Npas1⁺-Nkx2.1⁺ Neurons Are an Integral Part of the Cortico-pallido-cortical Loop

PV⁺ and Npas1⁺ neurons are distinct GPe neuron classes

Our laboratory and others have shown previously that the GPe contains two principal, near-exclusive classes of neurons distinguished by their respective expression of PV and Npas1 (Hegeman et al., 2016). However, the reported abundance of PV⁺ neurons across laboratories ranges from 30 to 60%, and their overlap with Npas1⁺ neurons within the GPe varies widely in the literature (Flandin et al., 2010; Nóbrega-Pereira et al., 2010; Mastro et al., 2014; Abdi et al., 2015; Dodson et al., 2015; Saunders et al., 2015; Oh et al., 2017; Abrahao and Lovinger, 2018). As multiple mouse lines and well characterized antibodies are available to study PV⁺ neurons, we sought to reevaluate the abundance of PV⁺ GPe neurons and reconcile the inconsistencies. First, we examined PV immunoreactivity (PV-ir) across all GPe neurons using HuCD as a neuronal marker; we observed PV-ir in ∼50% of GPe neurons (49 ± 4%, n = 2726 neurons, 19 sections) (Fig. 1c). Note that the range observed within the PV population data (Fig. 1c) mirrors the variance reported in the literature, suggesting that the discrepancy in the literature is likely of biological origin rather than methodological. Four PV antibodies were used throughout the study (Table 2). All PV antibodies yielded similar results and were therefore pooled. Although both PV-tdTom and PV-L-tdTom (PV-Cre;LSL-tdTomato) have been used consistently across laboratories for the identification of PV⁺ neurons, the utility of the PV-Flp line for studies of GPe neurons had not been established. To harness the potential of this mouse line, we crossed it with a Flp-reporter (FSF-tdTomato) line (see Materials and Methods). The resultant PV-F-tdTom (PV-Flp;FSF-tdTomato) line produced robust, cytoplasmic neuron labeling similar to that of the PV-L-tdTom and PV-tdTom lines. As expected, the PV-F-tdTom line showed prominent tdTomato expression (tdTomato⁺) not only in GPe PV⁺ neurons but also in PV⁺ neurons in the thalamic reticular nucleus (TRN) (Fig. 1d,e) and cerebellum, as well as in PV⁺ interneurons throughout the cortex, hippocampus, and striatum.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

GPe neuron diversity. a, The location of the GPe in a mouse brain is illustrated (side view). b, GPe neurons at three different lateral, intermediate, and medial levels (∼2.5, 2.1, and 1.7 lateral from bregma) were sampled. c, Using HuCD as a neuronal marker, population data for the relative abundance of GPe neuron markers was determined. Each circle represents a section. Inset: low-magnification confocal image of a sagittal brain section showing HuCD-labeling in the GPe with the dStr and ic defining the rostral and caudal borders, respectively. Note the low density of HuCD-labeled cells outside of the GPe. d, Low-magnification confocal images of the GPe and TRN in PV-L-tdTom (PV-Cre;LSL-tdTomato, top) and PV-tdTom (bottom) mice. e, PV-F-tdTom (PV-Flp;FSF-tdTomato) and Nkx2.1-F-tdTom (Nkx2.1-Flp;FSF-tdTomato) were used in this study. The PV-F-tdTom (PV-Flp;FSF-tdTomato) transgenic mouse line produces faithful reporting of PV⁺ neurons and similar cytoplasmic neuron labeling as the PV-L-tdTom and PV-tdTom lines (as shown in d). Top, low-magnification showing the PV-F-tdTom line produces prominent tdTomato expression (tdTomato⁺) in PV⁺ neurons in the TRN in addition to the GPe. To confirm the validity of the mouse line, tdTomato expression was compared against PV immunostaining. A higher magnification of the GPe shows nearly all tdTomato⁺ (magenta) neurons colocalize with PV-ir (green). Bottom, Nkx2.1-F-tdTom reliably captures neurons that arise from the Nkx2.1 lineage. Note that no cell bodies were found in the TRN (see also Fig. 5c). Double immunolabeling with tdTomato and Nkx2.1 demonstrated ∼90% colocalization. Arrowheads indicate neurons that do not colocalize. f, Triple immunostaining with PV, Npas1, and GFP on Lhx6-GFP brain sections confirmed the existence of a Lhx6⁺-PV⁻-Npas1⁻ GPe population. Circles indicate Lhx6⁺ neurons that colocalize with either PV or Npas1. Arrowheads point to unique Lhx6⁺ neurons. Note that there are both bright and dim populations of Lhx6⁺ neurons. g, Sox6⁺ neurons express established GPe markers. Note that there are both bright and dim populations of Sox6⁺ neurons. Bottom, arrowheads indicate Lhx6⁺-Sox6⁻ neurons. dStr = dorsal striatum; TRN, thalamic reticular nucleus; VPL/VPM, ventral posterior nucleus; ic, internal capsule.

Across the PV-tdTom, PV-L-tdTom, and PV-F-tdTom lines, the abundance of tdTomato⁺ neurons converged to ∼50% of the total GPe neuron population (PV-tdTom, 50 ± 9%, n = 571 neurons, 5 sections; PV-L-tdTom, 50 ± 13%, n = 848 neurons, 9 sections; PV-F-tdTom, 53 ± 15%, n = 859 neurons, 6 sections). Comparison of tdTomato expression with PV-ir demonstrated faithful reporting of PV⁺ neurons in all three transgenic lines (PV-tdTom, 100 ± 0%, n = 742 of 747 neurons, 6 sections; PV-L-tdTom, 100 ± 0%, n = 1380 of 1392 neurons, 12 sections; PV-F-tdTom, 100 ± 0%, n = 930 of 1023 neurons, 6 sections).

The Npas1-Cre-tdTom line was previously generated in our laboratory and labels ∼90% (88 ± 7%, n = 505 neurons) of neurons that endogenously express Npas1 in the GPe (Hernández et al., 2015). As this mouse line does not fully capture all Npas1⁺ neurons, we focused our quantifications on Npas1-immunoreactivity (Npas1-ir) with our previously characterized Npas1 antibodies (Hernández et al., 2015). We found that Npas1⁺ neurons make up ∼30% (32 ± 4%, n = 3361 neurons, 21 sections) of all GPe neurons. This result is consistent with our previous quantification of 27% of the entirety of the GPe. A reevaluation of the overlap between PV and Npas1 confirmed our prior observation (Hernández et al., 2015) that these two markers are expressed in almost completely segregated neuron populations. We observed a very low level of overlap across lateromedial GPe levels (2 ± 2%, n = 96 of 1777 neurons, 12 sections), with the most overlap observed at the intermediate level (lateral: 1 ± 1%, n = 17 neurons, 3 sections; intermediate: 3 ± 1%, n = 68 neurons, 5 sections; medial: 0 ± 0%, n = 11 neurons, 4 sections). This is slightly higher than our previous estimate and may be related to our previous reliance on the Npas1 mouse line for quantification (Hernández et al., 2015). Our data are at odds with those of Abrahao and Lovinger (2018), which showed a much higher (up to 12.6%) overlap between PV and Npas1 expression in the GPe. This discrepancy can be attributed to differences in immunolabeling protocols, quantification methods, and the age of mice used in the experiments.

As determined by immunolabeling, a small population of neurons expressed a cholinergic neuron marker, choline acetyltransferase (ChAT) (6 ± 2%, n = 267 neurons, 9 sections), a finding that is consistent with our previous reports (Gritti et al., 2006; Nóbrega-Pereira et al., 2010; Hernández et al., 2015) (Table 3). This leaves ∼15% of the GPe population that is unidentified by PV, Npas1, or ChAT.

Nkx2.1-F-tdTom mice label prototypic GPe neurons

Based on our previous data showing that Lhx6⁺ neurons overlap substantially with both PV⁺ neurons and Npas1⁺ neurons (Hernández et al., 2015), we hypothesized that a subset of Lhx6⁺ neurons may correspond to the 15% of GPe neurons that are PV⁻ and Npas1⁻ (and ChAT⁻) (Fig. 1f). However, the use of the Lhx6-GFP mouse line has resulted in highly inconsistent results across laboratories (Hegeman et al., 2016).

To glean insights into an improved GPe neuron classification, we systematically examined the GPe expression of Nkx2.1 and Sox6, signaling molecules that are upstream and downstream of Lhx6, respectively. This was motivated by the observation that Nkx2.1, Lhx6, and Sox6 work in concert to control the fate of forebrain GABAergic neurons (Du et al., 2008; Batista-Brito et al., 2009; Jaglin et al., 2012). Consistent with the fact that Nkx2.1 plays a crucial role in GPe development (Rubenstein et al., 1998; Sussel et al., 1999; Flandin et al., 2010; Nóbrega-Pereira et al., 2010), Nkx2.1 immunolabeling revealed that ∼60% (60 ± 2%, n = 5342 neurons, 16 sections) of GPe neurons are Nkx2.1-expressing (Nkx2.1⁺) (Fig. 1c,e, Table 3). In keeping with our previous analysis of Lhx6 and Foxp2 (Hernández et al., 2015), we observed no overlap between Nkx2.1⁺ neurons and Foxp2-expressing (Foxp2⁺) neurons in wild-type brain sections (n = 1 of 1131 Nkx2.1⁺ neurons, 3 sections; n = 1 of 414 Foxp2⁺ neurons, 3 sections). Subsequent immunohistological analysis of Nkx2.1 antibody labeling with established GPe markers revealed that the majority of Nkx2.1⁺ neurons are PV⁺ (71 ± 3%, n = 1105 neurons, 7 sections), while a smaller subset are Npas1⁺ (17 ± 2%, n = 383 neurons, 6 sections) or Lhx6⁺ (41 ± 7%, n = 1469 neurons, 9 sections) (Table 3). Importantly, Nkx2.1⁺ neurons are only ∼80% (83 ± 4%, n = 1105 neurons, 7 sections) of PV⁺ neurons, consistent with previous observations (Xu et al., 2008). Furthermore, Nkx2.1⁺ neurons represent ∼80% (78 ± 8%, n = 1469 neurons, 9 sections) of Lhx6⁺ neurons and only a subset of Npas1⁺ neurons (32 ± 1%, n = 383 neurons, 6 sections). More importantly, with triple immunostaining, we found that nearly all Npas1⁺-Nkx2.1⁺ neurons are Lhx6⁺ (90 ± 5%, n = 184 of 204 neurons, 3 sections). Accordingly, the Npas1⁺-Nkx2.1⁺ and Npas1⁺-Lhx6⁺ populations of Npas1⁺ neurons are near-identical (33 ± 5%, n = 204 of 609 neurons, 3 sections and 32 ± 3%, n = 233 of 609 neurons, 3 sections, respectively) (data not shown). This finding is in line with what we have described previously (Hernández et al., 2015).

To effectively identify all neurons derived from the Nkx2.1 lineage, we made a Nkx2.1-F-tdTom (Nkx2.1-Flp;FSF-tdTomato) genetic cross, which yielded robust tdTomato labeling (tdTomato⁺) in the GPe. In addition to tdTomato⁺ neurons, we observed tdTomato⁺ glia in the GPe that were distinguished from neurons based on their morphology. To assess the cumulative recombination events in Nkx2.1-F-tdTom, we compared tdTomato expression to Nkx2.1 immunolabeling. Immunoreactivity on Nkx2.1-F-tdTom brain sections with tdTomato, Nkx2.1, and HuCD antibodies revealed ∼90% (89 ± 11%, n = 795 neurons, 3 sections) of tdTomato⁺ neurons colocalized with Nkx2.1⁺ neurons (Fig. 1e). Although we observed significant tdTomato⁺ labeling throughout the entire cortex, Nkx2.1 immunoreactivity was absent. These data corroborate previous findings that Nkx2.1 is downregulated in cortical interneurons in adult mice (Butt et al., 2008; Nóbrega-Pereira et al., 2008) and that the majority of GPe neurons that are derived from the Nkx2.1 lineage maintain Nkx2.1 expression into adulthood. Similarly, we observed no tdTomato⁺ or Nkx2.1⁺ cell bodies in regions caudal to the forebrain, such as the TRN (Fig. 1e), STN, and SNr (data not shown).

Sox6 delineates GPe neuron subtypes

Previous literature demonstrates that the transcription factor Sox6 is present in most, if not all, medial ganglionic eminence (MGE)-derived neurons in the mature brain (Azim et al., 2009; Batista-Brito et al., 2009). Moreover, Sox6 and Lhx6 colocalize extensively within the MGE (Batista-Brito et al., 2009), consistent with the function of Sox6 as a downstream signaling partner of Lhx6 (Azim et al., 2009; Batista-Brito et al., 2009). In view of these findings, we set out to examine the Sox6 expression pattern in the GPe (Fig. 1g). Sox6 immunolabeling revealed ∼65% (64 ± 4%, n = 4681 neurons, 14 sections) of GPe neurons are Sox6-expressing (Sox6⁺). We first investigated Sox6 expression in Nkx2.1⁺ neurons; due to incompatibility of the Nkx2.1 and Sox6 antibodies, we relied on Sox6 immunolabeling in Nkx2.1-F-tdTom brain sections. We found substantial overlap between tdTomato⁺ neurons and Sox6⁺ neurons (68 ± 1% of Nkx2.1, n = 681 neurons, 3 sections; 58 ± 0% of Sox6, n = 681 neurons, 3 sections) (Table 3). Although both Sox6 and Nkx2.1 are expressed in ∼60–65% of GPe neurons, they do not represent the same pool of neurons, as their expression overlaps differently with other GPe neuron markers (Table 3). Nearly all Npas1⁺ neurons are Sox6⁺ (93 ± 6%, n = 1999 neurons, 14 sections), whereas only half of the PV⁺ population expresses Sox6 (53 ± 8%, n = 1675 neurons, 15 sections). As mentioned above, Nkx2.1⁺ neurons account for ∼32% of Npas1⁺ neurons and ∼83% of PV⁺ neurons.

Next, we examined the relationship between the Sox6 and Lhx6 GPe neuron populations, as this may yield important insights into targeting the otherwise difficult-to-identify Lhx6⁺-PV⁻-Npas1⁻ population. As we were unsuccessful in labeling endogenous Lhx6 protein with multiple Lhx6 antibodies (Santa Cruz Biotechnology SC-98607, Santa Cruz Biotechnology SC-271433, NeuroMab 73–241, and a custom antibody from Dr. John Rubenstein), we used the GFP expression in Lhx6-GFP mice as a proxy for Lhx6 expression. Overall, Lhx6⁺ neurons represent a third of the GPe (34 ± 8%, n = 2533 neurons, 12 sections); ∼30% of these neurons express PV (28 ± 7%, n = 654 neurons, 12 sections), 35% express Npas1 (35 ± 5%, n = 818 neurons, 12 sections), and 7% express ChAT (7 ± 1%, n = 48 neurons, 3 sections). These percentages confirm our previous reports (Hernández et al., 2015). To determine the relationship between Sox6 expression within the Lhx6⁺, PV⁺, and Npas1⁺ populations, immunolabeling on Lhx6-GFP sections showed that Sox6 was expressed in ∼75% (76 ± 7%, n = 2346 neurons, 15 sections) of Lhx6⁺ neurons (Fig. 1g) as well as in all PV⁺ neurons and Npas1⁺ neurons that were Lhx6⁺ (Fig. 2a). Note that similar to Lhx6, there were bright and dim populations of Sox6. We did not observe a relationship between Sox6 fluorescence levels and expression of PV and Npas1. Consistent with our previous observations (Hernández et al., 2015), we observed negligible overlap between Lhx6⁺ and Foxp2⁺ GPe neurons (8 of 1882 neurons, 3 sections).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Lhx6⁺ and Dbx1⁺ GPe neurons colocalize with established GPe markers. a, Left, low and high-magnification images of PV⁺, Sox6⁺, and Lhx6⁺ (GFP⁺) GPe neurons. Right, low and high-magnification images of Npas1⁺, Sox6⁺, and Lhx6⁺ (GFP⁺) GPe neurons. b, In situ hybridization signals from Dbx1-L-tdTom mouse line for tdTomato⁺ and Cre⁺ in the adult GPe and neighboring areas (left). Note the widespread tdTomato⁺ across brain areas (top right) resulted from the cumulative recombination from early developmental stages despite the absence of Cre⁺ expression in adult (bottom right). Data are adopted from Allen Brain Atlas. The Dbx1-L-tdTom mouse line labeled Dbx1⁺ (tdTomato⁺) neurons (HuCD, magenta) and glia in the GPe. c, Dbx1⁺ GPe neurons colocalized with established GPe markers and are largely PV⁺. Note that there was no overlap between Dbx1 and Foxp2. Circles indicate colocalization, star (bottom right) presents an example of astrocytic labeling in the Dbx1-L-tdTom mouse line. dStr, dorsal striatum; GPi, internal globus pallidus; SI, substantia innominata; TRN, thalamic reticular nucleus; ac, anterior commissure; ic, internal capsule.

Together, our data describe a Lhx6⁺ population (∼37%) that is devoid of both PV and Npas1 expression and accounts for ∼13% of the entire GPe. A portion of this Lhx6⁺-PV⁻-Npas1⁻ population does not express Sox6 and represents approximately a quarter of Lhx6⁺ neurons (24 ± 6%, n = 938 neurons, 15 sections) or 8% of all GPe neurons, whereas the Lhx6⁺-PV⁻-Npas1⁻ population that does express Sox6 accounts for only ∼4% of the entire GPe. Given that 7% of Lhx6⁺ neurons express ChAT, and we observed no overlap between ChAT and Sox6 (0 of 1674 neurons, 3 sections), we were able to pinpoint the molecular identity of the entirety of Lhx6⁺ neurons (see footnote in Table 3). Importantly, the Lhx6⁺-Nkx2.1⁺ population is likely equivalent to the Lhx6⁺-Sox6⁺ population, as both Nkx2.1 and Sox6 overlap similarly with Lhx6⁺ (Nkx2.1 of Lhx6, 78 ± 8%, n = 1469 neurons, 9 sections; Sox6 of Lhx6, 76 ± 7%, n = 2346 neurons, 15 sections), and the Lhx6⁺-Nkx2.1⁺ and Lhx6⁺-Sox6⁺ populations are similarly abundant within the entire GPe (28 ± 7% and 32 ± 6%, respectively). However, with our present tools, we cannot confirm the assertion that these two subtypes are identical.

Dbx1-L-tdTom mice label a heterogenous GPe neuron subset

Since Lhx6⁺-Sox6⁻ neurons constitute a substantial fraction (8%) of the GPe, we sought to identify an additional expression marker for this population. Neurons from the Dbx1 lineage originate from the preoptic area (PoA) and are known to populate the GPe (Nóbrega-Pereira et al., 2010). Considering Lhx6 is expressed in postmitotic neurons derived from both the MGE and the PoA (Fogarty et al., 2007; Du et al., 2008), we set out to investigate if neurons from the Dbx1 lineage correspond to Lhx6⁺-Sox6⁻ GPe neurons. Accordingly, we identified neurons that arise from the Dbx1 lineage and determined their relationship with the Lhx6⁺-Sox6⁻ GPe subset by using a Dbx1-L-tdTom (Dbx1-Cre;LSL-tdTomato) cross, which produced robust tdTomato expression in the GPe. In addition to neuronal labeling, tdTomato⁺ glia were present within the GPe and densely populated the ventral GPe border. For simplicity, we refer to tdTomato⁺ neurons in Dbx1-L-tdTom mice as Dbx1⁺ neurons. Using HuCD as a neuronal marker (Fig. 2b), we determined that Dbx1⁺ neurons account for ∼9% of the entire GPe population (9 ± 1%, n = 3002 neurons, 61 sections). Immunohistological analysis of Dbx1⁺ neurons showed substantial coexpression with Nkx2.1 (78 ± 10%, n = 329 neurons, 9 sections) and Sox6 (52 ± 8%, n = 442 neurons, 15 sections). Furthermore, Dbx1⁺ neurons are primarily PV⁺ (72 ± 8%, n = 493 neurons, 23 sections) and to a lesser extent Npas1⁺ (10 ± 5%, n = 72 neurons, 14 sections), ChAT⁺ (7 ± 3%, n = 34 neurons, 9 sections), and Foxp2⁺ (4 ± 2%, n = 13 neurons, 10 sections) (Fig. 2c). To summarize, we found Dbx1⁺ neurons do not correspond to the Lhx6⁺-Sox6⁻ unique GPe subset. Although both populations were of similar abundance in the GPe, they varied in their coexpression of PV, Npas1, and Sox6. Expression of Lhx6 in Dbx1⁺ neurons was not examined in this study for technical reasons; however, due to the close functional relationship between Nkx2.1, Sox6, and Lhx6, it can be inferred that Lhx6 represents a substantial fraction of Dbx1⁺ neurons.

Dbx1⁺-PV⁺ neurons exhibit canonical PV⁺ neuron projection patterns

Although Dbx1⁺ neurons account for only ∼10% of GPe neurons, it is possible that they target a unique area and serve an important function. As Cre is not expressed in adult Dbx1-Cre mice (Bielle et al., 2005) (Fig. 2b), we do not have true genetic access to this neuron population. Rather than relying on standard Cre-inducible viral approaches, Alexa-conjugated cholera toxin b (CTb), a widely used retrograde tracer, was used to map the axonal projection patterns of Dbx1⁺ neurons. Given the majority of Dbx1⁺ neurons are PV⁺, we first targeted the STN, the principal recipient of PV⁺ GPe input, in our connectome survey (Hernández et al., 2015; Hegeman et al., 2016). As expected, we found Dbx1⁺ neurons project primarily to the STN (23 ± 9%, n = 94 neurons, 9 sections) (Fig. 3c, Table 4). Additionally, but to a much lesser extent, we found Dbx1⁺ neurons project to the substantia nigra (SN, 2 ± 0%, n = 20 neurons, 9 sections). These numbers are likely an underestimation due to incomplete coverage of the target areas (i.e., STN and SN) with these injections. Importantly, although ∼10% of Dbx1⁺ GPe neurons coexpress Npas1, we observed no projections to the dorsal striatum (0 of 42 CTb⁺ neurons, 17 sections) (Table 4). The negative results were not due to poor labeling efficiency as neighboring Dbx1⁻ neurons were evidently labeled in the same section. To examine if Dbx1⁺ neurons account for the recently described noncholinergic, cortically-projecting GPe neurons (Van der Kooy and Kolb, 1985; Chen et al., 2015; Saunders et al., 2015; Schwarz et al., 2015; Ährlund-Richter et al., 2019; Sun et al., 2019), CTb was injected into various cortical regions of Dbx1-L-tdTom mice. Systematic analysis of multiple cortical areas including the somatomotor (MO, n = 4 mice), somatosensory (SS, n = 2 mice), anterior cingulate (AC, n = 1 mouse), agranular (AG) and orbital (ORB, n = 2 mice) cortices revealed Dbx1⁺ neurons do not project to any of these cortical regions (0 of 52 CTb⁺ neurons, 9 sections). The presence of CTb⁺ neurons within the GPe confirmed that the negative results were not due to failed injections (see below).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Retrograde tracing analysis. a, Representative injection sites from retrograde tracing connectome analysis. CTb (green) with or without LVretro-Cre (LV, red) was injected into dStr (top left), STN (top right), SNr (bottom right) and mounted with DAPI (blue) to visualize cytoarchitecture. b, Retrograde labeling of GPe-STN neurons with both CTb and LV tracing techniques. CTb (top left, green) labeled and LV (bottom left, magenta) GPe neurons from STN injection were primarily PV⁺ and did not colocalize with Npas1 immunostaining (top and bottom right). c, Retrograde labeling in Dbx1-L-tdTom mice shows Dbx1⁺ neurons (magenta) project to STN (top left) and SNr (top right) as indicated by colocalization with CTb (green). Circles denote colocalization. Arrowheads denote CTb⁺ STN projecting neurons that lack expression of Dbx1. Bottom, coronal view of a representative injection to the PF (left) along with expected positive cortical fluorescence (MO, right). No fluorescence was observed in the GPe. dStr, dorsal striatum; Ctx, cortex; GPe, external globus pallidus; GPi, internal globus pallidus; Hp, hippocampus; Pf, parafascicular nucleus; LGd, lateral geniculate, dorsal; MO, somatomotor cortex; ORB, orbital cortex; SI, substantia innominata; SNr, substantia nigra pars reticulata; STN, subthalamic nucleus; TRN, thalamic reticular nucleus; ZI, zona incerta; ac, anterior commissure; cpd, cerebral peduncle; ic, internal capsule; wm, white matter.

To confirm that the findings from retrograde tracing were not just due to obscure biases with the CTb, a pseudotyped-lentivirus was used for retrograde delivery of Cre recombinase (LVretro-Cre) (Knowland et al., 2017) in LSL-tdTomato mice. Unlike CTb, this strategy gives more robust, unambiguous cytoplasmic tdTomato expression. CTb and LVretro-Cre were injected into various known areas that receive GPe input, including dStr, STN, SNr, parafascicular nucleus (PF) (Kita, 2007). We observed comparable labeling patterns with LVretro-Cre and CTb injections into the dStr (Npas1⁺_LVretro: 75 ± 8%, n = 200 neurons, 9 sections; Npas1⁺_CTb: 100 ± 0%, n = 18 neurons, 9 sections) and STN (PV⁺_LVretro: 76 ± 7%, n = 278 neurons, 9 sections; PV⁺_CTb: 73 ± 11%, n = 263 neurons, 9 sections). A different scenario was observed with SNr injections—there was not a predominant neuron class that projected to the SNr (PV⁺_LVretro: 67 ± 26%, n = 51 neurons, 9 sections; PV⁺_CTb: 33 ± 11%, n = 14 neurons, 9 sections) (Fig. 3c, Table 4). Such a discrepancy between LVretro-Cre and CTb observed in the SNr data could be attributed to a number of factors including complex topographical organization of the GPe projection and CTb spreading into areas that are adjacent to the SNr and receive Npas1⁺ input (e.g., substantia nigra pars compacta). Furthermore, LVretro-Cre and CTb injections into the PF yielded labeling throughout the cortex, most prominently in the MO and AG regions (Fig. 3c), consistent with prior observations (Sherman, 2016; Mandelbaum et al., 2019). However, contrary to reports of a GPe-PF connection from other laboratories (Mastro et al., 2014), we observed no labeling in the GPe with these injections (LVretro-Cre, 4 mice; CTb, 9 mice). It is possible that CTb and LVretro-Cre both have low labeling efficiencies, making it difficult to conclude the absence of sparse projections.

Cortex and GPe are reciprocally connected

Although we found no Dbx1⁺ neurons projecting to the cortex, we identified a subset of GPe neurons that are cortex-projecting. CTb-based tracing revealed primary somatosensory (SSp, aka S1), primary somatomotor (MOp, aka M1), AG and ORB cortices as the primary targets of cortical-projecting GPe neurons (CTb, n = 116 neurons, 21 sections). To confirm these regions as the primary targets, we injected LVretro-Cre into frontal cortical regions of LSL-tdTomato mice. As expected, we observed a population of retrogradely labeled GPe neurons, i.e., tdTomato⁺ (LV, n = 293 neurons, 24 sections) (Fig. 4b).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Cortico-pallido-cortical macroscopic anatomy. a, Different cortical subregions examined in this study are highlighted. For clarity, frontal (top) and horizontal (bottom) views are shown. b, A confocal micrograph showing a representative example of retrogradely-labeled cortex-projecting GPe neurons (arrowhead) using LVretro-Cre in a LSL-tdTomato mouse; PV-immunolabeling (green) was performed in this example. Inset: experimental setup. LVretro-Cre and CTb were injected into different cortical areas mentioned in a, c, Top left, experimental setup: LVretro-Cre and CTb were injected into the GPe. Cortical inputs to the GPe were mapped using two-photon tomography. Top right, Two-photon image showing the location of the injection site. Bottom, representative two-photon images from coronal sections showing GPe-projecting cortical neurons were found primarily in layer 5 and 6 of MO and SS. d, Left, quantification of GPe-projecting neurons across the entire cortex. Medians, interquartile ranges, and 10^th to 90^th percentiles are represented in a graphical format. Right, laminar position of GPe-projecting neurons. e, Low-magnification image of Npas1⁺ pallido-cortical axons spanning across ORB, MO, and SS. Note the highest density occurred in layers 5 and 6 of MO followed by SS and dropped off precipitately rostrally in the ORB. Axons extend as far as layer 2/3. f, Local cortical infection in a Npas1-Cre-tdTom mouse confirmed axons visible in rostral cortical regions were from GPe projection and not ectopic infection of cortical neurons in the caudal areas. Injection site in SS (left) resulted in very low density of caudal to rostral cortico-cortical connectivity in MO and ORB (right). Arrowheads indicate the presence of cortical axons that arose from the more caudal regions. dStr, dorsal striatum; CLA, claustrum; MOp, primary somatomotor; MOs, secondary somatomotor; ORB, orbital; Sp, septum; SSp-n, primary somatosensory, nose; SSp-bfd, primary somatosensory, barrel field; SSp-ll, primary somatosensory, lower limb; SSp-m, primary somatosensory, mouth; SSp-ul, primary somatosensory, upper limb; SSp-tr, primary somatosensory, trunk; SSs, secondary somatosensory; VISC, visceral; ac, anterior commissure; fa, anterior forceps; wm, white matter.

Though cortical input is known to reach the GPe through the cortico-dStr-GPe and the cortico-STN-GPe pathways (Nambu et al., 2000; Kita, 2007; Jaeger and Kita, 2011; Iwamuro et al., 2017), there is increasing evidence that cortical input targets the GPe directly (Naito and Kita, 1994; Milardi et al., 2015; Smith and Wichmann, 2015; Karube et al., 2019). Using rabies virus tracing, we have recently confirmed the existence of cortical input to GPe neurons (Jeong et al., 2016; Hunt et al., 2018). However, we have not provided a complete representation of the input regions and their relative size. In this study, we sought to map the input from the entire cortical mantle with LVretro-Cre in LSL-tdTomato mice. Automated, serial two-photon tomography was used to provide an unbiased quantification and atlas registration (see Materials and Methods). As expected, input neurons (i.e., tdTomato⁺) from a wide array of brain regions to the GPe were evident (data not shown). A full description of the brain-wide input to the GPe will be detailed in a later publication. The cortical input amounts to ∼10% of the total input to the GPe (n = 4205 of 45223 neurons, 8 mice). Consistent with our previous observation, a notable input comes from the SSp followed by MOp. Additionally, but to a much lesser extent, input cells were found primarily in layers 5 and 6 of MOs, SSs, and lateral regions, such as gustatory area, visceral area, and claustrum. These results are summarized in Figure 4, c and d.

Npas1⁺-Nkx2.1⁺ neurons are cortex-projecting

A substantial amount of the cortex-projecting neurons displayed a distinctive larger soma and were located in the caudoventral regions of the GPe—features that are characteristic of ChAT⁺ neurons (Fig. 5a). Immunolabeling for ChAT revealed 50% (50 ± 0%, n = 19 of 42 neurons, 3 sections) of tdTomato⁺ neurons were ChAT⁺. The remaining cortex-projecting GPe neurons were ChAT⁻, i.e., noncholinergic. Our results are highly consistent with prior observations (Saunders et al., 2015; Ährlund-Richter et al., 2019). However, as the identity of noncholinergic, cortically projecting neurons remains elusive, we sought to characterize the molecular profile of these neurons. Through immunolabeling, we identified the ChAT⁻ neurons to be Nkx2.1⁺ (32 ± 6%, n = 20 of 68 neurons, 6 sections) and Npas1⁺ (25 ± 6%, n = 41 of 147 neurons, 13 sections). Furthermore, a population of cortex-projecting neurons was found to be both Nkx2.1⁺ and Npas1⁺ (20 ± 7%, n = 8 of 36 neurons, 3 sections) (Table 4). As expected, the same neurons that expressed Npas1 and Nkx2.1 did not express Foxp2 (0 of 36 neurons, 3 sections). Similarly, we observed a very small fraction of neurons that were immunoreactive for PV (0 ± 0%, n = 3 of 65 neurons, 8 sections). Although abundant retrogradely-labeled neurons were found with injections targeting MOp, SSp, AG and ORB, we observed low levels or nondetectable retrograde labeling with injection to neighboring frontal regions such as secondary somatomotor (MOs, aka M2) and secondary somatosensory (SSs, aka S2) cortices.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Cortex-projecting neuron properties. a, LV retrograde labeled (magenta) GPe neurons with PV immunostaining (green). Note that cortical projecting GPe neurons are not PV⁺. Arrowhead indicates a LV-labeled neuron with a large cell body characteristic of cholinergic neurons. b, Confocal micrograph showing the coexpression (dotted circles) of Npas1 (yellow) and Nkx2.1 (blue) in cortex-projecting GPe neurons (magenta). Inset: an example of a neuron (shown at the same magnification) that has a large cell body and low Nkx2.1 expression, features of cholinergic neurons within the confines of the GPe. c, High-magnification confocal micrographs of axons in the Ctx, dStr, and TRN with injection of a CreOn-ChR2 AAV into the GPe of a Npas1-Cre-tdTom mouse. Asterisks in the top left denote putative terminals. Bottom right, high density of synaptic boutons in the TRN of Nkx2.1-F-tdTom mice. d, Voltage-clamp recordings of the Npas1⁺ input in a cortical neuron within layers 5 and 6. The recorded neuron was held at -70 mV with a high Cl⁻ internal; IPSCs (IPSCs) were evoked from 20 Hz paired-pulse blue light stimulation (indicated by gray circles). Note the fast and depressing responses evoked. Inset: location of the recorded neuron (asterisk) in the Ctx is shown. IPSCs were attenuated with extracortical stimulation (data not shown) and abolished with tetrodotoxin (TTX, 1 μm). Application of 4-aminopyridine (4-AP, 100 μm) in the presence of TTX restored the response with intracortical stimulation. IPSCs were completely blocked with SR95531 (10 μm). e, Voltage-clamp recording of a TRN neuron with identical experimental setup shown in d, Note the facilitating responses evoked. Inset: location of the recorded neuron (asterisk) is shown. Responses were sensitive to the application of SR95531 (10 μm). f, Left, Pie charts summarizing the percentages of responders in Ctx and TRN. Right, Medians and interquartile ranges of IPSC amplitudes are represented in a graphical format. dStr, dorsal striatum; Ctx, cortex; TRN, thalamic reticular nucleus; ic, internal capsule; wm, white matter.

To further demonstrate that Npas1⁺ neurons project to the cortex, we injected Npas1-Cre-tdTom mice with a Cre-inducible ChR2-eYFP AAV (see Materials and Methods) into the GPe to trace the arborization patterns of Npas1⁺ axons. Consistent with our previous studies (Hernández et al., 2015; Glajch et al., 2016), dense Npas1⁺ axons were visible in the dStr. Moreover, we observed axons throughout the MO and SS, but its density dropped off precipitously once passed the ORB. This axonal projection pattern is distinct from that of ChAT⁺ neurons, which arborizes more broadly, including in more caudal regions of the cortex (Parent et al., 1981; Moriizumi and Hattori, 1992; Saunders et al., 2015). To confirm axons observed in the frontal cortex were not a result of ectopic infection of caudal cortical regions, we injected the same Cre-inducible ChR2-eYFP AAV into the SSp region directly above the GPe. We observed only sparse cortico-cortical axons running rostrocaudally (Fig. 4f), further confirming that the axons observed in the frontal regions were indeed projections from Npas1⁺ neurons.

The Npas1⁺ axons appeared to arborize heavily in layers 5 and 6 and were present as superficial as layer 2/3 (Fig. 4e). Under high magnification, perisomatic basket-like structures can be found in layer 5 of the MO. In addition, Npas1⁺ axons were found in the TRN, with more moderate projections in the zona incerta (ZI) and SN. Within the SN, we observed a higher density of fibers in the SNc (data not shown). The presence of Npas1⁺ axons in the TRN is consistent with the high density of Nkx2.1⁺ synaptic boutons in the area (Fig. 5c). As TRN neurons do not express Nkx2.1, the synaptic boutons observed arose from an extrinsic source. These observations suggest Npas1⁺-Nkx2.1⁺ GPe neurons form this projection, and thus provide new insights into the molecular identity of GPe neurons that project to the TRN (Asanuma, 1989, 1994; Cornwall et al., 1990; Hazrati and Parent, 1991; Gandia et al., 1993; Shammah-Lagnado et al., 1996; Kayahara and Nakano, 1998; Pazo et al., 2013; Clemente-Perez et al., 2017; Nakajima et al., 2019).

To evaluate the contact probability and the physiological properties of the synaptic connections from Npas1⁺-Nkx2.1⁺ neurons to the cortex and TRN, Npas1⁺ neurons in the GPe were infected with a ChR2-eYFP AAV; optogenetic-based mapping and patch-clamp recordings were performed from unidentified neurons in the cortex and TRN. As Npas1 axons primarily arborized in cortical layers 5 and 6, neurons in these layers were targeted for recording. In 3 of 37 cortical neurons recorded, large (438.5–821.7 pA, n = 3 neurons) IPSCs (IPSCs) were evoked by optogenetic activation of Npas1⁺ axons from the GPe (see Materials and Methods). These were not conducting events, as they were not abolished by the co-application of tetrodotoxin (1 μm) and 4-aminopyridine (100 μm). To confirm the GABAergic nature of the events, SR95531 (a GABA_A receptor antagonist, 10 μm) was applied, which completely abolished evoked IPSCs (Fig. 5d). In contrast to the apparent low connection probability in the cortex, large IPSCs were readily evoked optogenetically in all but one of the TRN neurons tested (707.8 ± 383.8 pA, n = 5 neurons) (Fig. 5f). Importantly, as identical optogenetic conditions were used for both experiments, these data argue that the low connection probability detected in the cortex is reflective of selective targeting of cortical neuron subtypes by Npas1⁺ axons. No photocurrents were observed in any of the recorded neurons, ruling out potential contributions of the synaptic events from ectopic infection of neurons local to the recorded areas.

Pyramidal tract-type cortical neurons target the GPe

Consistent with our two-photon tomography analysis (Fig. 4c,d), single-axon reconstruction data from the MouseLight project (Economo et al., 2016) suggest that GPe-projecting cortical neurons are predominantly pyramidal tract (PT)-type. They are thick tufted pyramidal neurons that are located in the lower layer 5 (L5B), do not have cross-hemispheric collaterals, and are multiprojectional (Kita and Kita, 2012; Harris and Shepherd, 2015; Kawaguchi, 2017; Hooks et al., 2018; Shibata et al., 2018) (Fig. 6a). This inference corroborated the anterograde viral-tracing data derived from a Sim1-Cre (PT-specific driver line) (Gerfen et al., 2013)—both single-axon (Fig. 6a) and bulk tracing data argue that cortical axons enter the GPe (Fig. 6d,e). In contrast, similar experiments with Tlx3-Cre (an intratelencephalic-specific driver line) suggest that intratelencephalic neurons do not provide input to the GPe (Fig. 6c). Overall, there were general similarities between the topographical organization of the cortico-striatal projections and the cortico-pallidal projections; both frontal and motor projections targeted more rostral areas of the GPe, whereas sensory projections targeted more posterior ones (Hintiryan et al., 2016; Hooks et al., 2018).

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Pyramidal-tract but not intratelencephalic axons collateralize in the GPe. a, Single-axon reconstruction of a layer 5 (L5) cortico-pallidal neuron (neuron #AA0122) in the motor cortex. Data are adapted from the MouseLight project (http://mouselight.janelia.org). The axonal projection pattern is consistent with a pyramidal-tract (PT)-type neuron. Inset: axonal arborization of 10 different cortical neurons. GPe and axonal endpoint were used as the target location and structure queries, respectively. b, Injection site center of mass of Sim1-Cre (L5-PT) plotted and spatially clustered (n = 62, triangles). These injection sites correspond to vibrissal, forelimb, and orofacial somatosensory cortices (vS1, fS1, and orfS1); vibrissal, forelimb, and lower limb motor cortices (vM1, fM1, and llM1); and frontal areas (anterior lateral motor cortex (ALM) and secondary motor cortex (M2)). Eight clusters shown in red (M2), orange (ALM), purple (vM1), burgundy (fM1), green (llM1), yellow (fS1), teal (vS1), and gray (orfS1). Indeterminate injection sites are white. Sites are superimposed on an image of the dorsal surface of mouse cortex. Black cross marks indicate midline and bregma. For simplicity, injection sites in Tlx3-Cre (L5-IT) are not shown (see Hooks et al., 2018, for further information). c, d, Tlx3-Cre (IT-type) projections from fM1 in dStr but not GPe. Sim1-Cre (PT-type) projections from fS1 (left) and fM1 (right) in dStr and GPe (pink arrows). Inset: Coronal images of injection sites in Tlx3-Cre and Sim1-Cre showing the cell body locations and their axonal projections. e, Coronal images of the average normalized PT-type projection to GPe from eight cortical areas. Each column is a cortical projection, with rows going from anterior (top) to posterior (bottom). Each projection is normalized for comparison within the projection.

To provide a confirmation of the existence of synaptic terminals formed by cortical axons in the GPe, we injected AAVretro-ChR2-eYFP into the GPe of Emx1-Cre mice (Gorski et al., 2002). This resulted in selective ChR2-eYFP expression in cortical neurons and their axons in the GPe, STN, and dStr (Fig. 7a). A high density of labeled neurons (and processes) were observed within the ORB, MO, and SS. This expression pattern is in agreement with our two-photon tomography data (Fig. 4c,d) and the axonal projection patterns revealed by Sim1-Cre (Fig. 6d,e). Furthermore, these data reinforce the idea that the cortical axons observed in the GPe in both the single axon and bulk tracking studies were not simply passage fibers. Consistent with this idea, colocalization of VGluT1 in ChR2-eYFP-labeled varicosities was readily observed (Fig. 7b). To provide more definitive evidence, ex vivo voltage-clamp recordings were performed; optogenetic activation of cortical input evoked EPSCs in 20 of 25 GPe neurons. However, the size of the EPSCs (29.0 ± 18.2 pA) spans over three orders of magnitude in the recorded neurons. Consistent with the single-axon and bulk tracing data, which both show that cortico-pallidal neurons produce collaterals within the dStr, EPSCs (100.8 ± 79.1 pA) were evoked readily across all striatal projection neurons tested (n = 6) (Fig. 7d).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Cortex neurons form functional synapses on GPe neurons. a, Top, Confocal micrograph of a sagittal section from an Emx1-Cre mouse, showing the neuronal elements expressing ChR2-eYFP delivered from an AAVretro injection into the GPe. Inset, ChR2-eYFP⁺ layer 5 neuron with morphology typical of pyramidal neurons is shown. Bottom left, Cortical axons were observed at the GPe level. An enrichment of axons was present at the rostral pole of the GPe, immediate adjacent to the dStr (pink arrows). Bottom right, High density of cortico-pallidal axons were observed to collateralize in the subthalamic nucleus (STN). b, A high-magnification image showing the colocalization (white arrowheads) of VGluT1 (magenta) in corticopallidal axons (green). c, Epifluorescence image from ex vivo tissue showing robust expression of ChR2-eYFP in cortical neurons. Top, AAVretro-CreOn-ChR2-eYFP was injected into the GPe of an Emx1-Cre mouse. Robust ChR2-eYFP expression in the MO was observed (left). ChR2-eYFP⁺ neurons were readily seen at high-magnification (right). Asterisk indicates the soma of a typical ChR2-eYFP⁺ neuron. Bottom, Cortico-pallidal axons were preserved in an ex vivo slice preparation. d, Top, Functional cortical inputs were recorded in GPe neurons (20 of 25) and dStr projection neurons (6 of 6). EPSCs were evoked with optogenetics. Bottom, Box and scatter plots summarizing EPSC amplitude recorded from GPe neurons (left) and dStr SPNs (right). Note the large variance in the data. Red arrowheads indicate cortical axon bundles; white arrowheads indicate apical dendrites. dStr, dorsal striatum; GPe, external globus pallidus; STN, subthalamic nucleus; ZI, zona incerta; ac, anterior commissure; cpd, cerebral peduncle; ic, internal capsule; wm, white matter.

Lhx6⁺ neuron subtypes have unique spatial patterns

Spatial distribution may vary with neuronal identity, as seen in the cortex, for example, where neurons are organized in a highly laminar-specific manner (Huang et al., 2007; Harris and Shepherd, 2015; Wamsley and Fishell, 2017). We therefore integrated spatial distribution as a metric to phenotype different GPe neuron types (Figs. 8, 9). Overall, more neurons populate toward the rostral pole of the GPe. We noted that ChAT⁺ neurons and Lhx6⁺-Sox6⁺ neurons are displaced more heavily toward the caudoventral regions and rostroventral regions, respectively. All other identified neurons are more evenly distributed throughout the GPe. This analysis, however, does not capture any lateromedial gradients. As neurons were routinely sampled from three different lateromedial planes across the GPe, the abundance of identified neurons were tallied. Consistent with previous studies (Mastro et al., 2014; Hernández et al., 2015), we observed a lateromedial gradient of PV⁺ neurons (lateral = 46 ± 10%, n = 4368 neurons, 24 sections; intermediate = 45 ± 11%, n = 5113 neurons, 27 sections; medial = 32 ± 7%, n = 2829 neurons, 20 sections) and Lhx6⁺ neurons (lateral = 28 ± 6%, n = 1422 neurons, 10 sections; intermediate = 42 ± 9%, n = 2050 neurons, 10 sections; medial = 45 ± 12%, n = 2190 neurons, 8 sections). PV⁺ neurons were more concentrated in the lateral than the medial level; the reverse pattern was found for Lhx6⁺ neurons. In Figure 9, we illustrate the distribution of Lhx6⁺-Sox6⁻ neurons (lateral = 8 ± 4%, n = 187 neurons, 5 sections; intermediate = 10 ± 1%, n = 292 neurons, 5 sections; medial = 21 ± 7%, n = 459 neurons, 5 sections), which follow the same pattern as pan-Lhx6⁺ neurons. Although PV⁺-Lhx6⁺ neurons displayed a similar pattern (lateral = 8 ± 2%, n = 116 neurons, 4 sections; intermediate = 11 ± 6%, n = 202 neurons, 3 sections; medial = 8 ± 1%, n = 175 neurons, 3 sections) as pan-Lhx6⁺ neurons, Npas1⁺-Lhx6⁺ neurons do not (lateral = 14 ± 2%, n = 231 neurons, 4 sections; intermediate = 15 ± 2%, n = 308 neurons, 4 sections; medial = 14 ± 2%, n = 284 neurons, 4 sections). The lateromedial gradients of different GPe neuron types are summarized in Figure 9.

• Download figure
• Open in new tab
• Download powerpoint

Figure 8.

Spatial distribution of GPe neuron subtypes. a, Spatial information of GPe neurons cannot be represented relative to bregma location (top, lower left) because of its complex geometry. To mathematically describe the spatial distribution of GPe neurons in each brain section, fixed mouse brains were sagittally sectioned and histologically processed. Images were manually aligned to a reference atlas. GPe neurons located at six different lateromedial levels (2.73 mm, 2.53 mm, 2.35 mm, 2.15 mm, 1.95 mm, and 1.73 mm) were charted and collapsed onto a single plane. As the GPe is similarly shaped across the lateromedial extent (lower right), both the rostrocaudal and dorsoventral extent were assigned to 0 and 1. The address of each neuron is defined by their x–y coordinates and represented with a marker (black). To capture the aggregate spatial distribution, a geometric centroid (red) of each neuron population was then determined to represent the center of mass in both x and y dimensions. Centroids are then used as the origin for the polar histograms in b, Size of each sector represents the relative neuron count as a function of direction. b, Representative data of neurons from two individual mice are shown in each case, except for retrogradely labeled cortically projecting neurons (n, 7 mice; 119 neurons; 15 sections). Each marker represents a neuron. The density of neurons are encoded with a yellow-blue gradient. Hash marks, which represent the dorsoventral and rostrocaudal axes, are presented with the centroids and polar histograms to facilitate comparison. Bin sizes in the polar histograms were chosen based on the size of each neuron population. The (x, y) centroid values for the respective GPe distributions were as follows: HuCD⁺ (0.3798, 0.4168); Nkx2.1⁺ (0.3599, 0.4439); Sox6⁺ (0.3587, 0.4529); PV⁺ (0.3205, 0.4699); Lhx6⁺ (0.3918, 0.3827); Lhx6⁺-Sox6⁻ (0.3755, 0.3164), Dbx1⁺ (0.3679, 0.3828); ChAT⁺ (0.6024, 0.3569); Npas1⁺ (0.4106, 0.4140); Npas1⁺-Foxp2⁺ (0.3695, 0.4676); Npas1⁺-Nkx2.1⁺ (0.4026, 0.4278); Ctx-projecting GPe neurons (0.5061, 0.2911).

• Download figure
• Open in new tab
• Download powerpoint

Figure 9.

Lateromedial gradients and relative abundance of different GPe neuron classes. a, Spatial maps of the pan-Lhx6⁺ and unique Lhx6⁺ Sox6⁻ GPe neuron populations. Both populations display a lateromedial gradient with more neurons populating the medial GPe. b, Relative abundance of neuron classes in different lateromedial subdivisions of the GPe (Sox6⁺, lateral: 60 ± 12%, n, 4457 neurons, intermediate: 63 ± 11%, n, 5112 neurons, medial: 50 ± 11%, n, 3286 neurons; Nkx2.1⁺, lateral: 56 ± 7%, n, 3365 neurons, intermediate: 53 ± 9%, n, 3878 neurons, medial: 64 ± 14%, n, 3265 neurons; PV⁺, lateral: 46 ± 10%, n, 4368 neurons, intermediate: 45 ± 11%, n, 5113 neurons, medial: 32 ± 7%, n, 2829 neurons; Lhx6⁺, lateral: 28 ± 6%, n, 1422 neurons, intermediate: 42 ± 9%, n, 2050 neurons, medial: 45 ± 12%, n, 2190 neurons; Npas1⁺, lateral: 32 ± 6%, n, 2635 neurons, intermediate: 31 ± 6%, n, 2903 neurons, medial: 27 ± 7%, n, 2252 neurons; Foxp2⁺, lateral: 24 ± 3%, n, 939 neurons, intermediate: 26 ± 4%, n, 1115 neurons, medial: 25 ± 6%, n, 686 neurons; Dbx1⁺, lateral: 10 ± 2%, n, 1219 neurons, intermediate: 9 ± 2%, n, 1540 neurons, medial: 8 ± 2%, n, 1121 neurons; ChAT⁺, lateral: 6 ± 1%, n, 100 neurons, intermediate: 4 ± 1%, n, 76 neurons, medial: 6 ± 3%, n, 91 neurons). Percentage total was calculated from HuCD⁺ cells within each section. Note that PV and Npas1 were expressed in a largely nonoverlapping fashion (2 ± 2%, n, 96 neurons, 12 sections). In contrast, considerable overlap between Lhx6 and PV (28 ± 7%, n, 654 neurons, 12 sections) or Npas1 (35 ± 5%, n, 818 neurons, 12 sections) was observed; the remaining fraction was uniquely labeled with Lhx6. Medians and interquartile ranges are represented in a graphical format. Asterisks denote statistical significance level: **p < 0.01, Mann–Whitney U test.

GPe neuron subtypes have distinct intrinsic properties

To further define GPe neuron subtypes, ex vivo electrophysiological analyses were performed systematically on genetically-identified GPe neuron subtypes, including the less well studied neuron subtypes. We used recording and analysis routines (see Materials and Methods) identical to those used in our previous study to facilitate cross-comparison between the two (Hernández et al., 2015).

To identify Foxp2⁺ neurons, we infected Foxp2-Cre mice in the GPe with a CreOn-mCherry AAV (see Materials and Methods). To confirm the validity of the approach, a subset of these mice were examined for cellular specificity of Cre-mediated mCherry expression (mCherry⁺). In nearly all mCherry⁺ neurons examined, Foxp2 was expressed (100 ± 0%, n = 473 of 485 neurons, 6 sections). No GPe neurons expressed mCherry when the same virus was injected in wild-type mice (n = 2 mice, 6 sections). For electrophysiological analysis, we qualitatively categorized Lhx6⁺ neurons as “bright” and “dim” based on their GFP expression level (Fig. 10a), though the definitive identities of Lhx6⁺_bright and Lhx6⁺_dim neurons were not confirmed post hoc. To identify PV⁺-Dbx1⁺ neurons, an intersectional cross was made—PV-Dbx1-FL-tdTom (PV-Flp;Dbx1-Cre;LSL-FSF-tdTomato)—to label PV⁺-Dbx1⁺ neurons (tdTomato⁺) (Fig. 10a). To unequivocally identify Npas1⁺-Lhx6⁺ neurons, which are equivalent to Npas1⁺-Nkx2.1⁺ neurons, we crossed Npas1-tdTom and Lhx6-GFP mice. Double-positive (tdTomato⁺ and GFP⁺) neurons were targeted for recordings. ChAT⁺ neurons, which can be identified based on their unique somatodendritic morphology, were not included in this analysis as we have previously established that they have a very distinct electrophysiological profile (Hernández et al., 2015). Last, because we do not have genetic access to Lhx6⁺-Sox6⁻ neurons, we did not have a means to target them for direct recording.

• Download figure
• Open in new tab
• Download powerpoint

Figure 10.

Genetically identified GPe neurons differ in their spontaneous activity. a, Representative bright-field and epifluorescence images of GPe neuron subtypes in ex vivo brain slices. Foxp2⁺ neuron (top, bright-field and mCherry), Lhx6⁺_bright neurons and Lhx6⁺_dim (bottom left, GFP), and PV⁺-Dbx1⁺ neurons (bottom right, tdTomato) were captured at 60× magnification. Note the difference in the morphology and GFP expression among Lhx6⁺ neurons. b, Box-plot summary of the electrophysiological properties of identified GPe neuron subtypes. Data are ranked based on the median values. See Tables 5 and 6 for median values, sample sizes, and statistical analysis. Medians, interquartile ranges, and 10^th to 90^th percentiles are represented in a graphical format. c, Top left, visualization of the clustered data on spontaneous activity (rate and CV_ISI) for k, 2 clusters. Centroid values for cluster 1 (teal circle) and 2 (tan squares) are as follows: 18.9 Hz, 0.16 and 7.9 Hz, 0.24. Middle left, Silhouette plots for different clusters. Bottom right, Silhouette values are plotted against cluster numbers showing an optima at k, 2. Large positive silhouette values indicate that the data point is close to its cluster's centroid, whereas negative silhouette values indicate that the data point is closer to the centroid of the other cluster. Right, a series of pie charts showing the membership assignment of different genetically defined GPe neuron subtypes. The membership assignment in cluster 1 (teal) and 2 (tan) for each neuron subtypes are as follows: PV⁺-Dbx1⁺ (81.3%, 18.8%, n, 16), PV⁺ (73.9%, 26.1%, n, 111), Lhx6⁺_dim (71.4%, 28.6%, n, 7), Npas1⁺-Lhx6⁺ (21.4%, 78.6%, n, 14), Npas1⁺ (17.7%, 82.3%, n, 62), Lhx6⁺_bright (13.3%, 86.7%, n, 15), Foxp2⁺ (0.0%, 100.0%, n, 20). Data are not shown for Dbx1⁺ (50.0%, 50.0%, n, 20) and Lhx6⁺ (61.1%, 38.9%, n, 18), which both contain a mixture of PV⁺ neurons and Npas1⁺ neurons. d, Bias-corrected and accelerated (BC_a) bootstrap estimation of effect sizes (median differences) and 95% confidence intervals. The median difference in spontaneous rate for seven comparisons against the PV⁺ neurons (left) and Npas1⁺ neurons (right) are shown. Median differences are plotted as bootstrap sampling distributions. Each median difference is depicted as a circle. Median differences are also encoded by color saturation. Lower and upper confidence interval bounds are indicated by the horizontal bars. Lhx6⁺_dim neurons and PV⁺-Dbx1⁺ neurons are statistically nonsignificant from PV⁺ neurons (p, 0.45 and 0.24). Lhx6⁺_bright neurons, Npas1⁺-Lhx6⁺ neurons, and Lhx6⁺ neurons are statistically nonsignificant from Npas1⁺ neurons (p = 0.44, 0.29, and 0.066).

To highlight, Foxp2⁺ neurons and PV⁺-Dbx1⁺ neurons were at the extremes in terms of their spontaneous activity level measured in cell-attached recordings (Foxp2⁺ = 6.1 ± 1.9 Hz, n = 20; PV⁺-Dbx1⁺ = 18.4 ± 3.6 Hz, n = 16), with the rest of the neuron subtypes displaying properties traversing the spectrum. Furthermore, our results corroborate findings from our prior studies (Hernández et al., 2015) that PV⁺ neurons fire at a higher rate than Npas1⁺ neurons (PV⁺ = 16.7 ± 3.4 Hz, n = 111; Npas1⁺ = 8.4 ± 3.1 Hz, n = 63, p < 0.0001). Lhx6⁺ neurons exhibited firing rates that are in between PV⁺ neurons and Npas1⁺ neurons (Lhx6⁺ = 10.5 ± 5.1 Hz, n = 43). Within the Lhx6⁺ population, Lhx6⁺_bright neurons and Lhx6⁺_dim neurons exhibited different firing rates (Lhx6⁺_bright = 8.1 ± 2.7 Hz, n = 18; Lhx6⁺_dim = 16.5 ± 1.2 Hz, n = 7, p = 0.0039). Additionally, the Npas1⁺-Lhx6⁺ neurons had a higher spontaneous firing than the Foxp2⁺ population (Npas1⁺-Lhx6⁺ = 10.4 ± 2.8 Hz, n = 14, p = 0.021) but was comparable to Lhx6⁺_bright neurons (p = 0.17) (Fig. 10b). As expected, Dbx1⁺ neurons exhibited firing behavior that was most consistent with its composition, with most neurons exhibiting similar firing rates to PV⁺ neurons and some more similar to Npas1⁺ neurons (Tables 5, 6, Dbx1⁺ = 13.2 ± 2.7 Hz, n = 21).

As we have previously shown that the regularity of the firing varied with GPe neuron types (Hernández et al., 2015), we measured the coefficient of variation in the interspike-interval (CV_ISI). Consistent with our prior observations, PV⁺ neurons exhibited a lower CV_ISI than Npas1⁺ neurons (PV⁺ = 0.14 ± 0.03, n = 111; Npas1⁺ = 0.27 ± 0.10, n = 63, p < 0.0001). Dbx1⁺ neurons and PV⁺-Dbx1⁺ neurons had CV_ISI that were statistically similar to those of PV⁺ neurons (Dbx1⁺ = 0.14 ± 0.03, n = 21; PV⁺-Dbx1⁺ = 0.16 ± 0.05, n = 16). Foxp2⁺ neurons had the largest variability in their firing rate, as indicated by the highest CV_ISI within the GPe neuronal population (Foxp2⁺ = 0.36 ± 0.09, n = 20). Npas1⁺-Lhx6⁺ neurons had CV_ISI statistically similar to that of the Lhx6⁺_bright neurons (Npas1⁺-Lhx6⁺ = 0.25 ± 0.07, n = 14, Lhx6⁺_bright = 0.21 ± 0.04 n = 18, p = 0.51).

To provide comprehensive electrophysiological profiles, we investigated a range of electrophysiological properties of GPe neurons in whole-cell current-clamp recordings. As seen in prior studies, PV⁺ neurons had the highest maximum firing rate (PV⁺ = 198 ± 24 Hz, n = 49). Lower maximum firing rates were observed in Npas1⁺ neurons and Foxp2⁺ neurons (Npas1⁺ = 96 ± 34 Hz, n = 33; Foxp2⁺ = 55 ± 23, n = 16) when compared with PV⁺, Dbx1⁺, and PV⁺-Dbx1⁺ neurons (p < 0.0001 for all). Within the Lhx6 population, Lhx6⁺_bright neurons and Lhx6⁺_dim neurons exhibited different maximum firing rates (Lhx6⁺_bright = 93 ± 31 Hz, n = 24; Lhx6⁺_dim = 161 ± 18 Hz, n = 10, p = 0.00054). GPe neuron subtypes also exhibited different responses to hyperpolarizing current injections (Fig. 10b). More negative trough potentials were noted in the Npas1⁺ neurons and Foxp2⁺ neurons (Npas1⁺ = −147 ± 17 mV, n = 45; Foxp2⁺ = −168 ± 8 mV, n = 16) when compared with PV⁺ neurons (PV⁺ = −96 ± 7 mV, n = 74, p < 0.0001 for both). Within Lhx6⁺ neurons, Lhx6⁺_bright neurons and Lhx6⁺_dim neurons exhibited different trough potentials (Lhx6⁺_bright= −129 ± 12 mV, n = 24, Lhx6⁺_dim = −111 ± 13 mV, n = 10, p = 0.010). The trough potential observed in Npas1⁺-Lhx6⁺ neurons (Npas1⁺-Lhx6⁺ = −119 ± 11 mV, n = 16) was less negative than both Npas1⁺ neurons (p = 0.0060) and Foxp2⁺ neurons (p = 0.00091). As seen in previous work from our lab (Hernández et al., 2015), PV⁺ and PV⁺-Dbx1⁺ neurons exhibited the least negative trough potential compared with the other studied GPe neuron types and were not different from one another (PV⁺ = −96 ± 7 mV, n = 74, PV⁺-Dbx1⁺ = −88 ± 16 mV, n = 21, p = 0.24). Dbx1⁺ neurons had trough potential (Dbx1⁺ = −104 ± 9 mV, n = 24) more negative than those seen in PV⁺ neurons (p = 0.0098) and PV⁺-Dbx1⁺ neurons (p = 0.0078). Similar to the trough potentials, higher sag ratios were observed for Npas1⁺ neurons and Foxp2⁺ neurons (Npas1⁺ = 1.23 ± 0.13, n = 45; Foxp2⁺ = 1.33 ± 0.13, n = 16). PV⁺ neurons had the lowest sag ratio (PV⁺ = 1.06 ± 0.03, n = 74). Within Lhx6, a difference in the sag ratio was observed between Lhx6⁺_bright neurons and Lhx6⁺_dim neurons (Lhx6⁺_bright = 1.20 ± 0.09, n = 24, Lhx6⁺_dim = 1.13 ± 0.06, n = 10, p = 0.020). The Npas1⁺-Lhx6⁺ neurons exhibited a lower sag ratio than both Npas1⁺ neurons (Npas1⁺-Lhx6⁺ = 1.06 ± 0.04, n = 16, p < 0.0001) and Foxp2⁺ neurons (p < 0.0001). As expected, the PV⁺-Dbx1⁺ neurons had statistically similar sag ratios to the PV⁺ neuron population (PV⁺-Dbx1⁺ = 1.13 ± 0.08, n = 21, p = 0.19). A full description of the electrophysiological characteristics of GPe neurons is listed in table form (Table 5).

We wanted to take advantage of the large dataset collected (n = 556 neurons) on the spontaneous activity of GPe neurons measured ex vivo as it may glean insights into the identification of GPe neurons in vivo. In particular, as the spontaneous activity level and regularity (as measured with CV_ISI) vary with neuron types, we examined if GPe neurons can be classified into electrophysiological subtypes. K-means analysis was implemented as it is one of the simplest and most commonly used strategies (Fig. 10c). This analysis returned an optima of k = 2 clusters with the centroids at (7.9 Hz, 0.24; n = 297 and 18.9 Hz, 0.16; n = 259), which approximate to the median values of Npas1⁺ neurons (8.4 Hz, 0.27) and PV⁺ neurons (16.7 Hz, 0.14), respectively. The inclusion of additional values for the number of clusters k (i.e., 3–6) returned suboptimal clustering as evaluated with silhouette analysis in Figure 10c. To evaluate the molecular correspondence of these two mathematically defined clusters, their compositions were visualized. Cluster 1 primarily consisted of PV⁺ neurons (88.2%) and cluster 2, Npas1⁺ neurons (63.8%). The mismatch between the molecular identity and the clustering membership assignment was not unexpected based on the population data of spontaneous activity level and regularity (Fig. 10b). As Foxp2 neurons are cleanly clustered within cluster 2, this analysis emphasizes the similarity between Npas1⁺-Nkx2.1⁺ and PV⁺ neurons—two primarily MGE-derived populations.

The k-means clustering further supports the notion that PV⁺ neurons and Npas1⁺ neurons are the two principal neuron classes in the GPe. A logistic regression with PV⁺ neurons ad Npas1⁺ neurons as the dependent variable and spontaneous rate and CV_ISI as the independent variables showed significant associations for both with neuron type—estimated odds ratio of 1.26 for spontaneous rate, p < 0.0001, indicating a higher value in PV⁺ neurons; estimated odds ratio of 0.08 for CV_ISI, p = 0.0020, indicating a higher value in Npas1⁺ neurons. As logistic regression indicates a strong association between spontaneous activity level and neuron type, bootstrap analysis was applied to estimate the differences between genetically-defined neuron subtypes and these two principal neuron classes (Fig. 10d). In general, 9 of the 10 characteristics considered—all except for AP height—showed differences among cell types based on the Kruskal–Wallis test, even after the Bonferroni multiple testing correction. It is notable that Npas1⁺-Lhx6⁺ neurons and Lhx6⁺_bright neurons have spontaneous activity that is statistically nonsignificant from Npas1⁺ neurons (p = 0.29 and 0.44, respectively). Furthermore, Dbx1⁺-PV⁺ neurons and Lhx6⁺_dim neurons exhibited spontaneous activity consistent with canonical PV⁺ neuron (p = 0.24 and 0.45, respectively). Pairwise comparisons of the electrophysiological characteristics across molecularly defined GPe neuron subtypes are tabulated in Table 6.

As multiple electrophysiological characteristics were obtained, we asked whether a combination of these quantitative features can statistically define GPe neurons. To determine at the single-cell level whether genetically-defined GPe neuron subtypes form separate or overlapping clusters, we applied a principal component analysis to the electrophysiological dataset (Fig. 11). Neurons with incomplete data were excluded from the analysis. As expected, and as illustrated by the dendrogram, neurons defined by PV and Npas1 transgenic lines were completely separated. Accordingly, they exhibited unique electrophysiological signatures as illustrated by the heatmap (Fig. 11a). This analysis established the correspondence of electrophysiological classes with predefined markers, i.e., PV and Npas1. Moreover, different genetically-defined GPe neurons that were targeted in this study have electrophysiological properties that can be broadly categorized into either PV⁺ neuron or Npas1⁺ neuron classes. This analysis reaffirmed our previous assertion regarding the utility of these transgenic lines and markers (Hernández et al., 2015). Though electrophysiological differences exist between Npas1⁺ neuron subclasses, hierarchical clustering of electrophysiological properties was insufficient to define GPe neuron subclasses.

• Download figure
• Open in new tab
• Download powerpoint

Figure 11.

Electrophysiological multivariate analysis of GPe neurons. a, Heatmap representation of electrical signatures of genetically identified GPe neuron subtypes. Dendrograms show the order and distances of neuron clusters and their electrical characteristics. A total of 130 neurons (n, PV⁺: 38, Npas1⁺: 19, Dbx1⁺: 19, Foxp2⁺: 16, PV⁺-Dbx1⁺: 16, Lhx6⁺_bright: 16, Lhx6⁺_dim: 7) were included in this analysis. Neurons with incomplete data were excluded from the analysis. b, Top, Scree plot (left) and pie chart (right) showing that the first three principal components (gray) capture 72.9% of the total variability in the data. Bottom, Principal component 1 and 2 account for 46.9% and 16.6% of the total variance, respectively. Prediction ellipses for PV⁺ neurons (pink) and Npas1⁺ neurons (brown) are shown. With probability 0.95, a new observation from the same group will fall inside the ellipse. c, Same dataset as a. Data are sorted by genetically identified neuron subtypes. Ordering of the clustering is the same as a.

The k-means and hierarchical clustering analyses provide complementary points of view for exploring our rich dataset. We used k-means with only spontaneous activity level and variability as the clustering variables, as they have the least amount of missing data, leading to a large dataset (n = 556 neurons). In addition, activity level and patterns have obvious appeal as they may glean new insights into the classification of GPe neurons in vivo. k-means clustering with these two variables thus provides a powerful and convenient visualization of the data. Although hierarchical clustering provides clusters by using more variables, it is not immediately apparent how each electrophysiological characteristic relates to the clusters.