Molecular Diversity of Neuron Types in the Salamander Amygdala and Implications for Amygdalar Evolution

The amygdala is a complex brain region that controls a variety of behaviors crucial for survival [Swanson and Petrovich, 1998]. In mammals, this part of the brain not only controls fear responses (i.e., escape) through direct connections with motor and autonomic systems but it is also critical for social behaviors and associative learning. The amygdala develops in the caudal telencephalon and includes nuclei derived from pallium and subpallium [Aerts and Seuntjens, 2021]. Furthermore, the amygdala harbors cells that migrate from the hypothalamus, and presumably also from the prethalamic eminence [García-Moreno et al., 2010; Morales et al., 2021; Garcia-Calero et al., 2021]. As a result of their complex developmental history, amygdalar nuclei have distinct neurochemical, hodological, and functional properties. Anatomical and molecular studies in mice suggest a great diversity of amygdalar neuron types; however, the characterization of cellular heterogeneity remains limited to a few nuclei and to the sampling of small numbers of cells [Wu et al., 2017; Zeisel et al., 2018; O’Leary et al., 2020].

Telencephalic areas with hodological and molecular similarities to the mammalian amygdala have been clearly identified in representatives of all the main classes of jawed vertebrates [Medina et al., 2017] (data on agnatha remain scarce). Therefore, the amygdala as a whole seems to be ancient in vertebrates, in line with its crucial functions. However, several questions on the evolution of the amygdala remain open: what are the exact boundaries and subdivisions of the amygdala in non-mammals? How do amygdalar subdivisions relate to each other across species? And ultimately, how did the mammalian amygdala, with its complex organization, emerge?

Unsurprisingly, the most controversial points concern the pallial amygdala, as the evolution of the pallium as a whole remains unclear. The well-known tetrapartite model for the pallium, introduced by Puelles and colleagues, postulated that the pallial amygdala is part of the so-called ventral pallium (VP), the pallial region defined by the absence of expression of the transcription factor Emx1 in its ventricular zone [Fernandez et al., 1998; Puelles et al., 2000; Gorski et al., 2002; Brox et al., 2004]. In the more recent versions of this model, the mammalian VP includes the pallial amygdala, the piriform cortex, and the ventral endopiriform nucleus, among others [Puelles, 2017]. Following this model, a VP has been identified in other vertebrates on the basis of topological criteria and conserved developmental gene expression patterns. In tetrapods, the neural territory originally defined as VP includes amygdalar regions, as well as other neural territories with controversial homologies. For example, the reptilian VP is heterogeneous; it includes the lateral cortex (an olfactory-recipient cortical region, considered the homolog of the mammalian piriform cortex) and the dorsal ventricular ridge (DVR), comprising anterior and posterior subdivisions (aDVR and pDVR). In birds, the nidopallium and arcopallium would correspond to the aDVR and pDVR, respectively [Medina et al., 2021]. While the homology of the pDVR and of the arcopallium with parts of the mammalian pallial amygdala has been reconciled, the nature of the aDVR and of the nidopallium remains debated [Bruce and Neary, 1995; Briscoe and Ragsdale, 2018; Tosches, 2021; Colquitt et al., 2021; Gedman et al., 2021; Medina et al., 2021].

Acknowledging the important differences in gene expression profiles in VP subdomains, Medina and colleagues proposed the partition of the original VP in two pallial sectors: the ventral and ventrocaudal pallium. The latter sector includes the pallial amygdala [Medina et al., 2017; Desfilis et al., 2018]. Taking these observations a step further, the new radial model for the pallium proposed by Puelles and colleagues abandons the classical tetrapartite scheme, and posits that the mammalian pallial amygdala arises from radial segments that are completely distinct from the rest of the pallium, which now takes the name of cortical pallium [Puelles et al., 2019].

As the sister group of amniotes, amphibians are extremely informative to reconstruct brain evolution in tetrapods. The amphibian pallium has a simple organization, as it consists of a periventricular cell layer and a plexiform layer, and is devoid of clearly distinct nuclei. Subdivisions of the amphibian pallium have been identified with the combination of tracing, neuroanatomical, and molecular studies. In amphibians, the terms medial, dorsal, lateral, and ventral pallium have been used to label these pallial subdivisions, and in this article, we will use this traditional terminology [González et al., 2020]. In this paper, the terms medial, dorsal, lateral, and ventral pallium are used for practical reasons just to refer to regions of the amphibian pallium, without endorsing any specific version of the tetrapartite model.

The neuroanatomical heterogeneity of the amphibian VP was already noticed by several investigators, who distinguished subdivisions of this area along the mediolateral axis even before the term VP was introduced [Northcutt and Kicliter, 1980; Neary, 1990; Bruce and Neary, 1995]. For example, Neary identified three regions of the “lateral pallium,” corresponding to the lateral and ventral pallium in modern terms [Neary, 1990], and in addition, Marín described a “subpallial-pallial transition area” (SPTA) [Marín et al., 1997a, b]. Molecular data confirmed the existence of a “ventral subdivision of the ventral pallium” of an amygdalar nature; this region was named “lateral amygdala” by Moreno, Gonzalez and colleagues [Moreno et al., 2004; Moreno and González, 2004]. The lateral amygdala is one of the three amygdalar subdivisions identified so far in amphibians. The other two, of subpallial origin, are the medial and the central amygdala [Laberge et al., 2006; Hall et al., 2013; González et al., 2020]. In Xenopus, the lateral amygdala is demarcated by the expression of the transcription factor Lhx9, and at mid-telencephalic levels, it is nested between the rest of the VP dorsally and the striatum ventrally [Brox et al., 2003; Moreno et al., 2004]. At caudal levels, the organization of the amphibian amygdala remains less clear. In urodeles, Northcutt and Kicliter [1980] identified an “amygdala pars lateralis” and proposed that this area has a pallial origin. However, Moreno and González [2007a] found expression of subpallial markers in the caudal telencephalon, and renamed this region medial amygdala, proposing its homology with the mammalian medial amygdala. These observations raise several questions about the organization of the amygdala in amphibians, and the evolution of the amygdala in tetrapods.

First, is the pallial amygdala distinct from the rest of the pallium, including the VP, in amphibians? The radial model proposes the existence of distinct radial sectors for the pallial amygdala and the cortical pallium. To confirm the existence of one (or multiple) pallial amygdala radial sector(s) in amphibians, it is necessary to clarify the boundaries with the VP and with the subpallium along the entire rostro-caudal axis of the telencephalon. Second, is the amphibian pallial amygdala heterogeneous, as it is in sauropsids and mammals? Recent work by Garcia-Calero et al. [2020], Puelles and colleagues demonstrated that the mammalian pallial amygdala develops from five radial sectors, which give rise to amygdalar nuclei with distinct gene expression profiles and connectivity. A single-cell RNA sequencing (scRNAseq) analysis of the turtle pallium identified five distinct clusters of single-cell transcriptomes (a proxy for cell types) in the pDVR, which were mapped to distinct subregions of the pDVR using in situ hybridization for marker genes [Tosches et al., 2018]. In contrast, it is unclear whether the amphibian pallial amygdala includes multiple types of neurons or distinct radial subdivisions. Finally, does the amphibian amygdala include neurons with extra-telencephalic origins, as it is the case in sauropsids and mammals? [García-Moreno et al., 2010; Garcia-Calero et al., 2021; Morales et al., 2021; Metwalli et al., 2022].

To address these questions, here we characterize glutamatergic neuron types in the amygdala of the urodele amphibian Pleurodeles waltl. In a recent study, we used scRNAseq to profile the entire telencephalon of P. waltl [Woych et al., 2022]. Building on this resource, we select molecular markers for clusters of amygdalar glutamatergic neurons and characterize the expression of these markers with in situ hybridization. Our results clarify the boundaries of pallial and subpallial amygdala, reveal the existence of cellular heterogeneity in the pallial amygdala, identify potentially homologous neuron types to the mammalian nucleus of the lateral olfactory tract (NLOT) and medial amygdala, and suggest the presence of a hypothalamus-amygdala migratory stream in amphibians.

Adult P. waltl were obtained from a breeding colony established at Columbia University and maintained in an aquatics facility at 20°C under a 12L: 12D cycle [Joven et al., 2015]. All experiments were conducted in accordance with the NIH guidelines and with the approval of the Columbia University Institutional Animal Care and Use Committee. Experiments were performed with adult (5–19 months) male and female salamanders.

The dataset was generated from scRNAseq of the P. waltl brain as described in Woych et al. [2022]. For in-depth analysis of the amygdala, the data were subsetted, and heterogeneous clusters (TEGLU25, 35 and 38) were subclustered based on differential gene expression. The cluster annotation was updated to reflect the findings from the work described in this paper.

Tissue preparation, in situ hybridization, immunochemistry, and imaging of brain sections were performed as described in Woych et al. [2022] and summarized in brief below. Adult animals were anesthetized by submersion in 0.1% MS-222, perfused with PBS and 4% PFA, and decapitated. Afterwards, brains were extracted, postfixed overnight at 4°C, and washed in PBS.

Following tissue preparation, coronal 70 μm sections were made with a Leica VT1200S vibratome. The sections were blocked in blocking solution for 1 h at RT, then incubated in rabbit anti-Foxg1 (1:1,000, Abcam ab18259) primary antibody solution for 72 h at 4°C. Tissue was washed and incubated for 2 h at room temperature in goat anti-rabbit IgG conjugated to Alexa 594 (1:500, Invitrogen). Sections were again washed, counterstained with DAPI, and mounted on glass slides with DAKO fluorescent mounting medium (Agilent Technologies). Imaging was performed at 20x with a confocal microscope (Zeiss LSM800), and images were processed in Fiji.

Fragments for ISH were either generated by PCR amplification of ∼1 kb sequences from a P. waltl brain cDNA library or ordered from Twist Bioscience and cloned into the pCRII-TOPO vector (Invitrogen) with the Hifi DNA Assembly Kit (NEB). Plasmids were verified by Sanger sequencing and linearized. Antisense RNA probes were then generated by in vitro transcription using Sp6 polymerase and purified with the Monarch RNA Cleanup Kit (NEB). Primers and gene block sequences are provided in online supplementary Table 1 and 2 (for all online suppl. material, see www.karger.com/doi/10.1159/000527899). Following tissue preparation, the sections were postfixed, and the pia mater removed. The sections were permeabilized with Proteinase K and acetylated. At 55–62°C, the sections were pre-hybridized for 1 h in hybridization mix. The sections were then incubated 1–2 nights in hybridization mix with denatured riboprobes (1–2 ng/μL). Two low stringency and two high stringency washes at 55–62°C, and two MABT washes at RT were followed by blocking, and overnight incubation in anti-DIG-Alkaline phosphatase antibody (Roche) solution at 4°C. Sections were washed, and then colorimetric (NBT/BCIP) signal was developed for 1–5 days, with fresh staining solution provided twice a day during development. Staining was stopped by washing in PBS (pH 7.4), and sections were mounted on glass slides with DAKO mounting medium. Imaging was performed at 10x with an upright bright-field microscope (Leica DMR with Basler color camera, ACCU-Slide MS software). Image background was subtracted using Photoshop PS6.

Whole-brain tissue was stained and cleared as described in Woych et al. [2022] using a method of iDISCO modified for HCR-v3.0 [Choi et al., 2018]. Probe pairs were either ordered from Molecular Instruments or designed using the insitu_probe_generator [Kuehn et al., 2021] and ordered from IDT (probe sequences used for generation of probe sets provided in online suppl. Table 3). Brains were incubated in 2–4 pmol of each probe set (Slc17a6, Gad1, Sox6, Sim1, Otp, Penk). Imaging was performed using a LaVision Ultramicroscope II light-sheet microscope at 4x magnification and 2 μm resolution. Images were visualized using ImarisViewer 9.8.0, and virtually downsized (0.4x) and resliced in Fiji [Schindelin et al., 2012]. Sample impurities within the ventricle or on the brain surface were masked and filtered out using ImarisViewer 9.8.0, according to fluorescence intensity.

In a recent scRNAseq study [Woych et al., 2022], we identified 114 clusters of neurons sampled from the adult telencephalon of the urodele amphibian P. waltl. Of these, 41 clusters were annotated as telencephalic glutamatergic neurons, on the basis of the expression of marker genes including the transcription factors Foxg1, Emx1, Tbr1, and Neurod2, the glutamate transporters Slc17a6 and Slc17a7, and for the lack of expression of diencephalic markers [Woych et al., 2022]. After staining brain sections or whole-mount brain preparations for cluster-specific marker genes, we assigned a regional identity to each of these clusters. We identified clusters that belong to the cortical pallium (TEGLU1-24, Fig. 1a [Woych et al., 2022]), and here, we describe the telencephalic glutamatergic clusters (TEGLU25 and TEGLU33-38) that map to distinct glutamatergic cell types in the P. waltl amygdala.

Glutamatergic neuron types in the cortical pallium and in the amygdala can be distinguished by the expression of genes in different combinations and levels, with only a few individual genes marking either group of neurons. Notably, the vesicular glutamate transporters Slc17a6 (Vglut2) and Slc17a7 (Vglut1) are differentially expressed among telencephalic glutamatergic neurons. Specifically, Slc17a7 is expressed at high levels in the entire cortical pallium, but also in the amygdala cluster TEGLU25. Clusters TEGLU33 to TEGLU38 do not express Slc17a7, but instead express Slc17a6 at high levels. Slc17a6 is also found in a few other glutamatergic clusters in the cortical pallium (Fig. 1b, c).

To determine the borders of the amygdala, we characterized the expression of Slc17a6 and Gad1 in the telencephalon in three dimensions. We combined whole-brain hybridization chain reaction (HCR) in situ hybridization, brain clearing with iDISCO, and light-sheet imaging. Here, we present these three-dimensional datasets by slicing them virtually at different angles (Fig. 1d).

From rostral to caudal levels, Slc17a6 is expressed at high levels in the mitral and tufted cells of the main and accessory olfactory bulb, in the post olfactory eminence (an olfactorecipient region in the rostroventral telencephalon previously mistaken for part of the septum, see also Endepols et al. [2005]), and in the medial septum. Starting just posterior to the accessory olfactory bulb, high Slc17a6 expression can be visualized as a longitudinal stripe of cells that extends to the most caudal levels of the telencephalic vesicle (Fig. 1d). Continuing posterior to the level of the interventricular foramen, Slc17a6 is expressed in the lateral cellular prominence (a characteristic thickening of the cellular layer at the border of pallium and subpallium, that corresponds to the caudal part of the lateral amygdala) and along the entire ventral wall of the telencephalic vesicles, in regions previously described as parts of the urodele amygdala. In addition, low expression of Slc17a6 is observed in the caudal region of the medial pallium and in the ventral region of the VP, in line with the scRNAseq data.

In comparison, Gad1, a marker of GABAergic neurons, is expressed rostrally in olfactory bulb interneurons and caudally throughout the subpallium (Fig. 1d). Together, we hypothesize that the lateral stripe of cells expressing high levels of Slc17a6 corresponds to the lateral amygdala (LA) and includes both glutamatergic (pallial) and GABAergic (subpallial) neurons. We propose to refer to these glutamatergic neuron types as lateral pallial amygdala, to distinguish them from the GABAergic neurons they are intermingled with.

To investigate this further and identify the boundary between the VP, the lateral amygdala, and the subpallium, we analyzed the expression of Sox6 (VP), Slc17a6 (VP low and LA high), Gad1 (GABAergic) and Penk (striatum and TEGLU35) (Fig. 1e–g) (see also Woych et al. [2022] for Gad1, Penk and Tac1 in the striatum). At mid-telencephalic levels, double HCR in situ hybridization shows Sox6 expression in the VP. On the other hand, Penk is expressed at high levels in the striatum, and is detected at lower levels in a band of cells just dorsal to the striatum (Fig. 1e, f). Similar to the Slc17a6 stripe, this band of cells starts posterior to the accessory olfactory bulb and ends in the caudal telencephalon (Fig. 1e). Interestingly, we observed that Gad1-expressing cells flank the Sox6-positive VP [Woych et al., 2022], at the same level of the highly-expressing Slc17a6 neurons (Fig. 1f). In more posterior sections, Slc17a6 and Gad1 are clearly coexpressed in the lateral cellular prominence (Fig. 1g). We conclude that the weakly expressing Penk cells correspond spatially to the stripe expressing high levels of Slc17a6. Confirming this, we find that Penk and Slc17a6 are coexpressed in cluster TEGLU35, indicating the existence of Penk-expressing glutamatergic neurons that are not part of the GABAergic striatum (Fig. 1c). Supporting this conclusion further, Lhx9, a well-characterized marker of the lateral amygdala [Moreno et al., 2004; Garcia-Calero and Puelles, 2021], is also expressed in TEGLU35 and stains neurons that occupy the same position as the weakly-expressing Penk cells (Fig. 2).

The spatial distribution of cells expressing these genes, in combination with the scRNAseq data, also clarify the distinction between VP and pallial amygdala. Neither Penk nor Lhx9 are coexpressed with Sox6 in any glutamatergic cluster of the scRNAseq dataset. In contrast, some cells in the VP clusters (TEGLU19, 22–24) do coexpress Sox6 and Slc17a6, consistent with the HCR data, where the Slc17a6 expression domain extends more dorsally than Penk and Lhx9 expression (Fig. 1c–e). We conclude that cluster TEGLU35 corresponds to the lateral amygdala as defined by Moreno, Gonzalez and colleagues, also referred to as the striato-pallial transition area by Marin and colleagues [Marín et al., 1997a; Marín et al., 1997b; Moreno et al., 2004; Moreno and González, 2004; Laberge et al., 2006]. Taken together, these data define the boundaries of the pallial amygdala at mid-telencephalic levels: rostrally with the accessory olfactory bulb, dorsally with the VP, and ventrally with the striatum.

To systematically explore the molecular diversity of the pallial amygdala, we subclustered glutamatergic amygdala cells (Fig. 2a). Some clusters were found to contain multiple distinct cell subtypes, and so we split clusters TEGLU25, TEGLU35, and TEGLU38 (indicated as TEGLU25.1 and 25.2, TEGLU35.1 and 35.2, and TEGLU38.1 and 38.2, respectively) (Fig. 2a–c). Together, our analysis of 1,417 single-cell transcriptomes suggests the existence of at least 10 distinct glutamatergic neuron types in the amygdala of P. waltl. To determine the exact spatial distribution of these amygdala neuron types, we drew on the scRNAseq dataset to identify cell-type specific marker genes (Fig. 2b, c, see below and Fig. 1c).

Focusing on the lateral pallial amygdala, our data show that this region is heterogeneous, as exemplified by Lhx9 and Meis1 expression at different levels in cells in clusters TEGLU35.1 and 35.2 (Fig. 2b, c). In situ hybridization for these genes shows that they are expressed in the Slc17a6/Penk stripe of the lateral amygdala (Fig. 2d), recognizable in sections posterior to the intraventricular foramen as the lateral cellular prominence. The neuropeptide precursor gene Tac1 is a specific marker of cluster TEGLU34. Expression analysis shows its presence in a subset of cells located in the lateral amygdala, further away from the ventricle compared to Lhx9. These Tac1-positive cells, or a subset of them, also express Fezf2 (Fig. 2c, d). Additionally, Calb1, detected in cluster TEGLU35.1, is expressed in scattered cells of the lateral amygdala (Fig. 2c and Morona and González [2008]).

Next, we investigated the expression of the transcription factor Sim1, found at high levels in clusters TEGLU25.1 and TEGLU25.2 and at low levels in clusters TEGLU37 and TEGLU38.1. To distinguish between these clusters, we also mapped the expression of Syt2 (Synaptotagmin 2), a gene coexpressed with Sim1 only in clusters TEGLU25.1 and TEGLU25.2 (Fig. 2c, d). In situ hybridization on brain sections and whole-mount preparations show expression of Sim1 at high levels in a large periventricular population of cells, along the ventrocaudal wall of the telencephalic vesicle, overlapping with the Syt2 expression domain (Fig. 2d, online suppl. File 1). This population is immediately caudal to the striatum, medial to the Lhx9/Penk sector of the lateral amygdala, and dorsolateral to the central amygdala (see also Fig. 1d; online suppl. Fig. S1). A previous analysis of transcriptomic similarity across telencephalic clusters indicated that TEGLU25 is more similar to the VP than to other clusters from the pallial amygdala [Woych et al., 2022]. However, the Sim1 expression pattern clearly indicates that this cluster maps to a region of the amygdala, and not to the cortical pallium. In addition, unlike the lateral amygdala clusters, TEGLU25 is expressing a series of genes also detected in the mouse nucleus of the lateral olfactory tract (NLOT2), such as Tbr1, Lhx2, Nr2f2 (Coup-tf2), Etv1, Ebf3, NeuroD2, and NeuroD6 (Fig. 2c) [Remedios et al., 2007; Tang et al., 2012; Garcia-Calero et al., 2020; Aerts and Seuntjens, 2021]. In light of this, here, we will refer to this newly defined amygdalar region as the NLOT-like area (see also Discussion).

Taken together, these results indicate that the salamander amygdala is heterogeneous and includes multiple neuron types. We refer to the glutamatergic cells of clusters TEGLU33-TEGLU36 as the lateral pallial amygdala (lpA), in order to distinguish these cell types from the GABAergic cells in the lateral amygdala.

The Sim1 expression of clusters TEGLU25.1 and TEGLU25.2 is mapped to a region here called the NLOT-like area. While all the glutamatergic neuron types in the amygdala express Slc17a6, Nr2f2, Foxg1, and Nos1, clusters TEGLU37-TEGLU38 are devoid of Tbr1 expression (Fig. 2c). As described below, these two clusters correspond to glutamatergic cells sampled from the medial amygdala (MeA).

From our analysis of the glutamatergic amygdala dataset, we found Sim1 expression at low levels in clusters TEGLU37 and TEGLU38.1 (Fig. 2b, c). In these clusters, Sim1 expression overlaps with high expression of the transcription factor Otp (Fig. 3a). Previous work in mammals [García-Moreno et al., 2010], lizards, and birds [Metwalli et al., 2022] identified a population of Sim1+ Otp+ neurons that migrate from the periventricular hypothalamus to the medial amygdala; these neurons are born in a territory called telencephalon-opto-hypothalamic domain [Morales et al., 2021] and migrate through the hypothalamo-amygdala corridor [Garcia-Calero et al., 2021]. We asked whether Sim1+ Otp+ neurons also migrate from the periventricular hypothalamus to the medial amygdala in P. waltl. Therefore, we analyzed the expression of Otp and Sim1 using whole-mount in situ hybridization in the P. waltl brain (Fig. 3b, c, online suppl. File 2). As visualized in optical horizontal sections through the telencephalon, Otp-expressing cells lie in the same position as the weakly expressing Sim1 cells in the medial amygdala. Using 3D rendering, we observed that the Otp expression domain is a continuous stripe of cells that goes from the periventricular hypothalamus to the medial amygdala and terminates just posterior to the cells expressing high levels of Sim1 in the NLOT-like area described above (Fig. 3c, online suppl. File 3).

According to our scRNAseq data, neurons in clusters TEGLU37 and TEGLU38 express Zic1 and Lhx5, markers that are also found in the frog medial amygdala (Fig. 3a) [Jiménez and Moreno, 2021]. Furthermore, these neurons express Foxg1 but not Tbr1 (Fig. 2c), similar to the Sim1+ neurons that populate the mouse medial amygdala [Garcia-Calero et al., 2021]. Finally, we find that the Bombesin-like receptor 3 gene (Brs3), a specific marker of Sim1+ neurons in the mouse medial amygdala, is also expressed in clusters TEGLU37 and TEGLU38.2 (Fig. 3a) [Xiao et al., 2020]. Taken together, these data indicate the existence of Sim1+ Otp+ neurons that might originate from the periventricular hypothalamus and populate the medial amygdala in P. waltl, with a molecular profile similar to the Sim1+ Otp+ neurons of the mouse medial amygdala.

In mammals, the amygdala is a heterogeneous collection of nuclei composed of neurons with diverse developmental origins and complex migratory histories. Reconstructing the evolution of the amygdala is further complicated by the fact that the amygdala is part of the caudal telencephalon, a brain region that underwent substantial changes during vertebrate evolution.

At a first glance, amphibians may seem easy to analyze because they have a “simple” brain. However, the absence of clear neuroanatomical landmarks in the amphibian telencephalon, where distinct neuron types lay next to each other in a continuous periventricular cell layer, complicates the identification of boundaries and subregions. This apparent continuum of cells hides a relatively large repertoire of neuron types, which we were able to identify in an unbiased manner by using scRNAseq data as the starting point for histological studies. After careful analysis of the spatial distribution of these neuron types, we propose the existence of boundaries that divide this seeming continuum of cells into specific subregions. Although some of these boundaries are not sharply defined (in comparison to amygdala nuclei in amniotes, for example), they still demarcate areas with distinct neuron types, and guide the readers to specific neuron types and regions in the P. waltl brain (see also Woych et al. [2022] for a brain atlas).

On the basis of this analysis, we propose that the VP territory (labeled by the expression of the transcription factor Sox6) and the adjacent pallial amygdala are two distinct subdivisions of the salamander pallium. This conclusion is primarily based on transcriptomics data, which indicate extensive molecular differences between glutamatergic neurons in the amygdala and cortical pallium [Woych et al., 2022]. The differences arise mostly from the expression of genes in different combinations, rather than from the presence of a large number of region-specific markers, as is the case for all other subdivisions of the pallium. Our conclusions are in line with recent models of the pallium in amniotes [Desfilis et al., 2018; Puelles et al., 2019]. However, the possibility that the salamander VP (Sox6+ cells, as defined here) and pallial amygdala belong to the same histogenic domain, as proposed earlier in the literature, cannot be fully excluded. Further work on the development of the amphibian pallium is needed to test whether VP and pallial amygdala neurons are born from distinct sets of neural progenitors.

Interestingly, glutamatergic neurons in the pallial amygdala and cortical pallium differ in their expression of vesicular glutamate transporters. In the amygdala, glutamatergic neurons express Slc17a6 (Vglut2), whereas in the cortical pallium, glutamatergic neurons express Slc17a7 (Vglut1). Exceptions to this rule include the weak expression of Slc17a6 in parts of the VP, and the strong expression of Slc17a7 in the newly-defined NLOT-like area (cluster TEGLU25). Although not as clearly distinct as in P. waltl, Slc17a6 and Slc17a7 are expressed in a largely complementary fashion in the mammalian brain. While Slc17a7 is predominantly expressed in the cerebral cortex (including the piriform cortex and the hippocampus) and in the cerebellum, Slc17a6 is expressed in specific cortical layers and regions, as well as subcortical regions, the thalamus, and the brainstem. In the amygdala specifically, medial and central nuclei predominantly express Slc17a6, while lateral and basolateral nuclei express Slc17a7 at higher levels [Fremeau et al., 2001; Wallén-Mackenzie et al., 2010; Vigneault et al., 2015]. Remarkably, in the turtle Trachemys scripta, Slc17a6 and Slc17a7 are coexpressed in all pallial glutamatergic neurons [Tosches et al., 2018], suggesting that the regulation of these two transporters in the pallium has changed during tetrapod evolution.

The expression of Slc17a6 helps to define the boundaries of the amygdala in urodeles. Mapping the expression of Slc17a6 and Gad1 reveals a complex regional distribution of glutamatergic and GABAergic neurons. Our data suggest that glutamatergic and GABAergic neurons are intermingled in the lateral amygdala (named by Moreno and Gonzalez [Moreno et al., 2004; Moreno and González, 2004]), as clearly indicated by the expression of both Slc17a6 and Gad1 in the lateral cellular prominence. Furthermore, the sharp boundary of Slc17a6 and Gad1 expression in the ventrocaudal telencephalic wall revealed the presence of a new region, the NLOT-like area, nested between the striatum anteriorly and the medial amygdala posteriorly. Finally, we confirm that the medial amygdala, which occupies the most caudal part of the telencephalon, is also a mixture of glutamatergic and GABAergic neuron types.

A peculiarity of the amniote amygdala is the presence of multiple nuclei. In amphibians, only one pallial nucleus (the lateral amygdala) and two subpallial nuclei (the medial amygdala and the central amygdala) have been reported in the literature. From a comparative perspective, this has two alternative explanations: either the amphibian amygdala is simpler than the amygdala of amniotes (because it is more primitive or as a result of secondary simplification) or neuronal and regional diversity were overlooked in the amphibian amygdala for the absence of clear morphological landmarks. The analysis of neuron types, enabled by the scRNAseq approach, reveals that although the amygdala in urodeles is not as complex as in amniotes, there is still more cellular diversity than previously anticipated.

In the lateral amygdala alone, we identified at least 5 distinct clusters or subclusters of glutamatergic neurons (clusters TEGLU33, 34, 35.1, 35.2, and 36). We propose to call these neuron types lateral pallial amygdala, to distinguish them from the GABAergic neurons they are intermingled with. These neurons are labeled by the differential expression of several markers, including Lhx9, Meis1, Calb1, Tac1, and Penk. These were all detected at the pallial-subpallial boundary (at mid-telencephalic levels) and in the lateral cellular prominence (from levels caudal to the intraventricular foramen), confirming the lateral amygdala assignment of these clusters. Importantly, the cell type analysis reveals that Lhx9, a marker that has been used to identify the lateral amygdala in amphibians [Moreno et al., 2004], is expressed only in a subset of the lateral amygdala glutamatergic neurons. Interestingly, some of these markers appear to label cells with different distances from the ventricle (superficial or deep), suggesting that distinct amygdalar glutamatergic types might be generated sequentially during development, as it is the case for the rest of the amphibian pallium [Moreno and González, 2017].

The recent radial model for the mammalian pallial amygdala posits that the pallial amygdala develops from five radial sectors (lateral, basal, anterior, posterior, and retroendopiriform) and that distinct amygdalar nuclei are formed by neurons generated sequentially within each of these sectors [Garcia-Calero and Puelles, 2021]. For example, the anterior basomedial amygdala (BMA) and the Anterior Cortical Amygdala (ACo) are generated within the same radial domain, where ACo neurons are born first, and BMA neurons are born later [Garcia-Calero et al., 2020]. The exact number of radial domains in the pallial amygdala of non-mammals, including amphibians, remains undetermined. On the basis of our data, we can speculate that five distinct radial units did not exist in tetrapod ancestors, because we cannot find evidence of radial units of the pallial amygdala in P. waltl. It is conceivable that new specialized subdivisions of the pallial amygdala emerged in amniote or in mammalian ancestors together with the expansion of the rest of the telencephalon. Work on amygdala development in a variety of vertebrate species will be needed to reconstruct the evolution of these radial domains.

Our scRNAseq data on the adult salamander add a cell type perspective to the comparison of amygdala nuclei across species. On the basis of hodological data, the amphibian lateral amygdala has been compared to several amygdalar nuclei in amniotes, including the mammalian basolateral complex and cortical amygdala [Moreno and González, 2006]. The amphibian lateral amygdala is extensively connected with other telencephalic areas (septum, bed nucleus of the stria terminalis, striatum), receives inputs from the thalamus and the parabrachial area, and projects to the ventromedial hypothalamus through the stria terminalis [Woych et al., 2022]. In addition, the amphibian lateral amygdala receives inputs from both the main and the accessory olfactory bulb [Woych et al., 2022]. In amniotes, olfactory inputs and other sensory inputs relayed through the thalamus reach distinct sets of amygdala nuclei. In mammals, the superficial cortical amygdala (early born neurons according to the radial model) receives olfactory inputs, whereas the basolateral complex (lateral, basolateral, and basomedial amygdala, late-born neurons in the radial model) receives multimodal thalamic inputs (lateral amygdala). The latter is reciprocally connected with the rest of the telencephalon, and projects to the ventromedial hypothalamus and other extra-telencephalic targets. In lizards, olfactory inputs reach the posterior lateral cortex and the nucleus sphericus, whereas the dorsolateral amygdala (DLA) and other parts of the posterior DVR have connections analogous to the mammalian basolateral complex [Voneida and Sligar, 1979; Bruce and Neary, 1995; Novejarque et al., 2004; Norimoto et al., 2020]. From an evolutionary perspective, these observations can be explained by at least two alternative hypotheses. Each neuron in the amphibian lateral amygdala might integrate both olfactory and nonolfactory information, and the separate processing of these modalities by distinct neurons and amygdalar nuclei may have evolved only in amniotes [Moreno and González, 2007b]. Alternatively, distinct neurons in the amphibian lateral amygdala would belong to separate subcircuits, which became segregated in distinct nuclei in amniotes. The surprising cellular heterogeneity of the P. waltl lateral amygdala supports the latter scenario. Lhx9, a transcription factor used previously to support the homology of the amphibian lateral amygdala with amniote pallial amygdala nuclei, is expressed only in a subset of the P. waltl lateral amygdala glutamatergic neurons. In E12.5 mice, Lhx9 is expressed at high levels in the anterior radial unit, which includes BMA and ACo, but not in the basal and lateral units (precursors of the basolateral and lateral amygdalar nuclei, among others) [Abellán et al., 2014; Garcia-Calero and Puelles, 2021]. Interestingly, BMA and ACo not only develop from the same radial unit but are also considered part of the same functional system [Petrovich et al., 1996]: ACo integrates inputs from the main and the accessory olfactory bulb, and is reciprocally connected with the BMA [Cádiz-Moretti et al., 2017]; the BMA instead is connected with olfactory cortical areas and sends projections to the ventromedial hypothalamus through the stria terminalis [Petrovich et al., 1996].

In contrast, the mammalian basolateral amygdala (BLA), which develops from the basal unit (not to be confused with the mammalian lateral amygdala, a derivative of the lateral unit; see [Garcia-Calero and Puelles, 2021] for terminology), expresses a distinct set of transcription factors, including the specific markers Fezf2 and Etv1 [Hirata-Fukae and Hirata, 2014; O’Leary et al., 2020]. In Fezf2 mutant mice, BLA glutamatergic neurons die after cell cycle exit, suggesting that Fezf2 is a terminal selector transcription factor for BLA identity [Hirata-Fukae and Hirata, 2014; O’Leary et al., 2020]. In P. waltl, the lateral amygdala region includes Fezf2-expressing neurons (clusters TEGLU33 and 34), which are distinct from the Lhx9-expressing neurons described above. However, these neurons do not coexpress Etv1, according to our scRNAseq and in situ hybridization data. In reptiles, neurons coexpressing Etv1 and Fezf2 have been identified in the pDVR of turtles and lizards [Tosches et al., 2018; Norimoto et al., 2020], suggesting that the gene regulatory program that establishes the coexpression of these two transcription factors exists only in amniotes, but not in amphibians (or at least not in salamanders; see Jiménez and Moreno [2021]). Additional cellular transcriptomics data in mammals, birds, reptiles, and other amphibians are needed to understand the evolutionary relationships of neuron types in the amphibian lateral amygdala, reptilian and avian pDVR, and mammalian cortical amygdala and basolateral complex.

Finally, the presence of GABAergic neurons in the lateral amygdala of P. waltl may provide new insights on the evolution of the mammalian intercalated cells (ITCs). In adult mice, ITCs surround the lateral and the basolateral amygdala, and populate the intercalated mass of the amygdala (IA), located between pallial and the subpallial amygdalar nuclei. Fate mapping experiments showed that unlike other amygdala interneurons, which originate in the ventral lateral ganglionic eminence (vLGE), ITCs develop from the dorsal LGE (dLGE), and therefore share their developmental origin with medium spiny neurons of the striatum [Waclaw et al., 2010; Bandler et al., 2022]. During development, the mammalian dLGE is immediately ventral to the pallium, in a position that is topologically homologous to the most dorsal part of the subpallium in amphibians. In mammals, ITCs migrate after birth in a ventrocaudal direction, through the lateral migratory stream, to reach their final destination in the amygdala.

Mouse ITCs express a unique set of transcription factors and marker genes, and the analysis of these markers in our scRNAseq data suggests the existence of ITC-like neurons in P. waltl. In mice, ITCs are Penk-positive, but unlike striatal Penk cells, they express the transcription factor Foxp2 instead of its paralog Foxp1. Furthermore, ITCs are distinct from the striatum for the expression of the transcription factors Tshz1 and Pax6 [Kuerbitz et al., 2018]; they also express Sp8 and Meis2 (like the striatum), but lack the expression of the striatal marker Isl1. In turtles and lizards, LGE-derived neurons with a ITC-like expression profile have been identified; these neurons coexpress Foxp2, Penk, Tshz1, Pbx3, and Meis2, and the Foxp2 expression pattern indicates their location in the pDVR [Tosches et al., 2018]. In P. waltl, a subcluster of Penk-expressing GABAergic neurons (TEGABA55 [Woych et al., 2022]) coexpress Penk, Foxp2, Tshz1, Pax6, Pbx3, and Meis2, but are negative for Foxp1 and Isl1 (online suppl. Fig. S2). Both Penk and Pax6 are expressed in the lateral amygdala [Joven et al., 2013], suggesting that the GABAergic neurons of the lateral amygdala correspond to this subcluster. Taken together, these data suggest that ITC GABAergic neurons originating from the dLGE are ancestral in tetrapods.

Here, we describe a new division of the amygdala in P. waltl, which we provisionally name NLOT-like area in light of its gene expression profile (cluster TEGLU25). The scRNAseq analysis indicates that these neurons are extremely different, at a molecular level, from neurons in the lateral amygdala. The most distinctive feature of cluster TEGLU25 is the expression of some cortical pallium markers, such as Slc17a7, NeuroD2, and NeuroD6, which initially misled us to think that TEGLU25 was part of the cortical pallium. However, the expression pattern of the transcription factor Sim1 in the adult brain shows unambiguously that TEGLU25 neurons are part of the amygdala, with a lateral boundary with the lateral amygdala, a rostral boundary with the striatum and a caudal boundary with the medial amygdala. This region of the amphibian brain is considered part of the medial amygdala in the literature [Moreno and González, 2007c]. Our scRNAseq data suggest an alternative interpretation. The expression of several genes reveals similarities of these P. waltl Sim1-expressing cells with cells in the mammalian NLOT. Like the mammalian NLOT, this region of the urodele telencephalon is receiving inputs from the main olfactory bulb [Laberge and Roth, 2005; Moreno and González, 2007c]. In mice, Sim1 is expressed in layer 2 of the NLOT (NLOT2), together with other markers such as Tbr1, Lhx2, Etv1, NeuroD2, NeuroD6, Zic2, and Dach1 [Remedios et al., 2007; Aerts, and Seuntjens, 2021; Garcia-Calero et al., 2021]. Mouse NLOT2 neurons arise from multiple progenitor domains: the hypothalamus (source of Sim1+ neurons), and the posterior amygdalar radial unit (source of TBR1+ neurons) [Remedios et al., 2007; Garcia-Calero et al., 2020]. From these birthplaces, Sim1+ (mRNA) and TBR1+ (protein) neurons are thought to migrate rostrally along converging routes to populate NLOT2 [Remedios et al., 2007; Aerts and Seuntjens, 2021; Garcia-Calero et al., 2021]. Interestingly, the P. waltl TEGLU25 cluster coexpresses Sim1 and Tbr1 mRNAs, together with other markers of NLOT2 (in TEGLU25.1, TEGLU25.2, or both) such as Lhx2, Nr2f2 (Coup-tf2), Etv1, Ebf3, NeuroD2, and NeuroD6 (but not Zic2 and Dach1) [Remedios et al., 2007; Tang et al., 2012; Garcia-Calero et al., 2020; Aerts and Seuntjens, 2021]. Sim1+ neurons have also been found in the chick LOT nucleus, putative homolog of the mammalian NLOT [García-Moreno et al., 2010; Morales et al., 2021; Garcia-Calero et al., 2021; Metwalli et al., 2022]. Taken together, these data suggest that a homolog of the mammalian NLOT nucleus, or some of its cell types, may have existed in tetrapod ancestors. However, the developmental origins of NLOT-like neurons in P. waltl remain obscure. The expression of “cortical pallium” markers (Tbr1, NeuroD2, NeuroD6) may suggest a pallial origin; although, the expression of Sim1 is in conflict with this interpretation. Furthermore, some “cortical pallium” markers are expressed in the mouse preoptic area and hypothalamus (Tbr1: [Bulfone et al., 1995; Puelles et al., 2000; Romanov et al., 2020], NeuroD6: [Caqueret et al., 2006]). However, it cannot be excluded that this is the result of cell migration from the pallium or the prethalamic eminence, at least in some cases. Further analysis of developmental stages in P. waltl and other amphibians is needed to determine the developmental origins of NLOT-like cells.

The NLOT-like area is distinct from a second population of Sim1-expressing neurons immediately caudal to it. These neurons co-express Sim1 and Otp (clusters TEGLU37 and TEGLU38) and lack Tbr1 expression; Otp whole-mount in situ hybridization reveals a continuous population of neurons from the periventricular hypothalamus to the caudal telencephalon. For their position and their characteristic molecular profile, we propose that these neurons are homologs of the Sim1+ Otp+ neurons of the mammalian medial amygdala, recently also identified in sauropsids [García-Moreno et al., 2010; Morales et al., 2021; Garcia-Calero et al., 2021; Metwalli et al., 2022]. Unlike in the amniote situation, where distinct Sim1+ Foxg1+ and Sim1+ Foxg1 negativeneurons have been identified, our staining data indicate so far that the medial amygdala Sim1+ Otp+ cells co-express Foxg1 in P. waltl, suggesting that the amniote amygdala harbors a larger variety of neuron types. Taken together, our results establish that Sim1+ Otp+ Foxg1+ neurons of the hypothalamic-amygdalar corridor [Garcia-Calero et al., 2021], also referred to as telencephalon-opto-hypothalamic domain [Morales et al., 2021], were present in the last common ancestor of tetrapods and contributed glutamatergic neurons to the medial amygdala. The existence of Otp+ neurons in the amygdala of zebrafish [Porter and Mueller, 2020] suggests that this neuron type may trace back to vertebrate ancestors, although additional work in fish and cyclostomes is needed to test this hypothesis.

There are mixed opinions in the literature on the relevance of salamanders for brain evolution studies. Herrick [1948] assumed that urodele amphibians retained the ancestral characters of the brains of tetrapod ancestors, while Roth et al. [1993] pointed out that the morphological simplicity of the salamander brain evolved secondarily as a result of paedomorphosis (retention of embryonic or juvenile characters in adulthood). These analyses, however, were limited to neuroanatomical traits, such as the migration of newly born neurons away from the ventricle. Our molecular analysis of neuron types in P. waltl indicates that the salamander telencephalon hosts an unanticipated variety of neuron types, which can be readily compared to neuron types in amniotes [Woych et al., 2022]. This suggests that brain morphology and neuronal diversity are partially uncoupled in evolution, and that brain structures with higher or lower degrees of morphological complexity can harbor homologous neuron types. The independent evolution of brain morphology and neuron types is an example of mosaic evolution, where biological traits evolve at different rates under different selective pressures [Northcutt and Kicliter, 1980]. In light of this, the simple organization of the salamander brain can be exploited to accelerate the discovery of neuron types and neural circuits shared across tetrapods, provided that their phylogenetic continuity can be demonstrated. In the case of the amygdala, we conclude that the amygdala of stem tetrapods had a heterogeneous repertoire of glutamatergic neurons, including distinct Lhx9-positive and Lhx9-negative neurons in the pallial amygdala, Sim1+ neurons putative homologs to the NLOT, and conserved Sim1+ Otp+ neurons in the medial amygdala. Further characterization of amygdalar neuron types in developmental and adult stages of amniotes and anamniotes is needed to reconstruct with higher precision the evolution of this incredibly complex part of the brain.

The authors are grateful to J. Barber, S. Cook, and the Columbia University Institute of Comparative Medicine for animal care, E. Gumnit for help with cloning, and A. Ortega Gurrola for critical feedback on the manuscript. Light-sheet imaging was performed with support from L. Hammond and the Zuckerman Institute’s Cellular Imaging platform (NIH 1S10OD023587-01).

This study protocol was reviewed and approved by the Columbia University Institutional Animal Care and Use Committee (IACUC).

The authors have no conflicts of interest to declare.

This work was supported by the McKnight Foundation (M.A.T.), the National Institute of Health (grant RM1HG011014 to M.A.T.), the European Molecular Biology Organization (Long-Term Fellowship ALTF 874-2021 to A.D.), and the National Science Foundation (Graduate Research Fellowship DGE 2036197 to E.C.J.).

Astrid Deryckere was responsible for scRNAseq analysis. Anatomy and histology data were collected by Jamie Woych, Astrid Deryckere, and Eliza C. B. Jaeger. The data were analyzed and manuscript drafted by Maria Antonietta Tosches, Astrid Deryckere, and Jamie Woych, with manuscript edits contributed by Eliza Jaeger. Maria Antonietta Tosches led management and supervision of the project.

The scRNA sequencing data used in this study have been deposited in the Gene Expression Omnibus (GEO), with accession numbers GSE197701, GSE197722, GSE197796, GSE197807, and GSE198363, as reported in Woych et al. [2022]. A website to access the data and search genes of interest is available at https://toscheslab.shinyapps.io/salamander_telencephalon/. Further inquiries can be directed to the corresponding author.

Astrid Deryckere and Jamie Woych contributed equally to this work.