude in a 50-msec post-stimulus window. Responses that were contaminated by direct somatodendritic stimulation were excluded (black pixels in maps). These synaptic input maps revealed regions of strong excitatory input (such as layer 2/3; Figure 2D, E), representing the locations of pyramidal neurons presynaptic to the recorded neuron.
Pixel values (Qcon) in synaptic input maps reflect a combination of (1) the total number of APs evoked in the small groups of presynaptic neurons at the stimulus location, and (2) the average connection strength (Icon) between the activated presynaptic neurons and the individually recorded postsynaptic neuron (Bureau et al., 2004; Shepherd et al., 2005; Bureau et al., 2008; Weiler et al., 2008). Icon reflects the product of the unitary connection probability (pcon) and unitary connection strength (icon):
The total numbers of neurons and APs activated per flash are influenced by neuronal density (rcell), and the resolution (e.g. excitation volume, Vexc) and intensity (number of APs, NAP) of photostimulation. Combining these terms,
To estimate Vexc and NAP for the conditions used here, we recorded ‘excitation profiles’ of layer 2/3 pyramidal neurons (Figure 2F–K), as these were the presynaptic neurons of interest (i.e., presynaptic to layer 5 neurons). We analyzed the excitation profile data and extracted parameters representing the resolution and intensity of photostimulation. The mean distance of AP-generating sites from the soma, an estimator of the resolution of photostimulation (closely related to Vexc), was not significantly different in WT and cKO slices (WT: 61.9 ± 5.3 µm, mean ± s.e.m., n = 25; cKO: 56.6 ± 5.2 µm, n = 17; p > 0.05, t-test; Figure 2H). This parameter also did not differ when the vertical and horizontal distances were considered separately. The total number of APs per excitation profile, an estimator of the intensity of photostimulation (closely related to NAP), also did not differ by genotype (WT: 7.9 ± 1.2, n = 25; cKO: 7.6 ± 2.1, n = 17; p > 0.05, t-test; Figure 2I). Neuronal density was assessed in Nissl-stained slices of AFC by counting the number of neurons in a defined region of interest in layer 2/3 (Figure 2J). Relative neuronal density, an estimator of rcell, did not differ between WT and cKO animals (WT: 37.1 ± 2.1 cells/100 µm2, n = 12 slices; cKO: 36.9 ± 1.0 cells/100 µm2, n = 12 slices; p > 0.05, t-test; Figure 2K).
These LSPS control experiments indicate that the efficacy of photostimulation was similar in WT and cKO slices. This implies that any differences observed in synaptic input maps (i.e., in pixel values, or Qcon) reflects differences in connectivity (Icon), rather than differences in LSPS excitation parameters (rcell · Vexc · NAP).
In this study we use the term ‘connectivity’ to refer to Icon, which, as the product of connection probability and amplitude, reflects the overall ‘strength’ or ‘impact’ of connections in a synaptic pathway (Lefort et al., 2009). The term ‘hyperconnectivity’ thus refers to an increase in Icon, resulting from an increase in either pcon or icon, or both.
We recorded from corticostriatal neurons in the AFC of Met-cKO and WT mice and mapped excitatory synaptic inputs with LSPS. Map data were analyzed in several ways to understand the circuit organization. First, we averaged all the maps in each group (equivalent to a ‘front view’ of the data set; Figure 3A). The average WT map (Figure 3B) showed strong excitatory input arising from layer 2/3, and weaker input from nearby in layer 5, much like the topography observed in primary motor cortex (Weiler et al., 2008; Anderson et al., 2010). The average cKO map (Figure 3C) showed similar topography, but a trend towards greater amplitude of layer 2/3 input; however, this difference did not reach statistical significance (WT: −16.6 ± 2.8 pA, n = 32 neurons; cKO: −20.8 ± 2.3 pA, n = 35 neurons; p > 0.05, t-test).
A potential complication in studying a particular class of cortical pyramidal neurons defined by their long-range projections alone is that local circuits can differ greatly depending on the precise sub-layer locations of the neurons’ somata (Otsuka and Kawaguchi, 2008; Anderson et al., 2010). We therefore further analyzed the maps of these corticostriatal neurons by sorting them according to soma location relative to the layer 5A/B border (superficial to deep). Projecting the sorted map data onto a single plane by averaging along the map rows gave a ‘side view’ of the maps (Figure 3A, 3D, 3E) providing a quantitative description of the local circuit in terms of the sub-layer locations of the presynaptic pyramidal neurons and postsynaptic corticostriatal neurons. In the side view maps, the apparent jitter in soma positions arises from variability in the distances of the layer 5A/B border from the pia.
Plotting the average strength of input from layer 2/3 to corticostriatal neurons as a function of the distance of the postsynaptic neuron’s soma from the layer 5A/B border (Figure 3F and 3G) showed that, on average, the amplitude of layer 2/3 excitatory input to corticostriatal neurons in upper layer 5B was greater in the cKO compared to WT by a factor of 2.3 (WT: −12.3 ± 3.3 pA, n = 11; cKO: −28.6 ± 5.7 pA, n = 12; p < 0.05, t-test). To visualize the circuit differences, we generated average maps for just the neurons located in upper layer 5B neurons (where the difference was maximal; n = 11 WT and 12 cKO neurons; Figure 3H), and computed a difference image by subtracting the WT map from the cKO map (Figure 3I). This revealed a locus of maximal difference in layer 2/3.
In the preceding experiments, we used Met-cKO mice engineered to eliminate Met signaling in tissue derived from the dorsal pallium. However, both rare loss-of-function copy number variation (Marshall et al., 2008; Pinto et al., 2010) and the common SNP allelic association of MET with ASD result in a reduction rather than complete absence of gene and protein expression (Campbell et al., 2006; Campbell et al., 2007). Therefore, to assess whether partial disruption of Met signaling in the forebrain affects cortical circuits, we also studied Met-heterozygous (HET) mice (Metfx/+/Emx1cre), in which Met expression in the forebrain is reduced by ~50% (determined by forebrain Western blot analysis; M. C. Judson and P. Levitt, unpublished observations).
In HET mice, corticostriatal neurons occupied the same laminar distribution as in WT and cKO (Figure 4A and 4B). Corticostriatal neurons in both layer 5A and 5B received inputs from layer 2/3, as seen in a side view projection of their input maps (Figure 4C). Plotting layer 2/3 input as a function of sub-layer location showed that, like the cKO (see above), the inputs in HET mice differed from WT in layer 5B (Figure 4D and 4E). Layer 2/3 inputs to both the HET and cKO layer 5B corticostriatal neurons were more than doubled compared to the WT (WT: −10.4 ± 2.1 pA; HET: −21.5 ± 3.9 pA; cKO: −23.0 ± 4.3 pA; one-way ANOVA, p < 0.05). These results indicate that even partial reduction of Met signaling is sufficient to alter local circuits in the frontal cortex of mice.
The LSPS data indicate that excitatory intracortical synaptic pathways are abnormally strengthened in cKO mice, but this provides indirect evidence about the strengths of unitary connections between individual pre- and postsynaptic neurons. Therefore, we performed paired recordings to measure the properties of unitary connections from layer 2/3 neurons onto radially subjacent layer 5B corticostriatal neurons (Figure 5A). APs were induced in the layer 2/3 neuron while recording from the corticostriatal neuron in voltage clamp mode (Figure 5B). In WT slices, excitatory connections were found for 8 of 76 paired recordings (i.e., pcon = 0.11), comparable to pcon for layer 2/3→5 connections in mouse barrel cortex (~0.1; (Lefort et al., 2009)). In cKO slices, excitatory connections were found for 7 of 41 paired recordings (pcon = 0.17). This difference between WT and cKO (a factor of 1.6; Figure 5C) did not reach statistical significance, presumably due to the low numbers of connected pairs (p > 0.05, d. f. = 1, two-sample χ2 test with Yates’ correction). On average, the strength of unitary EPSCs between connected pairs was significantly greater in cKO neurons compared to WT by a factor of 1.4 (WT: −27.1 ± 0.9 pA, n = 8 pairs; cKO: −37.1 ± 2.8, n = 7 pairs; p < 0.05, t-test; Figure 5D).
To relate these unitary connection measurements to LSPS mapping data, we considered that a pixel value is proportional to the average connection strength (i.e., (Qcon ~ Icon) (Equation 2), and that Icon is the product of the unitary connection probability (pcon) and amplitude (icon) (Equation 1) (see also (Lefort et al., 2009)). This is equivalent to pooling connected and unconnected pairs. In cKO, Icon was greater by a factor of 2.4 compared to WT (cKO: 6.3 pA; WT: 3.0 pA), a difference that was significant when calculated based on the product pcon × icon (p < 0.05, t-test), although not when based the pooled data (p > 0.05, Wilcoxon test). The greater difference in Icon than in icon reflects the higher pcon observed for cKO pairs. The magnitude of the difference (factor of 2.1) is consistent with the magnitude of the differences in Qcon measured by LSPS (factor of 2.3).
In principle, differences in short-term plasticity could also contribute to differences in LSPS maps, and moreover would implicate presynaptic mechanisms. We therefore tested the paired pulse ratios of the uEPSCs in the connected pairs, during which three current pulses (1 msec duration, 1 nA, 50 msec interval) were injected into the presynaptic layer 2/3 neurons to elicit single APs. The paired pulse ratios of uEPSCs in the postsynaptic L5B corticostriatal neurons were calculated as the amplitude ratios over the first uEPSC. We found that paired-pulse ratios for WT (n = 5) and cKO (n = 5) did not differ (two-way ANOVA, p > 0.05 for the effects of genotype). These results suggest that at least some functional presynaptic mechanisms for synaptic strengthening (e.g. involving changes in release probability) do not contribute directly to the circuit phenotype. The data, however, do not exclude the possibility of presynaptic involvement in the observed strengthening of the layer 2/3 inputs onto layer 5B corticostriatal neurons (for example, increased numbers of release sites).
In preceding experiments we focused on the population of contralaterally projecting corticostriatal neurons, a pyramidal neuronal class that normally expresses Met. However, Met is expressed in both layer 2/3 pyramidal neurons and layer 5B corticostriatal neurons, and cell-nonautonomous effects of Met deletion on postsynaptic neuronal structure can occur (Judson et al., 2010). Therefore the impact of Met deficiency on layer 2/3 input strength could be a result of mechanisms that are presynaptic, postsynaptic, or both. In particular, layer 2/3 input could be stronger to all layer 5B neurons, independent of their long-range projection targets. To test these ideas, we selected for additional physiological analyses a class of neurons that does not normally express Met, and therefore might be less affected by potential cell-autonomous effects of Met deficiency in the cKO mice. Based on immunocytochemical mapping, Met expression is low or absent in the axons of corticopontine neurons that project in the cerebral peduncle (Judson et al., 2009). Corticopontine neurons thus provide a comparison group by which to separate the effects of sublayer location from postsynaptic Met expression. Because the paired-pulse ratio data implicated postsynaptic rather than presynaptic effects of Met deficiency on circuits of (normally Met-expressing) layer 5B corticostriatal neurons (Figure 5F), we hypothesized that Met deficiency would not affect the circuits of (normally Met-non-expressing) layer 5B corticopontine neurons.
To label AFC neurons projecting to the pons, we injected fluorescent microspheres into the pontine nuclei and cerebral peduncle and prepared brain slices containing the ipsilateral AFC (Figure 6A). Because both Met-heterozygous (Metfx/+/Emx1cre) and Met-cKO (Metfx/fx/Emx1cre) corticostriatal neurons showed the same degree of local circuit strengthening (Figure 4), we grouped the two genotypes (hereafter, Metfx/*/Emx1cre) for the subsequent comparisons to controls (Metfx/*/WT). Fluorescently labeled corticopontine projection neurons were located exclusively in layer 5B (Figure 6B). LSPS front view map averages showed similar topography of input from layer 2/3 for both groups (Figure 6C and 6D). Quantification of this input as a function of soma position within layer 5 (Figure 6E) revealed that the layer 2/3 input strength was not increased for the Metfx/*/Emx1cre corticopontine neurons (Metfx/*/WT: −28.5 ± 3.3 pA, n = 34; Metfx/*/Emx1cre: −33.2 ± 5.4 pA, n = 19; p > 0.05, t-test; Figure 6F). These data indicate that, at least for cortical circuits, the synaptic abnormalities of local circuit excitatory connectivity in Met-deficient mice are projection class-specific.
While the number of candidate genes continues to increase (Abrahams and Geschwind, 2010), the functional link between genetic risk for ASD and cortical circuitry remains a significant challenge to define. The main finding of this study demonstrates a significant increase in intracortical connectivity in local excitatory circuits of AFC in mice with disrupted cortical expression of Met (Figure 7). Although MET typically is not included in models of key synaptic proteins that may contribute to ASD pathogenesis, the current electrophysiological data and our recent results of altered dendritic and spine development in Met-cKO mice (Judson et al., 2010) indicate that the contribution of MET to ASD risk is consistent with the model of disrupted synaptic development in specific circuits that underlie social behavior and communication (Zoghbi, 2003; Persico and Bourgeron, 2006; Sudhof, 2008).
We focused on characterizing the excitatory inputs from layer 2/3 pyramidal neurons to layer 5 corticostriatal neurons, as our mapping studies in WT animals revealed this pathway as the primary source of interlaminar input to these neurons. We found this pathway to be ‘hyperconnected’ both in Met-cKO and Met-HET mice. This was manifest as increased amplitude of synaptic inputs (Qcon) in LSPS maps (Figure 3), and increased synaptic connectivity (Icon) in paired recordings (Figure 5). Perhaps most surprising was the remarkably specific impact of Met deficiency on cortical microcircuits. A specific subpopulation of corticostriatal neurons exhibited hyperconnectivity (neurons just below the layer 5A/B border), and moreover, layer 5 pyramidal neurons that do not express Met (corticopontine neurons) failed to show changes in local circuits in our experiments (Figure 7). The data suggest that the spatial regulation of Met expression will be a key element in deciphering which circuits in the human brain are most vulnerable to MET genetic risk. Our recent study in the developing monkey, in which we demonstrate high levels of MET expression in restricted areas of frontal, temporal, and occipital cortex, amygdala, and hippocampus during synapse formation, suggest that in humans, vulnerability is likely to impact specific circuits that are involved in emotional face and complex image processing (Judson et al., 2011). In the affected circuits in the mouse, the hyperconnectivity was manifest as increased amplitude of average connectivity in LSPS maps (Figure 3), and as increased unitary connectivity in pair recordings (Figure 5). Although the behavioral significance of these microcircuit-level changes needs to be tested, the consistency of our data with predictions of local neocortical hyperconnectivity from human neuroimaging studies in clinical populations is striking (Frith, 2004; Just et al., 2004; Courchesne and Pierce, 2005; Geschwind and Levitt, 2007; Just et al., 2007; Scott-Van Zeeland et al., 2010).
Mouse models of ASD-related neurodevelopmental disorders are increasingly available, particularly for monogenic and syndromic disorders (reviewed in (Ehninger et al., 2008; Moy and Nadler, 2008)). Changes in cortical circuit organization have been described in some of these, and are interesting to compare to the present results. In FMR1-mutant mice, a model of Fragile X syndrome, excitatory drive in somatosensory microcircuits is reduced (Gibson et al., 2008; Pfeiffer and Huber, 2009), including a developmentally transient decrease in layer 4→2/3 connectivity that normalizes by early adulthood (Bureau et al., 2008). In MeCP2-mutant mice, a model of Rett syndrome, excitatory connections between layer 5 pyramidal neurons in somatosensory cortex show a developmentally enduring decrease in connectivity (Dani et al., 2005; Dani and Nelson, 2009), and ascending excitatory pathways to layer 2/3 pyramidal neurons in motor cortex also show reduced excitatory synaptic connectivity (Wood et al., 2009; Wood and Shepherd, 2010), consistent with reduced connectivity observed in monocultures (Chao et al., 2007). In contrast, in the present study using a mouse model with a disrupted risk gene of idiopathic ASD, we found evidence for hyperconnectivity in local excitatory circuits in the AFC. The differences may reflect a substantial level of complexity with which ASD risk genes contribute to abnormal cortical circuit development, which in turn may be relevant to the well-known heterogeneity of even the core behavioral triad of clinical phenotypes that comprise the ASD diagnosis in different populations.
Prenatal exposure to VPA is a relevant, non-genetic model related to ASD that can be compared to our findings. Exposure to VPA during a short critical period early in embryonic development significantly increases the risk of a child being diagnosed with ASD (Rasalam et al., 2005). In rats exposed in utero to VPA, layer 5 pyramidal neurons in somatosensory and frontal cortices exhibit recurrent excitatory hyperconnectivity in the form increased connection probability (Rinaldi et al., 2008a; Rinaldi et al., 2008b), similar to our findings of increased connectivity. However, in the VPA model the mean amplitude of unitary connections was reduced (here, increased); moreover, intrinsic excitability was strongly attenuated (here, unchanged). The underlying mechanism through which VPA induces local circuit changes is unknown, but the drug can alter gene expression in complex ways (Milutinovic et al., 2007), and furthermore may affect cortical excitation/inhibition balance by depleting specific populations of neocortical interneurons (Gogolla et al., 2009).
The consequences of local-circuit hyperconnectivity on cortical function remain unknown. It has been speculated that an excessive ratio of excitation to inhibition generates ‘noise’ in cortical signaling and sensory representations (Rubenstein and Merzenich, 2003; Levitt et al., 2004). Computational modeling has also been used to explore how hyperconnectivity in cortical circuits leads to greater transformation of the information being processed through the local circuit (Neymotin et al., 2010). Our results provide experimental data that could be used to constrain and extend such theoretical approaches. Moreover, some individuals with ASD display a ‘hyper-perception’ within a sensory domain (Iarocci and McDonald, 2006; Baron-Cohen et al., 2009), which might arise from local-circuit hyperconnectivity. These circuit-level and related changes could also lead to an “intense world syndrome” (Markram et al., 2007; Markram and Markram, 2010).
Our findings may also relate to the types motor and action planning abnormalities characteristic of ASD (Teitelbaum et al., 1998; Haswell et al., 2009; Mostofsky et al., 2009). The cell-class specificity – affecting the corticostriatal projection both at the level of local cortical circuits (present study) and at the level of medium spiny neurons in the striatum (Judson et al., 2010) – is consistent with the clinical picture of stereotypies and repetitive behaviors, which point to dysfunction in the extrapyramidal motor system including large-scale loops involving cortex, basal ganglia, and thalamus (Hallett et al., 1993; Haznedar et al., 2006). Thus, together with our recently reported influence of Met signaling on dendritic tree morphology and spine size (Judson et al., 2010), the present data provide a unique translation of the genetic risk of MET for ASD into a neurobiological mechanism with relevance to the core clinical phenotype (Campbell et al., 2010).
Our study raises many questions about other aspects of cortical organization that may be altered by Met deficiency. For example, the experimental paradigm employed here could be applied to other cell classes, cortical areas, and developmental time points. The possibility of abnormal inhibitory mechanisms also remains to be explored at the microcircuit level. Evaluation of even more restricted models of Met deficiency, such as single-cell knockdown approaches (Wood et al., 2009) would help to understand the extent to which the changes observed here are presynaptic versus postsynaptic, cell-autonomous versus cell-nonautonomous, and ‘primary’ versus ‘compensatory’ (‘adaptive’).
In summary, our analysis establishes a pattern of excitatory hyperconnectivity in specific AFC microcircuits caused by Met deficiency. We propose that these changes constitute a circuit-level physiological basis for Met-mediated ASD risk. The findings reported here provide a starting point, including a quantitative topographic map, to guide further exploration of how Met signaling influences cortical circuit development.
We thank Patrick Sheets for help and advice with LSPS experiments, Lisa Topolnik for advice on labeling and imaging, and Luke Trapp for two-photon imaging. We are grateful to Mac Hooks, Whit Tao, Barb Thompson, and Li Zhang for comments on the manuscript. This work was funded by a SFARI grant from the Simons Foundation (P.L., G.S.), and grants from the National Institutes of Health (C.A.: T32NS041234; S.Q.: K99MH087628; P.L.: R01MH067842; G.S.: R01NS061963).
AUTHOR CONTRIBUTIONSSQ and CA both performed the corticostriatal mapping experiments. SQ performed the paired recording and cell-counting experiments. CA performed the corticopontine mapping experiments. SQ, CA, PL, and GS conceived the design of the study and wrote the paper.